fs/xfs/xfs_itable.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 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_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dir2_sf.h"
  34 #include "xfs_attr_sf.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_ialloc.h"
  38 #include "xfs_itable.h"
  39 #include "xfs_error.h"
  40 #include "xfs_btree.h"
  41
  42 int
  43 xfs_internal_inum(
  44         xfs_mount_t     *mp,
  45         xfs_ino_t       ino)
  46 {
  47         return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
  48                 (xfs_sb_version_hasquota(&mp->m_sb) &&
  49                  (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino)));
  50 }
  51
  52 STATIC int
  53 xfs_bulkstat_one_iget(
  54         xfs_mount_t     *mp,            /* mount point for filesystem */
  55         xfs_ino_t       ino,            /* inode number to get data for */
  56         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  57         xfs_bstat_t     *buf,           /* return buffer */
  58         int             *stat)          /* BULKSTAT_RV_... */
  59 {
  60         xfs_icdinode_t  *dic;   /* dinode core info pointer */
  61         xfs_inode_t     *ip;            /* incore inode pointer */
  62         int             error;
  63
  64         error = xfs_iget(mp, NULL, ino,
  65                          XFS_IGET_BULKSTAT, XFS_ILOCK_SHARED, &ip, bno);
  66         if (error) {
  67                 *stat = BULKSTAT_RV_NOTHING;
  68                 return error;
  69         }
  70
  71         ASSERT(ip != NULL);
  72         ASSERT(ip->i_imap.im_blkno != 0);
  73
  74         dic = &ip->i_d;
  75
  76         /* xfs_iget returns the following without needing
  77          * further change.
  78          */
  79         buf->bs_nlink = dic->di_nlink;
  80         buf->bs_projid = dic->di_projid;
  81         buf->bs_ino = ino;
  82         buf->bs_mode = dic->di_mode;
  83         buf->bs_uid = dic->di_uid;
  84         buf->bs_gid = dic->di_gid;
  85         buf->bs_size = dic->di_size;
  86         vn_atime_to_bstime(VFS_I(ip), &buf->bs_atime);
  87         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
  88         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
  89         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
  90         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
  91         buf->bs_xflags = xfs_ip2xflags(ip);
  92         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
  93         buf->bs_extents = dic->di_nextents;
  94         buf->bs_gen = dic->di_gen;
  95         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
  96         buf->bs_dmevmask = dic->di_dmevmask;
  97         buf->bs_dmstate = dic->di_dmstate;
  98         buf->bs_aextents = dic->di_anextents;
  99
 100         switch (dic->di_format) {
 101         case XFS_DINODE_FMT_DEV:
 102                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 103                 buf->bs_blksize = BLKDEV_IOSIZE;
 104                 buf->bs_blocks = 0;
 105                 break;
 106         case XFS_DINODE_FMT_LOCAL:
 107         case XFS_DINODE_FMT_UUID:
 108                 buf->bs_rdev = 0;
 109                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 110                 buf->bs_blocks = 0;
 111                 break;
 112         case XFS_DINODE_FMT_EXTENTS:
 113         case XFS_DINODE_FMT_BTREE:
 114                 buf->bs_rdev = 0;
 115                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 116                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 117                 break;
 118         }
 119
 120         xfs_iput(ip, XFS_ILOCK_SHARED);
 121         return error;
 122 }
 123
 124 STATIC void
 125 xfs_bulkstat_one_dinode(
 126         xfs_mount_t     *mp,            /* mount point for filesystem */
 127         xfs_ino_t       ino,            /* inode number to get data for */
 128         xfs_dinode_t    *dic,           /* dinode inode pointer */
 129         xfs_bstat_t     *buf)           /* return buffer */
 130 {
 131         /*
 132          * The inode format changed when we moved the link count and
 133          * made it 32 bits long.  If this is an old format inode,
 134          * convert it in memory to look like a new one.  If it gets
 135          * flushed to disk we will convert back before flushing or
 136          * logging it.  We zero out the new projid field and the old link
 137          * count field.  We'll handle clearing the pad field (the remains
 138          * of the old uuid field) when we actually convert the inode to
 139          * the new format. We don't change the version number so that we
 140          * can distinguish this from a real new format inode.
 141          */
 142         if (dic->di_version == 1) {
 143                 buf->bs_nlink = be16_to_cpu(dic->di_onlink);
 144                 buf->bs_projid = 0;
 145         } else {
 146                 buf->bs_nlink = be32_to_cpu(dic->di_nlink);
 147                 buf->bs_projid = be16_to_cpu(dic->di_projid);
 148         }
 149
 150         buf->bs_ino = ino;
 151         buf->bs_mode = be16_to_cpu(dic->di_mode);
 152         buf->bs_uid = be32_to_cpu(dic->di_uid);
 153         buf->bs_gid = be32_to_cpu(dic->di_gid);
 154         buf->bs_size = be64_to_cpu(dic->di_size);
 155         buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
 156         buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
 157         buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
 158         buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
 159         buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
 160         buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
 161         buf->bs_xflags = xfs_dic2xflags(dic);
 162         buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
 163         buf->bs_extents = be32_to_cpu(dic->di_nextents);
 164         buf->bs_gen = be32_to_cpu(dic->di_gen);
 165         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 166         buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
 167         buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
 168         buf->bs_aextents = be16_to_cpu(dic->di_anextents);
 169
 170         switch (dic->di_format) {
 171         case XFS_DINODE_FMT_DEV:
 172                 buf->bs_rdev = xfs_dinode_get_rdev(dic);
 173                 buf->bs_blksize = BLKDEV_IOSIZE;
 174                 buf->bs_blocks = 0;
 175                 break;
 176         case XFS_DINODE_FMT_LOCAL:
 177         case XFS_DINODE_FMT_UUID:
 178                 buf->bs_rdev = 0;
 179                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 180                 buf->bs_blocks = 0;
 181                 break;
 182         case XFS_DINODE_FMT_EXTENTS:
 183         case XFS_DINODE_FMT_BTREE:
 184                 buf->bs_rdev = 0;
 185                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 186                 buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
 187                 break;
 188         }
 189 }
 190
 191 STATIC int
 192 xfs_bulkstat_one_fmt(
 193         void                    __user *ubuffer,
 194         const xfs_bstat_t       *buffer)
 195 {
 196         if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
 197                 return -EFAULT;
 198         return sizeof(*buffer);
 199 }
 200
 201 /*
 202  * Return stat information for one inode.
 203  * Return 0 if ok, else errno.
 204  */
 205 int                                     /* error status */
 206 xfs_bulkstat_one_int(
 207         xfs_mount_t     *mp,            /* mount point for filesystem */
 208         xfs_ino_t       ino,            /* inode number to get data for */
 209         void            __user *buffer, /* buffer to place output in */
 210         int             ubsize,         /* size of buffer */
 211         bulkstat_one_fmt_pf formatter,  /* formatter, copy to user */
 212         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 213         int             *ubused,        /* bytes used by me */
 214         void            *dibuff,        /* on-disk inode buffer */
 215         int             *stat)          /* BULKSTAT_RV_... */
 216 {
 217         xfs_bstat_t     *buf;           /* return buffer */
 218         int             error = 0;      /* error value */
 219         xfs_dinode_t    *dip;           /* dinode inode pointer */
 220
 221         dip = (xfs_dinode_t *)dibuff;
 222         *stat = BULKSTAT_RV_NOTHING;
 223
 224         if (!buffer || xfs_internal_inum(mp, ino))
 225                 return XFS_ERROR(EINVAL);
 226         if (ubsize < sizeof(*buf))
 227                 return XFS_ERROR(ENOMEM);
 228
 229         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 230
 231         if (dip == NULL) {
 232                 /* We're not being passed a pointer to a dinode.  This happens
 233                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 234                  */
 235                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 236                 if (error)
 237                         goto out_free;
 238         } else {
 239                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 240         }
 241
 242         error = formatter(buffer, buf);
 243         if (error < 0)  {
 244                 error = EFAULT;
 245                 goto out_free;
 246         }
 247
 248         *stat = BULKSTAT_RV_DIDONE;
 249         if (ubused)
 250                 *ubused = error;
 251
 252  out_free:
 253         kmem_free(buf);
 254         return error;
 255 }
 256
 257 int
 258 xfs_bulkstat_one(
 259         xfs_mount_t     *mp,            /* mount point for filesystem */
 260         xfs_ino_t       ino,            /* inode number to get data for */
 261         void            __user *buffer, /* buffer to place output in */
 262         int             ubsize,         /* size of buffer */
 263         void            *private_data,  /* my private data */
 264         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 265         int             *ubused,        /* bytes used by me */
 266         void            *dibuff,        /* on-disk inode buffer */
 267         int             *stat)          /* BULKSTAT_RV_... */
 268 {
 269         return xfs_bulkstat_one_int(mp, ino, buffer, ubsize,
 270                                     xfs_bulkstat_one_fmt, bno,
 271                                     ubused, dibuff, stat);
 272 }
 273
 274 /*
 275  * Test to see whether we can use the ondisk inode directly, based
 276  * on the given bulkstat flags, filling in dipp accordingly.
 277  * Returns zero if the inode is dodgey.
 278  */
 279 STATIC int
 280 xfs_bulkstat_use_dinode(
 281         xfs_mount_t     *mp,
 282         int             flags,
 283         xfs_buf_t       *bp,
 284         int             clustidx,
 285         xfs_dinode_t    **dipp)
 286 {
 287         xfs_dinode_t    *dip;
 288         unsigned int    aformat;
 289
 290         *dipp = NULL;
 291         if (!bp || (flags & BULKSTAT_FG_IGET))
 292                 return 1;
 293         dip = (xfs_dinode_t *)
 294                         xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
 295         /*
 296          * Check the buffer containing the on-disk inode for di_mode == 0.
 297          * This is to prevent xfs_bulkstat from picking up just reclaimed
 298          * inodes that have their in-core state initialized but not flushed
 299          * to disk yet. This is a temporary hack that would require a proper
 300          * fix in the future.
 301          */
 302         if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC ||
 303             !XFS_DINODE_GOOD_VERSION(dip->di_version) ||
 304             !dip->di_mode)
 305                 return 0;
 306         if (flags & BULKSTAT_FG_QUICK) {
 307                 *dipp = dip;
 308                 return 1;
 309         }
 310         /* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
 311         aformat = dip->di_aformat;
 312         if ((XFS_DFORK_Q(dip) == 0) ||
 313             (aformat == XFS_DINODE_FMT_LOCAL) ||
 314             (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_anextents)) {
 315                 *dipp = dip;
 316                 return 1;
 317         }
 318         return 1;
 319 }
 320
 321 #define XFS_BULKSTAT_UBLEFT(ubleft)     ((ubleft) >= statstruct_size)
 322
 323 /*
 324  * Return stat information in bulk (by-inode) for the filesystem.
 325  */
 326 int                                     /* error status */
 327 xfs_bulkstat(
 328         xfs_mount_t             *mp,    /* mount point for filesystem */
 329         xfs_ino_t               *lastinop, /* last inode returned */
 330         int                     *ubcountp, /* size of buffer/count returned */
 331         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 332         void                    *private_data,/* private data for formatter */
 333         size_t                  statstruct_size, /* sizeof struct filling */
 334         char                    __user *ubuffer, /* buffer with inode stats */
 335         int                     flags,  /* defined in xfs_itable.h */
 336         int                     *done)  /* 1 if there are more stats to get */
 337 {
 338         xfs_agblock_t           agbno=0;/* allocation group block number */
 339         xfs_buf_t               *agbp;  /* agi header buffer */
 340         xfs_agi_t               *agi;   /* agi header data */
 341         xfs_agino_t             agino;  /* inode # in allocation group */
 342         xfs_agnumber_t          agno;   /* allocation group number */
 343         xfs_daddr_t             bno;    /* inode cluster start daddr */
 344         int                     chunkidx; /* current index into inode chunk */
 345         int                     clustidx; /* current index into inode cluster */
 346         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 347         int                     end_of_ag; /* set if we've seen the ag end */
 348         int                     error;  /* error code */
 349         int                     fmterror;/* bulkstat formatter result */
 350         __int32_t               gcnt;   /* current btree rec's count */
 351         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 352         xfs_agino_t             gino;   /* current btree rec's start inode */
 353         int                     i;      /* loop index */
 354         int                     icount; /* count of inodes good in irbuf */
 355         size_t                  irbsize; /* size of irec buffer in bytes */
 356         xfs_ino_t               ino;    /* inode number (filesystem) */
 357         xfs_inobt_rec_incore_t  *irbp;  /* current irec buffer pointer */
 358         xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
 359         xfs_inobt_rec_incore_t  *irbufend; /* end of good irec buffer entries */
 360         xfs_ino_t               lastino; /* last inode number returned */
 361         int                     nbcluster; /* # of blocks in a cluster */
 362         int                     nicluster; /* # of inodes in a cluster */
 363         int                     nimask; /* mask for inode clusters */
 364         int                     nirbuf; /* size of irbuf */
 365         int                     rval;   /* return value error code */
 366         int                     tmp;    /* result value from btree calls */
 367         int                     ubcount; /* size of user's buffer */
 368         int                     ubleft; /* bytes left in user's buffer */
 369         char                    __user *ubufp;  /* pointer into user's buffer */
 370         int                     ubelem; /* spaces used in user's buffer */
 371         int                     ubused; /* bytes used by formatter */
 372         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 373         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 374
 375         /*
 376          * Get the last inode value, see if there's nothing to do.
 377          */
 378         ino = (xfs_ino_t)*lastinop;
 379         lastino = ino;
 380         dip = NULL;
 381         agno = XFS_INO_TO_AGNO(mp, ino);
 382         agino = XFS_INO_TO_AGINO(mp, ino);
 383         if (agno >= mp->m_sb.sb_agcount ||
 384             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 385                 *done = 1;
 386                 *ubcountp = 0;
 387                 return 0;
 388         }
 389         if (!ubcountp || *ubcountp <= 0) {
 390                 return EINVAL;
 391         }
 392         ubcount = *ubcountp; /* statstruct's */
 393         ubleft = ubcount * statstruct_size; /* bytes */
 394         *ubcountp = ubelem = 0;
 395         *done = 0;
 396         fmterror = 0;
 397         ubufp = ubuffer;
 398         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 399                 mp->m_sb.sb_inopblock :
 400                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 401         nimask = ~(nicluster - 1);
 402         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 403         irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
 404                                    KM_SLEEP | KM_MAYFAIL | KM_LARGE);
 405         nirbuf = irbsize / sizeof(*irbuf);
 406
 407         /*
 408          * Loop over the allocation groups, starting from the last
 409          * inode returned; 0 means start of the allocation group.
 410          */
 411         rval = 0;
 412         while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
 413                 cond_resched();
 414                 bp = NULL;
 415                 down_read(&mp->m_peraglock);
 416                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 417                 up_read(&mp->m_peraglock);
 418                 if (error) {
 419                         /*
 420                          * Skip this allocation group and go to the next one.
 421                          */
 422                         agno++;
 423                         agino = 0;
 424                         continue;
 425                 }
 426                 agi = XFS_BUF_TO_AGI(agbp);
 427                 /*
 428                  * Allocate and initialize a btree cursor for ialloc btree.
 429                  */
 430                 cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
 431                 irbp = irbuf;
 432                 irbufend = irbuf + nirbuf;
 433                 end_of_ag = 0;
 434                 /*
 435                  * If we're returning in the middle of an allocation group,
 436                  * we need to get the remainder of the chunk we're in.
 437                  */
 438                 if (agino > 0) {
 439                         /*
 440                          * Lookup the inode chunk that this inode lives in.
 441                          */
 442                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 443                         if (!error &&   /* no I/O error */
 444                             tmp &&      /* lookup succeeded */
 445                                         /* got the record, should always work */
 446                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 447                                     &gfree, &i)) &&
 448                             i == 1 &&
 449                                         /* this is the right chunk */
 450                             agino < gino + XFS_INODES_PER_CHUNK &&
 451                                         /* lastino was not last in chunk */
 452                             (chunkidx = agino - gino + 1) <
 453                                     XFS_INODES_PER_CHUNK &&
 454                                         /* there are some left allocated */
 455                             XFS_INOBT_MASKN(chunkidx,
 456                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 457                                 /*
 458                                  * Grab the chunk record.  Mark all the
 459                                  * uninteresting inodes (because they're
 460                                  * before our start point) free.
 461                                  */
 462                                 for (i = 0; i < chunkidx; i++) {
 463                                         if (XFS_INOBT_MASK(i) & ~gfree)
 464                                                 gcnt++;
 465                                 }
 466                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 467                                 irbp->ir_startino = gino;
 468                                 irbp->ir_freecount = gcnt;
 469                                 irbp->ir_free = gfree;
 470                                 irbp++;
 471                                 agino = gino + XFS_INODES_PER_CHUNK;
 472                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 473                         } else {
 474                                 /*
 475                                  * If any of those tests failed, bump the
 476                                  * inode number (just in case).
 477                                  */
 478                                 agino++;
 479                                 icount = 0;
 480                         }
 481                         /*
 482                          * In any case, increment to the next record.
 483                          */
 484                         if (!error)
 485                                 error = xfs_btree_increment(cur, 0, &tmp);
 486                 } else {
 487                         /*
 488                          * Start of ag.  Lookup the first inode chunk.
 489                          */
 490                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 491                         icount = 0;
 492                 }
 493                 /*
 494                  * Loop through inode btree records in this ag,
 495                  * until we run out of inodes or space in the buffer.
 496                  */
 497                 while (irbp < irbufend && icount < ubcount) {
 498                         /*
 499                          * Loop as long as we're unable to read the
 500                          * inode btree.
 501                          */
 502                         while (error) {
 503                                 agino += XFS_INODES_PER_CHUNK;
 504                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 505                                                 be32_to_cpu(agi->agi_length))
 506                                         break;
 507                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 508                                                             &tmp);
 509                                 cond_resched();
 510                         }
 511                         /*
 512                          * If ran off the end of the ag either with an error,
 513                          * or the normal way, set end and stop collecting.
 514                          */
 515                         if (error ||
 516                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 517                                     &gfree, &i)) ||
 518                             i == 0) {
 519                                 end_of_ag = 1;
 520                                 break;
 521                         }
 522                         /*
 523                          * If this chunk has any allocated inodes, save it.
 524                          * Also start read-ahead now for this chunk.
 525                          */
 526                         if (gcnt < XFS_INODES_PER_CHUNK) {
 527                                 /*
 528                                  * Loop over all clusters in the next chunk.
 529                                  * Do a readahead if there are any allocated
 530                                  * inodes in that cluster.
 531                                  */
 532                                 for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
 533                                      chunkidx = 0;
 534                                      chunkidx < XFS_INODES_PER_CHUNK;
 535                                      chunkidx += nicluster,
 536                                      agbno += nbcluster) {
 537                                         if (XFS_INOBT_MASKN(chunkidx,
 538                                                             nicluster) & ~gfree)
 539                                                 xfs_btree_reada_bufs(mp, agno,
 540                                                         agbno, nbcluster);
 541                                 }
 542                                 irbp->ir_startino = gino;
 543                                 irbp->ir_freecount = gcnt;
 544                                 irbp->ir_free = gfree;
 545                                 irbp++;
 546                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 547                         }
 548                         /*
 549                          * Set agino to after this chunk and bump the cursor.
 550                          */
 551                         agino = gino + XFS_INODES_PER_CHUNK;
 552                         error = xfs_btree_increment(cur, 0, &tmp);
 553                         cond_resched();
 554                 }
 555                 /*
 556                  * Drop the btree buffers and the agi buffer.
 557                  * We can't hold any of the locks these represent
 558                  * when calling iget.
 559                  */
 560                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 561                 xfs_buf_relse(agbp);
 562                 /*
 563                  * Now format all the good inodes into the user's buffer.
 564                  */
 565                 irbufend = irbp;
 566                 for (irbp = irbuf;
 567                      irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
 568                         /*
 569                          * Now process this chunk of inodes.
 570                          */
 571                         for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
 572                              XFS_BULKSTAT_UBLEFT(ubleft) &&
 573                                 irbp->ir_freecount < XFS_INODES_PER_CHUNK;
 574                              chunkidx++, clustidx++, agino++) {
 575                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 576                                 /*
 577                                  * Recompute agbno if this is the
 578                                  * first inode of the cluster.
 579                                  *
 580                                  * Careful with clustidx.   There can be
 581                                  * multple clusters per chunk, a single
 582                                  * cluster per chunk or a cluster that has
 583                                  * inodes represented from several different
 584                                  * chunks (if blocksize is large).
 585                                  *
 586                                  * Because of this, the starting clustidx is
 587                                  * initialized to zero in this loop but must
 588                                  * later be reset after reading in the cluster
 589                                  * buffer.
 590                                  */
 591                                 if ((chunkidx & (nicluster - 1)) == 0) {
 592                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 593                                                         irbp->ir_startino) +
 594                                                 ((chunkidx & nimask) >>
 595                                                  mp->m_sb.sb_inopblog);
 596
 597                                         if (flags & (BULKSTAT_FG_QUICK |
 598                                                      BULKSTAT_FG_INLINE)) {
 599                                                 int offset;
 600
 601                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 602                                                                        agino);
 603                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 604                                                                        agbno);
 605
 606                                                 /*
 607                                                  * Get the inode cluster buffer
 608                                                  */
 609                                                 if (bp)
 610                                                         xfs_buf_relse(bp);
 611
 612                                                 error = xfs_inotobp(mp, NULL, ino, &dip,
 613                                                                     &bp, &offset,
 614                                                                     XFS_IGET_BULKSTAT);
 615
 616                                                 if (!error)
 617                                                         clustidx = offset / mp->m_sb.sb_inodesize;
 618                                                 if (XFS_TEST_ERROR(error != 0,
 619                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 620                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 621                                                         bp = NULL;
 622                                                         ubleft = 0;
 623                                                         rval = error;
 624                                                         break;
 625                                                 }
 626                                         }
 627                                 }
 628                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 629                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 630                                 /*
 631                                  * Skip if this inode is free.
 632                                  */
 633                                 if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
 634                                         lastino = ino;
 635                                         continue;
 636                                 }
 637                                 /*
 638                                  * Count used inodes as free so we can tell
 639                                  * when the chunk is used up.
 640                                  */
 641                                 irbp->ir_freecount++;
 642                                 if (!xfs_bulkstat_use_dinode(mp, flags, bp,
 643                                                              clustidx, &dip)) {
 644                                         lastino = ino;
 645                                         continue;
 646                                 }
 647                                 /*
 648                                  * If we need to do an iget, cannot hold bp.
 649                                  * Drop it, until starting the next cluster.
 650                                  */
 651                                 if ((flags & BULKSTAT_FG_INLINE) && !dip) {
 652                                         if (bp)
 653                                                 xfs_buf_relse(bp);
 654                                         bp = NULL;
 655                                 }
 656
 657                                 /*
 658                                  * Get the inode and fill in a single buffer.
 659                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 660                                  * BULKSTAT_FG_IGET uses igets.
 661                                  * BULKSTAT_FG_INLINE uses dip if we have an
 662                                  * inline attr fork, else igets.
 663                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 664                                  * This is also used to count inodes/blks, etc
 665                                  * in xfs_qm_quotacheck.
 666                                  */
 667                                 ubused = statstruct_size;
 668                                 error = formatter(mp, ino, ubufp,
 669                                                 ubleft, private_data,
 670                                                 bno, &ubused, dip, &fmterror);
 671                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 672                                         if (error && error != ENOENT &&
 673                                                 error != EINVAL) {
 674                                                 ubleft = 0;
 675                                                 rval = error;
 676                                                 break;
 677                                         }
 678                                         lastino = ino;
 679                                         continue;
 680                                 }
 681                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 682                                         ubleft = 0;
 683                                         ASSERT(error);
 684                                         rval = error;
 685                                         break;
 686                                 }
 687                                 if (ubufp)
 688                                         ubufp += ubused;
 689                                 ubleft -= ubused;
 690                                 ubelem++;
 691                                 lastino = ino;
 692                         }
 693
 694                         cond_resched();
 695                 }
 696
 697                 if (bp)
 698                         xfs_buf_relse(bp);
 699
 700                 /*
 701                  * Set up for the next loop iteration.
 702                  */
 703                 if (XFS_BULKSTAT_UBLEFT(ubleft)) {
 704                         if (end_of_ag) {
 705                                 agno++;
 706                                 agino = 0;
 707                         } else
 708                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 709                 } else
 710                         break;
 711         }
 712         /*
 713          * Done, we're either out of filesystem or space to put the data.
 714          */
 715         kmem_free(irbuf);
 716         *ubcountp = ubelem;
 717         /*
 718          * Found some inodes, return them now and return the error next time.
 719          */
 720         if (ubelem)
 721                 rval = 0;
 722         if (agno >= mp->m_sb.sb_agcount) {
 723                 /*
 724                  * If we ran out of filesystem, mark lastino as off
 725                  * the end of the filesystem, so the next call
 726                  * will return immediately.
 727                  */
 728                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 729                 *done = 1;
 730         } else
 731                 *lastinop = (xfs_ino_t)lastino;
 732
 733         return rval;
 734 }
 735
 736 /*
 737  * Return stat information in bulk (by-inode) for the filesystem.
 738  * Special case for non-sequential one inode bulkstat.
 739  */
 740 int                                     /* error status */
 741 xfs_bulkstat_single(
 742         xfs_mount_t             *mp,    /* mount point for filesystem */
 743         xfs_ino_t               *lastinop, /* inode to return */
 744         char                    __user *buffer, /* buffer with inode stats */
 745         int                     *done)  /* 1 if there are more stats to get */
 746 {
 747         int                     count;  /* count value for bulkstat call */
 748         int                     error;  /* return value */
 749         xfs_ino_t               ino;    /* filesystem inode number */
 750         int                     res;    /* result from bs1 */
 751
 752         /*
 753          * note that requesting valid inode numbers which are not allocated
 754          * to inodes will most likely cause xfs_itobp to generate warning
 755          * messages about bad magic numbers. This is ok. The fact that
 756          * the inode isn't actually an inode is handled by the
 757          * error check below. Done this way to make the usual case faster
 758          * at the expense of the error case.
 759          */
 760
 761         ino = (xfs_ino_t)*lastinop;
 762         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 763                                  NULL, 0, NULL, NULL, &res);
 764         if (error) {
 765                 /*
 766                  * Special case way failed, do it the "long" way
 767                  * to see if that works.
 768                  */
 769                 (*lastinop)--;
 770                 count = 1;
 771                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 772                                 NULL, sizeof(xfs_bstat_t), buffer,
 773                                 BULKSTAT_FG_IGET, done))
 774                         return error;
 775                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 776                         return error == EFSCORRUPTED ?
 777                                 XFS_ERROR(EINVAL) : error;
 778                 else
 779                         return 0;
 780         }
 781         *done = 0;
 782         return 0;
 783 }
 784
 785 int
 786 xfs_inumbers_fmt(
 787         void                    __user *ubuffer, /* buffer to write to */
 788         const xfs_inogrp_t      *buffer,        /* buffer to read from */
 789         long                    count,          /* # of elements to read */
 790         long                    *written)       /* # of bytes written */
 791 {
 792         if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
 793                 return -EFAULT;
 794         *written = count * sizeof(*buffer);
 795         return 0;
 796 }
 797
 798 /*
 799  * Return inode number table for the filesystem.
 800  */
 801 int                                     /* error status */
 802 xfs_inumbers(
 803         xfs_mount_t     *mp,            /* mount point for filesystem */
 804         xfs_ino_t       *lastino,       /* last inode returned */
 805         int             *count,         /* size of buffer/count returned */
 806         void            __user *ubuffer,/* buffer with inode descriptions */
 807         inumbers_fmt_pf formatter)
 808 {
 809         xfs_buf_t       *agbp;
 810         xfs_agino_t     agino;
 811         xfs_agnumber_t  agno;
 812         int             bcount;
 813         xfs_inogrp_t    *buffer;
 814         int             bufidx;
 815         xfs_btree_cur_t *cur;
 816         int             error;
 817         __int32_t       gcnt;
 818         xfs_inofree_t   gfree;
 819         xfs_agino_t     gino;
 820         int             i;
 821         xfs_ino_t       ino;
 822         int             left;
 823         int             tmp;
 824
 825         ino = (xfs_ino_t)*lastino;
 826         agno = XFS_INO_TO_AGNO(mp, ino);
 827         agino = XFS_INO_TO_AGINO(mp, ino);
 828         left = *count;
 829         *count = 0;
 830         bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
 831         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 832         error = bufidx = 0;
 833         cur = NULL;
 834         agbp = NULL;
 835         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 836                 if (agbp == NULL) {
 837                         down_read(&mp->m_peraglock);
 838                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 839                         up_read(&mp->m_peraglock);
 840                         if (error) {
 841                                 /*
 842                                  * If we can't read the AGI of this ag,
 843                                  * then just skip to the next one.
 844                                  */
 845                                 ASSERT(cur == NULL);
 846                                 agbp = NULL;
 847                                 agno++;
 848                                 agino = 0;
 849                                 continue;
 850                         }
 851                         cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
 852                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 853                         if (error) {
 854                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 855                                 cur = NULL;
 856                                 xfs_buf_relse(agbp);
 857                                 agbp = NULL;
 858                                 /*
 859                                  * Move up the last inode in the current
 860                                  * chunk.  The lookup_ge will always get
 861                                  * us the first inode in the next chunk.
 862                                  */
 863                                 agino += XFS_INODES_PER_CHUNK - 1;
 864                                 continue;
 865                         }
 866                 }
 867                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 868                         &i)) ||
 869                     i == 0) {
 870                         xfs_buf_relse(agbp);
 871                         agbp = NULL;
 872                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 873                         cur = NULL;
 874                         agno++;
 875                         agino = 0;
 876                         continue;
 877                 }
 878                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 879                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 880                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 881                 buffer[bufidx].xi_allocmask = ~gfree;
 882                 bufidx++;
 883                 left--;
 884                 if (bufidx == bcount) {
 885                         long written;
 886                         if (formatter(ubuffer, buffer, bufidx, &written)) {
 887                                 error = XFS_ERROR(EFAULT);
 888                                 break;
 889                         }
 890                         ubuffer += written;
 891                         *count += bufidx;
 892                         bufidx = 0;
 893                 }
 894                 if (left) {
 895                         error = xfs_btree_increment(cur, 0, &tmp);
 896                         if (error) {
 897                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 898                                 cur = NULL;
 899                                 xfs_buf_relse(agbp);
 900                                 agbp = NULL;
 901                                 /*
 902                                  * The agino value has already been bumped.
 903                                  * Just try to skip up to it.
 904                                  */
 905                                 agino += XFS_INODES_PER_CHUNK;
 906                                 continue;
 907                         }
 908                 }
 909         }
 910         if (!error) {
 911                 if (bufidx) {
 912                         long written;
 913                         if (formatter(ubuffer, buffer, bufidx, &written))
 914                                 error = XFS_ERROR(EFAULT);
 915                         else
 916                                 *count += bufidx;
 917                 }
 918                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 919         }
 920         kmem_free(buffer);
 921         if (cur)
 922                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 923                                            XFS_BTREE_NOERROR));
 924         if (agbp)
 925                 xfs_buf_relse(agbp);
 926         return error;
 927 }