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_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir_sf.h"
  35 #include "xfs_dir2_sf.h"
  36 #include "xfs_attr_sf.h"
  37 #include "xfs_dinode.h"
  38 #include "xfs_inode.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_itable.h"
  41 #include "xfs_error.h"
  42 #include "xfs_btree.h"
  43
  44 #ifndef HAVE_USERACC
  45 #define useracc(ubuffer, size, flags, foo) (0)
  46 #define unuseracc(ubuffer, size, flags)
  47 #endif
  48
  49 STATIC int
  50 xfs_bulkstat_one_iget(
  51         xfs_mount_t     *mp,            /* mount point for filesystem */
  52         xfs_ino_t       ino,            /* inode number to get data for */
  53         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  54         xfs_bstat_t     *buf,           /* return buffer */
  55         int             *stat)          /* BULKSTAT_RV_... */
  56 {
  57         xfs_dinode_core_t *dic;         /* dinode core info pointer */
  58         xfs_inode_t     *ip;            /* incore inode pointer */
  59         vnode_t         *vp;
  60         int             error;
  61
  62         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
  63         if (error) {
  64                 *stat = BULKSTAT_RV_NOTHING;
  65                 return error;
  66         }
  67
  68         ASSERT(ip != NULL);
  69         ASSERT(ip->i_blkno != (xfs_daddr_t)0);
  70         if (ip->i_d.di_mode == 0) {
  71                 *stat = BULKSTAT_RV_NOTHING;
  72                 error = XFS_ERROR(ENOENT);
  73                 goto out_iput;
  74         }
  75
  76         vp = XFS_ITOV(ip);
  77         dic = &ip->i_d;
  78
  79         /* xfs_iget returns the following without needing
  80          * further change.
  81          */
  82         buf->bs_nlink = dic->di_nlink;
  83         buf->bs_projid = dic->di_projid;
  84         buf->bs_ino = ino;
  85         buf->bs_mode = dic->di_mode;
  86         buf->bs_uid = dic->di_uid;
  87         buf->bs_gid = dic->di_gid;
  88         buf->bs_xid = dic->di_xid;
  89         buf->bs_size = dic->di_size;
  90         vn_atime_to_bstime(vp, &buf->bs_atime);
  91         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
  92         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
  93         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
  94         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
  95         buf->bs_xflags = xfs_ip2xflags(ip);
  96         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
  97         buf->bs_extents = dic->di_nextents;
  98         buf->bs_gen = dic->di_gen;
  99         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 100         buf->bs_dmevmask = dic->di_dmevmask;
 101         buf->bs_dmstate = dic->di_dmstate;
 102         buf->bs_aextents = dic->di_anextents;
 103
 104         switch (dic->di_format) {
 105         case XFS_DINODE_FMT_DEV:
 106                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 107                 buf->bs_blksize = BLKDEV_IOSIZE;
 108                 buf->bs_blocks = 0;
 109                 break;
 110         case XFS_DINODE_FMT_LOCAL:
 111         case XFS_DINODE_FMT_UUID:
 112                 buf->bs_rdev = 0;
 113                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 114                 buf->bs_blocks = 0;
 115                 break;
 116         case XFS_DINODE_FMT_EXTENTS:
 117         case XFS_DINODE_FMT_BTREE:
 118                 buf->bs_rdev = 0;
 119                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 120                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 121                 break;
 122         }
 123
 124  out_iput:
 125         xfs_iput(ip, XFS_ILOCK_SHARED);
 126         return error;
 127 }
 128
 129 STATIC int
 130 xfs_bulkstat_one_dinode(
 131         xfs_mount_t     *mp,            /* mount point for filesystem */
 132         xfs_ino_t       ino,            /* inode number to get data for */
 133         xfs_dinode_t    *dip,           /* dinode inode pointer */
 134         xfs_bstat_t     *buf)           /* return buffer */
 135 {
 136         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 137
 138         dic = &dip->di_core;
 139
 140         /*
 141          * The inode format changed when we moved the link count and
 142          * made it 32 bits long.  If this is an old format inode,
 143          * convert it in memory to look like a new one.  If it gets
 144          * flushed to disk we will convert back before flushing or
 145          * logging it.  We zero out the new projid field and the old link
 146          * count field.  We'll handle clearing the pad field (the remains
 147          * of the old uuid field) when we actually convert the inode to
 148          * the new format. We don't change the version number so that we
 149          * can distinguish this from a real new format inode.
 150          */
 151         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 152                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 153                 buf->bs_projid = 0;
 154         } else {
 155                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 156                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 157         }
 158
 159         buf->bs_ino = ino;
 160         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 161         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 162         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 163         buf->bs_xid = INT_GET(dic->di_xid, ARCH_CONVERT);
 164         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 165         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 166         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 167         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 168         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 169         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 170         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 171         buf->bs_xflags = xfs_dic2xflags(dic);
 172         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 173         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 174         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 175         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 176         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 177         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 178         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 179
 180         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 181         case XFS_DINODE_FMT_DEV:
 182                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 183                 buf->bs_blksize = BLKDEV_IOSIZE;
 184                 buf->bs_blocks = 0;
 185                 break;
 186         case XFS_DINODE_FMT_LOCAL:
 187         case XFS_DINODE_FMT_UUID:
 188                 buf->bs_rdev = 0;
 189                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 190                 buf->bs_blocks = 0;
 191                 break;
 192         case XFS_DINODE_FMT_EXTENTS:
 193         case XFS_DINODE_FMT_BTREE:
 194                 buf->bs_rdev = 0;
 195                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 196                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 197                 break;
 198         }
 199
 200         return 0;
 201 }
 202
 203 /*
 204  * Return stat information for one inode.
 205  * Return 0 if ok, else errno.
 206  */
 207 int                             /* error status */
 208 xfs_bulkstat_one(
 209         xfs_mount_t     *mp,            /* mount point for filesystem */
 210         xfs_ino_t       ino,            /* inode number to get data for */
 211         void            __user *buffer, /* buffer to place output in */
 212         int             ubsize,         /* size of buffer */
 213         void            *private_data,  /* my private data */
 214         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 215         int             *ubused,        /* bytes used by me */
 216         void            *dibuff,        /* on-disk inode buffer */
 217         int             *stat)          /* BULKSTAT_RV_... */
 218 {
 219         xfs_bstat_t     *buf;           /* return buffer */
 220         int             error = 0;      /* error value */
 221         xfs_dinode_t    *dip;           /* dinode inode pointer */
 222
 223         dip = (xfs_dinode_t *)dibuff;
 224
 225         if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
 226             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 227              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
 228                 *stat = BULKSTAT_RV_NOTHING;
 229                 return XFS_ERROR(EINVAL);
 230         }
 231         if (ubsize < sizeof(*buf)) {
 232                 *stat = BULKSTAT_RV_NOTHING;
 233                 return XFS_ERROR(ENOMEM);
 234         }
 235
 236         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 237
 238         if (dip == NULL) {
 239                 /* We're not being passed a pointer to a dinode.  This happens
 240                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 241                  */
 242                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 243                 if (error)
 244                         goto out_free;
 245         } else {
 246                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 247         }
 248
 249         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 250                 *stat = BULKSTAT_RV_NOTHING;
 251                 error =  EFAULT;
 252                 goto out_free;
 253         }
 254
 255         *stat = BULKSTAT_RV_DIDONE;
 256         if (ubused)
 257                 *ubused = sizeof(*buf);
 258
 259  out_free:
 260         kmem_free(buf, sizeof(*buf));
 261         return error;
 262 }
 263
 264 /*
 265  * Return stat information in bulk (by-inode) for the filesystem.
 266  */
 267 int                                     /* error status */
 268 xfs_bulkstat(
 269         xfs_mount_t             *mp,    /* mount point for filesystem */
 270         xfs_ino_t               *lastinop, /* last inode returned */
 271         int                     *ubcountp, /* size of buffer/count returned */
 272         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 273         void                    *private_data,/* private data for formatter */
 274         size_t                  statstruct_size, /* sizeof struct filling */
 275         char                    __user *ubuffer, /* buffer with inode stats */
 276         int                     flags,  /* defined in xfs_itable.h */
 277         int                     *done)  /* 1 if there are more stats to get */
 278 {
 279         xfs_agblock_t           agbno=0;/* allocation group block number */
 280         xfs_buf_t               *agbp;  /* agi header buffer */
 281         xfs_agi_t               *agi;   /* agi header data */
 282         xfs_agino_t             agino;  /* inode # in allocation group */
 283         xfs_agnumber_t          agno;   /* allocation group number */
 284         xfs_daddr_t             bno;    /* inode cluster start daddr */
 285         int                     chunkidx; /* current index into inode chunk */
 286         int                     clustidx; /* current index into inode cluster */
 287         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 288         int                     end_of_ag; /* set if we've seen the ag end */
 289         int                     error;  /* error code */
 290         int                     fmterror;/* bulkstat formatter result */
 291         __int32_t               gcnt;   /* current btree rec's count */
 292         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 293         xfs_agino_t             gino;   /* current btree rec's start inode */
 294         int                     i;      /* loop index */
 295         int                     icount; /* count of inodes good in irbuf */
 296         xfs_ino_t               ino;    /* inode number (filesystem) */
 297         xfs_inobt_rec_t         *irbp;  /* current irec buffer pointer */
 298         xfs_inobt_rec_t         *irbuf; /* start of irec buffer */
 299         xfs_inobt_rec_t         *irbufend; /* end of good irec buffer entries */
 300         xfs_ino_t               lastino=0; /* last inode number returned */
 301         int                     nbcluster; /* # of blocks in a cluster */
 302         int                     nicluster; /* # of inodes in a cluster */
 303         int                     nimask; /* mask for inode clusters */
 304         int                     nirbuf; /* size of irbuf */
 305         int                     rval;   /* return value error code */
 306         int                     tmp;    /* result value from btree calls */
 307         int                     ubcount; /* size of user's buffer */
 308         int                     ubleft; /* bytes left in user's buffer */
 309         char                    __user *ubufp;  /* pointer into user's buffer */
 310         int                     ubelem; /* spaces used in user's buffer */
 311         int                     ubused; /* bytes used by formatter */
 312         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 313         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 314         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 315
 316         /*
 317          * Get the last inode value, see if there's nothing to do.
 318          */
 319         ino = (xfs_ino_t)*lastinop;
 320         dip = NULL;
 321         agno = XFS_INO_TO_AGNO(mp, ino);
 322         agino = XFS_INO_TO_AGINO(mp, ino);
 323         if (agno >= mp->m_sb.sb_agcount ||
 324             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 325                 *done = 1;
 326                 *ubcountp = 0;
 327                 return 0;
 328         }
 329         ubcount = *ubcountp; /* statstruct's */
 330         ubleft = ubcount * statstruct_size; /* bytes */
 331         *ubcountp = ubelem = 0;
 332         *done = 0;
 333         fmterror = 0;
 334         ubufp = ubuffer;
 335         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 336                 mp->m_sb.sb_inopblock :
 337                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 338         nimask = ~(nicluster - 1);
 339         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 340         /*
 341          * Lock down the user's buffer. If a buffer was not sent, as in the case
 342          * disk quota code calls here, we skip this.
 343          */
 344         if (ubuffer &&
 345             (error = useracc(ubuffer, ubcount * statstruct_size,
 346                         (B_READ|B_PHYS), NULL))) {
 347                 return error;
 348         }
 349         /*
 350          * Allocate a page-sized buffer for inode btree records.
 351          * We could try allocating something smaller, but for normal
 352          * calls we'll always (potentially) need the whole page.
 353          */
 354         irbuf = kmem_alloc(NBPC, KM_SLEEP);
 355         nirbuf = NBPC / sizeof(*irbuf);
 356         /*
 357          * Loop over the allocation groups, starting from the last
 358          * inode returned; 0 means start of the allocation group.
 359          */
 360         rval = 0;
 361         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 362                 bp = NULL;
 363                 down_read(&mp->m_peraglock);
 364                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 365                 up_read(&mp->m_peraglock);
 366                 if (error) {
 367                         /*
 368                          * Skip this allocation group and go to the next one.
 369                          */
 370                         agno++;
 371                         agino = 0;
 372                         continue;
 373                 }
 374                 agi = XFS_BUF_TO_AGI(agbp);
 375                 /*
 376                  * Allocate and initialize a btree cursor for ialloc btree.
 377                  */
 378                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 379                         (xfs_inode_t *)0, 0);
 380                 irbp = irbuf;
 381                 irbufend = irbuf + nirbuf;
 382                 end_of_ag = 0;
 383                 /*
 384                  * If we're returning in the middle of an allocation group,
 385                  * we need to get the remainder of the chunk we're in.
 386                  */
 387                 if (agino > 0) {
 388                         /*
 389                          * Lookup the inode chunk that this inode lives in.
 390                          */
 391                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 392                         if (!error &&   /* no I/O error */
 393                             tmp &&      /* lookup succeeded */
 394                                         /* got the record, should always work */
 395                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 396                                     &gfree, &i)) &&
 397                             i == 1 &&
 398                                         /* this is the right chunk */
 399                             agino < gino + XFS_INODES_PER_CHUNK &&
 400                                         /* lastino was not last in chunk */
 401                             (chunkidx = agino - gino + 1) <
 402                                     XFS_INODES_PER_CHUNK &&
 403                                         /* there are some left allocated */
 404                             XFS_INOBT_MASKN(chunkidx,
 405                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 406                                 /*
 407                                  * Grab the chunk record.  Mark all the
 408                                  * uninteresting inodes (because they're
 409                                  * before our start point) free.
 410                                  */
 411                                 for (i = 0; i < chunkidx; i++) {
 412                                         if (XFS_INOBT_MASK(i) & ~gfree)
 413                                                 gcnt++;
 414                                 }
 415                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 416                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 417                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 418                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 419                                 irbp++;
 420                                 agino = gino + XFS_INODES_PER_CHUNK;
 421                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 422                         } else {
 423                                 /*
 424                                  * If any of those tests failed, bump the
 425                                  * inode number (just in case).
 426                                  */
 427                                 agino++;
 428                                 icount = 0;
 429                         }
 430                         /*
 431                          * In any case, increment to the next record.
 432                          */
 433                         if (!error)
 434                                 error = xfs_inobt_increment(cur, 0, &tmp);
 435                 } else {
 436                         /*
 437                          * Start of ag.  Lookup the first inode chunk.
 438                          */
 439                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 440                         icount = 0;
 441                 }
 442                 /*
 443                  * Loop through inode btree records in this ag,
 444                  * until we run out of inodes or space in the buffer.
 445                  */
 446                 while (irbp < irbufend && icount < ubcount) {
 447                         /*
 448                          * Loop as long as we're unable to read the
 449                          * inode btree.
 450                          */
 451                         while (error) {
 452                                 agino += XFS_INODES_PER_CHUNK;
 453                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 454                                                 be32_to_cpu(agi->agi_length))
 455                                         break;
 456                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 457                                                             &tmp);
 458                         }
 459                         /*
 460                          * If ran off the end of the ag either with an error,
 461                          * or the normal way, set end and stop collecting.
 462                          */
 463                         if (error ||
 464                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 465                                     &gfree, &i)) ||
 466                             i == 0) {
 467                                 end_of_ag = 1;
 468                                 break;
 469                         }
 470                         /*
 471                          * If this chunk has any allocated inodes, save it.
 472                          */
 473                         if (gcnt < XFS_INODES_PER_CHUNK) {
 474                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 475                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 476                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 477                                 irbp++;
 478                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 479                         }
 480                         /*
 481                          * Set agino to after this chunk and bump the cursor.
 482                          */
 483                         agino = gino + XFS_INODES_PER_CHUNK;
 484                         error = xfs_inobt_increment(cur, 0, &tmp);
 485                 }
 486                 /*
 487                  * Drop the btree buffers and the agi buffer.
 488                  * We can't hold any of the locks these represent
 489                  * when calling iget.
 490                  */
 491                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 492                 xfs_buf_relse(agbp);
 493                 /*
 494                  * Now format all the good inodes into the user's buffer.
 495                  */
 496                 irbufend = irbp;
 497                 for (irbp = irbuf;
 498                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 499                         /*
 500                          * Read-ahead the next chunk's worth of inodes.
 501                          */
 502                         if (&irbp[1] < irbufend) {
 503                                 /*
 504                                  * Loop over all clusters in the next chunk.
 505                                  * Do a readahead if there are any allocated
 506                                  * inodes in that cluster.
 507                                  */
 508                                 for (agbno = XFS_AGINO_TO_AGBNO(mp,
 509                                                         INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
 510                                      chunkidx = 0;
 511                                      chunkidx < XFS_INODES_PER_CHUNK;
 512                                      chunkidx += nicluster,
 513                                      agbno += nbcluster) {
 514                                         if (XFS_INOBT_MASKN(chunkidx,
 515                                                             nicluster) &
 516                                             ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
 517                                                 xfs_btree_reada_bufs(mp, agno,
 518                                                         agbno, nbcluster);
 519                                 }
 520                         }
 521                         /*
 522                          * Now process this chunk of inodes.
 523                          */
 524                         for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
 525                              ubleft > 0 &&
 526                                 INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
 527                              chunkidx++, clustidx++, agino++) {
 528                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 529                                 /*
 530                                  * Recompute agbno if this is the
 531                                  * first inode of the cluster.
 532                                  *
 533                                  * Careful with clustidx.   There can be
 534                                  * multple clusters per chunk, a single
 535                                  * cluster per chunk or a cluster that has
 536                                  * inodes represented from several different
 537                                  * chunks (if blocksize is large).
 538                                  *
 539                                  * Because of this, the starting clustidx is
 540                                  * initialized to zero in this loop but must
 541                                  * later be reset after reading in the cluster
 542                                  * buffer.
 543                                  */
 544                                 if ((chunkidx & (nicluster - 1)) == 0) {
 545                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 546                                                         INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
 547                                                 ((chunkidx & nimask) >>
 548                                                  mp->m_sb.sb_inopblog);
 549
 550                                         if (flags & BULKSTAT_FG_QUICK) {
 551                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 552                                                                        agino);
 553                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 554                                                                        agbno);
 555
 556                                                 /*
 557                                                  * Get the inode cluster buffer
 558                                                  */
 559                                                 ASSERT(xfs_inode_zone != NULL);
 560                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 561                                                                       KM_SLEEP);
 562                                                 ip->i_ino = ino;
 563                                                 ip->i_mount = mp;
 564                                                 if (bp)
 565                                                         xfs_buf_relse(bp);
 566                                                 error = xfs_itobp(mp, NULL, ip,
 567                                                                 &dip, &bp, bno,
 568                                                                 XFS_IMAP_BULKSTAT);
 569                                                 if (!error)
 570                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 571                                                 kmem_zone_free(xfs_inode_zone, ip);
 572                                                 if (XFS_TEST_ERROR(error != 0,
 573                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 574                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 575                                                         bp = NULL;
 576                                                         ubleft = 0;
 577                                                         rval = error;
 578                                                         break;
 579                                                 }
 580                                         }
 581                                 }
 582                                 /*
 583                                  * Skip if this inode is free.
 584                                  */
 585                                 if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
 586                                         continue;
 587                                 /*
 588                                  * Count used inodes as free so we can tell
 589                                  * when the chunk is used up.
 590                                  */
 591                                 INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
 592                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 593                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 594                                 if (flags & BULKSTAT_FG_QUICK) {
 595                                         dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 596                                               (clustidx << mp->m_sb.sb_inodelog));
 597
 598                                         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
 599                                                     != XFS_DINODE_MAGIC
 600                                             || !XFS_DINODE_GOOD_VERSION(
 601                                                     INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 602                                                 continue;
 603                                 }
 604
 605                                 /*
 606                                  * Get the inode and fill in a single buffer.
 607                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 608                                  * BULKSTAT_FG_IGET uses igets.
 609                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 610                                  * This is also used to count inodes/blks, etc
 611                                  * in xfs_qm_quotacheck.
 612                                  */
 613                                 ubused = statstruct_size;
 614                                 error = formatter(mp, ino, ubufp,
 615                                                 ubleft, private_data,
 616                                                 bno, &ubused, dip, &fmterror);
 617                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 618                                         if (error == ENOMEM)
 619                                                 ubleft = 0;
 620                                         continue;
 621                                 }
 622                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 623                                         ubleft = 0;
 624                                         ASSERT(error);
 625                                         rval = error;
 626                                         break;
 627                                 }
 628                                 if (ubufp)
 629                                         ubufp += ubused;
 630                                 ubleft -= ubused;
 631                                 ubelem++;
 632                                 lastino = ino;
 633                         }
 634                 }
 635
 636                 if (bp)
 637                         xfs_buf_relse(bp);
 638
 639                 /*
 640                  * Set up for the next loop iteration.
 641                  */
 642                 if (ubleft > 0) {
 643                         if (end_of_ag) {
 644                                 agno++;
 645                                 agino = 0;
 646                         } else
 647                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 648                 } else
 649                         break;
 650         }
 651         /*
 652          * Done, we're either out of filesystem or space to put the data.
 653          */
 654         kmem_free(irbuf, NBPC);
 655         if (ubuffer)
 656                 unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
 657         *ubcountp = ubelem;
 658         if (agno >= mp->m_sb.sb_agcount) {
 659                 /*
 660                  * If we ran out of filesystem, mark lastino as off
 661                  * the end of the filesystem, so the next call
 662                  * will return immediately.
 663                  */
 664                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 665                 *done = 1;
 666         } else
 667                 *lastinop = (xfs_ino_t)lastino;
 668
 669         return rval;
 670 }
 671
 672 /*
 673  * Return stat information in bulk (by-inode) for the filesystem.
 674  * Special case for non-sequential one inode bulkstat.
 675  */
 676 int                                     /* error status */
 677 xfs_bulkstat_single(
 678         xfs_mount_t             *mp,    /* mount point for filesystem */
 679         xfs_ino_t               *lastinop, /* inode to return */
 680         char                    __user *buffer, /* buffer with inode stats */
 681         int                     *done)  /* 1 if there are more stats to get */
 682 {
 683         int                     count;  /* count value for bulkstat call */
 684         int                     error;  /* return value */
 685         xfs_ino_t               ino;    /* filesystem inode number */
 686         int                     res;    /* result from bs1 */
 687
 688         /*
 689          * note that requesting valid inode numbers which are not allocated
 690          * to inodes will most likely cause xfs_itobp to generate warning
 691          * messages about bad magic numbers. This is ok. The fact that
 692          * the inode isn't actually an inode is handled by the
 693          * error check below. Done this way to make the usual case faster
 694          * at the expense of the error case.
 695          */
 696
 697         ino = (xfs_ino_t)*lastinop;
 698         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 699                                  NULL, 0, NULL, NULL, &res);
 700         if (error) {
 701                 /*
 702                  * Special case way failed, do it the "long" way
 703                  * to see if that works.
 704                  */
 705                 (*lastinop)--;
 706                 count = 1;
 707                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 708                                 NULL, sizeof(xfs_bstat_t), buffer,
 709                                 BULKSTAT_FG_IGET, done))
 710                         return error;
 711                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 712                         return error == EFSCORRUPTED ?
 713                                 XFS_ERROR(EINVAL) : error;
 714                 else
 715                         return 0;
 716         }
 717         *done = 0;
 718         return 0;
 719 }
 720
 721 /*
 722  * Return inode number table for the filesystem.
 723  */
 724 int                                     /* error status */
 725 xfs_inumbers(
 726         xfs_mount_t     *mp,            /* mount point for filesystem */
 727         xfs_ino_t       *lastino,       /* last inode returned */
 728         int             *count,         /* size of buffer/count returned */
 729         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 730 {
 731         xfs_buf_t       *agbp;
 732         xfs_agino_t     agino;
 733         xfs_agnumber_t  agno;
 734         int             bcount;
 735         xfs_inogrp_t    *buffer;
 736         int             bufidx;
 737         xfs_btree_cur_t *cur;
 738         int             error;
 739         __int32_t       gcnt;
 740         xfs_inofree_t   gfree;
 741         xfs_agino_t     gino;
 742         int             i;
 743         xfs_ino_t       ino;
 744         int             left;
 745         int             tmp;
 746
 747         ino = (xfs_ino_t)*lastino;
 748         agno = XFS_INO_TO_AGNO(mp, ino);
 749         agino = XFS_INO_TO_AGINO(mp, ino);
 750         left = *count;
 751         *count = 0;
 752         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 753         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 754         error = bufidx = 0;
 755         cur = NULL;
 756         agbp = NULL;
 757         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 758                 if (agbp == NULL) {
 759                         down_read(&mp->m_peraglock);
 760                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 761                         up_read(&mp->m_peraglock);
 762                         if (error) {
 763                                 /*
 764                                  * If we can't read the AGI of this ag,
 765                                  * then just skip to the next one.
 766                                  */
 767                                 ASSERT(cur == NULL);
 768                                 agbp = NULL;
 769                                 agno++;
 770                                 agino = 0;
 771                                 continue;
 772                         }
 773                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 774                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 775                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 776                         if (error) {
 777                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 778                                 cur = NULL;
 779                                 xfs_buf_relse(agbp);
 780                                 agbp = NULL;
 781                                 /*
 782                                  * Move up the the last inode in the current
 783                                  * chunk.  The lookup_ge will always get
 784                                  * us the first inode in the next chunk.
 785                                  */
 786                                 agino += XFS_INODES_PER_CHUNK - 1;
 787                                 continue;
 788                         }
 789                 }
 790                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 791                         &i)) ||
 792                     i == 0) {
 793                         xfs_buf_relse(agbp);
 794                         agbp = NULL;
 795                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 796                         cur = NULL;
 797                         agno++;
 798                         agino = 0;
 799                         continue;
 800                 }
 801                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 802                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 803                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 804                 buffer[bufidx].xi_allocmask = ~gfree;
 805                 bufidx++;
 806                 left--;
 807                 if (bufidx == bcount) {
 808                         if (copy_to_user(ubuffer, buffer,
 809                                         bufidx * sizeof(*buffer))) {
 810                                 error = XFS_ERROR(EFAULT);
 811                                 break;
 812                         }
 813                         ubuffer += bufidx;
 814                         *count += bufidx;
 815                         bufidx = 0;
 816                 }
 817                 if (left) {
 818                         error = xfs_inobt_increment(cur, 0, &tmp);
 819                         if (error) {
 820                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 821                                 cur = NULL;
 822                                 xfs_buf_relse(agbp);
 823                                 agbp = NULL;
 824                                 /*
 825                                  * The agino value has already been bumped.
 826                                  * Just try to skip up to it.
 827                                  */
 828                                 agino += XFS_INODES_PER_CHUNK;
 829                                 continue;
 830                         }
 831                 }
 832         }
 833         if (!error) {
 834                 if (bufidx) {
 835                         if (copy_to_user(ubuffer, buffer,
 836                                         bufidx * sizeof(*buffer)))
 837                                 error = XFS_ERROR(EFAULT);
 838                         else
 839                                 *count += bufidx;
 840                 }
 841                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 842         }
 843         kmem_free(buffer, bcount * sizeof(*buffer));
 844         if (cur)
 845                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 846                                            XFS_BTREE_NOERROR));
 847         if (agbp)
 848                 xfs_buf_relse(agbp);
 849         return error;
 850 }