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
  57
  58 #ifdef DEBUG
  59 STATIC void
  60 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  61 #endif
  62
  63 kmem_zone_t             *xfs_bmap_free_item_zone;
  64
  65 /*
  66  * Prototypes for internal bmap routines.
  67  */
  68
  69
  70 /*
  71  * Called from xfs_bmap_add_attrfork to handle extents format files.
  72  */
  73 STATIC int                                      /* error */
  74 xfs_bmap_add_attrfork_extents(
  75         xfs_trans_t             *tp,            /* transaction pointer */
  76         xfs_inode_t             *ip,            /* incore inode pointer */
  77         xfs_fsblock_t           *firstblock,    /* first block allocated */
  78         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  79         int                     *flags);        /* inode logging flags */
  80
  81 /*
  82  * Called from xfs_bmap_add_attrfork to handle local format files.
  83  */
  84 STATIC int                                      /* error */
  85 xfs_bmap_add_attrfork_local(
  86         xfs_trans_t             *tp,            /* transaction pointer */
  87         xfs_inode_t             *ip,            /* incore inode pointer */
  88         xfs_fsblock_t           *firstblock,    /* first block allocated */
  89         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  90         int                     *flags);        /* inode logging flags */
  91
  92 /*
  93  * Called by xfs_bmapi to update file extent records and the btree
  94  * after allocating space (or doing a delayed allocation).
  95  */
  96 STATIC int                              /* error */
  97 xfs_bmap_add_extent(
  98         xfs_inode_t             *ip,    /* incore inode pointer */
  99         xfs_extnum_t            idx,    /* extent number to update/insert */
 100         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 101         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 102         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 103         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 104         int                     *logflagsp, /* inode logging flags */
 105         xfs_extdelta_t          *delta, /* Change made to incore extents */
 106         int                     whichfork, /* data or attr fork */
 107         int                     rsvd);  /* OK to allocate reserved blocks */
 108
 109 /*
 110  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 111  * allocation to a real allocation.
 112  */
 113 STATIC int                              /* error */
 114 xfs_bmap_add_extent_delay_real(
 115         xfs_inode_t             *ip,    /* incore inode pointer */
 116         xfs_extnum_t            idx,    /* extent number to update/insert */
 117         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 118         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 119         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 120         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 121         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 122         int                     *logflagsp, /* inode logging flags */
 123         xfs_extdelta_t          *delta, /* Change made to incore extents */
 124         int                     rsvd);  /* OK to allocate reserved blocks */
 125
 126 /*
 127  * Called by xfs_bmap_add_extent to handle cases converting a hole
 128  * to a delayed allocation.
 129  */
 130 STATIC int                              /* error */
 131 xfs_bmap_add_extent_hole_delay(
 132         xfs_inode_t             *ip,    /* incore inode pointer */
 133         xfs_extnum_t            idx,    /* extent number to update/insert */
 134         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 135         int                     *logflagsp,/* inode logging flags */
 136         xfs_extdelta_t          *delta, /* Change made to incore extents */
 137         int                     rsvd);  /* OK to allocate reserved blocks */
 138
 139 /*
 140  * Called by xfs_bmap_add_extent to handle cases converting a hole
 141  * to a real allocation.
 142  */
 143 STATIC int                              /* error */
 144 xfs_bmap_add_extent_hole_real(
 145         xfs_inode_t             *ip,    /* incore inode pointer */
 146         xfs_extnum_t            idx,    /* extent number to update/insert */
 147         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 148         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 149         int                     *logflagsp, /* inode logging flags */
 150         xfs_extdelta_t          *delta, /* Change made to incore extents */
 151         int                     whichfork); /* data or attr fork */
 152
 153 /*
 154  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 155  * allocation to a real allocation or vice versa.
 156  */
 157 STATIC int                              /* error */
 158 xfs_bmap_add_extent_unwritten_real(
 159         xfs_inode_t             *ip,    /* incore inode pointer */
 160         xfs_extnum_t            idx,    /* extent number to update/insert */
 161         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 162         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 163         int                     *logflagsp, /* inode logging flags */
 164         xfs_extdelta_t          *delta); /* Change made to incore extents */
 165
 166 /*
 167  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 168  * It figures out where to ask the underlying allocator to put the new extent.
 169  */
 170 STATIC int                              /* error */
 171 xfs_bmap_alloc(
 172         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 173
 174 /*
 175  * Transform a btree format file with only one leaf node, where the
 176  * extents list will fit in the inode, into an extents format file.
 177  * Since the file extents are already in-core, all we have to do is
 178  * give up the space for the btree root and pitch the leaf block.
 179  */
 180 STATIC int                              /* error */
 181 xfs_bmap_btree_to_extents(
 182         xfs_trans_t             *tp,    /* transaction pointer */
 183         xfs_inode_t             *ip,    /* incore inode pointer */
 184         xfs_btree_cur_t         *cur,   /* btree cursor */
 185         int                     *logflagsp, /* inode logging flags */
 186         int                     whichfork); /* data or attr fork */
 187
 188 /*
 189  * Called by xfs_bmapi to update file extent records and the btree
 190  * after removing space (or undoing a delayed allocation).
 191  */
 192 STATIC int                              /* error */
 193 xfs_bmap_del_extent(
 194         xfs_inode_t             *ip,    /* incore inode pointer */
 195         xfs_trans_t             *tp,    /* current trans pointer */
 196         xfs_extnum_t            idx,    /* extent number to update/insert */
 197         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 198         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 199         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 200         int                     *logflagsp,/* inode logging flags */
 201         xfs_extdelta_t          *delta, /* Change made to incore extents */
 202         int                     whichfork, /* data or attr fork */
 203         int                     rsvd);   /* OK to allocate reserved blocks */
 204
 205 /*
 206  * Remove the entry "free" from the free item list.  Prev points to the
 207  * previous entry, unless "free" is the head of the list.
 208  */
 209 STATIC void
 210 xfs_bmap_del_free(
 211         xfs_bmap_free_t         *flist, /* free item list header */
 212         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 213         xfs_bmap_free_item_t    *free); /* list item to be freed */
 214
 215 /*
 216  * Convert an extents-format file into a btree-format file.
 217  * The new file will have a root block (in the inode) and a single child block.
 218  */
 219 STATIC int                                      /* error */
 220 xfs_bmap_extents_to_btree(
 221         xfs_trans_t             *tp,            /* transaction pointer */
 222         xfs_inode_t             *ip,            /* incore inode pointer */
 223         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 224         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 225         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 226         int                     wasdel,         /* converting a delayed alloc */
 227         int                     *logflagsp,     /* inode logging flags */
 228         int                     whichfork);     /* data or attr fork */
 229
 230 /*
 231  * Convert a local file to an extents file.
 232  * This code is sort of bogus, since the file data needs to get
 233  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 234  */
 235 STATIC int                              /* error */
 236 xfs_bmap_local_to_extents(
 237         xfs_trans_t     *tp,            /* transaction pointer */
 238         xfs_inode_t     *ip,            /* incore inode pointer */
 239         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 240         xfs_extlen_t    total,          /* total blocks needed by transaction */
 241         int             *logflagsp,     /* inode logging flags */
 242         int             whichfork);     /* data or attr fork */
 243
 244 /*
 245  * Search the extents list for the inode, for the extent containing bno.
 246  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 247  * *eofp will be set, and *prevp will contain the last entry (null if none).
 248  * Else, *lastxp will be set to the index of the found
 249  * entry; *gotp will contain the entry.
 250  */
 251 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
 252 xfs_bmap_search_extents(
 253         xfs_inode_t     *ip,            /* incore inode pointer */
 254         xfs_fileoff_t   bno,            /* block number searched for */
 255         int             whichfork,      /* data or attr fork */
 256         int             *eofp,          /* out: end of file found */
 257         xfs_extnum_t    *lastxp,        /* out: last extent index */
 258         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 259         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 260
 261 /*
 262  * Check the last inode extent to determine whether this allocation will result
 263  * in blocks being allocated at the end of the file. When we allocate new data
 264  * blocks at the end of the file which do not start at the previous data block,
 265  * we will try to align the new blocks at stripe unit boundaries.
 266  */
 267 STATIC int                              /* error */
 268 xfs_bmap_isaeof(
 269         xfs_inode_t     *ip,            /* incore inode pointer */
 270         xfs_fileoff_t   off,            /* file offset in fsblocks */
 271         int             whichfork,      /* data or attribute fork */
 272         char            *aeof);         /* return value */
 273
 274 #ifdef XFS_BMAP_TRACE
 275 /*
 276  * Add a bmap trace buffer entry.  Base routine for the others.
 277  */
 278 STATIC void
 279 xfs_bmap_trace_addentry(
 280         int             opcode,         /* operation */
 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(ies) */
 285         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
 286         xfs_bmbt_rec_t  *r1,            /* first record */
 287         xfs_bmbt_rec_t  *r2,            /* second record or null */
 288         int             whichfork);     /* data or attr fork */
 289
 290 /*
 291  * Add bmap trace entry prior to a call to xfs_iext_remove.
 292  */
 293 STATIC void
 294 xfs_bmap_trace_delete(
 295         const char      *fname,         /* function name */
 296         char            *desc,          /* operation description */
 297         xfs_inode_t     *ip,            /* incore inode pointer */
 298         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
 299         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
 300         int             whichfork);     /* data or attr fork */
 301
 302 /*
 303  * Add bmap trace entry prior to a call to xfs_iext_insert, or
 304  * reading in the extents list from the disk (in the btree).
 305  */
 306 STATIC void
 307 xfs_bmap_trace_insert(
 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(entries) inserted */
 312         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
 313         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
 314         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
 315         int             whichfork);     /* data or attr fork */
 316
 317 /*
 318  * Add bmap trace entry after updating an extent record in place.
 319  */
 320 STATIC void
 321 xfs_bmap_trace_post_update(
 322         const char      *fname,         /* function name */
 323         char            *desc,          /* operation description */
 324         xfs_inode_t     *ip,            /* incore inode pointer */
 325         xfs_extnum_t    idx,            /* index of entry updated */
 326         int             whichfork);     /* data or attr fork */
 327
 328 /*
 329  * Add bmap trace entry prior to updating an extent record in place.
 330  */
 331 STATIC void
 332 xfs_bmap_trace_pre_update(
 333         const char      *fname,         /* function name */
 334         char            *desc,          /* operation description */
 335         xfs_inode_t     *ip,            /* incore inode pointer */
 336         xfs_extnum_t    idx,            /* index of entry to be updated */
 337         int             whichfork);     /* data or attr fork */
 338
 339 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)       \
 340         xfs_bmap_trace_delete(__FUNCTION__,d,ip,i,c,w)
 341 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w) \
 342         xfs_bmap_trace_insert(__FUNCTION__,d,ip,i,c,r1,r2,w)
 343 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)    \
 344         xfs_bmap_trace_post_update(__FUNCTION__,d,ip,i,w)
 345 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)     \
 346         xfs_bmap_trace_pre_update(__FUNCTION__,d,ip,i,w)
 347 #else
 348 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)
 349 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w)
 350 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)
 351 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)
 352 #endif  /* XFS_BMAP_TRACE */
 353
 354 /*
 355  * Compute the worst-case number of indirect blocks that will be used
 356  * for ip's delayed extent of length "len".
 357  */
 358 STATIC xfs_filblks_t
 359 xfs_bmap_worst_indlen(
 360         xfs_inode_t             *ip,    /* incore inode pointer */
 361         xfs_filblks_t           len);   /* delayed extent length */
 362
 363 #ifdef DEBUG
 364 /*
 365  * Perform various validation checks on the values being returned
 366  * from xfs_bmapi().
 367  */
 368 STATIC void
 369 xfs_bmap_validate_ret(
 370         xfs_fileoff_t           bno,
 371         xfs_filblks_t           len,
 372         int                     flags,
 373         xfs_bmbt_irec_t         *mval,
 374         int                     nmap,
 375         int                     ret_nmap);
 376 #else
 377 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 378 #endif /* DEBUG */
 379
 380 #if defined(XFS_RW_TRACE)
 381 STATIC void
 382 xfs_bunmap_trace(
 383         xfs_inode_t             *ip,
 384         xfs_fileoff_t           bno,
 385         xfs_filblks_t           len,
 386         int                     flags,
 387         inst_t                  *ra);
 388 #else
 389 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
 390 #endif  /* XFS_RW_TRACE */
 391
 392 STATIC int
 393 xfs_bmap_count_tree(
 394         xfs_mount_t     *mp,
 395         xfs_trans_t     *tp,
 396         xfs_ifork_t     *ifp,
 397         xfs_fsblock_t   blockno,
 398         int             levelin,
 399         int             *count);
 400
 401 STATIC int
 402 xfs_bmap_count_leaves(
 403         xfs_ifork_t             *ifp,
 404         xfs_extnum_t            idx,
 405         int                     numrecs,
 406         int                     *count);
 407
 408 STATIC int
 409 xfs_bmap_disk_count_leaves(
 410         xfs_extnum_t            idx,
 411         xfs_bmbt_block_t        *block,
 412         int                     numrecs,
 413         int                     *count);
 414
 415 /*
 416  * Bmap internal routines.
 417  */
 418
 419 /*
 420  * Called from xfs_bmap_add_attrfork to handle btree format files.
 421  */
 422 STATIC int                                      /* error */
 423 xfs_bmap_add_attrfork_btree(
 424         xfs_trans_t             *tp,            /* transaction pointer */
 425         xfs_inode_t             *ip,            /* incore inode pointer */
 426         xfs_fsblock_t           *firstblock,    /* first block allocated */
 427         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 428         int                     *flags)         /* inode logging flags */
 429 {
 430         xfs_btree_cur_t         *cur;           /* btree cursor */
 431         int                     error;          /* error return value */
 432         xfs_mount_t             *mp;            /* file system mount struct */
 433         int                     stat;           /* newroot status */
 434
 435         mp = ip->i_mount;
 436         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 437                 *flags |= XFS_ILOG_DBROOT;
 438         else {
 439                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
 440                         XFS_DATA_FORK);
 441                 cur->bc_private.b.flist = flist;
 442                 cur->bc_private.b.firstblock = *firstblock;
 443                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 444                         goto error0;
 445                 ASSERT(stat == 1);      /* must be at least one entry */
 446                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
 447                         goto error0;
 448                 if (stat == 0) {
 449                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 450                         return XFS_ERROR(ENOSPC);
 451                 }
 452                 *firstblock = cur->bc_private.b.firstblock;
 453                 cur->bc_private.b.allocated = 0;
 454                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 455         }
 456         return 0;
 457 error0:
 458         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 459         return error;
 460 }
 461
 462 /*
 463  * Called from xfs_bmap_add_attrfork to handle extents format files.
 464  */
 465 STATIC int                                      /* error */
 466 xfs_bmap_add_attrfork_extents(
 467         xfs_trans_t             *tp,            /* transaction pointer */
 468         xfs_inode_t             *ip,            /* incore inode pointer */
 469         xfs_fsblock_t           *firstblock,    /* first block allocated */
 470         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 471         int                     *flags)         /* inode logging flags */
 472 {
 473         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 474         int                     error;          /* error return value */
 475
 476         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 477                 return 0;
 478         cur = NULL;
 479         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 480                 flags, XFS_DATA_FORK);
 481         if (cur) {
 482                 cur->bc_private.b.allocated = 0;
 483                 xfs_btree_del_cursor(cur,
 484                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 485         }
 486         return error;
 487 }
 488
 489 /*
 490  * Called from xfs_bmap_add_attrfork to handle local format files.
 491  */
 492 STATIC int                                      /* error */
 493 xfs_bmap_add_attrfork_local(
 494         xfs_trans_t             *tp,            /* transaction pointer */
 495         xfs_inode_t             *ip,            /* incore inode pointer */
 496         xfs_fsblock_t           *firstblock,    /* first block allocated */
 497         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 498         int                     *flags)         /* inode logging flags */
 499 {
 500         xfs_da_args_t           dargs;          /* args for dir/attr code */
 501         int                     error;          /* error return value */
 502         xfs_mount_t             *mp;            /* mount structure pointer */
 503
 504         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 505                 return 0;
 506         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 507                 mp = ip->i_mount;
 508                 memset(&dargs, 0, sizeof(dargs));
 509                 dargs.dp = ip;
 510                 dargs.firstblock = firstblock;
 511                 dargs.flist = flist;
 512                 dargs.total = mp->m_dirblkfsbs;
 513                 dargs.whichfork = XFS_DATA_FORK;
 514                 dargs.trans = tp;
 515                 error = xfs_dir2_sf_to_block(&dargs);
 516         } else
 517                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 518                         XFS_DATA_FORK);
 519         return error;
 520 }
 521
 522 /*
 523  * Called by xfs_bmapi to update file extent records and the btree
 524  * after allocating space (or doing a delayed allocation).
 525  */
 526 STATIC int                              /* error */
 527 xfs_bmap_add_extent(
 528         xfs_inode_t             *ip,    /* incore inode pointer */
 529         xfs_extnum_t            idx,    /* extent number to update/insert */
 530         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 531         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 532         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 533         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 534         int                     *logflagsp, /* inode logging flags */
 535         xfs_extdelta_t          *delta, /* Change made to incore extents */
 536         int                     whichfork, /* data or attr fork */
 537         int                     rsvd)   /* OK to use reserved data blocks */
 538 {
 539         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 540         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 541         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 542         int                     error;  /* error return value */
 543         xfs_ifork_t             *ifp;   /* inode fork ptr */
 544         int                     logflags; /* returned value */
 545         xfs_extnum_t            nextents; /* number of extents in file now */
 546
 547         XFS_STATS_INC(xs_add_exlist);
 548         cur = *curp;
 549         ifp = XFS_IFORK_PTR(ip, whichfork);
 550         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 551         ASSERT(idx <= nextents);
 552         da_old = da_new = 0;
 553         error = 0;
 554         /*
 555          * This is the first extent added to a new/empty file.
 556          * Special case this one, so other routines get to assume there are
 557          * already extents in the list.
 558          */
 559         if (nextents == 0) {
 560                 XFS_BMAP_TRACE_INSERT("insert empty", ip, 0, 1, new, NULL,
 561                         whichfork);
 562                 xfs_iext_insert(ifp, 0, 1, new);
 563                 ASSERT(cur == NULL);
 564                 ifp->if_lastex = 0;
 565                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
 566                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 567                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
 568                 } else
 569                         logflags = 0;
 570                 /* DELTA: single new extent */
 571                 if (delta) {
 572                         if (delta->xed_startoff > new->br_startoff)
 573                                 delta->xed_startoff = new->br_startoff;
 574                         if (delta->xed_blockcount <
 575                                         new->br_startoff + new->br_blockcount)
 576                                 delta->xed_blockcount = new->br_startoff +
 577                                                 new->br_blockcount;
 578                 }
 579         }
 580         /*
 581          * Any kind of new delayed allocation goes here.
 582          */
 583         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
 584                 if (cur)
 585                         ASSERT((cur->bc_private.b.flags &
 586                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 587                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 588                                 &logflags, delta, rsvd)))
 589                         goto done;
 590         }
 591         /*
 592          * Real allocation off the end of the file.
 593          */
 594         else if (idx == nextents) {
 595                 if (cur)
 596                         ASSERT((cur->bc_private.b.flags &
 597                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 598                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 599                                 &logflags, delta, whichfork)))
 600                         goto done;
 601         } else {
 602                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 603
 604                 /*
 605                  * Get the record referred to by idx.
 606                  */
 607                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
 608                 /*
 609                  * If it's a real allocation record, and the new allocation ends
 610                  * after the start of the referred to record, then we're filling
 611                  * in a delayed or unwritten allocation with a real one, or
 612                  * converting real back to unwritten.
 613                  */
 614                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
 615                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 616                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 617                             ISNULLSTARTBLOCK(prev.br_startblock)) {
 618                                 da_old = STARTBLOCKVAL(prev.br_startblock);
 619                                 if (cur)
 620                                         ASSERT(cur->bc_private.b.flags &
 621                                                 XFS_BTCUR_BPRV_WASDEL);
 622                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 623                                         idx, &cur, new, &da_new, first, flist,
 624                                         &logflags, delta, rsvd)))
 625                                         goto done;
 626                         } else if (new->br_state == XFS_EXT_NORM) {
 627                                 ASSERT(new->br_state == XFS_EXT_NORM);
 628                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 629                                         ip, idx, &cur, new, &logflags, delta)))
 630                                         goto done;
 631                         } else {
 632                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 633                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 634                                         ip, idx, &cur, new, &logflags, delta)))
 635                                         goto done;
 636                         }
 637                         ASSERT(*curp == cur || *curp == NULL);
 638                 }
 639                 /*
 640                  * Otherwise we're filling in a hole with an allocation.
 641                  */
 642                 else {
 643                         if (cur)
 644                                 ASSERT((cur->bc_private.b.flags &
 645                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 646                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 647                                         new, &logflags, delta, whichfork)))
 648                                 goto done;
 649                 }
 650         }
 651
 652         ASSERT(*curp == cur || *curp == NULL);
 653         /*
 654          * Convert to a btree if necessary.
 655          */
 656         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 657             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 658                 int     tmp_logflags;   /* partial log flag return val */
 659
 660                 ASSERT(cur == NULL);
 661                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 662                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 663                 logflags |= tmp_logflags;
 664                 if (error)
 665                         goto done;
 666         }
 667         /*
 668          * Adjust for changes in reserved delayed indirect blocks.
 669          * Nothing to do for disk quotas here.
 670          */
 671         if (da_old || da_new) {
 672                 xfs_filblks_t   nblks;
 673
 674                 nblks = da_new;
 675                 if (cur)
 676                         nblks += cur->bc_private.b.allocated;
 677                 ASSERT(nblks <= da_old);
 678                 if (nblks < da_old)
 679                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 680                                 (int64_t)(da_old - nblks), rsvd);
 681         }
 682         /*
 683          * Clear out the allocated field, done with it now in any case.
 684          */
 685         if (cur) {
 686                 cur->bc_private.b.allocated = 0;
 687                 *curp = cur;
 688         }
 689 done:
 690 #ifdef DEBUG
 691         if (!error)
 692                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 693 #endif
 694         *logflagsp = logflags;
 695         return error;
 696 }
 697
 698 /*
 699  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 700  * allocation to a real allocation.
 701  */
 702 STATIC int                              /* error */
 703 xfs_bmap_add_extent_delay_real(
 704         xfs_inode_t             *ip,    /* incore inode pointer */
 705         xfs_extnum_t            idx,    /* extent number to update/insert */
 706         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 707         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 708         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 709         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 710         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 711         int                     *logflagsp, /* inode logging flags */
 712         xfs_extdelta_t          *delta, /* Change made to incore extents */
 713         int                     rsvd)   /* OK to use reserved data block allocation */
 714 {
 715         xfs_btree_cur_t         *cur;   /* btree cursor */
 716         int                     diff;   /* temp value */
 717         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
 718         int                     error;  /* error return value */
 719         int                     i;      /* temp state */
 720         xfs_ifork_t             *ifp;   /* inode fork pointer */
 721         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 722         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 723                                         /* left is 0, right is 1, prev is 2 */
 724         int                     rval=0; /* return value (logging flags) */
 725         int                     state = 0;/* state bits, accessed thru macros */
 726         xfs_filblks_t           temp=0; /* value for dnew calculations */
 727         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 728         int                     tmp_rval;       /* partial logging flags */
 729         enum {                          /* bit number definitions for state */
 730                 LEFT_CONTIG,    RIGHT_CONTIG,
 731                 LEFT_FILLING,   RIGHT_FILLING,
 732                 LEFT_DELAY,     RIGHT_DELAY,
 733                 LEFT_VALID,     RIGHT_VALID
 734         };
 735
 736 #define LEFT            r[0]
 737 #define RIGHT           r[1]
 738 #define PREV            r[2]
 739 #define MASK(b)         (1 << (b))
 740 #define MASK2(a,b)      (MASK(a) | MASK(b))
 741 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
 742 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
 743 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
 744 #define STATE_TEST(b)   (state & MASK(b))
 745 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
 746                                        ((state &= ~MASK(b)), 0))
 747 #define SWITCH_STATE            \
 748         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
 749
 750         /*
 751          * Set up a bunch of variables to make the tests simpler.
 752          */
 753         cur = *curp;
 754         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 755         ep = xfs_iext_get_ext(ifp, idx);
 756         xfs_bmbt_get_all(ep, &PREV);
 757         new_endoff = new->br_startoff + new->br_blockcount;
 758         ASSERT(PREV.br_startoff <= new->br_startoff);
 759         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 760         /*
 761          * Set flags determining what part of the previous delayed allocation
 762          * extent is being replaced by a real allocation.
 763          */
 764         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
 765         STATE_SET(RIGHT_FILLING,
 766                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
 767         /*
 768          * Check and set flags if this segment has a left neighbor.
 769          * Don't set contiguous if the combined extent would be too large.
 770          */
 771         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
 772                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
 773                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
 774         }
 775         STATE_SET(LEFT_CONTIG,
 776                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
 777                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 778                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 779                 LEFT.br_state == new->br_state &&
 780                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
 781         /*
 782          * Check and set flags if this segment has a right neighbor.
 783          * Don't set contiguous if the combined extent would be too large.
 784          * Also check for all-three-contiguous being too large.
 785          */
 786         if (STATE_SET_TEST(RIGHT_VALID,
 787                         idx <
 788                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
 789                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
 790                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
 791         }
 792         STATE_SET(RIGHT_CONTIG,
 793                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
 794                 new_endoff == RIGHT.br_startoff &&
 795                 new->br_startblock + new->br_blockcount ==
 796                     RIGHT.br_startblock &&
 797                 new->br_state == RIGHT.br_state &&
 798                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 799                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
 800                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
 801                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 802                      <= MAXEXTLEN));
 803         error = 0;
 804         /*
 805          * Switch out based on the FILLING and CONTIG state bits.
 806          */
 807         switch (SWITCH_STATE) {
 808
 809         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
 810                 /*
 811                  * Filling in all of a previously delayed allocation extent.
 812                  * The left and right neighbors are both contiguous with new.
 813                  */
 814                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 815                         XFS_DATA_FORK);
 816                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 817                         LEFT.br_blockcount + PREV.br_blockcount +
 818                         RIGHT.br_blockcount);
 819                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
 820                         XFS_DATA_FORK);
 821                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
 822                 xfs_iext_remove(ifp, idx, 2);
 823                 ip->i_df.if_lastex = idx - 1;
 824                 ip->i_d.di_nextents--;
 825                 if (cur == NULL)
 826                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 827                 else {
 828                         rval = XFS_ILOG_CORE;
 829                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 830                                         RIGHT.br_startblock,
 831                                         RIGHT.br_blockcount, &i)))
 832                                 goto done;
 833                         ASSERT(i == 1);
 834                         if ((error = xfs_bmbt_delete(cur, &i)))
 835                                 goto done;
 836                         ASSERT(i == 1);
 837                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
 838                                 goto done;
 839                         ASSERT(i == 1);
 840                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 841                                         LEFT.br_startblock,
 842                                         LEFT.br_blockcount +
 843                                         PREV.br_blockcount +
 844                                         RIGHT.br_blockcount, LEFT.br_state)))
 845                                 goto done;
 846                 }
 847                 *dnew = 0;
 848                 /* DELTA: Three in-core extents are replaced by one. */
 849                 temp = LEFT.br_startoff;
 850                 temp2 = LEFT.br_blockcount +
 851                         PREV.br_blockcount +
 852                         RIGHT.br_blockcount;
 853                 break;
 854
 855         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
 856                 /*
 857                  * Filling in all of a previously delayed allocation extent.
 858                  * The left neighbor is contiguous, the right is not.
 859                  */
 860                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
 861                         XFS_DATA_FORK);
 862                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 863                         LEFT.br_blockcount + PREV.br_blockcount);
 864                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
 865                         XFS_DATA_FORK);
 866                 ip->i_df.if_lastex = idx - 1;
 867                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
 868                 xfs_iext_remove(ifp, idx, 1);
 869                 if (cur == NULL)
 870                         rval = XFS_ILOG_DEXT;
 871                 else {
 872                         rval = 0;
 873                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 874                                         LEFT.br_startblock, LEFT.br_blockcount,
 875                                         &i)))
 876                                 goto done;
 877                         ASSERT(i == 1);
 878                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 879                                         LEFT.br_startblock,
 880                                         LEFT.br_blockcount +
 881                                         PREV.br_blockcount, LEFT.br_state)))
 882                                 goto done;
 883                 }
 884                 *dnew = 0;
 885                 /* DELTA: Two in-core extents are replaced by one. */
 886                 temp = LEFT.br_startoff;
 887                 temp2 = LEFT.br_blockcount +
 888                         PREV.br_blockcount;
 889                 break;
 890
 891         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
 892                 /*
 893                  * Filling in all of a previously delayed allocation extent.
 894                  * The right neighbor is contiguous, the left is not.
 895                  */
 896                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 897                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 898                 xfs_bmbt_set_blockcount(ep,
 899                         PREV.br_blockcount + RIGHT.br_blockcount);
 900                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
 901                 ip->i_df.if_lastex = idx;
 902                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
 903                 xfs_iext_remove(ifp, idx + 1, 1);
 904                 if (cur == NULL)
 905                         rval = XFS_ILOG_DEXT;
 906                 else {
 907                         rval = 0;
 908                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 909                                         RIGHT.br_startblock,
 910                                         RIGHT.br_blockcount, &i)))
 911                                 goto done;
 912                         ASSERT(i == 1);
 913                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 914                                         new->br_startblock,
 915                                         PREV.br_blockcount +
 916                                         RIGHT.br_blockcount, PREV.br_state)))
 917                                 goto done;
 918                 }
 919                 *dnew = 0;
 920                 /* DELTA: Two in-core extents are replaced by one. */
 921                 temp = PREV.br_startoff;
 922                 temp2 = PREV.br_blockcount +
 923                         RIGHT.br_blockcount;
 924                 break;
 925
 926         case MASK2(LEFT_FILLING, RIGHT_FILLING):
 927                 /*
 928                  * Filling in all of a previously delayed allocation extent.
 929                  * Neither the left nor right neighbors are contiguous with
 930                  * the new one.
 931                  */
 932                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 933                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 934                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
 935                 ip->i_df.if_lastex = idx;
 936                 ip->i_d.di_nextents++;
 937                 if (cur == NULL)
 938                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 939                 else {
 940                         rval = XFS_ILOG_CORE;
 941                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 942                                         new->br_startblock, new->br_blockcount,
 943                                         &i)))
 944                                 goto done;
 945                         ASSERT(i == 0);
 946                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 947                         if ((error = xfs_bmbt_insert(cur, &i)))
 948                                 goto done;
 949                         ASSERT(i == 1);
 950                 }
 951                 *dnew = 0;
 952                 /* DELTA: The in-core extent described by new changed type. */
 953                 temp = new->br_startoff;
 954                 temp2 = new->br_blockcount;
 955                 break;
 956
 957         case MASK2(LEFT_FILLING, LEFT_CONTIG):
 958                 /*
 959                  * Filling in the first part of a previous delayed allocation.
 960                  * The left neighbor is contiguous.
 961                  */
 962                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 963                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 964                         LEFT.br_blockcount + new->br_blockcount);
 965                 xfs_bmbt_set_startoff(ep,
 966                         PREV.br_startoff + new->br_blockcount);
 967                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
 968                 temp = PREV.br_blockcount - new->br_blockcount;
 969                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 970                 xfs_bmbt_set_blockcount(ep, temp);
 971                 ip->i_df.if_lastex = idx - 1;
 972                 if (cur == NULL)
 973                         rval = XFS_ILOG_DEXT;
 974                 else {
 975                         rval = 0;
 976                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 977                                         LEFT.br_startblock, LEFT.br_blockcount,
 978                                         &i)))
 979                                 goto done;
 980                         ASSERT(i == 1);
 981                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 982                                         LEFT.br_startblock,
 983                                         LEFT.br_blockcount +
 984                                         new->br_blockcount,
 985                                         LEFT.br_state)))
 986                                 goto done;
 987                 }
 988                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 989                         STARTBLOCKVAL(PREV.br_startblock));
 990                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
 991                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
 992                 *dnew = temp;
 993                 /* DELTA: The boundary between two in-core extents moved. */
 994                 temp = LEFT.br_startoff;
 995                 temp2 = LEFT.br_blockcount +
 996                         PREV.br_blockcount;
 997                 break;
 998
 999         case MASK(LEFT_FILLING):
1000                 /*
1001                  * Filling in the first part of a previous delayed allocation.
1002                  * The left neighbor is not contiguous.
1003                  */
1004                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1005                 xfs_bmbt_set_startoff(ep, new_endoff);
1006                 temp = PREV.br_blockcount - new->br_blockcount;
1007                 xfs_bmbt_set_blockcount(ep, temp);
1008                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1009                         XFS_DATA_FORK);
1010                 xfs_iext_insert(ifp, idx, 1, new);
1011                 ip->i_df.if_lastex = idx;
1012                 ip->i_d.di_nextents++;
1013                 if (cur == NULL)
1014                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1015                 else {
1016                         rval = XFS_ILOG_CORE;
1017                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1018                                         new->br_startblock, new->br_blockcount,
1019                                         &i)))
1020                                 goto done;
1021                         ASSERT(i == 0);
1022                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1023                         if ((error = xfs_bmbt_insert(cur, &i)))
1024                                 goto done;
1025                         ASSERT(i == 1);
1026                 }
1027                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1028                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1029                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1030                                         first, flist, &cur, 1, &tmp_rval,
1031                                         XFS_DATA_FORK);
1032                         rval |= tmp_rval;
1033                         if (error)
1034                                 goto done;
1035                 }
1036                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1037                         STARTBLOCKVAL(PREV.br_startblock) -
1038                         (cur ? cur->bc_private.b.allocated : 0));
1039                 ep = xfs_iext_get_ext(ifp, idx + 1);
1040                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1041                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx + 1, XFS_DATA_FORK);
1042                 *dnew = temp;
1043                 /* DELTA: One in-core extent is split in two. */
1044                 temp = PREV.br_startoff;
1045                 temp2 = PREV.br_blockcount;
1046                 break;
1047
1048         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1049                 /*
1050                  * Filling in the last part of a previous delayed allocation.
1051                  * The right neighbor is contiguous with the new allocation.
1052                  */
1053                 temp = PREV.br_blockcount - new->br_blockcount;
1054                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1055                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1056                 xfs_bmbt_set_blockcount(ep, temp);
1057                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1058                         new->br_startoff, new->br_startblock,
1059                         new->br_blockcount + RIGHT.br_blockcount,
1060                         RIGHT.br_state);
1061                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1062                 ip->i_df.if_lastex = idx + 1;
1063                 if (cur == NULL)
1064                         rval = XFS_ILOG_DEXT;
1065                 else {
1066                         rval = 0;
1067                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1068                                         RIGHT.br_startblock,
1069                                         RIGHT.br_blockcount, &i)))
1070                                 goto done;
1071                         ASSERT(i == 1);
1072                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1073                                         new->br_startblock,
1074                                         new->br_blockcount +
1075                                         RIGHT.br_blockcount,
1076                                         RIGHT.br_state)))
1077                                 goto done;
1078                 }
1079                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1080                         STARTBLOCKVAL(PREV.br_startblock));
1081                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1082                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1083                 *dnew = temp;
1084                 /* DELTA: The boundary between two in-core extents moved. */
1085                 temp = PREV.br_startoff;
1086                 temp2 = PREV.br_blockcount +
1087                         RIGHT.br_blockcount;
1088                 break;
1089
1090         case MASK(RIGHT_FILLING):
1091                 /*
1092                  * Filling in the last part of a previous delayed allocation.
1093                  * The right neighbor is not contiguous.
1094                  */
1095                 temp = PREV.br_blockcount - new->br_blockcount;
1096                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1097                 xfs_bmbt_set_blockcount(ep, temp);
1098                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1099                         XFS_DATA_FORK);
1100                 xfs_iext_insert(ifp, idx + 1, 1, new);
1101                 ip->i_df.if_lastex = idx + 1;
1102                 ip->i_d.di_nextents++;
1103                 if (cur == NULL)
1104                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1105                 else {
1106                         rval = XFS_ILOG_CORE;
1107                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1108                                         new->br_startblock, new->br_blockcount,
1109                                         &i)))
1110                                 goto done;
1111                         ASSERT(i == 0);
1112                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1113                         if ((error = xfs_bmbt_insert(cur, &i)))
1114                                 goto done;
1115                         ASSERT(i == 1);
1116                 }
1117                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1118                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1119                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1120                                 first, flist, &cur, 1, &tmp_rval,
1121                                 XFS_DATA_FORK);
1122                         rval |= tmp_rval;
1123                         if (error)
1124                                 goto done;
1125                 }
1126                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1127                         STARTBLOCKVAL(PREV.br_startblock) -
1128                         (cur ? cur->bc_private.b.allocated : 0));
1129                 ep = xfs_iext_get_ext(ifp, idx);
1130                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1131                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1132                 *dnew = temp;
1133                 /* DELTA: One in-core extent is split in two. */
1134                 temp = PREV.br_startoff;
1135                 temp2 = PREV.br_blockcount;
1136                 break;
1137
1138         case 0:
1139                 /*
1140                  * Filling in the middle part of a previous delayed allocation.
1141                  * Contiguity is impossible here.
1142                  * This case is avoided almost all the time.
1143                  */
1144                 temp = new->br_startoff - PREV.br_startoff;
1145                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1146                 xfs_bmbt_set_blockcount(ep, temp);
1147                 r[0] = *new;
1148                 r[1].br_state = PREV.br_state;
1149                 r[1].br_startblock = 0;
1150                 r[1].br_startoff = new_endoff;
1151                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1152                 r[1].br_blockcount = temp2;
1153                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1154                         XFS_DATA_FORK);
1155                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1156                 ip->i_df.if_lastex = idx + 1;
1157                 ip->i_d.di_nextents++;
1158                 if (cur == NULL)
1159                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1160                 else {
1161                         rval = XFS_ILOG_CORE;
1162                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1163                                         new->br_startblock, new->br_blockcount,
1164                                         &i)))
1165                                 goto done;
1166                         ASSERT(i == 0);
1167                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1168                         if ((error = xfs_bmbt_insert(cur, &i)))
1169                                 goto done;
1170                         ASSERT(i == 1);
1171                 }
1172                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1173                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1174                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1175                                         first, flist, &cur, 1, &tmp_rval,
1176                                         XFS_DATA_FORK);
1177                         rval |= tmp_rval;
1178                         if (error)
1179                                 goto done;
1180                 }
1181                 temp = xfs_bmap_worst_indlen(ip, temp);
1182                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1183                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1184                         (cur ? cur->bc_private.b.allocated : 0));
1185                 if (diff > 0 &&
1186                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1187                         /*
1188                          * Ick gross gag me with a spoon.
1189                          */
1190                         ASSERT(0);      /* want to see if this ever happens! */
1191                         while (diff > 0) {
1192                                 if (temp) {
1193                                         temp--;
1194                                         diff--;
1195                                         if (!diff ||
1196                                             !xfs_mod_incore_sb(ip->i_mount,
1197                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1198                                                 break;
1199                                 }
1200                                 if (temp2) {
1201                                         temp2--;
1202                                         diff--;
1203                                         if (!diff ||
1204                                             !xfs_mod_incore_sb(ip->i_mount,
1205                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1206                                                 break;
1207                                 }
1208                         }
1209                 }
1210                 ep = xfs_iext_get_ext(ifp, idx);
1211                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1212                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1213                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1214                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1215                         NULLSTARTBLOCK((int)temp2));
1216                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1217                 *dnew = temp + temp2;
1218                 /* DELTA: One in-core extent is split in three. */
1219                 temp = PREV.br_startoff;
1220                 temp2 = PREV.br_blockcount;
1221                 break;
1222
1223         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1224         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1225         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1226         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1227         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1228         case MASK(LEFT_CONTIG):
1229         case MASK(RIGHT_CONTIG):
1230                 /*
1231                  * These cases are all impossible.
1232                  */
1233                 ASSERT(0);
1234         }
1235         *curp = cur;
1236         if (delta) {
1237                 temp2 += temp;
1238                 if (delta->xed_startoff > temp)
1239                         delta->xed_startoff = temp;
1240                 if (delta->xed_blockcount < temp2)
1241                         delta->xed_blockcount = temp2;
1242         }
1243 done:
1244         *logflagsp = rval;
1245         return error;
1246 #undef  LEFT
1247 #undef  RIGHT
1248 #undef  PREV
1249 #undef  MASK
1250 #undef  MASK2
1251 #undef  MASK3
1252 #undef  MASK4
1253 #undef  STATE_SET
1254 #undef  STATE_TEST
1255 #undef  STATE_SET_TEST
1256 #undef  SWITCH_STATE
1257 }
1258
1259 /*
1260  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1261  * allocation to a real allocation or vice versa.
1262  */
1263 STATIC int                              /* error */
1264 xfs_bmap_add_extent_unwritten_real(
1265         xfs_inode_t             *ip,    /* incore inode pointer */
1266         xfs_extnum_t            idx,    /* extent number to update/insert */
1267         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1268         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1269         int                     *logflagsp, /* inode logging flags */
1270         xfs_extdelta_t          *delta) /* Change made to incore extents */
1271 {
1272         xfs_btree_cur_t         *cur;   /* btree cursor */
1273         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1274         int                     error;  /* error return value */
1275         int                     i;      /* temp state */
1276         xfs_ifork_t             *ifp;   /* inode fork pointer */
1277         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1278         xfs_exntst_t            newext; /* new extent state */
1279         xfs_exntst_t            oldext; /* old extent state */
1280         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1281                                         /* left is 0, right is 1, prev is 2 */
1282         int                     rval=0; /* return value (logging flags) */
1283         int                     state = 0;/* state bits, accessed thru macros */
1284         xfs_filblks_t           temp=0;
1285         xfs_filblks_t           temp2=0;
1286         enum {                          /* bit number definitions for state */
1287                 LEFT_CONTIG,    RIGHT_CONTIG,
1288                 LEFT_FILLING,   RIGHT_FILLING,
1289                 LEFT_DELAY,     RIGHT_DELAY,
1290                 LEFT_VALID,     RIGHT_VALID
1291         };
1292
1293 #define LEFT            r[0]
1294 #define RIGHT           r[1]
1295 #define PREV            r[2]
1296 #define MASK(b)         (1 << (b))
1297 #define MASK2(a,b)      (MASK(a) | MASK(b))
1298 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1299 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1300 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1301 #define STATE_TEST(b)   (state & MASK(b))
1302 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1303                                        ((state &= ~MASK(b)), 0))
1304 #define SWITCH_STATE            \
1305         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1306
1307         /*
1308          * Set up a bunch of variables to make the tests simpler.
1309          */
1310         error = 0;
1311         cur = *curp;
1312         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1313         ep = xfs_iext_get_ext(ifp, idx);
1314         xfs_bmbt_get_all(ep, &PREV);
1315         newext = new->br_state;
1316         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1317                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1318         ASSERT(PREV.br_state == oldext);
1319         new_endoff = new->br_startoff + new->br_blockcount;
1320         ASSERT(PREV.br_startoff <= new->br_startoff);
1321         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1322         /*
1323          * Set flags determining what part of the previous oldext allocation
1324          * extent is being replaced by a newext allocation.
1325          */
1326         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1327         STATE_SET(RIGHT_FILLING,
1328                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1329         /*
1330          * Check and set flags if this segment has a left neighbor.
1331          * Don't set contiguous if the combined extent would be too large.
1332          */
1333         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1334                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1335                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1336         }
1337         STATE_SET(LEFT_CONTIG,
1338                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1339                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1340                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1341                 LEFT.br_state == newext &&
1342                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1343         /*
1344          * Check and set flags if this segment has a right neighbor.
1345          * Don't set contiguous if the combined extent would be too large.
1346          * Also check for all-three-contiguous being too large.
1347          */
1348         if (STATE_SET_TEST(RIGHT_VALID,
1349                         idx <
1350                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1351                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1352                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1353         }
1354         STATE_SET(RIGHT_CONTIG,
1355                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1356                 new_endoff == RIGHT.br_startoff &&
1357                 new->br_startblock + new->br_blockcount ==
1358                     RIGHT.br_startblock &&
1359                 newext == RIGHT.br_state &&
1360                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1361                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1362                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1363                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1364                      <= MAXEXTLEN));
1365         /*
1366          * Switch out based on the FILLING and CONTIG state bits.
1367          */
1368         switch (SWITCH_STATE) {
1369
1370         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1371                 /*
1372                  * Setting all of a previous oldext extent to newext.
1373                  * The left and right neighbors are both contiguous with new.
1374                  */
1375                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1376                         XFS_DATA_FORK);
1377                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1378                         LEFT.br_blockcount + PREV.br_blockcount +
1379                         RIGHT.br_blockcount);
1380                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1381                         XFS_DATA_FORK);
1382                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
1383                 xfs_iext_remove(ifp, idx, 2);
1384                 ip->i_df.if_lastex = idx - 1;
1385                 ip->i_d.di_nextents -= 2;
1386                 if (cur == NULL)
1387                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1388                 else {
1389                         rval = XFS_ILOG_CORE;
1390                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1391                                         RIGHT.br_startblock,
1392                                         RIGHT.br_blockcount, &i)))
1393                                 goto done;
1394                         ASSERT(i == 1);
1395                         if ((error = xfs_bmbt_delete(cur, &i)))
1396                                 goto done;
1397                         ASSERT(i == 1);
1398                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1399                                 goto done;
1400                         ASSERT(i == 1);
1401                         if ((error = xfs_bmbt_delete(cur, &i)))
1402                                 goto done;
1403                         ASSERT(i == 1);
1404                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1405                                 goto done;
1406                         ASSERT(i == 1);
1407                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1408                                 LEFT.br_startblock,
1409                                 LEFT.br_blockcount + PREV.br_blockcount +
1410                                 RIGHT.br_blockcount, LEFT.br_state)))
1411                                 goto done;
1412                 }
1413                 /* DELTA: Three in-core extents are replaced by one. */
1414                 temp = LEFT.br_startoff;
1415                 temp2 = LEFT.br_blockcount +
1416                         PREV.br_blockcount +
1417                         RIGHT.br_blockcount;
1418                 break;
1419
1420         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1421                 /*
1422                  * Setting all of a previous oldext extent to newext.
1423                  * The left neighbor is contiguous, the right is not.
1424                  */
1425                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
1426                         XFS_DATA_FORK);
1427                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1428                         LEFT.br_blockcount + PREV.br_blockcount);
1429                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
1430                         XFS_DATA_FORK);
1431                 ip->i_df.if_lastex = idx - 1;
1432                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
1433                 xfs_iext_remove(ifp, idx, 1);
1434                 ip->i_d.di_nextents--;
1435                 if (cur == NULL)
1436                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1437                 else {
1438                         rval = XFS_ILOG_CORE;
1439                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1440                                         PREV.br_startblock, PREV.br_blockcount,
1441                                         &i)))
1442                                 goto done;
1443                         ASSERT(i == 1);
1444                         if ((error = xfs_bmbt_delete(cur, &i)))
1445                                 goto done;
1446                         ASSERT(i == 1);
1447                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1448                                 goto done;
1449                         ASSERT(i == 1);
1450                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1451                                 LEFT.br_startblock,
1452                                 LEFT.br_blockcount + PREV.br_blockcount,
1453                                 LEFT.br_state)))
1454                                 goto done;
1455                 }
1456                 /* DELTA: Two in-core extents are replaced by one. */
1457                 temp = LEFT.br_startoff;
1458                 temp2 = LEFT.br_blockcount +
1459                         PREV.br_blockcount;
1460                 break;
1461
1462         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1463                 /*
1464                  * Setting all of a previous oldext extent to newext.
1465                  * The right neighbor is contiguous, the left is not.
1466                  */
1467                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx,
1468                         XFS_DATA_FORK);
1469                 xfs_bmbt_set_blockcount(ep,
1470                         PREV.br_blockcount + RIGHT.br_blockcount);
1471                 xfs_bmbt_set_state(ep, newext);
1472                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx,
1473                         XFS_DATA_FORK);
1474                 ip->i_df.if_lastex = idx;
1475                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
1476                 xfs_iext_remove(ifp, idx + 1, 1);
1477                 ip->i_d.di_nextents--;
1478                 if (cur == NULL)
1479                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1480                 else {
1481                         rval = XFS_ILOG_CORE;
1482                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1483                                         RIGHT.br_startblock,
1484                                         RIGHT.br_blockcount, &i)))
1485                                 goto done;
1486                         ASSERT(i == 1);
1487                         if ((error = xfs_bmbt_delete(cur, &i)))
1488                                 goto done;
1489                         ASSERT(i == 1);
1490                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1491                                 goto done;
1492                         ASSERT(i == 1);
1493                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1494                                 new->br_startblock,
1495                                 new->br_blockcount + RIGHT.br_blockcount,
1496                                 newext)))
1497                                 goto done;
1498                 }
1499                 /* DELTA: Two in-core extents are replaced by one. */
1500                 temp = PREV.br_startoff;
1501                 temp2 = PREV.br_blockcount +
1502                         RIGHT.br_blockcount;
1503                 break;
1504
1505         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1506                 /*
1507                  * Setting all of a previous oldext extent to newext.
1508                  * Neither the left nor right neighbors are contiguous with
1509                  * the new one.
1510                  */
1511                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx,
1512                         XFS_DATA_FORK);
1513                 xfs_bmbt_set_state(ep, newext);
1514                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx,
1515                         XFS_DATA_FORK);
1516                 ip->i_df.if_lastex = idx;
1517                 if (cur == NULL)
1518                         rval = XFS_ILOG_DEXT;
1519                 else {
1520                         rval = 0;
1521                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1522                                         new->br_startblock, new->br_blockcount,
1523                                         &i)))
1524                                 goto done;
1525                         ASSERT(i == 1);
1526                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1527                                 new->br_startblock, new->br_blockcount,
1528                                 newext)))
1529                                 goto done;
1530                 }
1531                 /* DELTA: The in-core extent described by new changed type. */
1532                 temp = new->br_startoff;
1533                 temp2 = new->br_blockcount;
1534                 break;
1535
1536         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1537                 /*
1538                  * Setting the first part of a previous oldext extent to newext.
1539                  * The left neighbor is contiguous.
1540                  */
1541                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1,
1542                         XFS_DATA_FORK);
1543                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1544                         LEFT.br_blockcount + new->br_blockcount);
1545                 xfs_bmbt_set_startoff(ep,
1546                         PREV.br_startoff + new->br_blockcount);
1547                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1,
1548                         XFS_DATA_FORK);
1549                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx,
1550                         XFS_DATA_FORK);
1551                 xfs_bmbt_set_startblock(ep,
1552                         new->br_startblock + new->br_blockcount);
1553                 xfs_bmbt_set_blockcount(ep,
1554                         PREV.br_blockcount - new->br_blockcount);
1555                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx,
1556                         XFS_DATA_FORK);
1557                 ip->i_df.if_lastex = idx - 1;
1558                 if (cur == NULL)
1559                         rval = XFS_ILOG_DEXT;
1560                 else {
1561                         rval = 0;
1562                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1563                                         PREV.br_startblock, PREV.br_blockcount,
1564                                         &i)))
1565                                 goto done;
1566                         ASSERT(i == 1);
1567                         if ((error = xfs_bmbt_update(cur,
1568                                 PREV.br_startoff + new->br_blockcount,
1569                                 PREV.br_startblock + new->br_blockcount,
1570                                 PREV.br_blockcount - new->br_blockcount,
1571                                 oldext)))
1572                                 goto done;
1573                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1574                                 goto done;
1575                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1576                                 LEFT.br_startblock,
1577                                 LEFT.br_blockcount + new->br_blockcount,
1578                                 LEFT.br_state))
1579                                 goto done;
1580                 }
1581                 /* DELTA: The boundary between two in-core extents moved. */
1582                 temp = LEFT.br_startoff;
1583                 temp2 = LEFT.br_blockcount +
1584                         PREV.br_blockcount;
1585                 break;
1586
1587         case MASK(LEFT_FILLING):
1588                 /*
1589                  * Setting the first part of a previous oldext extent to newext.
1590                  * The left neighbor is not contiguous.
1591                  */
1592                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1593                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1594                 xfs_bmbt_set_startoff(ep, new_endoff);
1595                 xfs_bmbt_set_blockcount(ep,
1596                         PREV.br_blockcount - new->br_blockcount);
1597                 xfs_bmbt_set_startblock(ep,
1598                         new->br_startblock + new->br_blockcount);
1599                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1600                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1601                         XFS_DATA_FORK);
1602                 xfs_iext_insert(ifp, idx, 1, new);
1603                 ip->i_df.if_lastex = idx;
1604                 ip->i_d.di_nextents++;
1605                 if (cur == NULL)
1606                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1607                 else {
1608                         rval = XFS_ILOG_CORE;
1609                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1610                                         PREV.br_startblock, PREV.br_blockcount,
1611                                         &i)))
1612                                 goto done;
1613                         ASSERT(i == 1);
1614                         if ((error = xfs_bmbt_update(cur,
1615                                 PREV.br_startoff + new->br_blockcount,
1616                                 PREV.br_startblock + new->br_blockcount,
1617                                 PREV.br_blockcount - new->br_blockcount,
1618                                 oldext)))
1619                                 goto done;
1620                         cur->bc_rec.b = *new;
1621                         if ((error = xfs_bmbt_insert(cur, &i)))
1622                                 goto done;
1623                         ASSERT(i == 1);
1624                 }
1625                 /* DELTA: One in-core extent is split in two. */
1626                 temp = PREV.br_startoff;
1627                 temp2 = PREV.br_blockcount;
1628                 break;
1629
1630         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1631                 /*
1632                  * Setting the last part of a previous oldext extent to newext.
1633                  * The right neighbor is contiguous with the new allocation.
1634                  */
1635                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx,
1636                         XFS_DATA_FORK);
1637                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1,
1638                         XFS_DATA_FORK);
1639                 xfs_bmbt_set_blockcount(ep,
1640                         PREV.br_blockcount - new->br_blockcount);
1641                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx,
1642                         XFS_DATA_FORK);
1643                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1644                         new->br_startoff, new->br_startblock,
1645                         new->br_blockcount + RIGHT.br_blockcount, newext);
1646                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1,
1647                         XFS_DATA_FORK);
1648                 ip->i_df.if_lastex = idx + 1;
1649                 if (cur == NULL)
1650                         rval = XFS_ILOG_DEXT;
1651                 else {
1652                         rval = 0;
1653                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1654                                         PREV.br_startblock,
1655                                         PREV.br_blockcount, &i)))
1656                                 goto done;
1657                         ASSERT(i == 1);
1658                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1659                                 PREV.br_startblock,
1660                                 PREV.br_blockcount - new->br_blockcount,
1661                                 oldext)))
1662                                 goto done;
1663                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1664                                 goto done;
1665                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1666                                 new->br_startblock,
1667                                 new->br_blockcount + RIGHT.br_blockcount,
1668                                 newext)))
1669                                 goto done;
1670                 }
1671                 /* DELTA: The boundary between two in-core extents moved. */
1672                 temp = PREV.br_startoff;
1673                 temp2 = PREV.br_blockcount +
1674                         RIGHT.br_blockcount;
1675                 break;
1676
1677         case MASK(RIGHT_FILLING):
1678                 /*
1679                  * Setting the last part of a previous oldext extent to newext.
1680                  * The right neighbor is not contiguous.
1681                  */
1682                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1683                 xfs_bmbt_set_blockcount(ep,
1684                         PREV.br_blockcount - new->br_blockcount);
1685                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1686                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1687                         XFS_DATA_FORK);
1688                 xfs_iext_insert(ifp, idx + 1, 1, new);
1689                 ip->i_df.if_lastex = idx + 1;
1690                 ip->i_d.di_nextents++;
1691                 if (cur == NULL)
1692                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1693                 else {
1694                         rval = XFS_ILOG_CORE;
1695                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1696                                         PREV.br_startblock, PREV.br_blockcount,
1697                                         &i)))
1698                                 goto done;
1699                         ASSERT(i == 1);
1700                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1701                                 PREV.br_startblock,
1702                                 PREV.br_blockcount - new->br_blockcount,
1703                                 oldext)))
1704                                 goto done;
1705                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1706                                         new->br_startblock, new->br_blockcount,
1707                                         &i)))
1708                                 goto done;
1709                         ASSERT(i == 0);
1710                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1711                         if ((error = xfs_bmbt_insert(cur, &i)))
1712                                 goto done;
1713                         ASSERT(i == 1);
1714                 }
1715                 /* DELTA: One in-core extent is split in two. */
1716                 temp = PREV.br_startoff;
1717                 temp2 = PREV.br_blockcount;
1718                 break;
1719
1720         case 0:
1721                 /*
1722                  * Setting the middle part of a previous oldext extent to
1723                  * newext.  Contiguity is impossible here.
1724                  * One extent becomes three extents.
1725                  */
1726                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1727                 xfs_bmbt_set_blockcount(ep,
1728                         new->br_startoff - PREV.br_startoff);
1729                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1730                 r[0] = *new;
1731                 r[1].br_startoff = new_endoff;
1732                 r[1].br_blockcount =
1733                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1734                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1735                 r[1].br_state = oldext;
1736                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1737                         XFS_DATA_FORK);
1738                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1739                 ip->i_df.if_lastex = idx + 1;
1740                 ip->i_d.di_nextents += 2;
1741                 if (cur == NULL)
1742                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1743                 else {
1744                         rval = XFS_ILOG_CORE;
1745                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1746                                         PREV.br_startblock, PREV.br_blockcount,
1747                                         &i)))
1748                                 goto done;
1749                         ASSERT(i == 1);
1750                         /* new right extent - oldext */
1751                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1752                                 r[1].br_startblock, r[1].br_blockcount,
1753                                 r[1].br_state)))
1754                                 goto done;
1755                         /* new left extent - oldext */
1756                         PREV.br_blockcount =
1757                                 new->br_startoff - PREV.br_startoff;
1758                         cur->bc_rec.b = PREV;
1759                         if ((error = xfs_bmbt_insert(cur, &i)))
1760                                 goto done;
1761                         ASSERT(i == 1);
1762                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1763                                 goto done;
1764                         ASSERT(i == 1);
1765                         /* new middle extent - newext */
1766                         cur->bc_rec.b = *new;
1767                         if ((error = xfs_bmbt_insert(cur, &i)))
1768                                 goto done;
1769                         ASSERT(i == 1);
1770                 }
1771                 /* DELTA: One in-core extent is split in three. */
1772                 temp = PREV.br_startoff;
1773                 temp2 = PREV.br_blockcount;
1774                 break;
1775
1776         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1777         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1778         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1779         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1780         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1781         case MASK(LEFT_CONTIG):
1782         case MASK(RIGHT_CONTIG):
1783                 /*
1784                  * These cases are all impossible.
1785                  */
1786                 ASSERT(0);
1787         }
1788         *curp = cur;
1789         if (delta) {
1790                 temp2 += temp;
1791                 if (delta->xed_startoff > temp)
1792                         delta->xed_startoff = temp;
1793                 if (delta->xed_blockcount < temp2)
1794                         delta->xed_blockcount = temp2;
1795         }
1796 done:
1797         *logflagsp = rval;
1798         return error;
1799 #undef  LEFT
1800 #undef  RIGHT
1801 #undef  PREV
1802 #undef  MASK
1803 #undef  MASK2
1804 #undef  MASK3
1805 #undef  MASK4
1806 #undef  STATE_SET
1807 #undef  STATE_TEST
1808 #undef  STATE_SET_TEST
1809 #undef  SWITCH_STATE
1810 }
1811
1812 /*
1813  * Called by xfs_bmap_add_extent to handle cases converting a hole
1814  * to a delayed allocation.
1815  */
1816 /*ARGSUSED*/
1817 STATIC int                              /* error */
1818 xfs_bmap_add_extent_hole_delay(
1819         xfs_inode_t             *ip,    /* incore inode pointer */
1820         xfs_extnum_t            idx,    /* extent number to update/insert */
1821         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1822         int                     *logflagsp, /* inode logging flags */
1823         xfs_extdelta_t          *delta, /* Change made to incore extents */
1824         int                     rsvd)           /* OK to allocate reserved blocks */
1825 {
1826         xfs_bmbt_rec_t          *ep;    /* extent record for idx */
1827         xfs_ifork_t             *ifp;   /* inode fork pointer */
1828         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1829         xfs_filblks_t           newlen=0;       /* new indirect size */
1830         xfs_filblks_t           oldlen=0;       /* old indirect size */
1831         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1832         int                     state;  /* state bits, accessed thru macros */
1833         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1834         xfs_filblks_t           temp2=0;
1835         enum {                          /* bit number definitions for state */
1836                 LEFT_CONTIG,    RIGHT_CONTIG,
1837                 LEFT_DELAY,     RIGHT_DELAY,
1838                 LEFT_VALID,     RIGHT_VALID
1839         };
1840
1841 #define MASK(b)                 (1 << (b))
1842 #define MASK2(a,b)              (MASK(a) | MASK(b))
1843 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1844 #define STATE_TEST(b)           (state & MASK(b))
1845 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1846                                        ((state &= ~MASK(b)), 0))
1847 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1848
1849         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1850         ep = xfs_iext_get_ext(ifp, idx);
1851         state = 0;
1852         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1853         /*
1854          * Check and set flags if this segment has a left neighbor
1855          */
1856         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1857                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1858                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1859         }
1860         /*
1861          * Check and set flags if the current (right) segment exists.
1862          * If it doesn't exist, we're converting the hole at end-of-file.
1863          */
1864         if (STATE_SET_TEST(RIGHT_VALID,
1865                            idx <
1866                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1867                 xfs_bmbt_get_all(ep, &right);
1868                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1869         }
1870         /*
1871          * Set contiguity flags on the left and right neighbors.
1872          * Don't let extents get too large, even if the pieces are contiguous.
1873          */
1874         STATE_SET(LEFT_CONTIG,
1875                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1876                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1877                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1878         STATE_SET(RIGHT_CONTIG,
1879                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1880                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1881                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1882                 (!STATE_TEST(LEFT_CONTIG) ||
1883                  (left.br_blockcount + new->br_blockcount +
1884                      right.br_blockcount <= MAXEXTLEN)));
1885         /*
1886          * Switch out based on the contiguity flags.
1887          */
1888         switch (SWITCH_STATE) {
1889
1890         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1891                 /*
1892                  * New allocation is contiguous with delayed allocations
1893                  * on the left and on the right.
1894                  * Merge all three into a single extent record.
1895                  */
1896                 temp = left.br_blockcount + new->br_blockcount +
1897                         right.br_blockcount;
1898                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
1899                         XFS_DATA_FORK);
1900                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1901                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1902                         STARTBLOCKVAL(new->br_startblock) +
1903                         STARTBLOCKVAL(right.br_startblock);
1904                 newlen = xfs_bmap_worst_indlen(ip, temp);
1905                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1906                         NULLSTARTBLOCK((int)newlen));
1907                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
1908                         XFS_DATA_FORK);
1909                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, XFS_DATA_FORK);
1910                 xfs_iext_remove(ifp, idx, 1);
1911                 ip->i_df.if_lastex = idx - 1;
1912                 /* DELTA: Two in-core extents were replaced by one. */
1913                 temp2 = temp;
1914                 temp = left.br_startoff;
1915                 break;
1916
1917         case MASK(LEFT_CONTIG):
1918                 /*
1919                  * New allocation is contiguous with a delayed allocation
1920                  * on the left.
1921                  * Merge the new allocation with the left neighbor.
1922                  */
1923                 temp = left.br_blockcount + new->br_blockcount;
1924                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1,
1925                         XFS_DATA_FORK);
1926                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1927                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1928                         STARTBLOCKVAL(new->br_startblock);
1929                 newlen = xfs_bmap_worst_indlen(ip, temp);
1930                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1931                         NULLSTARTBLOCK((int)newlen));
1932                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1,
1933                         XFS_DATA_FORK);
1934                 ip->i_df.if_lastex = idx - 1;
1935                 /* DELTA: One in-core extent grew into a hole. */
1936                 temp2 = temp;
1937                 temp = left.br_startoff;
1938                 break;
1939
1940         case MASK(RIGHT_CONTIG):
1941                 /*
1942                  * New allocation is contiguous with a delayed allocation
1943                  * on the right.
1944                  * Merge the new allocation with the right neighbor.
1945                  */
1946                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1947                 temp = new->br_blockcount + right.br_blockcount;
1948                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1949                         STARTBLOCKVAL(right.br_startblock);
1950                 newlen = xfs_bmap_worst_indlen(ip, temp);
1951                 xfs_bmbt_set_allf(ep, new->br_startoff,
1952                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1953                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1954                 ip->i_df.if_lastex = idx;
1955                 /* DELTA: One in-core extent grew into a hole. */
1956                 temp2 = temp;
1957                 temp = new->br_startoff;
1958                 break;
1959
1960         case 0:
1961                 /*
1962                  * New allocation is not contiguous with another
1963                  * delayed allocation.
1964                  * Insert a new entry.
1965                  */
1966                 oldlen = newlen = 0;
1967                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL,
1968                         XFS_DATA_FORK);
1969                 xfs_iext_insert(ifp, idx, 1, new);
1970                 ip->i_df.if_lastex = idx;
1971                 /* DELTA: A new in-core extent was added in a hole. */
1972                 temp2 = new->br_blockcount;
1973                 temp = new->br_startoff;
1974                 break;
1975         }
1976         if (oldlen != newlen) {
1977                 ASSERT(oldlen > newlen);
1978                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1979                         (int64_t)(oldlen - newlen), rsvd);
1980                 /*
1981                  * Nothing to do for disk quota accounting here.
1982                  */
1983         }
1984         if (delta) {
1985                 temp2 += temp;
1986                 if (delta->xed_startoff > temp)
1987                         delta->xed_startoff = temp;
1988                 if (delta->xed_blockcount < temp2)
1989                         delta->xed_blockcount = temp2;
1990         }
1991         *logflagsp = 0;
1992         return 0;
1993 #undef  MASK
1994 #undef  MASK2
1995 #undef  STATE_SET
1996 #undef  STATE_TEST
1997 #undef  STATE_SET_TEST
1998 #undef  SWITCH_STATE
1999 }
2000
2001 /*
2002  * Called by xfs_bmap_add_extent to handle cases converting a hole
2003  * to a real allocation.
2004  */
2005 STATIC int                              /* error */
2006 xfs_bmap_add_extent_hole_real(
2007         xfs_inode_t             *ip,    /* incore inode pointer */
2008         xfs_extnum_t            idx,    /* extent number to update/insert */
2009         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2010         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2011         int                     *logflagsp, /* inode logging flags */
2012         xfs_extdelta_t          *delta, /* Change made to incore extents */
2013         int                     whichfork) /* data or attr fork */
2014 {
2015         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
2016         int                     error;  /* error return value */
2017         int                     i;      /* temp state */
2018         xfs_ifork_t             *ifp;   /* inode fork pointer */
2019         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2020         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2021         int                     rval=0; /* return value (logging flags) */
2022         int                     state;  /* state bits, accessed thru macros */
2023         xfs_filblks_t           temp=0;
2024         xfs_filblks_t           temp2=0;
2025         enum {                          /* bit number definitions for state */
2026                 LEFT_CONTIG,    RIGHT_CONTIG,
2027                 LEFT_DELAY,     RIGHT_DELAY,
2028                 LEFT_VALID,     RIGHT_VALID
2029         };
2030
2031 #define MASK(b)                 (1 << (b))
2032 #define MASK2(a,b)              (MASK(a) | MASK(b))
2033 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2034 #define STATE_TEST(b)           (state & MASK(b))
2035 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2036                                        ((state &= ~MASK(b)), 0))
2037 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2038
2039         ifp = XFS_IFORK_PTR(ip, whichfork);
2040         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2041         ep = xfs_iext_get_ext(ifp, idx);
2042         state = 0;
2043         /*
2044          * Check and set flags if this segment has a left neighbor.
2045          */
2046         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2047                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2048                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2049         }
2050         /*
2051          * Check and set flags if this segment has a current value.
2052          * Not true if we're inserting into the "hole" at eof.
2053          */
2054         if (STATE_SET_TEST(RIGHT_VALID,
2055                            idx <
2056                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2057                 xfs_bmbt_get_all(ep, &right);
2058                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2059         }
2060         /*
2061          * We're inserting a real allocation between "left" and "right".
2062          * Set the contiguity flags.  Don't let extents get too large.
2063          */
2064         STATE_SET(LEFT_CONTIG,
2065                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2066                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2067                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2068                 left.br_state == new->br_state &&
2069                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2070         STATE_SET(RIGHT_CONTIG,
2071                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2072                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2073                 new->br_startblock + new->br_blockcount ==
2074                     right.br_startblock &&
2075                 new->br_state == right.br_state &&
2076                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2077                 (!STATE_TEST(LEFT_CONTIG) ||
2078                  left.br_blockcount + new->br_blockcount +
2079                      right.br_blockcount <= MAXEXTLEN));
2080
2081         error = 0;
2082         /*
2083          * Select which case we're in here, and implement it.
2084          */
2085         switch (SWITCH_STATE) {
2086
2087         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2088                 /*
2089                  * New allocation is contiguous with real allocations on the
2090                  * left and on the right.
2091                  * Merge all three into a single extent record.
2092                  */
2093                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
2094                         whichfork);
2095                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2096                         left.br_blockcount + new->br_blockcount +
2097                         right.br_blockcount);
2098                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
2099                         whichfork);
2100                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, whichfork);
2101                 xfs_iext_remove(ifp, idx, 1);
2102                 ifp->if_lastex = idx - 1;
2103                 XFS_IFORK_NEXT_SET(ip, whichfork,
2104                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2105                 if (cur == NULL) {
2106                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2107                 } else {
2108                         rval = XFS_ILOG_CORE;
2109                         if ((error = xfs_bmbt_lookup_eq(cur,
2110                                         right.br_startoff,
2111                                         right.br_startblock,
2112                                         right.br_blockcount, &i)))
2113                                 goto done;
2114                         ASSERT(i == 1);
2115                         if ((error = xfs_bmbt_delete(cur, &i)))
2116                                 goto done;
2117                         ASSERT(i == 1);
2118                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2119                                 goto done;
2120                         ASSERT(i == 1);
2121                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2122                                         left.br_startblock,
2123                                         left.br_blockcount +
2124                                                 new->br_blockcount +
2125                                                 right.br_blockcount,
2126                                         left.br_state)))
2127                                 goto done;
2128                 }
2129                 /* DELTA: Two in-core extents were replaced by one. */
2130                 temp = left.br_startoff;
2131                 temp2 = left.br_blockcount +
2132                         new->br_blockcount +
2133                         right.br_blockcount;
2134                 break;
2135
2136         case MASK(LEFT_CONTIG):
2137                 /*
2138                  * New allocation is contiguous with a real allocation
2139                  * on the left.
2140                  * Merge the new allocation with the left neighbor.
2141                  */
2142                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1, whichfork);
2143                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2144                         left.br_blockcount + new->br_blockcount);
2145                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1, whichfork);
2146                 ifp->if_lastex = idx - 1;
2147                 if (cur == NULL) {
2148                         rval = XFS_ILOG_FEXT(whichfork);
2149                 } else {
2150                         rval = 0;
2151                         if ((error = xfs_bmbt_lookup_eq(cur,
2152                                         left.br_startoff,
2153                                         left.br_startblock,
2154                                         left.br_blockcount, &i)))
2155                                 goto done;
2156                         ASSERT(i == 1);
2157                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2158                                         left.br_startblock,
2159                                         left.br_blockcount +
2160                                                 new->br_blockcount,
2161                                         left.br_state)))
2162                                 goto done;
2163                 }
2164                 /* DELTA: One in-core extent grew. */
2165                 temp = left.br_startoff;
2166                 temp2 = left.br_blockcount +
2167                         new->br_blockcount;
2168                 break;
2169
2170         case MASK(RIGHT_CONTIG):
2171                 /*
2172                  * New allocation is contiguous with a real allocation
2173                  * on the right.
2174                  * Merge the new allocation with the right neighbor.
2175                  */
2176                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, whichfork);
2177                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2178                         new->br_blockcount + right.br_blockcount,
2179                         right.br_state);
2180                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, whichfork);
2181                 ifp->if_lastex = idx;
2182                 if (cur == NULL) {
2183                         rval = XFS_ILOG_FEXT(whichfork);
2184                 } else {
2185                         rval = 0;
2186                         if ((error = xfs_bmbt_lookup_eq(cur,
2187                                         right.br_startoff,
2188                                         right.br_startblock,
2189                                         right.br_blockcount, &i)))
2190                                 goto done;
2191                         ASSERT(i == 1);
2192                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2193                                         new->br_startblock,
2194                                         new->br_blockcount +
2195                                                 right.br_blockcount,
2196                                         right.br_state)))
2197                                 goto done;
2198                 }
2199                 /* DELTA: One in-core extent grew. */
2200                 temp = new->br_startoff;
2201                 temp2 = new->br_blockcount +
2202                         right.br_blockcount;
2203                 break;
2204
2205         case 0:
2206                 /*
2207                  * New allocation is not contiguous with another
2208                  * real allocation.
2209                  * Insert a new entry.
2210                  */
2211                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL, whichfork);
2212                 xfs_iext_insert(ifp, idx, 1, new);
2213                 ifp->if_lastex = idx;
2214                 XFS_IFORK_NEXT_SET(ip, whichfork,
2215                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2216                 if (cur == NULL) {
2217                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2218                 } else {
2219                         rval = XFS_ILOG_CORE;
2220                         if ((error = xfs_bmbt_lookup_eq(cur,
2221                                         new->br_startoff,
2222                                         new->br_startblock,
2223                                         new->br_blockcount, &i)))
2224                                 goto done;
2225                         ASSERT(i == 0);
2226                         cur->bc_rec.b.br_state = new->br_state;
2227                         if ((error = xfs_bmbt_insert(cur, &i)))
2228                                 goto done;
2229                         ASSERT(i == 1);
2230                 }
2231                 /* DELTA: A new extent was added in a hole. */
2232                 temp = new->br_startoff;
2233                 temp2 = new->br_blockcount;
2234                 break;
2235         }
2236         if (delta) {
2237                 temp2 += temp;
2238                 if (delta->xed_startoff > temp)
2239                         delta->xed_startoff = temp;
2240                 if (delta->xed_blockcount < temp2)
2241                         delta->xed_blockcount = temp2;
2242         }
2243 done:
2244         *logflagsp = rval;
2245         return error;
2246 #undef  MASK
2247 #undef  MASK2
2248 #undef  STATE_SET
2249 #undef  STATE_TEST
2250 #undef  STATE_SET_TEST
2251 #undef  SWITCH_STATE
2252 }
2253
2254 /*
2255  * Adjust the size of the new extent based on di_extsize and rt extsize.
2256  */
2257 STATIC int
2258 xfs_bmap_extsize_align(
2259         xfs_mount_t     *mp,
2260         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2261         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2262         xfs_extlen_t    extsz,          /* align to this extent size */
2263         int             rt,             /* is this a realtime inode? */
2264         int             eof,            /* is extent at end-of-file? */
2265         int             delay,          /* creating delalloc extent? */
2266         int             convert,        /* overwriting unwritten extent? */
2267         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2268         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2269 {
2270         xfs_fileoff_t   orig_off;       /* original offset */
2271         xfs_extlen_t    orig_alen;      /* original length */
2272         xfs_fileoff_t   orig_end;       /* original off+len */
2273         xfs_fileoff_t   nexto;          /* next file offset */
2274         xfs_fileoff_t   prevo;          /* previous file offset */
2275         xfs_fileoff_t   align_off;      /* temp for offset */
2276         xfs_extlen_t    align_alen;     /* temp for length */
2277         xfs_extlen_t    temp;           /* temp for calculations */
2278
2279         if (convert)
2280                 return 0;
2281
2282         orig_off = align_off = *offp;
2283         orig_alen = align_alen = *lenp;
2284         orig_end = orig_off + orig_alen;
2285
2286         /*
2287          * If this request overlaps an existing extent, then don't
2288          * attempt to perform any additional alignment.
2289          */
2290         if (!delay && !eof &&
2291             (orig_off >= gotp->br_startoff) &&
2292             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2293                 return 0;
2294         }
2295
2296         /*
2297          * If the file offset is unaligned vs. the extent size
2298          * we need to align it.  This will be possible unless
2299          * the file was previously written with a kernel that didn't
2300          * perform this alignment, or if a truncate shot us in the
2301          * foot.
2302          */
2303         temp = do_mod(orig_off, extsz);
2304         if (temp) {
2305                 align_alen += temp;
2306                 align_off -= temp;
2307         }
2308         /*
2309          * Same adjustment for the end of the requested area.
2310          */
2311         if ((temp = (align_alen % extsz))) {
2312                 align_alen += extsz - temp;
2313         }
2314         /*
2315          * If the previous block overlaps with this proposed allocation
2316          * then move the start forward without adjusting the length.
2317          */
2318         if (prevp->br_startoff != NULLFILEOFF) {
2319                 if (prevp->br_startblock == HOLESTARTBLOCK)
2320                         prevo = prevp->br_startoff;
2321                 else
2322                         prevo = prevp->br_startoff + prevp->br_blockcount;
2323         } else
2324                 prevo = 0;
2325         if (align_off != orig_off && align_off < prevo)
2326                 align_off = prevo;
2327         /*
2328          * If the next block overlaps with this proposed allocation
2329          * then move the start back without adjusting the length,
2330          * but not before offset 0.
2331          * This may of course make the start overlap previous block,
2332          * and if we hit the offset 0 limit then the next block
2333          * can still overlap too.
2334          */
2335         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2336                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2337                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2338                         nexto = gotp->br_startoff + gotp->br_blockcount;
2339                 else
2340                         nexto = gotp->br_startoff;
2341         } else
2342                 nexto = NULLFILEOFF;
2343         if (!eof &&
2344             align_off + align_alen != orig_end &&
2345             align_off + align_alen > nexto)
2346                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2347         /*
2348          * If we're now overlapping the next or previous extent that
2349          * means we can't fit an extsz piece in this hole.  Just move
2350          * the start forward to the first valid spot and set
2351          * the length so we hit the end.
2352          */
2353         if (align_off != orig_off && align_off < prevo)
2354                 align_off = prevo;
2355         if (align_off + align_alen != orig_end &&
2356             align_off + align_alen > nexto &&
2357             nexto != NULLFILEOFF) {
2358                 ASSERT(nexto > prevo);
2359                 align_alen = nexto - align_off;
2360         }
2361
2362         /*
2363          * If realtime, and the result isn't a multiple of the realtime
2364          * extent size we need to remove blocks until it is.
2365          */
2366         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2367                 /*
2368                  * We're not covering the original request, or
2369                  * we won't be able to once we fix the length.
2370                  */
2371                 if (orig_off < align_off ||
2372                     orig_end > align_off + align_alen ||
2373                     align_alen - temp < orig_alen)
2374                         return XFS_ERROR(EINVAL);
2375                 /*
2376                  * Try to fix it by moving the start up.
2377                  */
2378                 if (align_off + temp <= orig_off) {
2379                         align_alen -= temp;
2380                         align_off += temp;
2381                 }
2382                 /*
2383                  * Try to fix it by moving the end in.
2384                  */
2385                 else if (align_off + align_alen - temp >= orig_end)
2386                         align_alen -= temp;
2387                 /*
2388                  * Set the start to the minimum then trim the length.
2389                  */
2390                 else {
2391                         align_alen -= orig_off - align_off;
2392                         align_off = orig_off;
2393                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2394                 }
2395                 /*
2396                  * Result doesn't cover the request, fail it.
2397                  */
2398                 if (orig_off < align_off || orig_end > align_off + align_alen)
2399                         return XFS_ERROR(EINVAL);
2400         } else {
2401                 ASSERT(orig_off >= align_off);
2402                 ASSERT(orig_end <= align_off + align_alen);
2403         }
2404
2405 #ifdef DEBUG
2406         if (!eof && gotp->br_startoff != NULLFILEOFF)
2407                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2408         if (prevp->br_startoff != NULLFILEOFF)
2409                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2410 #endif
2411
2412         *lenp = align_alen;
2413         *offp = align_off;
2414         return 0;
2415 }
2416
2417 #define XFS_ALLOC_GAP_UNITS     4
2418
2419 STATIC int
2420 xfs_bmap_adjacent(
2421         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2422 {
2423         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2424         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2425         xfs_mount_t     *mp;            /* mount point structure */
2426         int             nullfb;         /* true if ap->firstblock isn't set */
2427         int             rt;             /* true if inode is realtime */
2428
2429 #define ISVALID(x,y)    \
2430         (rt ? \
2431                 (x) < mp->m_sb.sb_rblocks : \
2432                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2433                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2434                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2435
2436         mp = ap->ip->i_mount;
2437         nullfb = ap->firstblock == NULLFSBLOCK;
2438         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2439         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2440         /*
2441          * If allocating at eof, and there's a previous real block,
2442          * try to use it's last block as our starting point.
2443          */
2444         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2445             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2446             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2447                     ap->prevp->br_startblock)) {
2448                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2449                 /*
2450                  * Adjust for the gap between prevp and us.
2451                  */
2452                 adjust = ap->off -
2453                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2454                 if (adjust &&
2455                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2456                         ap->rval += adjust;
2457         }
2458         /*
2459          * If not at eof, then compare the two neighbor blocks.
2460          * Figure out whether either one gives us a good starting point,
2461          * and pick the better one.
2462          */
2463         else if (!ap->eof) {
2464                 xfs_fsblock_t   gotbno;         /* right side block number */
2465                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2466                 xfs_fsblock_t   prevbno;        /* left side block number */
2467                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2468
2469                 /*
2470                  * If there's a previous (left) block, select a requested
2471                  * start block based on it.
2472                  */
2473                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2474                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2475                     (prevbno = ap->prevp->br_startblock +
2476                                ap->prevp->br_blockcount) &&
2477                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2478                         /*
2479                          * Calculate gap to end of previous block.
2480                          */
2481                         adjust = prevdiff = ap->off -
2482                                 (ap->prevp->br_startoff +
2483                                  ap->prevp->br_blockcount);
2484                         /*
2485                          * Figure the startblock based on the previous block's
2486                          * end and the gap size.
2487                          * Heuristic!
2488                          * If the gap is large relative to the piece we're
2489                          * allocating, or using it gives us an invalid block
2490                          * number, then just use the end of the previous block.
2491                          */
2492                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2493                             ISVALID(prevbno + prevdiff,
2494                                     ap->prevp->br_startblock))
2495                                 prevbno += adjust;
2496                         else
2497                                 prevdiff += adjust;
2498                         /*
2499                          * If the firstblock forbids it, can't use it,
2500                          * must use default.
2501                          */
2502                         if (!rt && !nullfb &&
2503                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2504                                 prevbno = NULLFSBLOCK;
2505                 }
2506                 /*
2507                  * No previous block or can't follow it, just default.
2508                  */
2509                 else
2510                         prevbno = NULLFSBLOCK;
2511                 /*
2512                  * If there's a following (right) block, select a requested
2513                  * start block based on it.
2514                  */
2515                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2516                         /*
2517                          * Calculate gap to start of next block.
2518                          */
2519                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2520                         /*
2521                          * Figure the startblock based on the next block's
2522                          * start and the gap size.
2523                          */
2524                         gotbno = ap->gotp->br_startblock;
2525                         /*
2526                          * Heuristic!
2527                          * If the gap is large relative to the piece we're
2528                          * allocating, or using it gives us an invalid block
2529                          * number, then just use the start of the next block
2530                          * offset by our length.
2531                          */
2532                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2533                             ISVALID(gotbno - gotdiff, gotbno))
2534                                 gotbno -= adjust;
2535                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2536                                 gotbno -= ap->alen;
2537                                 gotdiff += adjust - ap->alen;
2538                         } else
2539                                 gotdiff += adjust;
2540                         /*
2541                          * If the firstblock forbids it, can't use it,
2542                          * must use default.
2543                          */
2544                         if (!rt && !nullfb &&
2545                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2546                                 gotbno = NULLFSBLOCK;
2547                 }
2548                 /*
2549                  * No next block, just default.
2550                  */
2551                 else
2552                         gotbno = NULLFSBLOCK;
2553                 /*
2554                  * If both valid, pick the better one, else the only good
2555                  * one, else ap->rval is already set (to 0 or the inode block).
2556                  */
2557                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2558                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2559                 else if (prevbno != NULLFSBLOCK)
2560                         ap->rval = prevbno;
2561                 else if (gotbno != NULLFSBLOCK)
2562                         ap->rval = gotbno;
2563         }
2564 #undef ISVALID
2565         return 0;
2566 }
2567
2568 STATIC int
2569 xfs_bmap_rtalloc(
2570         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2571 {
2572         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2573         int             error;          /* error return value */
2574         xfs_mount_t     *mp;            /* mount point structure */
2575         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2576         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2577         xfs_extlen_t    align;          /* minimum allocation alignment */
2578         xfs_rtblock_t   rtb;
2579
2580         mp = ap->ip->i_mount;
2581         align = xfs_get_extsz_hint(ap->ip);
2582         prod = align / mp->m_sb.sb_rextsize;
2583         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2584                                         align, 1, ap->eof, 0,
2585                                         ap->conv, &ap->off, &ap->alen);
2586         if (error)
2587                 return error;
2588         ASSERT(ap->alen);
2589         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2590
2591         /*
2592          * If the offset & length are not perfectly aligned
2593          * then kill prod, it will just get us in trouble.
2594          */
2595         if (do_mod(ap->off, align) || ap->alen % align)
2596                 prod = 1;
2597         /*
2598          * Set ralen to be the actual requested length in rtextents.
2599          */
2600         ralen = ap->alen / mp->m_sb.sb_rextsize;
2601         /*
2602          * If the old value was close enough to MAXEXTLEN that
2603          * we rounded up to it, cut it back so it's valid again.
2604          * Note that if it's a really large request (bigger than
2605          * MAXEXTLEN), we don't hear about that number, and can't
2606          * adjust the starting point to match it.
2607          */
2608         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2609                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2610         /*
2611          * If it's an allocation to an empty file at offset 0,
2612          * pick an extent that will space things out in the rt area.
2613          */
2614         if (ap->eof && ap->off == 0) {
2615                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2616
2617                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2618                 if (error)
2619                         return error;
2620                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2621         } else {
2622                 ap->rval = 0;
2623         }
2624
2625         xfs_bmap_adjacent(ap);
2626
2627         /*
2628          * Realtime allocation, done through xfs_rtallocate_extent.
2629          */
2630         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2631         do_div(ap->rval, mp->m_sb.sb_rextsize);
2632         rtb = ap->rval;
2633         ap->alen = ralen;
2634         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2635                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2636                 return error;
2637         if (rtb == NULLFSBLOCK && prod > 1 &&
2638             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2639                                            ap->alen, &ralen, atype,
2640                                            ap->wasdel, 1, &rtb)))
2641                 return error;
2642         ap->rval = rtb;
2643         if (ap->rval != NULLFSBLOCK) {
2644                 ap->rval *= mp->m_sb.sb_rextsize;
2645                 ralen *= mp->m_sb.sb_rextsize;
2646                 ap->alen = ralen;
2647                 ap->ip->i_d.di_nblocks += ralen;
2648                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2649                 if (ap->wasdel)
2650                         ap->ip->i_delayed_blks -= ralen;
2651                 /*
2652                  * Adjust the disk quota also. This was reserved
2653                  * earlier.
2654                  */
2655                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2656                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2657                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2658         } else {
2659                 ap->alen = 0;
2660         }
2661         return 0;
2662 }
2663
2664 STATIC int
2665 xfs_bmap_btalloc(
2666         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2667 {
2668         xfs_mount_t     *mp;            /* mount point structure */
2669         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2670         xfs_extlen_t    align;          /* minimum allocation alignment */
2671         xfs_agnumber_t  ag;
2672         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2673         xfs_agnumber_t  startag;
2674         xfs_alloc_arg_t args;
2675         xfs_extlen_t    blen;
2676         xfs_extlen_t    delta;
2677         xfs_extlen_t    longest;
2678         xfs_extlen_t    need;
2679         xfs_extlen_t    nextminlen = 0;
2680         xfs_perag_t     *pag;
2681         int             nullfb;         /* true if ap->firstblock isn't set */
2682         int             isaligned;
2683         int             notinit;
2684         int             tryagain;
2685         int             error;
2686
2687         mp = ap->ip->i_mount;
2688         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2689         if (unlikely(align)) {
2690                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2691                                                 align, 0, ap->eof, 0, ap->conv,
2692                                                 &ap->off, &ap->alen);
2693                 ASSERT(!error);
2694                 ASSERT(ap->alen);
2695         }
2696         nullfb = ap->firstblock == NULLFSBLOCK;
2697         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2698         if (nullfb) {
2699                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2700                         ag = xfs_filestream_lookup_ag(ap->ip);
2701                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2702                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2703                 } else {
2704                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2705                 }
2706         } else
2707                 ap->rval = ap->firstblock;
2708
2709         xfs_bmap_adjacent(ap);
2710
2711         /*
2712          * If allowed, use ap->rval; otherwise must use firstblock since
2713          * it's in the right allocation group.
2714          */
2715         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2716                 ;
2717         else
2718                 ap->rval = ap->firstblock;
2719         /*
2720          * Normal allocation, done through xfs_alloc_vextent.
2721          */
2722         tryagain = isaligned = 0;
2723         args.tp = ap->tp;
2724         args.mp = mp;
2725         args.fsbno = ap->rval;
2726         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2727         args.firstblock = ap->firstblock;
2728         blen = 0;
2729         if (nullfb) {
2730                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2731                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2732                 else
2733                         args.type = XFS_ALLOCTYPE_START_BNO;
2734                 args.total = ap->total;
2735
2736                 /*
2737                  * Search for an allocation group with a single extent
2738                  * large enough for the request.
2739                  *
2740                  * If one isn't found, then adjust the minimum allocation
2741                  * size to the largest space found.
2742                  */
2743                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2744                 if (startag == NULLAGNUMBER)
2745                         startag = ag = 0;
2746                 notinit = 0;
2747                 down_read(&mp->m_peraglock);
2748                 while (blen < ap->alen) {
2749                         pag = &mp->m_perag[ag];
2750                         if (!pag->pagf_init &&
2751                             (error = xfs_alloc_pagf_init(mp, args.tp,
2752                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2753                                 up_read(&mp->m_peraglock);
2754                                 return error;
2755                         }
2756                         /*
2757                          * See xfs_alloc_fix_freelist...
2758                          */
2759                         if (pag->pagf_init) {
2760                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2761                                 delta = need > pag->pagf_flcount ?
2762                                         need - pag->pagf_flcount : 0;
2763                                 longest = (pag->pagf_longest > delta) ?
2764                                         (pag->pagf_longest - delta) :
2765                                         (pag->pagf_flcount > 0 ||
2766                                          pag->pagf_longest > 0);
2767                                 if (blen < longest)
2768                                         blen = longest;
2769                         } else
2770                                 notinit = 1;
2771
2772                         if (xfs_inode_is_filestream(ap->ip)) {
2773                                 if (blen >= ap->alen)
2774                                         break;
2775
2776                                 if (ap->userdata) {
2777                                         /*
2778                                          * If startag is an invalid AG, we've
2779                                          * come here once before and
2780                                          * xfs_filestream_new_ag picked the
2781                                          * best currently available.
2782                                          *
2783                                          * Don't continue looping, since we
2784                                          * could loop forever.
2785                                          */
2786                                         if (startag == NULLAGNUMBER)
2787                                                 break;
2788
2789                                         error = xfs_filestream_new_ag(ap, &ag);
2790                                         if (error) {
2791                                                 up_read(&mp->m_peraglock);
2792                                                 return error;
2793                                         }
2794
2795                                         /* loop again to set 'blen'*/
2796                                         startag = NULLAGNUMBER;
2797                                         continue;
2798                                 }
2799                         }
2800                         if (++ag == mp->m_sb.sb_agcount)
2801                                 ag = 0;
2802                         if (ag == startag)
2803                                 break;
2804                 }
2805                 up_read(&mp->m_peraglock);
2806                 /*
2807                  * Since the above loop did a BUF_TRYLOCK, it is
2808                  * possible that there is space for this request.
2809                  */
2810                 if (notinit || blen < ap->minlen)
2811                         args.minlen = ap->minlen;
2812                 /*
2813                  * If the best seen length is less than the request
2814                  * length, use the best as the minimum.
2815                  */
2816                 else if (blen < ap->alen)
2817                         args.minlen = blen;
2818                 /*
2819                  * Otherwise we've seen an extent as big as alen,
2820                  * use that as the minimum.
2821                  */
2822                 else
2823                         args.minlen = ap->alen;
2824
2825                 /*
2826                  * set the failure fallback case to look in the selected
2827                  * AG as the stream may have moved.
2828                  */
2829                 if (xfs_inode_is_filestream(ap->ip))
2830                         ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2831         } else if (ap->low) {
2832                 if (xfs_inode_is_filestream(ap->ip))
2833                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2834                 else
2835                         args.type = XFS_ALLOCTYPE_START_BNO;
2836                 args.total = args.minlen = ap->minlen;
2837         } else {
2838                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2839                 args.total = ap->total;
2840                 args.minlen = ap->minlen;
2841         }
2842         /* apply extent size hints if obtained earlier */
2843         if (unlikely(align)) {
2844                 args.prod = align;
2845                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2846                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2847         } else if (mp->m_sb.sb_blocksize >= NBPP) {
2848                 args.prod = 1;
2849                 args.mod = 0;
2850         } else {
2851                 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2852                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2853                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2854         }
2855         /*
2856          * If we are not low on available data blocks, and the
2857          * underlying logical volume manager is a stripe, and
2858          * the file offset is zero then try to allocate data
2859          * blocks on stripe unit boundary.
2860          * NOTE: ap->aeof is only set if the allocation length
2861          * is >= the stripe unit and the allocation offset is
2862          * at the end of file.
2863          */
2864         if (!ap->low && ap->aeof) {
2865                 if (!ap->off) {
2866                         args.alignment = mp->m_dalign;
2867                         atype = args.type;
2868                         isaligned = 1;
2869                         /*
2870                          * Adjust for alignment
2871                          */
2872                         if (blen > args.alignment && blen <= ap->alen)
2873                                 args.minlen = blen - args.alignment;
2874                         args.minalignslop = 0;
2875                 } else {
2876                         /*
2877                          * First try an exact bno allocation.
2878                          * If it fails then do a near or start bno
2879                          * allocation with alignment turned on.
2880                          */
2881                         atype = args.type;
2882                         tryagain = 1;
2883                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2884                         args.alignment = 1;
2885                         /*
2886                          * Compute the minlen+alignment for the
2887                          * next case.  Set slop so that the value
2888                          * of minlen+alignment+slop doesn't go up
2889                          * between the calls.
2890                          */
2891                         if (blen > mp->m_dalign && blen <= ap->alen)
2892                                 nextminlen = blen - mp->m_dalign;
2893                         else
2894                                 nextminlen = args.minlen;
2895                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2896                                 args.minalignslop =
2897                                         nextminlen + mp->m_dalign -
2898                                         args.minlen - 1;
2899                         else
2900                                 args.minalignslop = 0;
2901                 }
2902         } else {
2903                 args.alignment = 1;
2904                 args.minalignslop = 0;
2905         }
2906         args.minleft = ap->minleft;
2907         args.wasdel = ap->wasdel;
2908         args.isfl = 0;
2909         args.userdata = ap->userdata;
2910         if ((error = xfs_alloc_vextent(&args)))
2911                 return error;
2912         if (tryagain && args.fsbno == NULLFSBLOCK) {
2913                 /*
2914                  * Exact allocation failed. Now try with alignment
2915                  * turned on.
2916                  */
2917                 args.type = atype;
2918                 args.fsbno = ap->rval;
2919                 args.alignment = mp->m_dalign;
2920                 args.minlen = nextminlen;
2921                 args.minalignslop = 0;
2922                 isaligned = 1;
2923                 if ((error = xfs_alloc_vextent(&args)))
2924                         return error;
2925         }
2926         if (isaligned && args.fsbno == NULLFSBLOCK) {
2927                 /*
2928                  * allocation failed, so turn off alignment and
2929                  * try again.
2930                  */
2931                 args.type = atype;
2932                 args.fsbno = ap->rval;
2933                 args.alignment = 0;
2934                 if ((error = xfs_alloc_vextent(&args)))
2935                         return error;
2936         }
2937         if (args.fsbno == NULLFSBLOCK && nullfb &&
2938             args.minlen > ap->minlen) {
2939                 args.minlen = ap->minlen;
2940                 args.type = XFS_ALLOCTYPE_START_BNO;
2941                 args.fsbno = ap->rval;
2942                 if ((error = xfs_alloc_vextent(&args)))
2943                         return error;
2944         }
2945         if (args.fsbno == NULLFSBLOCK && nullfb) {
2946                 args.fsbno = 0;
2947                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2948                 args.total = ap->minlen;
2949                 args.minleft = 0;
2950                 if ((error = xfs_alloc_vextent(&args)))
2951                         return error;
2952                 ap->low = 1;
2953         }
2954         if (args.fsbno != NULLFSBLOCK) {
2955                 ap->firstblock = ap->rval = args.fsbno;
2956                 ASSERT(nullfb || fb_agno == args.agno ||
2957                        (ap->low && fb_agno < args.agno));
2958                 ap->alen = args.len;
2959                 ap->ip->i_d.di_nblocks += args.len;
2960                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2961                 if (ap->wasdel)
2962                         ap->ip->i_delayed_blks -= args.len;
2963                 /*
2964                  * Adjust the disk quota also. This was reserved
2965                  * earlier.
2966                  */
2967                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2968                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2969                                         XFS_TRANS_DQ_BCOUNT,
2970                         (long) args.len);
2971         } else {
2972                 ap->rval = NULLFSBLOCK;
2973                 ap->alen = 0;
2974         }
2975         return 0;
2976 }
2977
2978 /*
2979  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2980  * It figures out where to ask the underlying allocator to put the new extent.
2981  */
2982 STATIC int
2983 xfs_bmap_alloc(
2984         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2985 {
2986         if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2987                 return xfs_bmap_rtalloc(ap);
2988         return xfs_bmap_btalloc(ap);
2989 }
2990
2991 /*
2992  * Transform a btree format file with only one leaf node, where the
2993  * extents list will fit in the inode, into an extents format file.
2994  * Since the file extents are already in-core, all we have to do is
2995  * give up the space for the btree root and pitch the leaf block.
2996  */
2997 STATIC int                              /* error */
2998 xfs_bmap_btree_to_extents(
2999         xfs_trans_t             *tp,    /* transaction pointer */
3000         xfs_inode_t             *ip,    /* incore inode pointer */
3001         xfs_btree_cur_t         *cur,   /* btree cursor */
3002         int                     *logflagsp, /* inode logging flags */
3003         int                     whichfork)  /* data or attr fork */
3004 {
3005         /* REFERENCED */
3006         xfs_bmbt_block_t        *cblock;/* child btree block */
3007         xfs_fsblock_t           cbno;   /* child block number */
3008         xfs_buf_t               *cbp;   /* child block's buffer */
3009         int                     error;  /* error return value */
3010         xfs_ifork_t             *ifp;   /* inode fork data */
3011         xfs_mount_t             *mp;    /* mount point structure */
3012         __be64                  *pp;    /* ptr to block address */
3013         xfs_bmbt_block_t        *rblock;/* root btree block */
3014
3015         ifp = XFS_IFORK_PTR(ip, whichfork);
3016         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3017         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3018         rblock = ifp->if_broot;
3019         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3020         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3021         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3022         mp = ip->i_mount;
3023         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3024         cbno = be64_to_cpu(*pp);
3025         *logflagsp = 0;
3026 #ifdef DEBUG
3027         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3028                 return error;
3029 #endif
3030         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3031                         XFS_BMAP_BTREE_REF)))
3032                 return error;
3033         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3034         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3035                 return error;
3036         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3037         ip->i_d.di_nblocks--;
3038         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3039         xfs_trans_binval(tp, cbp);
3040         if (cur->bc_bufs[0] == cbp)
3041                 cur->bc_bufs[0] = NULL;
3042         xfs_iroot_realloc(ip, -1, whichfork);
3043         ASSERT(ifp->if_broot == NULL);
3044         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3045         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3046         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3047         return 0;
3048 }
3049
3050 /*
3051  * Called by xfs_bmapi to update file extent records and the btree
3052  * after removing space (or undoing a delayed allocation).
3053  */
3054 STATIC int                              /* error */
3055 xfs_bmap_del_extent(
3056         xfs_inode_t             *ip,    /* incore inode pointer */
3057         xfs_trans_t             *tp,    /* current transaction pointer */
3058         xfs_extnum_t            idx,    /* extent number to update/delete */
3059         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3060         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3061         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3062         int                     *logflagsp, /* inode logging flags */
3063         xfs_extdelta_t          *delta, /* Change made to incore extents */
3064         int                     whichfork, /* data or attr fork */
3065         int                     rsvd)   /* OK to allocate reserved blocks */
3066 {
3067         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3068         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3069         xfs_fsblock_t           del_endblock=0; /* first block past del */
3070         xfs_fileoff_t           del_endoff;     /* first offset past del */
3071         int                     delay;  /* current block is delayed allocated */
3072         int                     do_fx;  /* free extent at end of routine */
3073         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
3074         int                     error;  /* error return value */
3075         int                     flags;  /* inode logging flags */
3076         xfs_bmbt_irec_t         got;    /* current extent entry */
3077         xfs_fileoff_t           got_endoff;     /* first offset past got */
3078         int                     i;      /* temp state */
3079         xfs_ifork_t             *ifp;   /* inode fork pointer */
3080         xfs_mount_t             *mp;    /* mount structure */
3081         xfs_filblks_t           nblks;  /* quota/sb block count */
3082         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3083         /* REFERENCED */
3084         uint                    qfield; /* quota field to update */
3085         xfs_filblks_t           temp;   /* for indirect length calculations */
3086         xfs_filblks_t           temp2;  /* for indirect length calculations */
3087
3088         XFS_STATS_INC(xs_del_exlist);
3089         mp = ip->i_mount;
3090         ifp = XFS_IFORK_PTR(ip, whichfork);
3091         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3092                 (uint)sizeof(xfs_bmbt_rec_t)));
3093         ASSERT(del->br_blockcount > 0);
3094         ep = xfs_iext_get_ext(ifp, idx);
3095         xfs_bmbt_get_all(ep, &got);
3096         ASSERT(got.br_startoff <= del->br_startoff);
3097         del_endoff = del->br_startoff + del->br_blockcount;
3098         got_endoff = got.br_startoff + got.br_blockcount;
3099         ASSERT(got_endoff >= del_endoff);
3100         delay = ISNULLSTARTBLOCK(got.br_startblock);
3101         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3102         flags = 0;
3103         qfield = 0;
3104         error = 0;
3105         /*
3106          * If deleting a real allocation, must free up the disk space.
3107          */
3108         if (!delay) {
3109                 flags = XFS_ILOG_CORE;
3110                 /*
3111                  * Realtime allocation.  Free it and record di_nblocks update.
3112                  */
3113                 if (whichfork == XFS_DATA_FORK &&
3114                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
3115                         xfs_fsblock_t   bno;
3116                         xfs_filblks_t   len;
3117
3118                         ASSERT(do_mod(del->br_blockcount,
3119                                       mp->m_sb.sb_rextsize) == 0);
3120                         ASSERT(do_mod(del->br_startblock,
3121                                       mp->m_sb.sb_rextsize) == 0);
3122                         bno = del->br_startblock;
3123                         len = del->br_blockcount;
3124                         do_div(bno, mp->m_sb.sb_rextsize);
3125                         do_div(len, mp->m_sb.sb_rextsize);
3126                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3127                                         (xfs_extlen_t)len)))
3128                                 goto done;
3129                         do_fx = 0;
3130                         nblks = len * mp->m_sb.sb_rextsize;
3131                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3132                 }
3133                 /*
3134                  * Ordinary allocation.
3135                  */
3136                 else {
3137                         do_fx = 1;
3138                         nblks = del->br_blockcount;
3139                         qfield = XFS_TRANS_DQ_BCOUNT;
3140                 }
3141                 /*
3142                  * Set up del_endblock and cur for later.
3143                  */
3144                 del_endblock = del->br_startblock + del->br_blockcount;
3145                 if (cur) {
3146                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3147                                         got.br_startblock, got.br_blockcount,
3148                                         &i)))
3149                                 goto done;
3150                         ASSERT(i == 1);
3151                 }
3152                 da_old = da_new = 0;
3153         } else {
3154                 da_old = STARTBLOCKVAL(got.br_startblock);
3155                 da_new = 0;
3156                 nblks = 0;
3157                 do_fx = 0;
3158         }
3159         /*
3160          * Set flag value to use in switch statement.
3161          * Left-contig is 2, right-contig is 1.
3162          */
3163         switch (((got.br_startoff == del->br_startoff) << 1) |
3164                 (got_endoff == del_endoff)) {
3165         case 3:
3166                 /*
3167                  * Matches the whole extent.  Delete the entry.
3168                  */
3169                 XFS_BMAP_TRACE_DELETE("3", ip, idx, 1, whichfork);
3170                 xfs_iext_remove(ifp, idx, 1);
3171                 ifp->if_lastex = idx;
3172                 if (delay)
3173                         break;
3174                 XFS_IFORK_NEXT_SET(ip, whichfork,
3175                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3176                 flags |= XFS_ILOG_CORE;
3177                 if (!cur) {
3178                         flags |= XFS_ILOG_FEXT(whichfork);
3179                         break;
3180                 }
3181                 if ((error = xfs_bmbt_delete(cur, &i)))
3182                         goto done;
3183                 ASSERT(i == 1);
3184                 break;
3185
3186         case 2:
3187                 /*
3188                  * Deleting the first part of the extent.
3189                  */
3190                 XFS_BMAP_TRACE_PRE_UPDATE("2", ip, idx, whichfork);
3191                 xfs_bmbt_set_startoff(ep, del_endoff);
3192                 temp = got.br_blockcount - del->br_blockcount;
3193                 xfs_bmbt_set_blockcount(ep, temp);
3194                 ifp->if_lastex = idx;
3195                 if (delay) {
3196                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3197                                 da_old);
3198                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3199                         XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx,
3200                                 whichfork);
3201                         da_new = temp;
3202                         break;
3203                 }
3204                 xfs_bmbt_set_startblock(ep, del_endblock);
3205                 XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx, whichfork);
3206                 if (!cur) {
3207                         flags |= XFS_ILOG_FEXT(whichfork);
3208                         break;
3209                 }
3210                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3211                                 got.br_blockcount - del->br_blockcount,
3212                                 got.br_state)))
3213                         goto done;
3214                 break;
3215
3216         case 1:
3217                 /*
3218                  * Deleting the last part of the extent.
3219                  */
3220                 temp = got.br_blockcount - del->br_blockcount;
3221                 XFS_BMAP_TRACE_PRE_UPDATE("1", ip, idx, whichfork);
3222                 xfs_bmbt_set_blockcount(ep, temp);
3223                 ifp->if_lastex = idx;
3224                 if (delay) {
3225                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3226                                 da_old);
3227                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3228                         XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx,
3229                                 whichfork);
3230                         da_new = temp;
3231                         break;
3232                 }
3233                 XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx, whichfork);
3234                 if (!cur) {
3235                         flags |= XFS_ILOG_FEXT(whichfork);
3236                         break;
3237                 }
3238                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3239                                 got.br_startblock,
3240                                 got.br_blockcount - del->br_blockcount,
3241                                 got.br_state)))
3242                         goto done;
3243                 break;
3244
3245         case 0:
3246                 /*
3247                  * Deleting the middle of the extent.
3248                  */
3249                 temp = del->br_startoff - got.br_startoff;
3250                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, whichfork);
3251                 xfs_bmbt_set_blockcount(ep, temp);
3252                 new.br_startoff = del_endoff;
3253                 temp2 = got_endoff - del_endoff;
3254                 new.br_blockcount = temp2;
3255                 new.br_state = got.br_state;
3256                 if (!delay) {
3257                         new.br_startblock = del_endblock;
3258                         flags |= XFS_ILOG_CORE;
3259                         if (cur) {
3260                                 if ((error = xfs_bmbt_update(cur,
3261                                                 got.br_startoff,
3262                                                 got.br_startblock, temp,
3263                                                 got.br_state)))
3264                                         goto done;
3265                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3266                                         goto done;
3267                                 cur->bc_rec.b = new;
3268                                 error = xfs_bmbt_insert(cur, &i);
3269                                 if (error && error != ENOSPC)
3270                                         goto done;
3271                                 /*
3272                                  * If get no-space back from btree insert,
3273                                  * it tried a split, and we have a zero
3274                                  * block reservation.
3275                                  * Fix up our state and return the error.
3276                                  */
3277                                 if (error == ENOSPC) {
3278                                         /*
3279                                          * Reset the cursor, don't trust
3280                                          * it after any insert operation.
3281                                          */
3282                                         if ((error = xfs_bmbt_lookup_eq(cur,
3283                                                         got.br_startoff,
3284                                                         got.br_startblock,
3285                                                         temp, &i)))
3286                                                 goto done;
3287                                         ASSERT(i == 1);
3288                                         /*
3289                                          * Update the btree record back
3290                                          * to the original value.
3291                                          */
3292                                         if ((error = xfs_bmbt_update(cur,
3293                                                         got.br_startoff,
3294                                                         got.br_startblock,
3295                                                         got.br_blockcount,
3296                                                         got.br_state)))
3297                                                 goto done;
3298                                         /*
3299                                          * Reset the extent record back
3300                                          * to the original value.
3301                                          */
3302                                         xfs_bmbt_set_blockcount(ep,
3303                                                 got.br_blockcount);
3304                                         flags = 0;
3305                                         error = XFS_ERROR(ENOSPC);
3306                                         goto done;
3307                                 }
3308                                 ASSERT(i == 1);
3309                         } else
3310                                 flags |= XFS_ILOG_FEXT(whichfork);
3311                         XFS_IFORK_NEXT_SET(ip, whichfork,
3312                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3313                 } else {
3314                         ASSERT(whichfork == XFS_DATA_FORK);
3315                         temp = xfs_bmap_worst_indlen(ip, temp);
3316                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3317                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3318                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3319                         da_new = temp + temp2;
3320                         while (da_new > da_old) {
3321                                 if (temp) {
3322                                         temp--;
3323                                         da_new--;
3324                                         xfs_bmbt_set_startblock(ep,
3325                                                 NULLSTARTBLOCK((int)temp));
3326                                 }
3327                                 if (da_new == da_old)
3328                                         break;
3329                                 if (temp2) {
3330                                         temp2--;
3331                                         da_new--;
3332                                         new.br_startblock =
3333                                                 NULLSTARTBLOCK((int)temp2);
3334                                 }
3335                         }
3336                 }
3337                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, whichfork);
3338                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 1, &new, NULL,
3339                         whichfork);
3340                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3341                 ifp->if_lastex = idx + 1;
3342                 break;
3343         }
3344         /*
3345          * If we need to, add to list of extents to delete.
3346          */
3347         if (do_fx)
3348                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3349                         mp);
3350         /*
3351          * Adjust inode # blocks in the file.
3352          */
3353         if (nblks)
3354                 ip->i_d.di_nblocks -= nblks;
3355         /*
3356          * Adjust quota data.
3357          */
3358         if (qfield)
3359                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3360
3361         /*
3362          * Account for change in delayed indirect blocks.
3363          * Nothing to do for disk quota accounting here.
3364          */
3365         ASSERT(da_old >= da_new);
3366         if (da_old > da_new)
3367                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3368                         rsvd);
3369         if (delta) {
3370                 /* DELTA: report the original extent. */
3371                 if (delta->xed_startoff > got.br_startoff)
3372                         delta->xed_startoff = got.br_startoff;
3373                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3374                         delta->xed_blockcount = got.br_startoff +
3375                                                         got.br_blockcount;
3376         }
3377 done:
3378         *logflagsp = flags;
3379         return error;
3380 }
3381
3382 /*
3383  * Remove the entry "free" from the free item list.  Prev points to the
3384  * previous entry, unless "free" is the head of the list.
3385  */
3386 STATIC void
3387 xfs_bmap_del_free(
3388         xfs_bmap_free_t         *flist, /* free item list header */
3389         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3390         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3391 {
3392         if (prev)
3393                 prev->xbfi_next = free->xbfi_next;
3394         else
3395                 flist->xbf_first = free->xbfi_next;
3396         flist->xbf_count--;
3397         kmem_zone_free(xfs_bmap_free_item_zone, free);
3398 }
3399
3400 /*
3401  * Convert an extents-format file into a btree-format file.
3402  * The new file will have a root block (in the inode) and a single child block.
3403  */
3404 STATIC int                                      /* error */
3405 xfs_bmap_extents_to_btree(
3406         xfs_trans_t             *tp,            /* transaction pointer */
3407         xfs_inode_t             *ip,            /* incore inode pointer */
3408         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3409         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3410         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3411         int                     wasdel,         /* converting a delayed alloc */
3412         int                     *logflagsp,     /* inode logging flags */
3413         int                     whichfork)      /* data or attr fork */
3414 {
3415         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3416         xfs_buf_t               *abp;           /* buffer for ablock */
3417         xfs_alloc_arg_t         args;           /* allocation arguments */
3418         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3419         xfs_bmbt_block_t        *block;         /* btree root block */
3420         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3421         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
3422         int                     error;          /* error return value */
3423         xfs_extnum_t            i, cnt;         /* extent record index */
3424         xfs_ifork_t             *ifp;           /* inode fork pointer */
3425         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3426         xfs_mount_t             *mp;            /* mount structure */
3427         xfs_extnum_t            nextents;       /* number of file extents */
3428         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3429
3430         ifp = XFS_IFORK_PTR(ip, whichfork);
3431         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3432         ASSERT(ifp->if_ext_max ==
3433                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3434         /*
3435          * Make space in the inode incore.
3436          */
3437         xfs_iroot_realloc(ip, 1, whichfork);
3438         ifp->if_flags |= XFS_IFBROOT;
3439         /*
3440          * Fill in the root.
3441          */
3442         block = ifp->if_broot;
3443         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3444         block->bb_level = cpu_to_be16(1);
3445         block->bb_numrecs = cpu_to_be16(1);
3446         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3447         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3448         /*
3449          * Need a cursor.  Can't allocate until bb_level is filled in.
3450          */
3451         mp = ip->i_mount;
3452         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3453                 whichfork);
3454         cur->bc_private.b.firstblock = *firstblock;
3455         cur->bc_private.b.flist = flist;
3456         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3457         /*
3458          * Convert to a btree with two levels, one record in root.
3459          */
3460         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3461         args.tp = tp;
3462         args.mp = mp;
3463         args.firstblock = *firstblock;
3464         if (*firstblock == NULLFSBLOCK) {
3465                 args.type = XFS_ALLOCTYPE_START_BNO;
3466                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3467         } else if (flist->xbf_low) {
3468                 args.type = XFS_ALLOCTYPE_START_BNO;
3469                 args.fsbno = *firstblock;
3470         } else {
3471                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3472                 args.fsbno = *firstblock;
3473         }
3474         args.minlen = args.maxlen = args.prod = 1;
3475         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3476                 args.minalignslop = 0;
3477         args.wasdel = wasdel;
3478         *logflagsp = 0;
3479         if ((error = xfs_alloc_vextent(&args))) {
3480                 xfs_iroot_realloc(ip, -1, whichfork);
3481                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3482                 return error;
3483         }
3484         /*
3485          * Allocation can't fail, the space was reserved.
3486          */
3487         ASSERT(args.fsbno != NULLFSBLOCK);
3488         ASSERT(*firstblock == NULLFSBLOCK ||
3489                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3490                (flist->xbf_low &&
3491                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3492         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3493         cur->bc_private.b.allocated++;
3494         ip->i_d.di_nblocks++;
3495         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3496         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3497         /*
3498          * Fill in the child block.
3499          */
3500         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3501         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3502         ablock->bb_level = 0;
3503         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3504         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3505         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3506         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3507         for (cnt = i = 0; i < nextents; i++) {
3508                 ep = xfs_iext_get_ext(ifp, i);
3509                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3510                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3511                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3512                         arp++; cnt++;
3513                 }
3514         }
3515         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3516         ablock->bb_numrecs = cpu_to_be16(cnt);
3517         /*
3518          * Fill in the root key and pointer.
3519          */
3520         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3521         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3522         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3523         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3524         *pp = cpu_to_be64(args.fsbno);
3525         /*
3526          * Do all this logging at the end so that
3527          * the root is at the right level.
3528          */
3529         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3530         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3531         ASSERT(*curp == NULL);
3532         *curp = cur;
3533         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3534         return 0;
3535 }
3536
3537 /*
3538  * Helper routine to reset inode di_forkoff field when switching
3539  * attribute fork from local to extent format - we reset it where
3540  * possible to make space available for inline data fork extents.
3541  */
3542 STATIC void
3543 xfs_bmap_forkoff_reset(
3544         xfs_mount_t     *mp,
3545         xfs_inode_t     *ip,
3546         int             whichfork)
3547 {
3548         if (whichfork == XFS_ATTR_FORK &&
3549             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3550             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3551             (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3552             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3553                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3554                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3555                                         (uint)sizeof(xfs_bmbt_rec_t);
3556                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3557                                         (uint)sizeof(xfs_bmbt_rec_t);
3558         }
3559 }
3560
3561 /*
3562  * Convert a local file to an extents file.
3563  * This code is out of bounds for data forks of regular files,
3564  * since the file data needs to get logged so things will stay consistent.
3565  * (The bmap-level manipulations are ok, though).
3566  */
3567 STATIC int                              /* error */
3568 xfs_bmap_local_to_extents(
3569         xfs_trans_t     *tp,            /* transaction pointer */
3570         xfs_inode_t     *ip,            /* incore inode pointer */
3571         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3572         xfs_extlen_t    total,          /* total blocks needed by transaction */
3573         int             *logflagsp,     /* inode logging flags */
3574         int             whichfork)      /* data or attr fork */
3575 {
3576         int             error;          /* error return value */
3577         int             flags;          /* logging flags returned */
3578         xfs_ifork_t     *ifp;           /* inode fork pointer */
3579
3580         /*
3581          * We don't want to deal with the case of keeping inode data inline yet.
3582          * So sending the data fork of a regular inode is invalid.
3583          */
3584         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3585                  whichfork == XFS_DATA_FORK));
3586         ifp = XFS_IFORK_PTR(ip, whichfork);
3587         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3588         flags = 0;
3589         error = 0;
3590         if (ifp->if_bytes) {
3591                 xfs_alloc_arg_t args;   /* allocation arguments */
3592                 xfs_buf_t       *bp;    /* buffer for extent block */
3593                 xfs_bmbt_rec_t  *ep;    /* extent record pointer */
3594
3595                 args.tp = tp;
3596                 args.mp = ip->i_mount;
3597                 args.firstblock = *firstblock;
3598                 ASSERT((ifp->if_flags &
3599                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3600                 /*
3601                  * Allocate a block.  We know we need only one, since the
3602                  * file currently fits in an inode.
3603                  */
3604                 if (*firstblock == NULLFSBLOCK) {
3605                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3606                         args.type = XFS_ALLOCTYPE_START_BNO;
3607                 } else {
3608                         args.fsbno = *firstblock;
3609                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3610                 }
3611                 args.total = total;
3612                 args.mod = args.minleft = args.alignment = args.wasdel =
3613                         args.isfl = args.minalignslop = 0;
3614                 args.minlen = args.maxlen = args.prod = 1;
3615                 if ((error = xfs_alloc_vextent(&args)))
3616                         goto done;
3617                 /*
3618                  * Can't fail, the space was reserved.
3619                  */
3620                 ASSERT(args.fsbno != NULLFSBLOCK);
3621                 ASSERT(args.len == 1);
3622                 *firstblock = args.fsbno;
3623                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3624                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3625                         ifp->if_bytes);
3626                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3627                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3628                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3629                 xfs_iext_add(ifp, 0, 1);
3630                 ep = xfs_iext_get_ext(ifp, 0);
3631                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3632                 XFS_BMAP_TRACE_POST_UPDATE("new", ip, 0, whichfork);
3633                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3634                 ip->i_d.di_nblocks = 1;
3635                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3636                         XFS_TRANS_DQ_BCOUNT, 1L);
3637                 flags |= XFS_ILOG_FEXT(whichfork);
3638         } else {
3639                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3640                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3641         }
3642         ifp->if_flags &= ~XFS_IFINLINE;
3643         ifp->if_flags |= XFS_IFEXTENTS;
3644         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3645         flags |= XFS_ILOG_CORE;
3646 done:
3647         *logflagsp = flags;
3648         return error;
3649 }
3650
3651 /*
3652  * Search the extent records for the entry containing block bno.
3653  * If bno lies in a hole, point to the next entry.  If bno lies
3654  * past eof, *eofp will be set, and *prevp will contain the last
3655  * entry (null if none).  Else, *lastxp will be set to the index
3656  * of the found entry; *gotp will contain the entry.
3657  */
3658 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3659 xfs_bmap_search_multi_extents(
3660         xfs_ifork_t     *ifp,           /* inode fork pointer */
3661         xfs_fileoff_t   bno,            /* block number searched for */
3662         int             *eofp,          /* out: end of file found */
3663         xfs_extnum_t    *lastxp,        /* out: last extent index */
3664         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3665         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3666 {
3667         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3668         xfs_extnum_t    lastx;          /* last extent index */
3669
3670         /*
3671          * Initialize the extent entry structure to catch access to
3672          * uninitialized br_startblock field.
3673          */
3674         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3675         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3676         gotp->br_state = XFS_EXT_INVALID;
3677 #if XFS_BIG_BLKNOS
3678         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3679 #else
3680         gotp->br_startblock = 0xffffa5a5;
3681 #endif
3682         prevp->br_startoff = NULLFILEOFF;
3683
3684         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3685         if (lastx > 0) {
3686                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3687         }
3688         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3689                 xfs_bmbt_get_all(ep, gotp);
3690                 *eofp = 0;
3691         } else {
3692                 if (lastx > 0) {
3693                         *gotp = *prevp;
3694                 }
3695                 *eofp = 1;
3696                 ep = NULL;
3697         }
3698         *lastxp = lastx;
3699         return ep;
3700 }
3701
3702 /*
3703  * Search the extents list for the inode, for the extent containing bno.
3704  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3705  * *eofp will be set, and *prevp will contain the last entry (null if none).
3706  * Else, *lastxp will be set to the index of the found
3707  * entry; *gotp will contain the entry.
3708  */
3709 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3710 xfs_bmap_search_extents(
3711         xfs_inode_t     *ip,            /* incore inode pointer */
3712         xfs_fileoff_t   bno,            /* block number searched for */
3713         int             fork,           /* data or attr fork */
3714         int             *eofp,          /* out: end of file found */
3715         xfs_extnum_t    *lastxp,        /* out: last extent index */
3716         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3717         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3718 {
3719         xfs_ifork_t     *ifp;           /* inode fork pointer */
3720         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3721
3722         XFS_STATS_INC(xs_look_exlist);
3723         ifp = XFS_IFORK_PTR(ip, fork);
3724
3725         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3726
3727         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3728                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3729                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3730                                 "Access to block zero in inode %llu "
3731                                 "start_block: %llx start_off: %llx "
3732                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3733                         (unsigned long long)ip->i_ino,
3734                         (unsigned long long)gotp->br_startblock,
3735                         (unsigned long long)gotp->br_startoff,
3736                         (unsigned long long)gotp->br_blockcount,
3737                         gotp->br_state, *lastxp);
3738                 *lastxp = NULLEXTNUM;
3739                 *eofp = 1;
3740                 return NULL;
3741         }
3742         return ep;
3743 }
3744
3745
3746 #ifdef XFS_BMAP_TRACE
3747 ktrace_t        *xfs_bmap_trace_buf;
3748
3749 /*
3750  * Add a bmap trace buffer entry.  Base routine for the others.
3751  */
3752 STATIC void
3753 xfs_bmap_trace_addentry(
3754         int             opcode,         /* operation */
3755         const char      *fname,         /* function name */
3756         char            *desc,          /* operation description */
3757         xfs_inode_t     *ip,            /* incore inode pointer */
3758         xfs_extnum_t    idx,            /* index of entry(ies) */
3759         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3760         xfs_bmbt_rec_t  *r1,            /* first record */
3761         xfs_bmbt_rec_t  *r2,            /* second record or null */
3762         int             whichfork)      /* data or attr fork */
3763 {
3764         xfs_bmbt_rec_t  tr2;
3765
3766         ASSERT(cnt == 1 || cnt == 2);
3767         ASSERT(r1 != NULL);
3768         if (cnt == 1) {
3769                 ASSERT(r2 == NULL);
3770                 r2 = &tr2;
3771                 memset(&tr2, 0, sizeof(tr2));
3772         } else
3773                 ASSERT(r2 != NULL);
3774         ktrace_enter(xfs_bmap_trace_buf,
3775                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3776                 (void *)fname, (void *)desc, (void *)ip,
3777                 (void *)(__psint_t)idx,
3778                 (void *)(__psint_t)cnt,
3779                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3780                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3781                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3782                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3783                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3784                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3785                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3786                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3787                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3788                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3789                 );
3790         ASSERT(ip->i_xtrace);
3791         ktrace_enter(ip->i_xtrace,
3792                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3793                 (void *)fname, (void *)desc, (void *)ip,
3794                 (void *)(__psint_t)idx,
3795                 (void *)(__psint_t)cnt,
3796                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3797                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3798                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3799                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3800                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3801                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3802                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3803                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3804                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3805                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3806                 );
3807 }
3808
3809 /*
3810  * Add bmap trace entry prior to a call to xfs_iext_remove.
3811  */
3812 STATIC void
3813 xfs_bmap_trace_delete(
3814         const char      *fname,         /* function name */
3815         char            *desc,          /* operation description */
3816         xfs_inode_t     *ip,            /* incore inode pointer */
3817         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3818         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3819         int             whichfork)      /* data or attr fork */
3820 {
3821         xfs_ifork_t     *ifp;           /* inode fork pointer */
3822
3823         ifp = XFS_IFORK_PTR(ip, whichfork);
3824         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3825                 cnt, xfs_iext_get_ext(ifp, idx),
3826                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3827                 whichfork);
3828 }
3829
3830 /*
3831  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3832  * reading in the extents list from the disk (in the btree).
3833  */
3834 STATIC void
3835 xfs_bmap_trace_insert(
3836         const char      *fname,         /* function name */
3837         char            *desc,          /* operation description */
3838         xfs_inode_t     *ip,            /* incore inode pointer */
3839         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3840         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3841         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3842         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3843         int             whichfork)      /* data or attr fork */
3844 {
3845         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3846         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3847
3848         xfs_bmbt_set_all(&tr1, r1);
3849         if (cnt == 2) {
3850                 ASSERT(r2 != NULL);
3851                 xfs_bmbt_set_all(&tr2, r2);
3852         } else {
3853                 ASSERT(cnt == 1);
3854                 ASSERT(r2 == NULL);
3855         }
3856         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3857                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3858 }
3859
3860 /*
3861  * Add bmap trace entry after updating an extent record in place.
3862  */
3863 STATIC void
3864 xfs_bmap_trace_post_update(
3865         const char      *fname,         /* function name */
3866         char            *desc,          /* operation description */
3867         xfs_inode_t     *ip,            /* incore inode pointer */
3868         xfs_extnum_t    idx,            /* index of entry updated */
3869         int             whichfork)      /* data or attr fork */
3870 {
3871         xfs_ifork_t     *ifp;           /* inode fork pointer */
3872
3873         ifp = XFS_IFORK_PTR(ip, whichfork);
3874         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3875                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3876 }
3877
3878 /*
3879  * Add bmap trace entry prior to updating an extent record in place.
3880  */
3881 STATIC void
3882 xfs_bmap_trace_pre_update(
3883         const char      *fname,         /* function name */
3884         char            *desc,          /* operation description */
3885         xfs_inode_t     *ip,            /* incore inode pointer */
3886         xfs_extnum_t    idx,            /* index of entry to be updated */
3887         int             whichfork)      /* data or attr fork */
3888 {
3889         xfs_ifork_t     *ifp;           /* inode fork pointer */
3890
3891         ifp = XFS_IFORK_PTR(ip, whichfork);
3892         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3893                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3894 }
3895 #endif  /* XFS_BMAP_TRACE */
3896
3897 /*
3898  * Compute the worst-case number of indirect blocks that will be used
3899  * for ip's delayed extent of length "len".
3900  */
3901 STATIC xfs_filblks_t
3902 xfs_bmap_worst_indlen(
3903         xfs_inode_t     *ip,            /* incore inode pointer */
3904         xfs_filblks_t   len)            /* delayed extent length */
3905 {
3906         int             level;          /* btree level number */
3907         int             maxrecs;        /* maximum record count at this level */
3908         xfs_mount_t     *mp;            /* mount structure */
3909         xfs_filblks_t   rval;           /* return value */
3910
3911         mp = ip->i_mount;
3912         maxrecs = mp->m_bmap_dmxr[0];
3913         for (level = 0, rval = 0;
3914              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3915              level++) {
3916                 len += maxrecs - 1;
3917                 do_div(len, maxrecs);
3918                 rval += len;
3919                 if (len == 1)
3920                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3921                                 level - 1;
3922                 if (level == 0)
3923                         maxrecs = mp->m_bmap_dmxr[1];
3924         }
3925         return rval;
3926 }
3927
3928 #if defined(XFS_RW_TRACE)
3929 STATIC void
3930 xfs_bunmap_trace(
3931         xfs_inode_t             *ip,
3932         xfs_fileoff_t           bno,
3933         xfs_filblks_t           len,
3934         int                     flags,
3935         inst_t                  *ra)
3936 {
3937         if (ip->i_rwtrace == NULL)
3938                 return;
3939         ktrace_enter(ip->i_rwtrace,
3940                 (void *)(__psint_t)XFS_BUNMAP,
3941                 (void *)ip,
3942                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3943                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3944                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3945                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3946                 (void *)(__psint_t)len,
3947                 (void *)(__psint_t)flags,
3948                 (void *)(unsigned long)current_cpu(),
3949                 (void *)ra,
3950                 (void *)0,
3951                 (void *)0,
3952                 (void *)0,
3953                 (void *)0,
3954                 (void *)0,
3955                 (void *)0);
3956 }
3957 #endif
3958
3959 /*
3960  * Convert inode from non-attributed to attributed.
3961  * Must not be in a transaction, ip must not be locked.
3962  */
3963 int                                             /* error code */
3964 xfs_bmap_add_attrfork(
3965         xfs_inode_t             *ip,            /* incore inode pointer */
3966         int                     size,           /* space new attribute needs */
3967         int                     rsvd)           /* xact may use reserved blks */
3968 {
3969         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3970         xfs_bmap_free_t         flist;          /* freed extent records */
3971         xfs_mount_t             *mp;            /* mount structure */
3972         xfs_trans_t             *tp;            /* transaction pointer */
3973         unsigned long           s;              /* spinlock spl value */
3974         int                     blks;           /* space reservation */
3975         int                     version = 1;    /* superblock attr version */
3976         int                     committed;      /* xaction was committed */
3977         int                     logflags;       /* logging flags */
3978         int                     error;          /* error return value */
3979
3980         ASSERT(XFS_IFORK_Q(ip) == 0);
3981         ASSERT(ip->i_df.if_ext_max ==
3982                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3983
3984         mp = ip->i_mount;
3985         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3986         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3987         blks = XFS_ADDAFORK_SPACE_RES(mp);
3988         if (rsvd)
3989                 tp->t_flags |= XFS_TRANS_RESERVE;
3990         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3991                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3992                 goto error0;
3993         xfs_ilock(ip, XFS_ILOCK_EXCL);
3994         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3995                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3996                         XFS_QMOPT_RES_REGBLKS);
3997         if (error) {
3998                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3999                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
4000                 return error;
4001         }
4002         if (XFS_IFORK_Q(ip))
4003                 goto error1;
4004         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
4005                 /*
4006                  * For inodes coming from pre-6.2 filesystems.
4007                  */
4008                 ASSERT(ip->i_d.di_aformat == 0);
4009                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
4010         }
4011         ASSERT(ip->i_d.di_anextents == 0);
4012         VN_HOLD(XFS_ITOV(ip));
4013         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4014         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4015         switch (ip->i_d.di_format) {
4016         case XFS_DINODE_FMT_DEV:
4017                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4018                 break;
4019         case XFS_DINODE_FMT_UUID:
4020                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4021                 break;
4022         case XFS_DINODE_FMT_LOCAL:
4023         case XFS_DINODE_FMT_EXTENTS:
4024         case XFS_DINODE_FMT_BTREE:
4025                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4026                 if (!ip->i_d.di_forkoff)
4027                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4028                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4029                         version = 2;
4030                 break;
4031         default:
4032                 ASSERT(0);
4033                 error = XFS_ERROR(EINVAL);
4034                 goto error1;
4035         }
4036         ip->i_df.if_ext_max =
4037                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4038         ASSERT(ip->i_afp == NULL);
4039         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4040         ip->i_afp->if_ext_max =
4041                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4042         ip->i_afp->if_flags = XFS_IFEXTENTS;
4043         logflags = 0;
4044         XFS_BMAP_INIT(&flist, &firstblock);
4045         switch (ip->i_d.di_format) {
4046         case XFS_DINODE_FMT_LOCAL:
4047                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4048                         &logflags);
4049                 break;
4050         case XFS_DINODE_FMT_EXTENTS:
4051                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4052                         &flist, &logflags);
4053                 break;
4054         case XFS_DINODE_FMT_BTREE:
4055                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4056                         &logflags);
4057                 break;
4058         default:
4059                 error = 0;
4060                 break;
4061         }
4062         if (logflags)
4063                 xfs_trans_log_inode(tp, ip, logflags);
4064         if (error)
4065                 goto error2;
4066         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
4067            (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
4068                 __int64_t sbfields = 0;
4069
4070                 s = XFS_SB_LOCK(mp);
4071                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
4072                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
4073                         sbfields |= XFS_SB_VERSIONNUM;
4074                 }
4075                 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
4076                         XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
4077                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4078                 }
4079                 if (sbfields) {
4080                         XFS_SB_UNLOCK(mp, s);
4081                         xfs_mod_sb(tp, sbfields);
4082                 } else
4083                         XFS_SB_UNLOCK(mp, s);
4084         }
4085         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4086                 goto error2;
4087         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4088         ASSERT(ip->i_df.if_ext_max ==
4089                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4090         return error;
4091 error2:
4092         xfs_bmap_cancel(&flist);
4093 error1:
4094         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
4095         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4096 error0:
4097         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4098         ASSERT(ip->i_df.if_ext_max ==
4099                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4100         return error;
4101 }
4102
4103 /*
4104  * Add the extent to the list of extents to be free at transaction end.
4105  * The list is maintained sorted (by block number).
4106  */
4107 /* ARGSUSED */
4108 void
4109 xfs_bmap_add_free(
4110         xfs_fsblock_t           bno,            /* fs block number of extent */
4111         xfs_filblks_t           len,            /* length of extent */
4112         xfs_bmap_free_t         *flist,         /* list of extents */
4113         xfs_mount_t             *mp)            /* mount point structure */
4114 {
4115         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4116         xfs_bmap_free_item_t    *new;           /* new element */
4117         xfs_bmap_free_item_t    *prev;          /* previous element */
4118 #ifdef DEBUG
4119         xfs_agnumber_t          agno;
4120         xfs_agblock_t           agbno;
4121
4122         ASSERT(bno != NULLFSBLOCK);
4123         ASSERT(len > 0);
4124         ASSERT(len <= MAXEXTLEN);
4125         ASSERT(!ISNULLSTARTBLOCK(bno));
4126         agno = XFS_FSB_TO_AGNO(mp, bno);
4127         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4128         ASSERT(agno < mp->m_sb.sb_agcount);
4129         ASSERT(agbno < mp->m_sb.sb_agblocks);
4130         ASSERT(len < mp->m_sb.sb_agblocks);
4131         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4132 #endif
4133         ASSERT(xfs_bmap_free_item_zone != NULL);
4134         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4135         new->xbfi_startblock = bno;
4136         new->xbfi_blockcount = (xfs_extlen_t)len;
4137         for (prev = NULL, cur = flist->xbf_first;
4138              cur != NULL;
4139              prev = cur, cur = cur->xbfi_next) {
4140                 if (cur->xbfi_startblock >= bno)
4141                         break;
4142         }
4143         if (prev)
4144                 prev->xbfi_next = new;
4145         else
4146                 flist->xbf_first = new;
4147         new->xbfi_next = cur;
4148         flist->xbf_count++;
4149 }
4150
4151 /*
4152  * Compute and fill in the value of the maximum depth of a bmap btree
4153  * in this filesystem.  Done once, during mount.
4154  */
4155 void
4156 xfs_bmap_compute_maxlevels(
4157         xfs_mount_t     *mp,            /* file system mount structure */
4158         int             whichfork)      /* data or attr fork */
4159 {
4160         int             level;          /* btree level */
4161         uint            maxblocks;      /* max blocks at this level */
4162         uint            maxleafents;    /* max leaf entries possible */
4163         int             maxrootrecs;    /* max records in root block */
4164         int             minleafrecs;    /* min records in leaf block */
4165         int             minnoderecs;    /* min records in node block */
4166         int             sz;             /* root block size */
4167
4168         /*
4169          * The maximum number of extents in a file, hence the maximum
4170          * number of leaf entries, is controlled by the type of di_nextents
4171          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4172          * (a signed 16-bit number, xfs_aextnum_t).
4173          */
4174         if (whichfork == XFS_DATA_FORK) {
4175                 maxleafents = MAXEXTNUM;
4176                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4177                         XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
4178         } else {
4179                 maxleafents = MAXAEXTNUM;
4180                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4181                         XFS_BMDR_SPACE_CALC(MINABTPTRS) :
4182                         mp->m_sb.sb_inodesize - mp->m_attroffset;
4183         }
4184         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4185         minleafrecs = mp->m_bmap_dmnr[0];
4186         minnoderecs = mp->m_bmap_dmnr[1];
4187         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4188         for (level = 1; maxblocks > 1; level++) {
4189                 if (maxblocks <= maxrootrecs)
4190                         maxblocks = 1;
4191                 else
4192                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4193         }
4194         mp->m_bm_maxlevels[whichfork] = level;
4195 }
4196
4197 /*
4198  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4199  * caller.  Frees all the extents that need freeing, which must be done
4200  * last due to locking considerations.  We never free any extents in
4201  * the first transaction.  This is to allow the caller to make the first
4202  * transaction a synchronous one so that the pointers to the data being
4203  * broken in this transaction will be permanent before the data is actually
4204  * freed.  This is necessary to prevent blocks from being reallocated
4205  * and written to before the free and reallocation are actually permanent.
4206  * We do not just make the first transaction synchronous here, because
4207  * there are more efficient ways to gain the same protection in some cases
4208  * (see the file truncation code).
4209  *
4210  * Return 1 if the given transaction was committed and a new one
4211  * started, and 0 otherwise in the committed parameter.
4212  */
4213 /*ARGSUSED*/
4214 int                                             /* error */
4215 xfs_bmap_finish(
4216         xfs_trans_t             **tp,           /* transaction pointer addr */
4217         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4218         int                     *committed)     /* xact committed or not */
4219 {
4220         xfs_efd_log_item_t      *efd;           /* extent free data */
4221         xfs_efi_log_item_t      *efi;           /* extent free intention */
4222         int                     error;          /* error return value */
4223         xfs_bmap_free_item_t    *free;          /* free extent item */
4224         unsigned int            logres;         /* new log reservation */
4225         unsigned int            logcount;       /* new log count */
4226         xfs_mount_t             *mp;            /* filesystem mount structure */
4227         xfs_bmap_free_item_t    *next;          /* next item on free list */
4228         xfs_trans_t             *ntp;           /* new transaction pointer */
4229
4230         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4231         if (flist->xbf_count == 0) {
4232                 *committed = 0;
4233                 return 0;
4234         }
4235         ntp = *tp;
4236         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4237         for (free = flist->xbf_first; free; free = free->xbfi_next)
4238                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4239                         free->xbfi_blockcount);
4240         logres = ntp->t_log_res;
4241         logcount = ntp->t_log_count;
4242         ntp = xfs_trans_dup(*tp);
4243         error = xfs_trans_commit(*tp, 0);
4244         *tp = ntp;
4245         *committed = 1;
4246         /*
4247          * We have a new transaction, so we should return committed=1,
4248          * even though we're returning an error.
4249          */
4250         if (error) {
4251                 return error;
4252         }
4253         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4254                         logcount)))
4255                 return error;
4256         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4257         for (free = flist->xbf_first; free != NULL; free = next) {
4258                 next = free->xbfi_next;
4259                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4260                                 free->xbfi_blockcount))) {
4261                         /*
4262                          * The bmap free list will be cleaned up at a
4263                          * higher level.  The EFI will be canceled when
4264                          * this transaction is aborted.
4265                          * Need to force shutdown here to make sure it
4266                          * happens, since this transaction may not be
4267                          * dirty yet.
4268                          */
4269                         mp = ntp->t_mountp;
4270                         if (!XFS_FORCED_SHUTDOWN(mp))
4271                                 xfs_force_shutdown(mp,
4272                                                    (error == EFSCORRUPTED) ?
4273                                                    SHUTDOWN_CORRUPT_INCORE :
4274                                                    SHUTDOWN_META_IO_ERROR);
4275                         return error;
4276                 }
4277                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4278                         free->xbfi_blockcount);
4279                 xfs_bmap_del_free(flist, NULL, free);
4280         }
4281         return 0;
4282 }
4283
4284 /*
4285  * Free up any items left in the list.
4286  */
4287 void
4288 xfs_bmap_cancel(
4289         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4290 {
4291         xfs_bmap_free_item_t    *free;  /* free list item */
4292         xfs_bmap_free_item_t    *next;
4293
4294         if (flist->xbf_count == 0)
4295                 return;
4296         ASSERT(flist->xbf_first != NULL);
4297         for (free = flist->xbf_first; free; free = next) {
4298                 next = free->xbfi_next;
4299                 xfs_bmap_del_free(flist, NULL, free);
4300         }
4301         ASSERT(flist->xbf_count == 0);
4302 }
4303
4304 /*
4305  * Returns the file-relative block number of the first unused block(s)
4306  * in the file with at least "len" logically contiguous blocks free.
4307  * This is the lowest-address hole if the file has holes, else the first block
4308  * past the end of file.
4309  * Return 0 if the file is currently local (in-inode).
4310  */
4311 int                                             /* error */
4312 xfs_bmap_first_unused(
4313         xfs_trans_t     *tp,                    /* transaction pointer */
4314         xfs_inode_t     *ip,                    /* incore inode */
4315         xfs_extlen_t    len,                    /* size of hole to find */
4316         xfs_fileoff_t   *first_unused,          /* unused block */
4317         int             whichfork)              /* data or attr fork */
4318 {
4319         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4320         int             error;                  /* error return value */
4321         int             idx;                    /* extent record index */
4322         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4323         xfs_fileoff_t   lastaddr;               /* last block number seen */
4324         xfs_fileoff_t   lowest;                 /* lowest useful block */
4325         xfs_fileoff_t   max;                    /* starting useful block */
4326         xfs_fileoff_t   off;                    /* offset for this block */
4327         xfs_extnum_t    nextents;               /* number of extent entries */
4328
4329         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4330                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4331                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4332         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4333                 *first_unused = 0;
4334                 return 0;
4335         }
4336         ifp = XFS_IFORK_PTR(ip, whichfork);
4337         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4338             (error = xfs_iread_extents(tp, ip, whichfork)))
4339                 return error;
4340         lowest = *first_unused;
4341         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4342         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4343                 ep = xfs_iext_get_ext(ifp, idx);
4344                 off = xfs_bmbt_get_startoff(ep);
4345                 /*
4346                  * See if the hole before this extent will work.
4347                  */
4348                 if (off >= lowest + len && off - max >= len) {
4349                         *first_unused = max;
4350                         return 0;
4351                 }
4352                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4353                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4354         }
4355         *first_unused = max;
4356         return 0;
4357 }
4358
4359 /*
4360  * Returns the file-relative block number of the last block + 1 before
4361  * last_block (input value) in the file.
4362  * This is not based on i_size, it is based on the extent records.
4363  * Returns 0 for local files, as they do not have extent records.
4364  */
4365 int                                             /* error */
4366 xfs_bmap_last_before(
4367         xfs_trans_t     *tp,                    /* transaction pointer */
4368         xfs_inode_t     *ip,                    /* incore inode */
4369         xfs_fileoff_t   *last_block,            /* last block */
4370         int             whichfork)              /* data or attr fork */
4371 {
4372         xfs_fileoff_t   bno;                    /* input file offset */
4373         int             eof;                    /* hit end of file */
4374         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4375         int             error;                  /* error return value */
4376         xfs_bmbt_irec_t got;                    /* current extent value */
4377         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4378         xfs_extnum_t    lastx;                  /* last extent used */
4379         xfs_bmbt_irec_t prev;                   /* previous extent value */
4380
4381         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4382             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4383             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4384                return XFS_ERROR(EIO);
4385         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4386                 *last_block = 0;
4387                 return 0;
4388         }
4389         ifp = XFS_IFORK_PTR(ip, whichfork);
4390         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4391             (error = xfs_iread_extents(tp, ip, whichfork)))
4392                 return error;
4393         bno = *last_block - 1;
4394         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4395                 &prev);
4396         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4397                 if (prev.br_startoff == NULLFILEOFF)
4398                         *last_block = 0;
4399                 else
4400                         *last_block = prev.br_startoff + prev.br_blockcount;
4401         }
4402         /*
4403          * Otherwise *last_block is already the right answer.
4404          */
4405         return 0;
4406 }
4407
4408 /*
4409  * Returns the file-relative block number of the first block past eof in
4410  * the file.  This is not based on i_size, it is based on the extent records.
4411  * Returns 0 for local files, as they do not have extent records.
4412  */
4413 int                                             /* error */
4414 xfs_bmap_last_offset(
4415         xfs_trans_t     *tp,                    /* transaction pointer */
4416         xfs_inode_t     *ip,                    /* incore inode */
4417         xfs_fileoff_t   *last_block,            /* last block */
4418         int             whichfork)              /* data or attr fork */
4419 {
4420         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4421         int             error;                  /* error return value */
4422         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4423         xfs_extnum_t    nextents;               /* number of extent entries */
4424
4425         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4426             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4427             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4428                return XFS_ERROR(EIO);
4429         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4430                 *last_block = 0;
4431                 return 0;
4432         }
4433         ifp = XFS_IFORK_PTR(ip, whichfork);
4434         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4435             (error = xfs_iread_extents(tp, ip, whichfork)))
4436                 return error;
4437         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4438         if (!nextents) {
4439                 *last_block = 0;
4440                 return 0;
4441         }
4442         ep = xfs_iext_get_ext(ifp, nextents - 1);
4443         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4444         return 0;
4445 }
4446
4447 /*
4448  * Returns whether the selected fork of the inode has exactly one
4449  * block or not.  For the data fork we check this matches di_size,
4450  * implying the file's range is 0..bsize-1.
4451  */
4452 int                                     /* 1=>1 block, 0=>otherwise */
4453 xfs_bmap_one_block(
4454         xfs_inode_t     *ip,            /* incore inode */
4455         int             whichfork)      /* data or attr fork */
4456 {
4457         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4458         xfs_ifork_t     *ifp;           /* inode fork pointer */
4459         int             rval;           /* return value */
4460         xfs_bmbt_irec_t s;              /* internal version of extent */
4461
4462 #ifndef DEBUG
4463         if (whichfork == XFS_DATA_FORK) {
4464                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4465                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4466                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4467         }
4468 #endif  /* !DEBUG */
4469         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4470                 return 0;
4471         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4472                 return 0;
4473         ifp = XFS_IFORK_PTR(ip, whichfork);
4474         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4475         ep = xfs_iext_get_ext(ifp, 0);
4476         xfs_bmbt_get_all(ep, &s);
4477         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4478         if (rval && whichfork == XFS_DATA_FORK)
4479                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4480         return rval;
4481 }
4482
4483 /*
4484  * Read in the extents to if_extents.
4485  * All inode fields are set up by caller, we just traverse the btree
4486  * and copy the records in. If the file system cannot contain unwritten
4487  * extents, the records are checked for no "state" flags.
4488  */
4489 int                                     /* error */
4490 xfs_bmap_read_extents(
4491         xfs_trans_t             *tp,    /* transaction pointer */
4492         xfs_inode_t             *ip,    /* incore inode */
4493         int                     whichfork) /* data or attr fork */
4494 {
4495         xfs_bmbt_block_t        *block; /* current btree block */
4496         xfs_fsblock_t           bno;    /* block # of "block" */
4497         xfs_buf_t               *bp;    /* buffer for "block" */
4498         int                     error;  /* error return value */
4499         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4500         xfs_extnum_t            i, j;   /* index into the extents list */
4501         xfs_ifork_t             *ifp;   /* fork structure */
4502         int                     level;  /* btree level, for checking */
4503         xfs_mount_t             *mp;    /* file system mount structure */
4504         __be64                  *pp;    /* pointer to block address */
4505         /* REFERENCED */
4506         xfs_extnum_t            room;   /* number of entries there's room for */
4507
4508         bno = NULLFSBLOCK;
4509         mp = ip->i_mount;
4510         ifp = XFS_IFORK_PTR(ip, whichfork);
4511         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4512                                         XFS_EXTFMT_INODE(ip);
4513         block = ifp->if_broot;
4514         /*
4515          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4516          */
4517         level = be16_to_cpu(block->bb_level);
4518         ASSERT(level > 0);
4519         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4520         bno = be64_to_cpu(*pp);
4521         ASSERT(bno != NULLDFSBNO);
4522         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4523         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4524         /*
4525          * Go down the tree until leaf level is reached, following the first
4526          * pointer (leftmost) at each level.
4527          */
4528         while (level-- > 0) {
4529                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4530                                 XFS_BMAP_BTREE_REF)))
4531                         return error;
4532                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4533                 XFS_WANT_CORRUPTED_GOTO(
4534                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4535                         error0);
4536                 if (level == 0)
4537                         break;
4538                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4539                 bno = be64_to_cpu(*pp);
4540                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4541                 xfs_trans_brelse(tp, bp);
4542         }
4543         /*
4544          * Here with bp and block set to the leftmost leaf node in the tree.
4545          */
4546         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4547         i = 0;
4548         /*
4549          * Loop over all leaf nodes.  Copy information to the extent records.
4550          */
4551         for (;;) {
4552                 xfs_bmbt_rec_t  *frp, *trp;
4553                 xfs_fsblock_t   nextbno;
4554                 xfs_extnum_t    num_recs;
4555                 xfs_extnum_t    start;
4556
4557
4558                 num_recs = be16_to_cpu(block->bb_numrecs);
4559                 if (unlikely(i + num_recs > room)) {
4560                         ASSERT(i + num_recs <= room);
4561                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4562                                 "corrupt dinode %Lu, (btree extents).",
4563                                 (unsigned long long) ip->i_ino);
4564                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4565                                          XFS_ERRLEVEL_LOW,
4566                                         ip->i_mount);
4567                         goto error0;
4568                 }
4569                 XFS_WANT_CORRUPTED_GOTO(
4570                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4571                         error0);
4572                 /*
4573                  * Read-ahead the next leaf block, if any.
4574                  */
4575                 nextbno = be64_to_cpu(block->bb_rightsib);
4576                 if (nextbno != NULLFSBLOCK)
4577                         xfs_btree_reada_bufl(mp, nextbno, 1);
4578                 /*
4579                  * Copy records into the extent records.
4580                  */
4581                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4582                 start = i;
4583                 for (j = 0; j < num_recs; j++, i++, frp++) {
4584                         trp = xfs_iext_get_ext(ifp, i);
4585                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4586                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4587                 }
4588                 if (exntf == XFS_EXTFMT_NOSTATE) {
4589                         /*
4590                          * Check all attribute bmap btree records and
4591                          * any "older" data bmap btree records for a
4592                          * set bit in the "extent flag" position.
4593                          */
4594                         if (unlikely(xfs_check_nostate_extents(ifp,
4595                                         start, num_recs))) {
4596                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4597                                                  XFS_ERRLEVEL_LOW,
4598                                                  ip->i_mount);
4599                                 goto error0;
4600                         }
4601                 }
4602                 xfs_trans_brelse(tp, bp);
4603                 bno = nextbno;
4604                 /*
4605                  * If we've reached the end, stop.
4606                  */
4607                 if (bno == NULLFSBLOCK)
4608                         break;
4609                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4610                                 XFS_BMAP_BTREE_REF)))
4611                         return error;
4612                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4613         }
4614         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4615         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4616         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4617         return 0;
4618 error0:
4619         xfs_trans_brelse(tp, bp);
4620         return XFS_ERROR(EFSCORRUPTED);
4621 }
4622
4623 #ifdef XFS_BMAP_TRACE
4624 /*
4625  * Add bmap trace insert entries for all the contents of the extent records.
4626  */
4627 void
4628 xfs_bmap_trace_exlist(
4629         const char      *fname,         /* function name */
4630         xfs_inode_t     *ip,            /* incore inode pointer */
4631         xfs_extnum_t    cnt,            /* count of entries in the list */
4632         int             whichfork)      /* data or attr fork */
4633 {
4634         xfs_bmbt_rec_t  *ep;            /* current extent record */
4635         xfs_extnum_t    idx;            /* extent record index */
4636         xfs_ifork_t     *ifp;           /* inode fork pointer */
4637         xfs_bmbt_irec_t s;              /* file extent record */
4638
4639         ifp = XFS_IFORK_PTR(ip, whichfork);
4640         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4641         for (idx = 0; idx < cnt; idx++) {
4642                 ep = xfs_iext_get_ext(ifp, idx);
4643                 xfs_bmbt_get_all(ep, &s);
4644                 XFS_BMAP_TRACE_INSERT("exlist", ip, idx, 1, &s, NULL,
4645                         whichfork);
4646         }
4647 }
4648 #endif
4649
4650 #ifdef DEBUG
4651 /*
4652  * Validate that the bmbt_irecs being returned from bmapi are valid
4653  * given the callers original parameters.  Specifically check the
4654  * ranges of the returned irecs to ensure that they only extent beyond
4655  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4656  */
4657 STATIC void
4658 xfs_bmap_validate_ret(
4659         xfs_fileoff_t           bno,
4660         xfs_filblks_t           len,
4661         int                     flags,
4662         xfs_bmbt_irec_t         *mval,
4663         int                     nmap,
4664         int                     ret_nmap)
4665 {
4666         int                     i;              /* index to map values */
4667
4668         ASSERT(ret_nmap <= nmap);
4669
4670         for (i = 0; i < ret_nmap; i++) {
4671                 ASSERT(mval[i].br_blockcount > 0);
4672                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4673                         ASSERT(mval[i].br_startoff >= bno);
4674                         ASSERT(mval[i].br_blockcount <= len);
4675                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4676                                bno + len);
4677                 } else {
4678                         ASSERT(mval[i].br_startoff < bno + len);
4679                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4680                                bno);
4681                 }
4682                 ASSERT(i == 0 ||
4683                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4684                        mval[i].br_startoff);
4685                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4686                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4687                                mval[i].br_startblock != HOLESTARTBLOCK);
4688                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4689                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4690         }
4691 }
4692 #endif /* DEBUG */
4693
4694
4695 /*
4696  * Map file blocks to filesystem blocks.
4697  * File range is given by the bno/len pair.
4698  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4699  * into a hole or past eof.
4700  * Only allocates blocks from a single allocation group,
4701  * to avoid locking problems.
4702  * The returned value in "firstblock" from the first call in a transaction
4703  * must be remembered and presented to subsequent calls in "firstblock".
4704  * An upper bound for the number of blocks to be allocated is supplied to
4705  * the first call in "total"; if no allocation group has that many free
4706  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4707  */
4708 int                                     /* error */
4709 xfs_bmapi(
4710         xfs_trans_t     *tp,            /* transaction pointer */
4711         xfs_inode_t     *ip,            /* incore inode */
4712         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4713         xfs_filblks_t   len,            /* length to map in file */
4714         int             flags,          /* XFS_BMAPI_... */
4715         xfs_fsblock_t   *firstblock,    /* first allocated block
4716                                            controls a.g. for allocs */
4717         xfs_extlen_t    total,          /* total blocks needed */
4718         xfs_bmbt_irec_t *mval,          /* output: map values */
4719         int             *nmap,          /* i/o: mval size/count */
4720         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4721         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4722 {
4723         xfs_fsblock_t   abno;           /* allocated block number */
4724         xfs_extlen_t    alen;           /* allocated extent length */
4725         xfs_fileoff_t   aoff;           /* allocated file offset */
4726         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4727         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4728         xfs_fileoff_t   end;            /* end of mapped file region */
4729         int             eof;            /* we've hit the end of extents */
4730         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
4731         int             error;          /* error return */
4732         xfs_bmbt_irec_t got;            /* current file extent record */
4733         xfs_ifork_t     *ifp;           /* inode fork pointer */
4734         xfs_extlen_t    indlen;         /* indirect blocks length */
4735         xfs_extnum_t    lastx;          /* last useful extent number */
4736         int             logflags;       /* flags for transaction logging */
4737         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4738         xfs_extlen_t    minlen;         /* min allocation size */
4739         xfs_mount_t     *mp;            /* xfs mount structure */
4740         int             n;              /* current extent index */
4741         int             nallocs;        /* number of extents alloc\'d */
4742         xfs_extnum_t    nextents;       /* number of extents in file */
4743         xfs_fileoff_t   obno;           /* old block number (offset) */
4744         xfs_bmbt_irec_t prev;           /* previous file extent record */
4745         int             tmp_logflags;   /* temp flags holder */
4746         int             whichfork;      /* data or attr fork */
4747         char            inhole;         /* current location is hole in file */
4748         char            wasdelay;       /* old extent was delayed */
4749         char            wr;             /* this is a write request */
4750         char            rt;             /* this is a realtime file */
4751 #ifdef DEBUG
4752         xfs_fileoff_t   orig_bno;       /* original block number value */
4753         int             orig_flags;     /* original flags arg value */
4754         xfs_filblks_t   orig_len;       /* original value of len arg */
4755         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4756         int             orig_nmap;      /* original value of *nmap */
4757
4758         orig_bno = bno;
4759         orig_len = len;
4760         orig_flags = flags;
4761         orig_mval = mval;
4762         orig_nmap = *nmap;
4763 #endif
4764         ASSERT(*nmap >= 1);
4765         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4766         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4767                 XFS_ATTR_FORK : XFS_DATA_FORK;
4768         mp = ip->i_mount;
4769         if (unlikely(XFS_TEST_ERROR(
4770             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4771              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4772              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4773              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4774                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4775                 return XFS_ERROR(EFSCORRUPTED);
4776         }
4777         if (XFS_FORCED_SHUTDOWN(mp))
4778                 return XFS_ERROR(EIO);
4779         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4780         ifp = XFS_IFORK_PTR(ip, whichfork);
4781         ASSERT(ifp->if_ext_max ==
4782                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4783         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4784                 XFS_STATS_INC(xs_blk_mapw);
4785         else
4786                 XFS_STATS_INC(xs_blk_mapr);
4787         /*
4788          * IGSTATE flag is used to combine extents which
4789          * differ only due to the state of the extents.
4790          * This technique is used from xfs_getbmap()
4791          * when the caller does not wish to see the
4792          * separation (which is the default).
4793          *
4794          * This technique is also used when writing a
4795          * buffer which has been partially written,
4796          * (usually by being flushed during a chunkread),
4797          * to ensure one write takes place. This also
4798          * prevents a change in the xfs inode extents at
4799          * this time, intentionally. This change occurs
4800          * on completion of the write operation, in
4801          * xfs_strat_comp(), where the xfs_bmapi() call
4802          * is transactioned, and the extents combined.
4803          */
4804         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4805                 wr = 0;                         /* no allocations are allowed */
4806         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4807         logflags = 0;
4808         nallocs = 0;
4809         cur = NULL;
4810         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4811                 ASSERT(wr && tp);
4812                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4813                                 firstblock, total, &logflags, whichfork)))
4814                         goto error0;
4815         }
4816         if (wr && *firstblock == NULLFSBLOCK) {
4817                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4818                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4819                 else
4820                         minleft = 1;
4821         } else
4822                 minleft = 0;
4823         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4824             (error = xfs_iread_extents(tp, ip, whichfork)))
4825                 goto error0;
4826         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4827                 &prev);
4828         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4829         n = 0;
4830         end = bno + len;
4831         obno = bno;
4832         bma.ip = NULL;
4833         if (delta) {
4834                 delta->xed_startoff = NULLFILEOFF;
4835                 delta->xed_blockcount = 0;
4836         }
4837         while (bno < end && n < *nmap) {
4838                 /*
4839                  * Reading past eof, act as though there's a hole
4840                  * up to end.
4841                  */
4842                 if (eof && !wr)
4843                         got.br_startoff = end;
4844                 inhole = eof || got.br_startoff > bno;
4845                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4846                         ISNULLSTARTBLOCK(got.br_startblock);
4847                 /*
4848                  * First, deal with the hole before the allocated space
4849                  * that we found, if any.
4850                  */
4851                 if (wr && (inhole || wasdelay)) {
4852                         /*
4853                          * For the wasdelay case, we could also just
4854                          * allocate the stuff asked for in this bmap call
4855                          * but that wouldn't be as good.
4856                          */
4857                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4858                                 alen = (xfs_extlen_t)got.br_blockcount;
4859                                 aoff = got.br_startoff;
4860                                 if (lastx != NULLEXTNUM && lastx) {
4861                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4862                                         xfs_bmbt_get_all(ep, &prev);
4863                                 }
4864                         } else if (wasdelay) {
4865                                 alen = (xfs_extlen_t)
4866                                         XFS_FILBLKS_MIN(len,
4867                                                 (got.br_startoff +
4868                                                  got.br_blockcount) - bno);
4869                                 aoff = bno;
4870                         } else {
4871                                 alen = (xfs_extlen_t)
4872                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4873                                 if (!eof)
4874                                         alen = (xfs_extlen_t)
4875                                                 XFS_FILBLKS_MIN(alen,
4876                                                         got.br_startoff - bno);
4877                                 aoff = bno;
4878                         }
4879                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4880                         if (flags & XFS_BMAPI_DELAY) {
4881                                 xfs_extlen_t    extsz;
4882
4883                                 /* Figure out the extent size, adjust alen */
4884                                 extsz = xfs_get_extsz_hint(ip);
4885                                 if (extsz) {
4886                                         error = xfs_bmap_extsize_align(mp,
4887                                                         &got, &prev, extsz,
4888                                                         rt, eof,
4889                                                         flags&XFS_BMAPI_DELAY,
4890                                                         flags&XFS_BMAPI_CONVERT,
4891                                                         &aoff, &alen);
4892                                         ASSERT(!error);
4893                                 }
4894
4895                                 if (rt)
4896                                         extsz = alen / mp->m_sb.sb_rextsize;
4897
4898                                 /*
4899                                  * Make a transaction-less quota reservation for
4900                                  * delayed allocation blocks. This number gets
4901                                  * adjusted later.  We return if we haven't
4902                                  * allocated blocks already inside this loop.
4903                                  */
4904                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4905                                                 mp, NULL, ip, (long)alen, 0,
4906                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4907                                                      XFS_QMOPT_RES_REGBLKS))) {
4908                                         if (n == 0) {
4909                                                 *nmap = 0;
4910                                                 ASSERT(cur == NULL);
4911                                                 return error;
4912                                         }
4913                                         break;
4914                                 }
4915
4916                                 /*
4917                                  * Split changing sb for alen and indlen since
4918                                  * they could be coming from different places.
4919                                  */
4920                                 indlen = (xfs_extlen_t)
4921                                         xfs_bmap_worst_indlen(ip, alen);
4922                                 ASSERT(indlen > 0);
4923
4924                                 if (rt) {
4925                                         error = xfs_mod_incore_sb(mp,
4926                                                         XFS_SBS_FREXTENTS,
4927                                                         -((int64_t)extsz), (flags &
4928                                                         XFS_BMAPI_RSVBLOCKS));
4929                                 } else {
4930                                         error = xfs_mod_incore_sb(mp,
4931                                                         XFS_SBS_FDBLOCKS,
4932                                                         -((int64_t)alen), (flags &
4933                                                         XFS_BMAPI_RSVBLOCKS));
4934                                 }
4935                                 if (!error) {
4936                                         error = xfs_mod_incore_sb(mp,
4937                                                         XFS_SBS_FDBLOCKS,
4938                                                         -((int64_t)indlen), (flags &
4939                                                         XFS_BMAPI_RSVBLOCKS));
4940                                         if (error && rt)
4941                                                 xfs_mod_incore_sb(mp,
4942                                                         XFS_SBS_FREXTENTS,
4943                                                         (int64_t)extsz, (flags &
4944                                                         XFS_BMAPI_RSVBLOCKS));
4945                                         else if (error)
4946                                                 xfs_mod_incore_sb(mp,
4947                                                         XFS_SBS_FDBLOCKS,
4948                                                         (int64_t)alen, (flags &
4949                                                         XFS_BMAPI_RSVBLOCKS));
4950                                 }
4951
4952                                 if (error) {
4953                                         if (XFS_IS_QUOTA_ON(mp))
4954                                                 /* unreserve the blocks now */
4955                                                 (void)
4956                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4957                                                         mp, NULL, ip,
4958                                                         (long)alen, 0, rt ?
4959                                                         XFS_QMOPT_RES_RTBLKS :
4960                                                         XFS_QMOPT_RES_REGBLKS);
4961                                         break;
4962                                 }
4963
4964                                 ip->i_delayed_blks += alen;
4965                                 abno = NULLSTARTBLOCK(indlen);
4966                         } else {
4967                                 /*
4968                                  * If first time, allocate and fill in
4969                                  * once-only bma fields.
4970                                  */
4971                                 if (bma.ip == NULL) {
4972                                         bma.tp = tp;
4973                                         bma.ip = ip;
4974                                         bma.prevp = &prev;
4975                                         bma.gotp = &got;
4976                                         bma.total = total;
4977                                         bma.userdata = 0;
4978                                 }
4979                                 /* Indicate if this is the first user data
4980                                  * in the file, or just any user data.
4981                                  */
4982                                 if (!(flags & XFS_BMAPI_METADATA)) {
4983                                         bma.userdata = (aoff == 0) ?
4984                                                 XFS_ALLOC_INITIAL_USER_DATA :
4985                                                 XFS_ALLOC_USERDATA;
4986                                 }
4987                                 /*
4988                                  * Fill in changeable bma fields.
4989                                  */
4990                                 bma.eof = eof;
4991                                 bma.firstblock = *firstblock;
4992                                 bma.alen = alen;
4993                                 bma.off = aoff;
4994                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4995                                 bma.wasdel = wasdelay;
4996                                 bma.minlen = minlen;
4997                                 bma.low = flist->xbf_low;
4998                                 bma.minleft = minleft;
4999                                 /*
5000                                  * Only want to do the alignment at the
5001                                  * eof if it is userdata and allocation length
5002                                  * is larger than a stripe unit.
5003                                  */
5004                                 if (mp->m_dalign && alen >= mp->m_dalign &&
5005                                     (!(flags & XFS_BMAPI_METADATA)) &&
5006                                     (whichfork == XFS_DATA_FORK)) {
5007                                         if ((error = xfs_bmap_isaeof(ip, aoff,
5008                                                         whichfork, &bma.aeof)))
5009                                                 goto error0;
5010                                 } else
5011                                         bma.aeof = 0;
5012                                 /*
5013                                  * Call allocator.
5014                                  */
5015                                 if ((error = xfs_bmap_alloc(&bma)))
5016                                         goto error0;
5017                                 /*
5018                                  * Copy out result fields.
5019                                  */
5020                                 abno = bma.rval;
5021                                 if ((flist->xbf_low = bma.low))
5022                                         minleft = 0;
5023                                 alen = bma.alen;
5024                                 aoff = bma.off;
5025                                 ASSERT(*firstblock == NULLFSBLOCK ||
5026                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5027                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5028                                        (flist->xbf_low &&
5029                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5030                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5031                                 *firstblock = bma.firstblock;
5032                                 if (cur)
5033                                         cur->bc_private.b.firstblock =
5034                                                 *firstblock;
5035                                 if (abno == NULLFSBLOCK)
5036                                         break;
5037                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5038                                         cur = xfs_btree_init_cursor(mp,
5039                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
5040                                                 ip, whichfork);
5041                                         cur->bc_private.b.firstblock =
5042                                                 *firstblock;
5043                                         cur->bc_private.b.flist = flist;
5044                                 }
5045                                 /*
5046                                  * Bump the number of extents we've allocated
5047                                  * in this call.
5048                                  */
5049                                 nallocs++;
5050                         }
5051                         if (cur)
5052                                 cur->bc_private.b.flags =
5053                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5054                         got.br_startoff = aoff;
5055                         got.br_startblock = abno;
5056                         got.br_blockcount = alen;
5057                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5058                         /*
5059                          * Determine state of extent, and the filesystem.
5060                          * A wasdelay extent has been initialized, so
5061                          * shouldn't be flagged as unwritten.
5062                          */
5063                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5064                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5065                                         got.br_state = XFS_EXT_UNWRITTEN;
5066                         }
5067                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5068                                 firstblock, flist, &tmp_logflags, delta,
5069                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5070                         logflags |= tmp_logflags;
5071                         if (error)
5072                                 goto error0;
5073                         lastx = ifp->if_lastex;
5074                         ep = xfs_iext_get_ext(ifp, lastx);
5075                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5076                         xfs_bmbt_get_all(ep, &got);
5077                         ASSERT(got.br_startoff <= aoff);
5078                         ASSERT(got.br_startoff + got.br_blockcount >=
5079                                 aoff + alen);
5080 #ifdef DEBUG
5081                         if (flags & XFS_BMAPI_DELAY) {
5082                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5083                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5084                         }
5085                         ASSERT(got.br_state == XFS_EXT_NORM ||
5086                                got.br_state == XFS_EXT_UNWRITTEN);
5087 #endif
5088                         /*
5089                          * Fall down into the found allocated space case.
5090                          */
5091                 } else if (inhole) {
5092                         /*
5093                          * Reading in a hole.
5094                          */
5095                         mval->br_startoff = bno;
5096                         mval->br_startblock = HOLESTARTBLOCK;
5097                         mval->br_blockcount =
5098                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5099                         mval->br_state = XFS_EXT_NORM;
5100                         bno += mval->br_blockcount;
5101                         len -= mval->br_blockcount;
5102                         mval++;
5103                         n++;
5104                         continue;
5105                 }
5106                 /*
5107                  * Then deal with the allocated space we found.
5108                  */
5109                 ASSERT(ep != NULL);
5110                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5111                     (got.br_startoff + got.br_blockcount > obno)) {
5112                         if (obno > bno)
5113                                 bno = obno;
5114                         ASSERT((bno >= obno) || (n == 0));
5115                         ASSERT(bno < end);
5116                         mval->br_startoff = bno;
5117                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5118                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5119                                 mval->br_startblock = DELAYSTARTBLOCK;
5120                         } else
5121                                 mval->br_startblock =
5122                                         got.br_startblock +
5123                                         (bno - got.br_startoff);
5124                         /*
5125                          * Return the minimum of what we got and what we
5126                          * asked for for the length.  We can use the len
5127                          * variable here because it is modified below
5128                          * and we could have been there before coming
5129                          * here if the first part of the allocation
5130                          * didn't overlap what was asked for.
5131                          */
5132                         mval->br_blockcount =
5133                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5134                                         (bno - got.br_startoff));
5135                         mval->br_state = got.br_state;
5136                         ASSERT(mval->br_blockcount <= len);
5137                 } else {
5138                         *mval = got;
5139                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5140                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5141                                 mval->br_startblock = DELAYSTARTBLOCK;
5142                         }
5143                 }
5144
5145                 /*
5146                  * Check if writing previously allocated but
5147                  * unwritten extents.
5148                  */
5149                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5150                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5151                         /*
5152                          * Modify (by adding) the state flag, if writing.
5153                          */
5154                         ASSERT(mval->br_blockcount <= len);
5155                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5156                                 cur = xfs_btree_init_cursor(mp,
5157                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5158                                         ip, whichfork);
5159                                 cur->bc_private.b.firstblock =
5160                                         *firstblock;
5161                                 cur->bc_private.b.flist = flist;
5162                         }
5163                         mval->br_state = XFS_EXT_NORM;
5164                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5165                                 firstblock, flist, &tmp_logflags, delta,
5166                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5167                         logflags |= tmp_logflags;
5168                         if (error)
5169                                 goto error0;
5170                         lastx = ifp->if_lastex;
5171                         ep = xfs_iext_get_ext(ifp, lastx);
5172                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5173                         xfs_bmbt_get_all(ep, &got);
5174                         /*
5175                          * We may have combined previously unwritten
5176                          * space with written space, so generate
5177                          * another request.
5178                          */
5179                         if (mval->br_blockcount < len)
5180                                 continue;
5181                 }
5182
5183                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5184                        ((mval->br_startoff + mval->br_blockcount) <= end));
5185                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5186                        (mval->br_blockcount <= len) ||
5187                        (mval->br_startoff < obno));
5188                 bno = mval->br_startoff + mval->br_blockcount;
5189                 len = end - bno;
5190                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5191                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5192                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5193                         ASSERT(mval->br_state == mval[-1].br_state);
5194                         mval[-1].br_blockcount = mval->br_blockcount;
5195                         mval[-1].br_state = mval->br_state;
5196                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5197                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5198                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5199                            mval->br_startblock ==
5200                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5201                            ((flags & XFS_BMAPI_IGSTATE) ||
5202                                 mval[-1].br_state == mval->br_state)) {
5203                         ASSERT(mval->br_startoff ==
5204                                mval[-1].br_startoff + mval[-1].br_blockcount);
5205                         mval[-1].br_blockcount += mval->br_blockcount;
5206                 } else if (n > 0 &&
5207                            mval->br_startblock == DELAYSTARTBLOCK &&
5208                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5209                            mval->br_startoff ==
5210                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5211                         mval[-1].br_blockcount += mval->br_blockcount;
5212                         mval[-1].br_state = mval->br_state;
5213                 } else if (!((n == 0) &&
5214                              ((mval->br_startoff + mval->br_blockcount) <=
5215                               obno))) {
5216                         mval++;
5217                         n++;
5218                 }
5219                 /*
5220                  * If we're done, stop now.  Stop when we've allocated
5221                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5222                  * the transaction may get too big.
5223                  */
5224                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5225                         break;
5226                 /*
5227                  * Else go on to the next record.
5228                  */
5229                 ep = xfs_iext_get_ext(ifp, ++lastx);
5230                 prev = got;
5231                 if (lastx >= nextents)
5232                         eof = 1;
5233                 else
5234                         xfs_bmbt_get_all(ep, &got);
5235         }
5236         ifp->if_lastex = lastx;
5237         *nmap = n;
5238         /*
5239          * Transform from btree to extents, give it cur.
5240          */
5241         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5242             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5243                 ASSERT(wr && cur);
5244                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5245                         &tmp_logflags, whichfork);
5246                 logflags |= tmp_logflags;
5247                 if (error)
5248                         goto error0;
5249         }
5250         ASSERT(ifp->if_ext_max ==
5251                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5252         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5253                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5254         error = 0;
5255         if (delta && delta->xed_startoff != NULLFILEOFF) {
5256                 /* A change was actually made.
5257                  * Note that delta->xed_blockount is an offset at this
5258                  * point and needs to be converted to a block count.
5259                  */
5260                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5261                 delta->xed_blockcount -= delta->xed_startoff;
5262         }
5263 error0:
5264         /*
5265          * Log everything.  Do this after conversion, there's no point in
5266          * logging the extent records if we've converted to btree format.
5267          */
5268         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5269             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5270                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5271         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5272                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5273                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5274         /*
5275          * Log whatever the flags say, even if error.  Otherwise we might miss
5276          * detecting a case where the data is changed, there's an error,
5277          * and it's not logged so we don't shutdown when we should.
5278          */
5279         if (logflags) {
5280                 ASSERT(tp && wr);
5281                 xfs_trans_log_inode(tp, ip, logflags);
5282         }
5283         if (cur) {
5284                 if (!error) {
5285                         ASSERT(*firstblock == NULLFSBLOCK ||
5286                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5287                                XFS_FSB_TO_AGNO(mp,
5288                                        cur->bc_private.b.firstblock) ||
5289                                (flist->xbf_low &&
5290                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5291                                 XFS_FSB_TO_AGNO(mp,
5292                                         cur->bc_private.b.firstblock)));
5293                         *firstblock = cur->bc_private.b.firstblock;
5294                 }
5295                 xfs_btree_del_cursor(cur,
5296                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5297         }
5298         if (!error)
5299                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5300                         orig_nmap, *nmap);
5301         return error;
5302 }
5303
5304 /*
5305  * Map file blocks to filesystem blocks, simple version.
5306  * One block (extent) only, read-only.
5307  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5308  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5309  * was set and all the others were clear.
5310  */
5311 int                                             /* error */
5312 xfs_bmapi_single(
5313         xfs_trans_t     *tp,            /* transaction pointer */
5314         xfs_inode_t     *ip,            /* incore inode */
5315         int             whichfork,      /* data or attr fork */
5316         xfs_fsblock_t   *fsb,           /* output: mapped block */
5317         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5318 {
5319         int             eof;            /* we've hit the end of extents */
5320         int             error;          /* error return */
5321         xfs_bmbt_irec_t got;            /* current file extent record */
5322         xfs_ifork_t     *ifp;           /* inode fork pointer */
5323         xfs_extnum_t    lastx;          /* last useful extent number */
5324         xfs_bmbt_irec_t prev;           /* previous file extent record */
5325
5326         ifp = XFS_IFORK_PTR(ip, whichfork);
5327         if (unlikely(
5328             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5329             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5330                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5331                                 ip->i_mount);
5332                return XFS_ERROR(EFSCORRUPTED);
5333         }
5334         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5335                 return XFS_ERROR(EIO);
5336         XFS_STATS_INC(xs_blk_mapr);
5337         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5338             (error = xfs_iread_extents(tp, ip, whichfork)))
5339                 return error;
5340         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5341                 &prev);
5342         /*
5343          * Reading past eof, act as though there's a hole
5344          * up to end.
5345          */
5346         if (eof || got.br_startoff > bno) {
5347                 *fsb = NULLFSBLOCK;
5348                 return 0;
5349         }
5350         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5351         ASSERT(bno < got.br_startoff + got.br_blockcount);
5352         *fsb = got.br_startblock + (bno - got.br_startoff);
5353         ifp->if_lastex = lastx;
5354         return 0;
5355 }
5356
5357 /*
5358  * Unmap (remove) blocks from a file.
5359  * If nexts is nonzero then the number of extents to remove is limited to
5360  * that value.  If not all extents in the block range can be removed then
5361  * *done is set.
5362  */
5363 int                                             /* error */
5364 xfs_bunmapi(
5365         xfs_trans_t             *tp,            /* transaction pointer */
5366         struct xfs_inode        *ip,            /* incore inode */
5367         xfs_fileoff_t           bno,            /* starting offset to unmap */
5368         xfs_filblks_t           len,            /* length to unmap in file */
5369         int                     flags,          /* misc flags */
5370         xfs_extnum_t            nexts,          /* number of extents max */
5371         xfs_fsblock_t           *firstblock,    /* first allocated block
5372                                                    controls a.g. for allocs */
5373         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5374         xfs_extdelta_t          *delta,         /* o: change made to incore
5375                                                    extents */
5376         int                     *done)          /* set if not done yet */
5377 {
5378         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5379         xfs_bmbt_irec_t         del;            /* extent being deleted */
5380         int                     eof;            /* is deleting at eof */
5381         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
5382         int                     error;          /* error return value */
5383         xfs_extnum_t            extno;          /* extent number in list */
5384         xfs_bmbt_irec_t         got;            /* current extent record */
5385         xfs_ifork_t             *ifp;           /* inode fork pointer */
5386         int                     isrt;           /* freeing in rt area */
5387         xfs_extnum_t            lastx;          /* last extent index used */
5388         int                     logflags;       /* transaction logging flags */
5389         xfs_extlen_t            mod;            /* rt extent offset */
5390         xfs_mount_t             *mp;            /* mount structure */
5391         xfs_extnum_t            nextents;       /* number of file extents */
5392         xfs_bmbt_irec_t         prev;           /* previous extent record */
5393         xfs_fileoff_t           start;          /* first file offset deleted */
5394         int                     tmp_logflags;   /* partial logging flags */
5395         int                     wasdel;         /* was a delayed alloc extent */
5396         int                     whichfork;      /* data or attribute fork */
5397         int                     rsvd;           /* OK to allocate reserved blocks */
5398         xfs_fsblock_t           sum;
5399
5400         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5401         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5402                 XFS_ATTR_FORK : XFS_DATA_FORK;
5403         ifp = XFS_IFORK_PTR(ip, whichfork);
5404         if (unlikely(
5405             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5406             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5407                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5408                                  ip->i_mount);
5409                 return XFS_ERROR(EFSCORRUPTED);
5410         }
5411         mp = ip->i_mount;
5412         if (XFS_FORCED_SHUTDOWN(mp))
5413                 return XFS_ERROR(EIO);
5414         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5415         ASSERT(len > 0);
5416         ASSERT(nexts >= 0);
5417         ASSERT(ifp->if_ext_max ==
5418                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5419         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5420             (error = xfs_iread_extents(tp, ip, whichfork)))
5421                 return error;
5422         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5423         if (nextents == 0) {
5424                 *done = 1;
5425                 return 0;
5426         }
5427         XFS_STATS_INC(xs_blk_unmap);
5428         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5429         start = bno;
5430         bno = start + len - 1;
5431         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5432                 &prev);
5433         if (delta) {
5434                 delta->xed_startoff = NULLFILEOFF;
5435                 delta->xed_blockcount = 0;
5436         }
5437         /*
5438          * Check to see if the given block number is past the end of the
5439          * file, back up to the last block if so...
5440          */
5441         if (eof) {
5442                 ep = xfs_iext_get_ext(ifp, --lastx);
5443                 xfs_bmbt_get_all(ep, &got);
5444                 bno = got.br_startoff + got.br_blockcount - 1;
5445         }
5446         logflags = 0;
5447         if (ifp->if_flags & XFS_IFBROOT) {
5448                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5449                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5450                         whichfork);
5451                 cur->bc_private.b.firstblock = *firstblock;
5452                 cur->bc_private.b.flist = flist;
5453                 cur->bc_private.b.flags = 0;
5454         } else
5455                 cur = NULL;
5456         extno = 0;
5457         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5458                (nexts == 0 || extno < nexts)) {
5459                 /*
5460                  * Is the found extent after a hole in which bno lives?
5461                  * Just back up to the previous extent, if so.
5462                  */
5463                 if (got.br_startoff > bno) {
5464                         if (--lastx < 0)
5465                                 break;
5466                         ep = xfs_iext_get_ext(ifp, lastx);
5467                         xfs_bmbt_get_all(ep, &got);
5468                 }
5469                 /*
5470                  * Is the last block of this extent before the range
5471                  * we're supposed to delete?  If so, we're done.
5472                  */
5473                 bno = XFS_FILEOFF_MIN(bno,
5474                         got.br_startoff + got.br_blockcount - 1);
5475                 if (bno < start)
5476                         break;
5477                 /*
5478                  * Then deal with the (possibly delayed) allocated space
5479                  * we found.
5480                  */
5481                 ASSERT(ep != NULL);
5482                 del = got;
5483                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5484                 if (got.br_startoff < start) {
5485                         del.br_startoff = start;
5486                         del.br_blockcount -= start - got.br_startoff;
5487                         if (!wasdel)
5488                                 del.br_startblock += start - got.br_startoff;
5489                 }
5490                 if (del.br_startoff + del.br_blockcount > bno + 1)
5491                         del.br_blockcount = bno + 1 - del.br_startoff;
5492                 sum = del.br_startblock + del.br_blockcount;
5493                 if (isrt &&
5494                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5495                         /*
5496                          * Realtime extent not lined up at the end.
5497                          * The extent could have been split into written
5498                          * and unwritten pieces, or we could just be
5499                          * unmapping part of it.  But we can't really
5500                          * get rid of part of a realtime extent.
5501                          */
5502                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5503                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5504                                 /*
5505                                  * This piece is unwritten, or we're not
5506                                  * using unwritten extents.  Skip over it.
5507                                  */
5508                                 ASSERT(bno >= mod);
5509                                 bno -= mod > del.br_blockcount ?
5510                                         del.br_blockcount : mod;
5511                                 if (bno < got.br_startoff) {
5512                                         if (--lastx >= 0)
5513                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5514                                                         ifp, lastx), &got);
5515                                 }
5516                                 continue;
5517                         }
5518                         /*
5519                          * It's written, turn it unwritten.
5520                          * This is better than zeroing it.
5521                          */
5522                         ASSERT(del.br_state == XFS_EXT_NORM);
5523                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5524                         /*
5525                          * If this spans a realtime extent boundary,
5526                          * chop it back to the start of the one we end at.
5527                          */
5528                         if (del.br_blockcount > mod) {
5529                                 del.br_startoff += del.br_blockcount - mod;
5530                                 del.br_startblock += del.br_blockcount - mod;
5531                                 del.br_blockcount = mod;
5532                         }
5533                         del.br_state = XFS_EXT_UNWRITTEN;
5534                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5535                                 firstblock, flist, &logflags, delta,
5536                                 XFS_DATA_FORK, 0);
5537                         if (error)
5538                                 goto error0;
5539                         goto nodelete;
5540                 }
5541                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5542                         /*
5543                          * Realtime extent is lined up at the end but not
5544                          * at the front.  We'll get rid of full extents if
5545                          * we can.
5546                          */
5547                         mod = mp->m_sb.sb_rextsize - mod;
5548                         if (del.br_blockcount > mod) {
5549                                 del.br_blockcount -= mod;
5550                                 del.br_startoff += mod;
5551                                 del.br_startblock += mod;
5552                         } else if ((del.br_startoff == start &&
5553                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5554                                      xfs_trans_get_block_res(tp) == 0)) ||
5555                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5556                                 /*
5557                                  * Can't make it unwritten.  There isn't
5558                                  * a full extent here so just skip it.
5559                                  */
5560                                 ASSERT(bno >= del.br_blockcount);
5561                                 bno -= del.br_blockcount;
5562                                 if (bno < got.br_startoff) {
5563                                         if (--lastx >= 0)
5564                                                 xfs_bmbt_get_all(--ep, &got);
5565                                 }
5566                                 continue;
5567                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5568                                 /*
5569                                  * This one is already unwritten.
5570                                  * It must have a written left neighbor.
5571                                  * Unwrite the killed part of that one and
5572                                  * try again.
5573                                  */
5574                                 ASSERT(lastx > 0);
5575                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5576                                                 lastx - 1), &prev);
5577                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5578                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5579                                 ASSERT(del.br_startblock ==
5580                                        prev.br_startblock + prev.br_blockcount);
5581                                 if (prev.br_startoff < start) {
5582                                         mod = start - prev.br_startoff;
5583                                         prev.br_blockcount -= mod;
5584                                         prev.br_startblock += mod;
5585                                         prev.br_startoff = start;
5586                                 }
5587                                 prev.br_state = XFS_EXT_UNWRITTEN;
5588                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5589                                         &prev, firstblock, flist, &logflags,
5590                                         delta, XFS_DATA_FORK, 0);
5591                                 if (error)
5592                                         goto error0;
5593                                 goto nodelete;
5594                         } else {
5595                                 ASSERT(del.br_state == XFS_EXT_NORM);
5596                                 del.br_state = XFS_EXT_UNWRITTEN;
5597                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5598                                         &del, firstblock, flist, &logflags,
5599                                         delta, XFS_DATA_FORK, 0);
5600                                 if (error)
5601                                         goto error0;
5602                                 goto nodelete;
5603                         }
5604                 }
5605                 if (wasdel) {
5606                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5607                         /* Update realtime/data freespace, unreserve quota */
5608                         if (isrt) {
5609                                 xfs_filblks_t rtexts;
5610
5611                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5612                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5613                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5614                                                 (int64_t)rtexts, rsvd);
5615                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5616                                         NULL, ip, -((long)del.br_blockcount), 0,
5617                                         XFS_QMOPT_RES_RTBLKS);
5618                         } else {
5619                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5620                                                 (int64_t)del.br_blockcount, rsvd);
5621                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5622                                         NULL, ip, -((long)del.br_blockcount), 0,
5623                                         XFS_QMOPT_RES_REGBLKS);
5624                         }
5625                         ip->i_delayed_blks -= del.br_blockcount;
5626                         if (cur)
5627                                 cur->bc_private.b.flags |=
5628                                         XFS_BTCUR_BPRV_WASDEL;
5629                 } else if (cur)
5630                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5631                 /*
5632                  * If it's the case where the directory code is running
5633                  * with no block reservation, and the deleted block is in
5634                  * the middle of its extent, and the resulting insert
5635                  * of an extent would cause transformation to btree format,
5636                  * then reject it.  The calling code will then swap
5637                  * blocks around instead.
5638                  * We have to do this now, rather than waiting for the
5639                  * conversion to btree format, since the transaction
5640                  * will be dirty.
5641                  */
5642                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5643                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5644                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5645                     del.br_startoff > got.br_startoff &&
5646                     del.br_startoff + del.br_blockcount <
5647                     got.br_startoff + got.br_blockcount) {
5648                         error = XFS_ERROR(ENOSPC);
5649                         goto error0;
5650                 }
5651                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5652                                 &tmp_logflags, delta, whichfork, rsvd);
5653                 logflags |= tmp_logflags;
5654                 if (error)
5655                         goto error0;
5656                 bno = del.br_startoff - 1;
5657 nodelete:
5658                 lastx = ifp->if_lastex;
5659                 /*
5660                  * If not done go on to the next (previous) record.
5661                  * Reset ep in case the extents array was re-alloced.
5662                  */
5663                 ep = xfs_iext_get_ext(ifp, lastx);
5664                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5665                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5666                             xfs_bmbt_get_startoff(ep) > bno) {
5667                                 if (--lastx >= 0)
5668                                         ep = xfs_iext_get_ext(ifp, lastx);
5669                         }
5670                         if (lastx >= 0)
5671                                 xfs_bmbt_get_all(ep, &got);
5672                         extno++;
5673                 }
5674         }
5675         ifp->if_lastex = lastx;
5676         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5677         ASSERT(ifp->if_ext_max ==
5678                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5679         /*
5680          * Convert to a btree if necessary.
5681          */
5682         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5683             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5684                 ASSERT(cur == NULL);
5685                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5686                         &cur, 0, &tmp_logflags, whichfork);
5687                 logflags |= tmp_logflags;
5688                 if (error)
5689                         goto error0;
5690         }
5691         /*
5692          * transform from btree to extents, give it cur
5693          */
5694         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5695                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5696                 ASSERT(cur != NULL);
5697                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5698                         whichfork);
5699                 logflags |= tmp_logflags;
5700                 if (error)
5701                         goto error0;
5702         }
5703         /*
5704          * transform from extents to local?
5705          */
5706         ASSERT(ifp->if_ext_max ==
5707                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5708         error = 0;
5709         if (delta && delta->xed_startoff != NULLFILEOFF) {
5710                 /* A change was actually made.
5711                  * Note that delta->xed_blockount is an offset at this
5712                  * point and needs to be converted to a block count.
5713                  */
5714                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5715                 delta->xed_blockcount -= delta->xed_startoff;
5716         }
5717 error0:
5718         /*
5719          * Log everything.  Do this after conversion, there's no point in
5720          * logging the extent records if we've converted to btree format.
5721          */
5722         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5723             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5724                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5725         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5726                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5727                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5728         /*
5729          * Log inode even in the error case, if the transaction
5730          * is dirty we'll need to shut down the filesystem.
5731          */
5732         if (logflags)
5733                 xfs_trans_log_inode(tp, ip, logflags);
5734         if (cur) {
5735                 if (!error) {
5736                         *firstblock = cur->bc_private.b.firstblock;
5737                         cur->bc_private.b.allocated = 0;
5738                 }
5739                 xfs_btree_del_cursor(cur,
5740                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5741         }
5742         return error;
5743 }
5744
5745 /*
5746  * Fcntl interface to xfs_bmapi.
5747  */
5748 int                                             /* error code */
5749 xfs_getbmap(
5750         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5751         struct getbmap          *bmv,           /* user bmap structure */
5752         void                    __user *ap,     /* pointer to user's array */
5753         int                     interface)      /* interface flags */
5754 {
5755         __int64_t               bmvend;         /* last block requested */
5756         int                     error;          /* return value */
5757         __int64_t               fixlen;         /* length for -1 case */
5758         int                     i;              /* extent number */
5759         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5760         bhv_vnode_t             *vp;            /* corresponding vnode */
5761         int                     lock;           /* lock state */
5762         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5763         xfs_mount_t             *mp;            /* file system mount point */
5764         int                     nex;            /* # of user extents can do */
5765         int                     nexleft;        /* # of user extents left */
5766         int                     subnex;         /* # of bmapi's can do */
5767         int                     nmap;           /* number of map entries */
5768         struct getbmap          out;            /* output structure */
5769         int                     whichfork;      /* data or attr fork */
5770         int                     prealloced;     /* this is a file with
5771                                                  * preallocated data space */
5772         int                     sh_unwritten;   /* true, if unwritten */
5773                                                 /* extents listed separately */
5774         int                     bmapi_flags;    /* flags for xfs_bmapi */
5775         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5776
5777         vp = BHV_TO_VNODE(bdp);
5778         ip = XFS_BHVTOI(bdp);
5779         mp = ip->i_mount;
5780
5781         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5782         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5783
5784         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5785          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5786          *      bit is set for the file, generate a read event in order
5787          *      that the DMAPI application may do its thing before we return
5788          *      the extents.  Usually this means restoring user file data to
5789          *      regions of the file that look like holes.
5790          *
5791          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5792          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5793          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5794          *      could misinterpret holes in a DMAPI file as true holes,
5795          *      when in fact they may represent offline user data.
5796          */
5797         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5798             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5799             && whichfork == XFS_DATA_FORK) {
5800
5801                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5802                 if (error)
5803                         return XFS_ERROR(error);
5804         }
5805
5806         if (whichfork == XFS_ATTR_FORK) {
5807                 if (XFS_IFORK_Q(ip)) {
5808                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5809                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5810                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5811                                 return XFS_ERROR(EINVAL);
5812                 } else if (unlikely(
5813                            ip->i_d.di_aformat != 0 &&
5814                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5815                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5816                                          ip->i_mount);
5817                         return XFS_ERROR(EFSCORRUPTED);
5818                 }
5819         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5820                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5821                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5822                 return XFS_ERROR(EINVAL);
5823         if (whichfork == XFS_DATA_FORK) {
5824                 if (xfs_get_extsz_hint(ip) ||
5825                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5826                         prealloced = 1;
5827                         fixlen = XFS_MAXIOFFSET(mp);
5828                 } else {
5829                         prealloced = 0;
5830                         fixlen = ip->i_size;
5831                 }
5832         } else {
5833                 prealloced = 0;
5834                 fixlen = 1LL << 32;
5835         }
5836
5837         if (bmv->bmv_length == -1) {
5838                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5839                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5840                                         (__int64_t)0);
5841         } else if (bmv->bmv_length < 0)
5842                 return XFS_ERROR(EINVAL);
5843         if (bmv->bmv_length == 0) {
5844                 bmv->bmv_entries = 0;
5845                 return 0;
5846         }
5847         nex = bmv->bmv_count - 1;
5848         if (nex <= 0)
5849                 return XFS_ERROR(EINVAL);
5850         bmvend = bmv->bmv_offset + bmv->bmv_length;
5851
5852         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5853
5854         if (whichfork == XFS_DATA_FORK &&
5855                 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5856                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5857                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
5858         }
5859
5860         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5861
5862         lock = xfs_ilock_map_shared(ip);
5863
5864         /*
5865          * Don't let nex be bigger than the number of extents
5866          * we can have assuming alternating holes and real extents.
5867          */
5868         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5869                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5870
5871         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5872                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5873
5874         /*
5875          * Allocate enough space to handle "subnex" maps at a time.
5876          */
5877         subnex = 16;
5878         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5879
5880         bmv->bmv_entries = 0;
5881
5882         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5883                 error = 0;
5884                 goto unlock_and_return;
5885         }
5886
5887         nexleft = nex;
5888
5889         do {
5890                 nmap = (nexleft > subnex) ? subnex : nexleft;
5891                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5892                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5893                                   bmapi_flags, NULL, 0, map, &nmap,
5894                                   NULL, NULL);
5895                 if (error)
5896                         goto unlock_and_return;
5897                 ASSERT(nmap <= subnex);
5898
5899                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5900                         nexleft--;
5901                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5902                                         BMV_OF_PREALLOC : 0;
5903                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5904                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5905                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5906                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5907                            ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5908                              whichfork == XFS_ATTR_FORK )) {
5909                                 /*
5910                                  * came to hole at end of file or the end of
5911                                    attribute fork
5912                                  */
5913                                 goto unlock_and_return;
5914                         } else {
5915                                 out.bmv_block =
5916                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5917                                         -1 :
5918                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5919
5920                                 /* return either getbmap/getbmapx structure. */
5921                                 if (interface & BMV_IF_EXTENDED) {
5922                                         struct  getbmapx        outx;
5923
5924                                         GETBMAP_CONVERT(out,outx);
5925                                         outx.bmv_oflags = oflags;
5926                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5927                                         if (copy_to_user(ap, &outx,
5928                                                         sizeof(outx))) {
5929                                                 error = XFS_ERROR(EFAULT);
5930                                                 goto unlock_and_return;
5931                                         }
5932                                 } else {
5933                                         if (copy_to_user(ap, &out,
5934                                                         sizeof(out))) {
5935                                                 error = XFS_ERROR(EFAULT);
5936                                                 goto unlock_and_return;
5937                                         }
5938                                 }
5939                                 bmv->bmv_offset =
5940                                         out.bmv_offset + out.bmv_length;
5941                                 bmv->bmv_length = MAX((__int64_t)0,
5942                                         (__int64_t)(bmvend - bmv->bmv_offset));
5943                                 bmv->bmv_entries++;
5944                                 ap = (interface & BMV_IF_EXTENDED) ?
5945                                                 (void __user *)
5946                                         ((struct getbmapx __user *)ap + 1) :
5947                                                 (void __user *)
5948                                         ((struct getbmap __user *)ap + 1);
5949                         }
5950                 }
5951         } while (nmap && nexleft && bmv->bmv_length);
5952
5953 unlock_and_return:
5954         xfs_iunlock_map_shared(ip, lock);
5955         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5956
5957         kmem_free(map, subnex * sizeof(*map));
5958
5959         return error;
5960 }
5961
5962 /*
5963  * Check the last inode extent to determine whether this allocation will result
5964  * in blocks being allocated at the end of the file. When we allocate new data
5965  * blocks at the end of the file which do not start at the previous data block,
5966  * we will try to align the new blocks at stripe unit boundaries.
5967  */
5968 STATIC int                              /* error */
5969 xfs_bmap_isaeof(
5970         xfs_inode_t     *ip,            /* incore inode pointer */
5971         xfs_fileoff_t   off,            /* file offset in fsblocks */
5972         int             whichfork,      /* data or attribute fork */
5973         char            *aeof)          /* return value */
5974 {
5975         int             error;          /* error return value */
5976         xfs_ifork_t     *ifp;           /* inode fork pointer */
5977         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
5978         xfs_extnum_t    nextents;       /* number of file extents */
5979         xfs_bmbt_irec_t s;              /* expanded extent record */
5980
5981         ASSERT(whichfork == XFS_DATA_FORK);
5982         ifp = XFS_IFORK_PTR(ip, whichfork);
5983         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5984             (error = xfs_iread_extents(NULL, ip, whichfork)))
5985                 return error;
5986         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5987         if (nextents == 0) {
5988                 *aeof = 1;
5989                 return 0;
5990         }
5991         /*
5992          * Go to the last extent
5993          */
5994         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5995         xfs_bmbt_get_all(lastrec, &s);
5996         /*
5997          * Check we are allocating in the last extent (for delayed allocations)
5998          * or past the last extent for non-delayed allocations.
5999          */
6000         *aeof = (off >= s.br_startoff &&
6001                  off < s.br_startoff + s.br_blockcount &&
6002                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6003                 off >= s.br_startoff + s.br_blockcount;
6004         return 0;
6005 }
6006
6007 /*
6008  * Check if the endoff is outside the last extent. If so the caller will grow
6009  * the allocation to a stripe unit boundary.
6010  */
6011 int                                     /* error */
6012 xfs_bmap_eof(
6013         xfs_inode_t     *ip,            /* incore inode pointer */
6014         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6015         int             whichfork,      /* data or attribute fork */
6016         int             *eof)           /* result value */
6017 {
6018         xfs_fsblock_t   blockcount;     /* extent block count */
6019         int             error;          /* error return value */
6020         xfs_ifork_t     *ifp;           /* inode fork pointer */
6021         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
6022         xfs_extnum_t    nextents;       /* number of file extents */
6023         xfs_fileoff_t   startoff;       /* extent starting file offset */
6024
6025         ASSERT(whichfork == XFS_DATA_FORK);
6026         ifp = XFS_IFORK_PTR(ip, whichfork);
6027         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6028             (error = xfs_iread_extents(NULL, ip, whichfork)))
6029                 return error;
6030         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6031         if (nextents == 0) {
6032                 *eof = 1;
6033                 return 0;
6034         }
6035         /*
6036          * Go to the last extent
6037          */
6038         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6039         startoff = xfs_bmbt_get_startoff(lastrec);
6040         blockcount = xfs_bmbt_get_blockcount(lastrec);
6041         *eof = endoff >= startoff + blockcount;
6042         return 0;
6043 }
6044
6045 #ifdef DEBUG
6046 STATIC
6047 xfs_buf_t *
6048 xfs_bmap_get_bp(
6049         xfs_btree_cur_t         *cur,
6050         xfs_fsblock_t           bno)
6051 {
6052         int i;
6053         xfs_buf_t *bp;
6054
6055         if (!cur)
6056                 return(NULL);
6057
6058         bp = NULL;
6059         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6060                 bp = cur->bc_bufs[i];
6061                 if (!bp) break;
6062                 if (XFS_BUF_ADDR(bp) == bno)
6063                         break;  /* Found it */
6064         }
6065         if (i == XFS_BTREE_MAXLEVELS)
6066                 bp = NULL;
6067
6068         if (!bp) { /* Chase down all the log items to see if the bp is there */
6069                 xfs_log_item_chunk_t    *licp;
6070                 xfs_trans_t             *tp;
6071
6072                 tp = cur->bc_tp;
6073                 licp = &tp->t_items;
6074                 while (!bp && licp != NULL) {
6075                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
6076                                 licp = licp->lic_next;
6077                                 continue;
6078                         }
6079                         for (i = 0; i < licp->lic_unused; i++) {
6080                                 xfs_log_item_desc_t     *lidp;
6081                                 xfs_log_item_t          *lip;
6082                                 xfs_buf_log_item_t      *bip;
6083                                 xfs_buf_t               *lbp;
6084
6085                                 if (XFS_LIC_ISFREE(licp, i)) {
6086                                         continue;
6087                                 }
6088
6089                                 lidp = XFS_LIC_SLOT(licp, i);
6090                                 lip = lidp->lid_item;
6091                                 if (lip->li_type != XFS_LI_BUF)
6092                                         continue;
6093
6094                                 bip = (xfs_buf_log_item_t *)lip;
6095                                 lbp = bip->bli_buf;
6096
6097                                 if (XFS_BUF_ADDR(lbp) == bno) {
6098                                         bp = lbp;
6099                                         break; /* Found it */
6100                                 }
6101                         }
6102                         licp = licp->lic_next;
6103                 }
6104         }
6105         return(bp);
6106 }
6107
6108 void
6109 xfs_check_block(
6110         xfs_bmbt_block_t        *block,
6111         xfs_mount_t             *mp,
6112         int                     root,
6113         short                   sz)
6114 {
6115         int                     i, j, dmxr;
6116         __be64                  *pp, *thispa;   /* pointer to block address */
6117         xfs_bmbt_key_t          *prevp, *keyp;
6118
6119         ASSERT(be16_to_cpu(block->bb_level) > 0);
6120
6121         prevp = NULL;
6122         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6123                 dmxr = mp->m_bmap_dmxr[0];
6124
6125                 if (root) {
6126                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6127                 } else {
6128                         keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6129                 }
6130
6131                 if (prevp) {
6132                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6133                 }
6134                 prevp = keyp;
6135
6136                 /*
6137                  * Compare the block numbers to see if there are dups.
6138                  */
6139
6140                 if (root) {
6141                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6142                 } else {
6143                         pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6144                 }
6145                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6146                         if (root) {
6147                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6148                         } else {
6149                                 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6150                                                             dmxr);
6151                         }
6152                         if (*thispa == *pp) {
6153                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6154                                         __FUNCTION__, j, i,
6155                                         (unsigned long long)be64_to_cpu(*thispa));
6156                                 panic("%s: ptrs are equal in node\n",
6157                                         __FUNCTION__);
6158                         }
6159                 }
6160         }
6161 }
6162
6163 /*
6164  * Check that the extents for the inode ip are in the right order in all
6165  * btree leaves.
6166  */
6167
6168 STATIC void
6169 xfs_bmap_check_leaf_extents(
6170         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6171         xfs_inode_t             *ip,            /* incore inode pointer */
6172         int                     whichfork)      /* data or attr fork */
6173 {
6174         xfs_bmbt_block_t        *block; /* current btree block */
6175         xfs_fsblock_t           bno;    /* block # of "block" */
6176         xfs_buf_t               *bp;    /* buffer for "block" */
6177         int                     error;  /* error return value */
6178         xfs_extnum_t            i=0, j; /* index into the extents list */
6179         xfs_ifork_t             *ifp;   /* fork structure */
6180         int                     level;  /* btree level, for checking */
6181         xfs_mount_t             *mp;    /* file system mount structure */
6182         __be64                  *pp;    /* pointer to block address */
6183         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6184         xfs_bmbt_rec_t          *lastp; /* pointer to previous extent */
6185         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6186         int                     bp_release = 0;
6187
6188         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6189                 return;
6190         }
6191
6192         bno = NULLFSBLOCK;
6193         mp = ip->i_mount;
6194         ifp = XFS_IFORK_PTR(ip, whichfork);
6195         block = ifp->if_broot;
6196         /*
6197          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6198          */
6199         level = be16_to_cpu(block->bb_level);
6200         ASSERT(level > 0);
6201         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6202         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6203         bno = be64_to_cpu(*pp);
6204
6205         ASSERT(bno != NULLDFSBNO);
6206         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6207         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6208
6209         /*
6210          * Go down the tree until leaf level is reached, following the first
6211          * pointer (leftmost) at each level.
6212          */
6213         while (level-- > 0) {
6214                 /* See if buf is in cur first */
6215                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6216                 if (bp) {
6217                         bp_release = 0;
6218                 } else {
6219                         bp_release = 1;
6220                 }
6221                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6222                                 XFS_BMAP_BTREE_REF)))
6223                         goto error_norelse;
6224                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6225                 XFS_WANT_CORRUPTED_GOTO(
6226                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6227                         error0);
6228                 if (level == 0)
6229                         break;
6230
6231                 /*
6232                  * Check this block for basic sanity (increasing keys and
6233                  * no duplicate blocks).
6234                  */
6235
6236                 xfs_check_block(block, mp, 0, 0);
6237                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6238                 bno = be64_to_cpu(*pp);
6239                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6240                 if (bp_release) {
6241                         bp_release = 0;
6242                         xfs_trans_brelse(NULL, bp);
6243                 }
6244         }
6245
6246         /*
6247          * Here with bp and block set to the leftmost leaf node in the tree.
6248          */
6249         i = 0;
6250
6251         /*
6252          * Loop over all leaf nodes checking that all extents are in the right order.
6253          */
6254         lastp = NULL;
6255         for (;;) {
6256                 xfs_fsblock_t   nextbno;
6257                 xfs_extnum_t    num_recs;
6258
6259
6260                 num_recs = be16_to_cpu(block->bb_numrecs);
6261
6262                 /*
6263                  * Read-ahead the next leaf block, if any.
6264                  */
6265
6266                 nextbno = be64_to_cpu(block->bb_rightsib);
6267
6268                 /*
6269                  * Check all the extents to make sure they are OK.
6270                  * If we had a previous block, the last entry should
6271                  * conform with the first entry in this one.
6272                  */
6273
6274                 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6275                 for (j = 1; j < num_recs; j++) {
6276                         nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6277                         if (lastp) {
6278                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6279                                         (void *)lastp, (void *)ep);
6280                         }
6281                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6282                                 (void *)(nextp));
6283                         lastp = ep;
6284                         ep = nextp;
6285                 }
6286
6287                 i += num_recs;
6288                 if (bp_release) {
6289                         bp_release = 0;
6290                         xfs_trans_brelse(NULL, bp);
6291                 }
6292                 bno = nextbno;
6293                 /*
6294                  * If we've reached the end, stop.
6295                  */
6296                 if (bno == NULLFSBLOCK)
6297                         break;
6298
6299                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6300                 if (bp) {
6301                         bp_release = 0;
6302                 } else {
6303                         bp_release = 1;
6304                 }
6305                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6306                                 XFS_BMAP_BTREE_REF)))
6307                         goto error_norelse;
6308                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6309         }
6310         if (bp_release) {
6311                 bp_release = 0;
6312                 xfs_trans_brelse(NULL, bp);
6313         }
6314         return;
6315
6316 error0:
6317         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6318         if (bp_release)
6319                 xfs_trans_brelse(NULL, bp);
6320 error_norelse:
6321         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6322                 __FUNCTION__, i);
6323         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6324         return;
6325 }
6326 #endif
6327
6328 /*
6329  * Count fsblocks of the given fork.
6330  */
6331 int                                             /* error */
6332 xfs_bmap_count_blocks(
6333         xfs_trans_t             *tp,            /* transaction pointer */
6334         xfs_inode_t             *ip,            /* incore inode */
6335         int                     whichfork,      /* data or attr fork */
6336         int                     *count)         /* out: count of blocks */
6337 {
6338         xfs_bmbt_block_t        *block; /* current btree block */
6339         xfs_fsblock_t           bno;    /* block # of "block" */
6340         xfs_ifork_t             *ifp;   /* fork structure */
6341         int                     level;  /* btree level, for checking */
6342         xfs_mount_t             *mp;    /* file system mount structure */
6343         __be64                  *pp;    /* pointer to block address */
6344
6345         bno = NULLFSBLOCK;
6346         mp = ip->i_mount;
6347         ifp = XFS_IFORK_PTR(ip, whichfork);
6348         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6349                 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6350                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6351                         count) < 0)) {
6352                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6353                                          XFS_ERRLEVEL_LOW, mp);
6354                         return XFS_ERROR(EFSCORRUPTED);
6355                 }
6356                 return 0;
6357         }
6358
6359         /*
6360          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6361          */
6362         block = ifp->if_broot;
6363         level = be16_to_cpu(block->bb_level);
6364         ASSERT(level > 0);
6365         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6366         bno = be64_to_cpu(*pp);
6367         ASSERT(bno != NULLDFSBNO);
6368         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6369         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6370
6371         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6372                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6373                                  mp);
6374                 return XFS_ERROR(EFSCORRUPTED);
6375         }
6376
6377         return 0;
6378 }
6379
6380 /*
6381  * Recursively walks each level of a btree
6382  * to count total fsblocks is use.
6383  */
6384 int                                     /* error */
6385 xfs_bmap_count_tree(
6386         xfs_mount_t     *mp,            /* file system mount point */
6387         xfs_trans_t     *tp,            /* transaction pointer */
6388         xfs_ifork_t     *ifp,           /* inode fork pointer */
6389         xfs_fsblock_t   blockno,        /* file system block number */
6390         int             levelin,        /* level in btree */
6391         int             *count)         /* Count of blocks */
6392 {
6393         int                     error;
6394         xfs_buf_t               *bp, *nbp;
6395         int                     level = levelin;
6396         __be64                  *pp;
6397         xfs_fsblock_t           bno = blockno;
6398         xfs_fsblock_t           nextbno;
6399         xfs_bmbt_block_t        *block, *nextblock;
6400         int                     numrecs;
6401
6402         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6403                 return error;
6404         *count += 1;
6405         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6406
6407         if (--level) {
6408                 /* Not at node above leafs, count this level of nodes */
6409                 nextbno = be64_to_cpu(block->bb_rightsib);
6410                 while (nextbno != NULLFSBLOCK) {
6411                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6412                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6413                                 return error;
6414                         *count += 1;
6415                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6416                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6417                         xfs_trans_brelse(tp, nbp);
6418                 }
6419
6420                 /* Dive to the next level */
6421                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6422                 bno = be64_to_cpu(*pp);
6423                 if (unlikely((error =
6424                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6425                         xfs_trans_brelse(tp, bp);
6426                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6427                                          XFS_ERRLEVEL_LOW, mp);
6428                         return XFS_ERROR(EFSCORRUPTED);
6429                 }
6430                 xfs_trans_brelse(tp, bp);
6431         } else {
6432                 /* count all level 1 nodes and their leaves */
6433                 for (;;) {
6434                         nextbno = be64_to_cpu(block->bb_rightsib);
6435                         numrecs = be16_to_cpu(block->bb_numrecs);
6436                         if (unlikely(xfs_bmap_disk_count_leaves(0,
6437                                         block, numrecs, count) < 0)) {
6438                                 xfs_trans_brelse(tp, bp);
6439                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6440                                                  XFS_ERRLEVEL_LOW, mp);
6441                                 return XFS_ERROR(EFSCORRUPTED);
6442                         }
6443                         xfs_trans_brelse(tp, bp);
6444                         if (nextbno == NULLFSBLOCK)
6445                                 break;
6446                         bno = nextbno;
6447                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6448                                 XFS_BMAP_BTREE_REF)))
6449                                 return error;
6450                         *count += 1;
6451                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6452                 }
6453         }
6454         return 0;
6455 }
6456
6457 /*
6458  * Count leaf blocks given a range of extent records.
6459  */
6460 int
6461 xfs_bmap_count_leaves(
6462         xfs_ifork_t             *ifp,
6463         xfs_extnum_t            idx,
6464         int                     numrecs,
6465         int                     *count)
6466 {
6467         int             b;
6468         xfs_bmbt_rec_t  *frp;
6469
6470         for (b = 0; b < numrecs; b++) {
6471                 frp = xfs_iext_get_ext(ifp, idx + b);
6472                 *count += xfs_bmbt_get_blockcount(frp);
6473         }
6474         return 0;
6475 }
6476
6477 /*
6478  * Count leaf blocks given a range of extent records originally
6479  * in btree format.
6480  */
6481 int
6482 xfs_bmap_disk_count_leaves(
6483         xfs_extnum_t            idx,
6484         xfs_bmbt_block_t        *block,
6485         int                     numrecs,
6486         int                     *count)
6487 {
6488         int             b;
6489         xfs_bmbt_rec_t  *frp;
6490
6491         for (b = 1; b <= numrecs; b++) {
6492                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6493                 *count += xfs_bmbt_disk_get_blockcount(frp);
6494         }
6495         return 0;
6496 }