fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dir2_sf.h"
  33 #include "xfs_attr_sf.h"
  34 #include "xfs_dinode.h"
  35 #include "xfs_inode.h"
  36 #include "xfs_btree.h"
  37 #include "xfs_dmapi.h"
  38 #include "xfs_mount.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_itable.h"
  41 #include "xfs_dir2_data.h"
  42 #include "xfs_dir2_leaf.h"
  43 #include "xfs_dir2_block.h"
  44 #include "xfs_inode_item.h"
  45 #include "xfs_extfree_item.h"
  46 #include "xfs_alloc.h"
  47 #include "xfs_bmap.h"
  48 #include "xfs_rtalloc.h"
  49 #include "xfs_error.h"
  50 #include "xfs_attr_leaf.h"
  51 #include "xfs_rw.h"
  52 #include "xfs_quota.h"
  53 #include "xfs_trans_space.h"
  54 #include "xfs_buf_item.h"
  55 #include "xfs_filestream.h"
  56 #include "xfs_vnodeops.h"
  57
  58
  59 #ifdef DEBUG
  60 STATIC void
  61 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  62 #endif
  63
  64 kmem_zone_t             *xfs_bmap_free_item_zone;
  65
  66 /*
  67  * Prototypes for internal bmap routines.
  68  */
  69
  70
  71 /*
  72  * Called from xfs_bmap_add_attrfork to handle extents format files.
  73  */
  74 STATIC int                                      /* error */
  75 xfs_bmap_add_attrfork_extents(
  76         xfs_trans_t             *tp,            /* transaction pointer */
  77         xfs_inode_t             *ip,            /* incore inode pointer */
  78         xfs_fsblock_t           *firstblock,    /* first block allocated */
  79         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  80         int                     *flags);        /* inode logging flags */
  81
  82 /*
  83  * Called from xfs_bmap_add_attrfork to handle local format files.
  84  */
  85 STATIC int                                      /* error */
  86 xfs_bmap_add_attrfork_local(
  87         xfs_trans_t             *tp,            /* transaction pointer */
  88         xfs_inode_t             *ip,            /* incore inode pointer */
  89         xfs_fsblock_t           *firstblock,    /* first block allocated */
  90         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  91         int                     *flags);        /* inode logging flags */
  92
  93 /*
  94  * Called by xfs_bmapi to update file extent records and the btree
  95  * after allocating space (or doing a delayed allocation).
  96  */
  97 STATIC int                              /* error */
  98 xfs_bmap_add_extent(
  99         xfs_inode_t             *ip,    /* incore inode pointer */
 100         xfs_extnum_t            idx,    /* extent number to update/insert */
 101         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 102         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 103         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 104         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 105         int                     *logflagsp, /* inode logging flags */
 106         xfs_extdelta_t          *delta, /* Change made to incore extents */
 107         int                     whichfork, /* data or attr fork */
 108         int                     rsvd);  /* OK to allocate reserved blocks */
 109
 110 /*
 111  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 112  * allocation to a real allocation.
 113  */
 114 STATIC int                              /* error */
 115 xfs_bmap_add_extent_delay_real(
 116         xfs_inode_t             *ip,    /* incore inode pointer */
 117         xfs_extnum_t            idx,    /* extent number to update/insert */
 118         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 119         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 120         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 121         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 122         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 123         int                     *logflagsp, /* inode logging flags */
 124         xfs_extdelta_t          *delta, /* Change made to incore extents */
 125         int                     rsvd);  /* OK to allocate reserved blocks */
 126
 127 /*
 128  * Called by xfs_bmap_add_extent to handle cases converting a hole
 129  * to a delayed allocation.
 130  */
 131 STATIC int                              /* error */
 132 xfs_bmap_add_extent_hole_delay(
 133         xfs_inode_t             *ip,    /* incore inode pointer */
 134         xfs_extnum_t            idx,    /* extent number to update/insert */
 135         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 136         int                     *logflagsp,/* inode logging flags */
 137         xfs_extdelta_t          *delta, /* Change made to incore extents */
 138         int                     rsvd);  /* OK to allocate reserved blocks */
 139
 140 /*
 141  * Called by xfs_bmap_add_extent to handle cases converting a hole
 142  * to a real allocation.
 143  */
 144 STATIC int                              /* error */
 145 xfs_bmap_add_extent_hole_real(
 146         xfs_inode_t             *ip,    /* incore inode pointer */
 147         xfs_extnum_t            idx,    /* extent number to update/insert */
 148         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 149         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 150         int                     *logflagsp, /* inode logging flags */
 151         xfs_extdelta_t          *delta, /* Change made to incore extents */
 152         int                     whichfork); /* data or attr fork */
 153
 154 /*
 155  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 156  * allocation to a real allocation or vice versa.
 157  */
 158 STATIC int                              /* error */
 159 xfs_bmap_add_extent_unwritten_real(
 160         xfs_inode_t             *ip,    /* incore inode pointer */
 161         xfs_extnum_t            idx,    /* extent number to update/insert */
 162         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 163         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 164         int                     *logflagsp, /* inode logging flags */
 165         xfs_extdelta_t          *delta); /* Change made to incore extents */
 166
 167 /*
 168  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 169  * It figures out where to ask the underlying allocator to put the new extent.
 170  */
 171 STATIC int                              /* error */
 172 xfs_bmap_alloc(
 173         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 174
 175 /*
 176  * Transform a btree format file with only one leaf node, where the
 177  * extents list will fit in the inode, into an extents format file.
 178  * Since the file extents are already in-core, all we have to do is
 179  * give up the space for the btree root and pitch the leaf block.
 180  */
 181 STATIC int                              /* error */
 182 xfs_bmap_btree_to_extents(
 183         xfs_trans_t             *tp,    /* transaction pointer */
 184         xfs_inode_t             *ip,    /* incore inode pointer */
 185         xfs_btree_cur_t         *cur,   /* btree cursor */
 186         int                     *logflagsp, /* inode logging flags */
 187         int                     whichfork); /* data or attr fork */
 188
 189 /*
 190  * Called by xfs_bmapi to update file extent records and the btree
 191  * after removing space (or undoing a delayed allocation).
 192  */
 193 STATIC int                              /* error */
 194 xfs_bmap_del_extent(
 195         xfs_inode_t             *ip,    /* incore inode pointer */
 196         xfs_trans_t             *tp,    /* current trans pointer */
 197         xfs_extnum_t            idx,    /* extent number to update/insert */
 198         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 199         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 200         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 201         int                     *logflagsp,/* inode logging flags */
 202         xfs_extdelta_t          *delta, /* Change made to incore extents */
 203         int                     whichfork, /* data or attr fork */
 204         int                     rsvd);   /* OK to allocate reserved blocks */
 205
 206 /*
 207  * Remove the entry "free" from the free item list.  Prev points to the
 208  * previous entry, unless "free" is the head of the list.
 209  */
 210 STATIC void
 211 xfs_bmap_del_free(
 212         xfs_bmap_free_t         *flist, /* free item list header */
 213         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 214         xfs_bmap_free_item_t    *free); /* list item to be freed */
 215
 216 /*
 217  * Convert an extents-format file into a btree-format file.
 218  * The new file will have a root block (in the inode) and a single child block.
 219  */
 220 STATIC int                                      /* error */
 221 xfs_bmap_extents_to_btree(
 222         xfs_trans_t             *tp,            /* transaction pointer */
 223         xfs_inode_t             *ip,            /* incore inode pointer */
 224         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 225         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 226         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 227         int                     wasdel,         /* converting a delayed alloc */
 228         int                     *logflagsp,     /* inode logging flags */
 229         int                     whichfork);     /* data or attr fork */
 230
 231 /*
 232  * Convert a local file to an extents file.
 233  * This code is sort of bogus, since the file data needs to get
 234  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 235  */
 236 STATIC int                              /* error */
 237 xfs_bmap_local_to_extents(
 238         xfs_trans_t     *tp,            /* transaction pointer */
 239         xfs_inode_t     *ip,            /* incore inode pointer */
 240         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 241         xfs_extlen_t    total,          /* total blocks needed by transaction */
 242         int             *logflagsp,     /* inode logging flags */
 243         int             whichfork);     /* data or attr fork */
 244
 245 /*
 246  * Search the extents list for the inode, for the extent containing bno.
 247  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 248  * *eofp will be set, and *prevp will contain the last entry (null if none).
 249  * Else, *lastxp will be set to the index of the found
 250  * entry; *gotp will contain the entry.
 251  */
 252 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 253 xfs_bmap_search_extents(
 254         xfs_inode_t     *ip,            /* incore inode pointer */
 255         xfs_fileoff_t   bno,            /* block number searched for */
 256         int             whichfork,      /* data or attr fork */
 257         int             *eofp,          /* out: end of file found */
 258         xfs_extnum_t    *lastxp,        /* out: last extent index */
 259         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 260         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 261
 262 /*
 263  * Check the last inode extent to determine whether this allocation will result
 264  * in blocks being allocated at the end of the file. When we allocate new data
 265  * blocks at the end of the file which do not start at the previous data block,
 266  * we will try to align the new blocks at stripe unit boundaries.
 267  */
 268 STATIC int                              /* error */
 269 xfs_bmap_isaeof(
 270         xfs_inode_t     *ip,            /* incore inode pointer */
 271         xfs_fileoff_t   off,            /* file offset in fsblocks */
 272         int             whichfork,      /* data or attribute fork */
 273         char            *aeof);         /* return value */
 274
 275 #ifdef XFS_BMAP_TRACE
 276 /*
 277  * Add bmap trace entry prior to a call to xfs_iext_remove.
 278  */
 279 STATIC void
 280 xfs_bmap_trace_delete(
 281         const char      *fname,         /* function name */
 282         char            *desc,          /* operation description */
 283         xfs_inode_t     *ip,            /* incore inode pointer */
 284         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
 285         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
 286         int             whichfork);     /* data or attr fork */
 287
 288 /*
 289  * Add bmap trace entry prior to a call to xfs_iext_insert, or
 290  * reading in the extents list from the disk (in the btree).
 291  */
 292 STATIC void
 293 xfs_bmap_trace_insert(
 294         const char      *fname,         /* function name */
 295         char            *desc,          /* operation description */
 296         xfs_inode_t     *ip,            /* incore inode pointer */
 297         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
 298         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
 299         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
 300         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
 301         int             whichfork);     /* data or attr fork */
 302
 303 /*
 304  * Add bmap trace entry after updating an extent record in place.
 305  */
 306 STATIC void
 307 xfs_bmap_trace_post_update(
 308         const char      *fname,         /* function name */
 309         char            *desc,          /* operation description */
 310         xfs_inode_t     *ip,            /* incore inode pointer */
 311         xfs_extnum_t    idx,            /* index of entry updated */
 312         int             whichfork);     /* data or attr fork */
 313
 314 /*
 315  * Add bmap trace entry prior to updating an extent record in place.
 316  */
 317 STATIC void
 318 xfs_bmap_trace_pre_update(
 319         const char      *fname,         /* function name */
 320         char            *desc,          /* operation description */
 321         xfs_inode_t     *ip,            /* incore inode pointer */
 322         xfs_extnum_t    idx,            /* index of entry to be updated */
 323         int             whichfork);     /* data or attr fork */
 324
 325 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)       \
 326         xfs_bmap_trace_delete(__func__,d,ip,i,c,w)
 327 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w) \
 328         xfs_bmap_trace_insert(__func__,d,ip,i,c,r1,r2,w)
 329 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)    \
 330         xfs_bmap_trace_post_update(__func__,d,ip,i,w)
 331 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)     \
 332         xfs_bmap_trace_pre_update(__func__,d,ip,i,w)
 333 #else
 334 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)
 335 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w)
 336 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)
 337 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)
 338 #endif  /* XFS_BMAP_TRACE */
 339
 340 /*
 341  * Compute the worst-case number of indirect blocks that will be used
 342  * for ip's delayed extent of length "len".
 343  */
 344 STATIC xfs_filblks_t
 345 xfs_bmap_worst_indlen(
 346         xfs_inode_t             *ip,    /* incore inode pointer */
 347         xfs_filblks_t           len);   /* delayed extent length */
 348
 349 #ifdef DEBUG
 350 /*
 351  * Perform various validation checks on the values being returned
 352  * from xfs_bmapi().
 353  */
 354 STATIC void
 355 xfs_bmap_validate_ret(
 356         xfs_fileoff_t           bno,
 357         xfs_filblks_t           len,
 358         int                     flags,
 359         xfs_bmbt_irec_t         *mval,
 360         int                     nmap,
 361         int                     ret_nmap);
 362 #else
 363 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 364 #endif /* DEBUG */
 365
 366 #if defined(XFS_RW_TRACE)
 367 STATIC void
 368 xfs_bunmap_trace(
 369         xfs_inode_t             *ip,
 370         xfs_fileoff_t           bno,
 371         xfs_filblks_t           len,
 372         int                     flags,
 373         inst_t                  *ra);
 374 #else
 375 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
 376 #endif  /* XFS_RW_TRACE */
 377
 378 STATIC int
 379 xfs_bmap_count_tree(
 380         xfs_mount_t     *mp,
 381         xfs_trans_t     *tp,
 382         xfs_ifork_t     *ifp,
 383         xfs_fsblock_t   blockno,
 384         int             levelin,
 385         int             *count);
 386
 387 STATIC int
 388 xfs_bmap_count_leaves(
 389         xfs_ifork_t             *ifp,
 390         xfs_extnum_t            idx,
 391         int                     numrecs,
 392         int                     *count);
 393
 394 STATIC int
 395 xfs_bmap_disk_count_leaves(
 396         xfs_extnum_t            idx,
 397         xfs_bmbt_block_t        *block,
 398         int                     numrecs,
 399         int                     *count);
 400
 401 /*
 402  * Bmap internal routines.
 403  */
 404
 405 /*
 406  * Called from xfs_bmap_add_attrfork to handle btree format files.
 407  */
 408 STATIC int                                      /* error */
 409 xfs_bmap_add_attrfork_btree(
 410         xfs_trans_t             *tp,            /* transaction pointer */
 411         xfs_inode_t             *ip,            /* incore inode pointer */
 412         xfs_fsblock_t           *firstblock,    /* first block allocated */
 413         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 414         int                     *flags)         /* inode logging flags */
 415 {
 416         xfs_btree_cur_t         *cur;           /* btree cursor */
 417         int                     error;          /* error return value */
 418         xfs_mount_t             *mp;            /* file system mount struct */
 419         int                     stat;           /* newroot status */
 420
 421         mp = ip->i_mount;
 422         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 423                 *flags |= XFS_ILOG_DBROOT;
 424         else {
 425                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
 426                         XFS_DATA_FORK);
 427                 cur->bc_private.b.flist = flist;
 428                 cur->bc_private.b.firstblock = *firstblock;
 429                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 430                         goto error0;
 431                 ASSERT(stat == 1);      /* must be at least one entry */
 432                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
 433                         goto error0;
 434                 if (stat == 0) {
 435                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 436                         return XFS_ERROR(ENOSPC);
 437                 }
 438                 *firstblock = cur->bc_private.b.firstblock;
 439                 cur->bc_private.b.allocated = 0;
 440                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 441         }
 442         return 0;
 443 error0:
 444         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 445         return error;
 446 }
 447
 448 /*
 449  * Called from xfs_bmap_add_attrfork to handle extents format files.
 450  */
 451 STATIC int                                      /* error */
 452 xfs_bmap_add_attrfork_extents(
 453         xfs_trans_t             *tp,            /* transaction pointer */
 454         xfs_inode_t             *ip,            /* incore inode pointer */
 455         xfs_fsblock_t           *firstblock,    /* first block allocated */
 456         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 457         int                     *flags)         /* inode logging flags */
 458 {
 459         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 460         int                     error;          /* error return value */
 461
 462         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 463                 return 0;
 464         cur = NULL;
 465         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 466                 flags, XFS_DATA_FORK);
 467         if (cur) {
 468                 cur->bc_private.b.allocated = 0;
 469                 xfs_btree_del_cursor(cur,
 470                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 471         }
 472         return error;
 473 }
 474
 475 /*
 476  * Called from xfs_bmap_add_attrfork to handle local format files.
 477  */
 478 STATIC int                                      /* error */
 479 xfs_bmap_add_attrfork_local(
 480         xfs_trans_t             *tp,            /* transaction pointer */
 481         xfs_inode_t             *ip,            /* incore inode pointer */
 482         xfs_fsblock_t           *firstblock,    /* first block allocated */
 483         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 484         int                     *flags)         /* inode logging flags */
 485 {
 486         xfs_da_args_t           dargs;          /* args for dir/attr code */
 487         int                     error;          /* error return value */
 488         xfs_mount_t             *mp;            /* mount structure pointer */
 489
 490         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 491                 return 0;
 492         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 493                 mp = ip->i_mount;
 494                 memset(&dargs, 0, sizeof(dargs));
 495                 dargs.dp = ip;
 496                 dargs.firstblock = firstblock;
 497                 dargs.flist = flist;
 498                 dargs.total = mp->m_dirblkfsbs;
 499                 dargs.whichfork = XFS_DATA_FORK;
 500                 dargs.trans = tp;
 501                 error = xfs_dir2_sf_to_block(&dargs);
 502         } else
 503                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 504                         XFS_DATA_FORK);
 505         return error;
 506 }
 507
 508 /*
 509  * Called by xfs_bmapi to update file extent records and the btree
 510  * after allocating space (or doing a delayed allocation).
 511  */
 512 STATIC int                              /* error */
 513 xfs_bmap_add_extent(
 514         xfs_inode_t             *ip,    /* incore inode pointer */
 515         xfs_extnum_t            idx,    /* extent number to update/insert */
 516         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 517         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 518         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 519         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 520         int                     *logflagsp, /* inode logging flags */
 521         xfs_extdelta_t          *delta, /* Change made to incore extents */
 522         int                     whichfork, /* data or attr fork */
 523         int                     rsvd)   /* OK to use reserved data blocks */
 524 {
 525         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 526         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 527         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 528         int                     error;  /* error return value */
 529         xfs_ifork_t             *ifp;   /* inode fork ptr */
 530         int                     logflags; /* returned value */
 531         xfs_extnum_t            nextents; /* number of extents in file now */
 532
 533         XFS_STATS_INC(xs_add_exlist);
 534         cur = *curp;
 535         ifp = XFS_IFORK_PTR(ip, whichfork);
 536         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 537         ASSERT(idx <= nextents);
 538         da_old = da_new = 0;
 539         error = 0;
 540         /*
 541          * This is the first extent added to a new/empty file.
 542          * Special case this one, so other routines get to assume there are
 543          * already extents in the list.
 544          */
 545         if (nextents == 0) {
 546                 XFS_BMAP_TRACE_INSERT("insert empty", ip, 0, 1, new, NULL,
 547                         whichfork);
 548                 xfs_iext_insert(ifp, 0, 1, new);
 549                 ASSERT(cur == NULL);
 550                 ifp->if_lastex = 0;
 551                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
 552                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 553                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
 554                 } else
 555                         logflags = 0;
 556                 /* DELTA: single new extent */
 557                 if (delta) {
 558                         if (delta->xed_startoff > new->br_startoff)
 559                                 delta->xed_startoff = new->br_startoff;
 560                         if (delta->xed_blockcount <
 561                                         new->br_startoff + new->br_blockcount)
 562                                 delta->xed_blockcount = new->br_startoff +
 563                                                 new->br_blockcount;
 564                 }
 565         }
 566         /*
 567          * Any kind of new delayed allocation goes here.
 568          */
 569         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
 570                 if (cur)
 571                         ASSERT((cur->bc_private.b.flags &
 572                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 573                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 574                                 &logflags, delta, rsvd)))
 575                         goto done;
 576         }
 577         /*
 578          * Real allocation off the end of the file.
 579          */
 580         else if (idx == nextents) {
 581                 if (cur)
 582                         ASSERT((cur->bc_private.b.flags &
 583                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 584                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 585                                 &logflags, delta, whichfork)))
 586                         goto done;
 587         } else {
 588                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 589
 590                 /*
 591                  * Get the record referred to by idx.
 592                  */
 593                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
 594                 /*
 595                  * If it's a real allocation record, and the new allocation ends
 596                  * after the start of the referred to record, then we're filling
 597                  * in a delayed or unwritten allocation with a real one, or
 598                  * converting real back to unwritten.
 599                  */
 600                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
 601                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 602                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 603                             ISNULLSTARTBLOCK(prev.br_startblock)) {
 604                                 da_old = STARTBLOCKVAL(prev.br_startblock);
 605                                 if (cur)
 606                                         ASSERT(cur->bc_private.b.flags &
 607                                                 XFS_BTCUR_BPRV_WASDEL);
 608                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 609                                         idx, &cur, new, &da_new, first, flist,
 610                                         &logflags, delta, rsvd)))
 611                                         goto done;
 612                         } else if (new->br_state == XFS_EXT_NORM) {
 613                                 ASSERT(new->br_state == XFS_EXT_NORM);
 614                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 615                                         ip, idx, &cur, new, &logflags, delta)))
 616                                         goto done;
 617                         } else {
 618                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 619                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 620                                         ip, idx, &cur, new, &logflags, delta)))
 621                                         goto done;
 622                         }
 623                         ASSERT(*curp == cur || *curp == NULL);
 624                 }
 625                 /*
 626                  * Otherwise we're filling in a hole with an allocation.
 627                  */
 628                 else {
 629                         if (cur)
 630                                 ASSERT((cur->bc_private.b.flags &
 631                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 632                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 633                                         new, &logflags, delta, whichfork)))
 634                                 goto done;
 635                 }
 636         }
 637
 638         ASSERT(*curp == cur || *curp == NULL);
 639         /*
 640          * Convert to a btree if necessary.
 641          */
 642         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 643             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 644                 int     tmp_logflags;   /* partial log flag return val */
 645
 646                 ASSERT(cur == NULL);
 647                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 648                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 649                 logflags |= tmp_logflags;
 650                 if (error)
 651                         goto done;
 652         }
 653         /*
 654          * Adjust for changes in reserved delayed indirect blocks.
 655          * Nothing to do for disk quotas here.
 656          */
 657         if (da_old || da_new) {
 658                 xfs_filblks_t   nblks;
 659
 660                 nblks = da_new;
 661                 if (cur)
 662                         nblks += cur->bc_private.b.allocated;
 663                 ASSERT(nblks <= da_old);
 664                 if (nblks < da_old)
 665                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 666                                 (int64_t)(da_old - nblks), rsvd);
 667         }
 668         /*
 669          * Clear out the allocated field, done with it now in any case.
 670          */
 671         if (cur) {
 672                 cur->bc_private.b.allocated = 0;
 673                 *curp = cur;
 674         }
 675 done:
 676 #ifdef DEBUG
 677         if (!error)
 678                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 679 #endif
 680         *logflagsp = logflags;
 681         return error;
 682 }
 683
 684 /*
 685  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 686  * allocation to a real allocation.
 687  */
 688 STATIC int                              /* error */
 689 xfs_bmap_add_extent_delay_real(
 690         xfs_inode_t             *ip,    /* incore inode pointer */
 691         xfs_extnum_t            idx,    /* extent number to update/insert */
 692         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 693         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 694         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 695         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 696         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 697         int                     *logflagsp, /* inode logging flags */
 698         xfs_extdelta_t          *delta, /* Change made to incore extents */
 699         int                     rsvd)   /* OK to use reserved data block allocation */
 700 {
 701         xfs_btree_cur_t         *cur;   /* btree cursor */
 702         int                     diff;   /* temp value */
 703         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 704         int                     error;  /* error return value */
 705         int                     i;      /* temp state */
 706         xfs_ifork_t             *ifp;   /* inode fork pointer */
 707         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 708         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 709                                         /* left is 0, right is 1, prev is 2 */
 710         int                     rval=0; /* return value (logging flags) */
 711         int                     state = 0;/* state bits, accessed thru macros */
 712         xfs_filblks_t           temp=0; /* value for dnew calculations */
 713         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 714         int                     tmp_rval;       /* partial logging flags */
 715         enum {                          /* bit number definitions for state */
 716                 LEFT_CONTIG,    RIGHT_CONTIG,
 717                 LEFT_FILLING,   RIGHT_FILLING,
 718                 LEFT_DELAY,     RIGHT_DELAY,
 719                 LEFT_VALID,     RIGHT_VALID
 720         };
 721
 722 #define LEFT            r[0]
 723 #define RIGHT           r[1]
 724 #define PREV            r[2]
 725 #define MASK(b)         (1 << (b))
 726 #define MASK2(a,b)      (MASK(a) | MASK(b))
 727 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
 728 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
 729 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
 730 #define STATE_TEST(b)   (state & MASK(b))
 731 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
 732                                        ((state &= ~MASK(b)), 0))
 733 #define SWITCH_STATE            \
 734         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
 735
 736         /*
 737          * Set up a bunch of variables to make the tests simpler.
 738          */
 739         cur = *curp;
 740         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 741         ep = xfs_iext_get_ext(ifp, idx);
 742         xfs_bmbt_get_all(ep, &PREV);
 743         new_endoff = new->br_startoff + new->br_blockcount;
 744         ASSERT(PREV.br_startoff <= new->br_startoff);
 745         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 746         /*
 747          * Set flags determining what part of the previous delayed allocation
 748          * extent is being replaced by a real allocation.
 749          */
 750         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
 751         STATE_SET(RIGHT_FILLING,
 752                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
 753         /*
 754          * Check and set flags if this segment has a left neighbor.
 755          * Don't set contiguous if the combined extent would be too large.
 756          */
 757         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
 758                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
 759                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
 760         }
 761         STATE_SET(LEFT_CONTIG,
 762                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
 763                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 764                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 765                 LEFT.br_state == new->br_state &&
 766                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
 767         /*
 768          * Check and set flags if this segment has a right neighbor.
 769          * Don't set contiguous if the combined extent would be too large.
 770          * Also check for all-three-contiguous being too large.
 771          */
 772         if (STATE_SET_TEST(RIGHT_VALID,
 773                         idx <
 774                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
 775                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
 776                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
 777         }
 778         STATE_SET(RIGHT_CONTIG,
 779                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
 780                 new_endoff == RIGHT.br_startoff &&
 781                 new->br_startblock + new->br_blockcount ==
 782                     RIGHT.br_startblock &&
 783                 new->br_state == RIGHT.br_state &&
 784                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 785                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
 786                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
 787                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 788                      <= MAXEXTLEN));
 789         error = 0;
 790         /*
 791          * Switch out based on the FILLING and CONTIG state bits.
 792          */
 793         switch (SWITCH_STATE) {
 794
 795         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
 796                 /*
 797                  * Filling in all of a previously delayed allocation extent.
 798                  * The left and right neighbors are both contiguous with new.
 799                  */
 800                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 801                         XFS_DATA_FORK);
 802                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 803                         LEFT.br_blockcount + PREV.br_blockcount +
 804                         RIGHT.br_blockcount);
 805                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 806                         XFS_DATA_FORK);
 807                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
 808                 xfs_iext_remove(ifp, idx, 2);
 809                 ip->i_df.if_lastex = idx - 1;
 810                 ip->i_d.di_nextents--;
 811                 if (cur == NULL)
 812                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 813                 else {
 814                         rval = XFS_ILOG_CORE;
 815                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 816                                         RIGHT.br_startblock,
 817                                         RIGHT.br_blockcount, &i)))
 818                                 goto done;
 819                         ASSERT(i == 1);
 820                         if ((error = xfs_bmbt_delete(cur, &i)))
 821                                 goto done;
 822                         ASSERT(i == 1);
 823                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
 824                                 goto done;
 825                         ASSERT(i == 1);
 826                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 827                                         LEFT.br_startblock,
 828                                         LEFT.br_blockcount +
 829                                         PREV.br_blockcount +
 830                                         RIGHT.br_blockcount, LEFT.br_state)))
 831                                 goto done;
 832                 }
 833                 *dnew = 0;
 834                 /* DELTA: Three in-core extents are replaced by one. */
 835                 temp = LEFT.br_startoff;
 836                 temp2 = LEFT.br_blockcount +
 837                         PREV.br_blockcount +
 838                         RIGHT.br_blockcount;
 839                 break;
 840
 841         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
 842                 /*
 843                  * Filling in all of a previously delayed allocation extent.
 844                  * The left neighbor is contiguous, the right is not.
 845                  */
 846                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
 847                         XFS_DATA_FORK);
 848                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 849                         LEFT.br_blockcount + PREV.br_blockcount);
 850                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
 851                         XFS_DATA_FORK);
 852                 ip->i_df.if_lastex = idx - 1;
 853                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
 854                 xfs_iext_remove(ifp, idx, 1);
 855                 if (cur == NULL)
 856                         rval = XFS_ILOG_DEXT;
 857                 else {
 858                         rval = 0;
 859                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 860                                         LEFT.br_startblock, LEFT.br_blockcount,
 861                                         &i)))
 862                                 goto done;
 863                         ASSERT(i == 1);
 864                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 865                                         LEFT.br_startblock,
 866                                         LEFT.br_blockcount +
 867                                         PREV.br_blockcount, LEFT.br_state)))
 868                                 goto done;
 869                 }
 870                 *dnew = 0;
 871                 /* DELTA: Two in-core extents are replaced by one. */
 872                 temp = LEFT.br_startoff;
 873                 temp2 = LEFT.br_blockcount +
 874                         PREV.br_blockcount;
 875                 break;
 876
 877         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
 878                 /*
 879                  * Filling in all of a previously delayed allocation extent.
 880                  * The right neighbor is contiguous, the left is not.
 881                  */
 882                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 883                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 884                 xfs_bmbt_set_blockcount(ep,
 885                         PREV.br_blockcount + RIGHT.br_blockcount);
 886                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 887                 ip->i_df.if_lastex = idx;
 888                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
 889                 xfs_iext_remove(ifp, idx + 1, 1);
 890                 if (cur == NULL)
 891                         rval = XFS_ILOG_DEXT;
 892                 else {
 893                         rval = 0;
 894                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 895                                         RIGHT.br_startblock,
 896                                         RIGHT.br_blockcount, &i)))
 897                                 goto done;
 898                         ASSERT(i == 1);
 899                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 900                                         new->br_startblock,
 901                                         PREV.br_blockcount +
 902                                         RIGHT.br_blockcount, PREV.br_state)))
 903                                 goto done;
 904                 }
 905                 *dnew = 0;
 906                 /* DELTA: Two in-core extents are replaced by one. */
 907                 temp = PREV.br_startoff;
 908                 temp2 = PREV.br_blockcount +
 909                         RIGHT.br_blockcount;
 910                 break;
 911
 912         case MASK2(LEFT_FILLING, RIGHT_FILLING):
 913                 /*
 914                  * Filling in all of a previously delayed allocation extent.
 915                  * Neither the left nor right neighbors are contiguous with
 916                  * the new one.
 917                  */
 918                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 919                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 920                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 921                 ip->i_df.if_lastex = idx;
 922                 ip->i_d.di_nextents++;
 923                 if (cur == NULL)
 924                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 925                 else {
 926                         rval = XFS_ILOG_CORE;
 927                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 928                                         new->br_startblock, new->br_blockcount,
 929                                         &i)))
 930                                 goto done;
 931                         ASSERT(i == 0);
 932                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 933                         if ((error = xfs_bmbt_insert(cur, &i)))
 934                                 goto done;
 935                         ASSERT(i == 1);
 936                 }
 937                 *dnew = 0;
 938                 /* DELTA: The in-core extent described by new changed type. */
 939                 temp = new->br_startoff;
 940                 temp2 = new->br_blockcount;
 941                 break;
 942
 943         case MASK2(LEFT_FILLING, LEFT_CONTIG):
 944                 /*
 945                  * Filling in the first part of a previous delayed allocation.
 946                  * The left neighbor is contiguous.
 947                  */
 948                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 949                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 950                         LEFT.br_blockcount + new->br_blockcount);
 951                 xfs_bmbt_set_startoff(ep,
 952                         PREV.br_startoff + new->br_blockcount);
 953                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 954                 temp = PREV.br_blockcount - new->br_blockcount;
 955                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 956                 xfs_bmbt_set_blockcount(ep, temp);
 957                 ip->i_df.if_lastex = idx - 1;
 958                 if (cur == NULL)
 959                         rval = XFS_ILOG_DEXT;
 960                 else {
 961                         rval = 0;
 962                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 963                                         LEFT.br_startblock, LEFT.br_blockcount,
 964                                         &i)))
 965                                 goto done;
 966                         ASSERT(i == 1);
 967                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 968                                         LEFT.br_startblock,
 969                                         LEFT.br_blockcount +
 970                                         new->br_blockcount,
 971                                         LEFT.br_state)))
 972                                 goto done;
 973                 }
 974                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 975                         STARTBLOCKVAL(PREV.br_startblock));
 976                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
 977                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 978                 *dnew = temp;
 979                 /* DELTA: The boundary between two in-core extents moved. */
 980                 temp = LEFT.br_startoff;
 981                 temp2 = LEFT.br_blockcount +
 982                         PREV.br_blockcount;
 983                 break;
 984
 985         case MASK(LEFT_FILLING):
 986                 /*
 987                  * Filling in the first part of a previous delayed allocation.
 988                  * The left neighbor is not contiguous.
 989                  */
 990                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
 991                 xfs_bmbt_set_startoff(ep, new_endoff);
 992                 temp = PREV.br_blockcount - new->br_blockcount;
 993                 xfs_bmbt_set_blockcount(ep, temp);
 994                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
 995                         XFS_DATA_FORK);
 996                 xfs_iext_insert(ifp, idx, 1, new);
 997                 ip->i_df.if_lastex = idx;
 998                 ip->i_d.di_nextents++;
 999                 if (cur == NULL)
1000                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1001                 else {
1002                         rval = XFS_ILOG_CORE;
1003                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1004                                         new->br_startblock, new->br_blockcount,
1005                                         &i)))
1006                                 goto done;
1007                         ASSERT(i == 0);
1008                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1009                         if ((error = xfs_bmbt_insert(cur, &i)))
1010                                 goto done;
1011                         ASSERT(i == 1);
1012                 }
1013                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1014                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1015                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1016                                         first, flist, &cur, 1, &tmp_rval,
1017                                         XFS_DATA_FORK);
1018                         rval |= tmp_rval;
1019                         if (error)
1020                                 goto done;
1021                 }
1022                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1023                         STARTBLOCKVAL(PREV.br_startblock) -
1024                         (cur ? cur->bc_private.b.allocated : 0));
1025                 ep = xfs_iext_get_ext(ifp, idx + 1);
1026                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1027                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx + 1, XFS_DATA_FORK);
1028                 *dnew = temp;
1029                 /* DELTA: One in-core extent is split in two. */
1030                 temp = PREV.br_startoff;
1031                 temp2 = PREV.br_blockcount;
1032                 break;
1033
1034         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1035                 /*
1036                  * Filling in the last part of a previous delayed allocation.
1037                  * The right neighbor is contiguous with the new allocation.
1038                  */
1039                 temp = PREV.br_blockcount - new->br_blockcount;
1040                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1041                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1042                 xfs_bmbt_set_blockcount(ep, temp);
1043                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1044                         new->br_startoff, new->br_startblock,
1045                         new->br_blockcount + RIGHT.br_blockcount,
1046                         RIGHT.br_state);
1047                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1048                 ip->i_df.if_lastex = idx + 1;
1049                 if (cur == NULL)
1050                         rval = XFS_ILOG_DEXT;
1051                 else {
1052                         rval = 0;
1053                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1054                                         RIGHT.br_startblock,
1055                                         RIGHT.br_blockcount, &i)))
1056                                 goto done;
1057                         ASSERT(i == 1);
1058                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1059                                         new->br_startblock,
1060                                         new->br_blockcount +
1061                                         RIGHT.br_blockcount,
1062                                         RIGHT.br_state)))
1063                                 goto done;
1064                 }
1065                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1066                         STARTBLOCKVAL(PREV.br_startblock));
1067                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1068                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1069                 *dnew = temp;
1070                 /* DELTA: The boundary between two in-core extents moved. */
1071                 temp = PREV.br_startoff;
1072                 temp2 = PREV.br_blockcount +
1073                         RIGHT.br_blockcount;
1074                 break;
1075
1076         case MASK(RIGHT_FILLING):
1077                 /*
1078                  * Filling in the last part of a previous delayed allocation.
1079                  * The right neighbor is not contiguous.
1080                  */
1081                 temp = PREV.br_blockcount - new->br_blockcount;
1082                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1083                 xfs_bmbt_set_blockcount(ep, temp);
1084                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1085                         XFS_DATA_FORK);
1086                 xfs_iext_insert(ifp, idx + 1, 1, new);
1087                 ip->i_df.if_lastex = idx + 1;
1088                 ip->i_d.di_nextents++;
1089                 if (cur == NULL)
1090                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1091                 else {
1092                         rval = XFS_ILOG_CORE;
1093                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1094                                         new->br_startblock, new->br_blockcount,
1095                                         &i)))
1096                                 goto done;
1097                         ASSERT(i == 0);
1098                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1099                         if ((error = xfs_bmbt_insert(cur, &i)))
1100                                 goto done;
1101                         ASSERT(i == 1);
1102                 }
1103                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1104                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1105                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1106                                 first, flist, &cur, 1, &tmp_rval,
1107                                 XFS_DATA_FORK);
1108                         rval |= tmp_rval;
1109                         if (error)
1110                                 goto done;
1111                 }
1112                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1113                         STARTBLOCKVAL(PREV.br_startblock) -
1114                         (cur ? cur->bc_private.b.allocated : 0));
1115                 ep = xfs_iext_get_ext(ifp, idx);
1116                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1117                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1118                 *dnew = temp;
1119                 /* DELTA: One in-core extent is split in two. */
1120                 temp = PREV.br_startoff;
1121                 temp2 = PREV.br_blockcount;
1122                 break;
1123
1124         case 0:
1125                 /*
1126                  * Filling in the middle part of a previous delayed allocation.
1127                  * Contiguity is impossible here.
1128                  * This case is avoided almost all the time.
1129                  */
1130                 temp = new->br_startoff - PREV.br_startoff;
1131                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1132                 xfs_bmbt_set_blockcount(ep, temp);
1133                 r[0] = *new;
1134                 r[1].br_state = PREV.br_state;
1135                 r[1].br_startblock = 0;
1136                 r[1].br_startoff = new_endoff;
1137                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1138                 r[1].br_blockcount = temp2;
1139                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1140                         XFS_DATA_FORK);
1141                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1142                 ip->i_df.if_lastex = idx + 1;
1143                 ip->i_d.di_nextents++;
1144                 if (cur == NULL)
1145                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1146                 else {
1147                         rval = XFS_ILOG_CORE;
1148                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1149                                         new->br_startblock, new->br_blockcount,
1150                                         &i)))
1151                                 goto done;
1152                         ASSERT(i == 0);
1153                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1154                         if ((error = xfs_bmbt_insert(cur, &i)))
1155                                 goto done;
1156                         ASSERT(i == 1);
1157                 }
1158                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1159                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1160                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1161                                         first, flist, &cur, 1, &tmp_rval,
1162                                         XFS_DATA_FORK);
1163                         rval |= tmp_rval;
1164                         if (error)
1165                                 goto done;
1166                 }
1167                 temp = xfs_bmap_worst_indlen(ip, temp);
1168                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1169                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1170                         (cur ? cur->bc_private.b.allocated : 0));
1171                 if (diff > 0 &&
1172                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1173                         /*
1174                          * Ick gross gag me with a spoon.
1175                          */
1176                         ASSERT(0);      /* want to see if this ever happens! */
1177                         while (diff > 0) {
1178                                 if (temp) {
1179                                         temp--;
1180                                         diff--;
1181                                         if (!diff ||
1182                                             !xfs_mod_incore_sb(ip->i_mount,
1183                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1184                                                 break;
1185                                 }
1186                                 if (temp2) {
1187                                         temp2--;
1188                                         diff--;
1189                                         if (!diff ||
1190                                             !xfs_mod_incore_sb(ip->i_mount,
1191                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1192                                                 break;
1193                                 }
1194                         }
1195                 }
1196                 ep = xfs_iext_get_ext(ifp, idx);
1197                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1198                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1199                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1200                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1201                         NULLSTARTBLOCK((int)temp2));
1202                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1203                 *dnew = temp + temp2;
1204                 /* DELTA: One in-core extent is split in three. */
1205                 temp = PREV.br_startoff;
1206                 temp2 = PREV.br_blockcount;
1207                 break;
1208
1209         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1210         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1211         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1212         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1213         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1214         case MASK(LEFT_CONTIG):
1215         case MASK(RIGHT_CONTIG):
1216                 /*
1217                  * These cases are all impossible.
1218                  */
1219                 ASSERT(0);
1220         }
1221         *curp = cur;
1222         if (delta) {
1223                 temp2 += temp;
1224                 if (delta->xed_startoff > temp)
1225                         delta->xed_startoff = temp;
1226                 if (delta->xed_blockcount < temp2)
1227                         delta->xed_blockcount = temp2;
1228         }
1229 done:
1230         *logflagsp = rval;
1231         return error;
1232 #undef  LEFT
1233 #undef  RIGHT
1234 #undef  PREV
1235 #undef  MASK
1236 #undef  MASK2
1237 #undef  MASK3
1238 #undef  MASK4
1239 #undef  STATE_SET
1240 #undef  STATE_TEST
1241 #undef  STATE_SET_TEST
1242 #undef  SWITCH_STATE
1243 }
1244
1245 /*
1246  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1247  * allocation to a real allocation or vice versa.
1248  */
1249 STATIC int                              /* error */
1250 xfs_bmap_add_extent_unwritten_real(
1251         xfs_inode_t             *ip,    /* incore inode pointer */
1252         xfs_extnum_t            idx,    /* extent number to update/insert */
1253         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1254         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1255         int                     *logflagsp, /* inode logging flags */
1256         xfs_extdelta_t          *delta) /* Change made to incore extents */
1257 {
1258         xfs_btree_cur_t         *cur;   /* btree cursor */
1259         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1260         int                     error;  /* error return value */
1261         int                     i;      /* temp state */
1262         xfs_ifork_t             *ifp;   /* inode fork pointer */
1263         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1264         xfs_exntst_t            newext; /* new extent state */
1265         xfs_exntst_t            oldext; /* old extent state */
1266         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1267                                         /* left is 0, right is 1, prev is 2 */
1268         int                     rval=0; /* return value (logging flags) */
1269         int                     state = 0;/* state bits, accessed thru macros */
1270         xfs_filblks_t           temp=0;
1271         xfs_filblks_t           temp2=0;
1272         enum {                          /* bit number definitions for state */
1273                 LEFT_CONTIG,    RIGHT_CONTIG,
1274                 LEFT_FILLING,   RIGHT_FILLING,
1275                 LEFT_DELAY,     RIGHT_DELAY,
1276                 LEFT_VALID,     RIGHT_VALID
1277         };
1278
1279 #define LEFT            r[0]
1280 #define RIGHT           r[1]
1281 #define PREV            r[2]
1282 #define MASK(b)         (1 << (b))
1283 #define MASK2(a,b)      (MASK(a) | MASK(b))
1284 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1285 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1286 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1287 #define STATE_TEST(b)   (state & MASK(b))
1288 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1289                                        ((state &= ~MASK(b)), 0))
1290 #define SWITCH_STATE            \
1291         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1292
1293         /*
1294          * Set up a bunch of variables to make the tests simpler.
1295          */
1296         error = 0;
1297         cur = *curp;
1298         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1299         ep = xfs_iext_get_ext(ifp, idx);
1300         xfs_bmbt_get_all(ep, &PREV);
1301         newext = new->br_state;
1302         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1303                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1304         ASSERT(PREV.br_state == oldext);
1305         new_endoff = new->br_startoff + new->br_blockcount;
1306         ASSERT(PREV.br_startoff <= new->br_startoff);
1307         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1308         /*
1309          * Set flags determining what part of the previous oldext allocation
1310          * extent is being replaced by a newext allocation.
1311          */
1312         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1313         STATE_SET(RIGHT_FILLING,
1314                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1315         /*
1316          * Check and set flags if this segment has a left neighbor.
1317          * Don't set contiguous if the combined extent would be too large.
1318          */
1319         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1320                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1321                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1322         }
1323         STATE_SET(LEFT_CONTIG,
1324                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1325                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1326                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1327                 LEFT.br_state == newext &&
1328                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1329         /*
1330          * Check and set flags if this segment has a right neighbor.
1331          * Don't set contiguous if the combined extent would be too large.
1332          * Also check for all-three-contiguous being too large.
1333          */
1334         if (STATE_SET_TEST(RIGHT_VALID,
1335                         idx <
1336                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1337                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1338                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1339         }
1340         STATE_SET(RIGHT_CONTIG,
1341                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1342                 new_endoff == RIGHT.br_startoff &&
1343                 new->br_startblock + new->br_blockcount ==
1344                     RIGHT.br_startblock &&
1345                 newext == RIGHT.br_state &&
1346                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1347                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1348                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1349                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1350                      <= MAXEXTLEN));
1351         /*
1352          * Switch out based on the FILLING and CONTIG state bits.
1353          */
1354         switch (SWITCH_STATE) {
1355
1356         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1357                 /*
1358                  * Setting all of a previous oldext extent to newext.
1359                  * The left and right neighbors are both contiguous with new.
1360                  */
1361                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1362                         XFS_DATA_FORK);
1363                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1364                         LEFT.br_blockcount + PREV.br_blockcount +
1365                         RIGHT.br_blockcount);
1366                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1367                         XFS_DATA_FORK);
1368                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
1369                 xfs_iext_remove(ifp, idx, 2);
1370                 ip->i_df.if_lastex = idx - 1;
1371                 ip->i_d.di_nextents -= 2;
1372                 if (cur == NULL)
1373                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1374                 else {
1375                         rval = XFS_ILOG_CORE;
1376                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1377                                         RIGHT.br_startblock,
1378                                         RIGHT.br_blockcount, &i)))
1379                                 goto done;
1380                         ASSERT(i == 1);
1381                         if ((error = xfs_bmbt_delete(cur, &i)))
1382                                 goto done;
1383                         ASSERT(i == 1);
1384                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1385                                 goto done;
1386                         ASSERT(i == 1);
1387                         if ((error = xfs_bmbt_delete(cur, &i)))
1388                                 goto done;
1389                         ASSERT(i == 1);
1390                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1391                                 goto done;
1392                         ASSERT(i == 1);
1393                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1394                                 LEFT.br_startblock,
1395                                 LEFT.br_blockcount + PREV.br_blockcount +
1396                                 RIGHT.br_blockcount, LEFT.br_state)))
1397                                 goto done;
1398                 }
1399                 /* DELTA: Three in-core extents are replaced by one. */
1400                 temp = LEFT.br_startoff;
1401                 temp2 = LEFT.br_blockcount +
1402                         PREV.br_blockcount +
1403                         RIGHT.br_blockcount;
1404                 break;
1405
1406         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1407                 /*
1408                  * Setting all of a previous oldext extent to newext.
1409                  * The left neighbor is contiguous, the right is not.
1410                  */
1411                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
1412                         XFS_DATA_FORK);
1413                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1414                         LEFT.br_blockcount + PREV.br_blockcount);
1415                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
1416                         XFS_DATA_FORK);
1417                 ip->i_df.if_lastex = idx - 1;
1418                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
1419                 xfs_iext_remove(ifp, idx, 1);
1420                 ip->i_d.di_nextents--;
1421                 if (cur == NULL)
1422                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1423                 else {
1424                         rval = XFS_ILOG_CORE;
1425                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1426                                         PREV.br_startblock, PREV.br_blockcount,
1427                                         &i)))
1428                                 goto done;
1429                         ASSERT(i == 1);
1430                         if ((error = xfs_bmbt_delete(cur, &i)))
1431                                 goto done;
1432                         ASSERT(i == 1);
1433                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1434                                 goto done;
1435                         ASSERT(i == 1);
1436                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1437                                 LEFT.br_startblock,
1438                                 LEFT.br_blockcount + PREV.br_blockcount,
1439                                 LEFT.br_state)))
1440                                 goto done;
1441                 }
1442                 /* DELTA: Two in-core extents are replaced by one. */
1443                 temp = LEFT.br_startoff;
1444                 temp2 = LEFT.br_blockcount +
1445                         PREV.br_blockcount;
1446                 break;
1447
1448         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1449                 /*
1450                  * Setting all of a previous oldext extent to newext.
1451                  * The right neighbor is contiguous, the left is not.
1452                  */
1453                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx,
1454                         XFS_DATA_FORK);
1455                 xfs_bmbt_set_blockcount(ep,
1456                         PREV.br_blockcount + RIGHT.br_blockcount);
1457                 xfs_bmbt_set_state(ep, newext);
1458                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx,
1459                         XFS_DATA_FORK);
1460                 ip->i_df.if_lastex = idx;
1461                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
1462                 xfs_iext_remove(ifp, idx + 1, 1);
1463                 ip->i_d.di_nextents--;
1464                 if (cur == NULL)
1465                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1466                 else {
1467                         rval = XFS_ILOG_CORE;
1468                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1469                                         RIGHT.br_startblock,
1470                                         RIGHT.br_blockcount, &i)))
1471                                 goto done;
1472                         ASSERT(i == 1);
1473                         if ((error = xfs_bmbt_delete(cur, &i)))
1474                                 goto done;
1475                         ASSERT(i == 1);
1476                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1477                                 goto done;
1478                         ASSERT(i == 1);
1479                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1480                                 new->br_startblock,
1481                                 new->br_blockcount + RIGHT.br_blockcount,
1482                                 newext)))
1483                                 goto done;
1484                 }
1485                 /* DELTA: Two in-core extents are replaced by one. */
1486                 temp = PREV.br_startoff;
1487                 temp2 = PREV.br_blockcount +
1488                         RIGHT.br_blockcount;
1489                 break;
1490
1491         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1492                 /*
1493                  * Setting all of a previous oldext extent to newext.
1494                  * Neither the left nor right neighbors are contiguous with
1495                  * the new one.
1496                  */
1497                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx,
1498                         XFS_DATA_FORK);
1499                 xfs_bmbt_set_state(ep, newext);
1500                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx,
1501                         XFS_DATA_FORK);
1502                 ip->i_df.if_lastex = idx;
1503                 if (cur == NULL)
1504                         rval = XFS_ILOG_DEXT;
1505                 else {
1506                         rval = 0;
1507                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1508                                         new->br_startblock, new->br_blockcount,
1509                                         &i)))
1510                                 goto done;
1511                         ASSERT(i == 1);
1512                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1513                                 new->br_startblock, new->br_blockcount,
1514                                 newext)))
1515                                 goto done;
1516                 }
1517                 /* DELTA: The in-core extent described by new changed type. */
1518                 temp = new->br_startoff;
1519                 temp2 = new->br_blockcount;
1520                 break;
1521
1522         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1523                 /*
1524                  * Setting the first part of a previous oldext extent to newext.
1525                  * The left neighbor is contiguous.
1526                  */
1527                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1,
1528                         XFS_DATA_FORK);
1529                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1530                         LEFT.br_blockcount + new->br_blockcount);
1531                 xfs_bmbt_set_startoff(ep,
1532                         PREV.br_startoff + new->br_blockcount);
1533                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1,
1534                         XFS_DATA_FORK);
1535                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx,
1536                         XFS_DATA_FORK);
1537                 xfs_bmbt_set_startblock(ep,
1538                         new->br_startblock + new->br_blockcount);
1539                 xfs_bmbt_set_blockcount(ep,
1540                         PREV.br_blockcount - new->br_blockcount);
1541                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx,
1542                         XFS_DATA_FORK);
1543                 ip->i_df.if_lastex = idx - 1;
1544                 if (cur == NULL)
1545                         rval = XFS_ILOG_DEXT;
1546                 else {
1547                         rval = 0;
1548                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1549                                         PREV.br_startblock, PREV.br_blockcount,
1550                                         &i)))
1551                                 goto done;
1552                         ASSERT(i == 1);
1553                         if ((error = xfs_bmbt_update(cur,
1554                                 PREV.br_startoff + new->br_blockcount,
1555                                 PREV.br_startblock + new->br_blockcount,
1556                                 PREV.br_blockcount - new->br_blockcount,
1557                                 oldext)))
1558                                 goto done;
1559                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1560                                 goto done;
1561                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1562                                 LEFT.br_startblock,
1563                                 LEFT.br_blockcount + new->br_blockcount,
1564                                 LEFT.br_state))
1565                                 goto done;
1566                 }
1567                 /* DELTA: The boundary between two in-core extents moved. */
1568                 temp = LEFT.br_startoff;
1569                 temp2 = LEFT.br_blockcount +
1570                         PREV.br_blockcount;
1571                 break;
1572
1573         case MASK(LEFT_FILLING):
1574                 /*
1575                  * Setting the first part of a previous oldext extent to newext.
1576                  * The left neighbor is not contiguous.
1577                  */
1578                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1579                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1580                 xfs_bmbt_set_startoff(ep, new_endoff);
1581                 xfs_bmbt_set_blockcount(ep,
1582                         PREV.br_blockcount - new->br_blockcount);
1583                 xfs_bmbt_set_startblock(ep,
1584                         new->br_startblock + new->br_blockcount);
1585                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1586                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1587                         XFS_DATA_FORK);
1588                 xfs_iext_insert(ifp, idx, 1, new);
1589                 ip->i_df.if_lastex = idx;
1590                 ip->i_d.di_nextents++;
1591                 if (cur == NULL)
1592                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1593                 else {
1594                         rval = XFS_ILOG_CORE;
1595                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1596                                         PREV.br_startblock, PREV.br_blockcount,
1597                                         &i)))
1598                                 goto done;
1599                         ASSERT(i == 1);
1600                         if ((error = xfs_bmbt_update(cur,
1601                                 PREV.br_startoff + new->br_blockcount,
1602                                 PREV.br_startblock + new->br_blockcount,
1603                                 PREV.br_blockcount - new->br_blockcount,
1604                                 oldext)))
1605                                 goto done;
1606                         cur->bc_rec.b = *new;
1607                         if ((error = xfs_bmbt_insert(cur, &i)))
1608                                 goto done;
1609                         ASSERT(i == 1);
1610                 }
1611                 /* DELTA: One in-core extent is split in two. */
1612                 temp = PREV.br_startoff;
1613                 temp2 = PREV.br_blockcount;
1614                 break;
1615
1616         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1617                 /*
1618                  * Setting the last part of a previous oldext extent to newext.
1619                  * The right neighbor is contiguous with the new allocation.
1620                  */
1621                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx,
1622                         XFS_DATA_FORK);
1623                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1,
1624                         XFS_DATA_FORK);
1625                 xfs_bmbt_set_blockcount(ep,
1626                         PREV.br_blockcount - new->br_blockcount);
1627                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx,
1628                         XFS_DATA_FORK);
1629                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1630                         new->br_startoff, new->br_startblock,
1631                         new->br_blockcount + RIGHT.br_blockcount, newext);
1632                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1,
1633                         XFS_DATA_FORK);
1634                 ip->i_df.if_lastex = idx + 1;
1635                 if (cur == NULL)
1636                         rval = XFS_ILOG_DEXT;
1637                 else {
1638                         rval = 0;
1639                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1640                                         PREV.br_startblock,
1641                                         PREV.br_blockcount, &i)))
1642                                 goto done;
1643                         ASSERT(i == 1);
1644                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1645                                 PREV.br_startblock,
1646                                 PREV.br_blockcount - new->br_blockcount,
1647                                 oldext)))
1648                                 goto done;
1649                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1650                                 goto done;
1651                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1652                                 new->br_startblock,
1653                                 new->br_blockcount + RIGHT.br_blockcount,
1654                                 newext)))
1655                                 goto done;
1656                 }
1657                 /* DELTA: The boundary between two in-core extents moved. */
1658                 temp = PREV.br_startoff;
1659                 temp2 = PREV.br_blockcount +
1660                         RIGHT.br_blockcount;
1661                 break;
1662
1663         case MASK(RIGHT_FILLING):
1664                 /*
1665                  * Setting the last part of a previous oldext extent to newext.
1666                  * The right neighbor is not contiguous.
1667                  */
1668                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1669                 xfs_bmbt_set_blockcount(ep,
1670                         PREV.br_blockcount - new->br_blockcount);
1671                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1672                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1673                         XFS_DATA_FORK);
1674                 xfs_iext_insert(ifp, idx + 1, 1, new);
1675                 ip->i_df.if_lastex = idx + 1;
1676                 ip->i_d.di_nextents++;
1677                 if (cur == NULL)
1678                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1679                 else {
1680                         rval = XFS_ILOG_CORE;
1681                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1682                                         PREV.br_startblock, PREV.br_blockcount,
1683                                         &i)))
1684                                 goto done;
1685                         ASSERT(i == 1);
1686                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1687                                 PREV.br_startblock,
1688                                 PREV.br_blockcount - new->br_blockcount,
1689                                 oldext)))
1690                                 goto done;
1691                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1692                                         new->br_startblock, new->br_blockcount,
1693                                         &i)))
1694                                 goto done;
1695                         ASSERT(i == 0);
1696                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1697                         if ((error = xfs_bmbt_insert(cur, &i)))
1698                                 goto done;
1699                         ASSERT(i == 1);
1700                 }
1701                 /* DELTA: One in-core extent is split in two. */
1702                 temp = PREV.br_startoff;
1703                 temp2 = PREV.br_blockcount;
1704                 break;
1705
1706         case 0:
1707                 /*
1708                  * Setting the middle part of a previous oldext extent to
1709                  * newext.  Contiguity is impossible here.
1710                  * One extent becomes three extents.
1711                  */
1712                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1713                 xfs_bmbt_set_blockcount(ep,
1714                         new->br_startoff - PREV.br_startoff);
1715                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1716                 r[0] = *new;
1717                 r[1].br_startoff = new_endoff;
1718                 r[1].br_blockcount =
1719                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1720                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1721                 r[1].br_state = oldext;
1722                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1723                         XFS_DATA_FORK);
1724                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1725                 ip->i_df.if_lastex = idx + 1;
1726                 ip->i_d.di_nextents += 2;
1727                 if (cur == NULL)
1728                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1729                 else {
1730                         rval = XFS_ILOG_CORE;
1731                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1732                                         PREV.br_startblock, PREV.br_blockcount,
1733                                         &i)))
1734                                 goto done;
1735                         ASSERT(i == 1);
1736                         /* new right extent - oldext */
1737                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1738                                 r[1].br_startblock, r[1].br_blockcount,
1739                                 r[1].br_state)))
1740                                 goto done;
1741                         /* new left extent - oldext */
1742                         PREV.br_blockcount =
1743                                 new->br_startoff - PREV.br_startoff;
1744                         cur->bc_rec.b = PREV;
1745                         if ((error = xfs_bmbt_insert(cur, &i)))
1746                                 goto done;
1747                         ASSERT(i == 1);
1748                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1749                                 goto done;
1750                         ASSERT(i == 1);
1751                         /* new middle extent - newext */
1752                         cur->bc_rec.b = *new;
1753                         if ((error = xfs_bmbt_insert(cur, &i)))
1754                                 goto done;
1755                         ASSERT(i == 1);
1756                 }
1757                 /* DELTA: One in-core extent is split in three. */
1758                 temp = PREV.br_startoff;
1759                 temp2 = PREV.br_blockcount;
1760                 break;
1761
1762         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1763         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1764         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1765         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1766         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1767         case MASK(LEFT_CONTIG):
1768         case MASK(RIGHT_CONTIG):
1769                 /*
1770                  * These cases are all impossible.
1771                  */
1772                 ASSERT(0);
1773         }
1774         *curp = cur;
1775         if (delta) {
1776                 temp2 += temp;
1777                 if (delta->xed_startoff > temp)
1778                         delta->xed_startoff = temp;
1779                 if (delta->xed_blockcount < temp2)
1780                         delta->xed_blockcount = temp2;
1781         }
1782 done:
1783         *logflagsp = rval;
1784         return error;
1785 #undef  LEFT
1786 #undef  RIGHT
1787 #undef  PREV
1788 #undef  MASK
1789 #undef  MASK2
1790 #undef  MASK3
1791 #undef  MASK4
1792 #undef  STATE_SET
1793 #undef  STATE_TEST
1794 #undef  STATE_SET_TEST
1795 #undef  SWITCH_STATE
1796 }
1797
1798 /*
1799  * Called by xfs_bmap_add_extent to handle cases converting a hole
1800  * to a delayed allocation.
1801  */
1802 /*ARGSUSED*/
1803 STATIC int                              /* error */
1804 xfs_bmap_add_extent_hole_delay(
1805         xfs_inode_t             *ip,    /* incore inode pointer */
1806         xfs_extnum_t            idx,    /* extent number to update/insert */
1807         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1808         int                     *logflagsp, /* inode logging flags */
1809         xfs_extdelta_t          *delta, /* Change made to incore extents */
1810         int                     rsvd)           /* OK to allocate reserved blocks */
1811 {
1812         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1813         xfs_ifork_t             *ifp;   /* inode fork pointer */
1814         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1815         xfs_filblks_t           newlen=0;       /* new indirect size */
1816         xfs_filblks_t           oldlen=0;       /* old indirect size */
1817         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1818         int                     state;  /* state bits, accessed thru macros */
1819         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1820         xfs_filblks_t           temp2=0;
1821         enum {                          /* bit number definitions for state */
1822                 LEFT_CONTIG,    RIGHT_CONTIG,
1823                 LEFT_DELAY,     RIGHT_DELAY,
1824                 LEFT_VALID,     RIGHT_VALID
1825         };
1826
1827 #define MASK(b)                 (1 << (b))
1828 #define MASK2(a,b)              (MASK(a) | MASK(b))
1829 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1830 #define STATE_TEST(b)           (state & MASK(b))
1831 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1832                                        ((state &= ~MASK(b)), 0))
1833 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1834
1835         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1836         ep = xfs_iext_get_ext(ifp, idx);
1837         state = 0;
1838         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1839         /*
1840          * Check and set flags if this segment has a left neighbor
1841          */
1842         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1843                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1844                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1845         }
1846         /*
1847          * Check and set flags if the current (right) segment exists.
1848          * If it doesn't exist, we're converting the hole at end-of-file.
1849          */
1850         if (STATE_SET_TEST(RIGHT_VALID,
1851                            idx <
1852                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1853                 xfs_bmbt_get_all(ep, &right);
1854                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1855         }
1856         /*
1857          * Set contiguity flags on the left and right neighbors.
1858          * Don't let extents get too large, even if the pieces are contiguous.
1859          */
1860         STATE_SET(LEFT_CONTIG,
1861                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1862                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1863                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1864         STATE_SET(RIGHT_CONTIG,
1865                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1866                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1867                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1868                 (!STATE_TEST(LEFT_CONTIG) ||
1869                  (left.br_blockcount + new->br_blockcount +
1870                      right.br_blockcount <= MAXEXTLEN)));
1871         /*
1872          * Switch out based on the contiguity flags.
1873          */
1874         switch (SWITCH_STATE) {
1875
1876         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1877                 /*
1878                  * New allocation is contiguous with delayed allocations
1879                  * on the left and on the right.
1880                  * Merge all three into a single extent record.
1881                  */
1882                 temp = left.br_blockcount + new->br_blockcount +
1883                         right.br_blockcount;
1884                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
1885                         XFS_DATA_FORK);
1886                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1887                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1888                         STARTBLOCKVAL(new->br_startblock) +
1889                         STARTBLOCKVAL(right.br_startblock);
1890                 newlen = xfs_bmap_worst_indlen(ip, temp);
1891                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1892                         NULLSTARTBLOCK((int)newlen));
1893                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
1894                         XFS_DATA_FORK);
1895                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, XFS_DATA_FORK);
1896                 xfs_iext_remove(ifp, idx, 1);
1897                 ip->i_df.if_lastex = idx - 1;
1898                 /* DELTA: Two in-core extents were replaced by one. */
1899                 temp2 = temp;
1900                 temp = left.br_startoff;
1901                 break;
1902
1903         case MASK(LEFT_CONTIG):
1904                 /*
1905                  * New allocation is contiguous with a delayed allocation
1906                  * on the left.
1907                  * Merge the new allocation with the left neighbor.
1908                  */
1909                 temp = left.br_blockcount + new->br_blockcount;
1910                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1,
1911                         XFS_DATA_FORK);
1912                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1913                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1914                         STARTBLOCKVAL(new->br_startblock);
1915                 newlen = xfs_bmap_worst_indlen(ip, temp);
1916                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1917                         NULLSTARTBLOCK((int)newlen));
1918                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1,
1919                         XFS_DATA_FORK);
1920                 ip->i_df.if_lastex = idx - 1;
1921                 /* DELTA: One in-core extent grew into a hole. */
1922                 temp2 = temp;
1923                 temp = left.br_startoff;
1924                 break;
1925
1926         case MASK(RIGHT_CONTIG):
1927                 /*
1928                  * New allocation is contiguous with a delayed allocation
1929                  * on the right.
1930                  * Merge the new allocation with the right neighbor.
1931                  */
1932                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1933                 temp = new->br_blockcount + right.br_blockcount;
1934                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1935                         STARTBLOCKVAL(right.br_startblock);
1936                 newlen = xfs_bmap_worst_indlen(ip, temp);
1937                 xfs_bmbt_set_allf(ep, new->br_startoff,
1938                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1939                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1940                 ip->i_df.if_lastex = idx;
1941                 /* DELTA: One in-core extent grew into a hole. */
1942                 temp2 = temp;
1943                 temp = new->br_startoff;
1944                 break;
1945
1946         case 0:
1947                 /*
1948                  * New allocation is not contiguous with another
1949                  * delayed allocation.
1950                  * Insert a new entry.
1951                  */
1952                 oldlen = newlen = 0;
1953                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL,
1954                         XFS_DATA_FORK);
1955                 xfs_iext_insert(ifp, idx, 1, new);
1956                 ip->i_df.if_lastex = idx;
1957                 /* DELTA: A new in-core extent was added in a hole. */
1958                 temp2 = new->br_blockcount;
1959                 temp = new->br_startoff;
1960                 break;
1961         }
1962         if (oldlen != newlen) {
1963                 ASSERT(oldlen > newlen);
1964                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1965                         (int64_t)(oldlen - newlen), rsvd);
1966                 /*
1967                  * Nothing to do for disk quota accounting here.
1968                  */
1969         }
1970         if (delta) {
1971                 temp2 += temp;
1972                 if (delta->xed_startoff > temp)
1973                         delta->xed_startoff = temp;
1974                 if (delta->xed_blockcount < temp2)
1975                         delta->xed_blockcount = temp2;
1976         }
1977         *logflagsp = 0;
1978         return 0;
1979 #undef  MASK
1980 #undef  MASK2
1981 #undef  STATE_SET
1982 #undef  STATE_TEST
1983 #undef  STATE_SET_TEST
1984 #undef  SWITCH_STATE
1985 }
1986
1987 /*
1988  * Called by xfs_bmap_add_extent to handle cases converting a hole
1989  * to a real allocation.
1990  */
1991 STATIC int                              /* error */
1992 xfs_bmap_add_extent_hole_real(
1993         xfs_inode_t             *ip,    /* incore inode pointer */
1994         xfs_extnum_t            idx,    /* extent number to update/insert */
1995         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1996         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1997         int                     *logflagsp, /* inode logging flags */
1998         xfs_extdelta_t          *delta, /* Change made to incore extents */
1999         int                     whichfork) /* data or attr fork */
2000 {
2001         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
2002         int                     error;  /* error return value */
2003         int                     i;      /* temp state */
2004         xfs_ifork_t             *ifp;   /* inode fork pointer */
2005         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2006         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2007         int                     rval=0; /* return value (logging flags) */
2008         int                     state;  /* state bits, accessed thru macros */
2009         xfs_filblks_t           temp=0;
2010         xfs_filblks_t           temp2=0;
2011         enum {                          /* bit number definitions for state */
2012                 LEFT_CONTIG,    RIGHT_CONTIG,
2013                 LEFT_DELAY,     RIGHT_DELAY,
2014                 LEFT_VALID,     RIGHT_VALID
2015         };
2016
2017 #define MASK(b)                 (1 << (b))
2018 #define MASK2(a,b)              (MASK(a) | MASK(b))
2019 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2020 #define STATE_TEST(b)           (state & MASK(b))
2021 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2022                                        ((state &= ~MASK(b)), 0))
2023 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2024
2025         ifp = XFS_IFORK_PTR(ip, whichfork);
2026         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2027         ep = xfs_iext_get_ext(ifp, idx);
2028         state = 0;
2029         /*
2030          * Check and set flags if this segment has a left neighbor.
2031          */
2032         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2033                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2034                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2035         }
2036         /*
2037          * Check and set flags if this segment has a current value.
2038          * Not true if we're inserting into the "hole" at eof.
2039          */
2040         if (STATE_SET_TEST(RIGHT_VALID,
2041                            idx <
2042                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2043                 xfs_bmbt_get_all(ep, &right);
2044                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2045         }
2046         /*
2047          * We're inserting a real allocation between "left" and "right".
2048          * Set the contiguity flags.  Don't let extents get too large.
2049          */
2050         STATE_SET(LEFT_CONTIG,
2051                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2052                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2053                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2054                 left.br_state == new->br_state &&
2055                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2056         STATE_SET(RIGHT_CONTIG,
2057                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2058                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2059                 new->br_startblock + new->br_blockcount ==
2060                     right.br_startblock &&
2061                 new->br_state == right.br_state &&
2062                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2063                 (!STATE_TEST(LEFT_CONTIG) ||
2064                  left.br_blockcount + new->br_blockcount +
2065                      right.br_blockcount <= MAXEXTLEN));
2066
2067         error = 0;
2068         /*
2069          * Select which case we're in here, and implement it.
2070          */
2071         switch (SWITCH_STATE) {
2072
2073         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2074                 /*
2075                  * New allocation is contiguous with real allocations on the
2076                  * left and on the right.
2077                  * Merge all three into a single extent record.
2078                  */
2079                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
2080                         whichfork);
2081                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2082                         left.br_blockcount + new->br_blockcount +
2083                         right.br_blockcount);
2084                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
2085                         whichfork);
2086                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, whichfork);
2087                 xfs_iext_remove(ifp, idx, 1);
2088                 ifp->if_lastex = idx - 1;
2089                 XFS_IFORK_NEXT_SET(ip, whichfork,
2090                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2091                 if (cur == NULL) {
2092                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2093                 } else {
2094                         rval = XFS_ILOG_CORE;
2095                         if ((error = xfs_bmbt_lookup_eq(cur,
2096                                         right.br_startoff,
2097                                         right.br_startblock,
2098                                         right.br_blockcount, &i)))
2099                                 goto done;
2100                         ASSERT(i == 1);
2101                         if ((error = xfs_bmbt_delete(cur, &i)))
2102                                 goto done;
2103                         ASSERT(i == 1);
2104                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2105                                 goto done;
2106                         ASSERT(i == 1);
2107                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2108                                         left.br_startblock,
2109                                         left.br_blockcount +
2110                                                 new->br_blockcount +
2111                                                 right.br_blockcount,
2112                                         left.br_state)))
2113                                 goto done;
2114                 }
2115                 /* DELTA: Two in-core extents were replaced by one. */
2116                 temp = left.br_startoff;
2117                 temp2 = left.br_blockcount +
2118                         new->br_blockcount +
2119                         right.br_blockcount;
2120                 break;
2121
2122         case MASK(LEFT_CONTIG):
2123                 /*
2124                  * New allocation is contiguous with a real allocation
2125                  * on the left.
2126                  * Merge the new allocation with the left neighbor.
2127                  */
2128                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1, whichfork);
2129                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2130                         left.br_blockcount + new->br_blockcount);
2131                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1, whichfork);
2132                 ifp->if_lastex = idx - 1;
2133                 if (cur == NULL) {
2134                         rval = XFS_ILOG_FEXT(whichfork);
2135                 } else {
2136                         rval = 0;
2137                         if ((error = xfs_bmbt_lookup_eq(cur,
2138                                         left.br_startoff,
2139                                         left.br_startblock,
2140                                         left.br_blockcount, &i)))
2141                                 goto done;
2142                         ASSERT(i == 1);
2143                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2144                                         left.br_startblock,
2145                                         left.br_blockcount +
2146                                                 new->br_blockcount,
2147                                         left.br_state)))
2148                                 goto done;
2149                 }
2150                 /* DELTA: One in-core extent grew. */
2151                 temp = left.br_startoff;
2152                 temp2 = left.br_blockcount +
2153                         new->br_blockcount;
2154                 break;
2155
2156         case MASK(RIGHT_CONTIG):
2157                 /*
2158                  * New allocation is contiguous with a real allocation
2159                  * on the right.
2160                  * Merge the new allocation with the right neighbor.
2161                  */
2162                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, whichfork);
2163                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2164                         new->br_blockcount + right.br_blockcount,
2165                         right.br_state);
2166                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, whichfork);
2167                 ifp->if_lastex = idx;
2168                 if (cur == NULL) {
2169                         rval = XFS_ILOG_FEXT(whichfork);
2170                 } else {
2171                         rval = 0;
2172                         if ((error = xfs_bmbt_lookup_eq(cur,
2173                                         right.br_startoff,
2174                                         right.br_startblock,
2175                                         right.br_blockcount, &i)))
2176                                 goto done;
2177                         ASSERT(i == 1);
2178                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2179                                         new->br_startblock,
2180                                         new->br_blockcount +
2181                                                 right.br_blockcount,
2182                                         right.br_state)))
2183                                 goto done;
2184                 }
2185                 /* DELTA: One in-core extent grew. */
2186                 temp = new->br_startoff;
2187                 temp2 = new->br_blockcount +
2188                         right.br_blockcount;
2189                 break;
2190
2191         case 0:
2192                 /*
2193                  * New allocation is not contiguous with another
2194                  * real allocation.
2195                  * Insert a new entry.
2196                  */
2197                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL, whichfork);
2198                 xfs_iext_insert(ifp, idx, 1, new);
2199                 ifp->if_lastex = idx;
2200                 XFS_IFORK_NEXT_SET(ip, whichfork,
2201                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2202                 if (cur == NULL) {
2203                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2204                 } else {
2205                         rval = XFS_ILOG_CORE;
2206                         if ((error = xfs_bmbt_lookup_eq(cur,
2207                                         new->br_startoff,
2208                                         new->br_startblock,
2209                                         new->br_blockcount, &i)))
2210                                 goto done;
2211                         ASSERT(i == 0);
2212                         cur->bc_rec.b.br_state = new->br_state;
2213                         if ((error = xfs_bmbt_insert(cur, &i)))
2214                                 goto done;
2215                         ASSERT(i == 1);
2216                 }
2217                 /* DELTA: A new extent was added in a hole. */
2218                 temp = new->br_startoff;
2219                 temp2 = new->br_blockcount;
2220                 break;
2221         }
2222         if (delta) {
2223                 temp2 += temp;
2224                 if (delta->xed_startoff > temp)
2225                         delta->xed_startoff = temp;
2226                 if (delta->xed_blockcount < temp2)
2227                         delta->xed_blockcount = temp2;
2228         }
2229 done:
2230         *logflagsp = rval;
2231         return error;
2232 #undef  MASK
2233 #undef  MASK2
2234 #undef  STATE_SET
2235 #undef  STATE_TEST
2236 #undef  STATE_SET_TEST
2237 #undef  SWITCH_STATE
2238 }
2239
2240 /*
2241  * Adjust the size of the new extent based on di_extsize and rt extsize.
2242  */
2243 STATIC int
2244 xfs_bmap_extsize_align(
2245         xfs_mount_t     *mp,
2246         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2247         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2248         xfs_extlen_t    extsz,          /* align to this extent size */
2249         int             rt,             /* is this a realtime inode? */
2250         int             eof,            /* is extent at end-of-file? */
2251         int             delay,          /* creating delalloc extent? */
2252         int             convert,        /* overwriting unwritten extent? */
2253         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2254         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2255 {
2256         xfs_fileoff_t   orig_off;       /* original offset */
2257         xfs_extlen_t    orig_alen;      /* original length */
2258         xfs_fileoff_t   orig_end;       /* original off+len */
2259         xfs_fileoff_t   nexto;          /* next file offset */
2260         xfs_fileoff_t   prevo;          /* previous file offset */
2261         xfs_fileoff_t   align_off;      /* temp for offset */
2262         xfs_extlen_t    align_alen;     /* temp for length */
2263         xfs_extlen_t    temp;           /* temp for calculations */
2264
2265         if (convert)
2266                 return 0;
2267
2268         orig_off = align_off = *offp;
2269         orig_alen = align_alen = *lenp;
2270         orig_end = orig_off + orig_alen;
2271
2272         /*
2273          * If this request overlaps an existing extent, then don't
2274          * attempt to perform any additional alignment.
2275          */
2276         if (!delay && !eof &&
2277             (orig_off >= gotp->br_startoff) &&
2278             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2279                 return 0;
2280         }
2281
2282         /*
2283          * If the file offset is unaligned vs. the extent size
2284          * we need to align it.  This will be possible unless
2285          * the file was previously written with a kernel that didn't
2286          * perform this alignment, or if a truncate shot us in the
2287          * foot.
2288          */
2289         temp = do_mod(orig_off, extsz);
2290         if (temp) {
2291                 align_alen += temp;
2292                 align_off -= temp;
2293         }
2294         /*
2295          * Same adjustment for the end of the requested area.
2296          */
2297         if ((temp = (align_alen % extsz))) {
2298                 align_alen += extsz - temp;
2299         }
2300         /*
2301          * If the previous block overlaps with this proposed allocation
2302          * then move the start forward without adjusting the length.
2303          */
2304         if (prevp->br_startoff != NULLFILEOFF) {
2305                 if (prevp->br_startblock == HOLESTARTBLOCK)
2306                         prevo = prevp->br_startoff;
2307                 else
2308                         prevo = prevp->br_startoff + prevp->br_blockcount;
2309         } else
2310                 prevo = 0;
2311         if (align_off != orig_off && align_off < prevo)
2312                 align_off = prevo;
2313         /*
2314          * If the next block overlaps with this proposed allocation
2315          * then move the start back without adjusting the length,
2316          * but not before offset 0.
2317          * This may of course make the start overlap previous block,
2318          * and if we hit the offset 0 limit then the next block
2319          * can still overlap too.
2320          */
2321         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2322                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2323                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2324                         nexto = gotp->br_startoff + gotp->br_blockcount;
2325                 else
2326                         nexto = gotp->br_startoff;
2327         } else
2328                 nexto = NULLFILEOFF;
2329         if (!eof &&
2330             align_off + align_alen != orig_end &&
2331             align_off + align_alen > nexto)
2332                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2333         /*
2334          * If we're now overlapping the next or previous extent that
2335          * means we can't fit an extsz piece in this hole.  Just move
2336          * the start forward to the first valid spot and set
2337          * the length so we hit the end.
2338          */
2339         if (align_off != orig_off && align_off < prevo)
2340                 align_off = prevo;
2341         if (align_off + align_alen != orig_end &&
2342             align_off + align_alen > nexto &&
2343             nexto != NULLFILEOFF) {
2344                 ASSERT(nexto > prevo);
2345                 align_alen = nexto - align_off;
2346         }
2347
2348         /*
2349          * If realtime, and the result isn't a multiple of the realtime
2350          * extent size we need to remove blocks until it is.
2351          */
2352         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2353                 /*
2354                  * We're not covering the original request, or
2355                  * we won't be able to once we fix the length.
2356                  */
2357                 if (orig_off < align_off ||
2358                     orig_end > align_off + align_alen ||
2359                     align_alen - temp < orig_alen)
2360                         return XFS_ERROR(EINVAL);
2361                 /*
2362                  * Try to fix it by moving the start up.
2363                  */
2364                 if (align_off + temp <= orig_off) {
2365                         align_alen -= temp;
2366                         align_off += temp;
2367                 }
2368                 /*
2369                  * Try to fix it by moving the end in.
2370                  */
2371                 else if (align_off + align_alen - temp >= orig_end)
2372                         align_alen -= temp;
2373                 /*
2374                  * Set the start to the minimum then trim the length.
2375                  */
2376                 else {
2377                         align_alen -= orig_off - align_off;
2378                         align_off = orig_off;
2379                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2380                 }
2381                 /*
2382                  * Result doesn't cover the request, fail it.
2383                  */
2384                 if (orig_off < align_off || orig_end > align_off + align_alen)
2385                         return XFS_ERROR(EINVAL);
2386         } else {
2387                 ASSERT(orig_off >= align_off);
2388                 ASSERT(orig_end <= align_off + align_alen);
2389         }
2390
2391 #ifdef DEBUG
2392         if (!eof && gotp->br_startoff != NULLFILEOFF)
2393                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2394         if (prevp->br_startoff != NULLFILEOFF)
2395                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2396 #endif
2397
2398         *lenp = align_alen;
2399         *offp = align_off;
2400         return 0;
2401 }
2402
2403 #define XFS_ALLOC_GAP_UNITS     4
2404
2405 STATIC void
2406 xfs_bmap_adjacent(
2407         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2408 {
2409         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2410         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2411         xfs_mount_t     *mp;            /* mount point structure */
2412         int             nullfb;         /* true if ap->firstblock isn't set */
2413         int             rt;             /* true if inode is realtime */
2414
2415 #define ISVALID(x,y)    \
2416         (rt ? \
2417                 (x) < mp->m_sb.sb_rblocks : \
2418                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2419                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2420                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2421
2422         mp = ap->ip->i_mount;
2423         nullfb = ap->firstblock == NULLFSBLOCK;
2424         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2425         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2426         /*
2427          * If allocating at eof, and there's a previous real block,
2428          * try to use it's last block as our starting point.
2429          */
2430         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2431             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2432             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2433                     ap->prevp->br_startblock)) {
2434                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2435                 /*
2436                  * Adjust for the gap between prevp and us.
2437                  */
2438                 adjust = ap->off -
2439                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2440                 if (adjust &&
2441                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2442                         ap->rval += adjust;
2443         }
2444         /*
2445          * If not at eof, then compare the two neighbor blocks.
2446          * Figure out whether either one gives us a good starting point,
2447          * and pick the better one.
2448          */
2449         else if (!ap->eof) {
2450                 xfs_fsblock_t   gotbno;         /* right side block number */
2451                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2452                 xfs_fsblock_t   prevbno;        /* left side block number */
2453                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2454
2455                 /*
2456                  * If there's a previous (left) block, select a requested
2457                  * start block based on it.
2458                  */
2459                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2460                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2461                     (prevbno = ap->prevp->br_startblock +
2462                                ap->prevp->br_blockcount) &&
2463                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2464                         /*
2465                          * Calculate gap to end of previous block.
2466                          */
2467                         adjust = prevdiff = ap->off -
2468                                 (ap->prevp->br_startoff +
2469                                  ap->prevp->br_blockcount);
2470                         /*
2471                          * Figure the startblock based on the previous block's
2472                          * end and the gap size.
2473                          * Heuristic!
2474                          * If the gap is large relative to the piece we're
2475                          * allocating, or using it gives us an invalid block
2476                          * number, then just use the end of the previous block.
2477                          */
2478                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2479                             ISVALID(prevbno + prevdiff,
2480                                     ap->prevp->br_startblock))
2481                                 prevbno += adjust;
2482                         else
2483                                 prevdiff += adjust;
2484                         /*
2485                          * If the firstblock forbids it, can't use it,
2486                          * must use default.
2487                          */
2488                         if (!rt && !nullfb &&
2489                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2490                                 prevbno = NULLFSBLOCK;
2491                 }
2492                 /*
2493                  * No previous block or can't follow it, just default.
2494                  */
2495                 else
2496                         prevbno = NULLFSBLOCK;
2497                 /*
2498                  * If there's a following (right) block, select a requested
2499                  * start block based on it.
2500                  */
2501                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2502                         /*
2503                          * Calculate gap to start of next block.
2504                          */
2505                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2506                         /*
2507                          * Figure the startblock based on the next block's
2508                          * start and the gap size.
2509                          */
2510                         gotbno = ap->gotp->br_startblock;
2511                         /*
2512                          * Heuristic!
2513                          * If the gap is large relative to the piece we're
2514                          * allocating, or using it gives us an invalid block
2515                          * number, then just use the start of the next block
2516                          * offset by our length.
2517                          */
2518                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2519                             ISVALID(gotbno - gotdiff, gotbno))
2520                                 gotbno -= adjust;
2521                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2522                                 gotbno -= ap->alen;
2523                                 gotdiff += adjust - ap->alen;
2524                         } else
2525                                 gotdiff += adjust;
2526                         /*
2527                          * If the firstblock forbids it, can't use it,
2528                          * must use default.
2529                          */
2530                         if (!rt && !nullfb &&
2531                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2532                                 gotbno = NULLFSBLOCK;
2533                 }
2534                 /*
2535                  * No next block, just default.
2536                  */
2537                 else
2538                         gotbno = NULLFSBLOCK;
2539                 /*
2540                  * If both valid, pick the better one, else the only good
2541                  * one, else ap->rval is already set (to 0 or the inode block).
2542                  */
2543                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2544                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2545                 else if (prevbno != NULLFSBLOCK)
2546                         ap->rval = prevbno;
2547                 else if (gotbno != NULLFSBLOCK)
2548                         ap->rval = gotbno;
2549         }
2550 #undef ISVALID
2551 }
2552
2553 STATIC int
2554 xfs_bmap_rtalloc(
2555         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2556 {
2557         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2558         int             error;          /* error return value */
2559         xfs_mount_t     *mp;            /* mount point structure */
2560         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2561         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2562         xfs_extlen_t    align;          /* minimum allocation alignment */
2563         xfs_rtblock_t   rtb;
2564
2565         mp = ap->ip->i_mount;
2566         align = xfs_get_extsz_hint(ap->ip);
2567         prod = align / mp->m_sb.sb_rextsize;
2568         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2569                                         align, 1, ap->eof, 0,
2570                                         ap->conv, &ap->off, &ap->alen);
2571         if (error)
2572                 return error;
2573         ASSERT(ap->alen);
2574         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2575
2576         /*
2577          * If the offset & length are not perfectly aligned
2578          * then kill prod, it will just get us in trouble.
2579          */
2580         if (do_mod(ap->off, align) || ap->alen % align)
2581                 prod = 1;
2582         /*
2583          * Set ralen to be the actual requested length in rtextents.
2584          */
2585         ralen = ap->alen / mp->m_sb.sb_rextsize;
2586         /*
2587          * If the old value was close enough to MAXEXTLEN that
2588          * we rounded up to it, cut it back so it's valid again.
2589          * Note that if it's a really large request (bigger than
2590          * MAXEXTLEN), we don't hear about that number, and can't
2591          * adjust the starting point to match it.
2592          */
2593         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2594                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2595         /*
2596          * If it's an allocation to an empty file at offset 0,
2597          * pick an extent that will space things out in the rt area.
2598          */
2599         if (ap->eof && ap->off == 0) {
2600                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2601
2602                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2603                 if (error)
2604                         return error;
2605                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2606         } else {
2607                 ap->rval = 0;
2608         }
2609
2610         xfs_bmap_adjacent(ap);
2611
2612         /*
2613          * Realtime allocation, done through xfs_rtallocate_extent.
2614          */
2615         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2616         do_div(ap->rval, mp->m_sb.sb_rextsize);
2617         rtb = ap->rval;
2618         ap->alen = ralen;
2619         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2620                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2621                 return error;
2622         if (rtb == NULLFSBLOCK && prod > 1 &&
2623             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2624                                            ap->alen, &ralen, atype,
2625                                            ap->wasdel, 1, &rtb)))
2626                 return error;
2627         ap->rval = rtb;
2628         if (ap->rval != NULLFSBLOCK) {
2629                 ap->rval *= mp->m_sb.sb_rextsize;
2630                 ralen *= mp->m_sb.sb_rextsize;
2631                 ap->alen = ralen;
2632                 ap->ip->i_d.di_nblocks += ralen;
2633                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2634                 if (ap->wasdel)
2635                         ap->ip->i_delayed_blks -= ralen;
2636                 /*
2637                  * Adjust the disk quota also. This was reserved
2638                  * earlier.
2639                  */
2640                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2641                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2642                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2643         } else {
2644                 ap->alen = 0;
2645         }
2646         return 0;
2647 }
2648
2649 STATIC int
2650 xfs_bmap_btalloc(
2651         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2652 {
2653         xfs_mount_t     *mp;            /* mount point structure */
2654         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2655         xfs_extlen_t    align;          /* minimum allocation alignment */
2656         xfs_agnumber_t  ag;
2657         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2658         xfs_agnumber_t  startag;
2659         xfs_alloc_arg_t args;
2660         xfs_extlen_t    blen;
2661         xfs_extlen_t    delta;
2662         xfs_extlen_t    longest;
2663         xfs_extlen_t    need;
2664         xfs_extlen_t    nextminlen = 0;
2665         xfs_perag_t     *pag;
2666         int             nullfb;         /* true if ap->firstblock isn't set */
2667         int             isaligned;
2668         int             notinit;
2669         int             tryagain;
2670         int             error;
2671
2672         mp = ap->ip->i_mount;
2673         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2674         if (unlikely(align)) {
2675                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2676                                                 align, 0, ap->eof, 0, ap->conv,
2677                                                 &ap->off, &ap->alen);
2678                 ASSERT(!error);
2679                 ASSERT(ap->alen);
2680         }
2681         nullfb = ap->firstblock == NULLFSBLOCK;
2682         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2683         if (nullfb) {
2684                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2685                         ag = xfs_filestream_lookup_ag(ap->ip);
2686                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2687                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2688                 } else {
2689                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2690                 }
2691         } else
2692                 ap->rval = ap->firstblock;
2693
2694         xfs_bmap_adjacent(ap);
2695
2696         /*
2697          * If allowed, use ap->rval; otherwise must use firstblock since
2698          * it's in the right allocation group.
2699          */
2700         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2701                 ;
2702         else
2703                 ap->rval = ap->firstblock;
2704         /*
2705          * Normal allocation, done through xfs_alloc_vextent.
2706          */
2707         tryagain = isaligned = 0;
2708         args.tp = ap->tp;
2709         args.mp = mp;
2710         args.fsbno = ap->rval;
2711         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2712         args.firstblock = ap->firstblock;
2713         blen = 0;
2714         if (nullfb) {
2715                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2716                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2717                 else
2718                         args.type = XFS_ALLOCTYPE_START_BNO;
2719                 args.total = ap->total;
2720
2721                 /*
2722                  * Search for an allocation group with a single extent
2723                  * large enough for the request.
2724                  *
2725                  * If one isn't found, then adjust the minimum allocation
2726                  * size to the largest space found.
2727                  */
2728                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2729                 if (startag == NULLAGNUMBER)
2730                         startag = ag = 0;
2731                 notinit = 0;
2732                 down_read(&mp->m_peraglock);
2733                 while (blen < ap->alen) {
2734                         pag = &mp->m_perag[ag];
2735                         if (!pag->pagf_init &&
2736                             (error = xfs_alloc_pagf_init(mp, args.tp,
2737                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2738                                 up_read(&mp->m_peraglock);
2739                                 return error;
2740                         }
2741                         /*
2742                          * See xfs_alloc_fix_freelist...
2743                          */
2744                         if (pag->pagf_init) {
2745                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2746                                 delta = need > pag->pagf_flcount ?
2747                                         need - pag->pagf_flcount : 0;
2748                                 longest = (pag->pagf_longest > delta) ?
2749                                         (pag->pagf_longest - delta) :
2750                                         (pag->pagf_flcount > 0 ||
2751                                          pag->pagf_longest > 0);
2752                                 if (blen < longest)
2753                                         blen = longest;
2754                         } else
2755                                 notinit = 1;
2756
2757                         if (xfs_inode_is_filestream(ap->ip)) {
2758                                 if (blen >= ap->alen)
2759                                         break;
2760
2761                                 if (ap->userdata) {
2762                                         /*
2763                                          * If startag is an invalid AG, we've
2764                                          * come here once before and
2765                                          * xfs_filestream_new_ag picked the
2766                                          * best currently available.
2767                                          *
2768                                          * Don't continue looping, since we
2769                                          * could loop forever.
2770                                          */
2771                                         if (startag == NULLAGNUMBER)
2772                                                 break;
2773
2774                                         error = xfs_filestream_new_ag(ap, &ag);
2775                                         if (error) {
2776                                                 up_read(&mp->m_peraglock);
2777                                                 return error;
2778                                         }
2779
2780                                         /* loop again to set 'blen'*/
2781                                         startag = NULLAGNUMBER;
2782                                         continue;
2783                                 }
2784                         }
2785                         if (++ag == mp->m_sb.sb_agcount)
2786                                 ag = 0;
2787                         if (ag == startag)
2788                                 break;
2789                 }
2790                 up_read(&mp->m_peraglock);
2791                 /*
2792                  * Since the above loop did a BUF_TRYLOCK, it is
2793                  * possible that there is space for this request.
2794                  */
2795                 if (notinit || blen < ap->minlen)
2796                         args.minlen = ap->minlen;
2797                 /*
2798                  * If the best seen length is less than the request
2799                  * length, use the best as the minimum.
2800                  */
2801                 else if (blen < ap->alen)
2802                         args.minlen = blen;
2803                 /*
2804                  * Otherwise we've seen an extent as big as alen,
2805                  * use that as the minimum.
2806                  */
2807                 else
2808                         args.minlen = ap->alen;
2809
2810                 /*
2811                  * set the failure fallback case to look in the selected
2812                  * AG as the stream may have moved.
2813                  */
2814                 if (xfs_inode_is_filestream(ap->ip))
2815                         ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2816         } else if (ap->low) {
2817                 if (xfs_inode_is_filestream(ap->ip))
2818                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2819                 else
2820                         args.type = XFS_ALLOCTYPE_START_BNO;
2821                 args.total = args.minlen = ap->minlen;
2822         } else {
2823                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2824                 args.total = ap->total;
2825                 args.minlen = ap->minlen;
2826         }
2827         /* apply extent size hints if obtained earlier */
2828         if (unlikely(align)) {
2829                 args.prod = align;
2830                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2831                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2832         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2833                 args.prod = 1;
2834                 args.mod = 0;
2835         } else {
2836                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2837                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2838                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2839         }
2840         /*
2841          * If we are not low on available data blocks, and the
2842          * underlying logical volume manager is a stripe, and
2843          * the file offset is zero then try to allocate data
2844          * blocks on stripe unit boundary.
2845          * NOTE: ap->aeof is only set if the allocation length
2846          * is >= the stripe unit and the allocation offset is
2847          * at the end of file.
2848          */
2849         if (!ap->low && ap->aeof) {
2850                 if (!ap->off) {
2851                         args.alignment = mp->m_dalign;
2852                         atype = args.type;
2853                         isaligned = 1;
2854                         /*
2855                          * Adjust for alignment
2856                          */
2857                         if (blen > args.alignment && blen <= ap->alen)
2858                                 args.minlen = blen - args.alignment;
2859                         args.minalignslop = 0;
2860                 } else {
2861                         /*
2862                          * First try an exact bno allocation.
2863                          * If it fails then do a near or start bno
2864                          * allocation with alignment turned on.
2865                          */
2866                         atype = args.type;
2867                         tryagain = 1;
2868                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2869                         args.alignment = 1;
2870                         /*
2871                          * Compute the minlen+alignment for the
2872                          * next case.  Set slop so that the value
2873                          * of minlen+alignment+slop doesn't go up
2874                          * between the calls.
2875                          */
2876                         if (blen > mp->m_dalign && blen <= ap->alen)
2877                                 nextminlen = blen - mp->m_dalign;
2878                         else
2879                                 nextminlen = args.minlen;
2880                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2881                                 args.minalignslop =
2882                                         nextminlen + mp->m_dalign -
2883                                         args.minlen - 1;
2884                         else
2885                                 args.minalignslop = 0;
2886                 }
2887         } else {
2888                 args.alignment = 1;
2889                 args.minalignslop = 0;
2890         }
2891         args.minleft = ap->minleft;
2892         args.wasdel = ap->wasdel;
2893         args.isfl = 0;
2894         args.userdata = ap->userdata;
2895         if ((error = xfs_alloc_vextent(&args)))
2896                 return error;
2897         if (tryagain && args.fsbno == NULLFSBLOCK) {
2898                 /*
2899                  * Exact allocation failed. Now try with alignment
2900                  * turned on.
2901                  */
2902                 args.type = atype;
2903                 args.fsbno = ap->rval;
2904                 args.alignment = mp->m_dalign;
2905                 args.minlen = nextminlen;
2906                 args.minalignslop = 0;
2907                 isaligned = 1;
2908                 if ((error = xfs_alloc_vextent(&args)))
2909                         return error;
2910         }
2911         if (isaligned && args.fsbno == NULLFSBLOCK) {
2912                 /*
2913                  * allocation failed, so turn off alignment and
2914                  * try again.
2915                  */
2916                 args.type = atype;
2917                 args.fsbno = ap->rval;
2918                 args.alignment = 0;
2919                 if ((error = xfs_alloc_vextent(&args)))
2920                         return error;
2921         }
2922         if (args.fsbno == NULLFSBLOCK && nullfb &&
2923             args.minlen > ap->minlen) {
2924                 args.minlen = ap->minlen;
2925                 args.type = XFS_ALLOCTYPE_START_BNO;
2926                 args.fsbno = ap->rval;
2927                 if ((error = xfs_alloc_vextent(&args)))
2928                         return error;
2929         }
2930         if (args.fsbno == NULLFSBLOCK && nullfb) {
2931                 args.fsbno = 0;
2932                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2933                 args.total = ap->minlen;
2934                 args.minleft = 0;
2935                 if ((error = xfs_alloc_vextent(&args)))
2936                         return error;
2937                 ap->low = 1;
2938         }
2939         if (args.fsbno != NULLFSBLOCK) {
2940                 ap->firstblock = ap->rval = args.fsbno;
2941                 ASSERT(nullfb || fb_agno == args.agno ||
2942                        (ap->low && fb_agno < args.agno));
2943                 ap->alen = args.len;
2944                 ap->ip->i_d.di_nblocks += args.len;
2945                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2946                 if (ap->wasdel)
2947                         ap->ip->i_delayed_blks -= args.len;
2948                 /*
2949                  * Adjust the disk quota also. This was reserved
2950                  * earlier.
2951                  */
2952                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2953                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2954                                         XFS_TRANS_DQ_BCOUNT,
2955                         (long) args.len);
2956         } else {
2957                 ap->rval = NULLFSBLOCK;
2958                 ap->alen = 0;
2959         }
2960         return 0;
2961 }
2962
2963 /*
2964  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2965  * It figures out where to ask the underlying allocator to put the new extent.
2966  */
2967 STATIC int
2968 xfs_bmap_alloc(
2969         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2970 {
2971         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2972                 return xfs_bmap_rtalloc(ap);
2973         return xfs_bmap_btalloc(ap);
2974 }
2975
2976 /*
2977  * Transform a btree format file with only one leaf node, where the
2978  * extents list will fit in the inode, into an extents format file.
2979  * Since the file extents are already in-core, all we have to do is
2980  * give up the space for the btree root and pitch the leaf block.
2981  */
2982 STATIC int                              /* error */
2983 xfs_bmap_btree_to_extents(
2984         xfs_trans_t             *tp,    /* transaction pointer */
2985         xfs_inode_t             *ip,    /* incore inode pointer */
2986         xfs_btree_cur_t         *cur,   /* btree cursor */
2987         int                     *logflagsp, /* inode logging flags */
2988         int                     whichfork)  /* data or attr fork */
2989 {
2990         /* REFERENCED */
2991         xfs_bmbt_block_t        *cblock;/* child btree block */
2992         xfs_fsblock_t           cbno;   /* child block number */
2993         xfs_buf_t               *cbp;   /* child block's buffer */
2994         int                     error;  /* error return value */
2995         xfs_ifork_t             *ifp;   /* inode fork data */
2996         xfs_mount_t             *mp;    /* mount point structure */
2997         __be64                  *pp;    /* ptr to block address */
2998         xfs_bmbt_block_t        *rblock;/* root btree block */
2999
3000         ifp = XFS_IFORK_PTR(ip, whichfork);
3001         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3002         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3003         rblock = ifp->if_broot;
3004         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3005         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3006         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3007         mp = ip->i_mount;
3008         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3009         cbno = be64_to_cpu(*pp);
3010         *logflagsp = 0;
3011 #ifdef DEBUG
3012         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3013                 return error;
3014 #endif
3015         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3016                         XFS_BMAP_BTREE_REF)))
3017                 return error;
3018         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3019         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3020                 return error;
3021         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3022         ip->i_d.di_nblocks--;
3023         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3024         xfs_trans_binval(tp, cbp);
3025         if (cur->bc_bufs[0] == cbp)
3026                 cur->bc_bufs[0] = NULL;
3027         xfs_iroot_realloc(ip, -1, whichfork);
3028         ASSERT(ifp->if_broot == NULL);
3029         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3030         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3031         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3032         return 0;
3033 }
3034
3035 /*
3036  * Called by xfs_bmapi to update file extent records and the btree
3037  * after removing space (or undoing a delayed allocation).
3038  */
3039 STATIC int                              /* error */
3040 xfs_bmap_del_extent(
3041         xfs_inode_t             *ip,    /* incore inode pointer */
3042         xfs_trans_t             *tp,    /* current transaction pointer */
3043         xfs_extnum_t            idx,    /* extent number to update/delete */
3044         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3045         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3046         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3047         int                     *logflagsp, /* inode logging flags */
3048         xfs_extdelta_t          *delta, /* Change made to incore extents */
3049         int                     whichfork, /* data or attr fork */
3050         int                     rsvd)   /* OK to allocate reserved blocks */
3051 {
3052         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3053         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3054         xfs_fsblock_t           del_endblock=0; /* first block past del */
3055         xfs_fileoff_t           del_endoff;     /* first offset past del */
3056         int                     delay;  /* current block is delayed allocated */
3057         int                     do_fx;  /* free extent at end of routine */
3058         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
3059         int                     error;  /* error return value */
3060         int                     flags;  /* inode logging flags */
3061         xfs_bmbt_irec_t         got;    /* current extent entry */
3062         xfs_fileoff_t           got_endoff;     /* first offset past got */
3063         int                     i;      /* temp state */
3064         xfs_ifork_t             *ifp;   /* inode fork pointer */
3065         xfs_mount_t             *mp;    /* mount structure */
3066         xfs_filblks_t           nblks;  /* quota/sb block count */
3067         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3068         /* REFERENCED */
3069         uint                    qfield; /* quota field to update */
3070         xfs_filblks_t           temp;   /* for indirect length calculations */
3071         xfs_filblks_t           temp2;  /* for indirect length calculations */
3072
3073         XFS_STATS_INC(xs_del_exlist);
3074         mp = ip->i_mount;
3075         ifp = XFS_IFORK_PTR(ip, whichfork);
3076         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3077                 (uint)sizeof(xfs_bmbt_rec_t)));
3078         ASSERT(del->br_blockcount > 0);
3079         ep = xfs_iext_get_ext(ifp, idx);
3080         xfs_bmbt_get_all(ep, &got);
3081         ASSERT(got.br_startoff <= del->br_startoff);
3082         del_endoff = del->br_startoff + del->br_blockcount;
3083         got_endoff = got.br_startoff + got.br_blockcount;
3084         ASSERT(got_endoff >= del_endoff);
3085         delay = ISNULLSTARTBLOCK(got.br_startblock);
3086         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3087         flags = 0;
3088         qfield = 0;
3089         error = 0;
3090         /*
3091          * If deleting a real allocation, must free up the disk space.
3092          */
3093         if (!delay) {
3094                 flags = XFS_ILOG_CORE;
3095                 /*
3096                  * Realtime allocation.  Free it and record di_nblocks update.
3097                  */
3098                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
3099                         xfs_fsblock_t   bno;
3100                         xfs_filblks_t   len;
3101
3102                         ASSERT(do_mod(del->br_blockcount,
3103                                       mp->m_sb.sb_rextsize) == 0);
3104                         ASSERT(do_mod(del->br_startblock,
3105                                       mp->m_sb.sb_rextsize) == 0);
3106                         bno = del->br_startblock;
3107                         len = del->br_blockcount;
3108                         do_div(bno, mp->m_sb.sb_rextsize);
3109                         do_div(len, mp->m_sb.sb_rextsize);
3110                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3111                                         (xfs_extlen_t)len)))
3112                                 goto done;
3113                         do_fx = 0;
3114                         nblks = len * mp->m_sb.sb_rextsize;
3115                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3116                 }
3117                 /*
3118                  * Ordinary allocation.
3119                  */
3120                 else {
3121                         do_fx = 1;
3122                         nblks = del->br_blockcount;
3123                         qfield = XFS_TRANS_DQ_BCOUNT;
3124                 }
3125                 /*
3126                  * Set up del_endblock and cur for later.
3127                  */
3128                 del_endblock = del->br_startblock + del->br_blockcount;
3129                 if (cur) {
3130                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3131                                         got.br_startblock, got.br_blockcount,
3132                                         &i)))
3133                                 goto done;
3134                         ASSERT(i == 1);
3135                 }
3136                 da_old = da_new = 0;
3137         } else {
3138                 da_old = STARTBLOCKVAL(got.br_startblock);
3139                 da_new = 0;
3140                 nblks = 0;
3141                 do_fx = 0;
3142         }
3143         /*
3144          * Set flag value to use in switch statement.
3145          * Left-contig is 2, right-contig is 1.
3146          */
3147         switch (((got.br_startoff == del->br_startoff) << 1) |
3148                 (got_endoff == del_endoff)) {
3149         case 3:
3150                 /*
3151                  * Matches the whole extent.  Delete the entry.
3152                  */
3153                 XFS_BMAP_TRACE_DELETE("3", ip, idx, 1, whichfork);
3154                 xfs_iext_remove(ifp, idx, 1);
3155                 ifp->if_lastex = idx;
3156                 if (delay)
3157                         break;
3158                 XFS_IFORK_NEXT_SET(ip, whichfork,
3159                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3160                 flags |= XFS_ILOG_CORE;
3161                 if (!cur) {
3162                         flags |= XFS_ILOG_FEXT(whichfork);
3163                         break;
3164                 }
3165                 if ((error = xfs_bmbt_delete(cur, &i)))
3166                         goto done;
3167                 ASSERT(i == 1);
3168                 break;
3169
3170         case 2:
3171                 /*
3172                  * Deleting the first part of the extent.
3173                  */
3174                 XFS_BMAP_TRACE_PRE_UPDATE("2", ip, idx, whichfork);
3175                 xfs_bmbt_set_startoff(ep, del_endoff);
3176                 temp = got.br_blockcount - del->br_blockcount;
3177                 xfs_bmbt_set_blockcount(ep, temp);
3178                 ifp->if_lastex = idx;
3179                 if (delay) {
3180                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3181                                 da_old);
3182                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3183                         XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx,
3184                                 whichfork);
3185                         da_new = temp;
3186                         break;
3187                 }
3188                 xfs_bmbt_set_startblock(ep, del_endblock);
3189                 XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx, whichfork);
3190                 if (!cur) {
3191                         flags |= XFS_ILOG_FEXT(whichfork);
3192                         break;
3193                 }
3194                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3195                                 got.br_blockcount - del->br_blockcount,
3196                                 got.br_state)))
3197                         goto done;
3198                 break;
3199
3200         case 1:
3201                 /*
3202                  * Deleting the last part of the extent.
3203                  */
3204                 temp = got.br_blockcount - del->br_blockcount;
3205                 XFS_BMAP_TRACE_PRE_UPDATE("1", ip, idx, whichfork);
3206                 xfs_bmbt_set_blockcount(ep, temp);
3207                 ifp->if_lastex = idx;
3208                 if (delay) {
3209                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3210                                 da_old);
3211                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3212                         XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx,
3213                                 whichfork);
3214                         da_new = temp;
3215                         break;
3216                 }
3217                 XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx, whichfork);
3218                 if (!cur) {
3219                         flags |= XFS_ILOG_FEXT(whichfork);
3220                         break;
3221                 }
3222                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3223                                 got.br_startblock,
3224                                 got.br_blockcount - del->br_blockcount,
3225                                 got.br_state)))
3226                         goto done;
3227                 break;
3228
3229         case 0:
3230                 /*
3231                  * Deleting the middle of the extent.
3232                  */
3233                 temp = del->br_startoff - got.br_startoff;
3234                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, whichfork);
3235                 xfs_bmbt_set_blockcount(ep, temp);
3236                 new.br_startoff = del_endoff;
3237                 temp2 = got_endoff - del_endoff;
3238                 new.br_blockcount = temp2;
3239                 new.br_state = got.br_state;
3240                 if (!delay) {
3241                         new.br_startblock = del_endblock;
3242                         flags |= XFS_ILOG_CORE;
3243                         if (cur) {
3244                                 if ((error = xfs_bmbt_update(cur,
3245                                                 got.br_startoff,
3246                                                 got.br_startblock, temp,
3247                                                 got.br_state)))
3248                                         goto done;
3249                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3250                                         goto done;
3251                                 cur->bc_rec.b = new;
3252                                 error = xfs_bmbt_insert(cur, &i);
3253                                 if (error && error != ENOSPC)
3254                                         goto done;
3255                                 /*
3256                                  * If get no-space back from btree insert,
3257                                  * it tried a split, and we have a zero
3258                                  * block reservation.
3259                                  * Fix up our state and return the error.
3260                                  */
3261                                 if (error == ENOSPC) {
3262                                         /*
3263                                          * Reset the cursor, don't trust
3264                                          * it after any insert operation.
3265                                          */
3266                                         if ((error = xfs_bmbt_lookup_eq(cur,
3267                                                         got.br_startoff,
3268                                                         got.br_startblock,
3269                                                         temp, &i)))
3270                                                 goto done;
3271                                         ASSERT(i == 1);
3272                                         /*
3273                                          * Update the btree record back
3274                                          * to the original value.
3275                                          */
3276                                         if ((error = xfs_bmbt_update(cur,
3277                                                         got.br_startoff,
3278                                                         got.br_startblock,
3279                                                         got.br_blockcount,
3280                                                         got.br_state)))
3281                                                 goto done;
3282                                         /*
3283                                          * Reset the extent record back
3284                                          * to the original value.
3285                                          */
3286                                         xfs_bmbt_set_blockcount(ep,
3287                                                 got.br_blockcount);
3288                                         flags = 0;
3289                                         error = XFS_ERROR(ENOSPC);
3290                                         goto done;
3291                                 }
3292                                 ASSERT(i == 1);
3293                         } else
3294                                 flags |= XFS_ILOG_FEXT(whichfork);
3295                         XFS_IFORK_NEXT_SET(ip, whichfork,
3296                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3297                 } else {
3298                         ASSERT(whichfork == XFS_DATA_FORK);
3299                         temp = xfs_bmap_worst_indlen(ip, temp);
3300                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3301                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3302                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3303                         da_new = temp + temp2;
3304                         while (da_new > da_old) {
3305                                 if (temp) {
3306                                         temp--;
3307                                         da_new--;
3308                                         xfs_bmbt_set_startblock(ep,
3309                                                 NULLSTARTBLOCK((int)temp));
3310                                 }
3311                                 if (da_new == da_old)
3312                                         break;
3313                                 if (temp2) {
3314                                         temp2--;
3315                                         da_new--;
3316                                         new.br_startblock =
3317                                                 NULLSTARTBLOCK((int)temp2);
3318                                 }
3319                         }
3320                 }
3321                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, whichfork);
3322                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 1, &new, NULL,
3323                         whichfork);
3324                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3325                 ifp->if_lastex = idx + 1;
3326                 break;
3327         }
3328         /*
3329          * If we need to, add to list of extents to delete.
3330          */
3331         if (do_fx)
3332                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3333                         mp);
3334         /*
3335          * Adjust inode # blocks in the file.
3336          */
3337         if (nblks)
3338                 ip->i_d.di_nblocks -= nblks;
3339         /*
3340          * Adjust quota data.
3341          */
3342         if (qfield)
3343                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3344
3345         /*
3346          * Account for change in delayed indirect blocks.
3347          * Nothing to do for disk quota accounting here.
3348          */
3349         ASSERT(da_old >= da_new);
3350         if (da_old > da_new)
3351                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3352                         rsvd);
3353         if (delta) {
3354                 /* DELTA: report the original extent. */
3355                 if (delta->xed_startoff > got.br_startoff)
3356                         delta->xed_startoff = got.br_startoff;
3357                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3358                         delta->xed_blockcount = got.br_startoff +
3359                                                         got.br_blockcount;
3360         }
3361 done:
3362         *logflagsp = flags;
3363         return error;
3364 }
3365
3366 /*
3367  * Remove the entry "free" from the free item list.  Prev points to the
3368  * previous entry, unless "free" is the head of the list.
3369  */
3370 STATIC void
3371 xfs_bmap_del_free(
3372         xfs_bmap_free_t         *flist, /* free item list header */
3373         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3374         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3375 {
3376         if (prev)
3377                 prev->xbfi_next = free->xbfi_next;
3378         else
3379                 flist->xbf_first = free->xbfi_next;
3380         flist->xbf_count--;
3381         kmem_zone_free(xfs_bmap_free_item_zone, free);
3382 }
3383
3384 /*
3385  * Convert an extents-format file into a btree-format file.
3386  * The new file will have a root block (in the inode) and a single child block.
3387  */
3388 STATIC int                                      /* error */
3389 xfs_bmap_extents_to_btree(
3390         xfs_trans_t             *tp,            /* transaction pointer */
3391         xfs_inode_t             *ip,            /* incore inode pointer */
3392         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3393         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3394         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3395         int                     wasdel,         /* converting a delayed alloc */
3396         int                     *logflagsp,     /* inode logging flags */
3397         int                     whichfork)      /* data or attr fork */
3398 {
3399         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3400         xfs_buf_t               *abp;           /* buffer for ablock */
3401         xfs_alloc_arg_t         args;           /* allocation arguments */
3402         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3403         xfs_bmbt_block_t        *block;         /* btree root block */
3404         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3405         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3406         int                     error;          /* error return value */
3407         xfs_extnum_t            i, cnt;         /* extent record index */
3408         xfs_ifork_t             *ifp;           /* inode fork pointer */
3409         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3410         xfs_mount_t             *mp;            /* mount structure */
3411         xfs_extnum_t            nextents;       /* number of file extents */
3412         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3413
3414         ifp = XFS_IFORK_PTR(ip, whichfork);
3415         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3416         ASSERT(ifp->if_ext_max ==
3417                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3418         /*
3419          * Make space in the inode incore.
3420          */
3421         xfs_iroot_realloc(ip, 1, whichfork);
3422         ifp->if_flags |= XFS_IFBROOT;
3423         /*
3424          * Fill in the root.
3425          */
3426         block = ifp->if_broot;
3427         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3428         block->bb_level = cpu_to_be16(1);
3429         block->bb_numrecs = cpu_to_be16(1);
3430         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3431         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3432         /*
3433          * Need a cursor.  Can't allocate until bb_level is filled in.
3434          */
3435         mp = ip->i_mount;
3436         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3437                 whichfork);
3438         cur->bc_private.b.firstblock = *firstblock;
3439         cur->bc_private.b.flist = flist;
3440         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3441         /*
3442          * Convert to a btree with two levels, one record in root.
3443          */
3444         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3445         args.tp = tp;
3446         args.mp = mp;
3447         args.firstblock = *firstblock;
3448         if (*firstblock == NULLFSBLOCK) {
3449                 args.type = XFS_ALLOCTYPE_START_BNO;
3450                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3451         } else if (flist->xbf_low) {
3452                 args.type = XFS_ALLOCTYPE_START_BNO;
3453                 args.fsbno = *firstblock;
3454         } else {
3455                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3456                 args.fsbno = *firstblock;
3457         }
3458         args.minlen = args.maxlen = args.prod = 1;
3459         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3460                 args.minalignslop = 0;
3461         args.wasdel = wasdel;
3462         *logflagsp = 0;
3463         if ((error = xfs_alloc_vextent(&args))) {
3464                 xfs_iroot_realloc(ip, -1, whichfork);
3465                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3466                 return error;
3467         }
3468         /*
3469          * Allocation can't fail, the space was reserved.
3470          */
3471         ASSERT(args.fsbno != NULLFSBLOCK);
3472         ASSERT(*firstblock == NULLFSBLOCK ||
3473                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3474                (flist->xbf_low &&
3475                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3476         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3477         cur->bc_private.b.allocated++;
3478         ip->i_d.di_nblocks++;
3479         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3480         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3481         /*
3482          * Fill in the child block.
3483          */
3484         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3485         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3486         ablock->bb_level = 0;
3487         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3488         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3489         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3490         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3491         for (cnt = i = 0; i < nextents; i++) {
3492                 ep = xfs_iext_get_ext(ifp, i);
3493                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3494                         arp->l0 = cpu_to_be64(ep->l0);
3495                         arp->l1 = cpu_to_be64(ep->l1);
3496                         arp++; cnt++;
3497                 }
3498         }
3499         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3500         ablock->bb_numrecs = cpu_to_be16(cnt);
3501         /*
3502          * Fill in the root key and pointer.
3503          */
3504         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3505         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3506         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3507         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3508         *pp = cpu_to_be64(args.fsbno);
3509         /*
3510          * Do all this logging at the end so that
3511          * the root is at the right level.
3512          */
3513         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3514         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3515         ASSERT(*curp == NULL);
3516         *curp = cur;
3517         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3518         return 0;
3519 }
3520
3521 /*
3522  * Helper routine to reset inode di_forkoff field when switching
3523  * attribute fork from local to extent format - we reset it where
3524  * possible to make space available for inline data fork extents.
3525  */
3526 STATIC void
3527 xfs_bmap_forkoff_reset(
3528         xfs_mount_t     *mp,
3529         xfs_inode_t     *ip,
3530         int             whichfork)
3531 {
3532         if (whichfork == XFS_ATTR_FORK &&
3533             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3534             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3535             (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3536             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3537                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3538                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3539                                         (uint)sizeof(xfs_bmbt_rec_t);
3540                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3541                                         (uint)sizeof(xfs_bmbt_rec_t);
3542         }
3543 }
3544
3545 /*
3546  * Convert a local file to an extents file.
3547  * This code is out of bounds for data forks of regular files,
3548  * since the file data needs to get logged so things will stay consistent.
3549  * (The bmap-level manipulations are ok, though).
3550  */
3551 STATIC int                              /* error */
3552 xfs_bmap_local_to_extents(
3553         xfs_trans_t     *tp,            /* transaction pointer */
3554         xfs_inode_t     *ip,            /* incore inode pointer */
3555         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3556         xfs_extlen_t    total,          /* total blocks needed by transaction */
3557         int             *logflagsp,     /* inode logging flags */
3558         int             whichfork)      /* data or attr fork */
3559 {
3560         int             error;          /* error return value */
3561         int             flags;          /* logging flags returned */
3562         xfs_ifork_t     *ifp;           /* inode fork pointer */
3563
3564         /*
3565          * We don't want to deal with the case of keeping inode data inline yet.
3566          * So sending the data fork of a regular inode is invalid.
3567          */
3568         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3569                  whichfork == XFS_DATA_FORK));
3570         ifp = XFS_IFORK_PTR(ip, whichfork);
3571         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3572         flags = 0;
3573         error = 0;
3574         if (ifp->if_bytes) {
3575                 xfs_alloc_arg_t args;   /* allocation arguments */
3576                 xfs_buf_t       *bp;    /* buffer for extent block */
3577                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3578
3579                 args.tp = tp;
3580                 args.mp = ip->i_mount;
3581                 args.firstblock = *firstblock;
3582                 ASSERT((ifp->if_flags &
3583                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3584                 /*
3585                  * Allocate a block.  We know we need only one, since the
3586                  * file currently fits in an inode.
3587                  */
3588                 if (*firstblock == NULLFSBLOCK) {
3589                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3590                         args.type = XFS_ALLOCTYPE_START_BNO;
3591                 } else {
3592                         args.fsbno = *firstblock;
3593                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3594                 }
3595                 args.total = total;
3596                 args.mod = args.minleft = args.alignment = args.wasdel =
3597                         args.isfl = args.minalignslop = 0;
3598                 args.minlen = args.maxlen = args.prod = 1;
3599                 if ((error = xfs_alloc_vextent(&args)))
3600                         goto done;
3601                 /*
3602                  * Can't fail, the space was reserved.
3603                  */
3604                 ASSERT(args.fsbno != NULLFSBLOCK);
3605                 ASSERT(args.len == 1);
3606                 *firstblock = args.fsbno;
3607                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3608                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3609                         ifp->if_bytes);
3610                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3611                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3612                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3613                 xfs_iext_add(ifp, 0, 1);
3614                 ep = xfs_iext_get_ext(ifp, 0);
3615                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3616                 XFS_BMAP_TRACE_POST_UPDATE("new", ip, 0, whichfork);
3617                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3618                 ip->i_d.di_nblocks = 1;
3619                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3620                         XFS_TRANS_DQ_BCOUNT, 1L);
3621                 flags |= XFS_ILOG_FEXT(whichfork);
3622         } else {
3623                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3624                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3625         }
3626         ifp->if_flags &= ~XFS_IFINLINE;
3627         ifp->if_flags |= XFS_IFEXTENTS;
3628         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3629         flags |= XFS_ILOG_CORE;
3630 done:
3631         *logflagsp = flags;
3632         return error;
3633 }
3634
3635 /*
3636  * Search the extent records for the entry containing block bno.
3637  * If bno lies in a hole, point to the next entry.  If bno lies
3638  * past eof, *eofp will be set, and *prevp will contain the last
3639  * entry (null if none).  Else, *lastxp will be set to the index
3640  * of the found entry; *gotp will contain the entry.
3641  */
3642 xfs_bmbt_rec_host_t *                   /* pointer to found extent entry */
3643 xfs_bmap_search_multi_extents(
3644         xfs_ifork_t     *ifp,           /* inode fork pointer */
3645         xfs_fileoff_t   bno,            /* block number searched for */
3646         int             *eofp,          /* out: end of file found */
3647         xfs_extnum_t    *lastxp,        /* out: last extent index */
3648         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3649         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3650 {
3651         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3652         xfs_extnum_t    lastx;          /* last extent index */
3653
3654         /*
3655          * Initialize the extent entry structure to catch access to
3656          * uninitialized br_startblock field.
3657          */
3658         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3659         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3660         gotp->br_state = XFS_EXT_INVALID;
3661 #if XFS_BIG_BLKNOS
3662         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3663 #else
3664         gotp->br_startblock = 0xffffa5a5;
3665 #endif
3666         prevp->br_startoff = NULLFILEOFF;
3667
3668         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3669         if (lastx > 0) {
3670                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3671         }
3672         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3673                 xfs_bmbt_get_all(ep, gotp);
3674                 *eofp = 0;
3675         } else {
3676                 if (lastx > 0) {
3677                         *gotp = *prevp;
3678                 }
3679                 *eofp = 1;
3680                 ep = NULL;
3681         }
3682         *lastxp = lastx;
3683         return ep;
3684 }
3685
3686 /*
3687  * Search the extents list for the inode, for the extent containing bno.
3688  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3689  * *eofp will be set, and *prevp will contain the last entry (null if none).
3690  * Else, *lastxp will be set to the index of the found
3691  * entry; *gotp will contain the entry.
3692  */
3693 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3694 xfs_bmap_search_extents(
3695         xfs_inode_t     *ip,            /* incore inode pointer */
3696         xfs_fileoff_t   bno,            /* block number searched for */
3697         int             fork,           /* data or attr fork */
3698         int             *eofp,          /* out: end of file found */
3699         xfs_extnum_t    *lastxp,        /* out: last extent index */
3700         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3701         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3702 {
3703         xfs_ifork_t     *ifp;           /* inode fork pointer */
3704         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3705
3706         XFS_STATS_INC(xs_look_exlist);
3707         ifp = XFS_IFORK_PTR(ip, fork);
3708
3709         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3710
3711         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3712                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3713                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3714                                 "Access to block zero in inode %llu "
3715                                 "start_block: %llx start_off: %llx "
3716                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3717                         (unsigned long long)ip->i_ino,
3718                         (unsigned long long)gotp->br_startblock,
3719                         (unsigned long long)gotp->br_startoff,
3720                         (unsigned long long)gotp->br_blockcount,
3721                         gotp->br_state, *lastxp);
3722                 *lastxp = NULLEXTNUM;
3723                 *eofp = 1;
3724                 return NULL;
3725         }
3726         return ep;
3727 }
3728
3729
3730 #ifdef XFS_BMAP_TRACE
3731 ktrace_t        *xfs_bmap_trace_buf;
3732
3733 /*
3734  * Add a bmap trace buffer entry.  Base routine for the others.
3735  */
3736 STATIC void
3737 xfs_bmap_trace_addentry(
3738         int             opcode,         /* operation */
3739         const char      *fname,         /* function name */
3740         char            *desc,          /* operation description */
3741         xfs_inode_t     *ip,            /* incore inode pointer */
3742         xfs_extnum_t    idx,            /* index of entry(ies) */
3743         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3744         xfs_bmbt_rec_host_t *r1,        /* first record */
3745         xfs_bmbt_rec_host_t *r2,        /* second record or null */
3746         int             whichfork)      /* data or attr fork */
3747 {
3748         xfs_bmbt_rec_host_t tr2;
3749
3750         ASSERT(cnt == 1 || cnt == 2);
3751         ASSERT(r1 != NULL);
3752         if (cnt == 1) {
3753                 ASSERT(r2 == NULL);
3754                 r2 = &tr2;
3755                 memset(&tr2, 0, sizeof(tr2));
3756         } else
3757                 ASSERT(r2 != NULL);
3758         ktrace_enter(xfs_bmap_trace_buf,
3759                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3760                 (void *)fname, (void *)desc, (void *)ip,
3761                 (void *)(__psint_t)idx,
3762                 (void *)(__psint_t)cnt,
3763                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3764                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3765                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3766                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3767                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3768                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3769                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3770                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3771                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3772                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3773                 );
3774         ASSERT(ip->i_xtrace);
3775         ktrace_enter(ip->i_xtrace,
3776                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3777                 (void *)fname, (void *)desc, (void *)ip,
3778                 (void *)(__psint_t)idx,
3779                 (void *)(__psint_t)cnt,
3780                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3781                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3782                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3783                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3784                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3785                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3786                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3787                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3788                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3789                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3790                 );
3791 }
3792
3793 /*
3794  * Add bmap trace entry prior to a call to xfs_iext_remove.
3795  */
3796 STATIC void
3797 xfs_bmap_trace_delete(
3798         const char      *fname,         /* function name */
3799         char            *desc,          /* operation description */
3800         xfs_inode_t     *ip,            /* incore inode pointer */
3801         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3802         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3803         int             whichfork)      /* data or attr fork */
3804 {
3805         xfs_ifork_t     *ifp;           /* inode fork pointer */
3806
3807         ifp = XFS_IFORK_PTR(ip, whichfork);
3808         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3809                 cnt, xfs_iext_get_ext(ifp, idx),
3810                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3811                 whichfork);
3812 }
3813
3814 /*
3815  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3816  * reading in the extents list from the disk (in the btree).
3817  */
3818 STATIC void
3819 xfs_bmap_trace_insert(
3820         const char      *fname,         /* function name */
3821         char            *desc,          /* operation description */
3822         xfs_inode_t     *ip,            /* incore inode pointer */
3823         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3824         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3825         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3826         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3827         int             whichfork)      /* data or attr fork */
3828 {
3829         xfs_bmbt_rec_host_t tr1;        /* compressed record 1 */
3830         xfs_bmbt_rec_host_t tr2;        /* compressed record 2 if needed */
3831
3832         xfs_bmbt_set_all(&tr1, r1);
3833         if (cnt == 2) {
3834                 ASSERT(r2 != NULL);
3835                 xfs_bmbt_set_all(&tr2, r2);
3836         } else {
3837                 ASSERT(cnt == 1);
3838                 ASSERT(r2 == NULL);
3839         }
3840         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3841                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3842 }
3843
3844 /*
3845  * Add bmap trace entry after updating an extent record in place.
3846  */
3847 STATIC void
3848 xfs_bmap_trace_post_update(
3849         const char      *fname,         /* function name */
3850         char            *desc,          /* operation description */
3851         xfs_inode_t     *ip,            /* incore inode pointer */
3852         xfs_extnum_t    idx,            /* index of entry updated */
3853         int             whichfork)      /* data or attr fork */
3854 {
3855         xfs_ifork_t     *ifp;           /* inode fork pointer */
3856
3857         ifp = XFS_IFORK_PTR(ip, whichfork);
3858         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3859                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3860 }
3861
3862 /*
3863  * Add bmap trace entry prior to updating an extent record in place.
3864  */
3865 STATIC void
3866 xfs_bmap_trace_pre_update(
3867         const char      *fname,         /* function name */
3868         char            *desc,          /* operation description */
3869         xfs_inode_t     *ip,            /* incore inode pointer */
3870         xfs_extnum_t    idx,            /* index of entry to be updated */
3871         int             whichfork)      /* data or attr fork */
3872 {
3873         xfs_ifork_t     *ifp;           /* inode fork pointer */
3874
3875         ifp = XFS_IFORK_PTR(ip, whichfork);
3876         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3877                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3878 }
3879 #endif  /* XFS_BMAP_TRACE */
3880
3881 /*
3882  * Compute the worst-case number of indirect blocks that will be used
3883  * for ip's delayed extent of length "len".
3884  */
3885 STATIC xfs_filblks_t
3886 xfs_bmap_worst_indlen(
3887         xfs_inode_t     *ip,            /* incore inode pointer */
3888         xfs_filblks_t   len)            /* delayed extent length */
3889 {
3890         int             level;          /* btree level number */
3891         int             maxrecs;        /* maximum record count at this level */
3892         xfs_mount_t     *mp;            /* mount structure */
3893         xfs_filblks_t   rval;           /* return value */
3894
3895         mp = ip->i_mount;
3896         maxrecs = mp->m_bmap_dmxr[0];
3897         for (level = 0, rval = 0;
3898              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3899              level++) {
3900                 len += maxrecs - 1;
3901                 do_div(len, maxrecs);
3902                 rval += len;
3903                 if (len == 1)
3904                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3905                                 level - 1;
3906                 if (level == 0)
3907                         maxrecs = mp->m_bmap_dmxr[1];
3908         }
3909         return rval;
3910 }
3911
3912 #if defined(XFS_RW_TRACE)
3913 STATIC void
3914 xfs_bunmap_trace(
3915         xfs_inode_t             *ip,
3916         xfs_fileoff_t           bno,
3917         xfs_filblks_t           len,
3918         int                     flags,
3919         inst_t                  *ra)
3920 {
3921         if (ip->i_rwtrace == NULL)
3922                 return;
3923         ktrace_enter(ip->i_rwtrace,
3924                 (void *)(__psint_t)XFS_BUNMAP,
3925                 (void *)ip,
3926                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3927                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3928                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3929                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3930                 (void *)(__psint_t)len,
3931                 (void *)(__psint_t)flags,
3932                 (void *)(unsigned long)current_cpu(),
3933                 (void *)ra,
3934                 (void *)0,
3935                 (void *)0,
3936                 (void *)0,
3937                 (void *)0,
3938                 (void *)0,
3939                 (void *)0);
3940 }
3941 #endif
3942
3943 /*
3944  * Convert inode from non-attributed to attributed.
3945  * Must not be in a transaction, ip must not be locked.
3946  */
3947 int                                             /* error code */
3948 xfs_bmap_add_attrfork(
3949         xfs_inode_t             *ip,            /* incore inode pointer */
3950         int                     size,           /* space new attribute needs */
3951         int                     rsvd)           /* xact may use reserved blks */
3952 {
3953         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3954         xfs_bmap_free_t         flist;          /* freed extent records */
3955         xfs_mount_t             *mp;            /* mount structure */
3956         xfs_trans_t             *tp;            /* transaction pointer */
3957         int                     blks;           /* space reservation */
3958         int                     version = 1;    /* superblock attr version */
3959         int                     committed;      /* xaction was committed */
3960         int                     logflags;       /* logging flags */
3961         int                     error;          /* error return value */
3962
3963         ASSERT(XFS_IFORK_Q(ip) == 0);
3964         ASSERT(ip->i_df.if_ext_max ==
3965                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3966
3967         mp = ip->i_mount;
3968         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3969         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3970         blks = XFS_ADDAFORK_SPACE_RES(mp);
3971         if (rsvd)
3972                 tp->t_flags |= XFS_TRANS_RESERVE;
3973         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3974                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3975                 goto error0;
3976         xfs_ilock(ip, XFS_ILOCK_EXCL);
3977         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3978                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3979                         XFS_QMOPT_RES_REGBLKS);
3980         if (error) {
3981                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3982                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3983                 return error;
3984         }
3985         if (XFS_IFORK_Q(ip))
3986                 goto error1;
3987         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3988                 /*
3989                  * For inodes coming from pre-6.2 filesystems.
3990                  */
3991                 ASSERT(ip->i_d.di_aformat == 0);
3992                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3993         }
3994         ASSERT(ip->i_d.di_anextents == 0);
3995         VN_HOLD(XFS_ITOV(ip));
3996         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3997         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3998         switch (ip->i_d.di_format) {
3999         case XFS_DINODE_FMT_DEV:
4000                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4001                 break;
4002         case XFS_DINODE_FMT_UUID:
4003                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4004                 break;
4005         case XFS_DINODE_FMT_LOCAL:
4006         case XFS_DINODE_FMT_EXTENTS:
4007         case XFS_DINODE_FMT_BTREE:
4008                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4009                 if (!ip->i_d.di_forkoff)
4010                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4011                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4012                         version = 2;
4013                 break;
4014         default:
4015                 ASSERT(0);
4016                 error = XFS_ERROR(EINVAL);
4017                 goto error1;
4018         }
4019         ip->i_df.if_ext_max =
4020                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4021         ASSERT(ip->i_afp == NULL);
4022         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4023         ip->i_afp->if_ext_max =
4024                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4025         ip->i_afp->if_flags = XFS_IFEXTENTS;
4026         logflags = 0;
4027         XFS_BMAP_INIT(&flist, &firstblock);
4028         switch (ip->i_d.di_format) {
4029         case XFS_DINODE_FMT_LOCAL:
4030                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4031                         &logflags);
4032                 break;
4033         case XFS_DINODE_FMT_EXTENTS:
4034                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4035                         &flist, &logflags);
4036                 break;
4037         case XFS_DINODE_FMT_BTREE:
4038                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4039                         &logflags);
4040                 break;
4041         default:
4042                 error = 0;
4043                 break;
4044         }
4045         if (logflags)
4046                 xfs_trans_log_inode(tp, ip, logflags);
4047         if (error)
4048                 goto error2;
4049         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
4050            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
4051                 __int64_t sbfields = 0;
4052
4053                 spin_lock(&mp->m_sb_lock);
4054                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
4055                         xfs_sb_version_addattr(&mp->m_sb);
4056                         sbfields |= XFS_SB_VERSIONNUM;
4057                 }
4058                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
4059                         xfs_sb_version_addattr2(&mp->m_sb);
4060                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4061                 }
4062                 if (sbfields) {
4063                         spin_unlock(&mp->m_sb_lock);
4064                         xfs_mod_sb(tp, sbfields);
4065                 } else
4066                         spin_unlock(&mp->m_sb_lock);
4067         }
4068         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4069                 goto error2;
4070         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4071         ASSERT(ip->i_df.if_ext_max ==
4072                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4073         return error;
4074 error2:
4075         xfs_bmap_cancel(&flist);
4076 error1:
4077         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4078 error0:
4079         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4080         ASSERT(ip->i_df.if_ext_max ==
4081                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4082         return error;
4083 }
4084
4085 /*
4086  * Add the extent to the list of extents to be free at transaction end.
4087  * The list is maintained sorted (by block number).
4088  */
4089 /* ARGSUSED */
4090 void
4091 xfs_bmap_add_free(
4092         xfs_fsblock_t           bno,            /* fs block number of extent */
4093         xfs_filblks_t           len,            /* length of extent */
4094         xfs_bmap_free_t         *flist,         /* list of extents */
4095         xfs_mount_t             *mp)            /* mount point structure */
4096 {
4097         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4098         xfs_bmap_free_item_t    *new;           /* new element */
4099         xfs_bmap_free_item_t    *prev;          /* previous element */
4100 #ifdef DEBUG
4101         xfs_agnumber_t          agno;
4102         xfs_agblock_t           agbno;
4103
4104         ASSERT(bno != NULLFSBLOCK);
4105         ASSERT(len > 0);
4106         ASSERT(len <= MAXEXTLEN);
4107         ASSERT(!ISNULLSTARTBLOCK(bno));
4108         agno = XFS_FSB_TO_AGNO(mp, bno);
4109         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4110         ASSERT(agno < mp->m_sb.sb_agcount);
4111         ASSERT(agbno < mp->m_sb.sb_agblocks);
4112         ASSERT(len < mp->m_sb.sb_agblocks);
4113         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4114 #endif
4115         ASSERT(xfs_bmap_free_item_zone != NULL);
4116         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4117         new->xbfi_startblock = bno;
4118         new->xbfi_blockcount = (xfs_extlen_t)len;
4119         for (prev = NULL, cur = flist->xbf_first;
4120              cur != NULL;
4121              prev = cur, cur = cur->xbfi_next) {
4122                 if (cur->xbfi_startblock >= bno)
4123                         break;
4124         }
4125         if (prev)
4126                 prev->xbfi_next = new;
4127         else
4128                 flist->xbf_first = new;
4129         new->xbfi_next = cur;
4130         flist->xbf_count++;
4131 }
4132
4133 /*
4134  * Compute and fill in the value of the maximum depth of a bmap btree
4135  * in this filesystem.  Done once, during mount.
4136  */
4137 void
4138 xfs_bmap_compute_maxlevels(
4139         xfs_mount_t     *mp,            /* file system mount structure */
4140         int             whichfork)      /* data or attr fork */
4141 {
4142         int             level;          /* btree level */
4143         uint            maxblocks;      /* max blocks at this level */
4144         uint            maxleafents;    /* max leaf entries possible */
4145         int             maxrootrecs;    /* max records in root block */
4146         int             minleafrecs;    /* min records in leaf block */
4147         int             minnoderecs;    /* min records in node block */
4148         int             sz;             /* root block size */
4149
4150         /*
4151          * The maximum number of extents in a file, hence the maximum
4152          * number of leaf entries, is controlled by the type of di_nextents
4153          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4154          * (a signed 16-bit number, xfs_aextnum_t).
4155          *
4156          * Note that we can no longer assume that if we are in ATTR1 that
4157          * the fork offset of all the inodes will be (m_attroffset >> 3)
4158          * because we could have mounted with ATTR2 and then mounted back
4159          * with ATTR1, keeping the di_forkoff's fixed but probably at
4160          * various positions. Therefore, for both ATTR1 and ATTR2
4161          * we have to assume the worst case scenario of a minimum size
4162          * available.
4163          */
4164         if (whichfork == XFS_DATA_FORK) {
4165                 maxleafents = MAXEXTNUM;
4166                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4167         } else {
4168                 maxleafents = MAXAEXTNUM;
4169                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
4170         }
4171         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4172         minleafrecs = mp->m_bmap_dmnr[0];
4173         minnoderecs = mp->m_bmap_dmnr[1];
4174         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4175         for (level = 1; maxblocks > 1; level++) {
4176                 if (maxblocks <= maxrootrecs)
4177                         maxblocks = 1;
4178                 else
4179                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4180         }
4181         mp->m_bm_maxlevels[whichfork] = level;
4182 }
4183
4184 /*
4185  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4186  * caller.  Frees all the extents that need freeing, which must be done
4187  * last due to locking considerations.  We never free any extents in
4188  * the first transaction.  This is to allow the caller to make the first
4189  * transaction a synchronous one so that the pointers to the data being
4190  * broken in this transaction will be permanent before the data is actually
4191  * freed.  This is necessary to prevent blocks from being reallocated
4192  * and written to before the free and reallocation are actually permanent.
4193  * We do not just make the first transaction synchronous here, because
4194  * there are more efficient ways to gain the same protection in some cases
4195  * (see the file truncation code).
4196  *
4197  * Return 1 if the given transaction was committed and a new one
4198  * started, and 0 otherwise in the committed parameter.
4199  */
4200 /*ARGSUSED*/
4201 int                                             /* error */
4202 xfs_bmap_finish(
4203         xfs_trans_t             **tp,           /* transaction pointer addr */
4204         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4205         int                     *committed)     /* xact committed or not */
4206 {
4207         xfs_efd_log_item_t      *efd;           /* extent free data */
4208         xfs_efi_log_item_t      *efi;           /* extent free intention */
4209         int                     error;          /* error return value */
4210         xfs_bmap_free_item_t    *free;          /* free extent item */
4211         unsigned int            logres;         /* new log reservation */
4212         unsigned int            logcount;       /* new log count */
4213         xfs_mount_t             *mp;            /* filesystem mount structure */
4214         xfs_bmap_free_item_t    *next;          /* next item on free list */
4215         xfs_trans_t             *ntp;           /* new transaction pointer */
4216
4217         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4218         if (flist->xbf_count == 0) {
4219                 *committed = 0;
4220                 return 0;
4221         }
4222         ntp = *tp;
4223         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4224         for (free = flist->xbf_first; free; free = free->xbfi_next)
4225                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4226                         free->xbfi_blockcount);
4227         logres = ntp->t_log_res;
4228         logcount = ntp->t_log_count;
4229         ntp = xfs_trans_dup(*tp);
4230         error = xfs_trans_commit(*tp, 0);
4231         *tp = ntp;
4232         *committed = 1;
4233         /*
4234          * We have a new transaction, so we should return committed=1,
4235          * even though we're returning an error.
4236          */
4237         if (error) {
4238                 return error;
4239         }
4240         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4241                         logcount)))
4242                 return error;
4243         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4244         for (free = flist->xbf_first; free != NULL; free = next) {
4245                 next = free->xbfi_next;
4246                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4247                                 free->xbfi_blockcount))) {
4248                         /*
4249                          * The bmap free list will be cleaned up at a
4250                          * higher level.  The EFI will be canceled when
4251                          * this transaction is aborted.
4252                          * Need to force shutdown here to make sure it
4253                          * happens, since this transaction may not be
4254                          * dirty yet.
4255                          */
4256                         mp = ntp->t_mountp;
4257                         if (!XFS_FORCED_SHUTDOWN(mp))
4258                                 xfs_force_shutdown(mp,
4259                                                    (error == EFSCORRUPTED) ?
4260                                                    SHUTDOWN_CORRUPT_INCORE :
4261                                                    SHUTDOWN_META_IO_ERROR);
4262                         return error;
4263                 }
4264                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4265                         free->xbfi_blockcount);
4266                 xfs_bmap_del_free(flist, NULL, free);
4267         }
4268         return 0;
4269 }
4270
4271 /*
4272  * Free up any items left in the list.
4273  */
4274 void
4275 xfs_bmap_cancel(
4276         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4277 {
4278         xfs_bmap_free_item_t    *free;  /* free list item */
4279         xfs_bmap_free_item_t    *next;
4280
4281         if (flist->xbf_count == 0)
4282                 return;
4283         ASSERT(flist->xbf_first != NULL);
4284         for (free = flist->xbf_first; free; free = next) {
4285                 next = free->xbfi_next;
4286                 xfs_bmap_del_free(flist, NULL, free);
4287         }
4288         ASSERT(flist->xbf_count == 0);
4289 }
4290
4291 /*
4292  * Returns the file-relative block number of the first unused block(s)
4293  * in the file with at least "len" logically contiguous blocks free.
4294  * This is the lowest-address hole if the file has holes, else the first block
4295  * past the end of file.
4296  * Return 0 if the file is currently local (in-inode).
4297  */
4298 int                                             /* error */
4299 xfs_bmap_first_unused(
4300         xfs_trans_t     *tp,                    /* transaction pointer */
4301         xfs_inode_t     *ip,                    /* incore inode */
4302         xfs_extlen_t    len,                    /* size of hole to find */
4303         xfs_fileoff_t   *first_unused,          /* unused block */
4304         int             whichfork)              /* data or attr fork */
4305 {
4306         int             error;                  /* error return value */
4307         int             idx;                    /* extent record index */
4308         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4309         xfs_fileoff_t   lastaddr;               /* last block number seen */
4310         xfs_fileoff_t   lowest;                 /* lowest useful block */
4311         xfs_fileoff_t   max;                    /* starting useful block */
4312         xfs_fileoff_t   off;                    /* offset for this block */
4313         xfs_extnum_t    nextents;               /* number of extent entries */
4314
4315         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4316                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4317                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4318         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4319                 *first_unused = 0;
4320                 return 0;
4321         }
4322         ifp = XFS_IFORK_PTR(ip, whichfork);
4323         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4324             (error = xfs_iread_extents(tp, ip, whichfork)))
4325                 return error;
4326         lowest = *first_unused;
4327         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4328         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4329                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
4330                 off = xfs_bmbt_get_startoff(ep);
4331                 /*
4332                  * See if the hole before this extent will work.
4333                  */
4334                 if (off >= lowest + len && off - max >= len) {
4335                         *first_unused = max;
4336                         return 0;
4337                 }
4338                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4339                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4340         }
4341         *first_unused = max;
4342         return 0;
4343 }
4344
4345 /*
4346  * Returns the file-relative block number of the last block + 1 before
4347  * last_block (input value) in the file.
4348  * This is not based on i_size, it is based on the extent records.
4349  * Returns 0 for local files, as they do not have extent records.
4350  */
4351 int                                             /* error */
4352 xfs_bmap_last_before(
4353         xfs_trans_t     *tp,                    /* transaction pointer */
4354         xfs_inode_t     *ip,                    /* incore inode */
4355         xfs_fileoff_t   *last_block,            /* last block */
4356         int             whichfork)              /* data or attr fork */
4357 {
4358         xfs_fileoff_t   bno;                    /* input file offset */
4359         int             eof;                    /* hit end of file */
4360         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4361         int             error;                  /* error return value */
4362         xfs_bmbt_irec_t got;                    /* current extent value */
4363         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4364         xfs_extnum_t    lastx;                  /* last extent used */
4365         xfs_bmbt_irec_t prev;                   /* previous extent value */
4366
4367         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4368             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4369             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4370                return XFS_ERROR(EIO);
4371         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4372                 *last_block = 0;
4373                 return 0;
4374         }
4375         ifp = XFS_IFORK_PTR(ip, whichfork);
4376         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4377             (error = xfs_iread_extents(tp, ip, whichfork)))
4378                 return error;
4379         bno = *last_block - 1;
4380         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4381                 &prev);
4382         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4383                 if (prev.br_startoff == NULLFILEOFF)
4384                         *last_block = 0;
4385                 else
4386                         *last_block = prev.br_startoff + prev.br_blockcount;
4387         }
4388         /*
4389          * Otherwise *last_block is already the right answer.
4390          */
4391         return 0;
4392 }
4393
4394 /*
4395  * Returns the file-relative block number of the first block past eof in
4396  * the file.  This is not based on i_size, it is based on the extent records.
4397  * Returns 0 for local files, as they do not have extent records.
4398  */
4399 int                                             /* error */
4400 xfs_bmap_last_offset(
4401         xfs_trans_t     *tp,                    /* transaction pointer */
4402         xfs_inode_t     *ip,                    /* incore inode */
4403         xfs_fileoff_t   *last_block,            /* last block */
4404         int             whichfork)              /* data or attr fork */
4405 {
4406         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4407         int             error;                  /* error return value */
4408         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4409         xfs_extnum_t    nextents;               /* number of extent entries */
4410
4411         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4412             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4413             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4414                return XFS_ERROR(EIO);
4415         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4416                 *last_block = 0;
4417                 return 0;
4418         }
4419         ifp = XFS_IFORK_PTR(ip, whichfork);
4420         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4421             (error = xfs_iread_extents(tp, ip, whichfork)))
4422                 return error;
4423         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4424         if (!nextents) {
4425                 *last_block = 0;
4426                 return 0;
4427         }
4428         ep = xfs_iext_get_ext(ifp, nextents - 1);
4429         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4430         return 0;
4431 }
4432
4433 /*
4434  * Returns whether the selected fork of the inode has exactly one
4435  * block or not.  For the data fork we check this matches di_size,
4436  * implying the file's range is 0..bsize-1.
4437  */
4438 int                                     /* 1=>1 block, 0=>otherwise */
4439 xfs_bmap_one_block(
4440         xfs_inode_t     *ip,            /* incore inode */
4441         int             whichfork)      /* data or attr fork */
4442 {
4443         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4444         xfs_ifork_t     *ifp;           /* inode fork pointer */
4445         int             rval;           /* return value */
4446         xfs_bmbt_irec_t s;              /* internal version of extent */
4447
4448 #ifndef DEBUG
4449         if (whichfork == XFS_DATA_FORK) {
4450                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4451                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4452                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4453         }
4454 #endif  /* !DEBUG */
4455         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4456                 return 0;
4457         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4458                 return 0;
4459         ifp = XFS_IFORK_PTR(ip, whichfork);
4460         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4461         ep = xfs_iext_get_ext(ifp, 0);
4462         xfs_bmbt_get_all(ep, &s);
4463         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4464         if (rval && whichfork == XFS_DATA_FORK)
4465                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4466         return rval;
4467 }
4468
4469 /*
4470  * Read in the extents to if_extents.
4471  * All inode fields are set up by caller, we just traverse the btree
4472  * and copy the records in. If the file system cannot contain unwritten
4473  * extents, the records are checked for no "state" flags.
4474  */
4475 int                                     /* error */
4476 xfs_bmap_read_extents(
4477         xfs_trans_t             *tp,    /* transaction pointer */
4478         xfs_inode_t             *ip,    /* incore inode */
4479         int                     whichfork) /* data or attr fork */
4480 {
4481         xfs_bmbt_block_t        *block; /* current btree block */
4482         xfs_fsblock_t           bno;    /* block # of "block" */
4483         xfs_buf_t               *bp;    /* buffer for "block" */
4484         int                     error;  /* error return value */
4485         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4486         xfs_extnum_t            i, j;   /* index into the extents list */
4487         xfs_ifork_t             *ifp;   /* fork structure */
4488         int                     level;  /* btree level, for checking */
4489         xfs_mount_t             *mp;    /* file system mount structure */
4490         __be64                  *pp;    /* pointer to block address */
4491         /* REFERENCED */
4492         xfs_extnum_t            room;   /* number of entries there's room for */
4493
4494         bno = NULLFSBLOCK;
4495         mp = ip->i_mount;
4496         ifp = XFS_IFORK_PTR(ip, whichfork);
4497         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4498                                         XFS_EXTFMT_INODE(ip);
4499         block = ifp->if_broot;
4500         /*
4501          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4502          */
4503         level = be16_to_cpu(block->bb_level);
4504         ASSERT(level > 0);
4505         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4506         bno = be64_to_cpu(*pp);
4507         ASSERT(bno != NULLDFSBNO);
4508         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4509         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4510         /*
4511          * Go down the tree until leaf level is reached, following the first
4512          * pointer (leftmost) at each level.
4513          */
4514         while (level-- > 0) {
4515                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4516                                 XFS_BMAP_BTREE_REF)))
4517                         return error;
4518                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4519                 XFS_WANT_CORRUPTED_GOTO(
4520                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4521                         error0);
4522                 if (level == 0)
4523                         break;
4524                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4525                 bno = be64_to_cpu(*pp);
4526                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4527                 xfs_trans_brelse(tp, bp);
4528         }
4529         /*
4530          * Here with bp and block set to the leftmost leaf node in the tree.
4531          */
4532         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4533         i = 0;
4534         /*
4535          * Loop over all leaf nodes.  Copy information to the extent records.
4536          */
4537         for (;;) {
4538                 xfs_bmbt_rec_t  *frp;
4539                 xfs_fsblock_t   nextbno;
4540                 xfs_extnum_t    num_recs;
4541                 xfs_extnum_t    start;
4542
4543
4544                 num_recs = be16_to_cpu(block->bb_numrecs);
4545                 if (unlikely(i + num_recs > room)) {
4546                         ASSERT(i + num_recs <= room);
4547                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4548                                 "corrupt dinode %Lu, (btree extents).",
4549                                 (unsigned long long) ip->i_ino);
4550                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4551                                          XFS_ERRLEVEL_LOW,
4552                                         ip->i_mount);
4553                         goto error0;
4554                 }
4555                 XFS_WANT_CORRUPTED_GOTO(
4556                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4557                         error0);
4558                 /*
4559                  * Read-ahead the next leaf block, if any.
4560                  */
4561                 nextbno = be64_to_cpu(block->bb_rightsib);
4562                 if (nextbno != NULLFSBLOCK)
4563                         xfs_btree_reada_bufl(mp, nextbno, 1);
4564                 /*
4565                  * Copy records into the extent records.
4566                  */
4567                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4568                 start = i;
4569                 for (j = 0; j < num_recs; j++, i++, frp++) {
4570                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4571                         trp->l0 = be64_to_cpu(frp->l0);
4572                         trp->l1 = be64_to_cpu(frp->l1);
4573                 }
4574                 if (exntf == XFS_EXTFMT_NOSTATE) {
4575                         /*
4576                          * Check all attribute bmap btree records and
4577                          * any "older" data bmap btree records for a
4578                          * set bit in the "extent flag" position.
4579                          */
4580                         if (unlikely(xfs_check_nostate_extents(ifp,
4581                                         start, num_recs))) {
4582                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4583                                                  XFS_ERRLEVEL_LOW,
4584                                                  ip->i_mount);
4585                                 goto error0;
4586                         }
4587                 }
4588                 xfs_trans_brelse(tp, bp);
4589                 bno = nextbno;
4590                 /*
4591                  * If we've reached the end, stop.
4592                  */
4593                 if (bno == NULLFSBLOCK)
4594                         break;
4595                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4596                                 XFS_BMAP_BTREE_REF)))
4597                         return error;
4598                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4599         }
4600         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4601         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4602         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4603         return 0;
4604 error0:
4605         xfs_trans_brelse(tp, bp);
4606         return XFS_ERROR(EFSCORRUPTED);
4607 }
4608
4609 #ifdef XFS_BMAP_TRACE
4610 /*
4611  * Add bmap trace insert entries for all the contents of the extent records.
4612  */
4613 void
4614 xfs_bmap_trace_exlist(
4615         const char      *fname,         /* function name */
4616         xfs_inode_t     *ip,            /* incore inode pointer */
4617         xfs_extnum_t    cnt,            /* count of entries in the list */
4618         int             whichfork)      /* data or attr fork */
4619 {
4620         xfs_bmbt_rec_host_t *ep;        /* current extent record */
4621         xfs_extnum_t    idx;            /* extent record index */
4622         xfs_ifork_t     *ifp;           /* inode fork pointer */
4623         xfs_bmbt_irec_t s;              /* file extent record */
4624
4625         ifp = XFS_IFORK_PTR(ip, whichfork);
4626         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4627         for (idx = 0; idx < cnt; idx++) {
4628                 ep = xfs_iext_get_ext(ifp, idx);
4629                 xfs_bmbt_get_all(ep, &s);
4630                 XFS_BMAP_TRACE_INSERT("exlist", ip, idx, 1, &s, NULL,
4631                         whichfork);
4632         }
4633 }
4634 #endif
4635
4636 #ifdef DEBUG
4637 /*
4638  * Validate that the bmbt_irecs being returned from bmapi are valid
4639  * given the callers original parameters.  Specifically check the
4640  * ranges of the returned irecs to ensure that they only extent beyond
4641  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4642  */
4643 STATIC void
4644 xfs_bmap_validate_ret(
4645         xfs_fileoff_t           bno,
4646         xfs_filblks_t           len,
4647         int                     flags,
4648         xfs_bmbt_irec_t         *mval,
4649         int                     nmap,
4650         int                     ret_nmap)
4651 {
4652         int                     i;              /* index to map values */
4653
4654         ASSERT(ret_nmap <= nmap);
4655
4656         for (i = 0; i < ret_nmap; i++) {
4657                 ASSERT(mval[i].br_blockcount > 0);
4658                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4659                         ASSERT(mval[i].br_startoff >= bno);
4660                         ASSERT(mval[i].br_blockcount <= len);
4661                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4662                                bno + len);
4663                 } else {
4664                         ASSERT(mval[i].br_startoff < bno + len);
4665                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4666                                bno);
4667                 }
4668                 ASSERT(i == 0 ||
4669                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4670                        mval[i].br_startoff);
4671                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4672                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4673                                mval[i].br_startblock != HOLESTARTBLOCK);
4674                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4675                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4676         }
4677 }
4678 #endif /* DEBUG */
4679
4680
4681 /*
4682  * Map file blocks to filesystem blocks.
4683  * File range is given by the bno/len pair.
4684  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4685  * into a hole or past eof.
4686  * Only allocates blocks from a single allocation group,
4687  * to avoid locking problems.
4688  * The returned value in "firstblock" from the first call in a transaction
4689  * must be remembered and presented to subsequent calls in "firstblock".
4690  * An upper bound for the number of blocks to be allocated is supplied to
4691  * the first call in "total"; if no allocation group has that many free
4692  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4693  */
4694 int                                     /* error */
4695 xfs_bmapi(
4696         xfs_trans_t     *tp,            /* transaction pointer */
4697         xfs_inode_t     *ip,            /* incore inode */
4698         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4699         xfs_filblks_t   len,            /* length to map in file */
4700         int             flags,          /* XFS_BMAPI_... */
4701         xfs_fsblock_t   *firstblock,    /* first allocated block
4702                                            controls a.g. for allocs */
4703         xfs_extlen_t    total,          /* total blocks needed */
4704         xfs_bmbt_irec_t *mval,          /* output: map values */
4705         int             *nmap,          /* i/o: mval size/count */
4706         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4707         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4708 {
4709         xfs_fsblock_t   abno;           /* allocated block number */
4710         xfs_extlen_t    alen;           /* allocated extent length */
4711         xfs_fileoff_t   aoff;           /* allocated file offset */
4712         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4713         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4714         xfs_fileoff_t   end;            /* end of mapped file region */
4715         int             eof;            /* we've hit the end of extents */
4716         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4717         int             error;          /* error return */
4718         xfs_bmbt_irec_t got;            /* current file extent record */
4719         xfs_ifork_t     *ifp;           /* inode fork pointer */
4720         xfs_extlen_t    indlen;         /* indirect blocks length */
4721         xfs_extnum_t    lastx;          /* last useful extent number */
4722         int             logflags;       /* flags for transaction logging */
4723         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4724         xfs_extlen_t    minlen;         /* min allocation size */
4725         xfs_mount_t     *mp;            /* xfs mount structure */
4726         int             n;              /* current extent index */
4727         int             nallocs;        /* number of extents alloc\'d */
4728         xfs_extnum_t    nextents;       /* number of extents in file */
4729         xfs_fileoff_t   obno;           /* old block number (offset) */
4730         xfs_bmbt_irec_t prev;           /* previous file extent record */
4731         int             tmp_logflags;   /* temp flags holder */
4732         int             whichfork;      /* data or attr fork */
4733         char            inhole;         /* current location is hole in file */
4734         char            wasdelay;       /* old extent was delayed */
4735         char            wr;             /* this is a write request */
4736         char            rt;             /* this is a realtime file */
4737 #ifdef DEBUG
4738         xfs_fileoff_t   orig_bno;       /* original block number value */
4739         int             orig_flags;     /* original flags arg value */
4740         xfs_filblks_t   orig_len;       /* original value of len arg */
4741         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4742         int             orig_nmap;      /* original value of *nmap */
4743
4744         orig_bno = bno;
4745         orig_len = len;
4746         orig_flags = flags;
4747         orig_mval = mval;
4748         orig_nmap = *nmap;
4749 #endif
4750         ASSERT(*nmap >= 1);
4751         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4752         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4753                 XFS_ATTR_FORK : XFS_DATA_FORK;
4754         mp = ip->i_mount;
4755         if (unlikely(XFS_TEST_ERROR(
4756             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4757              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4758              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4759              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4760                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4761                 return XFS_ERROR(EFSCORRUPTED);
4762         }
4763         if (XFS_FORCED_SHUTDOWN(mp))
4764                 return XFS_ERROR(EIO);
4765         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4766         ifp = XFS_IFORK_PTR(ip, whichfork);
4767         ASSERT(ifp->if_ext_max ==
4768                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4769         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4770                 XFS_STATS_INC(xs_blk_mapw);
4771         else
4772                 XFS_STATS_INC(xs_blk_mapr);
4773         /*
4774          * IGSTATE flag is used to combine extents which
4775          * differ only due to the state of the extents.
4776          * This technique is used from xfs_getbmap()
4777          * when the caller does not wish to see the
4778          * separation (which is the default).
4779          *
4780          * This technique is also used when writing a
4781          * buffer which has been partially written,
4782          * (usually by being flushed during a chunkread),
4783          * to ensure one write takes place. This also
4784          * prevents a change in the xfs inode extents at
4785          * this time, intentionally. This change occurs
4786          * on completion of the write operation, in
4787          * xfs_strat_comp(), where the xfs_bmapi() call
4788          * is transactioned, and the extents combined.
4789          */
4790         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4791                 wr = 0;                         /* no allocations are allowed */
4792         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4793         logflags = 0;
4794         nallocs = 0;
4795         cur = NULL;
4796         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4797                 ASSERT(wr && tp);
4798                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4799                                 firstblock, total, &logflags, whichfork)))
4800                         goto error0;
4801         }
4802         if (wr && *firstblock == NULLFSBLOCK) {
4803                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4804                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4805                 else
4806                         minleft = 1;
4807         } else
4808                 minleft = 0;
4809         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4810             (error = xfs_iread_extents(tp, ip, whichfork)))
4811                 goto error0;
4812         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4813                 &prev);
4814         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4815         n = 0;
4816         end = bno + len;
4817         obno = bno;
4818         bma.ip = NULL;
4819         if (delta) {
4820                 delta->xed_startoff = NULLFILEOFF;
4821                 delta->xed_blockcount = 0;
4822         }
4823         while (bno < end && n < *nmap) {
4824                 /*
4825                  * Reading past eof, act as though there's a hole
4826                  * up to end.
4827                  */
4828                 if (eof && !wr)
4829                         got.br_startoff = end;
4830                 inhole = eof || got.br_startoff > bno;
4831                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4832                         ISNULLSTARTBLOCK(got.br_startblock);
4833                 /*
4834                  * First, deal with the hole before the allocated space
4835                  * that we found, if any.
4836                  */
4837                 if (wr && (inhole || wasdelay)) {
4838                         /*
4839                          * For the wasdelay case, we could also just
4840                          * allocate the stuff asked for in this bmap call
4841                          * but that wouldn't be as good.
4842                          */
4843                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4844                                 alen = (xfs_extlen_t)got.br_blockcount;
4845                                 aoff = got.br_startoff;
4846                                 if (lastx != NULLEXTNUM && lastx) {
4847                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4848                                         xfs_bmbt_get_all(ep, &prev);
4849                                 }
4850                         } else if (wasdelay) {
4851                                 alen = (xfs_extlen_t)
4852                                         XFS_FILBLKS_MIN(len,
4853                                                 (got.br_startoff +
4854                                                  got.br_blockcount) - bno);
4855                                 aoff = bno;
4856                         } else {
4857                                 alen = (xfs_extlen_t)
4858                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4859                                 if (!eof)
4860                                         alen = (xfs_extlen_t)
4861                                                 XFS_FILBLKS_MIN(alen,
4862                                                         got.br_startoff - bno);
4863                                 aoff = bno;
4864                         }
4865                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4866                         if (flags & XFS_BMAPI_DELAY) {
4867                                 xfs_extlen_t    extsz;
4868
4869                                 /* Figure out the extent size, adjust alen */
4870                                 extsz = xfs_get_extsz_hint(ip);
4871                                 if (extsz) {
4872                                         error = xfs_bmap_extsize_align(mp,
4873                                                         &got, &prev, extsz,
4874                                                         rt, eof,
4875                                                         flags&XFS_BMAPI_DELAY,
4876                                                         flags&XFS_BMAPI_CONVERT,
4877                                                         &aoff, &alen);
4878                                         ASSERT(!error);
4879                                 }
4880
4881                                 if (rt)
4882                                         extsz = alen / mp->m_sb.sb_rextsize;
4883
4884                                 /*
4885                                  * Make a transaction-less quota reservation for
4886                                  * delayed allocation blocks. This number gets
4887                                  * adjusted later.  We return if we haven't
4888                                  * allocated blocks already inside this loop.
4889                                  */
4890                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4891                                                 mp, NULL, ip, (long)alen, 0,
4892                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4893                                                      XFS_QMOPT_RES_REGBLKS))) {
4894                                         if (n == 0) {
4895                                                 *nmap = 0;
4896                                                 ASSERT(cur == NULL);
4897                                                 return error;
4898                                         }
4899                                         break;
4900                                 }
4901
4902                                 /*
4903                                  * Split changing sb for alen and indlen since
4904                                  * they could be coming from different places.
4905                                  */
4906                                 indlen = (xfs_extlen_t)
4907                                         xfs_bmap_worst_indlen(ip, alen);
4908                                 ASSERT(indlen > 0);
4909
4910                                 if (rt) {
4911                                         error = xfs_mod_incore_sb(mp,
4912                                                         XFS_SBS_FREXTENTS,
4913                                                         -((int64_t)extsz), (flags &
4914                                                         XFS_BMAPI_RSVBLOCKS));
4915                                 } else {
4916                                         error = xfs_mod_incore_sb(mp,
4917                                                         XFS_SBS_FDBLOCKS,
4918                                                         -((int64_t)alen), (flags &
4919                                                         XFS_BMAPI_RSVBLOCKS));
4920                                 }
4921                                 if (!error) {
4922                                         error = xfs_mod_incore_sb(mp,
4923                                                         XFS_SBS_FDBLOCKS,
4924                                                         -((int64_t)indlen), (flags &
4925                                                         XFS_BMAPI_RSVBLOCKS));
4926                                         if (error && rt)
4927                                                 xfs_mod_incore_sb(mp,
4928                                                         XFS_SBS_FREXTENTS,
4929                                                         (int64_t)extsz, (flags &
4930                                                         XFS_BMAPI_RSVBLOCKS));
4931                                         else if (error)
4932                                                 xfs_mod_incore_sb(mp,
4933                                                         XFS_SBS_FDBLOCKS,
4934                                                         (int64_t)alen, (flags &
4935                                                         XFS_BMAPI_RSVBLOCKS));
4936                                 }
4937
4938                                 if (error) {
4939                                         if (XFS_IS_QUOTA_ON(mp))
4940                                                 /* unreserve the blocks now */
4941                                                 (void)
4942                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4943                                                         mp, NULL, ip,
4944                                                         (long)alen, 0, rt ?
4945                                                         XFS_QMOPT_RES_RTBLKS :
4946                                                         XFS_QMOPT_RES_REGBLKS);
4947                                         break;
4948                                 }
4949
4950                                 ip->i_delayed_blks += alen;
4951                                 abno = NULLSTARTBLOCK(indlen);
4952                         } else {
4953                                 /*
4954                                  * If first time, allocate and fill in
4955                                  * once-only bma fields.
4956                                  */
4957                                 if (bma.ip == NULL) {
4958                                         bma.tp = tp;
4959                                         bma.ip = ip;
4960                                         bma.prevp = &prev;
4961                                         bma.gotp = &got;
4962                                         bma.total = total;
4963                                         bma.userdata = 0;
4964                                 }
4965                                 /* Indicate if this is the first user data
4966                                  * in the file, or just any user data.
4967                                  */
4968                                 if (!(flags & XFS_BMAPI_METADATA)) {
4969                                         bma.userdata = (aoff == 0) ?
4970                                                 XFS_ALLOC_INITIAL_USER_DATA :
4971                                                 XFS_ALLOC_USERDATA;
4972                                 }
4973                                 /*
4974                                  * Fill in changeable bma fields.
4975                                  */
4976                                 bma.eof = eof;
4977                                 bma.firstblock = *firstblock;
4978                                 bma.alen = alen;
4979                                 bma.off = aoff;
4980                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4981                                 bma.wasdel = wasdelay;
4982                                 bma.minlen = minlen;
4983                                 bma.low = flist->xbf_low;
4984                                 bma.minleft = minleft;
4985                                 /*
4986                                  * Only want to do the alignment at the
4987                                  * eof if it is userdata and allocation length
4988                                  * is larger than a stripe unit.
4989                                  */
4990                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4991                                     (!(flags & XFS_BMAPI_METADATA)) &&
4992                                     (whichfork == XFS_DATA_FORK)) {
4993                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4994                                                         whichfork, &bma.aeof)))
4995                                                 goto error0;
4996                                 } else
4997                                         bma.aeof = 0;
4998                                 /*
4999                                  * Call allocator.
5000                                  */
5001                                 if ((error = xfs_bmap_alloc(&bma)))
5002                                         goto error0;
5003                                 /*
5004                                  * Copy out result fields.
5005                                  */
5006                                 abno = bma.rval;
5007                                 if ((flist->xbf_low = bma.low))
5008                                         minleft = 0;
5009                                 alen = bma.alen;
5010                                 aoff = bma.off;
5011                                 ASSERT(*firstblock == NULLFSBLOCK ||
5012                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5013                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5014                                        (flist->xbf_low &&
5015                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5016                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5017                                 *firstblock = bma.firstblock;
5018                                 if (cur)
5019                                         cur->bc_private.b.firstblock =
5020                                                 *firstblock;
5021                                 if (abno == NULLFSBLOCK)
5022                                         break;
5023                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5024                                         cur = xfs_btree_init_cursor(mp,
5025                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
5026                                                 ip, whichfork);
5027                                         cur->bc_private.b.firstblock =
5028                                                 *firstblock;
5029                                         cur->bc_private.b.flist = flist;
5030                                 }
5031                                 /*
5032                                  * Bump the number of extents we've allocated
5033                                  * in this call.
5034                                  */
5035                                 nallocs++;
5036                         }
5037                         if (cur)
5038                                 cur->bc_private.b.flags =
5039                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5040                         got.br_startoff = aoff;
5041                         got.br_startblock = abno;
5042                         got.br_blockcount = alen;
5043                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5044                         /*
5045                          * Determine state of extent, and the filesystem.
5046                          * A wasdelay extent has been initialized, so
5047                          * shouldn't be flagged as unwritten.
5048                          */
5049                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5050                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5051                                         got.br_state = XFS_EXT_UNWRITTEN;
5052                         }
5053                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5054                                 firstblock, flist, &tmp_logflags, delta,
5055                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5056                         logflags |= tmp_logflags;
5057                         if (error)
5058                                 goto error0;
5059                         lastx = ifp->if_lastex;
5060                         ep = xfs_iext_get_ext(ifp, lastx);
5061                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5062                         xfs_bmbt_get_all(ep, &got);
5063                         ASSERT(got.br_startoff <= aoff);
5064                         ASSERT(got.br_startoff + got.br_blockcount >=
5065                                 aoff + alen);
5066 #ifdef DEBUG
5067                         if (flags & XFS_BMAPI_DELAY) {
5068                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5069                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5070                         }
5071                         ASSERT(got.br_state == XFS_EXT_NORM ||
5072                                got.br_state == XFS_EXT_UNWRITTEN);
5073 #endif
5074                         /*
5075                          * Fall down into the found allocated space case.
5076                          */
5077                 } else if (inhole) {
5078                         /*
5079                          * Reading in a hole.
5080                          */
5081                         mval->br_startoff = bno;
5082                         mval->br_startblock = HOLESTARTBLOCK;
5083                         mval->br_blockcount =
5084                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5085                         mval->br_state = XFS_EXT_NORM;
5086                         bno += mval->br_blockcount;
5087                         len -= mval->br_blockcount;
5088                         mval++;
5089                         n++;
5090                         continue;
5091                 }
5092                 /*
5093                  * Then deal with the allocated space we found.
5094                  */
5095                 ASSERT(ep != NULL);
5096                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5097                     (got.br_startoff + got.br_blockcount > obno)) {
5098                         if (obno > bno)
5099                                 bno = obno;
5100                         ASSERT((bno >= obno) || (n == 0));
5101                         ASSERT(bno < end);
5102                         mval->br_startoff = bno;
5103                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5104                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5105                                 mval->br_startblock = DELAYSTARTBLOCK;
5106                         } else
5107                                 mval->br_startblock =
5108                                         got.br_startblock +
5109                                         (bno - got.br_startoff);
5110                         /*
5111                          * Return the minimum of what we got and what we
5112                          * asked for for the length.  We can use the len
5113                          * variable here because it is modified below
5114                          * and we could have been there before coming
5115                          * here if the first part of the allocation
5116                          * didn't overlap what was asked for.
5117                          */
5118                         mval->br_blockcount =
5119                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5120                                         (bno - got.br_startoff));
5121                         mval->br_state = got.br_state;
5122                         ASSERT(mval->br_blockcount <= len);
5123                 } else {
5124                         *mval = got;
5125                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5126                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5127                                 mval->br_startblock = DELAYSTARTBLOCK;
5128                         }
5129                 }
5130
5131                 /*
5132                  * Check if writing previously allocated but
5133                  * unwritten extents.
5134                  */
5135                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5136                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5137                         /*
5138                          * Modify (by adding) the state flag, if writing.
5139                          */
5140                         ASSERT(mval->br_blockcount <= len);
5141                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5142                                 cur = xfs_btree_init_cursor(mp,
5143                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5144                                         ip, whichfork);
5145                                 cur->bc_private.b.firstblock =
5146                                         *firstblock;
5147                                 cur->bc_private.b.flist = flist;
5148                         }
5149                         mval->br_state = XFS_EXT_NORM;
5150                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5151                                 firstblock, flist, &tmp_logflags, delta,
5152                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5153                         logflags |= tmp_logflags;
5154                         if (error)
5155                                 goto error0;
5156                         lastx = ifp->if_lastex;
5157                         ep = xfs_iext_get_ext(ifp, lastx);
5158                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5159                         xfs_bmbt_get_all(ep, &got);
5160                         /*
5161                          * We may have combined previously unwritten
5162                          * space with written space, so generate
5163                          * another request.
5164                          */
5165                         if (mval->br_blockcount < len)
5166                                 continue;
5167                 }
5168
5169                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5170                        ((mval->br_startoff + mval->br_blockcount) <= end));
5171                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5172                        (mval->br_blockcount <= len) ||
5173                        (mval->br_startoff < obno));
5174                 bno = mval->br_startoff + mval->br_blockcount;
5175                 len = end - bno;
5176                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5177                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5178                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5179                         ASSERT(mval->br_state == mval[-1].br_state);
5180                         mval[-1].br_blockcount = mval->br_blockcount;
5181                         mval[-1].br_state = mval->br_state;
5182                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5183                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5184                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5185                            mval->br_startblock ==
5186                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5187                            ((flags & XFS_BMAPI_IGSTATE) ||
5188                                 mval[-1].br_state == mval->br_state)) {
5189                         ASSERT(mval->br_startoff ==
5190                                mval[-1].br_startoff + mval[-1].br_blockcount);
5191                         mval[-1].br_blockcount += mval->br_blockcount;
5192                 } else if (n > 0 &&
5193                            mval->br_startblock == DELAYSTARTBLOCK &&
5194                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5195                            mval->br_startoff ==
5196                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5197                         mval[-1].br_blockcount += mval->br_blockcount;
5198                         mval[-1].br_state = mval->br_state;
5199                 } else if (!((n == 0) &&
5200                              ((mval->br_startoff + mval->br_blockcount) <=
5201                               obno))) {
5202                         mval++;
5203                         n++;
5204                 }
5205                 /*
5206                  * If we're done, stop now.  Stop when we've allocated
5207                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5208                  * the transaction may get too big.
5209                  */
5210                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5211                         break;
5212                 /*
5213                  * Else go on to the next record.
5214                  */
5215                 ep = xfs_iext_get_ext(ifp, ++lastx);
5216                 prev = got;
5217                 if (lastx >= nextents)
5218                         eof = 1;
5219                 else
5220                         xfs_bmbt_get_all(ep, &got);
5221         }
5222         ifp->if_lastex = lastx;
5223         *nmap = n;
5224         /*
5225          * Transform from btree to extents, give it cur.
5226          */
5227         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5228             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5229                 ASSERT(wr && cur);
5230                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5231                         &tmp_logflags, whichfork);
5232                 logflags |= tmp_logflags;
5233                 if (error)
5234                         goto error0;
5235         }
5236         ASSERT(ifp->if_ext_max ==
5237                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5238         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5239                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5240         error = 0;
5241         if (delta && delta->xed_startoff != NULLFILEOFF) {
5242                 /* A change was actually made.
5243                  * Note that delta->xed_blockount is an offset at this
5244                  * point and needs to be converted to a block count.
5245                  */
5246                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5247                 delta->xed_blockcount -= delta->xed_startoff;
5248         }
5249 error0:
5250         /*
5251          * Log everything.  Do this after conversion, there's no point in
5252          * logging the extent records if we've converted to btree format.
5253          */
5254         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5255             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5256                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5257         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5258                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5259                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5260         /*
5261          * Log whatever the flags say, even if error.  Otherwise we might miss
5262          * detecting a case where the data is changed, there's an error,
5263          * and it's not logged so we don't shutdown when we should.
5264          */
5265         if (logflags) {
5266                 ASSERT(tp && wr);
5267                 xfs_trans_log_inode(tp, ip, logflags);
5268         }
5269         if (cur) {
5270                 if (!error) {
5271                         ASSERT(*firstblock == NULLFSBLOCK ||
5272                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5273                                XFS_FSB_TO_AGNO(mp,
5274                                        cur->bc_private.b.firstblock) ||
5275                                (flist->xbf_low &&
5276                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5277                                 XFS_FSB_TO_AGNO(mp,
5278                                         cur->bc_private.b.firstblock)));
5279                         *firstblock = cur->bc_private.b.firstblock;
5280                 }
5281                 xfs_btree_del_cursor(cur,
5282                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5283         }
5284         if (!error)
5285                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5286                         orig_nmap, *nmap);
5287         return error;
5288 }
5289
5290 /*
5291  * Map file blocks to filesystem blocks, simple version.
5292  * One block (extent) only, read-only.
5293  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5294  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5295  * was set and all the others were clear.
5296  */
5297 int                                             /* error */
5298 xfs_bmapi_single(
5299         xfs_trans_t     *tp,            /* transaction pointer */
5300         xfs_inode_t     *ip,            /* incore inode */
5301         int             whichfork,      /* data or attr fork */
5302         xfs_fsblock_t   *fsb,           /* output: mapped block */
5303         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5304 {
5305         int             eof;            /* we've hit the end of extents */
5306         int             error;          /* error return */
5307         xfs_bmbt_irec_t got;            /* current file extent record */
5308         xfs_ifork_t     *ifp;           /* inode fork pointer */
5309         xfs_extnum_t    lastx;          /* last useful extent number */
5310         xfs_bmbt_irec_t prev;           /* previous file extent record */
5311
5312         ifp = XFS_IFORK_PTR(ip, whichfork);
5313         if (unlikely(
5314             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5315             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5316                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5317                                 ip->i_mount);
5318                return XFS_ERROR(EFSCORRUPTED);
5319         }
5320         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5321                 return XFS_ERROR(EIO);
5322         XFS_STATS_INC(xs_blk_mapr);
5323         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5324             (error = xfs_iread_extents(tp, ip, whichfork)))
5325                 return error;
5326         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5327                 &prev);
5328         /*
5329          * Reading past eof, act as though there's a hole
5330          * up to end.
5331          */
5332         if (eof || got.br_startoff > bno) {
5333                 *fsb = NULLFSBLOCK;
5334                 return 0;
5335         }
5336         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5337         ASSERT(bno < got.br_startoff + got.br_blockcount);
5338         *fsb = got.br_startblock + (bno - got.br_startoff);
5339         ifp->if_lastex = lastx;
5340         return 0;
5341 }
5342
5343 /*
5344  * Unmap (remove) blocks from a file.
5345  * If nexts is nonzero then the number of extents to remove is limited to
5346  * that value.  If not all extents in the block range can be removed then
5347  * *done is set.
5348  */
5349 int                                             /* error */
5350 xfs_bunmapi(
5351         xfs_trans_t             *tp,            /* transaction pointer */
5352         struct xfs_inode        *ip,            /* incore inode */
5353         xfs_fileoff_t           bno,            /* starting offset to unmap */
5354         xfs_filblks_t           len,            /* length to unmap in file */
5355         int                     flags,          /* misc flags */
5356         xfs_extnum_t            nexts,          /* number of extents max */
5357         xfs_fsblock_t           *firstblock,    /* first allocated block
5358                                                    controls a.g. for allocs */
5359         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5360         xfs_extdelta_t          *delta,         /* o: change made to incore
5361                                                    extents */
5362         int                     *done)          /* set if not done yet */
5363 {
5364         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5365         xfs_bmbt_irec_t         del;            /* extent being deleted */
5366         int                     eof;            /* is deleting at eof */
5367         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5368         int                     error;          /* error return value */
5369         xfs_extnum_t            extno;          /* extent number in list */
5370         xfs_bmbt_irec_t         got;            /* current extent record */
5371         xfs_ifork_t             *ifp;           /* inode fork pointer */
5372         int                     isrt;           /* freeing in rt area */
5373         xfs_extnum_t            lastx;          /* last extent index used */
5374         int                     logflags;       /* transaction logging flags */
5375         xfs_extlen_t            mod;            /* rt extent offset */
5376         xfs_mount_t             *mp;            /* mount structure */
5377         xfs_extnum_t            nextents;       /* number of file extents */
5378         xfs_bmbt_irec_t         prev;           /* previous extent record */
5379         xfs_fileoff_t           start;          /* first file offset deleted */
5380         int                     tmp_logflags;   /* partial logging flags */
5381         int                     wasdel;         /* was a delayed alloc extent */
5382         int                     whichfork;      /* data or attribute fork */
5383         int                     rsvd;           /* OK to allocate reserved blocks */
5384         xfs_fsblock_t           sum;
5385
5386         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5387         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5388                 XFS_ATTR_FORK : XFS_DATA_FORK;
5389         ifp = XFS_IFORK_PTR(ip, whichfork);
5390         if (unlikely(
5391             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5392             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5393                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5394                                  ip->i_mount);
5395                 return XFS_ERROR(EFSCORRUPTED);
5396         }
5397         mp = ip->i_mount;
5398         if (XFS_FORCED_SHUTDOWN(mp))
5399                 return XFS_ERROR(EIO);
5400         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5401         ASSERT(len > 0);
5402         ASSERT(nexts >= 0);
5403         ASSERT(ifp->if_ext_max ==
5404                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5405         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5406             (error = xfs_iread_extents(tp, ip, whichfork)))
5407                 return error;
5408         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5409         if (nextents == 0) {
5410                 *done = 1;
5411                 return 0;
5412         }
5413         XFS_STATS_INC(xs_blk_unmap);
5414         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5415         start = bno;
5416         bno = start + len - 1;
5417         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5418                 &prev);
5419         if (delta) {
5420                 delta->xed_startoff = NULLFILEOFF;
5421                 delta->xed_blockcount = 0;
5422         }
5423         /*
5424          * Check to see if the given block number is past the end of the
5425          * file, back up to the last block if so...
5426          */
5427         if (eof) {
5428                 ep = xfs_iext_get_ext(ifp, --lastx);
5429                 xfs_bmbt_get_all(ep, &got);
5430                 bno = got.br_startoff + got.br_blockcount - 1;
5431         }
5432         logflags = 0;
5433         if (ifp->if_flags & XFS_IFBROOT) {
5434                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5435                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5436                         whichfork);
5437                 cur->bc_private.b.firstblock = *firstblock;
5438                 cur->bc_private.b.flist = flist;
5439                 cur->bc_private.b.flags = 0;
5440         } else
5441                 cur = NULL;
5442         extno = 0;
5443         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5444                (nexts == 0 || extno < nexts)) {
5445                 /*
5446                  * Is the found extent after a hole in which bno lives?
5447                  * Just back up to the previous extent, if so.
5448                  */
5449                 if (got.br_startoff > bno) {
5450                         if (--lastx < 0)
5451                                 break;
5452                         ep = xfs_iext_get_ext(ifp, lastx);
5453                         xfs_bmbt_get_all(ep, &got);
5454                 }
5455                 /*
5456                  * Is the last block of this extent before the range
5457                  * we're supposed to delete?  If so, we're done.
5458                  */
5459                 bno = XFS_FILEOFF_MIN(bno,
5460                         got.br_startoff + got.br_blockcount - 1);
5461                 if (bno < start)
5462                         break;
5463                 /*
5464                  * Then deal with the (possibly delayed) allocated space
5465                  * we found.
5466                  */
5467                 ASSERT(ep != NULL);
5468                 del = got;
5469                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5470                 if (got.br_startoff < start) {
5471                         del.br_startoff = start;
5472                         del.br_blockcount -= start - got.br_startoff;
5473                         if (!wasdel)
5474                                 del.br_startblock += start - got.br_startoff;
5475                 }
5476                 if (del.br_startoff + del.br_blockcount > bno + 1)
5477                         del.br_blockcount = bno + 1 - del.br_startoff;
5478                 sum = del.br_startblock + del.br_blockcount;
5479                 if (isrt &&
5480                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5481                         /*
5482                          * Realtime extent not lined up at the end.
5483                          * The extent could have been split into written
5484                          * and unwritten pieces, or we could just be
5485                          * unmapping part of it.  But we can't really
5486                          * get rid of part of a realtime extent.
5487                          */
5488                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5489                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5490                                 /*
5491                                  * This piece is unwritten, or we're not
5492                                  * using unwritten extents.  Skip over it.
5493                                  */
5494                                 ASSERT(bno >= mod);
5495                                 bno -= mod > del.br_blockcount ?
5496                                         del.br_blockcount : mod;
5497                                 if (bno < got.br_startoff) {
5498                                         if (--lastx >= 0)
5499                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5500                                                         ifp, lastx), &got);
5501                                 }
5502                                 continue;
5503                         }
5504                         /*
5505                          * It's written, turn it unwritten.
5506                          * This is better than zeroing it.
5507                          */
5508                         ASSERT(del.br_state == XFS_EXT_NORM);
5509                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5510                         /*
5511                          * If this spans a realtime extent boundary,
5512                          * chop it back to the start of the one we end at.
5513                          */
5514                         if (del.br_blockcount > mod) {
5515                                 del.br_startoff += del.br_blockcount - mod;
5516                                 del.br_startblock += del.br_blockcount - mod;
5517                                 del.br_blockcount = mod;
5518                         }
5519                         del.br_state = XFS_EXT_UNWRITTEN;
5520                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5521                                 firstblock, flist, &logflags, delta,
5522                                 XFS_DATA_FORK, 0);
5523                         if (error)
5524                                 goto error0;
5525                         goto nodelete;
5526                 }
5527                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5528                         /*
5529                          * Realtime extent is lined up at the end but not
5530                          * at the front.  We'll get rid of full extents if
5531                          * we can.
5532                          */
5533                         mod = mp->m_sb.sb_rextsize - mod;
5534                         if (del.br_blockcount > mod) {
5535                                 del.br_blockcount -= mod;
5536                                 del.br_startoff += mod;
5537                                 del.br_startblock += mod;
5538                         } else if ((del.br_startoff == start &&
5539                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5540                                      xfs_trans_get_block_res(tp) == 0)) ||
5541                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5542                                 /*
5543                                  * Can't make it unwritten.  There isn't
5544                                  * a full extent here so just skip it.
5545                                  */
5546                                 ASSERT(bno >= del.br_blockcount);
5547                                 bno -= del.br_blockcount;
5548                                 if (bno < got.br_startoff) {
5549                                         if (--lastx >= 0)
5550                                                 xfs_bmbt_get_all(--ep, &got);
5551                                 }
5552                                 continue;
5553                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5554                                 /*
5555                                  * This one is already unwritten.
5556                                  * It must have a written left neighbor.
5557                                  * Unwrite the killed part of that one and
5558                                  * try again.
5559                                  */
5560                                 ASSERT(lastx > 0);
5561                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5562                                                 lastx - 1), &prev);
5563                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5564                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5565                                 ASSERT(del.br_startblock ==
5566                                        prev.br_startblock + prev.br_blockcount);
5567                                 if (prev.br_startoff < start) {
5568                                         mod = start - prev.br_startoff;
5569                                         prev.br_blockcount -= mod;
5570                                         prev.br_startblock += mod;
5571                                         prev.br_startoff = start;
5572                                 }
5573                                 prev.br_state = XFS_EXT_UNWRITTEN;
5574                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5575                                         &prev, firstblock, flist, &logflags,
5576                                         delta, XFS_DATA_FORK, 0);
5577                                 if (error)
5578                                         goto error0;
5579                                 goto nodelete;
5580                         } else {
5581                                 ASSERT(del.br_state == XFS_EXT_NORM);
5582                                 del.br_state = XFS_EXT_UNWRITTEN;
5583                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5584                                         &del, firstblock, flist, &logflags,
5585                                         delta, XFS_DATA_FORK, 0);
5586                                 if (error)
5587                                         goto error0;
5588                                 goto nodelete;
5589                         }
5590                 }
5591                 if (wasdel) {
5592                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5593                         /* Update realtime/data freespace, unreserve quota */
5594                         if (isrt) {
5595                                 xfs_filblks_t rtexts;
5596
5597                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5598                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5599                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5600                                                 (int64_t)rtexts, rsvd);
5601                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5602                                         NULL, ip, -((long)del.br_blockcount), 0,
5603                                         XFS_QMOPT_RES_RTBLKS);
5604                         } else {
5605                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5606                                                 (int64_t)del.br_blockcount, rsvd);
5607                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5608                                         NULL, ip, -((long)del.br_blockcount), 0,
5609                                         XFS_QMOPT_RES_REGBLKS);
5610                         }
5611                         ip->i_delayed_blks -= del.br_blockcount;
5612                         if (cur)
5613                                 cur->bc_private.b.flags |=
5614                                         XFS_BTCUR_BPRV_WASDEL;
5615                 } else if (cur)
5616                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5617                 /*
5618                  * If it's the case where the directory code is running
5619                  * with no block reservation, and the deleted block is in
5620                  * the middle of its extent, and the resulting insert
5621                  * of an extent would cause transformation to btree format,
5622                  * then reject it.  The calling code will then swap
5623                  * blocks around instead.
5624                  * We have to do this now, rather than waiting for the
5625                  * conversion to btree format, since the transaction
5626                  * will be dirty.
5627                  */
5628                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5629                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5630                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5631                     del.br_startoff > got.br_startoff &&
5632                     del.br_startoff + del.br_blockcount <
5633                     got.br_startoff + got.br_blockcount) {
5634                         error = XFS_ERROR(ENOSPC);
5635                         goto error0;
5636                 }
5637                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5638                                 &tmp_logflags, delta, whichfork, rsvd);
5639                 logflags |= tmp_logflags;
5640                 if (error)
5641                         goto error0;
5642                 bno = del.br_startoff - 1;
5643 nodelete:
5644                 lastx = ifp->if_lastex;
5645                 /*
5646                  * If not done go on to the next (previous) record.
5647                  * Reset ep in case the extents array was re-alloced.
5648                  */
5649                 ep = xfs_iext_get_ext(ifp, lastx);
5650                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5651                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5652                             xfs_bmbt_get_startoff(ep) > bno) {
5653                                 if (--lastx >= 0)
5654                                         ep = xfs_iext_get_ext(ifp, lastx);
5655                         }
5656                         if (lastx >= 0)
5657                                 xfs_bmbt_get_all(ep, &got);
5658                         extno++;
5659                 }
5660         }
5661         ifp->if_lastex = lastx;
5662         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5663         ASSERT(ifp->if_ext_max ==
5664                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5665         /*
5666          * Convert to a btree if necessary.
5667          */
5668         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5669             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5670                 ASSERT(cur == NULL);
5671                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5672                         &cur, 0, &tmp_logflags, whichfork);
5673                 logflags |= tmp_logflags;
5674                 if (error)
5675                         goto error0;
5676         }
5677         /*
5678          * transform from btree to extents, give it cur
5679          */
5680         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5681                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5682                 ASSERT(cur != NULL);
5683                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5684                         whichfork);
5685                 logflags |= tmp_logflags;
5686                 if (error)
5687                         goto error0;
5688         }
5689         /*
5690          * transform from extents to local?
5691          */
5692         ASSERT(ifp->if_ext_max ==
5693                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5694         error = 0;
5695         if (delta && delta->xed_startoff != NULLFILEOFF) {
5696                 /* A change was actually made.
5697                  * Note that delta->xed_blockount is an offset at this
5698                  * point and needs to be converted to a block count.
5699                  */
5700                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5701                 delta->xed_blockcount -= delta->xed_startoff;
5702         }
5703 error0:
5704         /*
5705          * Log everything.  Do this after conversion, there's no point in
5706          * logging the extent records if we've converted to btree format.
5707          */
5708         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5709             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5710                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5711         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5712                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5713                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5714         /*
5715          * Log inode even in the error case, if the transaction
5716          * is dirty we'll need to shut down the filesystem.
5717          */
5718         if (logflags)
5719                 xfs_trans_log_inode(tp, ip, logflags);
5720         if (cur) {
5721                 if (!error) {
5722                         *firstblock = cur->bc_private.b.firstblock;
5723                         cur->bc_private.b.allocated = 0;
5724                 }
5725                 xfs_btree_del_cursor(cur,
5726                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5727         }
5728         return error;
5729 }
5730
5731 /*
5732  * returns 1 for success, 0 if we failed to map the extent.
5733  */
5734 STATIC int
5735 xfs_getbmapx_fix_eof_hole(
5736         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5737         struct getbmap          *out,           /* output structure */
5738         int                     prealloced,     /* this is a file with
5739                                                 * preallocated data space */
5740         __int64_t               end,            /* last block requested */
5741         xfs_fsblock_t           startblock)
5742 {
5743         __int64_t               fixlen;
5744         xfs_mount_t             *mp;            /* file system mount point */
5745
5746         if (startblock == HOLESTARTBLOCK) {
5747                 mp = ip->i_mount;
5748                 out->bmv_block = -1;
5749                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5750                 fixlen -= out->bmv_offset;
5751                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5752                         /* Came to hole at EOF. Trim it. */
5753                         if (fixlen <= 0)
5754                                 return 0;
5755                         out->bmv_length = fixlen;
5756                 }
5757         } else {
5758                 out->bmv_block = XFS_FSB_TO_DB(ip, startblock);
5759         }
5760
5761         return 1;
5762 }
5763
5764 /*
5765  * Fcntl interface to xfs_bmapi.
5766  */
5767 int                                             /* error code */
5768 xfs_getbmap(
5769         xfs_inode_t             *ip,
5770         struct getbmap          *bmv,           /* user bmap structure */
5771         void                    __user *ap,     /* pointer to user's array */
5772         int                     interface)      /* interface flags */
5773 {
5774         __int64_t               bmvend;         /* last block requested */
5775         int                     error;          /* return value */
5776         __int64_t               fixlen;         /* length for -1 case */
5777         int                     i;              /* extent number */
5778         int                     lock;           /* lock state */
5779         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5780         xfs_mount_t             *mp;            /* file system mount point */
5781         int                     nex;            /* # of user extents can do */
5782         int                     nexleft;        /* # of user extents left */
5783         int                     subnex;         /* # of bmapi's can do */
5784         int                     nmap;           /* number of map entries */
5785         struct getbmap          out;            /* output structure */
5786         int                     whichfork;      /* data or attr fork */
5787         int                     prealloced;     /* this is a file with
5788                                                  * preallocated data space */
5789         int                     sh_unwritten;   /* true, if unwritten */
5790                                                 /* extents listed separately */
5791         int                     bmapi_flags;    /* flags for xfs_bmapi */
5792         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5793
5794         mp = ip->i_mount;
5795
5796         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5797         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5798
5799         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5800          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5801          *      bit is set for the file, generate a read event in order
5802          *      that the DMAPI application may do its thing before we return
5803          *      the extents.  Usually this means restoring user file data to
5804          *      regions of the file that look like holes.
5805          *
5806          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5807          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5808          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5809          *      could misinterpret holes in a DMAPI file as true holes,
5810          *      when in fact they may represent offline user data.
5811          */
5812         if ((interface & BMV_IF_NO_DMAPI_READ) == 0 &&
5813             DM_EVENT_ENABLED(ip, DM_EVENT_READ) &&
5814             whichfork == XFS_DATA_FORK) {
5815                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, 0, 0, 0, NULL);
5816                 if (error)
5817                         return XFS_ERROR(error);
5818         }
5819
5820         if (whichfork == XFS_ATTR_FORK) {
5821                 if (XFS_IFORK_Q(ip)) {
5822                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5823                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5824                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5825                                 return XFS_ERROR(EINVAL);
5826                 } else if (unlikely(
5827                            ip->i_d.di_aformat != 0 &&
5828                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5829                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5830                                          ip->i_mount);
5831                         return XFS_ERROR(EFSCORRUPTED);
5832                 }
5833         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5834                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5835                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5836                 return XFS_ERROR(EINVAL);
5837         if (whichfork == XFS_DATA_FORK) {
5838                 if (xfs_get_extsz_hint(ip) ||
5839                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5840                         prealloced = 1;
5841                         fixlen = XFS_MAXIOFFSET(mp);
5842                 } else {
5843                         prealloced = 0;
5844                         fixlen = ip->i_size;
5845                 }
5846         } else {
5847                 prealloced = 0;
5848                 fixlen = 1LL << 32;
5849         }
5850
5851         if (bmv->bmv_length == -1) {
5852                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5853                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5854                                         (__int64_t)0);
5855         } else if (bmv->bmv_length < 0)
5856                 return XFS_ERROR(EINVAL);
5857         if (bmv->bmv_length == 0) {
5858                 bmv->bmv_entries = 0;
5859                 return 0;
5860         }
5861         nex = bmv->bmv_count - 1;
5862         if (nex <= 0)
5863                 return XFS_ERROR(EINVAL);
5864         bmvend = bmv->bmv_offset + bmv->bmv_length;
5865
5866         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5867
5868         if (whichfork == XFS_DATA_FORK &&
5869                 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5870                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5871                 error = xfs_flush_pages(ip, (xfs_off_t)0,
5872                                                -1, 0, FI_REMAPF);
5873                 if (error) {
5874                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5875                 return error;
5876                 }
5877         }
5878
5879         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5880
5881         lock = xfs_ilock_map_shared(ip);
5882
5883         /*
5884          * Don't let nex be bigger than the number of extents
5885          * we can have assuming alternating holes and real extents.
5886          */
5887         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5888                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5889
5890         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5891                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5892
5893         /*
5894          * Allocate enough space to handle "subnex" maps at a time.
5895          */
5896         subnex = 16;
5897         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5898
5899         bmv->bmv_entries = 0;
5900
5901         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5902                 error = 0;
5903                 goto unlock_and_return;
5904         }
5905
5906         nexleft = nex;
5907
5908         do {
5909                 nmap = (nexleft > subnex) ? subnex : nexleft;
5910                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5911                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5912                                   bmapi_flags, NULL, 0, map, &nmap,
5913                                   NULL, NULL);
5914                 if (error)
5915                         goto unlock_and_return;
5916                 ASSERT(nmap <= subnex);
5917
5918                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5919                         nexleft--;
5920                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5921                                         BMV_OF_PREALLOC : 0;
5922                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5923                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5924                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5925                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5926                             whichfork == XFS_ATTR_FORK) {
5927                                 /* came to the end of attribute fork */
5928                                 goto unlock_and_return;
5929                         } else {
5930                                 if (!xfs_getbmapx_fix_eof_hole(ip, &out,
5931                                                         prealloced, bmvend,
5932                                                         map[i].br_startblock)) {
5933                                         goto unlock_and_return;
5934                                 }
5935
5936                                 /* return either getbmap/getbmapx structure. */
5937                                 if (interface & BMV_IF_EXTENDED) {
5938                                         struct  getbmapx        outx;
5939
5940                                         GETBMAP_CONVERT(out,outx);
5941                                         outx.bmv_oflags = oflags;
5942                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5943                                         if (copy_to_user(ap, &outx,
5944                                                         sizeof(outx))) {
5945                                                 error = XFS_ERROR(EFAULT);
5946                                                 goto unlock_and_return;
5947                                         }
5948                                 } else {
5949                                         if (copy_to_user(ap, &out,
5950                                                         sizeof(out))) {
5951                                                 error = XFS_ERROR(EFAULT);
5952                                                 goto unlock_and_return;
5953                                         }
5954                                 }
5955                                 bmv->bmv_offset =
5956                                         out.bmv_offset + out.bmv_length;
5957                                 bmv->bmv_length = MAX((__int64_t)0,
5958                                         (__int64_t)(bmvend - bmv->bmv_offset));
5959                                 bmv->bmv_entries++;
5960                                 ap = (interface & BMV_IF_EXTENDED) ?
5961                                                 (void __user *)
5962                                         ((struct getbmapx __user *)ap + 1) :
5963                                                 (void __user *)
5964                                         ((struct getbmap __user *)ap + 1);
5965                         }
5966                 }
5967         } while (nmap && nexleft && bmv->bmv_length);
5968
5969 unlock_and_return:
5970         xfs_iunlock_map_shared(ip, lock);
5971         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5972
5973         kmem_free(map, subnex * sizeof(*map));
5974
5975         return error;
5976 }
5977
5978 /*
5979  * Check the last inode extent to determine whether this allocation will result
5980  * in blocks being allocated at the end of the file. When we allocate new data
5981  * blocks at the end of the file which do not start at the previous data block,
5982  * we will try to align the new blocks at stripe unit boundaries.
5983  */
5984 STATIC int                              /* error */
5985 xfs_bmap_isaeof(
5986         xfs_inode_t     *ip,            /* incore inode pointer */
5987         xfs_fileoff_t   off,            /* file offset in fsblocks */
5988         int             whichfork,      /* data or attribute fork */
5989         char            *aeof)          /* return value */
5990 {
5991         int             error;          /* error return value */
5992         xfs_ifork_t     *ifp;           /* inode fork pointer */
5993         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5994         xfs_extnum_t    nextents;       /* number of file extents */
5995         xfs_bmbt_irec_t s;              /* expanded extent record */
5996
5997         ASSERT(whichfork == XFS_DATA_FORK);
5998         ifp = XFS_IFORK_PTR(ip, whichfork);
5999         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6000             (error = xfs_iread_extents(NULL, ip, whichfork)))
6001                 return error;
6002         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6003         if (nextents == 0) {
6004                 *aeof = 1;
6005                 return 0;
6006         }
6007         /*
6008          * Go to the last extent
6009          */
6010         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6011         xfs_bmbt_get_all(lastrec, &s);
6012         /*
6013          * Check we are allocating in the last extent (for delayed allocations)
6014          * or past the last extent for non-delayed allocations.
6015          */
6016         *aeof = (off >= s.br_startoff &&
6017                  off < s.br_startoff + s.br_blockcount &&
6018                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6019                 off >= s.br_startoff + s.br_blockcount;
6020         return 0;
6021 }
6022
6023 /*
6024  * Check if the endoff is outside the last extent. If so the caller will grow
6025  * the allocation to a stripe unit boundary.
6026  */
6027 int                                     /* error */
6028 xfs_bmap_eof(
6029         xfs_inode_t     *ip,            /* incore inode pointer */
6030         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6031         int             whichfork,      /* data or attribute fork */
6032         int             *eof)           /* result value */
6033 {
6034         xfs_fsblock_t   blockcount;     /* extent block count */
6035         int             error;          /* error return value */
6036         xfs_ifork_t     *ifp;           /* inode fork pointer */
6037         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6038         xfs_extnum_t    nextents;       /* number of file extents */
6039         xfs_fileoff_t   startoff;       /* extent starting file offset */
6040
6041         ASSERT(whichfork == XFS_DATA_FORK);
6042         ifp = XFS_IFORK_PTR(ip, whichfork);
6043         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6044             (error = xfs_iread_extents(NULL, ip, whichfork)))
6045                 return error;
6046         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6047         if (nextents == 0) {
6048                 *eof = 1;
6049                 return 0;
6050         }
6051         /*
6052          * Go to the last extent
6053          */
6054         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6055         startoff = xfs_bmbt_get_startoff(lastrec);
6056         blockcount = xfs_bmbt_get_blockcount(lastrec);
6057         *eof = endoff >= startoff + blockcount;
6058         return 0;
6059 }
6060
6061 #ifdef DEBUG
6062 STATIC
6063 xfs_buf_t *
6064 xfs_bmap_get_bp(
6065         xfs_btree_cur_t         *cur,
6066         xfs_fsblock_t           bno)
6067 {
6068         int i;
6069         xfs_buf_t *bp;
6070
6071         if (!cur)
6072                 return(NULL);
6073
6074         bp = NULL;
6075         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6076                 bp = cur->bc_bufs[i];
6077                 if (!bp) break;
6078                 if (XFS_BUF_ADDR(bp) == bno)
6079                         break;  /* Found it */
6080         }
6081         if (i == XFS_BTREE_MAXLEVELS)
6082                 bp = NULL;
6083
6084         if (!bp) { /* Chase down all the log items to see if the bp is there */
6085                 xfs_log_item_chunk_t    *licp;
6086                 xfs_trans_t             *tp;
6087
6088                 tp = cur->bc_tp;
6089                 licp = &tp->t_items;
6090                 while (!bp && licp != NULL) {
6091                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
6092                                 licp = licp->lic_next;
6093                                 continue;
6094                         }
6095                         for (i = 0; i < licp->lic_unused; i++) {
6096                                 xfs_log_item_desc_t     *lidp;
6097                                 xfs_log_item_t          *lip;
6098                                 xfs_buf_log_item_t      *bip;
6099                                 xfs_buf_t               *lbp;
6100
6101                                 if (XFS_LIC_ISFREE(licp, i)) {
6102                                         continue;
6103                                 }
6104
6105                                 lidp = XFS_LIC_SLOT(licp, i);
6106                                 lip = lidp->lid_item;
6107                                 if (lip->li_type != XFS_LI_BUF)
6108                                         continue;
6109
6110                                 bip = (xfs_buf_log_item_t *)lip;
6111                                 lbp = bip->bli_buf;
6112
6113                                 if (XFS_BUF_ADDR(lbp) == bno) {
6114                                         bp = lbp;
6115                                         break; /* Found it */
6116                                 }
6117                         }
6118                         licp = licp->lic_next;
6119                 }
6120         }
6121         return(bp);
6122 }
6123
6124 void
6125 xfs_check_block(
6126         xfs_bmbt_block_t        *block,
6127         xfs_mount_t             *mp,
6128         int                     root,
6129         short                   sz)
6130 {
6131         int                     i, j, dmxr;
6132         __be64                  *pp, *thispa;   /* pointer to block address */
6133         xfs_bmbt_key_t          *prevp, *keyp;
6134
6135         ASSERT(be16_to_cpu(block->bb_level) > 0);
6136
6137         prevp = NULL;
6138         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6139                 dmxr = mp->m_bmap_dmxr[0];
6140
6141                 if (root) {
6142                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6143                 } else {
6144                         keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6145                 }
6146
6147                 if (prevp) {
6148                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6149                 }
6150                 prevp = keyp;
6151
6152                 /*
6153                  * Compare the block numbers to see if there are dups.
6154                  */
6155
6156                 if (root) {
6157                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6158                 } else {
6159                         pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6160                 }
6161                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6162                         if (root) {
6163                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6164                         } else {
6165                                 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6166                                                             dmxr);
6167                         }
6168                         if (*thispa == *pp) {
6169                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6170                                         __func__, j, i,
6171                                         (unsigned long long)be64_to_cpu(*thispa));
6172                                 panic("%s: ptrs are equal in node\n",
6173                                         __func__);
6174                         }
6175                 }
6176         }
6177 }
6178
6179 /*
6180  * Check that the extents for the inode ip are in the right order in all
6181  * btree leaves.
6182  */
6183
6184 STATIC void
6185 xfs_bmap_check_leaf_extents(
6186         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6187         xfs_inode_t             *ip,            /* incore inode pointer */
6188         int                     whichfork)      /* data or attr fork */
6189 {
6190         xfs_bmbt_block_t        *block; /* current btree block */
6191         xfs_fsblock_t           bno;    /* block # of "block" */
6192         xfs_buf_t               *bp;    /* buffer for "block" */
6193         int                     error;  /* error return value */
6194         xfs_extnum_t            i=0, j; /* index into the extents list */
6195         xfs_ifork_t             *ifp;   /* fork structure */
6196         int                     level;  /* btree level, for checking */
6197         xfs_mount_t             *mp;    /* file system mount structure */
6198         __be64                  *pp;    /* pointer to block address */
6199         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6200         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
6201         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6202         int                     bp_release = 0;
6203
6204         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6205                 return;
6206         }
6207
6208         bno = NULLFSBLOCK;
6209         mp = ip->i_mount;
6210         ifp = XFS_IFORK_PTR(ip, whichfork);
6211         block = ifp->if_broot;
6212         /*
6213          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6214          */
6215         level = be16_to_cpu(block->bb_level);
6216         ASSERT(level > 0);
6217         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6218         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6219         bno = be64_to_cpu(*pp);
6220
6221         ASSERT(bno != NULLDFSBNO);
6222         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6223         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6224
6225         /*
6226          * Go down the tree until leaf level is reached, following the first
6227          * pointer (leftmost) at each level.
6228          */
6229         while (level-- > 0) {
6230                 /* See if buf is in cur first */
6231                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6232                 if (bp) {
6233                         bp_release = 0;
6234                 } else {
6235                         bp_release = 1;
6236                 }
6237                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6238                                 XFS_BMAP_BTREE_REF)))
6239                         goto error_norelse;
6240                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6241                 XFS_WANT_CORRUPTED_GOTO(
6242                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6243                         error0);
6244                 if (level == 0)
6245                         break;
6246
6247                 /*
6248                  * Check this block for basic sanity (increasing keys and
6249                  * no duplicate blocks).
6250                  */
6251
6252                 xfs_check_block(block, mp, 0, 0);
6253                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6254                 bno = be64_to_cpu(*pp);
6255                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6256                 if (bp_release) {
6257                         bp_release = 0;
6258                         xfs_trans_brelse(NULL, bp);
6259                 }
6260         }
6261
6262         /*
6263          * Here with bp and block set to the leftmost leaf node in the tree.
6264          */
6265         i = 0;
6266
6267         /*
6268          * Loop over all leaf nodes checking that all extents are in the right order.
6269          */
6270         for (;;) {
6271                 xfs_fsblock_t   nextbno;
6272                 xfs_extnum_t    num_recs;
6273
6274
6275                 num_recs = be16_to_cpu(block->bb_numrecs);
6276
6277                 /*
6278                  * Read-ahead the next leaf block, if any.
6279                  */
6280
6281                 nextbno = be64_to_cpu(block->bb_rightsib);
6282
6283                 /*
6284                  * Check all the extents to make sure they are OK.
6285                  * If we had a previous block, the last entry should
6286                  * conform with the first entry in this one.
6287                  */
6288
6289                 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6290                 if (i) {
6291                         xfs_btree_check_rec(XFS_BTNUM_BMAP, &last, ep);
6292                 }
6293                 for (j = 1; j < num_recs; j++) {
6294                         nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6295                         xfs_btree_check_rec(XFS_BTNUM_BMAP, ep, nextp);
6296                         ep = nextp;
6297                 }
6298
6299                 last = *ep;
6300                 i += num_recs;
6301                 if (bp_release) {
6302                         bp_release = 0;
6303                         xfs_trans_brelse(NULL, bp);
6304                 }
6305                 bno = nextbno;
6306                 /*
6307                  * If we've reached the end, stop.
6308                  */
6309                 if (bno == NULLFSBLOCK)
6310                         break;
6311
6312                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6313                 if (bp) {
6314                         bp_release = 0;
6315                 } else {
6316                         bp_release = 1;
6317                 }
6318                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6319                                 XFS_BMAP_BTREE_REF)))
6320                         goto error_norelse;
6321                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6322         }
6323         if (bp_release) {
6324                 bp_release = 0;
6325                 xfs_trans_brelse(NULL, bp);
6326         }
6327         return;
6328
6329 error0:
6330         cmn_err(CE_WARN, "%s: at error0", __func__);
6331         if (bp_release)
6332                 xfs_trans_brelse(NULL, bp);
6333 error_norelse:
6334         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6335                 __func__, i);
6336         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6337         return;
6338 }
6339 #endif
6340
6341 /*
6342  * Count fsblocks of the given fork.
6343  */
6344 int                                             /* error */
6345 xfs_bmap_count_blocks(
6346         xfs_trans_t             *tp,            /* transaction pointer */
6347         xfs_inode_t             *ip,            /* incore inode */
6348         int                     whichfork,      /* data or attr fork */
6349         int                     *count)         /* out: count of blocks */
6350 {
6351         xfs_bmbt_block_t        *block; /* current btree block */
6352         xfs_fsblock_t           bno;    /* block # of "block" */
6353         xfs_ifork_t             *ifp;   /* fork structure */
6354         int                     level;  /* btree level, for checking */
6355         xfs_mount_t             *mp;    /* file system mount structure */
6356         __be64                  *pp;    /* pointer to block address */
6357
6358         bno = NULLFSBLOCK;
6359         mp = ip->i_mount;
6360         ifp = XFS_IFORK_PTR(ip, whichfork);
6361         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6362                 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6363                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6364                         count) < 0)) {
6365                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6366                                          XFS_ERRLEVEL_LOW, mp);
6367                         return XFS_ERROR(EFSCORRUPTED);
6368                 }
6369                 return 0;
6370         }
6371
6372         /*
6373          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6374          */
6375         block = ifp->if_broot;
6376         level = be16_to_cpu(block->bb_level);
6377         ASSERT(level > 0);
6378         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6379         bno = be64_to_cpu(*pp);
6380         ASSERT(bno != NULLDFSBNO);
6381         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6382         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6383
6384         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6385                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6386                                  mp);
6387                 return XFS_ERROR(EFSCORRUPTED);
6388         }
6389
6390         return 0;
6391 }
6392
6393 /*
6394  * Recursively walks each level of a btree
6395  * to count total fsblocks is use.
6396  */
6397 STATIC int                                     /* error */
6398 xfs_bmap_count_tree(
6399         xfs_mount_t     *mp,            /* file system mount point */
6400         xfs_trans_t     *tp,            /* transaction pointer */
6401         xfs_ifork_t     *ifp,           /* inode fork pointer */
6402         xfs_fsblock_t   blockno,        /* file system block number */
6403         int             levelin,        /* level in btree */
6404         int             *count)         /* Count of blocks */
6405 {
6406         int                     error;
6407         xfs_buf_t               *bp, *nbp;
6408         int                     level = levelin;
6409         __be64                  *pp;
6410         xfs_fsblock_t           bno = blockno;
6411         xfs_fsblock_t           nextbno;
6412         xfs_bmbt_block_t        *block, *nextblock;
6413         int                     numrecs;
6414
6415         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6416                 return error;
6417         *count += 1;
6418         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6419
6420         if (--level) {
6421                 /* Not at node above leafs, count this level of nodes */
6422                 nextbno = be64_to_cpu(block->bb_rightsib);
6423                 while (nextbno != NULLFSBLOCK) {
6424                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6425                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6426                                 return error;
6427                         *count += 1;
6428                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6429                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6430                         xfs_trans_brelse(tp, nbp);
6431                 }
6432
6433                 /* Dive to the next level */
6434                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6435                 bno = be64_to_cpu(*pp);
6436                 if (unlikely((error =
6437                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6438                         xfs_trans_brelse(tp, bp);
6439                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6440                                          XFS_ERRLEVEL_LOW, mp);
6441                         return XFS_ERROR(EFSCORRUPTED);
6442                 }
6443                 xfs_trans_brelse(tp, bp);
6444         } else {
6445                 /* count all level 1 nodes and their leaves */
6446                 for (;;) {
6447                         nextbno = be64_to_cpu(block->bb_rightsib);
6448                         numrecs = be16_to_cpu(block->bb_numrecs);
6449                         if (unlikely(xfs_bmap_disk_count_leaves(0,
6450                                         block, numrecs, count) < 0)) {
6451                                 xfs_trans_brelse(tp, bp);
6452                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6453                                                  XFS_ERRLEVEL_LOW, mp);
6454                                 return XFS_ERROR(EFSCORRUPTED);
6455                         }
6456                         xfs_trans_brelse(tp, bp);
6457                         if (nextbno == NULLFSBLOCK)
6458                                 break;
6459                         bno = nextbno;
6460                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6461                                 XFS_BMAP_BTREE_REF)))
6462                                 return error;
6463                         *count += 1;
6464                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6465                 }
6466         }
6467         return 0;
6468 }
6469
6470 /*
6471  * Count leaf blocks given a range of extent records.
6472  */
6473 STATIC int
6474 xfs_bmap_count_leaves(
6475         xfs_ifork_t             *ifp,
6476         xfs_extnum_t            idx,
6477         int                     numrecs,
6478         int                     *count)
6479 {
6480         int             b;
6481
6482         for (b = 0; b < numrecs; b++) {
6483                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6484                 *count += xfs_bmbt_get_blockcount(frp);
6485         }
6486         return 0;
6487 }
6488
6489 /*
6490  * Count leaf blocks given a range of extent records originally
6491  * in btree format.
6492  */
6493 STATIC int
6494 xfs_bmap_disk_count_leaves(
6495         xfs_extnum_t            idx,
6496         xfs_bmbt_block_t        *block,
6497         int                     numrecs,
6498         int                     *count)
6499 {
6500         int             b;
6501         xfs_bmbt_rec_t  *frp;
6502
6503         for (b = 1; b <= numrecs; b++) {
6504                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6505                 *count += xfs_bmbt_disk_get_blockcount(frp);
6506         }
6507         return 0;
6508 }