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