fs/xfs/xfs_ialloc.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dir2_sf.h"
  34 #include "xfs_attr_sf.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_btree.h"
  38 #include "xfs_ialloc.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_rtalloc.h"
  41 #include "xfs_error.h"
  42 #include "xfs_bmap.h"
  43
  44
  45 /*
  46  * Allocation group level functions.
  47  */
  48 static inline int
  49 xfs_ialloc_cluster_alignment(
  50         xfs_alloc_arg_t *args)
  51 {
  52         if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
  53             args->mp->m_sb.sb_inoalignmt >=
  54              XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
  55                 return args->mp->m_sb.sb_inoalignmt;
  56         return 1;
  57 }
  58
  59 /*
  60  * Lookup a record by ino in the btree given by cur.
  61  */
  62 int                                     /* error */
  63 xfs_inobt_lookup(
  64         struct xfs_btree_cur    *cur,   /* btree cursor */
  65         xfs_agino_t             ino,    /* starting inode of chunk */
  66         xfs_lookup_t            dir,    /* <=, >=, == */
  67         int                     *stat)  /* success/failure */
  68 {
  69         cur->bc_rec.i.ir_startino = ino;
  70         cur->bc_rec.i.ir_freecount = 0;
  71         cur->bc_rec.i.ir_free = 0;
  72         return xfs_btree_lookup(cur, dir, stat);
  73 }
  74
  75 /*
  76  * Update the record referred to by cur to the value given.
  77  * This either works (return 0) or gets an EFSCORRUPTED error.
  78  */
  79 STATIC int                              /* error */
  80 xfs_inobt_update(
  81         struct xfs_btree_cur    *cur,   /* btree cursor */
  82         xfs_inobt_rec_incore_t  *irec)  /* btree record */
  83 {
  84         union xfs_btree_rec     rec;
  85
  86         rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
  87         rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
  88         rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
  89         return xfs_btree_update(cur, &rec);
  90 }
  91
  92 /*
  93  * Get the data from the pointed-to record.
  94  */
  95 int                                     /* error */
  96 xfs_inobt_get_rec(
  97         struct xfs_btree_cur    *cur,   /* btree cursor */
  98         xfs_inobt_rec_incore_t  *irec,  /* btree record */
  99         int                     *stat)  /* output: success/failure */
 100 {
 101         union xfs_btree_rec     *rec;
 102         int                     error;
 103
 104         error = xfs_btree_get_rec(cur, &rec, stat);
 105         if (!error && *stat == 1) {
 106                 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
 107                 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
 108                 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
 109         }
 110         return error;
 111 }
 112
 113 /*
 114  * Verify that the number of free inodes in the AGI is correct.
 115  */
 116 #ifdef DEBUG
 117 STATIC int
 118 xfs_check_agi_freecount(
 119         struct xfs_btree_cur    *cur,
 120         struct xfs_agi          *agi)
 121 {
 122         if (cur->bc_nlevels == 1) {
 123                 xfs_inobt_rec_incore_t rec;
 124                 int             freecount = 0;
 125                 int             error;
 126                 int             i;
 127
 128                 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
 129                 if (error)
 130                         return error;
 131
 132                 do {
 133                         error = xfs_inobt_get_rec(cur, &rec, &i);
 134                         if (error)
 135                                 return error;
 136
 137                         if (i) {
 138                                 freecount += rec.ir_freecount;
 139                                 error = xfs_btree_increment(cur, 0, &i);
 140                                 if (error)
 141                                         return error;
 142                         }
 143                 } while (i == 1);
 144
 145                 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
 146                         ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
 147         }
 148         return 0;
 149 }
 150 #else
 151 #define xfs_check_agi_freecount(cur, agi)       0
 152 #endif
 153
 154 /*
 155  * Initialise a new set of inodes.
 156  */
 157 STATIC void
 158 xfs_ialloc_inode_init(
 159         struct xfs_mount        *mp,
 160         struct xfs_trans        *tp,
 161         xfs_agnumber_t          agno,
 162         xfs_agblock_t           agbno,
 163         xfs_agblock_t           length,
 164         unsigned int            gen)
 165 {
 166         struct xfs_buf          *fbuf;
 167         struct xfs_dinode       *free;
 168         int                     blks_per_cluster, nbufs, ninodes;
 169         int                     version;
 170         int                     i, j;
 171         xfs_daddr_t             d;
 172
 173         /*
 174          * Loop over the new block(s), filling in the inodes.
 175          * For small block sizes, manipulate the inodes in buffers
 176          * which are multiples of the blocks size.
 177          */
 178         if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
 179                 blks_per_cluster = 1;
 180                 nbufs = length;
 181                 ninodes = mp->m_sb.sb_inopblock;
 182         } else {
 183                 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
 184                                    mp->m_sb.sb_blocksize;
 185                 nbufs = length / blks_per_cluster;
 186                 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
 187         }
 188
 189         /*
 190          * Figure out what version number to use in the inodes we create.
 191          * If the superblock version has caught up to the one that supports
 192          * the new inode format, then use the new inode version.  Otherwise
 193          * use the old version so that old kernels will continue to be
 194          * able to use the file system.
 195          */
 196         if (xfs_sb_version_hasnlink(&mp->m_sb))
 197                 version = 2;
 198         else
 199                 version = 1;
 200
 201         for (j = 0; j < nbufs; j++) {
 202                 /*
 203                  * Get the block.
 204                  */
 205                 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
 206                 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
 207                                          mp->m_bsize * blks_per_cluster,
 208                                          XFS_BUF_LOCK);
 209                 ASSERT(fbuf);
 210                 ASSERT(!XFS_BUF_GETERROR(fbuf));
 211
 212                 /*
 213                  * Initialize all inodes in this buffer and then log them.
 214                  *
 215                  * XXX: It would be much better if we had just one transaction
 216                  *      to log a whole cluster of inodes instead of all the
 217                  *      individual transactions causing a lot of log traffic.
 218                  */
 219                 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
 220                 for (i = 0; i < ninodes; i++) {
 221                         int     ioffset = i << mp->m_sb.sb_inodelog;
 222                         uint    isize = sizeof(struct xfs_dinode);
 223
 224                         free = xfs_make_iptr(mp, fbuf, i);
 225                         free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
 226                         free->di_version = version;
 227                         free->di_gen = cpu_to_be32(gen);
 228                         free->di_next_unlinked = cpu_to_be32(NULLAGINO);
 229                         xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
 230                 }
 231                 xfs_trans_inode_alloc_buf(tp, fbuf);
 232         }
 233 }
 234
 235 /*
 236  * Allocate new inodes in the allocation group specified by agbp.
 237  * Return 0 for success, else error code.
 238  */
 239 STATIC int                              /* error code or 0 */
 240 xfs_ialloc_ag_alloc(
 241         xfs_trans_t     *tp,            /* transaction pointer */
 242         xfs_buf_t       *agbp,          /* alloc group buffer */
 243         int             *alloc)
 244 {
 245         xfs_agi_t       *agi;           /* allocation group header */
 246         xfs_alloc_arg_t args;           /* allocation argument structure */
 247         xfs_btree_cur_t *cur;           /* inode btree cursor */
 248         xfs_agnumber_t  agno;
 249         int             error;
 250         int             i;
 251         xfs_agino_t     newino;         /* new first inode's number */
 252         xfs_agino_t     newlen;         /* new number of inodes */
 253         xfs_agino_t     thisino;        /* current inode number, for loop */
 254         int             isaligned = 0;  /* inode allocation at stripe unit */
 255                                         /* boundary */
 256
 257         args.tp = tp;
 258         args.mp = tp->t_mountp;
 259
 260         /*
 261          * Locking will ensure that we don't have two callers in here
 262          * at one time.
 263          */
 264         newlen = XFS_IALLOC_INODES(args.mp);
 265         if (args.mp->m_maxicount &&
 266             args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
 267                 return XFS_ERROR(ENOSPC);
 268         args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
 269         /*
 270          * First try to allocate inodes contiguous with the last-allocated
 271          * chunk of inodes.  If the filesystem is striped, this will fill
 272          * an entire stripe unit with inodes.
 273          */
 274         agi = XFS_BUF_TO_AGI(agbp);
 275         newino = be32_to_cpu(agi->agi_newino);
 276         agno = be32_to_cpu(agi->agi_seqno);
 277         args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
 278                         XFS_IALLOC_BLOCKS(args.mp);
 279         if (likely(newino != NULLAGINO &&
 280                   (args.agbno < be32_to_cpu(agi->agi_length)))) {
 281                 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 282                 args.type = XFS_ALLOCTYPE_THIS_BNO;
 283                 args.mod = args.total = args.wasdel = args.isfl =
 284                         args.userdata = args.minalignslop = 0;
 285                 args.prod = 1;
 286
 287                 /*
 288                  * We need to take into account alignment here to ensure that
 289                  * we don't modify the free list if we fail to have an exact
 290                  * block. If we don't have an exact match, and every oher
 291                  * attempt allocation attempt fails, we'll end up cancelling
 292                  * a dirty transaction and shutting down.
 293                  *
 294                  * For an exact allocation, alignment must be 1,
 295                  * however we need to take cluster alignment into account when
 296                  * fixing up the freelist. Use the minalignslop field to
 297                  * indicate that extra blocks might be required for alignment,
 298                  * but not to use them in the actual exact allocation.
 299                  */
 300                 args.alignment = 1;
 301                 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
 302
 303                 /* Allow space for the inode btree to split. */
 304                 args.minleft = args.mp->m_in_maxlevels - 1;
 305                 if ((error = xfs_alloc_vextent(&args)))
 306                         return error;
 307         } else
 308                 args.fsbno = NULLFSBLOCK;
 309
 310         if (unlikely(args.fsbno == NULLFSBLOCK)) {
 311                 /*
 312                  * Set the alignment for the allocation.
 313                  * If stripe alignment is turned on then align at stripe unit
 314                  * boundary.
 315                  * If the cluster size is smaller than a filesystem block
 316                  * then we're doing I/O for inodes in filesystem block size
 317                  * pieces, so don't need alignment anyway.
 318                  */
 319                 isaligned = 0;
 320                 if (args.mp->m_sinoalign) {
 321                         ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
 322                         args.alignment = args.mp->m_dalign;
 323                         isaligned = 1;
 324                 } else
 325                         args.alignment = xfs_ialloc_cluster_alignment(&args);
 326                 /*
 327                  * Need to figure out where to allocate the inode blocks.
 328                  * Ideally they should be spaced out through the a.g.
 329                  * For now, just allocate blocks up front.
 330                  */
 331                 args.agbno = be32_to_cpu(agi->agi_root);
 332                 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 333                 /*
 334                  * Allocate a fixed-size extent of inodes.
 335                  */
 336                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 337                 args.mod = args.total = args.wasdel = args.isfl =
 338                         args.userdata = args.minalignslop = 0;
 339                 args.prod = 1;
 340                 /*
 341                  * Allow space for the inode btree to split.
 342                  */
 343                 args.minleft = args.mp->m_in_maxlevels - 1;
 344                 if ((error = xfs_alloc_vextent(&args)))
 345                         return error;
 346         }
 347
 348         /*
 349          * If stripe alignment is turned on, then try again with cluster
 350          * alignment.
 351          */
 352         if (isaligned && args.fsbno == NULLFSBLOCK) {
 353                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
 354                 args.agbno = be32_to_cpu(agi->agi_root);
 355                 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 356                 args.alignment = xfs_ialloc_cluster_alignment(&args);
 357                 if ((error = xfs_alloc_vextent(&args)))
 358                         return error;
 359         }
 360
 361         if (args.fsbno == NULLFSBLOCK) {
 362                 *alloc = 0;
 363                 return 0;
 364         }
 365         ASSERT(args.len == args.minlen);
 366
 367         /*
 368          * Stamp and write the inode buffers.
 369          *
 370          * Seed the new inode cluster with a random generation number. This
 371          * prevents short-term reuse of generation numbers if a chunk is
 372          * freed and then immediately reallocated. We use random numbers
 373          * rather than a linear progression to prevent the next generation
 374          * number from being easily guessable.
 375          */
 376         xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
 377                               random32());
 378
 379         /*
 380          * Convert the results.
 381          */
 382         newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
 383         be32_add_cpu(&agi->agi_count, newlen);
 384         be32_add_cpu(&agi->agi_freecount, newlen);
 385         down_read(&args.mp->m_peraglock);
 386         args.mp->m_perag[agno].pagi_freecount += newlen;
 387         up_read(&args.mp->m_peraglock);
 388         agi->agi_newino = cpu_to_be32(newino);
 389
 390         /*
 391          * Insert records describing the new inode chunk into the btree.
 392          */
 393         cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
 394         for (thisino = newino;
 395              thisino < newino + newlen;
 396              thisino += XFS_INODES_PER_CHUNK) {
 397                 cur->bc_rec.i.ir_startino = thisino;
 398                 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
 399                 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
 400                 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
 401                 if (error) {
 402                         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 403                         return error;
 404                 }
 405                 ASSERT(i == 0);
 406                 error = xfs_btree_insert(cur, &i);
 407                 if (error) {
 408                         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 409                         return error;
 410                 }
 411                 ASSERT(i == 1);
 412         }
 413         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 414         /*
 415          * Log allocation group header fields
 416          */
 417         xfs_ialloc_log_agi(tp, agbp,
 418                 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
 419         /*
 420          * Modify/log superblock values for inode count and inode free count.
 421          */
 422         xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
 423         xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
 424         *alloc = 1;
 425         return 0;
 426 }
 427
 428 STATIC_INLINE xfs_agnumber_t
 429 xfs_ialloc_next_ag(
 430         xfs_mount_t     *mp)
 431 {
 432         xfs_agnumber_t  agno;
 433
 434         spin_lock(&mp->m_agirotor_lock);
 435         agno = mp->m_agirotor;
 436         if (++mp->m_agirotor == mp->m_maxagi)
 437                 mp->m_agirotor = 0;
 438         spin_unlock(&mp->m_agirotor_lock);
 439
 440         return agno;
 441 }
 442
 443 /*
 444  * Select an allocation group to look for a free inode in, based on the parent
 445  * inode and then mode.  Return the allocation group buffer.
 446  */
 447 STATIC xfs_buf_t *                      /* allocation group buffer */
 448 xfs_ialloc_ag_select(
 449         xfs_trans_t     *tp,            /* transaction pointer */
 450         xfs_ino_t       parent,         /* parent directory inode number */
 451         mode_t          mode,           /* bits set to indicate file type */
 452         int             okalloc)        /* ok to allocate more space */
 453 {
 454         xfs_buf_t       *agbp;          /* allocation group header buffer */
 455         xfs_agnumber_t  agcount;        /* number of ag's in the filesystem */
 456         xfs_agnumber_t  agno;           /* current ag number */
 457         int             flags;          /* alloc buffer locking flags */
 458         xfs_extlen_t    ineed;          /* blocks needed for inode allocation */
 459         xfs_extlen_t    longest = 0;    /* longest extent available */
 460         xfs_mount_t     *mp;            /* mount point structure */
 461         int             needspace;      /* file mode implies space allocated */
 462         xfs_perag_t     *pag;           /* per allocation group data */
 463         xfs_agnumber_t  pagno;          /* parent (starting) ag number */
 464
 465         /*
 466          * Files of these types need at least one block if length > 0
 467          * (and they won't fit in the inode, but that's hard to figure out).
 468          */
 469         needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
 470         mp = tp->t_mountp;
 471         agcount = mp->m_maxagi;
 472         if (S_ISDIR(mode))
 473                 pagno = xfs_ialloc_next_ag(mp);
 474         else {
 475                 pagno = XFS_INO_TO_AGNO(mp, parent);
 476                 if (pagno >= agcount)
 477                         pagno = 0;
 478         }
 479         ASSERT(pagno < agcount);
 480         /*
 481          * Loop through allocation groups, looking for one with a little
 482          * free space in it.  Note we don't look for free inodes, exactly.
 483          * Instead, we include whether there is a need to allocate inodes
 484          * to mean that blocks must be allocated for them,
 485          * if none are currently free.
 486          */
 487         agno = pagno;
 488         flags = XFS_ALLOC_FLAG_TRYLOCK;
 489         down_read(&mp->m_peraglock);
 490         for (;;) {
 491                 pag = &mp->m_perag[agno];
 492                 if (!pag->pagi_init) {
 493                         if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
 494                                 agbp = NULL;
 495                                 goto nextag;
 496                         }
 497                 } else
 498                         agbp = NULL;
 499
 500                 if (!pag->pagi_inodeok) {
 501                         xfs_ialloc_next_ag(mp);
 502                         goto unlock_nextag;
 503                 }
 504
 505                 /*
 506                  * Is there enough free space for the file plus a block
 507                  * of inodes (if we need to allocate some)?
 508                  */
 509                 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
 510                 if (ineed && !pag->pagf_init) {
 511                         if (agbp == NULL &&
 512                             xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
 513                                 agbp = NULL;
 514                                 goto nextag;
 515                         }
 516                         (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
 517                 }
 518                 if (!ineed || pag->pagf_init) {
 519                         if (ineed && !(longest = pag->pagf_longest))
 520                                 longest = pag->pagf_flcount > 0;
 521                         if (!ineed ||
 522                             (pag->pagf_freeblks >= needspace + ineed &&
 523                              longest >= ineed &&
 524                              okalloc)) {
 525                                 if (agbp == NULL &&
 526                                     xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
 527                                         agbp = NULL;
 528                                         goto nextag;
 529                                 }
 530                                 up_read(&mp->m_peraglock);
 531                                 return agbp;
 532                         }
 533                 }
 534 unlock_nextag:
 535                 if (agbp)
 536                         xfs_trans_brelse(tp, agbp);
 537 nextag:
 538                 /*
 539                  * No point in iterating over the rest, if we're shutting
 540                  * down.
 541                  */
 542                 if (XFS_FORCED_SHUTDOWN(mp)) {
 543                         up_read(&mp->m_peraglock);
 544                         return NULL;
 545                 }
 546                 agno++;
 547                 if (agno >= agcount)
 548                         agno = 0;
 549                 if (agno == pagno) {
 550                         if (flags == 0) {
 551                                 up_read(&mp->m_peraglock);
 552                                 return NULL;
 553                         }
 554                         flags = 0;
 555                 }
 556         }
 557 }
 558
 559 /*
 560  * Try to retrieve the next record to the left/right from the current one.
 561  */
 562 STATIC int
 563 xfs_ialloc_next_rec(
 564         struct xfs_btree_cur    *cur,
 565         xfs_inobt_rec_incore_t  *rec,
 566         int                     *done,
 567         int                     left)
 568 {
 569         int                     error;
 570         int                     i;
 571
 572         if (left)
 573                 error = xfs_btree_decrement(cur, 0, &i);
 574         else
 575                 error = xfs_btree_increment(cur, 0, &i);
 576
 577         if (error)
 578                 return error;
 579         *done = !i;
 580         if (i) {
 581                 error = xfs_inobt_get_rec(cur, rec, &i);
 582                 if (error)
 583                         return error;
 584                 XFS_WANT_CORRUPTED_RETURN(i == 1);
 585         }
 586
 587         return 0;
 588 }
 589
 590 STATIC int
 591 xfs_ialloc_get_rec(
 592         struct xfs_btree_cur    *cur,
 593         xfs_agino_t             agino,
 594         xfs_inobt_rec_incore_t  *rec,
 595         int                     *done,
 596         int                     left)
 597 {
 598         int                     error;
 599         int                     i;
 600
 601         error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
 602         if (error)
 603                 return error;
 604         *done = !i;
 605         if (i) {
 606                 error = xfs_inobt_get_rec(cur, rec, &i);
 607                 if (error)
 608                         return error;
 609                 XFS_WANT_CORRUPTED_RETURN(i == 1);
 610         }
 611
 612         return 0;
 613 }
 614
 615 /*
 616  * Visible inode allocation functions.
 617  */
 618
 619 /*
 620  * Allocate an inode on disk.
 621  * Mode is used to tell whether the new inode will need space, and whether
 622  * it is a directory.
 623  *
 624  * The arguments IO_agbp and alloc_done are defined to work within
 625  * the constraint of one allocation per transaction.
 626  * xfs_dialloc() is designed to be called twice if it has to do an
 627  * allocation to make more free inodes.  On the first call,
 628  * IO_agbp should be set to NULL. If an inode is available,
 629  * i.e., xfs_dialloc() did not need to do an allocation, an inode
 630  * number is returned.  In this case, IO_agbp would be set to the
 631  * current ag_buf and alloc_done set to false.
 632  * If an allocation needed to be done, xfs_dialloc would return
 633  * the current ag_buf in IO_agbp and set alloc_done to true.
 634  * The caller should then commit the current transaction, allocate a new
 635  * transaction, and call xfs_dialloc() again, passing in the previous
 636  * value of IO_agbp.  IO_agbp should be held across the transactions.
 637  * Since the agbp is locked across the two calls, the second call is
 638  * guaranteed to have a free inode available.
 639  *
 640  * Once we successfully pick an inode its number is returned and the
 641  * on-disk data structures are updated.  The inode itself is not read
 642  * in, since doing so would break ordering constraints with xfs_reclaim.
 643  */
 644 int
 645 xfs_dialloc(
 646         xfs_trans_t     *tp,            /* transaction pointer */
 647         xfs_ino_t       parent,         /* parent inode (directory) */
 648         mode_t          mode,           /* mode bits for new inode */
 649         int             okalloc,        /* ok to allocate more space */
 650         xfs_buf_t       **IO_agbp,      /* in/out ag header's buffer */
 651         boolean_t       *alloc_done,    /* true if we needed to replenish
 652                                            inode freelist */
 653         xfs_ino_t       *inop)          /* inode number allocated */
 654 {
 655         xfs_agnumber_t  agcount;        /* number of allocation groups */
 656         xfs_buf_t       *agbp;          /* allocation group header's buffer */
 657         xfs_agnumber_t  agno;           /* allocation group number */
 658         xfs_agi_t       *agi;           /* allocation group header structure */
 659         xfs_btree_cur_t *cur;           /* inode allocation btree cursor */
 660         int             error;          /* error return value */
 661         int             i;              /* result code */
 662         int             ialloced;       /* inode allocation status */
 663         int             noroom = 0;     /* no space for inode blk allocation */
 664         xfs_ino_t       ino;            /* fs-relative inode to be returned */
 665         /* REFERENCED */
 666         int             j;              /* result code */
 667         xfs_mount_t     *mp;            /* file system mount structure */
 668         int             offset;         /* index of inode in chunk */
 669         xfs_agino_t     pagino;         /* parent's AG relative inode # */
 670         xfs_agnumber_t  pagno;          /* parent's AG number */
 671         xfs_inobt_rec_incore_t rec;     /* inode allocation record */
 672         xfs_agnumber_t  tagno;          /* testing allocation group number */
 673         xfs_btree_cur_t *tcur;          /* temp cursor */
 674         xfs_inobt_rec_incore_t trec;    /* temp inode allocation record */
 675
 676
 677         if (*IO_agbp == NULL) {
 678                 /*
 679                  * We do not have an agbp, so select an initial allocation
 680                  * group for inode allocation.
 681                  */
 682                 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
 683                 /*
 684                  * Couldn't find an allocation group satisfying the
 685                  * criteria, give up.
 686                  */
 687                 if (!agbp) {
 688                         *inop = NULLFSINO;
 689                         return 0;
 690                 }
 691                 agi = XFS_BUF_TO_AGI(agbp);
 692                 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
 693         } else {
 694                 /*
 695                  * Continue where we left off before.  In this case, we
 696                  * know that the allocation group has free inodes.
 697                  */
 698                 agbp = *IO_agbp;
 699                 agi = XFS_BUF_TO_AGI(agbp);
 700                 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
 701                 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
 702         }
 703         mp = tp->t_mountp;
 704         agcount = mp->m_sb.sb_agcount;
 705         agno = be32_to_cpu(agi->agi_seqno);
 706         tagno = agno;
 707         pagno = XFS_INO_TO_AGNO(mp, parent);
 708         pagino = XFS_INO_TO_AGINO(mp, parent);
 709
 710         /*
 711          * If we have already hit the ceiling of inode blocks then clear
 712          * okalloc so we scan all available agi structures for a free
 713          * inode.
 714          */
 715
 716         if (mp->m_maxicount &&
 717             mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
 718                 noroom = 1;
 719                 okalloc = 0;
 720         }
 721
 722         /*
 723          * Loop until we find an allocation group that either has free inodes
 724          * or in which we can allocate some inodes.  Iterate through the
 725          * allocation groups upward, wrapping at the end.
 726          */
 727         *alloc_done = B_FALSE;
 728         while (!agi->agi_freecount) {
 729                 /*
 730                  * Don't do anything if we're not supposed to allocate
 731                  * any blocks, just go on to the next ag.
 732                  */
 733                 if (okalloc) {
 734                         /*
 735                          * Try to allocate some new inodes in the allocation
 736                          * group.
 737                          */
 738                         if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
 739                                 xfs_trans_brelse(tp, agbp);
 740                                 if (error == ENOSPC) {
 741                                         *inop = NULLFSINO;
 742                                         return 0;
 743                                 } else
 744                                         return error;
 745                         }
 746                         if (ialloced) {
 747                                 /*
 748                                  * We successfully allocated some inodes, return
 749                                  * the current context to the caller so that it
 750                                  * can commit the current transaction and call
 751                                  * us again where we left off.
 752                                  */
 753                                 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
 754                                 *alloc_done = B_TRUE;
 755                                 *IO_agbp = agbp;
 756                                 *inop = NULLFSINO;
 757                                 return 0;
 758                         }
 759                 }
 760                 /*
 761                  * If it failed, give up on this ag.
 762                  */
 763                 xfs_trans_brelse(tp, agbp);
 764                 /*
 765                  * Go on to the next ag: get its ag header.
 766                  */
 767 nextag:
 768                 if (++tagno == agcount)
 769                         tagno = 0;
 770                 if (tagno == agno) {
 771                         *inop = NULLFSINO;
 772                         return noroom ? ENOSPC : 0;
 773                 }
 774                 down_read(&mp->m_peraglock);
 775                 if (mp->m_perag[tagno].pagi_inodeok == 0) {
 776                         up_read(&mp->m_peraglock);
 777                         goto nextag;
 778                 }
 779                 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
 780                 up_read(&mp->m_peraglock);
 781                 if (error)
 782                         goto nextag;
 783                 agi = XFS_BUF_TO_AGI(agbp);
 784                 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
 785         }
 786         /*
 787          * Here with an allocation group that has a free inode.
 788          * Reset agno since we may have chosen a new ag in the
 789          * loop above.
 790          */
 791         agno = tagno;
 792         *IO_agbp = NULL;
 793
 794  restart_pagno:
 795         cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
 796         /*
 797          * If pagino is 0 (this is the root inode allocation) use newino.
 798          * This must work because we've just allocated some.
 799          */
 800         if (!pagino)
 801                 pagino = be32_to_cpu(agi->agi_newino);
 802
 803         error = xfs_check_agi_freecount(cur, agi);
 804         if (error)
 805                 goto error0;
 806
 807         /*
 808          * If in the same AG as the parent, try to get near the parent.
 809          */
 810         if (pagno == agno) {
 811                 xfs_perag_t     *pag = &mp->m_perag[agno];
 812                 int             doneleft;       /* done, to the left */
 813                 int             doneright;      /* done, to the right */
 814                 int             searchdistance = 10;
 815
 816                 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
 817                 if (error)
 818                         goto error0;
 819                 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 820
 821                 error = xfs_inobt_get_rec(cur, &rec, &j);
 822                 if (error)
 823                         goto error0;
 824                 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 825
 826                 if (rec.ir_freecount > 0) {
 827                         /*
 828                          * Found a free inode in the same chunk
 829                          * as the parent, done.
 830                          */
 831                         goto alloc_inode;
 832                 }
 833
 834
 835                 /*
 836                  * In the same AG as parent, but parent's chunk is full.
 837                  */
 838
 839                 /* duplicate the cursor, search left & right simultaneously */
 840                 error = xfs_btree_dup_cursor(cur, &tcur);
 841                 if (error)
 842                         goto error0;
 843
 844                 /*
 845                  * Skip to last blocks looked up if same parent inode.
 846                  */
 847                 if (pagino != NULLAGINO &&
 848                     pag->pagl_pagino == pagino &&
 849                     pag->pagl_leftrec != NULLAGINO &&
 850                     pag->pagl_rightrec != NULLAGINO) {
 851                         error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
 852                                                    &trec, &doneleft, 1);
 853                         if (error)
 854                                 goto error1;
 855
 856                         error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
 857                                                    &rec, &doneright, 0);
 858                         if (error)
 859                                 goto error1;
 860                 } else {
 861                         /* search left with tcur, back up 1 record */
 862                         error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
 863                         if (error)
 864                                 goto error1;
 865
 866                         /* search right with cur, go forward 1 record. */
 867                         error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
 868                         if (error)
 869                                 goto error1;
 870                 }
 871
 872                 /*
 873                  * Loop until we find an inode chunk with a free inode.
 874                  */
 875                 while (!doneleft || !doneright) {
 876                         int     useleft;  /* using left inode chunk this time */
 877
 878                         if (!--searchdistance) {
 879                                 /*
 880                                  * Not in range - save last search
 881                                  * location and allocate a new inode
 882                                  */
 883                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 884                                 pag->pagl_leftrec = trec.ir_startino;
 885                                 pag->pagl_rightrec = rec.ir_startino;
 886                                 pag->pagl_pagino = pagino;
 887                                 goto newino;
 888                         }
 889
 890                         /* figure out the closer block if both are valid. */
 891                         if (!doneleft && !doneright) {
 892                                 useleft = pagino -
 893                                  (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
 894                                   rec.ir_startino - pagino;
 895                         } else {
 896                                 useleft = !doneleft;
 897                         }
 898
 899                         /* free inodes to the left? */
 900                         if (useleft && trec.ir_freecount) {
 901                                 rec = trec;
 902                                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 903                                 cur = tcur;
 904
 905                                 pag->pagl_leftrec = trec.ir_startino;
 906                                 pag->pagl_rightrec = rec.ir_startino;
 907                                 pag->pagl_pagino = pagino;
 908                                 goto alloc_inode;
 909                         }
 910
 911                         /* free inodes to the right? */
 912                         if (!useleft && rec.ir_freecount) {
 913                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 914
 915                                 pag->pagl_leftrec = trec.ir_startino;
 916                                 pag->pagl_rightrec = rec.ir_startino;
 917                                 pag->pagl_pagino = pagino;
 918                                 goto alloc_inode;
 919                         }
 920
 921                         /* get next record to check */
 922                         if (useleft) {
 923                                 error = xfs_ialloc_next_rec(tcur, &trec,
 924                                                                  &doneleft, 1);
 925                         } else {
 926                                 error = xfs_ialloc_next_rec(cur, &rec,
 927                                                                  &doneright, 0);
 928                         }
 929                         if (error)
 930                                 goto error1;
 931                 }
 932
 933                 /*
 934                  * We've reached the end of the btree. because
 935                  * we are only searching a small chunk of the
 936                  * btree each search, there is obviously free
 937                  * inodes closer to the parent inode than we
 938                  * are now. restart the search again.
 939                  */
 940                 pag->pagl_pagino = NULLAGINO;
 941                 pag->pagl_leftrec = NULLAGINO;
 942                 pag->pagl_rightrec = NULLAGINO;
 943                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 944                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 945                 goto restart_pagno;
 946         }
 947
 948         /*
 949          * In a different AG from the parent.
 950          * See if the most recently allocated block has any free.
 951          */
 952 newino:
 953         if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
 954                 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
 955                                          XFS_LOOKUP_EQ, &i);
 956                 if (error)
 957                         goto error0;
 958
 959                 if (i == 1) {
 960                         error = xfs_inobt_get_rec(cur, &rec, &j);
 961                         if (error)
 962                                 goto error0;
 963
 964                         if (j == 1 && rec.ir_freecount > 0) {
 965                                 /*
 966                                  * The last chunk allocated in the group
 967                                  * still has a free inode.
 968                                  */
 969                                 goto alloc_inode;
 970                         }
 971                 }
 972         }
 973
 974         /*
 975          * None left in the last group, search the whole AG
 976          */
 977         error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
 978         if (error)
 979                 goto error0;
 980         XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 981
 982         for (;;) {
 983                 error = xfs_inobt_get_rec(cur, &rec, &i);
 984                 if (error)
 985                         goto error0;
 986                 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 987                 if (rec.ir_freecount > 0)
 988                         break;
 989                 error = xfs_btree_increment(cur, 0, &i);
 990                 if (error)
 991                         goto error0;
 992                 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 993         }
 994
 995 alloc_inode:
 996         offset = xfs_ialloc_find_free(&rec.ir_free);
 997         ASSERT(offset >= 0);
 998         ASSERT(offset < XFS_INODES_PER_CHUNK);
 999         ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1000                                    XFS_INODES_PER_CHUNK) == 0);
1001         ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1002         rec.ir_free &= ~XFS_INOBT_MASK(offset);
1003         rec.ir_freecount--;
1004         error = xfs_inobt_update(cur, &rec);
1005         if (error)
1006                 goto error0;
1007         be32_add_cpu(&agi->agi_freecount, -1);
1008         xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1009         down_read(&mp->m_peraglock);
1010         mp->m_perag[tagno].pagi_freecount--;
1011         up_read(&mp->m_peraglock);
1012
1013         error = xfs_check_agi_freecount(cur, agi);
1014         if (error)
1015                 goto error0;
1016
1017         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1018         xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1019         *inop = ino;
1020         return 0;
1021 error1:
1022         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1023 error0:
1024         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1025         return error;
1026 }
1027
1028 /*
1029  * Free disk inode.  Carefully avoids touching the incore inode, all
1030  * manipulations incore are the caller's responsibility.
1031  * The on-disk inode is not changed by this operation, only the
1032  * btree (free inode mask) is changed.
1033  */
1034 int
1035 xfs_difree(
1036         xfs_trans_t     *tp,            /* transaction pointer */
1037         xfs_ino_t       inode,          /* inode to be freed */
1038         xfs_bmap_free_t *flist,         /* extents to free */
1039         int             *delete,        /* set if inode cluster was deleted */
1040         xfs_ino_t       *first_ino)     /* first inode in deleted cluster */
1041 {
1042         /* REFERENCED */
1043         xfs_agblock_t   agbno;  /* block number containing inode */
1044         xfs_buf_t       *agbp;  /* buffer containing allocation group header */
1045         xfs_agino_t     agino;  /* inode number relative to allocation group */
1046         xfs_agnumber_t  agno;   /* allocation group number */
1047         xfs_agi_t       *agi;   /* allocation group header */
1048         xfs_btree_cur_t *cur;   /* inode btree cursor */
1049         int             error;  /* error return value */
1050         int             i;      /* result code */
1051         int             ilen;   /* inodes in an inode cluster */
1052         xfs_mount_t     *mp;    /* mount structure for filesystem */
1053         int             off;    /* offset of inode in inode chunk */
1054         xfs_inobt_rec_incore_t rec;     /* btree record */
1055
1056         mp = tp->t_mountp;
1057
1058         /*
1059          * Break up inode number into its components.
1060          */
1061         agno = XFS_INO_TO_AGNO(mp, inode);
1062         if (agno >= mp->m_sb.sb_agcount)  {
1063                 cmn_err(CE_WARN,
1064                         "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s.  Returning EINVAL.",
1065                         agno, mp->m_sb.sb_agcount, mp->m_fsname);
1066                 ASSERT(0);
1067                 return XFS_ERROR(EINVAL);
1068         }
1069         agino = XFS_INO_TO_AGINO(mp, inode);
1070         if (inode != XFS_AGINO_TO_INO(mp, agno, agino))  {
1071                 cmn_err(CE_WARN,
1072                         "xfs_difree: inode != XFS_AGINO_TO_INO() "
1073                         "(%llu != %llu) on %s.  Returning EINVAL.",
1074                         (unsigned long long)inode,
1075                         (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1076                         mp->m_fsname);
1077                 ASSERT(0);
1078                 return XFS_ERROR(EINVAL);
1079         }
1080         agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1081         if (agbno >= mp->m_sb.sb_agblocks)  {
1082                 cmn_err(CE_WARN,
1083                         "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s.  Returning EINVAL.",
1084                         agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1085                 ASSERT(0);
1086                 return XFS_ERROR(EINVAL);
1087         }
1088         /*
1089          * Get the allocation group header.
1090          */
1091         down_read(&mp->m_peraglock);
1092         error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1093         up_read(&mp->m_peraglock);
1094         if (error) {
1095                 cmn_err(CE_WARN,
1096                         "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s.  Returning error.",
1097                         error, mp->m_fsname);
1098                 return error;
1099         }
1100         agi = XFS_BUF_TO_AGI(agbp);
1101         ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1102         ASSERT(agbno < be32_to_cpu(agi->agi_length));
1103         /*
1104          * Initialize the cursor.
1105          */
1106         cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1107
1108         error = xfs_check_agi_freecount(cur, agi);
1109         if (error)
1110                 goto error0;
1111
1112         /*
1113          * Look for the entry describing this inode.
1114          */
1115         if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1116                 cmn_err(CE_WARN,
1117                         "xfs_difree: xfs_inobt_lookup returned()  an error %d on %s.  Returning error.",
1118                         error, mp->m_fsname);
1119                 goto error0;
1120         }
1121         XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1122         error = xfs_inobt_get_rec(cur, &rec, &i);
1123         if (error) {
1124                 cmn_err(CE_WARN,
1125                         "xfs_difree: xfs_inobt_get_rec()  returned an error %d on %s.  Returning error.",
1126                         error, mp->m_fsname);
1127                 goto error0;
1128         }
1129         XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1130         /*
1131          * Get the offset in the inode chunk.
1132          */
1133         off = agino - rec.ir_startino;
1134         ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1135         ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1136         /*
1137          * Mark the inode free & increment the count.
1138          */
1139         rec.ir_free |= XFS_INOBT_MASK(off);
1140         rec.ir_freecount++;
1141
1142         /*
1143          * When an inode cluster is free, it becomes eligible for removal
1144          */
1145         if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1146             (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1147
1148                 *delete = 1;
1149                 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1150
1151                 /*
1152                  * Remove the inode cluster from the AGI B+Tree, adjust the
1153                  * AGI and Superblock inode counts, and mark the disk space
1154                  * to be freed when the transaction is committed.
1155                  */
1156                 ilen = XFS_IALLOC_INODES(mp);
1157                 be32_add_cpu(&agi->agi_count, -ilen);
1158                 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1159                 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1160                 down_read(&mp->m_peraglock);
1161                 mp->m_perag[agno].pagi_freecount -= ilen - 1;
1162                 up_read(&mp->m_peraglock);
1163                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1164                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1165
1166                 if ((error = xfs_btree_delete(cur, &i))) {
1167                         cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1168                                 error, mp->m_fsname);
1169                         goto error0;
1170                 }
1171
1172                 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1173                                 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1174                                 XFS_IALLOC_BLOCKS(mp), flist, mp);
1175         } else {
1176                 *delete = 0;
1177
1178                 error = xfs_inobt_update(cur, &rec);
1179                 if (error) {
1180                         cmn_err(CE_WARN,
1181         "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1182                                 error, mp->m_fsname);
1183                         goto error0;
1184                 }
1185
1186                 /*
1187                  * Change the inode free counts and log the ag/sb changes.
1188                  */
1189                 be32_add_cpu(&agi->agi_freecount, 1);
1190                 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1191                 down_read(&mp->m_peraglock);
1192                 mp->m_perag[agno].pagi_freecount++;
1193                 up_read(&mp->m_peraglock);
1194                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1195         }
1196
1197         error = xfs_check_agi_freecount(cur, agi);
1198         if (error)
1199                 goto error0;
1200
1201         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1202         return 0;
1203
1204 error0:
1205         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1206         return error;
1207 }
1208
1209 STATIC int
1210 xfs_imap_lookup(
1211         struct xfs_mount        *mp,
1212         struct xfs_trans        *tp,
1213         xfs_agnumber_t          agno,
1214         xfs_agino_t             agino,
1215         xfs_agblock_t           agbno,
1216         xfs_agblock_t           *chunk_agbno,
1217         xfs_agblock_t           *offset_agbno,
1218         int                     flags)
1219 {
1220         struct xfs_inobt_rec_incore rec;
1221         struct xfs_btree_cur    *cur;
1222         struct xfs_buf          *agbp;
1223         int                     error;
1224         int                     i;
1225
1226         down_read(&mp->m_peraglock);
1227         error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1228         up_read(&mp->m_peraglock);
1229         if (error) {
1230                 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1231                                 "xfs_ialloc_read_agi() returned "
1232                                 "error %d, agno %d",
1233                                 error, agno);
1234                 return error;
1235         }
1236
1237         /*
1238          * Lookup the inode record for the given agino. If the record cannot be
1239          * found, then it's an invalid inode number and we should abort. Once
1240          * we have a record, we need to ensure it contains the inode number
1241          * we are looking up.
1242          */
1243         cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1244         error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1245         if (!error) {
1246                 if (i)
1247                         error = xfs_inobt_get_rec(cur, &rec, &i);
1248                 if (!error && i == 0)
1249                         error = EINVAL;
1250         }
1251
1252         xfs_trans_brelse(tp, agbp);
1253         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1254         if (error)
1255                 return error;
1256
1257         /* check that the returned record contains the required inode */
1258         if (rec.ir_startino > agino ||
1259             rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1260                 return EINVAL;
1261
1262         /* for untrusted inodes check it is allocated first */
1263         if ((flags & XFS_IGET_UNTRUSTED) &&
1264             (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1265                 return EINVAL;
1266
1267         *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1268         *offset_agbno = agbno - *chunk_agbno;
1269         return 0;
1270 }
1271
1272 /*
1273  * Return the location of the inode in imap, for mapping it into a buffer.
1274  */
1275 int
1276 xfs_imap(
1277         xfs_mount_t      *mp,   /* file system mount structure */
1278         xfs_trans_t      *tp,   /* transaction pointer */
1279         xfs_ino_t       ino,    /* inode to locate */
1280         struct xfs_imap *imap,  /* location map structure */
1281         uint            flags)  /* flags for inode btree lookup */
1282 {
1283         xfs_agblock_t   agbno;  /* block number of inode in the alloc group */
1284         xfs_agino_t     agino;  /* inode number within alloc group */
1285         xfs_agnumber_t  agno;   /* allocation group number */
1286         int             blks_per_cluster; /* num blocks per inode cluster */
1287         xfs_agblock_t   chunk_agbno;    /* first block in inode chunk */
1288         xfs_agblock_t   cluster_agbno;  /* first block in inode cluster */
1289         int             error;  /* error code */
1290         int             offset; /* index of inode in its buffer */
1291         int             offset_agbno;   /* blks from chunk start to inode */
1292
1293         ASSERT(ino != NULLFSINO);
1294
1295         /*
1296          * Split up the inode number into its parts.
1297          */
1298         agno = XFS_INO_TO_AGNO(mp, ino);
1299         agino = XFS_INO_TO_AGINO(mp, ino);
1300         agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1301         if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1302             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1303 #ifdef DEBUG
1304                 /*
1305                  * Don't output diagnostic information for untrusted inodes
1306                  * as they can be invalid without implying corruption.
1307                  */
1308                 if (flags & XFS_IGET_UNTRUSTED)
1309                         return XFS_ERROR(EINVAL);
1310                 if (agno >= mp->m_sb.sb_agcount) {
1311                         xfs_fs_cmn_err(CE_ALERT, mp,
1312                                         "xfs_imap: agno (%d) >= "
1313                                         "mp->m_sb.sb_agcount (%d)",
1314                                         agno,  mp->m_sb.sb_agcount);
1315                 }
1316                 if (agbno >= mp->m_sb.sb_agblocks) {
1317                         xfs_fs_cmn_err(CE_ALERT, mp,
1318                                         "xfs_imap: agbno (0x%llx) >= "
1319                                         "mp->m_sb.sb_agblocks (0x%lx)",
1320                                         (unsigned long long) agbno,
1321                                         (unsigned long) mp->m_sb.sb_agblocks);
1322                 }
1323                 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1324                         xfs_fs_cmn_err(CE_ALERT, mp,
1325                                         "xfs_imap: ino (0x%llx) != "
1326                                         "XFS_AGINO_TO_INO(mp, agno, agino) "
1327                                         "(0x%llx)",
1328                                         ino, XFS_AGINO_TO_INO(mp, agno, agino));
1329                 }
1330                 xfs_stack_trace();
1331 #endif /* DEBUG */
1332                 return XFS_ERROR(EINVAL);
1333         }
1334
1335         blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1336
1337         /*
1338          * For bulkstat and handle lookups, we have an untrusted inode number
1339          * that we have to verify is valid. We cannot do this just by reading
1340          * the inode buffer as it may have been unlinked and removed leaving
1341          * inodes in stale state on disk. Hence we have to do a btree lookup
1342          * in all cases where an untrusted inode number is passed.
1343          */
1344         if (flags & XFS_IGET_UNTRUSTED) {
1345                 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1346                                         &chunk_agbno, &offset_agbno, flags);
1347                 if (error)
1348                         return error;
1349                 goto out_map;
1350         }
1351
1352         /*
1353          * If the inode cluster size is the same as the blocksize or
1354          * smaller we get to the buffer by simple arithmetics.
1355          */
1356         if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1357                 offset = XFS_INO_TO_OFFSET(mp, ino);
1358                 ASSERT(offset < mp->m_sb.sb_inopblock);
1359
1360                 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1361                 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1362                 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1363                 return 0;
1364         }
1365
1366         /*
1367          * If the inode chunks are aligned then use simple maths to
1368          * find the location. Otherwise we have to do a btree
1369          * lookup to find the location.
1370          */
1371         if (mp->m_inoalign_mask) {
1372                 offset_agbno = agbno & mp->m_inoalign_mask;
1373                 chunk_agbno = agbno - offset_agbno;
1374         } else {
1375                 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1376                                         &chunk_agbno, &offset_agbno, flags);
1377                 if (error)
1378                         return error;
1379         }
1380
1381 out_map:
1382         ASSERT(agbno >= chunk_agbno);
1383         cluster_agbno = chunk_agbno +
1384                 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1385         offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1386                 XFS_INO_TO_OFFSET(mp, ino);
1387
1388         imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1389         imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1390         imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1391
1392         /*
1393          * If the inode number maps to a block outside the bounds
1394          * of the file system then return NULL rather than calling
1395          * read_buf and panicing when we get an error from the
1396          * driver.
1397          */
1398         if ((imap->im_blkno + imap->im_len) >
1399             XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1400                 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1401                         "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1402                         " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1403                         (unsigned long long) imap->im_blkno,
1404                         (unsigned long long) imap->im_len,
1405                         XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1406                 return XFS_ERROR(EINVAL);
1407         }
1408
1409         return 0;
1410 }
1411
1412 /*
1413  * Compute and fill in value of m_in_maxlevels.
1414  */
1415 void
1416 xfs_ialloc_compute_maxlevels(
1417         xfs_mount_t     *mp)            /* file system mount structure */
1418 {
1419         int             level;
1420         uint            maxblocks;
1421         uint            maxleafents;
1422         int             minleafrecs;
1423         int             minnoderecs;
1424
1425         maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1426                 XFS_INODES_PER_CHUNK_LOG;
1427         minleafrecs = mp->m_alloc_mnr[0];
1428         minnoderecs = mp->m_alloc_mnr[1];
1429         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1430         for (level = 1; maxblocks > 1; level++)
1431                 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1432         mp->m_in_maxlevels = level;
1433 }
1434
1435 /*
1436  * Log specified fields for the ag hdr (inode section)
1437  */
1438 void
1439 xfs_ialloc_log_agi(
1440         xfs_trans_t     *tp,            /* transaction pointer */
1441         xfs_buf_t       *bp,            /* allocation group header buffer */
1442         int             fields)         /* bitmask of fields to log */
1443 {
1444         int                     first;          /* first byte number */
1445         int                     last;           /* last byte number */
1446         static const short      offsets[] = {   /* field starting offsets */
1447                                         /* keep in sync with bit definitions */
1448                 offsetof(xfs_agi_t, agi_magicnum),
1449                 offsetof(xfs_agi_t, agi_versionnum),
1450                 offsetof(xfs_agi_t, agi_seqno),
1451                 offsetof(xfs_agi_t, agi_length),
1452                 offsetof(xfs_agi_t, agi_count),
1453                 offsetof(xfs_agi_t, agi_root),
1454                 offsetof(xfs_agi_t, agi_level),
1455                 offsetof(xfs_agi_t, agi_freecount),
1456                 offsetof(xfs_agi_t, agi_newino),
1457                 offsetof(xfs_agi_t, agi_dirino),
1458                 offsetof(xfs_agi_t, agi_unlinked),
1459                 sizeof(xfs_agi_t)
1460         };
1461 #ifdef DEBUG
1462         xfs_agi_t               *agi;   /* allocation group header */
1463
1464         agi = XFS_BUF_TO_AGI(bp);
1465         ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1466 #endif
1467         /*
1468          * Compute byte offsets for the first and last fields.
1469          */
1470         xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1471         /*
1472          * Log the allocation group inode header buffer.
1473          */
1474         xfs_trans_log_buf(tp, bp, first, last);
1475 }
1476
1477 #ifdef DEBUG
1478 STATIC void
1479 xfs_check_agi_unlinked(
1480         struct xfs_agi          *agi)
1481 {
1482         int                     i;
1483
1484         for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1485                 ASSERT(agi->agi_unlinked[i]);
1486 }
1487 #else
1488 #define xfs_check_agi_unlinked(agi)
1489 #endif
1490
1491 /*
1492  * Read in the allocation group header (inode allocation section)
1493  */
1494 int
1495 xfs_read_agi(
1496         struct xfs_mount        *mp,    /* file system mount structure */
1497         struct xfs_trans        *tp,    /* transaction pointer */
1498         xfs_agnumber_t          agno,   /* allocation group number */
1499         struct xfs_buf          **bpp)  /* allocation group hdr buf */
1500 {
1501         struct xfs_agi          *agi;   /* allocation group header */
1502         int                     agi_ok; /* agi is consistent */
1503         int                     error;
1504
1505         ASSERT(agno != NULLAGNUMBER);
1506
1507         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1508                         XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1509                         XFS_FSS_TO_BB(mp, 1), 0, bpp);
1510         if (error)
1511                 return error;
1512
1513         ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1514         agi = XFS_BUF_TO_AGI(*bpp);
1515
1516         /*
1517          * Validate the magic number of the agi block.
1518          */
1519         agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1520                 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1521                 be32_to_cpu(agi->agi_seqno) == agno;
1522         if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1523                         XFS_RANDOM_IALLOC_READ_AGI))) {
1524                 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1525                                      mp, agi);
1526                 xfs_trans_brelse(tp, *bpp);
1527                 return XFS_ERROR(EFSCORRUPTED);
1528         }
1529
1530         XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1531
1532         xfs_check_agi_unlinked(agi);
1533         return 0;
1534 }
1535
1536 int
1537 xfs_ialloc_read_agi(
1538         struct xfs_mount        *mp,    /* file system mount structure */
1539         struct xfs_trans        *tp,    /* transaction pointer */
1540         xfs_agnumber_t          agno,   /* allocation group number */
1541         struct xfs_buf          **bpp)  /* allocation group hdr buf */
1542 {
1543         struct xfs_agi          *agi;   /* allocation group header */
1544         struct xfs_perag        *pag;   /* per allocation group data */
1545         int                     error;
1546
1547         error = xfs_read_agi(mp, tp, agno, bpp);
1548         if (error)
1549                 return error;
1550
1551         agi = XFS_BUF_TO_AGI(*bpp);
1552         pag = &mp->m_perag[agno];
1553
1554         if (!pag->pagi_init) {
1555                 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1556                 pag->pagi_count = be32_to_cpu(agi->agi_count);
1557                 pag->pagi_init = 1;
1558         }
1559
1560         /*
1561          * It's possible for these to be out of sync if
1562          * we are in the middle of a forced shutdown.
1563          */
1564         ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1565                 XFS_FORCED_SHUTDOWN(mp));
1566         return 0;
1567 }
1568
1569 /*
1570  * Read in the agi to initialise the per-ag data in the mount structure
1571  */
1572 int
1573 xfs_ialloc_pagi_init(
1574         xfs_mount_t     *mp,            /* file system mount structure */
1575         xfs_trans_t     *tp,            /* transaction pointer */
1576         xfs_agnumber_t  agno)           /* allocation group number */
1577 {
1578         xfs_buf_t       *bp = NULL;
1579         int             error;
1580
1581         error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1582         if (error)
1583                 return error;
1584         if (bp)
1585                 xfs_trans_brelse(tp, bp);
1586         return 0;
1587 }