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_dinode_core_t *dic;         /* dinode core info pointer */
  61         xfs_inode_t     *ip;            /* incore inode pointer */
  62         bhv_vnode_t     *vp;
  63         int             error;
  64
  65         error = xfs_iget(mp, NULL, ino,
  66                          XFS_IGET_BULKSTAT, XFS_ILOCK_SHARED, &ip, bno);
  67         if (error) {
  68                 *stat = BULKSTAT_RV_NOTHING;
  69                 return error;
  70         }
  71
  72         ASSERT(ip != NULL);
  73         ASSERT(ip->i_blkno != (xfs_daddr_t)0);
  74         if (ip->i_d.di_mode == 0) {
  75                 *stat = BULKSTAT_RV_NOTHING;
  76                 error = XFS_ERROR(ENOENT);
  77                 goto out_iput;
  78         }
  79
  80         vp = XFS_ITOV(ip);
  81         dic = &ip->i_d;
  82
  83         /* xfs_iget returns the following without needing
  84          * further change.
  85          */
  86         buf->bs_nlink = dic->di_nlink;
  87         buf->bs_projid = dic->di_projid;
  88         buf->bs_ino = ino;
  89         buf->bs_mode = dic->di_mode;
  90         buf->bs_uid = dic->di_uid;
  91         buf->bs_gid = dic->di_gid;
  92         buf->bs_tag = dic->di_tag;
  93         buf->bs_size = dic->di_size;
  94         vn_atime_to_bstime(vp, &buf->bs_atime);
  95         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
  96         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
  97         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
  98         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
  99         buf->bs_xflags = xfs_ip2xflags(ip);
 100         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
 101         buf->bs_extents = dic->di_nextents;
 102         buf->bs_gen = dic->di_gen;
 103         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 104         buf->bs_dmevmask = dic->di_dmevmask;
 105         buf->bs_dmstate = dic->di_dmstate;
 106         buf->bs_aextents = dic->di_anextents;
 107
 108         switch (dic->di_format) {
 109         case XFS_DINODE_FMT_DEV:
 110                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 111                 buf->bs_blksize = BLKDEV_IOSIZE;
 112                 buf->bs_blocks = 0;
 113                 break;
 114         case XFS_DINODE_FMT_LOCAL:
 115         case XFS_DINODE_FMT_UUID:
 116                 buf->bs_rdev = 0;
 117                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 118                 buf->bs_blocks = 0;
 119                 break;
 120         case XFS_DINODE_FMT_EXTENTS:
 121         case XFS_DINODE_FMT_BTREE:
 122                 buf->bs_rdev = 0;
 123                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 124                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 125                 break;
 126         }
 127
 128  out_iput:
 129         xfs_iput(ip, XFS_ILOCK_SHARED);
 130         return error;
 131 }
 132
 133 STATIC int
 134 xfs_bulkstat_one_dinode(
 135         xfs_mount_t     *mp,            /* mount point for filesystem */
 136         xfs_ino_t       ino,            /* inode number to get data for */
 137         xfs_dinode_t    *dip,           /* dinode inode pointer */
 138         xfs_bstat_t     *buf)           /* return buffer */
 139 {
 140         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 141
 142         dic = &dip->di_core;
 143
 144         /*
 145          * The inode format changed when we moved the link count and
 146          * made it 32 bits long.  If this is an old format inode,
 147          * convert it in memory to look like a new one.  If it gets
 148          * flushed to disk we will convert back before flushing or
 149          * logging it.  We zero out the new projid field and the old link
 150          * count field.  We'll handle clearing the pad field (the remains
 151          * of the old uuid field) when we actually convert the inode to
 152          * the new format. We don't change the version number so that we
 153          * can distinguish this from a real new format inode.
 154          */
 155         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 156                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 157                 buf->bs_projid = 0;
 158         } else {
 159                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 160                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 161         }
 162
 163         buf->bs_ino = ino;
 164         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 165         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 166         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 167         buf->bs_tag = INT_GET(dic->di_tag, ARCH_CONVERT);
 168         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 169         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 170         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 171         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 172         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 173         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 174         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 175         buf->bs_xflags = xfs_dic2xflags(dic);
 176         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 177         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 178         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 179         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 180         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 181         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 182         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 183
 184         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 185         case XFS_DINODE_FMT_DEV:
 186                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 187                 buf->bs_blksize = BLKDEV_IOSIZE;
 188                 buf->bs_blocks = 0;
 189                 break;
 190         case XFS_DINODE_FMT_LOCAL:
 191         case XFS_DINODE_FMT_UUID:
 192                 buf->bs_rdev = 0;
 193                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 194                 buf->bs_blocks = 0;
 195                 break;
 196         case XFS_DINODE_FMT_EXTENTS:
 197         case XFS_DINODE_FMT_BTREE:
 198                 buf->bs_rdev = 0;
 199                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 200                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 201                 break;
 202         }
 203
 204         return 0;
 205 }
 206
 207 /*
 208  * Return stat information for one inode.
 209  * Return 0 if ok, else errno.
 210  */
 211 int                             /* error status */
 212 xfs_bulkstat_one(
 213         xfs_mount_t     *mp,            /* mount point for filesystem */
 214         xfs_ino_t       ino,            /* inode number to get data for */
 215         void            __user *buffer, /* buffer to place output in */
 216         int             ubsize,         /* size of buffer */
 217         void            *private_data,  /* my private data */
 218         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 219         int             *ubused,        /* bytes used by me */
 220         void            *dibuff,        /* on-disk inode buffer */
 221         int             *stat)          /* BULKSTAT_RV_... */
 222 {
 223         xfs_bstat_t     *buf;           /* return buffer */
 224         int             error = 0;      /* error value */
 225         xfs_dinode_t    *dip;           /* dinode inode pointer */
 226
 227         dip = (xfs_dinode_t *)dibuff;
 228         *stat = BULKSTAT_RV_NOTHING;
 229
 230         if (!buffer || xfs_internal_inum(mp, ino))
 231                 return XFS_ERROR(EINVAL);
 232         if (ubsize < sizeof(*buf))
 233                 return XFS_ERROR(ENOMEM);
 234
 235         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 236
 237         if (dip == NULL) {
 238                 /* We're not being passed a pointer to a dinode.  This happens
 239                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 240                  */
 241                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 242                 if (error)
 243                         goto out_free;
 244         } else {
 245                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 246         }
 247
 248         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 249                 error = EFAULT;
 250                 goto out_free;
 251         }
 252
 253         *stat = BULKSTAT_RV_DIDONE;
 254         if (ubused)
 255                 *ubused = sizeof(*buf);
 256
 257  out_free:
 258         kmem_free(buf, sizeof(*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         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC ||
 284             !XFS_DINODE_GOOD_VERSION(
 285                         INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 286                 return 0;
 287         if (flags & BULKSTAT_FG_QUICK) {
 288                 *dipp = dip;
 289                 return 1;
 290         }
 291         /* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
 292         aformat = INT_GET(dip->di_core.di_aformat, ARCH_CONVERT);
 293         if ((XFS_CFORK_Q(&dip->di_core) == 0) ||
 294             (aformat == XFS_DINODE_FMT_LOCAL) ||
 295             (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
 296                 *dipp = dip;
 297                 return 1;
 298         }
 299         return 1;
 300 }
 301
 302 /*
 303  * Return stat information in bulk (by-inode) for the filesystem.
 304  */
 305 int                                     /* error status */
 306 xfs_bulkstat(
 307         xfs_mount_t             *mp,    /* mount point for filesystem */
 308         xfs_ino_t               *lastinop, /* last inode returned */
 309         int                     *ubcountp, /* size of buffer/count returned */
 310         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 311         void                    *private_data,/* private data for formatter */
 312         size_t                  statstruct_size, /* sizeof struct filling */
 313         char                    __user *ubuffer, /* buffer with inode stats */
 314         int                     flags,  /* defined in xfs_itable.h */
 315         int                     *done)  /* 1 if there are more stats to get */
 316 {
 317         xfs_agblock_t           agbno=0;/* allocation group block number */
 318         xfs_buf_t               *agbp;  /* agi header buffer */
 319         xfs_agi_t               *agi;   /* agi header data */
 320         xfs_agino_t             agino;  /* inode # in allocation group */
 321         xfs_agnumber_t          agno;   /* allocation group number */
 322         xfs_daddr_t             bno;    /* inode cluster start daddr */
 323         int                     chunkidx; /* current index into inode chunk */
 324         int                     clustidx; /* current index into inode cluster */
 325         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 326         int                     end_of_ag; /* set if we've seen the ag end */
 327         int                     error;  /* error code */
 328         int                     fmterror;/* bulkstat formatter result */
 329         __int32_t               gcnt;   /* current btree rec's count */
 330         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 331         xfs_agino_t             gino;   /* current btree rec's start inode */
 332         int                     i;      /* loop index */
 333         int                     icount; /* count of inodes good in irbuf */
 334         size_t                  irbsize; /* size of irec buffer in bytes */
 335         xfs_ino_t               ino;    /* inode number (filesystem) */
 336         xfs_inobt_rec_incore_t  *irbp;  /* current irec buffer pointer */
 337         xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
 338         xfs_inobt_rec_incore_t  *irbufend; /* end of good irec buffer entries */
 339         xfs_ino_t               lastino=0; /* last inode number returned */
 340         int                     nbcluster; /* # of blocks in a cluster */
 341         int                     nicluster; /* # of inodes in a cluster */
 342         int                     nimask; /* mask for inode clusters */
 343         int                     nirbuf; /* size of irbuf */
 344         int                     rval;   /* return value error code */
 345         int                     tmp;    /* result value from btree calls */
 346         int                     ubcount; /* size of user's buffer */
 347         int                     ubleft; /* bytes left in user's buffer */
 348         char                    __user *ubufp;  /* pointer into user's buffer */
 349         int                     ubelem; /* spaces used in user's buffer */
 350         int                     ubused; /* bytes used by formatter */
 351         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 352         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 353         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 354
 355         /*
 356          * Get the last inode value, see if there's nothing to do.
 357          */
 358         ino = (xfs_ino_t)*lastinop;
 359         dip = NULL;
 360         agno = XFS_INO_TO_AGNO(mp, ino);
 361         agino = XFS_INO_TO_AGINO(mp, ino);
 362         if (agno >= mp->m_sb.sb_agcount ||
 363             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 364                 *done = 1;
 365                 *ubcountp = 0;
 366                 return 0;
 367         }
 368         ubcount = *ubcountp; /* statstruct's */
 369         ubleft = ubcount * statstruct_size; /* bytes */
 370         *ubcountp = ubelem = 0;
 371         *done = 0;
 372         fmterror = 0;
 373         ubufp = ubuffer;
 374         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 375                 mp->m_sb.sb_inopblock :
 376                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 377         nimask = ~(nicluster - 1);
 378         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 379         irbuf = kmem_zalloc_greedy(&irbsize, NBPC, NBPC * 4,
 380                                    KM_SLEEP | KM_MAYFAIL | KM_LARGE);
 381         nirbuf = irbsize / sizeof(*irbuf);
 382
 383         /*
 384          * Loop over the allocation groups, starting from the last
 385          * inode returned; 0 means start of the allocation group.
 386          */
 387         rval = 0;
 388         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 389                 bp = NULL;
 390                 down_read(&mp->m_peraglock);
 391                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 392                 up_read(&mp->m_peraglock);
 393                 if (error) {
 394                         /*
 395                          * Skip this allocation group and go to the next one.
 396                          */
 397                         agno++;
 398                         agino = 0;
 399                         continue;
 400                 }
 401                 agi = XFS_BUF_TO_AGI(agbp);
 402                 /*
 403                  * Allocate and initialize a btree cursor for ialloc btree.
 404                  */
 405                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 406                                                 (xfs_inode_t *)0, 0);
 407                 irbp = irbuf;
 408                 irbufend = irbuf + nirbuf;
 409                 end_of_ag = 0;
 410                 /*
 411                  * If we're returning in the middle of an allocation group,
 412                  * we need to get the remainder of the chunk we're in.
 413                  */
 414                 if (agino > 0) {
 415                         /*
 416                          * Lookup the inode chunk that this inode lives in.
 417                          */
 418                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 419                         if (!error &&   /* no I/O error */
 420                             tmp &&      /* lookup succeeded */
 421                                         /* got the record, should always work */
 422                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 423                                     &gfree, &i)) &&
 424                             i == 1 &&
 425                                         /* this is the right chunk */
 426                             agino < gino + XFS_INODES_PER_CHUNK &&
 427                                         /* lastino was not last in chunk */
 428                             (chunkidx = agino - gino + 1) <
 429                                     XFS_INODES_PER_CHUNK &&
 430                                         /* there are some left allocated */
 431                             XFS_INOBT_MASKN(chunkidx,
 432                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 433                                 /*
 434                                  * Grab the chunk record.  Mark all the
 435                                  * uninteresting inodes (because they're
 436                                  * before our start point) free.
 437                                  */
 438                                 for (i = 0; i < chunkidx; i++) {
 439                                         if (XFS_INOBT_MASK(i) & ~gfree)
 440                                                 gcnt++;
 441                                 }
 442                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 443                                 irbp->ir_startino = gino;
 444                                 irbp->ir_freecount = gcnt;
 445                                 irbp->ir_free = gfree;
 446                                 irbp++;
 447                                 agino = gino + XFS_INODES_PER_CHUNK;
 448                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 449                         } else {
 450                                 /*
 451                                  * If any of those tests failed, bump the
 452                                  * inode number (just in case).
 453                                  */
 454                                 agino++;
 455                                 icount = 0;
 456                         }
 457                         /*
 458                          * In any case, increment to the next record.
 459                          */
 460                         if (!error)
 461                                 error = xfs_inobt_increment(cur, 0, &tmp);
 462                 } else {
 463                         /*
 464                          * Start of ag.  Lookup the first inode chunk.
 465                          */
 466                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 467                         icount = 0;
 468                 }
 469                 /*
 470                  * Loop through inode btree records in this ag,
 471                  * until we run out of inodes or space in the buffer.
 472                  */
 473                 while (irbp < irbufend && icount < ubcount) {
 474                         /*
 475                          * Loop as long as we're unable to read the
 476                          * inode btree.
 477                          */
 478                         while (error) {
 479                                 agino += XFS_INODES_PER_CHUNK;
 480                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 481                                                 be32_to_cpu(agi->agi_length))
 482                                         break;
 483                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 484                                                             &tmp);
 485                         }
 486                         /*
 487                          * If ran off the end of the ag either with an error,
 488                          * or the normal way, set end and stop collecting.
 489                          */
 490                         if (error ||
 491                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 492                                     &gfree, &i)) ||
 493                             i == 0) {
 494                                 end_of_ag = 1;
 495                                 break;
 496                         }
 497                         /*
 498                          * If this chunk has any allocated inodes, save it.
 499                          * Also start read-ahead now for this chunk.
 500                          */
 501                         if (gcnt < XFS_INODES_PER_CHUNK) {
 502                                 /*
 503                                  * Loop over all clusters in the next chunk.
 504                                  * Do a readahead if there are any allocated
 505                                  * inodes in that cluster.
 506                                  */
 507                                 for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
 508                                      chunkidx = 0;
 509                                      chunkidx < XFS_INODES_PER_CHUNK;
 510                                      chunkidx += nicluster,
 511                                      agbno += nbcluster) {
 512                                         if (XFS_INOBT_MASKN(chunkidx,
 513                                                             nicluster) & ~gfree)
 514                                                 xfs_btree_reada_bufs(mp, agno,
 515                                                         agbno, nbcluster);
 516                                 }
 517                                 irbp->ir_startino = gino;
 518                                 irbp->ir_freecount = gcnt;
 519                                 irbp->ir_free = gfree;
 520                                 irbp++;
 521                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 522                         }
 523                         /*
 524                          * Set agino to after this chunk and bump the cursor.
 525                          */
 526                         agino = gino + XFS_INODES_PER_CHUNK;
 527                         error = xfs_inobt_increment(cur, 0, &tmp);
 528                 }
 529                 /*
 530                  * Drop the btree buffers and the agi buffer.
 531                  * We can't hold any of the locks these represent
 532                  * when calling iget.
 533                  */
 534                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 535                 xfs_buf_relse(agbp);
 536                 /*
 537                  * Now format all the good inodes into the user's buffer.
 538                  */
 539                 irbufend = irbp;
 540                 for (irbp = irbuf;
 541                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 542                         /*
 543                          * Now process this chunk of inodes.
 544                          */
 545                         for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
 546                              ubleft > 0 &&
 547                                 irbp->ir_freecount < XFS_INODES_PER_CHUNK;
 548                              chunkidx++, clustidx++, agino++) {
 549                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 550                                 /*
 551                                  * Recompute agbno if this is the
 552                                  * first inode of the cluster.
 553                                  *
 554                                  * Careful with clustidx.   There can be
 555                                  * multple clusters per chunk, a single
 556                                  * cluster per chunk or a cluster that has
 557                                  * inodes represented from several different
 558                                  * chunks (if blocksize is large).
 559                                  *
 560                                  * Because of this, the starting clustidx is
 561                                  * initialized to zero in this loop but must
 562                                  * later be reset after reading in the cluster
 563                                  * buffer.
 564                                  */
 565                                 if ((chunkidx & (nicluster - 1)) == 0) {
 566                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 567                                                         irbp->ir_startino) +
 568                                                 ((chunkidx & nimask) >>
 569                                                  mp->m_sb.sb_inopblog);
 570
 571                                         if (flags & (BULKSTAT_FG_QUICK |
 572                                                      BULKSTAT_FG_INLINE)) {
 573                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 574                                                                        agino);
 575                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 576                                                                        agbno);
 577
 578                                                 /*
 579                                                  * Get the inode cluster buffer
 580                                                  */
 581                                                 ASSERT(xfs_inode_zone != NULL);
 582                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 583                                                                       KM_SLEEP);
 584                                                 ip->i_ino = ino;
 585                                                 ip->i_mount = mp;
 586                                                 spin_lock_init(&ip->i_flags_lock);
 587                                                 if (bp)
 588                                                         xfs_buf_relse(bp);
 589                                                 error = xfs_itobp(mp, NULL, ip,
 590                                                                 &dip, &bp, bno,
 591                                                                 XFS_IMAP_BULKSTAT);
 592                                                 if (!error)
 593                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 594                                                 kmem_zone_free(xfs_inode_zone, ip);
 595                                                 if (XFS_TEST_ERROR(error != 0,
 596                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 597                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 598                                                         bp = NULL;
 599                                                         ubleft = 0;
 600                                                         rval = error;
 601                                                         break;
 602                                                 }
 603                                         }
 604                                 }
 605                                 /*
 606                                  * Skip if this inode is free.
 607                                  */
 608                                 if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free)
 609                                         continue;
 610                                 /*
 611                                  * Count used inodes as free so we can tell
 612                                  * when the chunk is used up.
 613                                  */
 614                                 irbp->ir_freecount++;
 615                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 616                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 617                                 if (!xfs_bulkstat_use_dinode(mp, flags, bp,
 618                                                              clustidx, &dip))
 619                                         continue;
 620                                 /*
 621                                  * If we need to do an iget, cannot hold bp.
 622                                  * Drop it, until starting the next cluster.
 623                                  */
 624                                 if ((flags & BULKSTAT_FG_INLINE) && !dip) {
 625                                         if (bp)
 626                                                 xfs_buf_relse(bp);
 627                                         bp = NULL;
 628                                 }
 629
 630                                 /*
 631                                  * Get the inode and fill in a single buffer.
 632                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 633                                  * BULKSTAT_FG_IGET uses igets.
 634                                  * BULKSTAT_FG_INLINE uses dip if we have an
 635                                  * inline attr fork, else igets.
 636                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 637                                  * This is also used to count inodes/blks, etc
 638                                  * in xfs_qm_quotacheck.
 639                                  */
 640                                 ubused = statstruct_size;
 641                                 error = formatter(mp, ino, ubufp,
 642                                                 ubleft, private_data,
 643                                                 bno, &ubused, dip, &fmterror);
 644                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 645                                         if (error == EFAULT) {
 646                                                 ubleft = 0;
 647                                                 rval = error;
 648                                                 break;
 649                                         }
 650                                         else if (error == ENOMEM)
 651                                                 ubleft = 0;
 652                                         else
 653                                                 lastino = ino;
 654                                         continue;
 655                                 }
 656                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 657                                         ubleft = 0;
 658                                         ASSERT(error);
 659                                         rval = error;
 660                                         break;
 661                                 }
 662                                 if (ubufp)
 663                                         ubufp += ubused;
 664                                 ubleft -= ubused;
 665                                 ubelem++;
 666                                 lastino = ino;
 667                         }
 668                 }
 669
 670                 if (bp)
 671                         xfs_buf_relse(bp);
 672
 673                 /*
 674                  * Set up for the next loop iteration.
 675                  */
 676                 if (ubleft > 0) {
 677                         if (end_of_ag) {
 678                                 agno++;
 679                                 agino = 0;
 680                         } else
 681                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 682                 } else
 683                         break;
 684         }
 685         /*
 686          * Done, we're either out of filesystem or space to put the data.
 687          */
 688         kmem_free(irbuf, irbsize);
 689         *ubcountp = ubelem;
 690         if (agno >= mp->m_sb.sb_agcount) {
 691                 /*
 692                  * If we ran out of filesystem, mark lastino as off
 693                  * the end of the filesystem, so the next call
 694                  * will return immediately.
 695                  */
 696                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 697                 *done = 1;
 698         } else
 699                 *lastinop = (xfs_ino_t)lastino;
 700
 701         return rval;
 702 }
 703
 704 /*
 705  * Return stat information in bulk (by-inode) for the filesystem.
 706  * Special case for non-sequential one inode bulkstat.
 707  */
 708 int                                     /* error status */
 709 xfs_bulkstat_single(
 710         xfs_mount_t             *mp,    /* mount point for filesystem */
 711         xfs_ino_t               *lastinop, /* inode to return */
 712         char                    __user *buffer, /* buffer with inode stats */
 713         int                     *done)  /* 1 if there are more stats to get */
 714 {
 715         int                     count;  /* count value for bulkstat call */
 716         int                     error;  /* return value */
 717         xfs_ino_t               ino;    /* filesystem inode number */
 718         int                     res;    /* result from bs1 */
 719
 720         /*
 721          * note that requesting valid inode numbers which are not allocated
 722          * to inodes will most likely cause xfs_itobp to generate warning
 723          * messages about bad magic numbers. This is ok. The fact that
 724          * the inode isn't actually an inode is handled by the
 725          * error check below. Done this way to make the usual case faster
 726          * at the expense of the error case.
 727          */
 728
 729         ino = (xfs_ino_t)*lastinop;
 730         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 731                                  NULL, 0, NULL, NULL, &res);
 732         if (error) {
 733                 /*
 734                  * Special case way failed, do it the "long" way
 735                  * to see if that works.
 736                  */
 737                 (*lastinop)--;
 738                 count = 1;
 739                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 740                                 NULL, sizeof(xfs_bstat_t), buffer,
 741                                 BULKSTAT_FG_IGET, done))
 742                         return error;
 743                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 744                         return error == EFSCORRUPTED ?
 745                                 XFS_ERROR(EINVAL) : error;
 746                 else
 747                         return 0;
 748         }
 749         *done = 0;
 750         return 0;
 751 }
 752
 753 /*
 754  * Return inode number table for the filesystem.
 755  */
 756 int                                     /* error status */
 757 xfs_inumbers(
 758         xfs_mount_t     *mp,            /* mount point for filesystem */
 759         xfs_ino_t       *lastino,       /* last inode returned */
 760         int             *count,         /* size of buffer/count returned */
 761         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 762 {
 763         xfs_buf_t       *agbp;
 764         xfs_agino_t     agino;
 765         xfs_agnumber_t  agno;
 766         int             bcount;
 767         xfs_inogrp_t    *buffer;
 768         int             bufidx;
 769         xfs_btree_cur_t *cur;
 770         int             error;
 771         __int32_t       gcnt;
 772         xfs_inofree_t   gfree;
 773         xfs_agino_t     gino;
 774         int             i;
 775         xfs_ino_t       ino;
 776         int             left;
 777         int             tmp;
 778
 779         ino = (xfs_ino_t)*lastino;
 780         agno = XFS_INO_TO_AGNO(mp, ino);
 781         agino = XFS_INO_TO_AGINO(mp, ino);
 782         left = *count;
 783         *count = 0;
 784         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 785         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 786         error = bufidx = 0;
 787         cur = NULL;
 788         agbp = NULL;
 789         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 790                 if (agbp == NULL) {
 791                         down_read(&mp->m_peraglock);
 792                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 793                         up_read(&mp->m_peraglock);
 794                         if (error) {
 795                                 /*
 796                                  * If we can't read the AGI of this ag,
 797                                  * then just skip to the next one.
 798                                  */
 799                                 ASSERT(cur == NULL);
 800                                 agbp = NULL;
 801                                 agno++;
 802                                 agino = 0;
 803                                 continue;
 804                         }
 805                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 806                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 807                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 808                         if (error) {
 809                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 810                                 cur = NULL;
 811                                 xfs_buf_relse(agbp);
 812                                 agbp = NULL;
 813                                 /*
 814                                  * Move up the the last inode in the current
 815                                  * chunk.  The lookup_ge will always get
 816                                  * us the first inode in the next chunk.
 817                                  */
 818                                 agino += XFS_INODES_PER_CHUNK - 1;
 819                                 continue;
 820                         }
 821                 }
 822                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 823                         &i)) ||
 824                     i == 0) {
 825                         xfs_buf_relse(agbp);
 826                         agbp = NULL;
 827                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 828                         cur = NULL;
 829                         agno++;
 830                         agino = 0;
 831                         continue;
 832                 }
 833                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 834                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 835                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 836                 buffer[bufidx].xi_allocmask = ~gfree;
 837                 bufidx++;
 838                 left--;
 839                 if (bufidx == bcount) {
 840                         if (copy_to_user(ubuffer, buffer,
 841                                         bufidx * sizeof(*buffer))) {
 842                                 error = XFS_ERROR(EFAULT);
 843                                 break;
 844                         }
 845                         ubuffer += bufidx;
 846                         *count += bufidx;
 847                         bufidx = 0;
 848                 }
 849                 if (left) {
 850                         error = xfs_inobt_increment(cur, 0, &tmp);
 851                         if (error) {
 852                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 853                                 cur = NULL;
 854                                 xfs_buf_relse(agbp);
 855                                 agbp = NULL;
 856                                 /*
 857                                  * The agino value has already been bumped.
 858                                  * Just try to skip up to it.
 859                                  */
 860                                 agino += XFS_INODES_PER_CHUNK;
 861                                 continue;
 862                         }
 863                 }
 864         }
 865         if (!error) {
 866                 if (bufidx) {
 867                         if (copy_to_user(ubuffer, buffer,
 868                                         bufidx * sizeof(*buffer)))
 869                                 error = XFS_ERROR(EFAULT);
 870                         else
 871                                 *count += bufidx;
 872                 }
 873                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 874         }
 875         kmem_free(buffer, bcount * sizeof(*buffer));
 876         if (cur)
 877                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 878                                            XFS_BTREE_NOERROR));
 879         if (agbp)
 880                 xfs_buf_relse(agbp);
 881         return error;
 882 }