git.oblomov.eu Git - linux-2.6/blob - 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_size = dic->di_size;
 101         buf->bs_atime.tv_sec = dic->di_atime.t_sec;
 102         buf->bs_atime.tv_nsec = dic->di_atime.t_nsec;
 103         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
 104         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
 105         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
 106         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
 107         buf->bs_xflags = xfs_ip2xflags(ip);
 108         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
 109         buf->bs_extents = dic->di_nextents;
 110         buf->bs_gen = dic->di_gen;
 111         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 112         buf->bs_dmevmask = dic->di_dmevmask;
 113         buf->bs_dmstate = dic->di_dmstate;
 114         buf->bs_aextents = dic->di_anextents;
 115
 116         switch (dic->di_format) {
 117         case XFS_DINODE_FMT_DEV:
 118                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
 119                 buf->bs_blksize = BLKDEV_IOSIZE;
 120                 buf->bs_blocks = 0;
 121                 break;
 122         case XFS_DINODE_FMT_LOCAL:
 123         case XFS_DINODE_FMT_UUID:
 124                 buf->bs_rdev = 0;
 125                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 126                 buf->bs_blocks = 0;
 127                 break;
 128         case XFS_DINODE_FMT_EXTENTS:
 129         case XFS_DINODE_FMT_BTREE:
 130                 buf->bs_rdev = 0;
 131                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 132                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 133                 break;
 134         }
 135
 136  out_iput:
 137         xfs_iput(ip, XFS_ILOCK_SHARED);
 138         return error;
 139 }
 140
 141 STATIC int
 142 xfs_bulkstat_one_dinode(
 143         xfs_mount_t     *mp,            /* mount point for filesystem */
 144         xfs_ino_t       ino,            /* inode number to get data for */
 145         xfs_dinode_t    *dip,           /* dinode inode pointer */
 146         xfs_bstat_t     *buf)           /* return buffer */
 147 {
 148         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 149
 150         dic = &dip->di_core;
 151
 152         /*
 153          * The inode format changed when we moved the link count and
 154          * made it 32 bits long.  If this is an old format inode,
 155          * convert it in memory to look like a new one.  If it gets
 156          * flushed to disk we will convert back before flushing or
 157          * logging it.  We zero out the new projid field and the old link
 158          * count field.  We'll handle clearing the pad field (the remains
 159          * of the old uuid field) when we actually convert the inode to
 160          * the new format. We don't change the version number so that we
 161          * can distinguish this from a real new format inode.
 162          */
 163         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 164                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 165                 buf->bs_projid = 0;
 166         } else {
 167                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 168                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 169         }
 170
 171         buf->bs_ino = ino;
 172         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 173         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 174         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 175         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 176         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 177         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 178         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 179         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 180         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 181         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 182         buf->bs_xflags = xfs_dic2xflags(dic);
 183         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 184         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 185         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 186         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 187         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 188         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 189         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 190
 191         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 192         case XFS_DINODE_FMT_DEV:
 193                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 194                 buf->bs_blksize = BLKDEV_IOSIZE;
 195                 buf->bs_blocks = 0;
 196                 break;
 197         case XFS_DINODE_FMT_LOCAL:
 198         case XFS_DINODE_FMT_UUID:
 199                 buf->bs_rdev = 0;
 200                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 201                 buf->bs_blocks = 0;
 202                 break;
 203         case XFS_DINODE_FMT_EXTENTS:
 204         case XFS_DINODE_FMT_BTREE:
 205                 buf->bs_rdev = 0;
 206                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 207                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 208                 break;
 209         }
 210
 211         return 0;
 212 }
 213
 214 /*
 215  * Return stat information for one inode.
 216  * Return 0 if ok, else errno.
 217  */
 218 int                             /* error status */
 219 xfs_bulkstat_one(
 220         xfs_mount_t     *mp,            /* mount point for filesystem */
 221         xfs_ino_t       ino,            /* inode number to get data for */
 222         void            __user *buffer, /* buffer to place output in */
 223         int             ubsize,         /* size of buffer */
 224         void            *private_data,  /* my private data */
 225         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 226         int             *ubused,        /* bytes used by me */
 227         void            *dibuff,        /* on-disk inode buffer */
 228         int             *stat)          /* BULKSTAT_RV_... */
 229 {
 230         xfs_bstat_t     *buf;           /* return buffer */
 231         int             error = 0;      /* error value */
 232         xfs_dinode_t    *dip;           /* dinode inode pointer */
 233
 234         dip = (xfs_dinode_t *)dibuff;
 235
 236         if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
 237             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 238              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
 239                 *stat = BULKSTAT_RV_NOTHING;
 240                 return XFS_ERROR(EINVAL);
 241         }
 242         if (ubsize < sizeof(*buf)) {
 243                 *stat = BULKSTAT_RV_NOTHING;
 244                 return XFS_ERROR(ENOMEM);
 245         }
 246
 247         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 248
 249         if (dip == NULL) {
 250                 /* We're not being passed a pointer to a dinode.  This happens
 251                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 252                  */
 253                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 254                 if (error)
 255                         goto out_free;
 256         } else {
 257                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 258         }
 259
 260         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 261                 *stat = BULKSTAT_RV_NOTHING;
 262                 error =  EFAULT;
 263                 goto out_free;
 264         }
 265
 266         *stat = BULKSTAT_RV_DIDONE;
 267         if (ubused)
 268                 *ubused = sizeof(*buf);
 269
 270  out_free:
 271         kmem_free(buf, sizeof(*buf));
 272         return error;
 273 }
 274
 275 /*
 276  * Return stat information in bulk (by-inode) for the filesystem.
 277  */
 278 int                                     /* error status */
 279 xfs_bulkstat(
 280         xfs_mount_t             *mp,    /* mount point for filesystem */
 281         xfs_ino_t               *lastinop, /* last inode returned */
 282         int                     *ubcountp, /* size of buffer/count returned */
 283         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 284         void                    *private_data,/* private data for formatter */
 285         size_t                  statstruct_size, /* sizeof struct filling */
 286         char                    __user *ubuffer, /* buffer with inode stats */
 287         int                     flags,  /* defined in xfs_itable.h */
 288         int                     *done)  /* 1 if there're more stats to get */
 289 {
 290         xfs_agblock_t           agbno=0;/* allocation group block number */
 291         xfs_buf_t               *agbp;  /* agi header buffer */
 292         xfs_agi_t               *agi;   /* agi header data */
 293         xfs_agino_t             agino;  /* inode # in allocation group */
 294         xfs_agnumber_t          agno;   /* allocation group number */
 295         xfs_daddr_t             bno;    /* inode cluster start daddr */
 296         int                     chunkidx; /* current index into inode chunk */
 297         int                     clustidx; /* current index into inode cluster */
 298         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 299         int                     end_of_ag; /* set if we've seen the ag end */
 300         int                     error;  /* error code */
 301         int                     fmterror;/* bulkstat formatter result */
 302         __int32_t               gcnt;   /* current btree rec's count */
 303         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 304         xfs_agino_t             gino;   /* current btree rec's start inode */
 305         int                     i;      /* loop index */
 306         int                     icount; /* count of inodes good in irbuf */
 307         xfs_ino_t               ino;    /* inode number (filesystem) */
 308         xfs_inobt_rec_t         *irbp;  /* current irec buffer pointer */
 309         xfs_inobt_rec_t         *irbuf; /* start of irec buffer */
 310         xfs_inobt_rec_t         *irbufend; /* end of good irec buffer entries */
 311         xfs_ino_t               lastino=0; /* last inode number returned */
 312         int                     nbcluster; /* # of blocks in a cluster */
 313         int                     nicluster; /* # of inodes in a cluster */
 314         int                     nimask; /* mask for inode clusters */
 315         int                     nirbuf; /* size of irbuf */
 316         int                     rval;   /* return value error code */
 317         int                     tmp;    /* result value from btree calls */
 318         int                     ubcount; /* size of user's buffer */
 319         int                     ubleft; /* bytes left in user's buffer */
 320         char                    __user *ubufp;  /* pointer into user's buffer */
 321         int                     ubelem; /* spaces used in user's buffer */
 322         int                     ubused; /* bytes used by formatter */
 323         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 324         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 325         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 326
 327         /*
 328          * Get the last inode value, see if there's nothing to do.
 329          */
 330         ino = (xfs_ino_t)*lastinop;
 331         dip = NULL;
 332         agno = XFS_INO_TO_AGNO(mp, ino);
 333         agino = XFS_INO_TO_AGINO(mp, ino);
 334         if (agno >= mp->m_sb.sb_agcount ||
 335             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 336                 *done = 1;
 337                 *ubcountp = 0;
 338                 return 0;
 339         }
 340         ubcount = *ubcountp; /* statstruct's */
 341         ubleft = ubcount * statstruct_size; /* bytes */
 342         *ubcountp = ubelem = 0;
 343         *done = 0;
 344         fmterror = 0;
 345         ubufp = ubuffer;
 346         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 347                 mp->m_sb.sb_inopblock :
 348                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 349         nimask = ~(nicluster - 1);
 350         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 351         /*
 352          * Lock down the user's buffer. If a buffer was not sent, as in the case
 353          * disk quota code calls here, we skip this.
 354          */
 355         if (ubuffer &&
 356             (error = useracc(ubuffer, ubcount * statstruct_size,
 357                         (B_READ|B_PHYS), NULL))) {
 358                 return error;
 359         }
 360         /*
 361          * Allocate a page-sized buffer for inode btree records.
 362          * We could try allocating something smaller, but for normal
 363          * calls we'll always (potentially) need the whole page.
 364          */
 365         irbuf = kmem_alloc(NBPC, KM_SLEEP);
 366         nirbuf = NBPC / sizeof(*irbuf);
 367         /*
 368          * Loop over the allocation groups, starting from the last
 369          * inode returned; 0 means start of the allocation group.
 370          */
 371         rval = 0;
 372         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 373                 bp = NULL;
 374                 down_read(&mp->m_peraglock);
 375                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 376                 up_read(&mp->m_peraglock);
 377                 if (error) {
 378                         /*
 379                          * Skip this allocation group and go to the next one.
 380                          */
 381                         agno++;
 382                         agino = 0;
 383                         continue;
 384                 }
 385                 agi = XFS_BUF_TO_AGI(agbp);
 386                 /*
 387                  * Allocate and initialize a btree cursor for ialloc btree.
 388                  */
 389                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 390                         (xfs_inode_t *)0, 0);
 391                 irbp = irbuf;
 392                 irbufend = irbuf + nirbuf;
 393                 end_of_ag = 0;
 394                 /*
 395                  * If we're returning in the middle of an allocation group,
 396                  * we need to get the remainder of the chunk we're in.
 397                  */
 398                 if (agino > 0) {
 399                         /*
 400                          * Lookup the inode chunk that this inode lives in.
 401                          */
 402                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 403                         if (!error &&   /* no I/O error */
 404                             tmp &&      /* lookup succeeded */
 405                                         /* got the record, should always work */
 406                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 407                                     &gfree, &i)) &&
 408                             i == 1 &&
 409                                         /* this is the right chunk */
 410                             agino < gino + XFS_INODES_PER_CHUNK &&
 411                                         /* lastino was not last in chunk */
 412                             (chunkidx = agino - gino + 1) <
 413                                     XFS_INODES_PER_CHUNK &&
 414                                         /* there are some left allocated */
 415                             XFS_INOBT_MASKN(chunkidx,
 416                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 417                                 /*
 418                                  * Grab the chunk record.  Mark all the
 419                                  * uninteresting inodes (because they're
 420                                  * before our start point) free.
 421                                  */
 422                                 for (i = 0; i < chunkidx; i++) {
 423                                         if (XFS_INOBT_MASK(i) & ~gfree)
 424                                                 gcnt++;
 425                                 }
 426                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 427                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 428                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 429                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 430                                 irbp++;
 431                                 agino = gino + XFS_INODES_PER_CHUNK;
 432                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 433                         } else {
 434                                 /*
 435                                  * If any of those tests failed, bump the
 436                                  * inode number (just in case).
 437                                  */
 438                                 agino++;
 439                                 icount = 0;
 440                         }
 441                         /*
 442                          * In any case, increment to the next record.
 443                          */
 444                         if (!error)
 445                                 error = xfs_inobt_increment(cur, 0, &tmp);
 446                 } else {
 447                         /*
 448                          * Start of ag.  Lookup the first inode chunk.
 449                          */
 450                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 451                         icount = 0;
 452                 }
 453                 /*
 454                  * Loop through inode btree records in this ag,
 455                  * until we run out of inodes or space in the buffer.
 456                  */
 457                 while (irbp < irbufend && icount < ubcount) {
 458                         /*
 459                          * Loop as long as we're unable to read the
 460                          * inode btree.
 461                          */
 462                         while (error) {
 463                                 agino += XFS_INODES_PER_CHUNK;
 464                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 465                                                 INT_GET(agi->agi_length, ARCH_CONVERT))
 466                                         break;
 467                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 468                                                             &tmp);
 469                         }
 470                         /*
 471                          * If ran off the end of the ag either with an error,
 472                          * or the normal way, set end and stop collecting.
 473                          */
 474                         if (error ||
 475                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 476                                     &gfree, &i)) ||
 477                             i == 0) {
 478                                 end_of_ag = 1;
 479                                 break;
 480                         }
 481                         /*
 482                          * If this chunk has any allocated inodes, save it.
 483                          */
 484                         if (gcnt < XFS_INODES_PER_CHUNK) {
 485                                 INT_SET(irbp->ir_startino, ARCH_CONVERT, gino);
 486                                 INT_SET(irbp->ir_freecount, ARCH_CONVERT, gcnt);
 487                                 INT_SET(irbp->ir_free, ARCH_CONVERT, gfree);
 488                                 irbp++;
 489                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 490                         }
 491                         /*
 492                          * Set agino to after this chunk and bump the cursor.
 493                          */
 494                         agino = gino + XFS_INODES_PER_CHUNK;
 495                         error = xfs_inobt_increment(cur, 0, &tmp);
 496                 }
 497                 /*
 498                  * Drop the btree buffers and the agi buffer.
 499                  * We can't hold any of the locks these represent
 500                  * when calling iget.
 501                  */
 502                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 503                 xfs_buf_relse(agbp);
 504                 /*
 505                  * Now format all the good inodes into the user's buffer.
 506                  */
 507                 irbufend = irbp;
 508                 for (irbp = irbuf;
 509                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 510                         /*
 511                          * Read-ahead the next chunk's worth of inodes.
 512                          */
 513                         if (&irbp[1] < irbufend) {
 514                                 /*
 515                                  * Loop over all clusters in the next chunk.
 516                                  * Do a readahead if there are any allocated
 517                                  * inodes in that cluster.
 518                                  */
 519                                 for (agbno = XFS_AGINO_TO_AGBNO(mp,
 520                                                         INT_GET(irbp[1].ir_startino, ARCH_CONVERT)),
 521                                      chunkidx = 0;
 522                                      chunkidx < XFS_INODES_PER_CHUNK;
 523                                      chunkidx += nicluster,
 524                                      agbno += nbcluster) {
 525                                         if (XFS_INOBT_MASKN(chunkidx,
 526                                                             nicluster) &
 527                                             ~(INT_GET(irbp[1].ir_free, ARCH_CONVERT)))
 528                                                 xfs_btree_reada_bufs(mp, agno,
 529                                                         agbno, nbcluster);
 530                                 }
 531                         }
 532                         /*
 533                          * Now process this chunk of inodes.
 534                          */
 535                         for (agino = INT_GET(irbp->ir_startino, ARCH_CONVERT), chunkidx = 0, clustidx = 0;
 536                              ubleft > 0 &&
 537                                 INT_GET(irbp->ir_freecount, ARCH_CONVERT) < XFS_INODES_PER_CHUNK;
 538                              chunkidx++, clustidx++, agino++) {
 539                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 540                                 /*
 541                                  * Recompute agbno if this is the
 542                                  * first inode of the cluster.
 543                                  *
 544                                  * Careful with clustidx.   There can be
 545                                  * multple clusters per chunk, a single
 546                                  * cluster per chunk or a cluster that has
 547                                  * inodes represented from several different
 548                                  * chunks (if blocksize is large).
 549                                  *
 550                                  * Because of this, the starting clustidx is
 551                                  * initialized to zero in this loop but must
 552                                  * later be reset after reading in the cluster
 553                                  * buffer.
 554                                  */
 555                                 if ((chunkidx & (nicluster - 1)) == 0) {
 556                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 557                                                         INT_GET(irbp->ir_startino, ARCH_CONVERT)) +
 558                                                 ((chunkidx & nimask) >>
 559                                                  mp->m_sb.sb_inopblog);
 560
 561                                         if (flags & BULKSTAT_FG_QUICK) {
 562                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 563                                                                        agino);
 564                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 565                                                                        agbno);
 566
 567                                                 /*
 568                                                  * Get the inode cluster buffer
 569                                                  */
 570                                                 ASSERT(xfs_inode_zone != NULL);
 571                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 572                                                                       KM_SLEEP);
 573                                                 ip->i_ino = ino;
 574                                                 ip->i_mount = mp;
 575                                                 if (bp)
 576                                                         xfs_buf_relse(bp);
 577                                                 error = xfs_itobp(mp, NULL, ip,
 578                                                                   &dip, &bp, bno);
 579                                                 if (!error)
 580                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 581                                                 kmem_zone_free(xfs_inode_zone, ip);
 582                                                 if (XFS_TEST_ERROR(error != 0,
 583                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 584                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 585                                                         bp = NULL;
 586                                                         break;
 587                                                 }
 588                                         }
 589                                 }
 590                                 /*
 591                                  * Skip if this inode is free.
 592                                  */
 593                                 if (XFS_INOBT_MASK(chunkidx) & INT_GET(irbp->ir_free, ARCH_CONVERT))
 594                                         continue;
 595                                 /*
 596                                  * Count used inodes as free so we can tell
 597                                  * when the chunk is used up.
 598                                  */
 599                                 INT_MOD(irbp->ir_freecount, ARCH_CONVERT, +1);
 600                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 601                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 602                                 if (flags & BULKSTAT_FG_QUICK) {
 603                                         dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 604                                               (clustidx << mp->m_sb.sb_inodelog));
 605
 606                                         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT)
 607                                                     != XFS_DINODE_MAGIC
 608                                             || !XFS_DINODE_GOOD_VERSION(
 609                                                     INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 610                                                 continue;
 611                                 }
 612
 613                                 /*
 614                                  * Get the inode and fill in a single buffer.
 615                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 616                                  * BULKSTAT_FG_IGET uses igets.
 617                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 618                                  * This is also used to count inodes/blks, etc
 619                                  * in xfs_qm_quotacheck.
 620                                  */
 621                                 ubused = statstruct_size;
 622                                 error = formatter(mp, ino, ubufp,
 623                                                 ubleft, private_data,
 624                                                 bno, &ubused, dip, &fmterror);
 625                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 626                                         if (error == ENOMEM)
 627                                                 ubleft = 0;
 628                                         continue;
 629                                 }
 630                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 631                                         ubleft = 0;
 632                                         ASSERT(error);
 633                                         rval = error;
 634                                         break;
 635                                 }
 636                                 if (ubufp)
 637                                         ubufp += ubused;
 638                                 ubleft -= ubused;
 639                                 ubelem++;
 640                                 lastino = ino;
 641                         }
 642                 }
 643
 644                 if (bp)
 645                         xfs_buf_relse(bp);
 646
 647                 /*
 648                  * Set up for the next loop iteration.
 649                  */
 650                 if (ubleft > 0) {
 651                         if (end_of_ag) {
 652                                 agno++;
 653                                 agino = 0;
 654                         } else
 655                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 656                 } else
 657                         break;
 658         }
 659         /*
 660          * Done, we're either out of filesystem or space to put the data.
 661          */
 662         kmem_free(irbuf, NBPC);
 663         if (ubuffer)
 664                 unuseracc(ubuffer, ubcount * statstruct_size, (B_READ|B_PHYS));
 665         *ubcountp = ubelem;
 666         if (agno >= mp->m_sb.sb_agcount) {
 667                 /*
 668                  * If we ran out of filesystem, mark lastino as off
 669                  * the end of the filesystem, so the next call
 670                  * will return immediately.
 671                  */
 672                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 673                 *done = 1;
 674         } else
 675                 *lastinop = (xfs_ino_t)lastino;
 676
 677         return rval;
 678 }
 679
 680 /*
 681  * Return stat information in bulk (by-inode) for the filesystem.
 682  * Special case for non-sequential one inode bulkstat.
 683  */
 684 int                                     /* error status */
 685 xfs_bulkstat_single(
 686         xfs_mount_t             *mp,    /* mount point for filesystem */
 687         xfs_ino_t               *lastinop, /* inode to return */
 688         char                    __user *buffer, /* buffer with inode stats */
 689         int                     *done)  /* 1 if there're more stats to get */
 690 {
 691         int                     count;  /* count value for bulkstat call */
 692         int                     error;  /* return value */
 693         xfs_ino_t               ino;    /* filesystem inode number */
 694         int                     res;    /* result from bs1 */
 695
 696         /*
 697          * note that requesting valid inode numbers which are not allocated
 698          * to inodes will most likely cause xfs_itobp to generate warning
 699          * messages about bad magic numbers. This is ok. The fact that
 700          * the inode isn't actually an inode is handled by the
 701          * error check below. Done this way to make the usual case faster
 702          * at the expense of the error case.
 703          */
 704
 705         ino = (xfs_ino_t)*lastinop;
 706         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 707                                  NULL, 0, NULL, NULL, &res);
 708         if (error) {
 709                 /*
 710                  * Special case way failed, do it the "long" way
 711                  * to see if that works.
 712                  */
 713                 (*lastinop)--;
 714                 count = 1;
 715                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 716                                 NULL, sizeof(xfs_bstat_t), buffer,
 717                                 BULKSTAT_FG_IGET, done))
 718                         return error;
 719                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 720                         return error == EFSCORRUPTED ?
 721                                 XFS_ERROR(EINVAL) : error;
 722                 else
 723                         return 0;
 724         }
 725         *done = 0;
 726         return 0;
 727 }
 728
 729 /*
 730  * Return inode number table for the filesystem.
 731  */
 732 int                                     /* error status */
 733 xfs_inumbers(
 734         xfs_mount_t     *mp,            /* mount point for filesystem */
 735         xfs_ino_t       *lastino,       /* last inode returned */
 736         int             *count,         /* size of buffer/count returned */
 737         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 738 {
 739         xfs_buf_t       *agbp;
 740         xfs_agino_t     agino;
 741         xfs_agnumber_t  agno;
 742         int             bcount;
 743         xfs_inogrp_t    *buffer;
 744         int             bufidx;
 745         xfs_btree_cur_t *cur;
 746         int             error;
 747         __int32_t       gcnt;
 748         xfs_inofree_t   gfree;
 749         xfs_agino_t     gino;
 750         int             i;
 751         xfs_ino_t       ino;
 752         int             left;
 753         int             tmp;
 754
 755         ino = (xfs_ino_t)*lastino;
 756         agno = XFS_INO_TO_AGNO(mp, ino);
 757         agino = XFS_INO_TO_AGINO(mp, ino);
 758         left = *count;
 759         *count = 0;
 760         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 761         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 762         error = bufidx = 0;
 763         cur = NULL;
 764         agbp = NULL;
 765         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 766                 if (agbp == NULL) {
 767                         down_read(&mp->m_peraglock);
 768                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 769                         up_read(&mp->m_peraglock);
 770                         if (error) {
 771                                 /*
 772                                  * If we can't read the AGI of this ag,
 773                                  * then just skip to the next one.
 774                                  */
 775                                 ASSERT(cur == NULL);
 776                                 agbp = NULL;
 777                                 agno++;
 778                                 agino = 0;
 779                                 continue;
 780                         }
 781                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 782                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 783                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 784                         if (error) {
 785                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 786                                 cur = NULL;
 787                                 xfs_buf_relse(agbp);
 788                                 agbp = NULL;
 789                                 /*
 790                                  * Move up the the last inode in the current
 791                                  * chunk.  The lookup_ge will always get
 792                                  * us the first inode in the next chunk.
 793                                  */
 794                                 agino += XFS_INODES_PER_CHUNK - 1;
 795                                 continue;
 796                         }
 797                 }
 798                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 799                         &i)) ||
 800                     i == 0) {
 801                         xfs_buf_relse(agbp);
 802                         agbp = NULL;
 803                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 804                         cur = NULL;
 805                         agno++;
 806                         agino = 0;
 807                         continue;
 808                 }
 809                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 810                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 811                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 812                 buffer[bufidx].xi_allocmask = ~gfree;
 813                 bufidx++;
 814                 left--;
 815                 if (bufidx == bcount) {
 816                         if (copy_to_user(ubuffer, buffer,
 817                                         bufidx * sizeof(*buffer))) {
 818                                 error = XFS_ERROR(EFAULT);
 819                                 break;
 820                         }
 821                         ubuffer += bufidx;
 822                         *count += bufidx;
 823                         bufidx = 0;
 824                 }
 825                 if (left) {
 826                         error = xfs_inobt_increment(cur, 0, &tmp);
 827                         if (error) {
 828                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 829                                 cur = NULL;
 830                                 xfs_buf_relse(agbp);
 831                                 agbp = NULL;
 832                                 /*
 833                                  * The agino value has already been bumped.
 834                                  * Just try to skip up to it.
 835                                  */
 836                                 agino += XFS_INODES_PER_CHUNK;
 837                                 continue;
 838                         }
 839                 }
 840         }
 841         if (!error) {
 842                 if (bufidx) {
 843                         if (copy_to_user(ubuffer, buffer,
 844                                         bufidx * sizeof(*buffer)))
 845                                 error = XFS_ERROR(EFAULT);
 846                         else
 847                                 *count += bufidx;
 848                 }
 849                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 850         }
 851         kmem_free(buffer, bcount * sizeof(*buffer));
 852         if (cur)
 853                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 854                                            XFS_BTREE_NOERROR));
 855         if (agbp)
 856                 xfs_buf_relse(agbp);
 857         return error;
 858 }