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_blkno != (xfs_daddr_t)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    *dip,           /* dinode inode pointer */
 129         xfs_bstat_t     *buf)           /* return buffer */
 130 {
 131         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 132
 133         dic = &dip->di_core;
 134
 135         /*
 136          * The inode format changed when we moved the link count and
 137          * made it 32 bits long.  If this is an old format inode,
 138          * convert it in memory to look like a new one.  If it gets
 139          * flushed to disk we will convert back before flushing or
 140          * logging it.  We zero out the new projid field and the old link
 141          * count field.  We'll handle clearing the pad field (the remains
 142          * of the old uuid field) when we actually convert the inode to
 143          * the new format. We don't change the version number so that we
 144          * can distinguish this from a real new format inode.
 145          */
 146         if (dic->di_version == XFS_DINODE_VERSION_1) {
 147                 buf->bs_nlink = be16_to_cpu(dic->di_onlink);
 148                 buf->bs_projid = 0;
 149         } else {
 150                 buf->bs_nlink = be32_to_cpu(dic->di_nlink);
 151                 buf->bs_projid = be16_to_cpu(dic->di_projid);
 152         }
 153
 154         buf->bs_ino = ino;
 155         buf->bs_mode = be16_to_cpu(dic->di_mode);
 156         buf->bs_uid = be32_to_cpu(dic->di_uid);
 157         buf->bs_gid = be32_to_cpu(dic->di_gid);
 158         buf->bs_size = be64_to_cpu(dic->di_size);
 159         buf->bs_atime.tv_sec = be32_to_cpu(dic->di_atime.t_sec);
 160         buf->bs_atime.tv_nsec = be32_to_cpu(dic->di_atime.t_nsec);
 161         buf->bs_mtime.tv_sec = be32_to_cpu(dic->di_mtime.t_sec);
 162         buf->bs_mtime.tv_nsec = be32_to_cpu(dic->di_mtime.t_nsec);
 163         buf->bs_ctime.tv_sec = be32_to_cpu(dic->di_ctime.t_sec);
 164         buf->bs_ctime.tv_nsec = be32_to_cpu(dic->di_ctime.t_nsec);
 165         buf->bs_xflags = xfs_dic2xflags(dip);
 166         buf->bs_extsize = be32_to_cpu(dic->di_extsize) << mp->m_sb.sb_blocklog;
 167         buf->bs_extents = be32_to_cpu(dic->di_nextents);
 168         buf->bs_gen = be32_to_cpu(dic->di_gen);
 169         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 170         buf->bs_dmevmask = be32_to_cpu(dic->di_dmevmask);
 171         buf->bs_dmstate = be16_to_cpu(dic->di_dmstate);
 172         buf->bs_aextents = be16_to_cpu(dic->di_anextents);
 173
 174         switch (dic->di_format) {
 175         case XFS_DINODE_FMT_DEV:
 176                 buf->bs_rdev = be32_to_cpu(dip->di_u.di_dev);
 177                 buf->bs_blksize = BLKDEV_IOSIZE;
 178                 buf->bs_blocks = 0;
 179                 break;
 180         case XFS_DINODE_FMT_LOCAL:
 181         case XFS_DINODE_FMT_UUID:
 182                 buf->bs_rdev = 0;
 183                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 184                 buf->bs_blocks = 0;
 185                 break;
 186         case XFS_DINODE_FMT_EXTENTS:
 187         case XFS_DINODE_FMT_BTREE:
 188                 buf->bs_rdev = 0;
 189                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 190                 buf->bs_blocks = be64_to_cpu(dic->di_nblocks);
 191                 break;
 192         }
 193 }
 194
 195 STATIC int
 196 xfs_bulkstat_one_fmt(
 197         void                    __user *ubuffer,
 198         const xfs_bstat_t       *buffer)
 199 {
 200         if (copy_to_user(ubuffer, buffer, sizeof(*buffer)))
 201                 return -EFAULT;
 202         return sizeof(*buffer);
 203 }
 204
 205 /*
 206  * Return stat information for one inode.
 207  * Return 0 if ok, else errno.
 208  */
 209 int                             /* error status */
 210 xfs_bulkstat_one(
 211         xfs_mount_t     *mp,            /* mount point for filesystem */
 212         xfs_ino_t       ino,            /* inode number to get data for */
 213         void            __user *buffer, /* buffer to place output in */
 214         int             ubsize,         /* size of buffer */
 215         void            *private_data,  /* my private data */
 216         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 217         int             *ubused,        /* bytes used by me */
 218         void            *dibuff,        /* on-disk inode buffer */
 219         int             *stat)          /* BULKSTAT_RV_... */
 220 {
 221         xfs_bstat_t     *buf;           /* return buffer */
 222         int             error = 0;      /* error value */
 223         xfs_dinode_t    *dip;           /* dinode inode pointer */
 224         bulkstat_one_fmt_pf formatter = private_data ? : xfs_bulkstat_one_fmt;
 225
 226         dip = (xfs_dinode_t *)dibuff;
 227         *stat = BULKSTAT_RV_NOTHING;
 228
 229         if (!buffer || xfs_internal_inum(mp, ino))
 230                 return XFS_ERROR(EINVAL);
 231         if (ubsize < sizeof(*buf))
 232                 return XFS_ERROR(ENOMEM);
 233
 234         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 235
 236         if (dip == NULL) {
 237                 /* We're not being passed a pointer to a dinode.  This happens
 238                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 239                  */
 240                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 241                 if (error)
 242                         goto out_free;
 243         } else {
 244                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 245         }
 246
 247         error = formatter(buffer, buf);
 248         if (error < 0)  {
 249                 error = EFAULT;
 250                 goto out_free;
 251         }
 252
 253         *stat = BULKSTAT_RV_DIDONE;
 254         if (ubused)
 255                 *ubused = error;
 256
 257  out_free:
 258         kmem_free(buf);
 259         return error;
 260 }
 261
 262 /*
 263  * Test to see whether we can use the ondisk inode directly, based
 264  * on the given bulkstat flags, filling in dipp accordingly.
 265  * Returns zero if the inode is dodgey.
 266  */
 267 STATIC int
 268 xfs_bulkstat_use_dinode(
 269         xfs_mount_t     *mp,
 270         int             flags,
 271         xfs_buf_t       *bp,
 272         int             clustidx,
 273         xfs_dinode_t    **dipp)
 274 {
 275         xfs_dinode_t    *dip;
 276         unsigned int    aformat;
 277
 278         *dipp = NULL;
 279         if (!bp || (flags & BULKSTAT_FG_IGET))
 280                 return 1;
 281         dip = (xfs_dinode_t *)
 282                         xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
 283         /*
 284          * Check the buffer containing the on-disk inode for di_mode == 0.
 285          * This is to prevent xfs_bulkstat from picking up just reclaimed
 286          * inodes that have their in-core state initialized but not flushed
 287          * to disk yet. This is a temporary hack that would require a proper
 288          * fix in the future.
 289          */
 290         if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC ||
 291             !XFS_DINODE_GOOD_VERSION(dip->di_core.di_version) ||
 292             !dip->di_core.di_mode)
 293                 return 0;
 294         if (flags & BULKSTAT_FG_QUICK) {
 295                 *dipp = dip;
 296                 return 1;
 297         }
 298         /* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
 299         aformat = dip->di_core.di_aformat;
 300         if ((XFS_DFORK_Q(dip) == 0) ||
 301             (aformat == XFS_DINODE_FMT_LOCAL) ||
 302             (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
 303                 *dipp = dip;
 304                 return 1;
 305         }
 306         return 1;
 307 }
 308
 309 #define XFS_BULKSTAT_UBLEFT(ubleft)     ((ubleft) >= statstruct_size)
 310
 311 /*
 312  * Return stat information in bulk (by-inode) for the filesystem.
 313  */
 314 int                                     /* error status */
 315 xfs_bulkstat(
 316         xfs_mount_t             *mp,    /* mount point for filesystem */
 317         xfs_ino_t               *lastinop, /* last inode returned */
 318         int                     *ubcountp, /* size of buffer/count returned */
 319         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 320         void                    *private_data,/* private data for formatter */
 321         size_t                  statstruct_size, /* sizeof struct filling */
 322         char                    __user *ubuffer, /* buffer with inode stats */
 323         int                     flags,  /* defined in xfs_itable.h */
 324         int                     *done)  /* 1 if there are more stats to get */
 325 {
 326         xfs_agblock_t           agbno=0;/* allocation group block number */
 327         xfs_buf_t               *agbp;  /* agi header buffer */
 328         xfs_agi_t               *agi;   /* agi header data */
 329         xfs_agino_t             agino;  /* inode # in allocation group */
 330         xfs_agnumber_t          agno;   /* allocation group number */
 331         xfs_daddr_t             bno;    /* inode cluster start daddr */
 332         int                     chunkidx; /* current index into inode chunk */
 333         int                     clustidx; /* current index into inode cluster */
 334         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 335         int                     end_of_ag; /* set if we've seen the ag end */
 336         int                     error;  /* error code */
 337         int                     fmterror;/* bulkstat formatter result */
 338         __int32_t               gcnt;   /* current btree rec's count */
 339         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 340         xfs_agino_t             gino;   /* current btree rec's start inode */
 341         int                     i;      /* loop index */
 342         int                     icount; /* count of inodes good in irbuf */
 343         size_t                  irbsize; /* size of irec buffer in bytes */
 344         xfs_ino_t               ino;    /* inode number (filesystem) */
 345         xfs_inobt_rec_incore_t  *irbp;  /* current irec buffer pointer */
 346         xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
 347         xfs_inobt_rec_incore_t  *irbufend; /* end of good irec buffer entries */
 348         xfs_ino_t               lastino; /* last inode number returned */
 349         int                     nbcluster; /* # of blocks in a cluster */
 350         int                     nicluster; /* # of inodes in a cluster */
 351         int                     nimask; /* mask for inode clusters */
 352         int                     nirbuf; /* size of irbuf */
 353         int                     rval;   /* return value error code */
 354         int                     tmp;    /* result value from btree calls */
 355         int                     ubcount; /* size of user's buffer */
 356         int                     ubleft; /* bytes left in user's buffer */
 357         char                    __user *ubufp;  /* pointer into user's buffer */
 358         int                     ubelem; /* spaces used in user's buffer */
 359         int                     ubused; /* bytes used by formatter */
 360         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 361         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 362
 363         /*
 364          * Get the last inode value, see if there's nothing to do.
 365          */
 366         ino = (xfs_ino_t)*lastinop;
 367         lastino = ino;
 368         dip = NULL;
 369         agno = XFS_INO_TO_AGNO(mp, ino);
 370         agino = XFS_INO_TO_AGINO(mp, ino);
 371         if (agno >= mp->m_sb.sb_agcount ||
 372             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 373                 *done = 1;
 374                 *ubcountp = 0;
 375                 return 0;
 376         }
 377         if (!ubcountp || *ubcountp <= 0) {
 378                 return EINVAL;
 379         }
 380         ubcount = *ubcountp; /* statstruct's */
 381         ubleft = ubcount * statstruct_size; /* bytes */
 382         *ubcountp = ubelem = 0;
 383         *done = 0;
 384         fmterror = 0;
 385         ubufp = ubuffer;
 386         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 387                 mp->m_sb.sb_inopblock :
 388                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 389         nimask = ~(nicluster - 1);
 390         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 391         irbuf = kmem_zalloc_greedy(&irbsize, PAGE_SIZE, PAGE_SIZE * 4,
 392                                    KM_SLEEP | KM_MAYFAIL | KM_LARGE);
 393         nirbuf = irbsize / sizeof(*irbuf);
 394
 395         /*
 396          * Loop over the allocation groups, starting from the last
 397          * inode returned; 0 means start of the allocation group.
 398          */
 399         rval = 0;
 400         while (XFS_BULKSTAT_UBLEFT(ubleft) && agno < mp->m_sb.sb_agcount) {
 401                 cond_resched();
 402                 bp = NULL;
 403                 down_read(&mp->m_peraglock);
 404                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 405                 up_read(&mp->m_peraglock);
 406                 if (error) {
 407                         /*
 408                          * Skip this allocation group and go to the next one.
 409                          */
 410                         agno++;
 411                         agino = 0;
 412                         continue;
 413                 }
 414                 agi = XFS_BUF_TO_AGI(agbp);
 415                 /*
 416                  * Allocate and initialize a btree cursor for ialloc btree.
 417                  */
 418                 cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
 419                 irbp = irbuf;
 420                 irbufend = irbuf + nirbuf;
 421                 end_of_ag = 0;
 422                 /*
 423                  * If we're returning in the middle of an allocation group,
 424                  * we need to get the remainder of the chunk we're in.
 425                  */
 426                 if (agino > 0) {
 427                         /*
 428                          * Lookup the inode chunk that this inode lives in.
 429                          */
 430                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 431                         if (!error &&   /* no I/O error */
 432                             tmp &&      /* lookup succeeded */
 433                                         /* got the record, should always work */
 434                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 435                                     &gfree, &i)) &&
 436                             i == 1 &&
 437                                         /* this is the right chunk */
 438                             agino < gino + XFS_INODES_PER_CHUNK &&
 439                                         /* lastino was not last in chunk */
 440                             (chunkidx = agino - gino + 1) <
 441                                     XFS_INODES_PER_CHUNK &&
 442                                         /* there are some left allocated */
 443                             XFS_INOBT_MASKN(chunkidx,
 444                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 445                                 /*
 446                                  * Grab the chunk record.  Mark all the
 447                                  * uninteresting inodes (because they're
 448                                  * before our start point) free.
 449                                  */
 450                                 for (i = 0; i < chunkidx; i++) {
 451                                         if (XFS_INOBT_MASK(i) & ~gfree)
 452                                                 gcnt++;
 453                                 }
 454                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 455                                 irbp->ir_startino = gino;
 456                                 irbp->ir_freecount = gcnt;
 457                                 irbp->ir_free = gfree;
 458                                 irbp++;
 459                                 agino = gino + XFS_INODES_PER_CHUNK;
 460                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 461                         } else {
 462                                 /*
 463                                  * If any of those tests failed, bump the
 464                                  * inode number (just in case).
 465                                  */
 466                                 agino++;
 467                                 icount = 0;
 468                         }
 469                         /*
 470                          * In any case, increment to the next record.
 471                          */
 472                         if (!error)
 473                                 error = xfs_btree_increment(cur, 0, &tmp);
 474                 } else {
 475                         /*
 476                          * Start of ag.  Lookup the first inode chunk.
 477                          */
 478                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 479                         icount = 0;
 480                 }
 481                 /*
 482                  * Loop through inode btree records in this ag,
 483                  * until we run out of inodes or space in the buffer.
 484                  */
 485                 while (irbp < irbufend && icount < ubcount) {
 486                         /*
 487                          * Loop as long as we're unable to read the
 488                          * inode btree.
 489                          */
 490                         while (error) {
 491                                 agino += XFS_INODES_PER_CHUNK;
 492                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 493                                                 be32_to_cpu(agi->agi_length))
 494                                         break;
 495                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 496                                                             &tmp);
 497                                 cond_resched();
 498                         }
 499                         /*
 500                          * If ran off the end of the ag either with an error,
 501                          * or the normal way, set end and stop collecting.
 502                          */
 503                         if (error ||
 504                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 505                                     &gfree, &i)) ||
 506                             i == 0) {
 507                                 end_of_ag = 1;
 508                                 break;
 509                         }
 510                         /*
 511                          * If this chunk has any allocated inodes, save it.
 512                          * Also start read-ahead now for this chunk.
 513                          */
 514                         if (gcnt < XFS_INODES_PER_CHUNK) {
 515                                 /*
 516                                  * Loop over all clusters in the next chunk.
 517                                  * Do a readahead if there are any allocated
 518                                  * inodes in that cluster.
 519                                  */
 520                                 for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
 521                                      chunkidx = 0;
 522                                      chunkidx < XFS_INODES_PER_CHUNK;
 523                                      chunkidx += nicluster,
 524                                      agbno += nbcluster) {
 525                                         if (XFS_INOBT_MASKN(chunkidx,
 526                                                             nicluster) & ~gfree)
 527                                                 xfs_btree_reada_bufs(mp, agno,
 528                                                         agbno, nbcluster);
 529                                 }
 530                                 irbp->ir_startino = gino;
 531                                 irbp->ir_freecount = gcnt;
 532                                 irbp->ir_free = gfree;
 533                                 irbp++;
 534                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 535                         }
 536                         /*
 537                          * Set agino to after this chunk and bump the cursor.
 538                          */
 539                         agino = gino + XFS_INODES_PER_CHUNK;
 540                         error = xfs_btree_increment(cur, 0, &tmp);
 541                         cond_resched();
 542                 }
 543                 /*
 544                  * Drop the btree buffers and the agi buffer.
 545                  * We can't hold any of the locks these represent
 546                  * when calling iget.
 547                  */
 548                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 549                 xfs_buf_relse(agbp);
 550                 /*
 551                  * Now format all the good inodes into the user's buffer.
 552                  */
 553                 irbufend = irbp;
 554                 for (irbp = irbuf;
 555                      irbp < irbufend && XFS_BULKSTAT_UBLEFT(ubleft); irbp++) {
 556                         /*
 557                          * Now process this chunk of inodes.
 558                          */
 559                         for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
 560                              XFS_BULKSTAT_UBLEFT(ubleft) &&
 561                                 irbp->ir_freecount < XFS_INODES_PER_CHUNK;
 562                              chunkidx++, clustidx++, agino++) {
 563                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 564                                 /*
 565                                  * Recompute agbno if this is the
 566                                  * first inode of the cluster.
 567                                  *
 568                                  * Careful with clustidx.   There can be
 569                                  * multple clusters per chunk, a single
 570                                  * cluster per chunk or a cluster that has
 571                                  * inodes represented from several different
 572                                  * chunks (if blocksize is large).
 573                                  *
 574                                  * Because of this, the starting clustidx is
 575                                  * initialized to zero in this loop but must
 576                                  * later be reset after reading in the cluster
 577                                  * buffer.
 578                                  */
 579                                 if ((chunkidx & (nicluster - 1)) == 0) {
 580                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 581                                                         irbp->ir_startino) +
 582                                                 ((chunkidx & nimask) >>
 583                                                  mp->m_sb.sb_inopblog);
 584
 585                                         if (flags & (BULKSTAT_FG_QUICK |
 586                                                      BULKSTAT_FG_INLINE)) {
 587                                                 int offset;
 588
 589                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 590                                                                        agino);
 591                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 592                                                                        agbno);
 593
 594                                                 /*
 595                                                  * Get the inode cluster buffer
 596                                                  */
 597                                                 if (bp)
 598                                                         xfs_buf_relse(bp);
 599
 600                                                 error = xfs_inotobp(mp, NULL, ino, &dip,
 601                                                                     &bp, &offset,
 602                                                                     XFS_IMAP_BULKSTAT);
 603
 604                                                 if (!error)
 605                                                         clustidx = offset / mp->m_sb.sb_inodesize;
 606                                                 if (XFS_TEST_ERROR(error != 0,
 607                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 608                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 609                                                         bp = NULL;
 610                                                         ubleft = 0;
 611                                                         rval = error;
 612                                                         break;
 613                                                 }
 614                                         }
 615                                 }
 616                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 617                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 618                                 /*
 619                                  * Skip if this inode is free.
 620                                  */
 621                                 if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free) {
 622                                         lastino = ino;
 623                                         continue;
 624                                 }
 625                                 /*
 626                                  * Count used inodes as free so we can tell
 627                                  * when the chunk is used up.
 628                                  */
 629                                 irbp->ir_freecount++;
 630                                 if (!xfs_bulkstat_use_dinode(mp, flags, bp,
 631                                                              clustidx, &dip)) {
 632                                         lastino = ino;
 633                                         continue;
 634                                 }
 635                                 /*
 636                                  * If we need to do an iget, cannot hold bp.
 637                                  * Drop it, until starting the next cluster.
 638                                  */
 639                                 if ((flags & BULKSTAT_FG_INLINE) && !dip) {
 640                                         if (bp)
 641                                                 xfs_buf_relse(bp);
 642                                         bp = NULL;
 643                                 }
 644
 645                                 /*
 646                                  * Get the inode and fill in a single buffer.
 647                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 648                                  * BULKSTAT_FG_IGET uses igets.
 649                                  * BULKSTAT_FG_INLINE uses dip if we have an
 650                                  * inline attr fork, else igets.
 651                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 652                                  * This is also used to count inodes/blks, etc
 653                                  * in xfs_qm_quotacheck.
 654                                  */
 655                                 ubused = statstruct_size;
 656                                 error = formatter(mp, ino, ubufp,
 657                                                 ubleft, private_data,
 658                                                 bno, &ubused, dip, &fmterror);
 659                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 660                                         if (error && error != ENOENT &&
 661                                                 error != EINVAL) {
 662                                                 ubleft = 0;
 663                                                 rval = error;
 664                                                 break;
 665                                         }
 666                                         lastino = ino;
 667                                         continue;
 668                                 }
 669                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 670                                         ubleft = 0;
 671                                         ASSERT(error);
 672                                         rval = error;
 673                                         break;
 674                                 }
 675                                 if (ubufp)
 676                                         ubufp += ubused;
 677                                 ubleft -= ubused;
 678                                 ubelem++;
 679                                 lastino = ino;
 680                         }
 681
 682                         cond_resched();
 683                 }
 684
 685                 if (bp)
 686                         xfs_buf_relse(bp);
 687
 688                 /*
 689                  * Set up for the next loop iteration.
 690                  */
 691                 if (XFS_BULKSTAT_UBLEFT(ubleft)) {
 692                         if (end_of_ag) {
 693                                 agno++;
 694                                 agino = 0;
 695                         } else
 696                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 697                 } else
 698                         break;
 699         }
 700         /*
 701          * Done, we're either out of filesystem or space to put the data.
 702          */
 703         kmem_free(irbuf);
 704         *ubcountp = ubelem;
 705         /*
 706          * Found some inodes, return them now and return the error next time.
 707          */
 708         if (ubelem)
 709                 rval = 0;
 710         if (agno >= mp->m_sb.sb_agcount) {
 711                 /*
 712                  * If we ran out of filesystem, mark lastino as off
 713                  * the end of the filesystem, so the next call
 714                  * will return immediately.
 715                  */
 716                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 717                 *done = 1;
 718         } else
 719                 *lastinop = (xfs_ino_t)lastino;
 720
 721         return rval;
 722 }
 723
 724 /*
 725  * Return stat information in bulk (by-inode) for the filesystem.
 726  * Special case for non-sequential one inode bulkstat.
 727  */
 728 int                                     /* error status */
 729 xfs_bulkstat_single(
 730         xfs_mount_t             *mp,    /* mount point for filesystem */
 731         xfs_ino_t               *lastinop, /* inode to return */
 732         char                    __user *buffer, /* buffer with inode stats */
 733         int                     *done)  /* 1 if there are more stats to get */
 734 {
 735         int                     count;  /* count value for bulkstat call */
 736         int                     error;  /* return value */
 737         xfs_ino_t               ino;    /* filesystem inode number */
 738         int                     res;    /* result from bs1 */
 739
 740         /*
 741          * note that requesting valid inode numbers which are not allocated
 742          * to inodes will most likely cause xfs_itobp to generate warning
 743          * messages about bad magic numbers. This is ok. The fact that
 744          * the inode isn't actually an inode is handled by the
 745          * error check below. Done this way to make the usual case faster
 746          * at the expense of the error case.
 747          */
 748
 749         ino = (xfs_ino_t)*lastinop;
 750         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 751                                  NULL, 0, NULL, NULL, &res);
 752         if (error) {
 753                 /*
 754                  * Special case way failed, do it the "long" way
 755                  * to see if that works.
 756                  */
 757                 (*lastinop)--;
 758                 count = 1;
 759                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 760                                 NULL, sizeof(xfs_bstat_t), buffer,
 761                                 BULKSTAT_FG_IGET, done))
 762                         return error;
 763                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 764                         return error == EFSCORRUPTED ?
 765                                 XFS_ERROR(EINVAL) : error;
 766                 else
 767                         return 0;
 768         }
 769         *done = 0;
 770         return 0;
 771 }
 772
 773 int
 774 xfs_inumbers_fmt(
 775         void                    __user *ubuffer, /* buffer to write to */
 776         const xfs_inogrp_t      *buffer,        /* buffer to read from */
 777         long                    count,          /* # of elements to read */
 778         long                    *written)       /* # of bytes written */
 779 {
 780         if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
 781                 return -EFAULT;
 782         *written = count * sizeof(*buffer);
 783         return 0;
 784 }
 785
 786 /*
 787  * Return inode number table for the filesystem.
 788  */
 789 int                                     /* error status */
 790 xfs_inumbers(
 791         xfs_mount_t     *mp,            /* mount point for filesystem */
 792         xfs_ino_t       *lastino,       /* last inode returned */
 793         int             *count,         /* size of buffer/count returned */
 794         void            __user *ubuffer,/* buffer with inode descriptions */
 795         inumbers_fmt_pf formatter)
 796 {
 797         xfs_buf_t       *agbp;
 798         xfs_agino_t     agino;
 799         xfs_agnumber_t  agno;
 800         int             bcount;
 801         xfs_inogrp_t    *buffer;
 802         int             bufidx;
 803         xfs_btree_cur_t *cur;
 804         int             error;
 805         __int32_t       gcnt;
 806         xfs_inofree_t   gfree;
 807         xfs_agino_t     gino;
 808         int             i;
 809         xfs_ino_t       ino;
 810         int             left;
 811         int             tmp;
 812
 813         ino = (xfs_ino_t)*lastino;
 814         agno = XFS_INO_TO_AGNO(mp, ino);
 815         agino = XFS_INO_TO_AGINO(mp, ino);
 816         left = *count;
 817         *count = 0;
 818         bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
 819         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 820         error = bufidx = 0;
 821         cur = NULL;
 822         agbp = NULL;
 823         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 824                 if (agbp == NULL) {
 825                         down_read(&mp->m_peraglock);
 826                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 827                         up_read(&mp->m_peraglock);
 828                         if (error) {
 829                                 /*
 830                                  * If we can't read the AGI of this ag,
 831                                  * then just skip to the next one.
 832                                  */
 833                                 ASSERT(cur == NULL);
 834                                 agbp = NULL;
 835                                 agno++;
 836                                 agino = 0;
 837                                 continue;
 838                         }
 839                         cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
 840                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 841                         if (error) {
 842                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 843                                 cur = NULL;
 844                                 xfs_buf_relse(agbp);
 845                                 agbp = NULL;
 846                                 /*
 847                                  * Move up the last inode in the current
 848                                  * chunk.  The lookup_ge will always get
 849                                  * us the first inode in the next chunk.
 850                                  */
 851                                 agino += XFS_INODES_PER_CHUNK - 1;
 852                                 continue;
 853                         }
 854                 }
 855                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 856                         &i)) ||
 857                     i == 0) {
 858                         xfs_buf_relse(agbp);
 859                         agbp = NULL;
 860                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 861                         cur = NULL;
 862                         agno++;
 863                         agino = 0;
 864                         continue;
 865                 }
 866                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 867                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 868                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 869                 buffer[bufidx].xi_allocmask = ~gfree;
 870                 bufidx++;
 871                 left--;
 872                 if (bufidx == bcount) {
 873                         long written;
 874                         if (formatter(ubuffer, buffer, bufidx, &written)) {
 875                                 error = XFS_ERROR(EFAULT);
 876                                 break;
 877                         }
 878                         ubuffer += written;
 879                         *count += bufidx;
 880                         bufidx = 0;
 881                 }
 882                 if (left) {
 883                         error = xfs_btree_increment(cur, 0, &tmp);
 884                         if (error) {
 885                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 886                                 cur = NULL;
 887                                 xfs_buf_relse(agbp);
 888                                 agbp = NULL;
 889                                 /*
 890                                  * The agino value has already been bumped.
 891                                  * Just try to skip up to it.
 892                                  */
 893                                 agino += XFS_INODES_PER_CHUNK;
 894                                 continue;
 895                         }
 896                 }
 897         }
 898         if (!error) {
 899                 if (bufidx) {
 900                         long written;
 901                         if (formatter(ubuffer, buffer, bufidx, &written))
 902                                 error = XFS_ERROR(EFAULT);
 903                         else
 904                                 *count += bufidx;
 905                 }
 906                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 907         }
 908         kmem_free(buffer);
 909         if (cur)
 910                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 911                                            XFS_BTREE_NOERROR));
 912         if (agbp)
 913                 xfs_buf_relse(agbp);
 914         return error;
 915 }