libxfs/xfs_alloc_btree.c

   1 /*
   2  * Copyright (c) 2000-2001,2005 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 "libxfs_priv.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_btree.h"
  27 #include "xfs_alloc_btree.h"
  28 #include "xfs_alloc.h"
  29 #include "xfs_trace.h"
  30 #include "xfs_cksum.h"
  31 #include "xfs_trans.h"
  32
  33
  34 STATIC struct xfs_btree_cur *
  35 xfs_allocbt_dup_cursor(
  36         struct xfs_btree_cur    *cur)
  37 {
  38         return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
  39                         cur->bc_private.a.agbp, cur->bc_private.a.agno,
  40                         cur->bc_btnum);
  41 }
  42
  43 STATIC void
  44 xfs_allocbt_set_root(
  45         struct xfs_btree_cur    *cur,
  46         union xfs_btree_ptr     *ptr,
  47         int                     inc)
  48 {
  49         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
  50         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
  51         xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
  52         int                     btnum = cur->bc_btnum;
  53         struct xfs_perag        *pag = xfs_perag_get(cur->bc_mp, seqno);
  54
  55         ASSERT(ptr->s != 0);
  56
  57         agf->agf_roots[btnum] = ptr->s;
  58         be32_add_cpu(&agf->agf_levels[btnum], inc);
  59         pag->pagf_levels[btnum] += inc;
  60         xfs_perag_put(pag);
  61
  62         xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
  63 }
  64
  65 STATIC int
  66 xfs_allocbt_alloc_block(
  67         struct xfs_btree_cur    *cur,
  68         union xfs_btree_ptr     *start,
  69         union xfs_btree_ptr     *new,
  70         int                     *stat)
  71 {
  72         int                     error;
  73         xfs_agblock_t           bno;
  74
  75         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
  76
  77         /* Allocate the new block from the freelist. If we can't, give up.  */
  78         error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
  79                                        &bno, 1);
  80         if (error) {
  81                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
  82                 return error;
  83         }
  84
  85         if (bno == NULLAGBLOCK) {
  86                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
  87                 *stat = 0;
  88                 return 0;
  89         }
  90
  91         xfs_extent_busy_reuse(cur->bc_mp, cur->bc_private.a.agno, bno, 1, false);
  92
  93         xfs_trans_agbtree_delta(cur->bc_tp, 1);
  94         new->s = cpu_to_be32(bno);
  95
  96         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
  97         *stat = 1;
  98         return 0;
  99 }
 100
 101 STATIC int
 102 xfs_allocbt_free_block(
 103         struct xfs_btree_cur    *cur,
 104         struct xfs_buf          *bp)
 105 {
 106         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
 107         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
 108         xfs_agblock_t           bno;
 109         int                     error;
 110
 111         bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
 112         error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
 113         if (error)
 114                 return error;
 115
 116         xfs_extent_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1,
 117                               XFS_EXTENT_BUSY_SKIP_DISCARD);
 118         xfs_trans_agbtree_delta(cur->bc_tp, -1);
 119         return 0;
 120 }
 121
 122 /*
 123  * Update the longest extent in the AGF
 124  */
 125 STATIC void
 126 xfs_allocbt_update_lastrec(
 127         struct xfs_btree_cur    *cur,
 128         struct xfs_btree_block  *block,
 129         union xfs_btree_rec     *rec,
 130         int                     ptr,
 131         int                     reason)
 132 {
 133         struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
 134         xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
 135         struct xfs_perag        *pag;
 136         __be32                  len;
 137         int                     numrecs;
 138
 139         ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
 140
 141         switch (reason) {
 142         case LASTREC_UPDATE:
 143                 /*
 144                  * If this is the last leaf block and it's the last record,
 145                  * then update the size of the longest extent in the AG.
 146                  */
 147                 if (ptr != xfs_btree_get_numrecs(block))
 148                         return;
 149                 len = rec->alloc.ar_blockcount;
 150                 break;
 151         case LASTREC_INSREC:
 152                 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
 153                     be32_to_cpu(agf->agf_longest))
 154                         return;
 155                 len = rec->alloc.ar_blockcount;
 156                 break;
 157         case LASTREC_DELREC:
 158                 numrecs = xfs_btree_get_numrecs(block);
 159                 if (ptr <= numrecs)
 160                         return;
 161                 ASSERT(ptr == numrecs + 1);
 162
 163                 if (numrecs) {
 164                         xfs_alloc_rec_t *rrp;
 165
 166                         rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
 167                         len = rrp->ar_blockcount;
 168                 } else {
 169                         len = 0;
 170                 }
 171
 172                 break;
 173         default:
 174                 ASSERT(0);
 175                 return;
 176         }
 177
 178         agf->agf_longest = len;
 179         pag = xfs_perag_get(cur->bc_mp, seqno);
 180         pag->pagf_longest = be32_to_cpu(len);
 181         xfs_perag_put(pag);
 182         xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, XFS_AGF_LONGEST);
 183 }
 184
 185 STATIC int
 186 xfs_allocbt_get_minrecs(
 187         struct xfs_btree_cur    *cur,
 188         int                     level)
 189 {
 190         return cur->bc_mp->m_alloc_mnr[level != 0];
 191 }
 192
 193 STATIC int
 194 xfs_allocbt_get_maxrecs(
 195         struct xfs_btree_cur    *cur,
 196         int                     level)
 197 {
 198         return cur->bc_mp->m_alloc_mxr[level != 0];
 199 }
 200
 201 STATIC void
 202 xfs_allocbt_init_key_from_rec(
 203         union xfs_btree_key     *key,
 204         union xfs_btree_rec     *rec)
 205 {
 206         key->alloc.ar_startblock = rec->alloc.ar_startblock;
 207         key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
 208 }
 209
 210 STATIC void
 211 xfs_bnobt_init_high_key_from_rec(
 212         union xfs_btree_key     *key,
 213         union xfs_btree_rec     *rec)
 214 {
 215         __u32                   x;
 216
 217         x = be32_to_cpu(rec->alloc.ar_startblock);
 218         x += be32_to_cpu(rec->alloc.ar_blockcount) - 1;
 219         key->alloc.ar_startblock = cpu_to_be32(x);
 220         key->alloc.ar_blockcount = 0;
 221 }
 222
 223 STATIC void
 224 xfs_cntbt_init_high_key_from_rec(
 225         union xfs_btree_key     *key,
 226         union xfs_btree_rec     *rec)
 227 {
 228         key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
 229         key->alloc.ar_startblock = 0;
 230 }
 231
 232 STATIC void
 233 xfs_allocbt_init_rec_from_cur(
 234         struct xfs_btree_cur    *cur,
 235         union xfs_btree_rec     *rec)
 236 {
 237         rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
 238         rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
 239 }
 240
 241 STATIC void
 242 xfs_allocbt_init_ptr_from_cur(
 243         struct xfs_btree_cur    *cur,
 244         union xfs_btree_ptr     *ptr)
 245 {
 246         struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
 247
 248         ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
 249         ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
 250
 251         ptr->s = agf->agf_roots[cur->bc_btnum];
 252 }
 253
 254 STATIC int64_t
 255 xfs_bnobt_key_diff(
 256         struct xfs_btree_cur    *cur,
 257         union xfs_btree_key     *key)
 258 {
 259         xfs_alloc_rec_incore_t  *rec = &cur->bc_rec.a;
 260         xfs_alloc_key_t         *kp = &key->alloc;
 261
 262         return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
 263 }
 264
 265 STATIC int64_t
 266 xfs_cntbt_key_diff(
 267         struct xfs_btree_cur    *cur,
 268         union xfs_btree_key     *key)
 269 {
 270         xfs_alloc_rec_incore_t  *rec = &cur->bc_rec.a;
 271         xfs_alloc_key_t         *kp = &key->alloc;
 272         int64_t                 diff;
 273
 274         diff = (int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
 275         if (diff)
 276                 return diff;
 277
 278         return (int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
 279 }
 280
 281 STATIC int64_t
 282 xfs_bnobt_diff_two_keys(
 283         struct xfs_btree_cur    *cur,
 284         union xfs_btree_key     *k1,
 285         union xfs_btree_key     *k2)
 286 {
 287         return (int64_t)be32_to_cpu(k1->alloc.ar_startblock) -
 288                           be32_to_cpu(k2->alloc.ar_startblock);
 289 }
 290
 291 STATIC int64_t
 292 xfs_cntbt_diff_two_keys(
 293         struct xfs_btree_cur    *cur,
 294         union xfs_btree_key     *k1,
 295         union xfs_btree_key     *k2)
 296 {
 297         int64_t                 diff;
 298
 299         diff =  be32_to_cpu(k1->alloc.ar_blockcount) -
 300                 be32_to_cpu(k2->alloc.ar_blockcount);
 301         if (diff)
 302                 return diff;
 303
 304         return  be32_to_cpu(k1->alloc.ar_startblock) -
 305                 be32_to_cpu(k2->alloc.ar_startblock);
 306 }
 307
 308 static xfs_failaddr_t
 309 xfs_allocbt_verify(
 310         struct xfs_buf          *bp)
 311 {
 312         struct xfs_mount        *mp = bp->b_target->bt_mount;
 313         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 314         struct xfs_perag        *pag = bp->b_pag;
 315         xfs_failaddr_t          fa;
 316         unsigned int            level;
 317
 318         /*
 319          * magic number and level verification
 320          *
 321          * During growfs operations, we can't verify the exact level or owner as
 322          * the perag is not fully initialised and hence not attached to the
 323          * buffer.  In this case, check against the maximum tree depth.
 324          *
 325          * Similarly, during log recovery we will have a perag structure
 326          * attached, but the agf information will not yet have been initialised
 327          * from the on disk AGF. Again, we can only check against maximum limits
 328          * in this case.
 329          */
 330         level = be16_to_cpu(block->bb_level);
 331         switch (block->bb_magic) {
 332         case cpu_to_be32(XFS_ABTB_CRC_MAGIC):
 333                 fa = xfs_btree_sblock_v5hdr_verify(bp);
 334                 if (fa)
 335                         return fa;
 336                 /* fall through */
 337         case cpu_to_be32(XFS_ABTB_MAGIC):
 338                 if (pag && pag->pagf_init) {
 339                         if (level >= pag->pagf_levels[XFS_BTNUM_BNOi])
 340                                 return __this_address;
 341                 } else if (level >= mp->m_ag_maxlevels)
 342                         return __this_address;
 343                 break;
 344         case cpu_to_be32(XFS_ABTC_CRC_MAGIC):
 345                 fa = xfs_btree_sblock_v5hdr_verify(bp);
 346                 if (fa)
 347                         return fa;
 348                 /* fall through */
 349         case cpu_to_be32(XFS_ABTC_MAGIC):
 350                 if (pag && pag->pagf_init) {
 351                         if (level >= pag->pagf_levels[XFS_BTNUM_CNTi])
 352                                 return __this_address;
 353                 } else if (level >= mp->m_ag_maxlevels)
 354                         return __this_address;
 355                 break;
 356         default:
 357                 return __this_address;
 358         }
 359
 360         return xfs_btree_sblock_verify(bp, mp->m_alloc_mxr[level != 0]);
 361 }
 362
 363 static void
 364 xfs_allocbt_read_verify(
 365         struct xfs_buf  *bp)
 366 {
 367         xfs_failaddr_t  fa;
 368
 369         if (!xfs_btree_sblock_verify_crc(bp))
 370                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 371         else {
 372                 fa = xfs_allocbt_verify(bp);
 373                 if (fa)
 374                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 375         }
 376
 377         if (bp->b_error)
 378                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 379 }
 380
 381 static void
 382 xfs_allocbt_write_verify(
 383         struct xfs_buf  *bp)
 384 {
 385         xfs_failaddr_t  fa;
 386
 387         fa = xfs_allocbt_verify(bp);
 388         if (fa) {
 389                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 390                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 391                 return;
 392         }
 393         xfs_btree_sblock_calc_crc(bp);
 394
 395 }
 396
 397 const struct xfs_buf_ops xfs_allocbt_buf_ops = {
 398         .name = "xfs_allocbt",
 399         .verify_read = xfs_allocbt_read_verify,
 400         .verify_write = xfs_allocbt_write_verify,
 401         .verify_struct = xfs_allocbt_verify,
 402 };
 403
 404
 405 STATIC int
 406 xfs_bnobt_keys_inorder(
 407         struct xfs_btree_cur    *cur,
 408         union xfs_btree_key     *k1,
 409         union xfs_btree_key     *k2)
 410 {
 411         return be32_to_cpu(k1->alloc.ar_startblock) <
 412                be32_to_cpu(k2->alloc.ar_startblock);
 413 }
 414
 415 STATIC int
 416 xfs_bnobt_recs_inorder(
 417         struct xfs_btree_cur    *cur,
 418         union xfs_btree_rec     *r1,
 419         union xfs_btree_rec     *r2)
 420 {
 421         return be32_to_cpu(r1->alloc.ar_startblock) +
 422                 be32_to_cpu(r1->alloc.ar_blockcount) <=
 423                 be32_to_cpu(r2->alloc.ar_startblock);
 424 }
 425
 426 STATIC int
 427 xfs_cntbt_keys_inorder(
 428         struct xfs_btree_cur    *cur,
 429         union xfs_btree_key     *k1,
 430         union xfs_btree_key     *k2)
 431 {
 432         return be32_to_cpu(k1->alloc.ar_blockcount) <
 433                 be32_to_cpu(k2->alloc.ar_blockcount) ||
 434                 (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
 435                  be32_to_cpu(k1->alloc.ar_startblock) <
 436                  be32_to_cpu(k2->alloc.ar_startblock));
 437 }
 438
 439 STATIC int
 440 xfs_cntbt_recs_inorder(
 441         struct xfs_btree_cur    *cur,
 442         union xfs_btree_rec     *r1,
 443         union xfs_btree_rec     *r2)
 444 {
 445         return be32_to_cpu(r1->alloc.ar_blockcount) <
 446                 be32_to_cpu(r2->alloc.ar_blockcount) ||
 447                 (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
 448                  be32_to_cpu(r1->alloc.ar_startblock) <
 449                  be32_to_cpu(r2->alloc.ar_startblock));
 450 }
 451
 452 static const struct xfs_btree_ops xfs_bnobt_ops = {
 453         .rec_len                = sizeof(xfs_alloc_rec_t),
 454         .key_len                = sizeof(xfs_alloc_key_t),
 455
 456         .dup_cursor             = xfs_allocbt_dup_cursor,
 457         .set_root               = xfs_allocbt_set_root,
 458         .alloc_block            = xfs_allocbt_alloc_block,
 459         .free_block             = xfs_allocbt_free_block,
 460         .update_lastrec         = xfs_allocbt_update_lastrec,
 461         .get_minrecs            = xfs_allocbt_get_minrecs,
 462         .get_maxrecs            = xfs_allocbt_get_maxrecs,
 463         .init_key_from_rec      = xfs_allocbt_init_key_from_rec,
 464         .init_high_key_from_rec = xfs_bnobt_init_high_key_from_rec,
 465         .init_rec_from_cur      = xfs_allocbt_init_rec_from_cur,
 466         .init_ptr_from_cur      = xfs_allocbt_init_ptr_from_cur,
 467         .key_diff               = xfs_bnobt_key_diff,
 468         .buf_ops                = &xfs_allocbt_buf_ops,
 469         .diff_two_keys          = xfs_bnobt_diff_two_keys,
 470         .keys_inorder           = xfs_bnobt_keys_inorder,
 471         .recs_inorder           = xfs_bnobt_recs_inorder,
 472 };
 473
 474 static const struct xfs_btree_ops xfs_cntbt_ops = {
 475         .rec_len                = sizeof(xfs_alloc_rec_t),
 476         .key_len                = sizeof(xfs_alloc_key_t),
 477
 478         .dup_cursor             = xfs_allocbt_dup_cursor,
 479         .set_root               = xfs_allocbt_set_root,
 480         .alloc_block            = xfs_allocbt_alloc_block,
 481         .free_block             = xfs_allocbt_free_block,
 482         .update_lastrec         = xfs_allocbt_update_lastrec,
 483         .get_minrecs            = xfs_allocbt_get_minrecs,
 484         .get_maxrecs            = xfs_allocbt_get_maxrecs,
 485         .init_key_from_rec      = xfs_allocbt_init_key_from_rec,
 486         .init_high_key_from_rec = xfs_cntbt_init_high_key_from_rec,
 487         .init_rec_from_cur      = xfs_allocbt_init_rec_from_cur,
 488         .init_ptr_from_cur      = xfs_allocbt_init_ptr_from_cur,
 489         .key_diff               = xfs_cntbt_key_diff,
 490         .buf_ops                = &xfs_allocbt_buf_ops,
 491         .diff_two_keys          = xfs_cntbt_diff_two_keys,
 492         .keys_inorder           = xfs_cntbt_keys_inorder,
 493         .recs_inorder           = xfs_cntbt_recs_inorder,
 494 };
 495
 496 /*
 497  * Allocate a new allocation btree cursor.
 498  */
 499 struct xfs_btree_cur *                  /* new alloc btree cursor */
 500 xfs_allocbt_init_cursor(
 501         struct xfs_mount        *mp,            /* file system mount point */
 502         struct xfs_trans        *tp,            /* transaction pointer */
 503         struct xfs_buf          *agbp,          /* buffer for agf structure */
 504         xfs_agnumber_t          agno,           /* allocation group number */
 505         xfs_btnum_t             btnum)          /* btree identifier */
 506 {
 507         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
 508         struct xfs_btree_cur    *cur;
 509
 510         ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
 511
 512         cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
 513
 514         cur->bc_tp = tp;
 515         cur->bc_mp = mp;
 516         cur->bc_btnum = btnum;
 517         cur->bc_blocklog = mp->m_sb.sb_blocklog;
 518
 519         if (btnum == XFS_BTNUM_CNT) {
 520                 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2);
 521                 cur->bc_ops = &xfs_cntbt_ops;
 522                 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]);
 523                 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
 524         } else {
 525                 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2);
 526                 cur->bc_ops = &xfs_bnobt_ops;
 527                 cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]);
 528         }
 529
 530         cur->bc_private.a.agbp = agbp;
 531         cur->bc_private.a.agno = agno;
 532
 533         if (xfs_sb_version_hascrc(&mp->m_sb))
 534                 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
 535
 536         return cur;
 537 }
 538
 539 /*
 540  * Calculate number of records in an alloc btree block.
 541  */
 542 int
 543 xfs_allocbt_maxrecs(
 544         struct xfs_mount        *mp,
 545         int                     blocklen,
 546         int                     leaf)
 547 {
 548         blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
 549
 550         if (leaf)
 551                 return blocklen / sizeof(xfs_alloc_rec_t);
 552         return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
 553 }