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 /*
  59  * Return stat information for one inode.
  60  * Return 0 if ok, else errno.
  61  */
  62 int                                     /* error status */
  63 xfs_bulkstat_one(
  64         xfs_mount_t     *mp,            /* mount point for filesystem */
  65         xfs_trans_t     *tp,            /* transaction pointer */
  66         xfs_ino_t       ino,            /* inode number to get data for */
  67         void            *buffer,        /* buffer to place output in */
  68         int             ubsize,         /* size of buffer */
  69         void            *private_data,  /* my private data */
  70         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  71         int             *ubused,        /* bytes used by me */
  72         void            *dibuff,        /* on-disk inode buffer */
  73         int             *stat)          /* BULKSTAT_RV_... */
  74 {
  75         xfs_bstat_t     *buf;           /* return buffer */
  76         int             error;          /* error value */
  77         xfs_dinode_t    *dip;           /* dinode inode pointer */
  78         xfs_dinode_core_t *dic;         /* dinode core info pointer */
  79         xfs_inode_t     *ip = NULL;     /* incore inode pointer */
  80         xfs_arch_t      arch;           /* these are set according to      */
  81         __uint16_t      di_flags;       /* temp */
  82
  83         buf = (xfs_bstat_t *)buffer;
  84         dip = (xfs_dinode_t *)dibuff;
  85
  86         if (! buf || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
  87             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
  88              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
  89                 *stat = BULKSTAT_RV_NOTHING;
  90                 return XFS_ERROR(EINVAL);
  91         }
  92         if (ubsize < sizeof(*buf)) {
  93                 *stat = BULKSTAT_RV_NOTHING;
  94                 return XFS_ERROR(ENOMEM);
  95         }
  96
  97         if (dip == NULL) {
  98                 /* We're not being passed a pointer to a dinode.  This happens
  99                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 100                  */
 101                 error = xfs_iget(mp, tp, ino, XFS_ILOCK_SHARED, &ip, bno);
 102                 if (error) {
 103                         *stat = BULKSTAT_RV_NOTHING;
 104                         return error;
 105                 }
 106                 ASSERT(ip != NULL);
 107                 ASSERT(ip->i_blkno != (xfs_daddr_t)0);
 108                 if (ip->i_d.di_mode == 0) {
 109                         xfs_iput_new(ip, XFS_ILOCK_SHARED);
 110                         *stat = BULKSTAT_RV_NOTHING;
 111                         return XFS_ERROR(ENOENT);
 112                 }
 113                 dic = &ip->i_d;
 114                 arch = ARCH_NOCONVERT;          /* in-core! */
 115                 ASSERT(dic != NULL);
 116
 117                 /* xfs_iget returns the following without needing
 118                  * further change.
 119                  */
 120                 buf->bs_nlink = dic->di_nlink;
 121                 buf->bs_projid = dic->di_projid;
 122
 123         } else {
 124                 dic = &dip->di_core;
 125                 ASSERT(dic != NULL);
 126
 127                 /* buffer dinode_core is in on-disk arch */
 128                 arch = ARCH_CONVERT;
 129
 130                 /*
 131                  * The inode format changed when we moved the link count and
 132                  * made it 32 bits long.  If this is an old format inode,
 133                  * convert it in memory to look like a new one.  If it gets
 134                  * flushed to disk we will convert back before flushing or
 135                  * logging it.  We zero out the new projid field and the old link
 136                  * count field.  We'll handle clearing the pad field (the remains
 137                  * of the old uuid field) when we actually convert the inode to
 138                  * the new format. We don't change the version number so that we
 139                  * can distinguish this from a real new format inode.
 140                  */
 141                 if (INT_GET(dic->di_version, arch) == XFS_DINODE_VERSION_1) {
 142                         buf->bs_nlink = INT_GET(dic->di_onlink, arch);
 143                         buf->bs_projid = 0;
 144                 }
 145                 else {
 146                         buf->bs_nlink = INT_GET(dic->di_nlink, arch);
 147                         buf->bs_projid = INT_GET(dic->di_projid, arch);
 148                 }
 149
 150         }
 151
 152         buf->bs_ino = ino;
 153         buf->bs_mode = INT_GET(dic->di_mode, arch);
 154         buf->bs_uid = INT_GET(dic->di_uid, arch);
 155         buf->bs_gid = INT_GET(dic->di_gid, arch);
 156         buf->bs_size = INT_GET(dic->di_size, arch);
 157         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, arch);
 158         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, arch);
 159         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, arch);
 160         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, arch);
 161         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, arch);
 162         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, arch);
 163         /*
 164          * convert di_flags to bs_xflags.
 165          */
 166         di_flags = INT_GET(dic->di_flags, arch);
 167
 168         buf->bs_xflags =
 169                 ((di_flags & XFS_DIFLAG_REALTIME) ?
 170                         XFS_XFLAG_REALTIME : 0) |
 171                 ((di_flags & XFS_DIFLAG_PREALLOC) ?
 172                         XFS_XFLAG_PREALLOC : 0) |
 173                 ((di_flags & XFS_DIFLAG_IMMUTABLE) ?
 174                         XFS_XFLAG_IMMUTABLE : 0) |
 175                 ((di_flags & XFS_DIFLAG_APPEND) ?
 176                         XFS_XFLAG_APPEND : 0) |
 177                 ((di_flags & XFS_DIFLAG_SYNC) ?
 178                         XFS_XFLAG_SYNC : 0) |
 179                 ((di_flags & XFS_DIFLAG_NOATIME) ?
 180                         XFS_XFLAG_NOATIME : 0) |
 181                 ((di_flags & XFS_DIFLAG_NODUMP) ?
 182                         XFS_XFLAG_NODUMP : 0) |
 183                 (XFS_CFORK_Q_ARCH(dic, arch) ?
 184                         XFS_XFLAG_HASATTR : 0);
 185
 186         buf->bs_extsize = INT_GET(dic->di_extsize, arch) << mp->m_sb.sb_blocklog;
 187         buf->bs_extents = INT_GET(dic->di_nextents, arch);
 188         buf->bs_gen = INT_GET(dic->di_gen, arch);
 189         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 190         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, arch);
 191         buf->bs_dmstate = INT_GET(dic->di_dmstate, arch);
 192         buf->bs_aextents = INT_GET(dic->di_anextents, arch);
 193
 194         switch (INT_GET(dic->di_format, arch)) {
 195         case XFS_DINODE_FMT_DEV:
 196                 if ( ip ) {
 197                         buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 198                 } else {
 199                         buf->bs_rdev = INT_GET(dip->di_u.di_dev, arch);
 200                 }
 201
 202                 buf->bs_blksize = BLKDEV_IOSIZE;
 203                 buf->bs_blocks = 0;
 204                 break;
 205         case XFS_DINODE_FMT_LOCAL:
 206         case XFS_DINODE_FMT_UUID:
 207                 buf->bs_rdev = 0;
 208                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 209                 buf->bs_blocks = 0;
 210                 break;
 211         case XFS_DINODE_FMT_EXTENTS:
 212         case XFS_DINODE_FMT_BTREE:
 213                 buf->bs_rdev = 0;
 214                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 215                 if ( ip ) {
 216                         buf->bs_blocks = INT_GET(dic->di_nblocks, arch) + ip->i_delayed_blks;
 217                 } else {
 218                         buf->bs_blocks = INT_GET(dic->di_nblocks, arch);
 219                 }
 220                 break;
 221         }
 222
 223         if (ip) {
 224                 xfs_iput(ip, XFS_ILOCK_SHARED);
 225         }
 226
 227         *stat = BULKSTAT_RV_DIDONE;
 228         if (ubused)
 229                 *ubused = sizeof(*buf);
 230         return 0;
 231 }
 232
 233 /*
 234  * Return stat information in bulk (by-inode) for the filesystem.
 235  */
 236 int                                     /* error status */
 237 xfs_bulkstat(
 238         xfs_mount_t             *mp,    /* mount point for filesystem */
 239         xfs_trans_t             *tp,    /* transaction pointer */
 240         xfs_ino_t               *lastinop, /* last inode returned */
 241         int                     *ubcountp, /* size of buffer/count returned */
 242         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 243         void                    *private_data,/* private data for formatter */
 244         size_t                  statstruct_size, /* sizeof struct filling */
 245         xfs_caddr_t             ubuffer, /* buffer with inode stats */
 246         int                     flags,  /* defined in xfs_itable.h */
 247         int                     *done)  /* 1 if there're more stats to get */
 248 {
 249         xfs_agblock_t           agbno=0;/* allocation group block number */
 250         xfs_buf_t               *agbp;  /* agi header buffer */
 251         xfs_agi_t               *agi;   /* agi header data */
 252         xfs_agino_t             agino;  /* inode # in allocation group */
 253         xfs_agnumber_t          agno;   /* allocation group number */
 254         xfs_daddr_t             bno;    /* inode cluster start daddr */
 255         int                     chunkidx; /* current index into inode chunk */
 256         int                     clustidx; /* current index into inode cluster */
 257         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 258         int                     end_of_ag; /* set if we've seen the ag end */
 259         int                     error;  /* error code */
 260         int                     fmterror;/* bulkstat formatter result */
 261         __int32_t               gcnt;   /* current btree rec's count */
 262         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 263         xfs_agino_t             gino;   /* current btree rec's start inode */
 264         int                     i;      /* loop index */
 265         int                     icount; /* count of inodes good in irbuf */
 266         xfs_ino_t               ino;    /* inode number (filesystem) */
 267         xfs_inobt_rec_t         *irbp;  /* current irec buffer pointer */
 268         xfs_inobt_rec_t         *irbuf; /* start of irec buffer */
 269         xfs_inobt_rec_t         *irbufend; /* end of good irec buffer entries */
 270         xfs_ino_t               lastino=0; /* last inode number returned */
 271         int                     nbcluster; /* # of blocks in a cluster */
 272         int                     nicluster; /* # of inodes in a cluster */
 273         int                     nimask; /* mask for inode clusters */
 274         int                     nirbuf; /* size of irbuf */
 275         int                     rval;   /* return value error code */
 276         int                     tmp;    /* result value from btree calls */
 277         int                     ubcount; /* size of user's buffer */
 278         int                     ubleft; /* bytes left in user's buffer */
 279         xfs_caddr_t             ubufp;  /* current pointer into user's buffer */
 280         int                     ubelem; /* spaces used in user's buffer */
 281         int                     ubused; /* bytes used by formatter */
 282         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 283         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 284         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 285
 286         /*
 287          * Get the last inode value, see if there's nothing to do.
 288          */
 289         ino = (xfs_ino_t)*lastinop;
 290         dip = NULL;
 291         agno = XFS_INO_TO_AGNO(mp, ino);
 292         agino = XFS_INO_TO_AGINO(mp, ino);
 293         if (agno >= mp->m_sb.sb_agcount ||
 294             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 295                 *done = 1;
 296                 *ubcountp = 0;
 297                 return 0;
 298         }
 299         ubcount = *ubcountp; /* statstruct's */
 300         ubleft = ubcount * statstruct_size; /* bytes */
 301         *ubcountp = ubelem = 0;
 302         *done = 0;
 303         fmterror = 0;
 304         ubufp = ubuffer;
 305         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 306                 mp->m_sb.sb_inopblock :
 307                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 308         nimask = ~(nicluster - 1);
 309         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 310         /*
 311          * Lock down the user's buffer. If a buffer was not sent, as in the case
 312          * disk quota code calls here, we skip this.
 313          */
 314 #if defined(HAVE_USERACC)
 315         if (ubuffer &&
 316             (error = useracc(ubuffer, ubcount * statstruct_size,
 317                         (B_READ|B_PHYS), NULL))) {
 318                 return error;
 319         }
 320 #endif
 321         /*
 322          * Allocate a page-sized buffer for inode btree records.
 323          * We could try allocating something smaller, but for normal
 324          * calls we'll always (potentially) need the whole page.
 325          */
 326         irbuf = kmem_alloc(NBPC, KM_SLEEP);
 327         nirbuf = NBPC / sizeof(*irbuf);
 328         /*
 329          * Loop over the allocation groups, starting from the last
 330          * inode returned; 0 means start of the allocation group.
 331          */
 332         rval = 0;
 333         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 334                 bp = NULL;
 335                 down_read(&mp->m_peraglock);
 336                 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
 337                 up_read(&mp->m_peraglock);
 338                 if (error) {
 339                         /*
 340                          * Skip this allocation group and go to the next one.
 341                          */
 342                         agno++;
 343                         agino = 0;
 344                         continue;
 345                 }
 346                 agi = XFS_BUF_TO_AGI(agbp);
 347                 /*
 348                  * Allocate and initialize a btree cursor for ialloc btree.
 349                  */
 350                 cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO,
 351                         (xfs_inode_t *)0, 0);
 352                 irbp = irbuf;
 353                 irbufend = irbuf + nirbuf;
 354                 end_of_ag = 0;
 355                 /*
 356                  * If we're returning in the middle of an allocation group,
 357                  * we need to get the remainder of the chunk we're in.
 358                  */
 359                 if (agino > 0) {
 360                         /*
 361                          * Lookup the inode chunk that this inode lives in.
 362                          */
 363                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 364                         if (!error &&   /* no I/O error */
 365                             tmp &&      /* lookup succeeded */
 366                                         /* got the record, should always work */
 367                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 368                                     &gfree, &i, ARCH_NOCONVERT)) &&
 369                             i == 1 &&
 370                                         /* this is the right chunk */
 371                             agino < gino + XFS_INODES_PER_CHUNK &&
 372                                         /* lastino was not last in chunk */
 373                             (chunkidx = agino - gino + 1) <
 374                                     XFS_INODES_PER_CHUNK &&
 375                                         /* there are some left allocated */
 376                             XFS_INOBT_MASKN(chunkidx,
 377                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 378                                 /*
 379                                  * Grab the chunk record.  Mark all the
 380                                  * uninteresting inodes (because they're
 381                                  * before our start point) free.
 382                                  */
 383                                 for (i = 0; i < chunkidx; i++) {
 384                                         if (XFS_INOBT_MASK(i) & ~gfree)
 385                                                 gcnt++;
 386                                 }
 387                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 388                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 389                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 390                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 391                                 irbp++;
 392                                 agino = gino + XFS_INODES_PER_CHUNK;
 393                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 394                         } else {
 395                                 /*
 396                                  * If any of those tests failed, bump the
 397                                  * inode number (just in case).
 398                                  */
 399                                 agino++;
 400                                 icount = 0;
 401                         }
 402                         /*
 403                          * In any case, increment to the next record.
 404                          */
 405                         if (!error)
 406                                 error = xfs_inobt_increment(cur, 0, &tmp);
 407                 } else {
 408                         /*
 409                          * Start of ag.  Lookup the first inode chunk.
 410                          */
 411                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 412                         icount = 0;
 413                 }
 414                 /*
 415                  * Loop through inode btree records in this ag,
 416                  * until we run out of inodes or space in the buffer.
 417                  */
 418                 while (irbp < irbufend && icount < ubcount) {
 419                         /*
 420                          * Loop as long as we're unable to read the
 421                          * inode btree.
 422                          */
 423                         while (error) {
 424                                 agino += XFS_INODES_PER_CHUNK;
 425                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 426                                                 INT_GET(agi->agi_length, ARCH_CONVERT))
 427                                         break;
 428                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 429                                                             &tmp);
 430                         }
 431                         /*
 432                          * If ran off the end of the ag either with an error,
 433                          * or the normal way, set end and stop collecting.
 434                          */
 435                         if (error ||
 436                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 437                                     &gfree, &i, ARCH_NOCONVERT)) ||
 438                             i == 0) {
 439                                 end_of_ag = 1;
 440                                 break;
 441                         }
 442                         /*
 443                          * If this chunk has any allocated inodes, save it.
 444                          */
 445                         if (gcnt < XFS_INODES_PER_CHUNK) {
 446                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 447                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 448                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 449                                 irbp++;
 450                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 451                         }
 452                         /*
 453                          * Set agino to after this chunk and bump the cursor.
 454                          */
 455                         agino = gino + XFS_INODES_PER_CHUNK;
 456                         error = xfs_inobt_increment(cur, 0, &tmp);
 457                 }
 458                 /*
 459                  * Drop the btree buffers and the agi buffer.
 460                  * We can't hold any of the locks these represent
 461                  * when calling iget.
 462                  */
 463                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 464                 xfs_trans_brelse(tp, agbp);
 465                 /*
 466                  * Now format all the good inodes into the user's buffer.
 467                  */
 468                 irbufend = irbp;
 469                 for (irbp = irbuf;
 470                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 471                         /*
 472                          * Read-ahead the next chunk's worth of inodes.
 473                          */
 474                         if (&irbp[1] < irbufend) {
 475                                 /*
 476                                  * Loop over all clusters in the next chunk.
 477                                  * Do a readahead if there are any allocated
 478                                  * inodes in that cluster.
 479                                  */
 480                                 for (agbno = XFS_AGINO_TO_AGBNO(mp,
 481                                                         INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
 482                                      chunkidx = 0;
 483                                      chunkidx < XFS_INODES_PER_CHUNK;
 484                                      chunkidx += nicluster,
 485                                      agbno += nbcluster) {
 486                                         if (XFS_INOBT_MASKN(chunkidx,
 487                                                             nicluster) &
 488                                             ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
 489                                                 xfs_btree_reada_bufs(mp, agno,
 490                                                         agbno, nbcluster);
 491                                 }
 492                         }
 493                         /*
 494                          * Now process this chunk of inodes.
 495                          */
 496                         for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
 497                              ubleft > 0 &&
 498                                 INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
 499                              chunkidx++, clustidx++, agino++) {
 500                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 501                                 /*
 502                                  * Recompute agbno if this is the
 503                                  * first inode of the cluster.
 504                                  *
 505                                  * Careful with clustidx.   There can be
 506                                  * multple clusters per chunk, a single
 507                                  * cluster per chunk or a cluster that has
 508                                  * inodes represented from several different
 509                                  * chunks (if blocksize is large).
 510                                  *
 511                                  * Because of this, the starting clustidx is
 512                                  * initialized to zero in this loop but must
 513                                  * later be reset after reading in the cluster
 514                                  * buffer.
 515                                  */
 516                                 if ((chunkidx & (nicluster - 1)) == 0) {
 517                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 518                                                         INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
 519                                                 ((chunkidx & nimask) >>
 520                                                  mp->m_sb.sb_inopblog);
 521
 522                                         if (flags & BULKSTAT_FG_QUICK) {
 523                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 524                                                                        agino);
 525                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 526                                                                        agbno);
 527
 528                                                 /*
 529                                                  * Get the inode cluster buffer
 530                                                  */
 531                                                 ASSERT(xfs_inode_zone != NULL);
 532                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 533                                                                       KM_SLEEP);
 534                                                 ip->i_ino = ino;
 535                                                 ip->i_mount = mp;
 536                                                 if (bp)
 537                                                         xfs_trans_brelse(tp, bp);
 538                                                 error = xfs_itobp(mp, tp, ip,
 539                                                                   &dip, &bp, bno);
 540                                                 if (!error)
 541                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 542                                                 kmem_zone_free(xfs_inode_zone, ip);
 543                                                 if (XFS_TEST_ERROR(error != 0,
 544                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 545                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 546                                                         bp = NULL;
 547                                                         break;
 548                                                 }
 549                                         }
 550                                 }
 551                                 /*
 552                                  * Skip if this inode is free.
 553                                  */
 554                                 if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
 555                                         continue;
 556                                 /*
 557                                  * Count used inodes as free so we can tell
 558                                  * when the chunk is used up.
 559                                  */
 560                                 INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
 561                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 562                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 563                                 if (flags & BULKSTAT_FG_QUICK) {
 564                                         dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 565                                               (clustidx << mp->m_sb.sb_inodelog));
 566
 567                                         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
 568                                                     != XFS_DINODE_MAGIC
 569                                             || !XFS_DINODE_GOOD_VERSION(
 570                                                     INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 571                                                 continue;
 572                                 }
 573
 574                                 /*
 575                                  * Get the inode and fill in a single buffer.
 576                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 577                                  * BULKSTAT_FG_IGET uses igets.
 578                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 579                                  * This is also used to count inodes/blks, etc
 580                                  * in xfs_qm_quotacheck.
 581                                  */
 582                                 ubused = statstruct_size;
 583                                 error = formatter(mp, tp, ino, ubufp,
 584                                                 ubleft, private_data,
 585                                                 bno, &ubused, dip, &fmterror);
 586                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 587                                         if (error == ENOMEM)
 588                                                 ubleft = 0;
 589                                         continue;
 590                                 }
 591                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 592                                         ubleft = 0;
 593                                         ASSERT(error);
 594                                         rval = error;
 595                                         break;
 596                                 }
 597                                 if (ubufp)
 598                                         ubufp += ubused;
 599                                 ubleft -= ubused;
 600                                 ubelem++;
 601                                 lastino = ino;
 602                         }
 603                 }
 604
 605                 if (bp)
 606                         xfs_trans_brelse(tp, bp);
 607
 608                 /*
 609                  * Set up for the next loop iteration.
 610                  */
 611                 if (ubleft > 0) {
 612                         if (end_of_ag) {
 613                                 agno++;
 614                                 agino = 0;
 615                         } else
 616                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 617                 } else
 618                         break;
 619         }
 620         /*
 621          * Done, we're either out of filesystem or space to put the data.
 622          */
 623         kmem_free(irbuf, NBPC);
 624 #if defined(HAVE_USERACC)
 625         if (ubuffer)
 626                 unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
 627 #endif
 628         *ubcountp = ubelem;
 629         if (agno >= mp->m_sb.sb_agcount) {
 630                 /*
 631                  * If we ran out of filesystem, mark lastino as off
 632                  * the end of the filesystem, so the next call
 633                  * will return immediately.
 634                  */
 635                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 636                 *done = 1;
 637         } else
 638                 *lastinop = (xfs_ino_t)lastino;
 639
 640         return rval;
 641 }
 642
 643 /*
 644  * Return stat information in bulk (by-inode) for the filesystem.
 645  * Special case for non-sequential one inode bulkstat.
 646  */
 647 int                                     /* error status */
 648 xfs_bulkstat_single(
 649         xfs_mount_t             *mp,    /* mount point for filesystem */
 650         xfs_ino_t               *lastinop, /* inode to return */
 651         xfs_caddr_t             buffer, /* buffer with inode stats */
 652         int                     *done)  /* 1 if there're more stats to get */
 653 {
 654         xfs_bstat_t             bstat;  /* one bulkstat result structure */
 655         int                     count;  /* count value for bulkstat call */
 656         int                     error;  /* return value */
 657         xfs_ino_t               ino;    /* filesystem inode number */
 658         int                     res;    /* result from bs1 */
 659
 660         /*
 661          * note that requesting valid inode numbers which are not allocated
 662          * to inodes will most likely cause xfs_itobp to generate warning
 663          * messages about bad magic numbers. This is ok. The fact that
 664          * the inode isn't actually an inode is handled by the
 665          * error check below. Done this way to make the usual case faster
 666          * at the expense of the error case.
 667          */
 668
 669         ino = (xfs_ino_t)*lastinop;
 670         error = xfs_bulkstat_one(mp, NULL, ino, &bstat, sizeof(bstat),
 671                                  NULL, 0, NULL, NULL, &res);
 672         if (error) {
 673                 /*
 674                  * Special case way failed, do it the "long" way
 675                  * to see if that works.
 676                  */
 677                 (*lastinop)--;
 678                 count = 1;
 679                 if (xfs_bulkstat(mp, NULL, lastinop, &count, xfs_bulkstat_one,
 680                                 NULL,
 681                                 sizeof(bstat), buffer, BULKSTAT_FG_IGET, done))
 682                         return error;
 683                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 684                         return error == EFSCORRUPTED ?
 685                                 XFS_ERROR(EINVAL) : error;
 686                 else
 687                         return 0;
 688         }
 689         *done = 0;
 690         if (copy_to_user(buffer, &bstat, sizeof(bstat)))
 691                 return XFS_ERROR(EFAULT);
 692         return 0;
 693 }
 694
 695 /*
 696  * Return inode number table for the filesystem.
 697  */
 698 int                                     /* error status */
 699 xfs_inumbers(
 700         xfs_mount_t     *mp,            /* mount point for filesystem */
 701         xfs_trans_t     *tp,            /* transaction pointer */
 702         xfs_ino_t       *lastino,       /* last inode returned */
 703         int             *count,         /* size of buffer/count returned */
 704         xfs_caddr_t     ubuffer)        /* buffer with inode descriptions */
 705 {
 706         xfs_buf_t       *agbp;
 707         xfs_agino_t     agino;
 708         xfs_agnumber_t  agno;
 709         int             bcount;
 710         xfs_inogrp_t    *buffer;
 711         int             bufidx;
 712         xfs_btree_cur_t *cur;
 713         int             error;
 714         __int32_t       gcnt;
 715         xfs_inofree_t   gfree;
 716         xfs_agino_t     gino;
 717         int             i;
 718         xfs_ino_t       ino;
 719         int             left;
 720         int             tmp;
 721
 722         ino = (xfs_ino_t)*lastino;
 723         agno = XFS_INO_TO_AGNO(mp, ino);
 724         agino = XFS_INO_TO_AGINO(mp, ino);
 725         left = *count;
 726         *count = 0;
 727         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 728         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 729         error = bufidx = 0;
 730         cur = NULL;
 731         agbp = NULL;
 732         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 733                 if (agbp == NULL) {
 734                         down_read(&mp->m_peraglock);
 735                         error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
 736                         up_read(&mp->m_peraglock);
 737                         if (error) {
 738                                 /*
 739                                  * If we can't read the AGI of this ag,
 740                                  * then just skip to the next one.
 741                                  */
 742                                 ASSERT(cur == NULL);
 743                                 agbp = NULL;
 744                                 agno++;
 745                                 agino = 0;
 746                                 continue;
 747                         }
 748                         cur = xfs_btree_init_cursor(mp, tp, agbp, agno,
 749                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 750                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 751                         if (error) {
 752                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 753                                 cur = NULL;
 754                                 xfs_trans_brelse(tp, agbp);
 755                                 agbp = NULL;
 756                                 /*
 757                                  * Move up the the last inode in the current
 758                                  * chunk.  The lookup_ge will always get
 759                                  * us the first inode in the next chunk.
 760                                  */
 761                                 agino += XFS_INODES_PER_CHUNK - 1;
 762                                 continue;
 763                         }
 764                 }
 765                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 766                         &i, ARCH_NOCONVERT)) ||
 767                     i == 0) {
 768                         xfs_trans_brelse(tp, agbp);
 769                         agbp = NULL;
 770                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 771                         cur = NULL;
 772                         agno++;
 773                         agino = 0;
 774                         continue;
 775                 }
 776                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 777                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 778                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 779                 buffer[bufidx].xi_allocmask = ~gfree;
 780                 bufidx++;
 781                 left--;
 782                 if (bufidx == bcount) {
 783                         if (copy_to_user(ubuffer, buffer,
 784                                         bufidx * sizeof(*buffer))) {
 785                                 error = XFS_ERROR(EFAULT);
 786                                 break;
 787                         }
 788                         ubuffer += bufidx * sizeof(*buffer);
 789                         *count += bufidx;
 790                         bufidx = 0;
 791                 }
 792                 if (left) {
 793                         error = xfs_inobt_increment(cur, 0, &tmp);
 794                         if (error) {
 795                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 796                                 cur = NULL;
 797                                 xfs_trans_brelse(tp, agbp);
 798                                 agbp = NULL;
 799                                 /*
 800                                  * The agino value has already been bumped.
 801                                  * Just try to skip up to it.
 802                                  */
 803                                 agino += XFS_INODES_PER_CHUNK;
 804                                 continue;
 805                         }
 806                 }
 807         }
 808         if (!error) {
 809                 if (bufidx) {
 810                         if (copy_to_user(ubuffer, buffer,
 811                                         bufidx * sizeof(*buffer)))
 812                                 error = XFS_ERROR(EFAULT);
 813                         else
 814                                 *count += bufidx;
 815                 }
 816                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 817         }
 818         kmem_free(buffer, bcount * sizeof(*buffer));
 819         if (cur)
 820                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 821                                            XFS_BTREE_NOERROR));
 822         if (agbp)
 823                 xfs_trans_brelse(tp, agbp);
 824         return error;
 825 }