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