fs/xfs/xfs_itable.c

   1 /*
   2  * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34 #include "xfs_macros.h"
  35 #include "xfs_types.h"
  36 #include "xfs_inum.h"
  37 #include "xfs_log.h"
  38 #include "xfs_trans.h"
  39 #include "xfs_sb.h"
  40 #include "xfs_dir.h"
  41 #include "xfs_dir2.h"
  42 #include "xfs_dmapi.h"
  43 #include "xfs_mount.h"
  44 #include "xfs_ag.h"
  45 #include "xfs_alloc_btree.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_btree.h"
  49 #include "xfs_attr_sf.h"
  50 #include "xfs_dir_sf.h"
  51 #include "xfs_dir2_sf.h"
  52 #include "xfs_dinode.h"
  53 #include "xfs_inode.h"
  54 #include "xfs_ialloc.h"
  55 #include "xfs_itable.h"
  56 #include "xfs_error.h"
  57
  58 #ifndef HAVE_USERACC
  59 #define useracc(ubuffer, size, flags, foo) (0)
  60 #define unuseracc(ubuffer, size, flags)
  61 #endif
  62
  63 STATIC int
  64 xfs_bulkstat_one_iget(
  65         xfs_mount_t     *mp,            /* mount point for filesystem */
  66         xfs_ino_t       ino,            /* inode number to get data for */
  67         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  68         xfs_bstat_t     *buf,           /* return buffer */
  69         int             *stat)          /* BULKSTAT_RV_... */
  70 {
  71         xfs_dinode_core_t *dic;         /* dinode core info pointer */
  72         xfs_inode_t     *ip;            /* incore inode pointer */
  73         int             error;
  74
  75         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
  76         if (error) {
  77                 *stat = BULKSTAT_RV_NOTHING;
  78                 return error;
  79         }
  80
  81         ASSERT(ip != NULL);
  82         ASSERT(ip->i_blkno != (xfs_daddr_t)0);
  83         if (ip->i_d.di_mode == 0) {
  84                 *stat = BULKSTAT_RV_NOTHING;
  85                 error = XFS_ERROR(ENOENT);
  86                 goto out_iput;
  87         }
  88
  89         dic = &ip->i_d;
  90
  91         /* xfs_iget returns the following without needing
  92          * further change.
  93          */
  94         buf->bs_nlink = dic->di_nlink;
  95         buf->bs_projid = dic->di_projid;
  96         buf->bs_ino = ino;
  97         buf->bs_mode = dic->di_mode;
  98         buf->bs_uid = dic->di_uid;
  99         buf->bs_gid = dic->di_gid;
 100         buf->bs_xid = dic->di_xid;
 101         buf->bs_size = dic->di_size;
 102         buf->bs_atime.tv_sec = dic->di_atime.t_sec;
 103         buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
 104         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
 105         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
 106         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
 107         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
 108         buf->bs_xflags = xfs_ip2xflags(ip);
 109         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
 110         buf->bs_extents = dic->di_nextents;
 111         buf->bs_gen = dic->di_gen;
 112         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 113         buf->bs_dmevmask = dic->di_dmevmask;
 114         buf->bs_dmstate = dic->di_dmstate;
 115         buf->bs_aextents = dic->di_anextents;
 116
 117         switch (dic->di_format) {
 118         case XFS_DINODE_FMT_DEV:
 119                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 120                 buf->bs_blksize = BLKDEV_IOSIZE;
 121                 buf->bs_blocks = 0;
 122                 break;
 123         case XFS_DINODE_FMT_LOCAL:
 124         case XFS_DINODE_FMT_UUID:
 125                 buf->bs_rdev = 0;
 126                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 127                 buf->bs_blocks = 0;
 128                 break;
 129         case XFS_DINODE_FMT_EXTENTS:
 130         case XFS_DINODE_FMT_BTREE:
 131                 buf->bs_rdev = 0;
 132                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 133                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 134                 break;
 135         }
 136
 137  out_iput:
 138         xfs_iput(ip, XFS_ILOCK_SHARED);
 139         return error;
 140 }
 141
 142 STATIC int
 143 xfs_bulkstat_one_dinode(
 144         xfs_mount_t     *mp,            /* mount point for filesystem */
 145         xfs_ino_t       ino,            /* inode number to get data for */
 146         xfs_dinode_t    *dip,           /* dinode inode pointer */
 147         xfs_bstat_t     *buf)           /* return buffer */
 148 {
 149         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 150
 151         dic = &dip->di_core;
 152
 153         /*
 154          * The inode format changed when we moved the link count and
 155          * made it 32 bits long.  If this is an old format inode,
 156          * convert it in memory to look like a new one.  If it gets
 157          * flushed to disk we will convert back before flushing or
 158          * logging it.  We zero out the new projid field and the old link
 159          * count field.  We'll handle clearing the pad field (the remains
 160          * of the old uuid field) when we actually convert the inode to
 161          * the new format. We don't change the version number so that we
 162          * can distinguish this from a real new format inode.
 163          */
 164         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 165                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 166                 buf->bs_projid = 0;
 167         } else {
 168                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 169                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 170         }
 171
 172         buf->bs_ino = ino;
 173         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 174         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 175         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 176         buf->bs_xid = INT_GET(dic->di_xid, ARCH_CONVERT);
 177         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 178         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 179         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 180         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 181         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 182         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 183         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 184         buf->bs_xflags = xfs_dic2xflags(dic);
 185         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 186         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 187         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 188         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 189         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 190         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 191         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 192
 193         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 194         case XFS_DINODE_FMT_DEV:
 195                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 196                 buf->bs_blksize = BLKDEV_IOSIZE;
 197                 buf->bs_blocks = 0;
 198                 break;
 199         case XFS_DINODE_FMT_LOCAL:
 200         case XFS_DINODE_FMT_UUID:
 201                 buf->bs_rdev = 0;
 202                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 203                 buf->bs_blocks = 0;
 204                 break;
 205         case XFS_DINODE_FMT_EXTENTS:
 206         case XFS_DINODE_FMT_BTREE:
 207                 buf->bs_rdev = 0;
 208                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 209                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 210                 break;
 211         }
 212
 213         return 0;
 214 }
 215
 216 /*
 217  * Return stat information for one inode.
 218  * Return 0 if ok, else errno.
 219  */
 220 int                             /* error status */
 221 xfs_bulkstat_one(
 222         xfs_mount_t     *mp,            /* mount point for filesystem */
 223         xfs_ino_t       ino,            /* inode number to get data for */
 224         void            __user *buffer, /* buffer to place output in */
 225         int             ubsize,         /* size of buffer */
 226         void            *private_data,  /* my private data */
 227         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 228         int             *ubused,        /* bytes used by me */
 229         void            *dibuff,        /* on-disk inode buffer */
 230         int             *stat)          /* BULKSTAT_RV_... */
 231 {
 232         xfs_bstat_t     *buf;           /* return buffer */
 233         int             error = 0;      /* error value */
 234         xfs_dinode_t    *dip;           /* dinode inode pointer */
 235
 236         dip = (xfs_dinode_t *)dibuff;
 237
 238         if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
 239             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 240              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
 241                 *stat = BULKSTAT_RV_NOTHING;
 242                 return XFS_ERROR(EINVAL);
 243         }
 244         if (ubsize < sizeof(*buf)) {
 245                 *stat = BULKSTAT_RV_NOTHING;
 246                 return XFS_ERROR(ENOMEM);
 247         }
 248
 249         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 250
 251         if (dip == NULL) {
 252                 /* We're not being passed a pointer to a dinode.  This happens
 253                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 254                  */
 255                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 256                 if (error)
 257                         goto out_free;
 258         } else {
 259                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 260         }
 261
 262         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 263                 *stat = BULKSTAT_RV_NOTHING;
 264                 error =  EFAULT;
 265                 goto out_free;
 266         }
 267
 268         *stat = BULKSTAT_RV_DIDONE;
 269         if (ubused)
 270                 *ubused = sizeof(*buf);
 271
 272  out_free:
 273         kmem_free(buf, sizeof(*buf));
 274         return error;
 275 }
 276
 277 /*
 278  * Return stat information in bulk (by-inode) for the filesystem.
 279  */
 280 int                                     /* error status */
 281 xfs_bulkstat(
 282         xfs_mount_t             *mp,    /* mount point for filesystem */
 283         xfs_ino_t               *lastinop, /* last inode returned */
 284         int                     *ubcountp, /* size of buffer/count returned */
 285         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 286         void                    *private_data,/* private data for formatter */
 287         size_t                  statstruct_size, /* sizeof struct filling */
 288         char                    __user *ubuffer, /* buffer with inode stats */
 289         int                     flags,  /* defined in xfs_itable.h */
 290         int                     *done)  /* 1 if there're more stats to get */
 291 {
 292         xfs_agblock_t           agbno=0;/* allocation group block number */
 293         xfs_buf_t               *agbp;  /* agi header buffer */
 294         xfs_agi_t               *agi;   /* agi header data */
 295         xfs_agino_t             agino;  /* inode # in allocation group */
 296         xfs_agnumber_t          agno;   /* allocation group number */
 297         xfs_daddr_t             bno;    /* inode cluster start daddr */
 298         int                     chunkidx; /* current index into inode chunk */
 299         int                     clustidx; /* current index into inode cluster */
 300         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 301         int                     end_of_ag; /* set if we've seen the ag end */
 302         int                     error;  /* error code */
 303         int                     fmterror;/* bulkstat formatter result */
 304         __int32_t               gcnt;   /* current btree rec's count */
 305         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 306         xfs_agino_t             gino;   /* current btree rec's start inode */
 307         int                     i;      /* loop index */
 308         int                     icount; /* count of inodes good in irbuf */
 309         xfs_ino_t               ino;    /* inode number (filesystem) */
 310         xfs_inobt_rec_t         *irbp;  /* current irec buffer pointer */
 311         xfs_inobt_rec_t         *irbuf; /* start of irec buffer */
 312         xfs_inobt_rec_t         *irbufend; /* end of good irec buffer entries */
 313         xfs_ino_t               lastino=0; /* last inode number returned */
 314         int                     nbcluster; /* # of blocks in a cluster */
 315         int                     nicluster; /* # of inodes in a cluster */
 316         int                     nimask; /* mask for inode clusters */
 317         int                     nirbuf; /* size of irbuf */
 318         int                     rval;   /* return value error code */
 319         int                     tmp;    /* result value from btree calls */
 320         int                     ubcount; /* size of user's buffer */
 321         int                     ubleft; /* bytes left in user's buffer */
 322         char                    __user *ubufp;  /* pointer into user's buffer */
 323         int                     ubelem; /* spaces used in user's buffer */
 324         int                     ubused; /* bytes used by formatter */
 325         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 326         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 327         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 328
 329         /*
 330          * Get the last inode value, see if there's nothing to do.
 331          */
 332         ino = (xfs_ino_t)*lastinop;
 333         dip = NULL;
 334         agno = XFS_INO_TO_AGNO(mp, ino);
 335         agino = XFS_INO_TO_AGINO(mp, ino);
 336         if (agno >= mp->m_sb.sb_agcount ||
 337             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 338                 *done = 1;
 339                 *ubcountp = 0;
 340                 return 0;
 341         }
 342         ubcount = *ubcountp; /* statstruct's */
 343         ubleft = ubcount * statstruct_size; /* bytes */
 344         *ubcountp = ubelem = 0;
 345         *done = 0;
 346         fmterror = 0;
 347         ubufp = ubuffer;
 348         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 349                 mp->m_sb.sb_inopblock :
 350                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 351         nimask = ~(nicluster - 1);
 352         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 353         /*
 354          * Lock down the user's buffer. If a buffer was not sent, as in the case
 355          * disk quota code calls here, we skip this.
 356          */
 357         if (ubuffer &&
 358             (error = useracc(ubuffer, ubcount * statstruct_size,
 359                         (B_READ|B_PHYS), NULL))) {
 360                 return error;
 361         }
 362         /*
 363          * Allocate a page-sized buffer for inode btree records.
 364          * We could try allocating something smaller, but for normal
 365          * calls we'll always (potentially) need the whole page.
 366          */
 367         irbuf = kmem_alloc(NBPC, KM_SLEEP);
 368         nirbuf = NBPC / sizeof(*irbuf);
 369         /*
 370          * Loop over the allocation groups, starting from the last
 371          * inode returned; 0 means start of the allocation group.
 372          */
 373         rval = 0;
 374         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 375                 bp = NULL;
 376                 down_read(&mp->m_peraglock);
 377                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 378                 up_read(&mp->m_peraglock);
 379                 if (error) {
 380                         /*
 381                          * Skip this allocation group and go to the next one.
 382                          */
 383                         agno++;
 384                         agino = 0;
 385                         continue;
 386                 }
 387                 agi = XFS_BUF_TO_AGI(agbp);
 388                 /*
 389                  * Allocate and initialize a btree cursor for ialloc btree.
 390                  */
 391                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 392                         (xfs_inode_t *)0, 0);
 393                 irbp = irbuf;
 394                 irbufend = irbuf + nirbuf;
 395                 end_of_ag = 0;
 396                 /*
 397                  * If we're returning in the middle of an allocation group,
 398                  * we need to get the remainder of the chunk we're in.
 399                  */
 400                 if (agino > 0) {
 401                         /*
 402                          * Lookup the inode chunk that this inode lives in.
 403                          */
 404                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 405                         if (!error &&   /* no I/O error */
 406                             tmp &&      /* lookup succeeded */
 407                                         /* got the record, should always work */
 408                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 409                                     &gfree, &i)) &&
 410                             i == 1 &&
 411                                         /* this is the right chunk */
 412                             agino < gino + XFS_INODES_PER_CHUNK &&
 413                                         /* lastino was not last in chunk */
 414                             (chunkidx = agino - gino + 1) <
 415                                     XFS_INODES_PER_CHUNK &&
 416                                         /* there are some left allocated */
 417                             XFS_INOBT_MASKN(chunkidx,
 418                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 419                                 /*
 420                                  * Grab the chunk record.  Mark all the
 421                                  * uninteresting inodes (because they're
 422                                  * before our start point) free.
 423                                  */
 424                                 for (i = 0; i < chunkidx; i++) {
 425                                         if (XFS_INOBT_MASK(i) & ~gfree)
 426                                                 gcnt++;
 427                                 }
 428                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 429                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 430                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 431                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 432                                 irbp++;
 433                                 agino = gino + XFS_INODES_PER_CHUNK;
 434                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 435                         } else {
 436                                 /*
 437                                  * If any of those tests failed, bump the
 438                                  * inode number (just in case).
 439                                  */
 440                                 agino++;
 441                                 icount = 0;
 442                         }
 443                         /*
 444                          * In any case, increment to the next record.
 445                          */
 446                         if (!error)
 447                                 error = xfs_inobt_increment(cur, 0, &tmp);
 448                 } else {
 449                         /*
 450                          * Start of ag.  Lookup the first inode chunk.
 451                          */
 452                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 453                         icount = 0;
 454                 }
 455                 /*
 456                  * Loop through inode btree records in this ag,
 457                  * until we run out of inodes or space in the buffer.
 458                  */
 459                 while (irbp < irbufend && icount < ubcount) {
 460                         /*
 461                          * Loop as long as we're unable to read the
 462                          * inode btree.
 463                          */
 464                         while (error) {
 465                                 agino += XFS_INODES_PER_CHUNK;
 466                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 467                                                 INT_GET(agi->agi_length, ARCH_CONVERT))
 468                                         break;
 469                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 470                                                             &tmp);
 471                         }
 472                         /*
 473                          * If ran off the end of the ag either with an error,
 474                          * or the normal way, set end and stop collecting.
 475                          */
 476                         if (error ||
 477                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 478                                     &gfree, &i)) ||
 479                             i == 0) {
 480                                 end_of_ag = 1;
 481                                 break;
 482                         }
 483                         /*
 484                          * If this chunk has any allocated inodes, save it.
 485                          */
 486                         if (gcnt < XFS_INODES_PER_CHUNK) {
 487                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 488                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 489                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 490                                 irbp++;
 491                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 492                         }
 493                         /*
 494                          * Set agino to after this chunk and bump the cursor.
 495                          */
 496                         agino = gino + XFS_INODES_PER_CHUNK;
 497                         error = xfs_inobt_increment(cur, 0, &tmp);
 498                 }
 499                 /*
 500                  * Drop the btree buffers and the agi buffer.
 501                  * We can't hold any of the locks these represent
 502                  * when calling iget.
 503                  */
 504                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 505                 xfs_buf_relse(agbp);
 506                 /*
 507                  * Now format all the good inodes into the user's buffer.
 508                  */
 509                 irbufend = irbp;
 510                 for (irbp = irbuf;
 511                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 512                         /*
 513                          * Read-ahead the next chunk's worth of inodes.
 514                          */
 515                         if (&irbp[1] < irbufend) {
 516                                 /*
 517                                  * Loop over all clusters in the next chunk.
 518                                  * Do a readahead if there are any allocated
 519                                  * inodes in that cluster.
 520                                  */
 521                                 for (agbno = XFS_AGINO_TO_AGBNO(mp,
 522                                                         INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
 523                                      chunkidx = 0;
 524                                      chunkidx < XFS_INODES_PER_CHUNK;
 525                                      chunkidx += nicluster,
 526                                      agbno += nbcluster) {
 527                                         if (XFS_INOBT_MASKN(chunkidx,
 528                                                             nicluster) &
 529                                             ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
 530                                                 xfs_btree_reada_bufs(mp, agno,
 531                                                         agbno, nbcluster);
 532                                 }
 533                         }
 534                         /*
 535                          * Now process this chunk of inodes.
 536                          */
 537                         for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
 538                              ubleft > 0 &&
 539                                 INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
 540                              chunkidx++, clustidx++, agino++) {
 541                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 542                                 /*
 543                                  * Recompute agbno if this is the
 544                                  * first inode of the cluster.
 545                                  *
 546                                  * Careful with clustidx.   There can be
 547                                  * multple clusters per chunk, a single
 548                                  * cluster per chunk or a cluster that has
 549                                  * inodes represented from several different
 550                                  * chunks (if blocksize is large).
 551                                  *
 552                                  * Because of this, the starting clustidx is
 553                                  * initialized to zero in this loop but must
 554                                  * later be reset after reading in the cluster
 555                                  * buffer.
 556                                  */
 557                                 if ((chunkidx & (nicluster - 1)) == 0) {
 558                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 559                                                         INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
 560                                                 ((chunkidx & nimask) >>
 561                                                  mp->m_sb.sb_inopblog);
 562
 563                                         if (flags & BULKSTAT_FG_QUICK) {
 564                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 565                                                                        agino);
 566                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 567                                                                        agbno);
 568
 569                                                 /*
 570                                                  * Get the inode cluster buffer
 571                                                  */
 572                                                 ASSERT(xfs_inode_zone != NULL);
 573                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 574                                                                       KM_SLEEP);
 575                                                 ip->i_ino = ino;
 576                                                 ip->i_mount = mp;
 577                                                 if (bp)
 578                                                         xfs_buf_relse(bp);
 579                                                 error = xfs_itobp(mp, NULL, ip,
 580                                                                   &dip, &bp, bno);
 581                                                 if (!error)
 582                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 583                                                 kmem_zone_free(xfs_inode_zone, ip);
 584                                                 if (XFS_TEST_ERROR(error != 0,
 585                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 586                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 587                                                         bp = NULL;
 588                                                         break;
 589                                                 }
 590                                         }
 591                                 }
 592                                 /*
 593                                  * Skip if this inode is free.
 594                                  */
 595                                 if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
 596                                         continue;
 597                                 /*
 598                                  * Count used inodes as free so we can tell
 599                                  * when the chunk is used up.
 600                                  */
 601                                 INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
 602                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 603                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 604                                 if (flags & BULKSTAT_FG_QUICK) {
 605                                         dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 606                                               (clustidx << mp->m_sb.sb_inodelog));
 607
 608                                         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
 609                                                     != XFS_DINODE_MAGIC
 610                                             || !XFS_DINODE_GOOD_VERSION(
 611                                                     INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 612                                                 continue;
 613                                 }
 614
 615                                 /*
 616                                  * Get the inode and fill in a single buffer.
 617                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 618                                  * BULKSTAT_FG_IGET uses igets.
 619                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 620                                  * This is also used to count inodes/blks, etc
 621                                  * in xfs_qm_quotacheck.
 622                                  */
 623                                 ubused = statstruct_size;
 624                                 error = formatter(mp, ino, ubufp,
 625                                                 ubleft, private_data,
 626                                                 bno, &ubused, dip, &fmterror);
 627                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 628                                         if (error == ENOMEM)
 629                                                 ubleft = 0;
 630                                         continue;
 631                                 }
 632                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 633                                         ubleft = 0;
 634                                         ASSERT(error);
 635                                         rval = error;
 636                                         break;
 637                                 }
 638                                 if (ubufp)
 639                                         ubufp += ubused;
 640                                 ubleft -= ubused;
 641                                 ubelem++;
 642                                 lastino = ino;
 643                         }
 644                 }
 645
 646                 if (bp)
 647                         xfs_buf_relse(bp);
 648
 649                 /*
 650                  * Set up for the next loop iteration.
 651                  */
 652                 if (ubleft > 0) {
 653                         if (end_of_ag) {
 654                                 agno++;
 655                                 agino = 0;
 656                         } else
 657                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 658                 } else
 659                         break;
 660         }
 661         /*
 662          * Done, we're either out of filesystem or space to put the data.
 663          */
 664         kmem_free(irbuf, NBPC);
 665         if (ubuffer)
 666                 unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
 667         *ubcountp = ubelem;
 668         if (agno >= mp->m_sb.sb_agcount) {
 669                 /*
 670                  * If we ran out of filesystem, mark lastino as off
 671                  * the end of the filesystem, so the next call
 672                  * will return immediately.
 673                  */
 674                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 675                 *done = 1;
 676         } else
 677                 *lastinop = (xfs_ino_t)lastino;
 678
 679         return rval;
 680 }
 681
 682 /*
 683  * Return stat information in bulk (by-inode) for the filesystem.
 684  * Special case for non-sequential one inode bulkstat.
 685  */
 686 int                                     /* error status */
 687 xfs_bulkstat_single(
 688         xfs_mount_t             *mp,    /* mount point for filesystem */
 689         xfs_ino_t               *lastinop, /* inode to return */
 690         char                    __user *buffer, /* buffer with inode stats */
 691         int                     *done)  /* 1 if there're more stats to get */
 692 {
 693         int                     count;  /* count value for bulkstat call */
 694         int                     error;  /* return value */
 695         xfs_ino_t               ino;    /* filesystem inode number */
 696         int                     res;    /* result from bs1 */
 697
 698         /*
 699          * note that requesting valid inode numbers which are not allocated
 700          * to inodes will most likely cause xfs_itobp to generate warning
 701          * messages about bad magic numbers. This is ok. The fact that
 702          * the inode isn't actually an inode is handled by the
 703          * error check below. Done this way to make the usual case faster
 704          * at the expense of the error case.
 705          */
 706
 707         ino = (xfs_ino_t)*lastinop;
 708         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 709                                  NULL, 0, NULL, NULL, &res);
 710         if (error) {
 711                 /*
 712                  * Special case way failed, do it the "long" way
 713                  * to see if that works.
 714                  */
 715                 (*lastinop)--;
 716                 count = 1;
 717                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 718                                 NULL, sizeof(xfs_bstat_t), buffer,
 719                                 BULKSTAT_FG_IGET, done))
 720                         return error;
 721                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 722                         return error == EFSCORRUPTED ?
 723                                 XFS_ERROR(EINVAL) : error;
 724                 else
 725                         return 0;
 726         }
 727         *done = 0;
 728         return 0;
 729 }
 730
 731 /*
 732  * Return inode number table for the filesystem.
 733  */
 734 int                                     /* error status */
 735 xfs_inumbers(
 736         xfs_mount_t     *mp,            /* mount point for filesystem */
 737         xfs_ino_t       *lastino,       /* last inode returned */
 738         int             *count,         /* size of buffer/count returned */
 739         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 740 {
 741         xfs_buf_t       *agbp;
 742         xfs_agino_t     agino;
 743         xfs_agnumber_t  agno;
 744         int             bcount;
 745         xfs_inogrp_t    *buffer;
 746         int             bufidx;
 747         xfs_btree_cur_t *cur;
 748         int             error;
 749         __int32_t       gcnt;
 750         xfs_inofree_t   gfree;
 751         xfs_agino_t     gino;
 752         int             i;
 753         xfs_ino_t       ino;
 754         int             left;
 755         int             tmp;
 756
 757         ino = (xfs_ino_t)*lastino;
 758         agno = XFS_INO_TO_AGNO(mp, ino);
 759         agino = XFS_INO_TO_AGINO(mp, ino);
 760         left = *count;
 761         *count = 0;
 762         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 763         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 764         error = bufidx = 0;
 765         cur = NULL;
 766         agbp = NULL;
 767         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 768                 if (agbp == NULL) {
 769                         down_read(&mp->m_peraglock);
 770                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 771                         up_read(&mp->m_peraglock);
 772                         if (error) {
 773                                 /*
 774                                  * If we can't read the AGI of this ag,
 775                                  * then just skip to the next one.
 776                                  */
 777                                 ASSERT(cur == NULL);
 778                                 agbp = NULL;
 779                                 agno++;
 780                                 agino = 0;
 781                                 continue;
 782                         }
 783                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 784                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 785                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 786                         if (error) {
 787                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 788                                 cur = NULL;
 789                                 xfs_buf_relse(agbp);
 790                                 agbp = NULL;
 791                                 /*
 792                                  * Move up the the last inode in the current
 793                                  * chunk.  The lookup_ge will always get
 794                                  * us the first inode in the next chunk.
 795                                  */
 796                                 agino += XFS_INODES_PER_CHUNK - 1;
 797                                 continue;
 798                         }
 799                 }
 800                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 801                         &i)) ||
 802                     i == 0) {
 803                         xfs_buf_relse(agbp);
 804                         agbp = NULL;
 805                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 806                         cur = NULL;
 807                         agno++;
 808                         agino = 0;
 809                         continue;
 810                 }
 811                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 812                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 813                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 814                 buffer[bufidx].xi_allocmask = ~gfree;
 815                 bufidx++;
 816                 left--;
 817                 if (bufidx == bcount) {
 818                         if (copy_to_user(ubuffer, buffer,
 819                                         bufidx * sizeof(*buffer))) {
 820                                 error = XFS_ERROR(EFAULT);
 821                                 break;
 822                         }
 823                         ubuffer += bufidx;
 824                         *count += bufidx;
 825                         bufidx = 0;
 826                 }
 827                 if (left) {
 828                         error = xfs_inobt_increment(cur, 0, &tmp);
 829                         if (error) {
 830                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 831                                 cur = NULL;
 832                                 xfs_buf_relse(agbp);
 833                                 agbp = NULL;
 834                                 /*
 835                                  * The agino value has already been bumped.
 836                                  * Just try to skip up to it.
 837                                  */
 838                                 agino += XFS_INODES_PER_CHUNK;
 839                                 continue;
 840                         }
 841                 }
 842         }
 843         if (!error) {
 844                 if (bufidx) {
 845                         if (copy_to_user(ubuffer, buffer,
 846                                         bufidx * sizeof(*buffer)))
 847                                 error = XFS_ERROR(EFAULT);
 848                         else
 849                                 *count += bufidx;
 850                 }
 851                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 852         }
 853         kmem_free(buffer, bcount * sizeof(*buffer));
 854         if (cur)
 855                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 856                                            XFS_BTREE_NOERROR));
 857         if (agbp)
 858                 xfs_buf_relse(agbp);
 859         return error;
 860 }