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 STATIC int
  43 xfs_bulkstat_one_iget(
  44         xfs_mount_t     *mp,            /* mount point for filesystem */
  45         xfs_ino_t       ino,            /* inode number to get data for */
  46         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  47         xfs_bstat_t     *buf,           /* return buffer */
  48         int             *stat)          /* BULKSTAT_RV_... */
  49 {
  50         xfs_dinode_core_t *dic;         /* dinode core info pointer */
  51         xfs_inode_t     *ip;            /* incore inode pointer */
  52         bhv_vnode_t     *vp;
  53         int             error;
  54
  55         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
  56         if (error) {
  57                 *stat = BULKSTAT_RV_NOTHING;
  58                 return error;
  59         }
  60
  61         ASSERT(ip != NULL);
  62         ASSERT(ip->i_blkno != (xfs_daddr_t)0);
  63         if (ip->i_d.di_mode == 0) {
  64                 *stat = BULKSTAT_RV_NOTHING;
  65                 error = XFS_ERROR(ENOENT);
  66                 goto out_iput;
  67         }
  68
  69         vp = XFS_ITOV(ip);
  70         dic = &ip->i_d;
  71
  72         /* xfs_iget returns the following without needing
  73          * further change.
  74          */
  75         buf->bs_nlink = dic->di_nlink;
  76         buf->bs_projid = dic->di_projid;
  77         buf->bs_ino = ino;
  78         buf->bs_mode = dic->di_mode;
  79         buf->bs_uid = dic->di_uid;
  80         buf->bs_gid = dic->di_gid;
  81         buf->bs_xid = dic->di_xid;
  82         buf->bs_size = dic->di_size;
  83         vn_atime_to_bstime(vp, &buf->bs_atime);
  84         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
  85         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
  86         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
  87         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
  88         buf->bs_xflags = xfs_ip2xflags(ip);
  89         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
  90         buf->bs_extents = dic->di_nextents;
  91         buf->bs_gen = dic->di_gen;
  92         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
  93         buf->bs_dmevmask = dic->di_dmevmask;
  94         buf->bs_dmstate = dic->di_dmstate;
  95         buf->bs_aextents = dic->di_anextents;
  96
  97         switch (dic->di_format) {
  98         case XFS_DINODE_FMT_DEV:
  99                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 100                 buf->bs_blksize = BLKDEV_IOSIZE;
 101                 buf->bs_blocks = 0;
 102                 break;
 103         case XFS_DINODE_FMT_LOCAL:
 104         case XFS_DINODE_FMT_UUID:
 105                 buf->bs_rdev = 0;
 106                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 107                 buf->bs_blocks = 0;
 108                 break;
 109         case XFS_DINODE_FMT_EXTENTS:
 110         case XFS_DINODE_FMT_BTREE:
 111                 buf->bs_rdev = 0;
 112                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 113                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 114                 break;
 115         }
 116
 117  out_iput:
 118         xfs_iput(ip, XFS_ILOCK_SHARED);
 119         return error;
 120 }
 121
 122 STATIC int
 123 xfs_bulkstat_one_dinode(
 124         xfs_mount_t     *mp,            /* mount point for filesystem */
 125         xfs_ino_t       ino,            /* inode number to get data for */
 126         xfs_dinode_t    *dip,           /* dinode inode pointer */
 127         xfs_bstat_t     *buf)           /* return buffer */
 128 {
 129         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 130
 131         dic = &dip->di_core;
 132
 133         /*
 134          * The inode format changed when we moved the link count and
 135          * made it 32 bits long.  If this is an old format inode,
 136          * convert it in memory to look like a new one.  If it gets
 137          * flushed to disk we will convert back before flushing or
 138          * logging it.  We zero out the new projid field and the old link
 139          * count field.  We'll handle clearing the pad field (the remains
 140          * of the old uuid field) when we actually convert the inode to
 141          * the new format. We don't change the version number so that we
 142          * can distinguish this from a real new format inode.
 143          */
 144         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 145                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 146                 buf->bs_projid = 0;
 147         } else {
 148                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 149                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 150         }
 151
 152         buf->bs_ino = ino;
 153         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 154         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 155         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 156         buf->bs_xid = INT_GET(dic->di_xid, ARCH_CONVERT);
 157         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 158         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 159         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 160         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 161         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 162         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 163         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 164         buf->bs_xflags = xfs_dic2xflags(dic);
 165         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 166         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 167         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 168         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 169         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 170         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 171         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 172
 173         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 174         case XFS_DINODE_FMT_DEV:
 175                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 176                 buf->bs_blksize = BLKDEV_IOSIZE;
 177                 buf->bs_blocks = 0;
 178                 break;
 179         case XFS_DINODE_FMT_LOCAL:
 180         case XFS_DINODE_FMT_UUID:
 181                 buf->bs_rdev = 0;
 182                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 183                 buf->bs_blocks = 0;
 184                 break;
 185         case XFS_DINODE_FMT_EXTENTS:
 186         case XFS_DINODE_FMT_BTREE:
 187                 buf->bs_rdev = 0;
 188                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 189                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 190                 break;
 191         }
 192
 193         return 0;
 194 }
 195
 196 /*
 197  * Return stat information for one inode.
 198  * Return 0 if ok, else errno.
 199  */
 200 int                             /* error status */
 201 xfs_bulkstat_one(
 202         xfs_mount_t     *mp,            /* mount point for filesystem */
 203         xfs_ino_t       ino,            /* inode number to get data for */
 204         void            __user *buffer, /* buffer to place output in */
 205         int             ubsize,         /* size of buffer */
 206         void            *private_data,  /* my private data */
 207         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 208         int             *ubused,        /* bytes used by me */
 209         void            *dibuff,        /* on-disk inode buffer */
 210         int             *stat)          /* BULKSTAT_RV_... */
 211 {
 212         xfs_bstat_t     *buf;           /* return buffer */
 213         int             error = 0;      /* error value */
 214         xfs_dinode_t    *dip;           /* dinode inode pointer */
 215
 216         dip = (xfs_dinode_t *)dibuff;
 217
 218         if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
 219             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 220              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
 221                 *stat = BULKSTAT_RV_NOTHING;
 222                 return XFS_ERROR(EINVAL);
 223         }
 224         if (ubsize < sizeof(*buf)) {
 225                 *stat = BULKSTAT_RV_NOTHING;
 226                 return XFS_ERROR(ENOMEM);
 227         }
 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         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 243                 *stat = BULKSTAT_RV_NOTHING;
 244                 error =  EFAULT;
 245                 goto out_free;
 246         }
 247
 248         *stat = BULKSTAT_RV_DIDONE;
 249         if (ubused)
 250                 *ubused = sizeof(*buf);
 251
 252  out_free:
 253         kmem_free(buf, sizeof(*buf));
 254         return error;
 255 }
 256
 257 /*
 258  * Return stat information in bulk (by-inode) for the filesystem.
 259  */
 260 int                                     /* error status */
 261 xfs_bulkstat(
 262         xfs_mount_t             *mp,    /* mount point for filesystem */
 263         xfs_ino_t               *lastinop, /* last inode returned */
 264         int                     *ubcountp, /* size of buffer/count returned */
 265         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 266         void                    *private_data,/* private data for formatter */
 267         size_t                  statstruct_size, /* sizeof struct filling */
 268         char                    __user *ubuffer, /* buffer with inode stats */
 269         int                     flags,  /* defined in xfs_itable.h */
 270         int                     *done)  /* 1 if there are more stats to get */
 271 {
 272         xfs_agblock_t           agbno=0;/* allocation group block number */
 273         xfs_buf_t               *agbp;  /* agi header buffer */
 274         xfs_agi_t               *agi;   /* agi header data */
 275         xfs_agino_t             agino;  /* inode # in allocation group */
 276         xfs_agnumber_t          agno;   /* allocation group number */
 277         xfs_daddr_t             bno;    /* inode cluster start daddr */
 278         int                     chunkidx; /* current index into inode chunk */
 279         int                     clustidx; /* current index into inode cluster */
 280         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 281         int                     end_of_ag; /* set if we've seen the ag end */
 282         int                     error;  /* error code */
 283         int                     fmterror;/* bulkstat formatter result */
 284         __int32_t               gcnt;   /* current btree rec's count */
 285         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 286         xfs_agino_t             gino;   /* current btree rec's start inode */
 287         int                     i;      /* loop index */
 288         int                     icount; /* count of inodes good in irbuf */
 289         xfs_ino_t               ino;    /* inode number (filesystem) */
 290         xfs_inobt_rec_t         *irbp;  /* current irec buffer pointer */
 291         xfs_inobt_rec_t         *irbuf; /* start of irec buffer */
 292         xfs_inobt_rec_t         *irbufend; /* end of good irec buffer entries */
 293         xfs_ino_t               lastino=0; /* last inode number returned */
 294         int                     nbcluster; /* # of blocks in a cluster */
 295         int                     nicluster; /* # of inodes in a cluster */
 296         int                     nimask; /* mask for inode clusters */
 297         int                     nirbuf; /* size of irbuf */
 298         int                     rval;   /* return value error code */
 299         int                     tmp;    /* result value from btree calls */
 300         int                     ubcount; /* size of user's buffer */
 301         int                     ubleft; /* bytes left in user's buffer */
 302         char                    __user *ubufp;  /* pointer into user's buffer */
 303         int                     ubelem; /* spaces used in user's buffer */
 304         int                     ubused; /* bytes used by formatter */
 305         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 306         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 307         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 308
 309         /*
 310          * Get the last inode value, see if there's nothing to do.
 311          */
 312         ino = (xfs_ino_t)*lastinop;
 313         dip = NULL;
 314         agno = XFS_INO_TO_AGNO(mp, ino);
 315         agino = XFS_INO_TO_AGINO(mp, ino);
 316         if (agno >= mp->m_sb.sb_agcount ||
 317             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 318                 *done = 1;
 319                 *ubcountp = 0;
 320                 return 0;
 321         }
 322         ubcount = *ubcountp; /* statstruct's */
 323         ubleft = ubcount * statstruct_size; /* bytes */
 324         *ubcountp = ubelem = 0;
 325         *done = 0;
 326         fmterror = 0;
 327         ubufp = ubuffer;
 328         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 329                 mp->m_sb.sb_inopblock :
 330                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 331         nimask = ~(nicluster - 1);
 332         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 333         /*
 334          * Allocate a page-sized buffer for inode btree records.
 335          * We could try allocating something smaller, but for normal
 336          * calls we'll always (potentially) need the whole page.
 337          */
 338         irbuf = kmem_alloc(NBPC, KM_SLEEP);
 339         nirbuf = NBPC / sizeof(*irbuf);
 340         /*
 341          * Loop over the allocation groups, starting from the last
 342          * inode returned; 0 means start of the allocation group.
 343          */
 344         rval = 0;
 345         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 346                 bp = NULL;
 347                 down_read(&mp->m_peraglock);
 348                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 349                 up_read(&mp->m_peraglock);
 350                 if (error) {
 351                         /*
 352                          * Skip this allocation group and go to the next one.
 353                          */
 354                         agno++;
 355                         agino = 0;
 356                         continue;
 357                 }
 358                 agi = XFS_BUF_TO_AGI(agbp);
 359                 /*
 360                  * Allocate and initialize a btree cursor for ialloc btree.
 361                  */
 362                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 363                         (xfs_inode_t *)0, 0);
 364                 irbp = irbuf;
 365                 irbufend = irbuf + nirbuf;
 366                 end_of_ag = 0;
 367                 /*
 368                  * If we're returning in the middle of an allocation group,
 369                  * we need to get the remainder of the chunk we're in.
 370                  */
 371                 if (agino > 0) {
 372                         /*
 373                          * Lookup the inode chunk that this inode lives in.
 374                          */
 375                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 376                         if (!error &&   /* no I/O error */
 377                             tmp &&      /* lookup succeeded */
 378                                         /* got the record, should always work */
 379                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 380                                     &gfree, &i)) &&
 381                             i == 1 &&
 382                                         /* this is the right chunk */
 383                             agino < gino + XFS_INODES_PER_CHUNK &&
 384                                         /* lastino was not last in chunk */
 385                             (chunkidx = agino - gino + 1) <
 386                                     XFS_INODES_PER_CHUNK &&
 387                                         /* there are some left allocated */
 388                             XFS_INOBT_MASKN(chunkidx,
 389                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 390                                 /*
 391                                  * Grab the chunk record.  Mark all the
 392                                  * uninteresting inodes (because they're
 393                                  * before our start point) free.
 394                                  */
 395                                 for (i = 0; i < chunkidx; i++) {
 396                                         if (XFS_INOBT_MASK(i) & ~gfree)
 397                                                 gcnt++;
 398                                 }
 399                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 400                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 401                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 402                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 403                                 irbp++;
 404                                 agino = gino + XFS_INODES_PER_CHUNK;
 405                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 406                         } else {
 407                                 /*
 408                                  * If any of those tests failed, bump the
 409                                  * inode number (just in case).
 410                                  */
 411                                 agino++;
 412                                 icount = 0;
 413                         }
 414                         /*
 415                          * In any case, increment to the next record.
 416                          */
 417                         if (!error)
 418                                 error = xfs_inobt_increment(cur, 0, &tmp);
 419                 } else {
 420                         /*
 421                          * Start of ag.  Lookup the first inode chunk.
 422                          */
 423                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 424                         icount = 0;
 425                 }
 426                 /*
 427                  * Loop through inode btree records in this ag,
 428                  * until we run out of inodes or space in the buffer.
 429                  */
 430                 while (irbp < irbufend && icount < ubcount) {
 431                         /*
 432                          * Loop as long as we're unable to read the
 433                          * inode btree.
 434                          */
 435                         while (error) {
 436                                 agino += XFS_INODES_PER_CHUNK;
 437                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 438                                                 be32_to_cpu(agi->agi_length))
 439                                         break;
 440                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 441                                                             &tmp);
 442                         }
 443                         /*
 444                          * If ran off the end of the ag either with an error,
 445                          * or the normal way, set end and stop collecting.
 446                          */
 447                         if (error ||
 448                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 449                                     &gfree, &i)) ||
 450                             i == 0) {
 451                                 end_of_ag = 1;
 452                                 break;
 453                         }
 454                         /*
 455                          * If this chunk has any allocated inodes, save it.
 456                          */
 457                         if (gcnt < XFS_INODES_PER_CHUNK) {
 458                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 459                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 460                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 461                                 irbp++;
 462                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 463                         }
 464                         /*
 465                          * Set agino to after this chunk and bump the cursor.
 466                          */
 467                         agino = gino + XFS_INODES_PER_CHUNK;
 468                         error = xfs_inobt_increment(cur, 0, &tmp);
 469                 }
 470                 /*
 471                  * Drop the btree buffers and the agi buffer.
 472                  * We can't hold any of the locks these represent
 473                  * when calling iget.
 474                  */
 475                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 476                 xfs_buf_relse(agbp);
 477                 /*
 478                  * Now format all the good inodes into the user's buffer.
 479                  */
 480                 irbufend = irbp;
 481                 for (irbp = irbuf;
 482                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 483                         /*
 484                          * Read-ahead the next chunk's worth of inodes.
 485                          */
 486                         if (&irbp[1] < irbufend) {
 487                                 /*
 488                                  * Loop over all clusters in the next chunk.
 489                                  * Do a readahead if there are any allocated
 490                                  * inodes in that cluster.
 491                                  */
 492                                 for (agbno = XFS_AGINO_TO_AGBNO(mp,
 493                                                         INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
 494                                      chunkidx = 0;
 495                                      chunkidx < XFS_INODES_PER_CHUNK;
 496                                      chunkidx += nicluster,
 497                                      agbno += nbcluster) {
 498                                         if (XFS_INOBT_MASKN(chunkidx,
 499                                                             nicluster) &
 500                                             ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
 501                                                 xfs_btree_reada_bufs(mp, agno,
 502                                                         agbno, nbcluster);
 503                                 }
 504                         }
 505                         /*
 506                          * Now process this chunk of inodes.
 507                          */
 508                         for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
 509                              ubleft > 0 &&
 510                                 INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
 511                              chunkidx++, clustidx++, agino++) {
 512                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 513                                 /*
 514                                  * Recompute agbno if this is the
 515                                  * first inode of the cluster.
 516                                  *
 517                                  * Careful with clustidx.   There can be
 518                                  * multple clusters per chunk, a single
 519                                  * cluster per chunk or a cluster that has
 520                                  * inodes represented from several different
 521                                  * chunks (if blocksize is large).
 522                                  *
 523                                  * Because of this, the starting clustidx is
 524                                  * initialized to zero in this loop but must
 525                                  * later be reset after reading in the cluster
 526                                  * buffer.
 527                                  */
 528                                 if ((chunkidx & (nicluster - 1)) == 0) {
 529                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 530                                                         INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
 531                                                 ((chunkidx & nimask) >>
 532                                                  mp->m_sb.sb_inopblog);
 533
 534                                         if (flags & BULKSTAT_FG_QUICK) {
 535                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 536                                                                        agino);
 537                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 538                                                                        agbno);
 539
 540                                                 /*
 541                                                  * Get the inode cluster buffer
 542                                                  */
 543                                                 ASSERT(xfs_inode_zone != NULL);
 544                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 545                                                                       KM_SLEEP);
 546                                                 ip->i_ino = ino;
 547                                                 ip->i_mount = mp;
 548                                                 if (bp)
 549                                                         xfs_buf_relse(bp);
 550                                                 error = xfs_itobp(mp, NULL, ip,
 551                                                                 &dip, &bp, bno,
 552                                                                 XFS_IMAP_BULKSTAT);
 553                                                 if (!error)
 554                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 555                                                 kmem_zone_free(xfs_inode_zone, ip);
 556                                                 if (XFS_TEST_ERROR(error != 0,
 557                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 558                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 559                                                         bp = NULL;
 560                                                         ubleft = 0;
 561                                                         rval = error;
 562                                                         break;
 563                                                 }
 564                                         }
 565                                 }
 566                                 /*
 567                                  * Skip if this inode is free.
 568                                  */
 569                                 if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
 570                                         continue;
 571                                 /*
 572                                  * Count used inodes as free so we can tell
 573                                  * when the chunk is used up.
 574                                  */
 575                                 INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
 576                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 577                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 578                                 if (flags & BULKSTAT_FG_QUICK) {
 579                                         dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 580                                               (clustidx << mp->m_sb.sb_inodelog));
 581
 582                                         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
 583                                                     != XFS_DINODE_MAGIC
 584                                             || !XFS_DINODE_GOOD_VERSION(
 585                                                     INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 586                                                 continue;
 587                                 }
 588
 589                                 /*
 590                                  * Get the inode and fill in a single buffer.
 591                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 592                                  * BULKSTAT_FG_IGET uses igets.
 593                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 594                                  * This is also used to count inodes/blks, etc
 595                                  * in xfs_qm_quotacheck.
 596                                  */
 597                                 ubused = statstruct_size;
 598                                 error = formatter(mp, ino, ubufp,
 599                                                 ubleft, private_data,
 600                                                 bno, &ubused, dip, &fmterror);
 601                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 602                                         if (error == ENOMEM)
 603                                                 ubleft = 0;
 604                                         continue;
 605                                 }
 606                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 607                                         ubleft = 0;
 608                                         ASSERT(error);
 609                                         rval = error;
 610                                         break;
 611                                 }
 612                                 if (ubufp)
 613                                         ubufp += ubused;
 614                                 ubleft -= ubused;
 615                                 ubelem++;
 616                                 lastino = ino;
 617                         }
 618                 }
 619
 620                 if (bp)
 621                         xfs_buf_relse(bp);
 622
 623                 /*
 624                  * Set up for the next loop iteration.
 625                  */
 626                 if (ubleft > 0) {
 627                         if (end_of_ag) {
 628                                 agno++;
 629                                 agino = 0;
 630                         } else
 631                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 632                 } else
 633                         break;
 634         }
 635         /*
 636          * Done, we're either out of filesystem or space to put the data.
 637          */
 638         kmem_free(irbuf, NBPC);
 639         *ubcountp = ubelem;
 640         if (agno >= mp->m_sb.sb_agcount) {
 641                 /*
 642                  * If we ran out of filesystem, mark lastino as off
 643                  * the end of the filesystem, so the next call
 644                  * will return immediately.
 645                  */
 646                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 647                 *done = 1;
 648         } else
 649                 *lastinop = (xfs_ino_t)lastino;
 650
 651         return rval;
 652 }
 653
 654 /*
 655  * Return stat information in bulk (by-inode) for the filesystem.
 656  * Special case for non-sequential one inode bulkstat.
 657  */
 658 int                                     /* error status */
 659 xfs_bulkstat_single(
 660         xfs_mount_t             *mp,    /* mount point for filesystem */
 661         xfs_ino_t               *lastinop, /* inode to return */
 662         char                    __user *buffer, /* buffer with inode stats */
 663         int                     *done)  /* 1 if there are more stats to get */
 664 {
 665         int                     count;  /* count value for bulkstat call */
 666         int                     error;  /* return value */
 667         xfs_ino_t               ino;    /* filesystem inode number */
 668         int                     res;    /* result from bs1 */
 669
 670         /*
 671          * note that requesting valid inode numbers which are not allocated
 672          * to inodes will most likely cause xfs_itobp to generate warning
 673          * messages about bad magic numbers. This is ok. The fact that
 674          * the inode isn't actually an inode is handled by the
 675          * error check below. Done this way to make the usual case faster
 676          * at the expense of the error case.
 677          */
 678
 679         ino = (xfs_ino_t)*lastinop;
 680         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 681                                  NULL, 0, NULL, NULL, &res);
 682         if (error) {
 683                 /*
 684                  * Special case way failed, do it the "long" way
 685                  * to see if that works.
 686                  */
 687                 (*lastinop)--;
 688                 count = 1;
 689                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 690                                 NULL, sizeof(xfs_bstat_t), buffer,
 691                                 BULKSTAT_FG_IGET, done))
 692                         return error;
 693                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 694                         return error == EFSCORRUPTED ?
 695                                 XFS_ERROR(EINVAL) : error;
 696                 else
 697                         return 0;
 698         }
 699         *done = 0;
 700         return 0;
 701 }
 702
 703 /*
 704  * Return inode number table for the filesystem.
 705  */
 706 int                                     /* error status */
 707 xfs_inumbers(
 708         xfs_mount_t     *mp,            /* mount point for filesystem */
 709         xfs_ino_t       *lastino,       /* last inode returned */
 710         int             *count,         /* size of buffer/count returned */
 711         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 712 {
 713         xfs_buf_t       *agbp;
 714         xfs_agino_t     agino;
 715         xfs_agnumber_t  agno;
 716         int             bcount;
 717         xfs_inogrp_t    *buffer;
 718         int             bufidx;
 719         xfs_btree_cur_t *cur;
 720         int             error;
 721         __int32_t       gcnt;
 722         xfs_inofree_t   gfree;
 723         xfs_agino_t     gino;
 724         int             i;
 725         xfs_ino_t       ino;
 726         int             left;
 727         int             tmp;
 728
 729         ino = (xfs_ino_t)*lastino;
 730         agno = XFS_INO_TO_AGNO(mp, ino);
 731         agino = XFS_INO_TO_AGINO(mp, ino);
 732         left = *count;
 733         *count = 0;
 734         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 735         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 736         error = bufidx = 0;
 737         cur = NULL;
 738         agbp = NULL;
 739         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 740                 if (agbp == NULL) {
 741                         down_read(&mp->m_peraglock);
 742                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 743                         up_read(&mp->m_peraglock);
 744                         if (error) {
 745                                 /*
 746                                  * If we can't read the AGI of this ag,
 747                                  * then just skip to the next one.
 748                                  */
 749                                 ASSERT(cur == NULL);
 750                                 agbp = NULL;
 751                                 agno++;
 752                                 agino = 0;
 753                                 continue;
 754                         }
 755                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 756                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 757                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 758                         if (error) {
 759                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 760                                 cur = NULL;
 761                                 xfs_buf_relse(agbp);
 762                                 agbp = NULL;
 763                                 /*
 764                                  * Move up the the last inode in the current
 765                                  * chunk.  The lookup_ge will always get
 766                                  * us the first inode in the next chunk.
 767                                  */
 768                                 agino += XFS_INODES_PER_CHUNK - 1;
 769                                 continue;
 770                         }
 771                 }
 772                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 773                         &i)) ||
 774                     i == 0) {
 775                         xfs_buf_relse(agbp);
 776                         agbp = NULL;
 777                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 778                         cur = NULL;
 779                         agno++;
 780                         agino = 0;
 781                         continue;
 782                 }
 783                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 784                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 785                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 786                 buffer[bufidx].xi_allocmask = ~gfree;
 787                 bufidx++;
 788                 left--;
 789                 if (bufidx == bcount) {
 790                         if (copy_to_user(ubuffer, buffer,
 791                                         bufidx * sizeof(*buffer))) {
 792                                 error = XFS_ERROR(EFAULT);
 793                                 break;
 794                         }
 795                         ubuffer += bufidx;
 796                         *count += bufidx;
 797                         bufidx = 0;
 798                 }
 799                 if (left) {
 800                         error = xfs_inobt_increment(cur, 0, &tmp);
 801                         if (error) {
 802                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 803                                 cur = NULL;
 804                                 xfs_buf_relse(agbp);
 805                                 agbp = NULL;
 806                                 /*
 807                                  * The agino value has already been bumped.
 808                                  * Just try to skip up to it.
 809                                  */
 810                                 agino += XFS_INODES_PER_CHUNK;
 811                                 continue;
 812                         }
 813                 }
 814         }
 815         if (!error) {
 816                 if (bufidx) {
 817                         if (copy_to_user(ubuffer, buffer,
 818                                         bufidx * sizeof(*buffer)))
 819                                 error = XFS_ERROR(EFAULT);
 820                         else
 821                                 *count += bufidx;
 822                 }
 823                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 824         }
 825         kmem_free(buffer, bcount * sizeof(*buffer));
 826         if (cur)
 827                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 828                                            XFS_BTREE_NOERROR));
 829         if (agbp)
 830                 xfs_buf_relse(agbp);
 831         return error;
 832 }