| 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_shared.h" |
| 21 | #include "xfs_format.h" |
| 22 | #include "xfs_log_format.h" |
| 23 | #include "xfs_trans_resv.h" |
| 24 | #include "xfs_bit.h" |
| 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_mount.h" |
| 27 | #include "xfs_defer.h" |
| 28 | #include "xfs_inode.h" |
| 29 | #include "xfs_btree.h" |
| 30 | #include "xfs_ialloc.h" |
| 31 | #include "xfs_ialloc_btree.h" |
| 32 | #include "xfs_alloc.h" |
| 33 | #include "xfs_rtalloc.h" |
| 34 | #include "xfs_error.h" |
| 35 | #include "xfs_bmap.h" |
| 36 | #include "xfs_cksum.h" |
| 37 | #include "xfs_trans.h" |
| 38 | #include "xfs_buf_item.h" |
| 39 | #include "xfs_icreate_item.h" |
| 40 | #include "xfs_icache.h" |
| 41 | #include "xfs_trace.h" |
| 42 | #include "xfs_log.h" |
| 43 | #include "xfs_rmap.h" |
| 44 | |
| 45 | |
| 46 | /* |
| 47 | * Allocation group level functions. |
| 48 | */ |
| 49 | int |
| 50 | xfs_ialloc_cluster_alignment( |
| 51 | struct xfs_mount *mp) |
| 52 | { |
| 53 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
| 54 | mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp)) |
| 55 | return mp->m_sb.sb_inoalignmt; |
| 56 | return 1; |
| 57 | } |
| 58 | |
| 59 | /* |
| 60 | * Lookup a record by ino in the btree given by cur. |
| 61 | */ |
| 62 | int /* error */ |
| 63 | xfs_inobt_lookup( |
| 64 | struct xfs_btree_cur *cur, /* btree cursor */ |
| 65 | xfs_agino_t ino, /* starting inode of chunk */ |
| 66 | xfs_lookup_t dir, /* <=, >=, == */ |
| 67 | int *stat) /* success/failure */ |
| 68 | { |
| 69 | cur->bc_rec.i.ir_startino = ino; |
| 70 | cur->bc_rec.i.ir_holemask = 0; |
| 71 | cur->bc_rec.i.ir_count = 0; |
| 72 | cur->bc_rec.i.ir_freecount = 0; |
| 73 | cur->bc_rec.i.ir_free = 0; |
| 74 | return xfs_btree_lookup(cur, dir, stat); |
| 75 | } |
| 76 | |
| 77 | /* |
| 78 | * Update the record referred to by cur to the value given. |
| 79 | * This either works (return 0) or gets an EFSCORRUPTED error. |
| 80 | */ |
| 81 | STATIC int /* error */ |
| 82 | xfs_inobt_update( |
| 83 | struct xfs_btree_cur *cur, /* btree cursor */ |
| 84 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
| 85 | { |
| 86 | union xfs_btree_rec rec; |
| 87 | |
| 88 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
| 89 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
| 90 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); |
| 91 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; |
| 92 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; |
| 93 | } else { |
| 94 | /* ir_holemask/ir_count not supported on-disk */ |
| 95 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); |
| 96 | } |
| 97 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
| 98 | return xfs_btree_update(cur, &rec); |
| 99 | } |
| 100 | |
| 101 | /* Convert on-disk btree record to incore inobt record. */ |
| 102 | void |
| 103 | xfs_inobt_btrec_to_irec( |
| 104 | struct xfs_mount *mp, |
| 105 | union xfs_btree_rec *rec, |
| 106 | struct xfs_inobt_rec_incore *irec) |
| 107 | { |
| 108 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
| 109 | if (xfs_sb_version_hassparseinodes(&mp->m_sb)) { |
| 110 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
| 111 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; |
| 112 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; |
| 113 | } else { |
| 114 | /* |
| 115 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded |
| 116 | * values for full inode chunks. |
| 117 | */ |
| 118 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; |
| 119 | irec->ir_count = XFS_INODES_PER_CHUNK; |
| 120 | irec->ir_freecount = |
| 121 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); |
| 122 | } |
| 123 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * Get the data from the pointed-to record. |
| 128 | */ |
| 129 | int |
| 130 | xfs_inobt_get_rec( |
| 131 | struct xfs_btree_cur *cur, |
| 132 | struct xfs_inobt_rec_incore *irec, |
| 133 | int *stat) |
| 134 | { |
| 135 | union xfs_btree_rec *rec; |
| 136 | int error; |
| 137 | |
| 138 | error = xfs_btree_get_rec(cur, &rec, stat); |
| 139 | if (error || *stat == 0) |
| 140 | return error; |
| 141 | |
| 142 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, irec); |
| 143 | |
| 144 | return 0; |
| 145 | } |
| 146 | |
| 147 | /* |
| 148 | * Insert a single inobt record. Cursor must already point to desired location. |
| 149 | */ |
| 150 | STATIC int |
| 151 | xfs_inobt_insert_rec( |
| 152 | struct xfs_btree_cur *cur, |
| 153 | uint16_t holemask, |
| 154 | uint8_t count, |
| 155 | int32_t freecount, |
| 156 | xfs_inofree_t free, |
| 157 | int *stat) |
| 158 | { |
| 159 | cur->bc_rec.i.ir_holemask = holemask; |
| 160 | cur->bc_rec.i.ir_count = count; |
| 161 | cur->bc_rec.i.ir_freecount = freecount; |
| 162 | cur->bc_rec.i.ir_free = free; |
| 163 | return xfs_btree_insert(cur, stat); |
| 164 | } |
| 165 | |
| 166 | /* |
| 167 | * Insert records describing a newly allocated inode chunk into the inobt. |
| 168 | */ |
| 169 | STATIC int |
| 170 | xfs_inobt_insert( |
| 171 | struct xfs_mount *mp, |
| 172 | struct xfs_trans *tp, |
| 173 | struct xfs_buf *agbp, |
| 174 | xfs_agino_t newino, |
| 175 | xfs_agino_t newlen, |
| 176 | xfs_btnum_t btnum) |
| 177 | { |
| 178 | struct xfs_btree_cur *cur; |
| 179 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 180 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 181 | xfs_agino_t thisino; |
| 182 | int i; |
| 183 | int error; |
| 184 | |
| 185 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); |
| 186 | |
| 187 | for (thisino = newino; |
| 188 | thisino < newino + newlen; |
| 189 | thisino += XFS_INODES_PER_CHUNK) { |
| 190 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); |
| 191 | if (error) { |
| 192 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 193 | return error; |
| 194 | } |
| 195 | ASSERT(i == 0); |
| 196 | |
| 197 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
| 198 | XFS_INODES_PER_CHUNK, |
| 199 | XFS_INODES_PER_CHUNK, |
| 200 | XFS_INOBT_ALL_FREE, &i); |
| 201 | if (error) { |
| 202 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 203 | return error; |
| 204 | } |
| 205 | ASSERT(i == 1); |
| 206 | } |
| 207 | |
| 208 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 209 | |
| 210 | return 0; |
| 211 | } |
| 212 | |
| 213 | /* |
| 214 | * Verify that the number of free inodes in the AGI is correct. |
| 215 | */ |
| 216 | #ifdef DEBUG |
| 217 | STATIC int |
| 218 | xfs_check_agi_freecount( |
| 219 | struct xfs_btree_cur *cur, |
| 220 | struct xfs_agi *agi) |
| 221 | { |
| 222 | if (cur->bc_nlevels == 1) { |
| 223 | xfs_inobt_rec_incore_t rec; |
| 224 | int freecount = 0; |
| 225 | int error; |
| 226 | int i; |
| 227 | |
| 228 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
| 229 | if (error) |
| 230 | return error; |
| 231 | |
| 232 | do { |
| 233 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 234 | if (error) |
| 235 | return error; |
| 236 | |
| 237 | if (i) { |
| 238 | freecount += rec.ir_freecount; |
| 239 | error = xfs_btree_increment(cur, 0, &i); |
| 240 | if (error) |
| 241 | return error; |
| 242 | } |
| 243 | } while (i == 1); |
| 244 | |
| 245 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) |
| 246 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); |
| 247 | } |
| 248 | return 0; |
| 249 | } |
| 250 | #else |
| 251 | #define xfs_check_agi_freecount(cur, agi) 0 |
| 252 | #endif |
| 253 | |
| 254 | /* |
| 255 | * Initialise a new set of inodes. When called without a transaction context |
| 256 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather |
| 257 | * than logging them (which in a transaction context puts them into the AIL |
| 258 | * for writeback rather than the xfsbufd queue). |
| 259 | */ |
| 260 | int |
| 261 | xfs_ialloc_inode_init( |
| 262 | struct xfs_mount *mp, |
| 263 | struct xfs_trans *tp, |
| 264 | struct list_head *buffer_list, |
| 265 | int icount, |
| 266 | xfs_agnumber_t agno, |
| 267 | xfs_agblock_t agbno, |
| 268 | xfs_agblock_t length, |
| 269 | unsigned int gen) |
| 270 | { |
| 271 | struct xfs_buf *fbuf; |
| 272 | struct xfs_dinode *free; |
| 273 | int nbufs, blks_per_cluster, inodes_per_cluster; |
| 274 | int version; |
| 275 | int i, j; |
| 276 | xfs_daddr_t d; |
| 277 | xfs_ino_t ino = 0; |
| 278 | |
| 279 | /* |
| 280 | * Loop over the new block(s), filling in the inodes. For small block |
| 281 | * sizes, manipulate the inodes in buffers which are multiples of the |
| 282 | * blocks size. |
| 283 | */ |
| 284 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
| 285 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; |
| 286 | nbufs = length / blks_per_cluster; |
| 287 | |
| 288 | /* |
| 289 | * Figure out what version number to use in the inodes we create. If |
| 290 | * the superblock version has caught up to the one that supports the new |
| 291 | * inode format, then use the new inode version. Otherwise use the old |
| 292 | * version so that old kernels will continue to be able to use the file |
| 293 | * system. |
| 294 | * |
| 295 | * For v3 inodes, we also need to write the inode number into the inode, |
| 296 | * so calculate the first inode number of the chunk here as |
| 297 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not |
| 298 | * across multiple filesystem blocks (such as a cluster) and so cannot |
| 299 | * be used in the cluster buffer loop below. |
| 300 | * |
| 301 | * Further, because we are writing the inode directly into the buffer |
| 302 | * and calculating a CRC on the entire inode, we have ot log the entire |
| 303 | * inode so that the entire range the CRC covers is present in the log. |
| 304 | * That means for v3 inode we log the entire buffer rather than just the |
| 305 | * inode cores. |
| 306 | */ |
| 307 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 308 | version = 3; |
| 309 | ino = XFS_AGINO_TO_INO(mp, agno, |
| 310 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); |
| 311 | |
| 312 | /* |
| 313 | * log the initialisation that is about to take place as an |
| 314 | * logical operation. This means the transaction does not |
| 315 | * need to log the physical changes to the inode buffers as log |
| 316 | * recovery will know what initialisation is actually needed. |
| 317 | * Hence we only need to log the buffers as "ordered" buffers so |
| 318 | * they track in the AIL as if they were physically logged. |
| 319 | */ |
| 320 | if (tp) |
| 321 | xfs_icreate_log(tp, agno, agbno, icount, |
| 322 | mp->m_sb.sb_inodesize, length, gen); |
| 323 | } else |
| 324 | version = 2; |
| 325 | |
| 326 | for (j = 0; j < nbufs; j++) { |
| 327 | /* |
| 328 | * Get the block. |
| 329 | */ |
| 330 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); |
| 331 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, |
| 332 | mp->m_bsize * blks_per_cluster, |
| 333 | XBF_UNMAPPED); |
| 334 | if (!fbuf) |
| 335 | return -ENOMEM; |
| 336 | |
| 337 | /* Initialize the inode buffers and log them appropriately. */ |
| 338 | fbuf->b_ops = &xfs_inode_buf_ops; |
| 339 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
| 340 | for (i = 0; i < inodes_per_cluster; i++) { |
| 341 | int ioffset = i << mp->m_sb.sb_inodelog; |
| 342 | uint isize = xfs_dinode_size(version); |
| 343 | |
| 344 | free = xfs_make_iptr(mp, fbuf, i); |
| 345 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); |
| 346 | free->di_version = version; |
| 347 | free->di_gen = cpu_to_be32(gen); |
| 348 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); |
| 349 | |
| 350 | if (version == 3) { |
| 351 | free->di_ino = cpu_to_be64(ino); |
| 352 | ino++; |
| 353 | uuid_copy(&free->di_uuid, |
| 354 | &mp->m_sb.sb_meta_uuid); |
| 355 | xfs_dinode_calc_crc(mp, free); |
| 356 | } else if (tp) { |
| 357 | /* just log the inode core */ |
| 358 | xfs_trans_log_buf(tp, fbuf, ioffset, |
| 359 | ioffset + isize - 1); |
| 360 | } |
| 361 | } |
| 362 | |
| 363 | if (tp) { |
| 364 | /* |
| 365 | * Mark the buffer as an inode allocation buffer so it |
| 366 | * sticks in AIL at the point of this allocation |
| 367 | * transaction. This ensures the they are on disk before |
| 368 | * the tail of the log can be moved past this |
| 369 | * transaction (i.e. by preventing relogging from moving |
| 370 | * it forward in the log). |
| 371 | */ |
| 372 | xfs_trans_inode_alloc_buf(tp, fbuf); |
| 373 | if (version == 3) { |
| 374 | /* |
| 375 | * Mark the buffer as ordered so that they are |
| 376 | * not physically logged in the transaction but |
| 377 | * still tracked in the AIL as part of the |
| 378 | * transaction and pin the log appropriately. |
| 379 | */ |
| 380 | xfs_trans_ordered_buf(tp, fbuf); |
| 381 | xfs_trans_log_buf(tp, fbuf, 0, |
| 382 | BBTOB(fbuf->b_length) - 1); |
| 383 | } |
| 384 | } else { |
| 385 | fbuf->b_flags |= XBF_DONE; |
| 386 | xfs_buf_delwri_queue(fbuf, buffer_list); |
| 387 | xfs_buf_relse(fbuf); |
| 388 | } |
| 389 | } |
| 390 | return 0; |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * Align startino and allocmask for a recently allocated sparse chunk such that |
| 395 | * they are fit for insertion (or merge) into the on-disk inode btrees. |
| 396 | * |
| 397 | * Background: |
| 398 | * |
| 399 | * When enabled, sparse inode support increases the inode alignment from cluster |
| 400 | * size to inode chunk size. This means that the minimum range between two |
| 401 | * non-adjacent inode records in the inobt is large enough for a full inode |
| 402 | * record. This allows for cluster sized, cluster aligned block allocation |
| 403 | * without need to worry about whether the resulting inode record overlaps with |
| 404 | * another record in the tree. Without this basic rule, we would have to deal |
| 405 | * with the consequences of overlap by potentially undoing recent allocations in |
| 406 | * the inode allocation codepath. |
| 407 | * |
| 408 | * Because of this alignment rule (which is enforced on mount), there are two |
| 409 | * inobt possibilities for newly allocated sparse chunks. One is that the |
| 410 | * aligned inode record for the chunk covers a range of inodes not already |
| 411 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The |
| 412 | * other is that a record already exists at the aligned startino that considers |
| 413 | * the newly allocated range as sparse. In the latter case, record content is |
| 414 | * merged in hope that sparse inode chunks fill to full chunks over time. |
| 415 | */ |
| 416 | STATIC void |
| 417 | xfs_align_sparse_ino( |
| 418 | struct xfs_mount *mp, |
| 419 | xfs_agino_t *startino, |
| 420 | uint16_t *allocmask) |
| 421 | { |
| 422 | xfs_agblock_t agbno; |
| 423 | xfs_agblock_t mod; |
| 424 | int offset; |
| 425 | |
| 426 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); |
| 427 | mod = agbno % mp->m_sb.sb_inoalignmt; |
| 428 | if (!mod) |
| 429 | return; |
| 430 | |
| 431 | /* calculate the inode offset and align startino */ |
| 432 | offset = mod << mp->m_sb.sb_inopblog; |
| 433 | *startino -= offset; |
| 434 | |
| 435 | /* |
| 436 | * Since startino has been aligned down, left shift allocmask such that |
| 437 | * it continues to represent the same physical inodes relative to the |
| 438 | * new startino. |
| 439 | */ |
| 440 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; |
| 441 | } |
| 442 | |
| 443 | /* |
| 444 | * Determine whether the source inode record can merge into the target. Both |
| 445 | * records must be sparse, the inode ranges must match and there must be no |
| 446 | * allocation overlap between the records. |
| 447 | */ |
| 448 | STATIC bool |
| 449 | __xfs_inobt_can_merge( |
| 450 | struct xfs_inobt_rec_incore *trec, /* tgt record */ |
| 451 | struct xfs_inobt_rec_incore *srec) /* src record */ |
| 452 | { |
| 453 | uint64_t talloc; |
| 454 | uint64_t salloc; |
| 455 | |
| 456 | /* records must cover the same inode range */ |
| 457 | if (trec->ir_startino != srec->ir_startino) |
| 458 | return false; |
| 459 | |
| 460 | /* both records must be sparse */ |
| 461 | if (!xfs_inobt_issparse(trec->ir_holemask) || |
| 462 | !xfs_inobt_issparse(srec->ir_holemask)) |
| 463 | return false; |
| 464 | |
| 465 | /* both records must track some inodes */ |
| 466 | if (!trec->ir_count || !srec->ir_count) |
| 467 | return false; |
| 468 | |
| 469 | /* can't exceed capacity of a full record */ |
| 470 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) |
| 471 | return false; |
| 472 | |
| 473 | /* verify there is no allocation overlap */ |
| 474 | talloc = xfs_inobt_irec_to_allocmask(trec); |
| 475 | salloc = xfs_inobt_irec_to_allocmask(srec); |
| 476 | if (talloc & salloc) |
| 477 | return false; |
| 478 | |
| 479 | return true; |
| 480 | } |
| 481 | |
| 482 | /* |
| 483 | * Merge the source inode record into the target. The caller must call |
| 484 | * __xfs_inobt_can_merge() to ensure the merge is valid. |
| 485 | */ |
| 486 | STATIC void |
| 487 | __xfs_inobt_rec_merge( |
| 488 | struct xfs_inobt_rec_incore *trec, /* target */ |
| 489 | struct xfs_inobt_rec_incore *srec) /* src */ |
| 490 | { |
| 491 | ASSERT(trec->ir_startino == srec->ir_startino); |
| 492 | |
| 493 | /* combine the counts */ |
| 494 | trec->ir_count += srec->ir_count; |
| 495 | trec->ir_freecount += srec->ir_freecount; |
| 496 | |
| 497 | /* |
| 498 | * Merge the holemask and free mask. For both fields, 0 bits refer to |
| 499 | * allocated inodes. We combine the allocated ranges with bitwise AND. |
| 500 | */ |
| 501 | trec->ir_holemask &= srec->ir_holemask; |
| 502 | trec->ir_free &= srec->ir_free; |
| 503 | } |
| 504 | |
| 505 | /* |
| 506 | * Insert a new sparse inode chunk into the associated inode btree. The inode |
| 507 | * record for the sparse chunk is pre-aligned to a startino that should match |
| 508 | * any pre-existing sparse inode record in the tree. This allows sparse chunks |
| 509 | * to fill over time. |
| 510 | * |
| 511 | * This function supports two modes of handling preexisting records depending on |
| 512 | * the merge flag. If merge is true, the provided record is merged with the |
| 513 | * existing record and updated in place. The merged record is returned in nrec. |
| 514 | * If merge is false, an existing record is replaced with the provided record. |
| 515 | * If no preexisting record exists, the provided record is always inserted. |
| 516 | * |
| 517 | * It is considered corruption if a merge is requested and not possible. Given |
| 518 | * the sparse inode alignment constraints, this should never happen. |
| 519 | */ |
| 520 | STATIC int |
| 521 | xfs_inobt_insert_sprec( |
| 522 | struct xfs_mount *mp, |
| 523 | struct xfs_trans *tp, |
| 524 | struct xfs_buf *agbp, |
| 525 | int btnum, |
| 526 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ |
| 527 | bool merge) /* merge or replace */ |
| 528 | { |
| 529 | struct xfs_btree_cur *cur; |
| 530 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 531 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 532 | int error; |
| 533 | int i; |
| 534 | struct xfs_inobt_rec_incore rec; |
| 535 | |
| 536 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); |
| 537 | |
| 538 | /* the new record is pre-aligned so we know where to look */ |
| 539 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); |
| 540 | if (error) |
| 541 | goto error; |
| 542 | /* if nothing there, insert a new record and return */ |
| 543 | if (i == 0) { |
| 544 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, |
| 545 | nrec->ir_count, nrec->ir_freecount, |
| 546 | nrec->ir_free, &i); |
| 547 | if (error) |
| 548 | goto error; |
| 549 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
| 550 | |
| 551 | goto out; |
| 552 | } |
| 553 | |
| 554 | /* |
| 555 | * A record exists at this startino. Merge or replace the record |
| 556 | * depending on what we've been asked to do. |
| 557 | */ |
| 558 | if (merge) { |
| 559 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 560 | if (error) |
| 561 | goto error; |
| 562 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
| 563 | XFS_WANT_CORRUPTED_GOTO(mp, |
| 564 | rec.ir_startino == nrec->ir_startino, |
| 565 | error); |
| 566 | |
| 567 | /* |
| 568 | * This should never fail. If we have coexisting records that |
| 569 | * cannot merge, something is seriously wrong. |
| 570 | */ |
| 571 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), |
| 572 | error); |
| 573 | |
| 574 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, |
| 575 | rec.ir_holemask, nrec->ir_startino, |
| 576 | nrec->ir_holemask); |
| 577 | |
| 578 | /* merge to nrec to output the updated record */ |
| 579 | __xfs_inobt_rec_merge(nrec, &rec); |
| 580 | |
| 581 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, |
| 582 | nrec->ir_holemask); |
| 583 | |
| 584 | error = xfs_inobt_rec_check_count(mp, nrec); |
| 585 | if (error) |
| 586 | goto error; |
| 587 | } |
| 588 | |
| 589 | error = xfs_inobt_update(cur, nrec); |
| 590 | if (error) |
| 591 | goto error; |
| 592 | |
| 593 | out: |
| 594 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 595 | return 0; |
| 596 | error: |
| 597 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 598 | return error; |
| 599 | } |
| 600 | |
| 601 | /* |
| 602 | * Allocate new inodes in the allocation group specified by agbp. |
| 603 | * Return 0 for success, else error code. |
| 604 | */ |
| 605 | STATIC int /* error code or 0 */ |
| 606 | xfs_ialloc_ag_alloc( |
| 607 | xfs_trans_t *tp, /* transaction pointer */ |
| 608 | xfs_buf_t *agbp, /* alloc group buffer */ |
| 609 | int *alloc) |
| 610 | { |
| 611 | xfs_agi_t *agi; /* allocation group header */ |
| 612 | xfs_alloc_arg_t args; /* allocation argument structure */ |
| 613 | xfs_agnumber_t agno; |
| 614 | int error; |
| 615 | xfs_agino_t newino; /* new first inode's number */ |
| 616 | xfs_agino_t newlen; /* new number of inodes */ |
| 617 | int isaligned = 0; /* inode allocation at stripe unit */ |
| 618 | /* boundary */ |
| 619 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
| 620 | struct xfs_inobt_rec_incore rec; |
| 621 | struct xfs_perag *pag; |
| 622 | int do_sparse = 0; |
| 623 | |
| 624 | memset(&args, 0, sizeof(args)); |
| 625 | args.tp = tp; |
| 626 | args.mp = tp->t_mountp; |
| 627 | args.fsbno = NULLFSBLOCK; |
| 628 | xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES); |
| 629 | |
| 630 | #ifdef DEBUG |
| 631 | /* randomly do sparse inode allocations */ |
| 632 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && |
| 633 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) |
| 634 | do_sparse = prandom_u32() & 1; |
| 635 | #endif |
| 636 | |
| 637 | /* |
| 638 | * Locking will ensure that we don't have two callers in here |
| 639 | * at one time. |
| 640 | */ |
| 641 | newlen = args.mp->m_ialloc_inos; |
| 642 | if (args.mp->m_maxicount && |
| 643 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
| 644 | args.mp->m_maxicount) |
| 645 | return -ENOSPC; |
| 646 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
| 647 | /* |
| 648 | * First try to allocate inodes contiguous with the last-allocated |
| 649 | * chunk of inodes. If the filesystem is striped, this will fill |
| 650 | * an entire stripe unit with inodes. |
| 651 | */ |
| 652 | agi = XFS_BUF_TO_AGI(agbp); |
| 653 | newino = be32_to_cpu(agi->agi_newino); |
| 654 | agno = be32_to_cpu(agi->agi_seqno); |
| 655 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
| 656 | args.mp->m_ialloc_blks; |
| 657 | if (do_sparse) |
| 658 | goto sparse_alloc; |
| 659 | if (likely(newino != NULLAGINO && |
| 660 | (args.agbno < be32_to_cpu(agi->agi_length)))) { |
| 661 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
| 662 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
| 663 | args.prod = 1; |
| 664 | |
| 665 | /* |
| 666 | * We need to take into account alignment here to ensure that |
| 667 | * we don't modify the free list if we fail to have an exact |
| 668 | * block. If we don't have an exact match, and every oher |
| 669 | * attempt allocation attempt fails, we'll end up cancelling |
| 670 | * a dirty transaction and shutting down. |
| 671 | * |
| 672 | * For an exact allocation, alignment must be 1, |
| 673 | * however we need to take cluster alignment into account when |
| 674 | * fixing up the freelist. Use the minalignslop field to |
| 675 | * indicate that extra blocks might be required for alignment, |
| 676 | * but not to use them in the actual exact allocation. |
| 677 | */ |
| 678 | args.alignment = 1; |
| 679 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
| 680 | |
| 681 | /* Allow space for the inode btree to split. */ |
| 682 | args.minleft = args.mp->m_in_maxlevels - 1; |
| 683 | if ((error = xfs_alloc_vextent(&args))) |
| 684 | return error; |
| 685 | |
| 686 | /* |
| 687 | * This request might have dirtied the transaction if the AG can |
| 688 | * satisfy the request, but the exact block was not available. |
| 689 | * If the allocation did fail, subsequent requests will relax |
| 690 | * the exact agbno requirement and increase the alignment |
| 691 | * instead. It is critical that the total size of the request |
| 692 | * (len + alignment + slop) does not increase from this point |
| 693 | * on, so reset minalignslop to ensure it is not included in |
| 694 | * subsequent requests. |
| 695 | */ |
| 696 | args.minalignslop = 0; |
| 697 | } |
| 698 | |
| 699 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
| 700 | /* |
| 701 | * Set the alignment for the allocation. |
| 702 | * If stripe alignment is turned on then align at stripe unit |
| 703 | * boundary. |
| 704 | * If the cluster size is smaller than a filesystem block |
| 705 | * then we're doing I/O for inodes in filesystem block size |
| 706 | * pieces, so don't need alignment anyway. |
| 707 | */ |
| 708 | isaligned = 0; |
| 709 | if (args.mp->m_sinoalign) { |
| 710 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); |
| 711 | args.alignment = args.mp->m_dalign; |
| 712 | isaligned = 1; |
| 713 | } else |
| 714 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
| 715 | /* |
| 716 | * Need to figure out where to allocate the inode blocks. |
| 717 | * Ideally they should be spaced out through the a.g. |
| 718 | * For now, just allocate blocks up front. |
| 719 | */ |
| 720 | args.agbno = be32_to_cpu(agi->agi_root); |
| 721 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
| 722 | /* |
| 723 | * Allocate a fixed-size extent of inodes. |
| 724 | */ |
| 725 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
| 726 | args.prod = 1; |
| 727 | /* |
| 728 | * Allow space for the inode btree to split. |
| 729 | */ |
| 730 | args.minleft = args.mp->m_in_maxlevels - 1; |
| 731 | if ((error = xfs_alloc_vextent(&args))) |
| 732 | return error; |
| 733 | } |
| 734 | |
| 735 | /* |
| 736 | * If stripe alignment is turned on, then try again with cluster |
| 737 | * alignment. |
| 738 | */ |
| 739 | if (isaligned && args.fsbno == NULLFSBLOCK) { |
| 740 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
| 741 | args.agbno = be32_to_cpu(agi->agi_root); |
| 742 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
| 743 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
| 744 | if ((error = xfs_alloc_vextent(&args))) |
| 745 | return error; |
| 746 | } |
| 747 | |
| 748 | /* |
| 749 | * Finally, try a sparse allocation if the filesystem supports it and |
| 750 | * the sparse allocation length is smaller than a full chunk. |
| 751 | */ |
| 752 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && |
| 753 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && |
| 754 | args.fsbno == NULLFSBLOCK) { |
| 755 | sparse_alloc: |
| 756 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
| 757 | args.agbno = be32_to_cpu(agi->agi_root); |
| 758 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
| 759 | args.alignment = args.mp->m_sb.sb_spino_align; |
| 760 | args.prod = 1; |
| 761 | |
| 762 | args.minlen = args.mp->m_ialloc_min_blks; |
| 763 | args.maxlen = args.minlen; |
| 764 | |
| 765 | /* |
| 766 | * The inode record will be aligned to full chunk size. We must |
| 767 | * prevent sparse allocation from AG boundaries that result in |
| 768 | * invalid inode records, such as records that start at agbno 0 |
| 769 | * or extend beyond the AG. |
| 770 | * |
| 771 | * Set min agbno to the first aligned, non-zero agbno and max to |
| 772 | * the last aligned agbno that is at least one full chunk from |
| 773 | * the end of the AG. |
| 774 | */ |
| 775 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; |
| 776 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, |
| 777 | args.mp->m_sb.sb_inoalignmt) - |
| 778 | args.mp->m_ialloc_blks; |
| 779 | |
| 780 | error = xfs_alloc_vextent(&args); |
| 781 | if (error) |
| 782 | return error; |
| 783 | |
| 784 | newlen = args.len << args.mp->m_sb.sb_inopblog; |
| 785 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
| 786 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
| 787 | } |
| 788 | |
| 789 | if (args.fsbno == NULLFSBLOCK) { |
| 790 | *alloc = 0; |
| 791 | return 0; |
| 792 | } |
| 793 | ASSERT(args.len == args.minlen); |
| 794 | |
| 795 | /* |
| 796 | * Stamp and write the inode buffers. |
| 797 | * |
| 798 | * Seed the new inode cluster with a random generation number. This |
| 799 | * prevents short-term reuse of generation numbers if a chunk is |
| 800 | * freed and then immediately reallocated. We use random numbers |
| 801 | * rather than a linear progression to prevent the next generation |
| 802 | * number from being easily guessable. |
| 803 | */ |
| 804 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
| 805 | args.agbno, args.len, prandom_u32()); |
| 806 | |
| 807 | if (error) |
| 808 | return error; |
| 809 | /* |
| 810 | * Convert the results. |
| 811 | */ |
| 812 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); |
| 813 | |
| 814 | if (xfs_inobt_issparse(~allocmask)) { |
| 815 | /* |
| 816 | * We've allocated a sparse chunk. Align the startino and mask. |
| 817 | */ |
| 818 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); |
| 819 | |
| 820 | rec.ir_startino = newino; |
| 821 | rec.ir_holemask = ~allocmask; |
| 822 | rec.ir_count = newlen; |
| 823 | rec.ir_freecount = newlen; |
| 824 | rec.ir_free = XFS_INOBT_ALL_FREE; |
| 825 | |
| 826 | /* |
| 827 | * Insert the sparse record into the inobt and allow for a merge |
| 828 | * if necessary. If a merge does occur, rec is updated to the |
| 829 | * merged record. |
| 830 | */ |
| 831 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, |
| 832 | &rec, true); |
| 833 | if (error == -EFSCORRUPTED) { |
| 834 | xfs_alert(args.mp, |
| 835 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", |
| 836 | XFS_AGINO_TO_INO(args.mp, agno, |
| 837 | rec.ir_startino), |
| 838 | rec.ir_holemask, rec.ir_count); |
| 839 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); |
| 840 | } |
| 841 | if (error) |
| 842 | return error; |
| 843 | |
| 844 | /* |
| 845 | * We can't merge the part we've just allocated as for the inobt |
| 846 | * due to finobt semantics. The original record may or may not |
| 847 | * exist independent of whether physical inodes exist in this |
| 848 | * sparse chunk. |
| 849 | * |
| 850 | * We must update the finobt record based on the inobt record. |
| 851 | * rec contains the fully merged and up to date inobt record |
| 852 | * from the previous call. Set merge false to replace any |
| 853 | * existing record with this one. |
| 854 | */ |
| 855 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { |
| 856 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, |
| 857 | XFS_BTNUM_FINO, &rec, |
| 858 | false); |
| 859 | if (error) |
| 860 | return error; |
| 861 | } |
| 862 | } else { |
| 863 | /* full chunk - insert new records to both btrees */ |
| 864 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, |
| 865 | XFS_BTNUM_INO); |
| 866 | if (error) |
| 867 | return error; |
| 868 | |
| 869 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { |
| 870 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, |
| 871 | newlen, XFS_BTNUM_FINO); |
| 872 | if (error) |
| 873 | return error; |
| 874 | } |
| 875 | } |
| 876 | |
| 877 | /* |
| 878 | * Update AGI counts and newino. |
| 879 | */ |
| 880 | be32_add_cpu(&agi->agi_count, newlen); |
| 881 | be32_add_cpu(&agi->agi_freecount, newlen); |
| 882 | pag = xfs_perag_get(args.mp, agno); |
| 883 | pag->pagi_freecount += newlen; |
| 884 | xfs_perag_put(pag); |
| 885 | agi->agi_newino = cpu_to_be32(newino); |
| 886 | |
| 887 | /* |
| 888 | * Log allocation group header fields |
| 889 | */ |
| 890 | xfs_ialloc_log_agi(tp, agbp, |
| 891 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); |
| 892 | /* |
| 893 | * Modify/log superblock values for inode count and inode free count. |
| 894 | */ |
| 895 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); |
| 896 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); |
| 897 | *alloc = 1; |
| 898 | return 0; |
| 899 | } |
| 900 | |
| 901 | STATIC xfs_agnumber_t |
| 902 | xfs_ialloc_next_ag( |
| 903 | xfs_mount_t *mp) |
| 904 | { |
| 905 | xfs_agnumber_t agno; |
| 906 | |
| 907 | spin_lock(&mp->m_agirotor_lock); |
| 908 | agno = mp->m_agirotor; |
| 909 | if (++mp->m_agirotor >= mp->m_maxagi) |
| 910 | mp->m_agirotor = 0; |
| 911 | spin_unlock(&mp->m_agirotor_lock); |
| 912 | |
| 913 | return agno; |
| 914 | } |
| 915 | |
| 916 | /* |
| 917 | * Select an allocation group to look for a free inode in, based on the parent |
| 918 | * inode and the mode. Return the allocation group buffer. |
| 919 | */ |
| 920 | STATIC xfs_agnumber_t |
| 921 | xfs_ialloc_ag_select( |
| 922 | xfs_trans_t *tp, /* transaction pointer */ |
| 923 | xfs_ino_t parent, /* parent directory inode number */ |
| 924 | umode_t mode, /* bits set to indicate file type */ |
| 925 | int okalloc) /* ok to allocate more space */ |
| 926 | { |
| 927 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
| 928 | xfs_agnumber_t agno; /* current ag number */ |
| 929 | int flags; /* alloc buffer locking flags */ |
| 930 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ |
| 931 | xfs_extlen_t longest = 0; /* longest extent available */ |
| 932 | xfs_mount_t *mp; /* mount point structure */ |
| 933 | int needspace; /* file mode implies space allocated */ |
| 934 | xfs_perag_t *pag; /* per allocation group data */ |
| 935 | xfs_agnumber_t pagno; /* parent (starting) ag number */ |
| 936 | int error; |
| 937 | |
| 938 | /* |
| 939 | * Files of these types need at least one block if length > 0 |
| 940 | * (and they won't fit in the inode, but that's hard to figure out). |
| 941 | */ |
| 942 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); |
| 943 | mp = tp->t_mountp; |
| 944 | agcount = mp->m_maxagi; |
| 945 | if (S_ISDIR(mode)) |
| 946 | pagno = xfs_ialloc_next_ag(mp); |
| 947 | else { |
| 948 | pagno = XFS_INO_TO_AGNO(mp, parent); |
| 949 | if (pagno >= agcount) |
| 950 | pagno = 0; |
| 951 | } |
| 952 | |
| 953 | ASSERT(pagno < agcount); |
| 954 | |
| 955 | /* |
| 956 | * Loop through allocation groups, looking for one with a little |
| 957 | * free space in it. Note we don't look for free inodes, exactly. |
| 958 | * Instead, we include whether there is a need to allocate inodes |
| 959 | * to mean that blocks must be allocated for them, |
| 960 | * if none are currently free. |
| 961 | */ |
| 962 | agno = pagno; |
| 963 | flags = XFS_ALLOC_FLAG_TRYLOCK; |
| 964 | for (;;) { |
| 965 | pag = xfs_perag_get(mp, agno); |
| 966 | if (!pag->pagi_inodeok) { |
| 967 | xfs_ialloc_next_ag(mp); |
| 968 | goto nextag; |
| 969 | } |
| 970 | |
| 971 | if (!pag->pagi_init) { |
| 972 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
| 973 | if (error) |
| 974 | goto nextag; |
| 975 | } |
| 976 | |
| 977 | if (pag->pagi_freecount) { |
| 978 | xfs_perag_put(pag); |
| 979 | return agno; |
| 980 | } |
| 981 | |
| 982 | if (!okalloc) |
| 983 | goto nextag; |
| 984 | |
| 985 | if (!pag->pagf_init) { |
| 986 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); |
| 987 | if (error) |
| 988 | goto nextag; |
| 989 | } |
| 990 | |
| 991 | /* |
| 992 | * Check that there is enough free space for the file plus a |
| 993 | * chunk of inodes if we need to allocate some. If this is the |
| 994 | * first pass across the AGs, take into account the potential |
| 995 | * space needed for alignment of inode chunks when checking the |
| 996 | * longest contiguous free space in the AG - this prevents us |
| 997 | * from getting ENOSPC because we have free space larger than |
| 998 | * m_ialloc_blks but alignment constraints prevent us from using |
| 999 | * it. |
| 1000 | * |
| 1001 | * If we can't find an AG with space for full alignment slack to |
| 1002 | * be taken into account, we must be near ENOSPC in all AGs. |
| 1003 | * Hence we don't include alignment for the second pass and so |
| 1004 | * if we fail allocation due to alignment issues then it is most |
| 1005 | * likely a real ENOSPC condition. |
| 1006 | */ |
| 1007 | ineed = mp->m_ialloc_min_blks; |
| 1008 | if (flags && ineed > 1) |
| 1009 | ineed += xfs_ialloc_cluster_alignment(mp); |
| 1010 | longest = pag->pagf_longest; |
| 1011 | if (!longest) |
| 1012 | longest = pag->pagf_flcount > 0; |
| 1013 | |
| 1014 | if (pag->pagf_freeblks >= needspace + ineed && |
| 1015 | longest >= ineed) { |
| 1016 | xfs_perag_put(pag); |
| 1017 | return agno; |
| 1018 | } |
| 1019 | nextag: |
| 1020 | xfs_perag_put(pag); |
| 1021 | /* |
| 1022 | * No point in iterating over the rest, if we're shutting |
| 1023 | * down. |
| 1024 | */ |
| 1025 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 1026 | return NULLAGNUMBER; |
| 1027 | agno++; |
| 1028 | if (agno >= agcount) |
| 1029 | agno = 0; |
| 1030 | if (agno == pagno) { |
| 1031 | if (flags == 0) |
| 1032 | return NULLAGNUMBER; |
| 1033 | flags = 0; |
| 1034 | } |
| 1035 | } |
| 1036 | } |
| 1037 | |
| 1038 | /* |
| 1039 | * Try to retrieve the next record to the left/right from the current one. |
| 1040 | */ |
| 1041 | STATIC int |
| 1042 | xfs_ialloc_next_rec( |
| 1043 | struct xfs_btree_cur *cur, |
| 1044 | xfs_inobt_rec_incore_t *rec, |
| 1045 | int *done, |
| 1046 | int left) |
| 1047 | { |
| 1048 | int error; |
| 1049 | int i; |
| 1050 | |
| 1051 | if (left) |
| 1052 | error = xfs_btree_decrement(cur, 0, &i); |
| 1053 | else |
| 1054 | error = xfs_btree_increment(cur, 0, &i); |
| 1055 | |
| 1056 | if (error) |
| 1057 | return error; |
| 1058 | *done = !i; |
| 1059 | if (i) { |
| 1060 | error = xfs_inobt_get_rec(cur, rec, &i); |
| 1061 | if (error) |
| 1062 | return error; |
| 1063 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1064 | } |
| 1065 | |
| 1066 | return 0; |
| 1067 | } |
| 1068 | |
| 1069 | STATIC int |
| 1070 | xfs_ialloc_get_rec( |
| 1071 | struct xfs_btree_cur *cur, |
| 1072 | xfs_agino_t agino, |
| 1073 | xfs_inobt_rec_incore_t *rec, |
| 1074 | int *done) |
| 1075 | { |
| 1076 | int error; |
| 1077 | int i; |
| 1078 | |
| 1079 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); |
| 1080 | if (error) |
| 1081 | return error; |
| 1082 | *done = !i; |
| 1083 | if (i) { |
| 1084 | error = xfs_inobt_get_rec(cur, rec, &i); |
| 1085 | if (error) |
| 1086 | return error; |
| 1087 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1088 | } |
| 1089 | |
| 1090 | return 0; |
| 1091 | } |
| 1092 | |
| 1093 | /* |
| 1094 | * Return the offset of the first free inode in the record. If the inode chunk |
| 1095 | * is sparsely allocated, we convert the record holemask to inode granularity |
| 1096 | * and mask off the unallocated regions from the inode free mask. |
| 1097 | */ |
| 1098 | STATIC int |
| 1099 | xfs_inobt_first_free_inode( |
| 1100 | struct xfs_inobt_rec_incore *rec) |
| 1101 | { |
| 1102 | xfs_inofree_t realfree; |
| 1103 | |
| 1104 | /* if there are no holes, return the first available offset */ |
| 1105 | if (!xfs_inobt_issparse(rec->ir_holemask)) |
| 1106 | return xfs_lowbit64(rec->ir_free); |
| 1107 | |
| 1108 | realfree = xfs_inobt_irec_to_allocmask(rec); |
| 1109 | realfree &= rec->ir_free; |
| 1110 | |
| 1111 | return xfs_lowbit64(realfree); |
| 1112 | } |
| 1113 | |
| 1114 | /* |
| 1115 | * Allocate an inode using the inobt-only algorithm. |
| 1116 | */ |
| 1117 | STATIC int |
| 1118 | xfs_dialloc_ag_inobt( |
| 1119 | struct xfs_trans *tp, |
| 1120 | struct xfs_buf *agbp, |
| 1121 | xfs_ino_t parent, |
| 1122 | xfs_ino_t *inop) |
| 1123 | { |
| 1124 | struct xfs_mount *mp = tp->t_mountp; |
| 1125 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 1126 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 1127 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); |
| 1128 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); |
| 1129 | struct xfs_perag *pag; |
| 1130 | struct xfs_btree_cur *cur, *tcur; |
| 1131 | struct xfs_inobt_rec_incore rec, trec; |
| 1132 | xfs_ino_t ino; |
| 1133 | int error; |
| 1134 | int offset; |
| 1135 | int i, j; |
| 1136 | |
| 1137 | pag = xfs_perag_get(mp, agno); |
| 1138 | |
| 1139 | ASSERT(pag->pagi_init); |
| 1140 | ASSERT(pag->pagi_inodeok); |
| 1141 | ASSERT(pag->pagi_freecount > 0); |
| 1142 | |
| 1143 | restart_pagno: |
| 1144 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
| 1145 | /* |
| 1146 | * If pagino is 0 (this is the root inode allocation) use newino. |
| 1147 | * This must work because we've just allocated some. |
| 1148 | */ |
| 1149 | if (!pagino) |
| 1150 | pagino = be32_to_cpu(agi->agi_newino); |
| 1151 | |
| 1152 | error = xfs_check_agi_freecount(cur, agi); |
| 1153 | if (error) |
| 1154 | goto error0; |
| 1155 | |
| 1156 | /* |
| 1157 | * If in the same AG as the parent, try to get near the parent. |
| 1158 | */ |
| 1159 | if (pagno == agno) { |
| 1160 | int doneleft; /* done, to the left */ |
| 1161 | int doneright; /* done, to the right */ |
| 1162 | int searchdistance = 10; |
| 1163 | |
| 1164 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
| 1165 | if (error) |
| 1166 | goto error0; |
| 1167 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1168 | |
| 1169 | error = xfs_inobt_get_rec(cur, &rec, &j); |
| 1170 | if (error) |
| 1171 | goto error0; |
| 1172 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
| 1173 | |
| 1174 | if (rec.ir_freecount > 0) { |
| 1175 | /* |
| 1176 | * Found a free inode in the same chunk |
| 1177 | * as the parent, done. |
| 1178 | */ |
| 1179 | goto alloc_inode; |
| 1180 | } |
| 1181 | |
| 1182 | |
| 1183 | /* |
| 1184 | * In the same AG as parent, but parent's chunk is full. |
| 1185 | */ |
| 1186 | |
| 1187 | /* duplicate the cursor, search left & right simultaneously */ |
| 1188 | error = xfs_btree_dup_cursor(cur, &tcur); |
| 1189 | if (error) |
| 1190 | goto error0; |
| 1191 | |
| 1192 | /* |
| 1193 | * Skip to last blocks looked up if same parent inode. |
| 1194 | */ |
| 1195 | if (pagino != NULLAGINO && |
| 1196 | pag->pagl_pagino == pagino && |
| 1197 | pag->pagl_leftrec != NULLAGINO && |
| 1198 | pag->pagl_rightrec != NULLAGINO) { |
| 1199 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, |
| 1200 | &trec, &doneleft); |
| 1201 | if (error) |
| 1202 | goto error1; |
| 1203 | |
| 1204 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
| 1205 | &rec, &doneright); |
| 1206 | if (error) |
| 1207 | goto error1; |
| 1208 | } else { |
| 1209 | /* search left with tcur, back up 1 record */ |
| 1210 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); |
| 1211 | if (error) |
| 1212 | goto error1; |
| 1213 | |
| 1214 | /* search right with cur, go forward 1 record. */ |
| 1215 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); |
| 1216 | if (error) |
| 1217 | goto error1; |
| 1218 | } |
| 1219 | |
| 1220 | /* |
| 1221 | * Loop until we find an inode chunk with a free inode. |
| 1222 | */ |
| 1223 | while (!doneleft || !doneright) { |
| 1224 | int useleft; /* using left inode chunk this time */ |
| 1225 | |
| 1226 | if (!--searchdistance) { |
| 1227 | /* |
| 1228 | * Not in range - save last search |
| 1229 | * location and allocate a new inode |
| 1230 | */ |
| 1231 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
| 1232 | pag->pagl_leftrec = trec.ir_startino; |
| 1233 | pag->pagl_rightrec = rec.ir_startino; |
| 1234 | pag->pagl_pagino = pagino; |
| 1235 | goto newino; |
| 1236 | } |
| 1237 | |
| 1238 | /* figure out the closer block if both are valid. */ |
| 1239 | if (!doneleft && !doneright) { |
| 1240 | useleft = pagino - |
| 1241 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < |
| 1242 | rec.ir_startino - pagino; |
| 1243 | } else { |
| 1244 | useleft = !doneleft; |
| 1245 | } |
| 1246 | |
| 1247 | /* free inodes to the left? */ |
| 1248 | if (useleft && trec.ir_freecount) { |
| 1249 | rec = trec; |
| 1250 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 1251 | cur = tcur; |
| 1252 | |
| 1253 | pag->pagl_leftrec = trec.ir_startino; |
| 1254 | pag->pagl_rightrec = rec.ir_startino; |
| 1255 | pag->pagl_pagino = pagino; |
| 1256 | goto alloc_inode; |
| 1257 | } |
| 1258 | |
| 1259 | /* free inodes to the right? */ |
| 1260 | if (!useleft && rec.ir_freecount) { |
| 1261 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
| 1262 | |
| 1263 | pag->pagl_leftrec = trec.ir_startino; |
| 1264 | pag->pagl_rightrec = rec.ir_startino; |
| 1265 | pag->pagl_pagino = pagino; |
| 1266 | goto alloc_inode; |
| 1267 | } |
| 1268 | |
| 1269 | /* get next record to check */ |
| 1270 | if (useleft) { |
| 1271 | error = xfs_ialloc_next_rec(tcur, &trec, |
| 1272 | &doneleft, 1); |
| 1273 | } else { |
| 1274 | error = xfs_ialloc_next_rec(cur, &rec, |
| 1275 | &doneright, 0); |
| 1276 | } |
| 1277 | if (error) |
| 1278 | goto error1; |
| 1279 | } |
| 1280 | |
| 1281 | /* |
| 1282 | * We've reached the end of the btree. because |
| 1283 | * we are only searching a small chunk of the |
| 1284 | * btree each search, there is obviously free |
| 1285 | * inodes closer to the parent inode than we |
| 1286 | * are now. restart the search again. |
| 1287 | */ |
| 1288 | pag->pagl_pagino = NULLAGINO; |
| 1289 | pag->pagl_leftrec = NULLAGINO; |
| 1290 | pag->pagl_rightrec = NULLAGINO; |
| 1291 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
| 1292 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 1293 | goto restart_pagno; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * In a different AG from the parent. |
| 1298 | * See if the most recently allocated block has any free. |
| 1299 | */ |
| 1300 | newino: |
| 1301 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
| 1302 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
| 1303 | XFS_LOOKUP_EQ, &i); |
| 1304 | if (error) |
| 1305 | goto error0; |
| 1306 | |
| 1307 | if (i == 1) { |
| 1308 | error = xfs_inobt_get_rec(cur, &rec, &j); |
| 1309 | if (error) |
| 1310 | goto error0; |
| 1311 | |
| 1312 | if (j == 1 && rec.ir_freecount > 0) { |
| 1313 | /* |
| 1314 | * The last chunk allocated in the group |
| 1315 | * still has a free inode. |
| 1316 | */ |
| 1317 | goto alloc_inode; |
| 1318 | } |
| 1319 | } |
| 1320 | } |
| 1321 | |
| 1322 | /* |
| 1323 | * None left in the last group, search the whole AG |
| 1324 | */ |
| 1325 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
| 1326 | if (error) |
| 1327 | goto error0; |
| 1328 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1329 | |
| 1330 | for (;;) { |
| 1331 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 1332 | if (error) |
| 1333 | goto error0; |
| 1334 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1335 | if (rec.ir_freecount > 0) |
| 1336 | break; |
| 1337 | error = xfs_btree_increment(cur, 0, &i); |
| 1338 | if (error) |
| 1339 | goto error0; |
| 1340 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1341 | } |
| 1342 | |
| 1343 | alloc_inode: |
| 1344 | offset = xfs_inobt_first_free_inode(&rec); |
| 1345 | ASSERT(offset >= 0); |
| 1346 | ASSERT(offset < XFS_INODES_PER_CHUNK); |
| 1347 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % |
| 1348 | XFS_INODES_PER_CHUNK) == 0); |
| 1349 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); |
| 1350 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
| 1351 | rec.ir_freecount--; |
| 1352 | error = xfs_inobt_update(cur, &rec); |
| 1353 | if (error) |
| 1354 | goto error0; |
| 1355 | be32_add_cpu(&agi->agi_freecount, -1); |
| 1356 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
| 1357 | pag->pagi_freecount--; |
| 1358 | |
| 1359 | error = xfs_check_agi_freecount(cur, agi); |
| 1360 | if (error) |
| 1361 | goto error0; |
| 1362 | |
| 1363 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 1364 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); |
| 1365 | xfs_perag_put(pag); |
| 1366 | *inop = ino; |
| 1367 | return 0; |
| 1368 | error1: |
| 1369 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); |
| 1370 | error0: |
| 1371 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 1372 | xfs_perag_put(pag); |
| 1373 | return error; |
| 1374 | } |
| 1375 | |
| 1376 | /* |
| 1377 | * Use the free inode btree to allocate an inode based on distance from the |
| 1378 | * parent. Note that the provided cursor may be deleted and replaced. |
| 1379 | */ |
| 1380 | STATIC int |
| 1381 | xfs_dialloc_ag_finobt_near( |
| 1382 | xfs_agino_t pagino, |
| 1383 | struct xfs_btree_cur **ocur, |
| 1384 | struct xfs_inobt_rec_incore *rec) |
| 1385 | { |
| 1386 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ |
| 1387 | struct xfs_btree_cur *rcur; /* right search cursor */ |
| 1388 | struct xfs_inobt_rec_incore rrec; |
| 1389 | int error; |
| 1390 | int i, j; |
| 1391 | |
| 1392 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); |
| 1393 | if (error) |
| 1394 | return error; |
| 1395 | |
| 1396 | if (i == 1) { |
| 1397 | error = xfs_inobt_get_rec(lcur, rec, &i); |
| 1398 | if (error) |
| 1399 | return error; |
| 1400 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
| 1401 | |
| 1402 | /* |
| 1403 | * See if we've landed in the parent inode record. The finobt |
| 1404 | * only tracks chunks with at least one free inode, so record |
| 1405 | * existence is enough. |
| 1406 | */ |
| 1407 | if (pagino >= rec->ir_startino && |
| 1408 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) |
| 1409 | return 0; |
| 1410 | } |
| 1411 | |
| 1412 | error = xfs_btree_dup_cursor(lcur, &rcur); |
| 1413 | if (error) |
| 1414 | return error; |
| 1415 | |
| 1416 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); |
| 1417 | if (error) |
| 1418 | goto error_rcur; |
| 1419 | if (j == 1) { |
| 1420 | error = xfs_inobt_get_rec(rcur, &rrec, &j); |
| 1421 | if (error) |
| 1422 | goto error_rcur; |
| 1423 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
| 1424 | } |
| 1425 | |
| 1426 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
| 1427 | if (i == 1 && j == 1) { |
| 1428 | /* |
| 1429 | * Both the left and right records are valid. Choose the closer |
| 1430 | * inode chunk to the target. |
| 1431 | */ |
| 1432 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > |
| 1433 | (rrec.ir_startino - pagino)) { |
| 1434 | *rec = rrec; |
| 1435 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); |
| 1436 | *ocur = rcur; |
| 1437 | } else { |
| 1438 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); |
| 1439 | } |
| 1440 | } else if (j == 1) { |
| 1441 | /* only the right record is valid */ |
| 1442 | *rec = rrec; |
| 1443 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); |
| 1444 | *ocur = rcur; |
| 1445 | } else if (i == 1) { |
| 1446 | /* only the left record is valid */ |
| 1447 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); |
| 1448 | } |
| 1449 | |
| 1450 | return 0; |
| 1451 | |
| 1452 | error_rcur: |
| 1453 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); |
| 1454 | return error; |
| 1455 | } |
| 1456 | |
| 1457 | /* |
| 1458 | * Use the free inode btree to find a free inode based on a newino hint. If |
| 1459 | * the hint is NULL, find the first free inode in the AG. |
| 1460 | */ |
| 1461 | STATIC int |
| 1462 | xfs_dialloc_ag_finobt_newino( |
| 1463 | struct xfs_agi *agi, |
| 1464 | struct xfs_btree_cur *cur, |
| 1465 | struct xfs_inobt_rec_incore *rec) |
| 1466 | { |
| 1467 | int error; |
| 1468 | int i; |
| 1469 | |
| 1470 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
| 1471 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
| 1472 | XFS_LOOKUP_EQ, &i); |
| 1473 | if (error) |
| 1474 | return error; |
| 1475 | if (i == 1) { |
| 1476 | error = xfs_inobt_get_rec(cur, rec, &i); |
| 1477 | if (error) |
| 1478 | return error; |
| 1479 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1480 | return 0; |
| 1481 | } |
| 1482 | } |
| 1483 | |
| 1484 | /* |
| 1485 | * Find the first inode available in the AG. |
| 1486 | */ |
| 1487 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
| 1488 | if (error) |
| 1489 | return error; |
| 1490 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1491 | |
| 1492 | error = xfs_inobt_get_rec(cur, rec, &i); |
| 1493 | if (error) |
| 1494 | return error; |
| 1495 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1496 | |
| 1497 | return 0; |
| 1498 | } |
| 1499 | |
| 1500 | /* |
| 1501 | * Update the inobt based on a modification made to the finobt. Also ensure that |
| 1502 | * the records from both trees are equivalent post-modification. |
| 1503 | */ |
| 1504 | STATIC int |
| 1505 | xfs_dialloc_ag_update_inobt( |
| 1506 | struct xfs_btree_cur *cur, /* inobt cursor */ |
| 1507 | struct xfs_inobt_rec_incore *frec, /* finobt record */ |
| 1508 | int offset) /* inode offset */ |
| 1509 | { |
| 1510 | struct xfs_inobt_rec_incore rec; |
| 1511 | int error; |
| 1512 | int i; |
| 1513 | |
| 1514 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); |
| 1515 | if (error) |
| 1516 | return error; |
| 1517 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1518 | |
| 1519 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 1520 | if (error) |
| 1521 | return error; |
| 1522 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
| 1523 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
| 1524 | XFS_INODES_PER_CHUNK) == 0); |
| 1525 | |
| 1526 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
| 1527 | rec.ir_freecount--; |
| 1528 | |
| 1529 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
| 1530 | (rec.ir_freecount == frec->ir_freecount)); |
| 1531 | |
| 1532 | return xfs_inobt_update(cur, &rec); |
| 1533 | } |
| 1534 | |
| 1535 | /* |
| 1536 | * Allocate an inode using the free inode btree, if available. Otherwise, fall |
| 1537 | * back to the inobt search algorithm. |
| 1538 | * |
| 1539 | * The caller selected an AG for us, and made sure that free inodes are |
| 1540 | * available. |
| 1541 | */ |
| 1542 | STATIC int |
| 1543 | xfs_dialloc_ag( |
| 1544 | struct xfs_trans *tp, |
| 1545 | struct xfs_buf *agbp, |
| 1546 | xfs_ino_t parent, |
| 1547 | xfs_ino_t *inop) |
| 1548 | { |
| 1549 | struct xfs_mount *mp = tp->t_mountp; |
| 1550 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 1551 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 1552 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); |
| 1553 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); |
| 1554 | struct xfs_perag *pag; |
| 1555 | struct xfs_btree_cur *cur; /* finobt cursor */ |
| 1556 | struct xfs_btree_cur *icur; /* inobt cursor */ |
| 1557 | struct xfs_inobt_rec_incore rec; |
| 1558 | xfs_ino_t ino; |
| 1559 | int error; |
| 1560 | int offset; |
| 1561 | int i; |
| 1562 | |
| 1563 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) |
| 1564 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); |
| 1565 | |
| 1566 | pag = xfs_perag_get(mp, agno); |
| 1567 | |
| 1568 | /* |
| 1569 | * If pagino is 0 (this is the root inode allocation) use newino. |
| 1570 | * This must work because we've just allocated some. |
| 1571 | */ |
| 1572 | if (!pagino) |
| 1573 | pagino = be32_to_cpu(agi->agi_newino); |
| 1574 | |
| 1575 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); |
| 1576 | |
| 1577 | error = xfs_check_agi_freecount(cur, agi); |
| 1578 | if (error) |
| 1579 | goto error_cur; |
| 1580 | |
| 1581 | /* |
| 1582 | * The search algorithm depends on whether we're in the same AG as the |
| 1583 | * parent. If so, find the closest available inode to the parent. If |
| 1584 | * not, consider the agi hint or find the first free inode in the AG. |
| 1585 | */ |
| 1586 | if (agno == pagno) |
| 1587 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); |
| 1588 | else |
| 1589 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); |
| 1590 | if (error) |
| 1591 | goto error_cur; |
| 1592 | |
| 1593 | offset = xfs_inobt_first_free_inode(&rec); |
| 1594 | ASSERT(offset >= 0); |
| 1595 | ASSERT(offset < XFS_INODES_PER_CHUNK); |
| 1596 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % |
| 1597 | XFS_INODES_PER_CHUNK) == 0); |
| 1598 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); |
| 1599 | |
| 1600 | /* |
| 1601 | * Modify or remove the finobt record. |
| 1602 | */ |
| 1603 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
| 1604 | rec.ir_freecount--; |
| 1605 | if (rec.ir_freecount) |
| 1606 | error = xfs_inobt_update(cur, &rec); |
| 1607 | else |
| 1608 | error = xfs_btree_delete(cur, &i); |
| 1609 | if (error) |
| 1610 | goto error_cur; |
| 1611 | |
| 1612 | /* |
| 1613 | * The finobt has now been updated appropriately. We haven't updated the |
| 1614 | * agi and superblock yet, so we can create an inobt cursor and validate |
| 1615 | * the original freecount. If all is well, make the equivalent update to |
| 1616 | * the inobt using the finobt record and offset information. |
| 1617 | */ |
| 1618 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
| 1619 | |
| 1620 | error = xfs_check_agi_freecount(icur, agi); |
| 1621 | if (error) |
| 1622 | goto error_icur; |
| 1623 | |
| 1624 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); |
| 1625 | if (error) |
| 1626 | goto error_icur; |
| 1627 | |
| 1628 | /* |
| 1629 | * Both trees have now been updated. We must update the perag and |
| 1630 | * superblock before we can check the freecount for each btree. |
| 1631 | */ |
| 1632 | be32_add_cpu(&agi->agi_freecount, -1); |
| 1633 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
| 1634 | pag->pagi_freecount--; |
| 1635 | |
| 1636 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); |
| 1637 | |
| 1638 | error = xfs_check_agi_freecount(icur, agi); |
| 1639 | if (error) |
| 1640 | goto error_icur; |
| 1641 | error = xfs_check_agi_freecount(cur, agi); |
| 1642 | if (error) |
| 1643 | goto error_icur; |
| 1644 | |
| 1645 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); |
| 1646 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 1647 | xfs_perag_put(pag); |
| 1648 | *inop = ino; |
| 1649 | return 0; |
| 1650 | |
| 1651 | error_icur: |
| 1652 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); |
| 1653 | error_cur: |
| 1654 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 1655 | xfs_perag_put(pag); |
| 1656 | return error; |
| 1657 | } |
| 1658 | |
| 1659 | /* |
| 1660 | * Allocate an inode on disk. |
| 1661 | * |
| 1662 | * Mode is used to tell whether the new inode will need space, and whether it |
| 1663 | * is a directory. |
| 1664 | * |
| 1665 | * This function is designed to be called twice if it has to do an allocation |
| 1666 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. |
| 1667 | * If an inode is available without having to performn an allocation, an inode |
| 1668 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
| 1669 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. |
| 1670 | * The caller should then commit the current transaction, allocate a |
| 1671 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
| 1672 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI |
| 1673 | * buffer is locked across the two calls, the second call is guaranteed to have |
| 1674 | * a free inode available. |
| 1675 | * |
| 1676 | * Once we successfully pick an inode its number is returned and the on-disk |
| 1677 | * data structures are updated. The inode itself is not read in, since doing so |
| 1678 | * would break ordering constraints with xfs_reclaim. |
| 1679 | */ |
| 1680 | int |
| 1681 | xfs_dialloc( |
| 1682 | struct xfs_trans *tp, |
| 1683 | xfs_ino_t parent, |
| 1684 | umode_t mode, |
| 1685 | int okalloc, |
| 1686 | struct xfs_buf **IO_agbp, |
| 1687 | xfs_ino_t *inop) |
| 1688 | { |
| 1689 | struct xfs_mount *mp = tp->t_mountp; |
| 1690 | struct xfs_buf *agbp; |
| 1691 | xfs_agnumber_t agno; |
| 1692 | int error; |
| 1693 | int ialloced; |
| 1694 | int noroom = 0; |
| 1695 | xfs_agnumber_t start_agno; |
| 1696 | struct xfs_perag *pag; |
| 1697 | |
| 1698 | if (*IO_agbp) { |
| 1699 | /* |
| 1700 | * If the caller passes in a pointer to the AGI buffer, |
| 1701 | * continue where we left off before. In this case, we |
| 1702 | * know that the allocation group has free inodes. |
| 1703 | */ |
| 1704 | agbp = *IO_agbp; |
| 1705 | goto out_alloc; |
| 1706 | } |
| 1707 | |
| 1708 | /* |
| 1709 | * We do not have an agbp, so select an initial allocation |
| 1710 | * group for inode allocation. |
| 1711 | */ |
| 1712 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
| 1713 | if (start_agno == NULLAGNUMBER) { |
| 1714 | *inop = NULLFSINO; |
| 1715 | return 0; |
| 1716 | } |
| 1717 | |
| 1718 | /* |
| 1719 | * If we have already hit the ceiling of inode blocks then clear |
| 1720 | * okalloc so we scan all available agi structures for a free |
| 1721 | * inode. |
| 1722 | * |
| 1723 | * Read rough value of mp->m_icount by percpu_counter_read_positive, |
| 1724 | * which will sacrifice the preciseness but improve the performance. |
| 1725 | */ |
| 1726 | if (mp->m_maxicount && |
| 1727 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
| 1728 | > mp->m_maxicount) { |
| 1729 | noroom = 1; |
| 1730 | okalloc = 0; |
| 1731 | } |
| 1732 | |
| 1733 | /* |
| 1734 | * Loop until we find an allocation group that either has free inodes |
| 1735 | * or in which we can allocate some inodes. Iterate through the |
| 1736 | * allocation groups upward, wrapping at the end. |
| 1737 | */ |
| 1738 | agno = start_agno; |
| 1739 | for (;;) { |
| 1740 | pag = xfs_perag_get(mp, agno); |
| 1741 | if (!pag->pagi_inodeok) { |
| 1742 | xfs_ialloc_next_ag(mp); |
| 1743 | goto nextag; |
| 1744 | } |
| 1745 | |
| 1746 | if (!pag->pagi_init) { |
| 1747 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
| 1748 | if (error) |
| 1749 | goto out_error; |
| 1750 | } |
| 1751 | |
| 1752 | /* |
| 1753 | * Do a first racy fast path check if this AG is usable. |
| 1754 | */ |
| 1755 | if (!pag->pagi_freecount && !okalloc) |
| 1756 | goto nextag; |
| 1757 | |
| 1758 | /* |
| 1759 | * Then read in the AGI buffer and recheck with the AGI buffer |
| 1760 | * lock held. |
| 1761 | */ |
| 1762 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
| 1763 | if (error) |
| 1764 | goto out_error; |
| 1765 | |
| 1766 | if (pag->pagi_freecount) { |
| 1767 | xfs_perag_put(pag); |
| 1768 | goto out_alloc; |
| 1769 | } |
| 1770 | |
| 1771 | if (!okalloc) |
| 1772 | goto nextag_relse_buffer; |
| 1773 | |
| 1774 | |
| 1775 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); |
| 1776 | if (error) { |
| 1777 | xfs_trans_brelse(tp, agbp); |
| 1778 | |
| 1779 | if (error != -ENOSPC) |
| 1780 | goto out_error; |
| 1781 | |
| 1782 | xfs_perag_put(pag); |
| 1783 | *inop = NULLFSINO; |
| 1784 | return 0; |
| 1785 | } |
| 1786 | |
| 1787 | if (ialloced) { |
| 1788 | /* |
| 1789 | * We successfully allocated some inodes, return |
| 1790 | * the current context to the caller so that it |
| 1791 | * can commit the current transaction and call |
| 1792 | * us again where we left off. |
| 1793 | */ |
| 1794 | ASSERT(pag->pagi_freecount > 0); |
| 1795 | xfs_perag_put(pag); |
| 1796 | |
| 1797 | *IO_agbp = agbp; |
| 1798 | *inop = NULLFSINO; |
| 1799 | return 0; |
| 1800 | } |
| 1801 | |
| 1802 | nextag_relse_buffer: |
| 1803 | xfs_trans_brelse(tp, agbp); |
| 1804 | nextag: |
| 1805 | xfs_perag_put(pag); |
| 1806 | if (++agno == mp->m_sb.sb_agcount) |
| 1807 | agno = 0; |
| 1808 | if (agno == start_agno) { |
| 1809 | *inop = NULLFSINO; |
| 1810 | return noroom ? -ENOSPC : 0; |
| 1811 | } |
| 1812 | } |
| 1813 | |
| 1814 | out_alloc: |
| 1815 | *IO_agbp = NULL; |
| 1816 | return xfs_dialloc_ag(tp, agbp, parent, inop); |
| 1817 | out_error: |
| 1818 | xfs_perag_put(pag); |
| 1819 | return error; |
| 1820 | } |
| 1821 | |
| 1822 | /* |
| 1823 | * Free the blocks of an inode chunk. We must consider that the inode chunk |
| 1824 | * might be sparse and only free the regions that are allocated as part of the |
| 1825 | * chunk. |
| 1826 | */ |
| 1827 | STATIC void |
| 1828 | xfs_difree_inode_chunk( |
| 1829 | struct xfs_mount *mp, |
| 1830 | xfs_agnumber_t agno, |
| 1831 | struct xfs_inobt_rec_incore *rec, |
| 1832 | struct xfs_defer_ops *dfops) |
| 1833 | { |
| 1834 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); |
| 1835 | int startidx, endidx; |
| 1836 | int nextbit; |
| 1837 | xfs_agblock_t agbno; |
| 1838 | int contigblk; |
| 1839 | struct xfs_owner_info oinfo; |
| 1840 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
| 1841 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES); |
| 1842 | |
| 1843 | if (!xfs_inobt_issparse(rec->ir_holemask)) { |
| 1844 | /* not sparse, calculate extent info directly */ |
| 1845 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno), |
| 1846 | mp->m_ialloc_blks, &oinfo); |
| 1847 | return; |
| 1848 | } |
| 1849 | |
| 1850 | /* holemask is only 16-bits (fits in an unsigned long) */ |
| 1851 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); |
| 1852 | holemask[0] = rec->ir_holemask; |
| 1853 | |
| 1854 | /* |
| 1855 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the |
| 1856 | * holemask and convert the start/end index of each range to an extent. |
| 1857 | * We start with the start and end index both pointing at the first 0 in |
| 1858 | * the mask. |
| 1859 | */ |
| 1860 | startidx = endidx = find_first_zero_bit(holemask, |
| 1861 | XFS_INOBT_HOLEMASK_BITS); |
| 1862 | nextbit = startidx + 1; |
| 1863 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { |
| 1864 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, |
| 1865 | nextbit); |
| 1866 | /* |
| 1867 | * If the next zero bit is contiguous, update the end index of |
| 1868 | * the current range and continue. |
| 1869 | */ |
| 1870 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && |
| 1871 | nextbit == endidx + 1) { |
| 1872 | endidx = nextbit; |
| 1873 | goto next; |
| 1874 | } |
| 1875 | |
| 1876 | /* |
| 1877 | * nextbit is not contiguous with the current end index. Convert |
| 1878 | * the current start/end to an extent and add it to the free |
| 1879 | * list. |
| 1880 | */ |
| 1881 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / |
| 1882 | mp->m_sb.sb_inopblock; |
| 1883 | contigblk = ((endidx - startidx + 1) * |
| 1884 | XFS_INODES_PER_HOLEMASK_BIT) / |
| 1885 | mp->m_sb.sb_inopblock; |
| 1886 | |
| 1887 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); |
| 1888 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); |
| 1889 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno), |
| 1890 | contigblk, &oinfo); |
| 1891 | |
| 1892 | /* reset range to current bit and carry on... */ |
| 1893 | startidx = endidx = nextbit; |
| 1894 | |
| 1895 | next: |
| 1896 | nextbit++; |
| 1897 | } |
| 1898 | } |
| 1899 | |
| 1900 | STATIC int |
| 1901 | xfs_difree_inobt( |
| 1902 | struct xfs_mount *mp, |
| 1903 | struct xfs_trans *tp, |
| 1904 | struct xfs_buf *agbp, |
| 1905 | xfs_agino_t agino, |
| 1906 | struct xfs_defer_ops *dfops, |
| 1907 | struct xfs_icluster *xic, |
| 1908 | struct xfs_inobt_rec_incore *orec) |
| 1909 | { |
| 1910 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 1911 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 1912 | struct xfs_perag *pag; |
| 1913 | struct xfs_btree_cur *cur; |
| 1914 | struct xfs_inobt_rec_incore rec; |
| 1915 | int ilen; |
| 1916 | int error; |
| 1917 | int i; |
| 1918 | int off; |
| 1919 | |
| 1920 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
| 1921 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
| 1922 | |
| 1923 | /* |
| 1924 | * Initialize the cursor. |
| 1925 | */ |
| 1926 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
| 1927 | |
| 1928 | error = xfs_check_agi_freecount(cur, agi); |
| 1929 | if (error) |
| 1930 | goto error0; |
| 1931 | |
| 1932 | /* |
| 1933 | * Look for the entry describing this inode. |
| 1934 | */ |
| 1935 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
| 1936 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
| 1937 | __func__, error); |
| 1938 | goto error0; |
| 1939 | } |
| 1940 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1941 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 1942 | if (error) { |
| 1943 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
| 1944 | __func__, error); |
| 1945 | goto error0; |
| 1946 | } |
| 1947 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
| 1948 | /* |
| 1949 | * Get the offset in the inode chunk. |
| 1950 | */ |
| 1951 | off = agino - rec.ir_startino; |
| 1952 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); |
| 1953 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
| 1954 | /* |
| 1955 | * Mark the inode free & increment the count. |
| 1956 | */ |
| 1957 | rec.ir_free |= XFS_INOBT_MASK(off); |
| 1958 | rec.ir_freecount++; |
| 1959 | |
| 1960 | /* |
| 1961 | * When an inode chunk is free, it becomes eligible for removal. Don't |
| 1962 | * remove the chunk if the block size is large enough for multiple inode |
| 1963 | * chunks (that might not be free). |
| 1964 | */ |
| 1965 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
| 1966 | rec.ir_free == XFS_INOBT_ALL_FREE && |
| 1967 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { |
| 1968 | xic->deleted = 1; |
| 1969 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
| 1970 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); |
| 1971 | |
| 1972 | /* |
| 1973 | * Remove the inode cluster from the AGI B+Tree, adjust the |
| 1974 | * AGI and Superblock inode counts, and mark the disk space |
| 1975 | * to be freed when the transaction is committed. |
| 1976 | */ |
| 1977 | ilen = rec.ir_freecount; |
| 1978 | be32_add_cpu(&agi->agi_count, -ilen); |
| 1979 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); |
| 1980 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
| 1981 | pag = xfs_perag_get(mp, agno); |
| 1982 | pag->pagi_freecount -= ilen - 1; |
| 1983 | xfs_perag_put(pag); |
| 1984 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
| 1985 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); |
| 1986 | |
| 1987 | if ((error = xfs_btree_delete(cur, &i))) { |
| 1988 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
| 1989 | __func__, error); |
| 1990 | goto error0; |
| 1991 | } |
| 1992 | |
| 1993 | xfs_difree_inode_chunk(mp, agno, &rec, dfops); |
| 1994 | } else { |
| 1995 | xic->deleted = 0; |
| 1996 | |
| 1997 | error = xfs_inobt_update(cur, &rec); |
| 1998 | if (error) { |
| 1999 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
| 2000 | __func__, error); |
| 2001 | goto error0; |
| 2002 | } |
| 2003 | |
| 2004 | /* |
| 2005 | * Change the inode free counts and log the ag/sb changes. |
| 2006 | */ |
| 2007 | be32_add_cpu(&agi->agi_freecount, 1); |
| 2008 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
| 2009 | pag = xfs_perag_get(mp, agno); |
| 2010 | pag->pagi_freecount++; |
| 2011 | xfs_perag_put(pag); |
| 2012 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
| 2013 | } |
| 2014 | |
| 2015 | error = xfs_check_agi_freecount(cur, agi); |
| 2016 | if (error) |
| 2017 | goto error0; |
| 2018 | |
| 2019 | *orec = rec; |
| 2020 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 2021 | return 0; |
| 2022 | |
| 2023 | error0: |
| 2024 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 2025 | return error; |
| 2026 | } |
| 2027 | |
| 2028 | /* |
| 2029 | * Free an inode in the free inode btree. |
| 2030 | */ |
| 2031 | STATIC int |
| 2032 | xfs_difree_finobt( |
| 2033 | struct xfs_mount *mp, |
| 2034 | struct xfs_trans *tp, |
| 2035 | struct xfs_buf *agbp, |
| 2036 | xfs_agino_t agino, |
| 2037 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ |
| 2038 | { |
| 2039 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
| 2040 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); |
| 2041 | struct xfs_btree_cur *cur; |
| 2042 | struct xfs_inobt_rec_incore rec; |
| 2043 | int offset = agino - ibtrec->ir_startino; |
| 2044 | int error; |
| 2045 | int i; |
| 2046 | |
| 2047 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); |
| 2048 | |
| 2049 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); |
| 2050 | if (error) |
| 2051 | goto error; |
| 2052 | if (i == 0) { |
| 2053 | /* |
| 2054 | * If the record does not exist in the finobt, we must have just |
| 2055 | * freed an inode in a previously fully allocated chunk. If not, |
| 2056 | * something is out of sync. |
| 2057 | */ |
| 2058 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
| 2059 | |
| 2060 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
| 2061 | ibtrec->ir_count, |
| 2062 | ibtrec->ir_freecount, |
| 2063 | ibtrec->ir_free, &i); |
| 2064 | if (error) |
| 2065 | goto error; |
| 2066 | ASSERT(i == 1); |
| 2067 | |
| 2068 | goto out; |
| 2069 | } |
| 2070 | |
| 2071 | /* |
| 2072 | * Read and update the existing record. We could just copy the ibtrec |
| 2073 | * across here, but that would defeat the purpose of having redundant |
| 2074 | * metadata. By making the modifications independently, we can catch |
| 2075 | * corruptions that we wouldn't see if we just copied from one record |
| 2076 | * to another. |
| 2077 | */ |
| 2078 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 2079 | if (error) |
| 2080 | goto error; |
| 2081 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
| 2082 | |
| 2083 | rec.ir_free |= XFS_INOBT_MASK(offset); |
| 2084 | rec.ir_freecount++; |
| 2085 | |
| 2086 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
| 2087 | (rec.ir_freecount == ibtrec->ir_freecount), |
| 2088 | error); |
| 2089 | |
| 2090 | /* |
| 2091 | * The content of inobt records should always match between the inobt |
| 2092 | * and finobt. The lifecycle of records in the finobt is different from |
| 2093 | * the inobt in that the finobt only tracks records with at least one |
| 2094 | * free inode. Hence, if all of the inodes are free and we aren't |
| 2095 | * keeping inode chunks permanently on disk, remove the record. |
| 2096 | * Otherwise, update the record with the new information. |
| 2097 | * |
| 2098 | * Note that we currently can't free chunks when the block size is large |
| 2099 | * enough for multiple chunks. Leave the finobt record to remain in sync |
| 2100 | * with the inobt. |
| 2101 | */ |
| 2102 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
| 2103 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && |
| 2104 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
| 2105 | error = xfs_btree_delete(cur, &i); |
| 2106 | if (error) |
| 2107 | goto error; |
| 2108 | ASSERT(i == 1); |
| 2109 | } else { |
| 2110 | error = xfs_inobt_update(cur, &rec); |
| 2111 | if (error) |
| 2112 | goto error; |
| 2113 | } |
| 2114 | |
| 2115 | out: |
| 2116 | error = xfs_check_agi_freecount(cur, agi); |
| 2117 | if (error) |
| 2118 | goto error; |
| 2119 | |
| 2120 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
| 2121 | return 0; |
| 2122 | |
| 2123 | error: |
| 2124 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
| 2125 | return error; |
| 2126 | } |
| 2127 | |
| 2128 | /* |
| 2129 | * Free disk inode. Carefully avoids touching the incore inode, all |
| 2130 | * manipulations incore are the caller's responsibility. |
| 2131 | * The on-disk inode is not changed by this operation, only the |
| 2132 | * btree (free inode mask) is changed. |
| 2133 | */ |
| 2134 | int |
| 2135 | xfs_difree( |
| 2136 | struct xfs_trans *tp, /* transaction pointer */ |
| 2137 | xfs_ino_t inode, /* inode to be freed */ |
| 2138 | struct xfs_defer_ops *dfops, /* extents to free */ |
| 2139 | struct xfs_icluster *xic) /* cluster info if deleted */ |
| 2140 | { |
| 2141 | /* REFERENCED */ |
| 2142 | xfs_agblock_t agbno; /* block number containing inode */ |
| 2143 | struct xfs_buf *agbp; /* buffer for allocation group header */ |
| 2144 | xfs_agino_t agino; /* allocation group inode number */ |
| 2145 | xfs_agnumber_t agno; /* allocation group number */ |
| 2146 | int error; /* error return value */ |
| 2147 | struct xfs_mount *mp; /* mount structure for filesystem */ |
| 2148 | struct xfs_inobt_rec_incore rec;/* btree record */ |
| 2149 | |
| 2150 | mp = tp->t_mountp; |
| 2151 | |
| 2152 | /* |
| 2153 | * Break up inode number into its components. |
| 2154 | */ |
| 2155 | agno = XFS_INO_TO_AGNO(mp, inode); |
| 2156 | if (agno >= mp->m_sb.sb_agcount) { |
| 2157 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", |
| 2158 | __func__, agno, mp->m_sb.sb_agcount); |
| 2159 | ASSERT(0); |
| 2160 | return -EINVAL; |
| 2161 | } |
| 2162 | agino = XFS_INO_TO_AGINO(mp, inode); |
| 2163 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { |
| 2164 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", |
| 2165 | __func__, (unsigned long long)inode, |
| 2166 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); |
| 2167 | ASSERT(0); |
| 2168 | return -EINVAL; |
| 2169 | } |
| 2170 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); |
| 2171 | if (agbno >= mp->m_sb.sb_agblocks) { |
| 2172 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", |
| 2173 | __func__, agbno, mp->m_sb.sb_agblocks); |
| 2174 | ASSERT(0); |
| 2175 | return -EINVAL; |
| 2176 | } |
| 2177 | /* |
| 2178 | * Get the allocation group header. |
| 2179 | */ |
| 2180 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
| 2181 | if (error) { |
| 2182 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", |
| 2183 | __func__, error); |
| 2184 | return error; |
| 2185 | } |
| 2186 | |
| 2187 | /* |
| 2188 | * Fix up the inode allocation btree. |
| 2189 | */ |
| 2190 | error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec); |
| 2191 | if (error) |
| 2192 | goto error0; |
| 2193 | |
| 2194 | /* |
| 2195 | * Fix up the free inode btree. |
| 2196 | */ |
| 2197 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { |
| 2198 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); |
| 2199 | if (error) |
| 2200 | goto error0; |
| 2201 | } |
| 2202 | |
| 2203 | return 0; |
| 2204 | |
| 2205 | error0: |
| 2206 | return error; |
| 2207 | } |
| 2208 | |
| 2209 | STATIC int |
| 2210 | xfs_imap_lookup( |
| 2211 | struct xfs_mount *mp, |
| 2212 | struct xfs_trans *tp, |
| 2213 | xfs_agnumber_t agno, |
| 2214 | xfs_agino_t agino, |
| 2215 | xfs_agblock_t agbno, |
| 2216 | xfs_agblock_t *chunk_agbno, |
| 2217 | xfs_agblock_t *offset_agbno, |
| 2218 | int flags) |
| 2219 | { |
| 2220 | struct xfs_inobt_rec_incore rec; |
| 2221 | struct xfs_btree_cur *cur; |
| 2222 | struct xfs_buf *agbp; |
| 2223 | int error; |
| 2224 | int i; |
| 2225 | |
| 2226 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
| 2227 | if (error) { |
| 2228 | xfs_alert(mp, |
| 2229 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", |
| 2230 | __func__, error, agno); |
| 2231 | return error; |
| 2232 | } |
| 2233 | |
| 2234 | /* |
| 2235 | * Lookup the inode record for the given agino. If the record cannot be |
| 2236 | * found, then it's an invalid inode number and we should abort. Once |
| 2237 | * we have a record, we need to ensure it contains the inode number |
| 2238 | * we are looking up. |
| 2239 | */ |
| 2240 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
| 2241 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
| 2242 | if (!error) { |
| 2243 | if (i) |
| 2244 | error = xfs_inobt_get_rec(cur, &rec, &i); |
| 2245 | if (!error && i == 0) |
| 2246 | error = -EINVAL; |
| 2247 | } |
| 2248 | |
| 2249 | xfs_trans_brelse(tp, agbp); |
| 2250 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
| 2251 | if (error) |
| 2252 | return error; |
| 2253 | |
| 2254 | /* check that the returned record contains the required inode */ |
| 2255 | if (rec.ir_startino > agino || |
| 2256 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
| 2257 | return -EINVAL; |
| 2258 | |
| 2259 | /* for untrusted inodes check it is allocated first */ |
| 2260 | if ((flags & XFS_IGET_UNTRUSTED) && |
| 2261 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
| 2262 | return -EINVAL; |
| 2263 | |
| 2264 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); |
| 2265 | *offset_agbno = agbno - *chunk_agbno; |
| 2266 | return 0; |
| 2267 | } |
| 2268 | |
| 2269 | /* |
| 2270 | * Return the location of the inode in imap, for mapping it into a buffer. |
| 2271 | */ |
| 2272 | int |
| 2273 | xfs_imap( |
| 2274 | xfs_mount_t *mp, /* file system mount structure */ |
| 2275 | xfs_trans_t *tp, /* transaction pointer */ |
| 2276 | xfs_ino_t ino, /* inode to locate */ |
| 2277 | struct xfs_imap *imap, /* location map structure */ |
| 2278 | uint flags) /* flags for inode btree lookup */ |
| 2279 | { |
| 2280 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ |
| 2281 | xfs_agino_t agino; /* inode number within alloc group */ |
| 2282 | xfs_agnumber_t agno; /* allocation group number */ |
| 2283 | int blks_per_cluster; /* num blocks per inode cluster */ |
| 2284 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ |
| 2285 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
| 2286 | int error; /* error code */ |
| 2287 | int offset; /* index of inode in its buffer */ |
| 2288 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
| 2289 | |
| 2290 | ASSERT(ino != NULLFSINO); |
| 2291 | |
| 2292 | /* |
| 2293 | * Split up the inode number into its parts. |
| 2294 | */ |
| 2295 | agno = XFS_INO_TO_AGNO(mp, ino); |
| 2296 | agino = XFS_INO_TO_AGINO(mp, ino); |
| 2297 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); |
| 2298 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || |
| 2299 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
| 2300 | #ifdef DEBUG |
| 2301 | /* |
| 2302 | * Don't output diagnostic information for untrusted inodes |
| 2303 | * as they can be invalid without implying corruption. |
| 2304 | */ |
| 2305 | if (flags & XFS_IGET_UNTRUSTED) |
| 2306 | return -EINVAL; |
| 2307 | if (agno >= mp->m_sb.sb_agcount) { |
| 2308 | xfs_alert(mp, |
| 2309 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", |
| 2310 | __func__, agno, mp->m_sb.sb_agcount); |
| 2311 | } |
| 2312 | if (agbno >= mp->m_sb.sb_agblocks) { |
| 2313 | xfs_alert(mp, |
| 2314 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", |
| 2315 | __func__, (unsigned long long)agbno, |
| 2316 | (unsigned long)mp->m_sb.sb_agblocks); |
| 2317 | } |
| 2318 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
| 2319 | xfs_alert(mp, |
| 2320 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", |
| 2321 | __func__, ino, |
| 2322 | XFS_AGINO_TO_INO(mp, agno, agino)); |
| 2323 | } |
| 2324 | xfs_stack_trace(); |
| 2325 | #endif /* DEBUG */ |
| 2326 | return -EINVAL; |
| 2327 | } |
| 2328 | |
| 2329 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
| 2330 | |
| 2331 | /* |
| 2332 | * For bulkstat and handle lookups, we have an untrusted inode number |
| 2333 | * that we have to verify is valid. We cannot do this just by reading |
| 2334 | * the inode buffer as it may have been unlinked and removed leaving |
| 2335 | * inodes in stale state on disk. Hence we have to do a btree lookup |
| 2336 | * in all cases where an untrusted inode number is passed. |
| 2337 | */ |
| 2338 | if (flags & XFS_IGET_UNTRUSTED) { |
| 2339 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
| 2340 | &chunk_agbno, &offset_agbno, flags); |
| 2341 | if (error) |
| 2342 | return error; |
| 2343 | goto out_map; |
| 2344 | } |
| 2345 | |
| 2346 | /* |
| 2347 | * If the inode cluster size is the same as the blocksize or |
| 2348 | * smaller we get to the buffer by simple arithmetics. |
| 2349 | */ |
| 2350 | if (blks_per_cluster == 1) { |
| 2351 | offset = XFS_INO_TO_OFFSET(mp, ino); |
| 2352 | ASSERT(offset < mp->m_sb.sb_inopblock); |
| 2353 | |
| 2354 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); |
| 2355 | imap->im_len = XFS_FSB_TO_BB(mp, 1); |
| 2356 | imap->im_boffset = (unsigned short)(offset << |
| 2357 | mp->m_sb.sb_inodelog); |
| 2358 | return 0; |
| 2359 | } |
| 2360 | |
| 2361 | /* |
| 2362 | * If the inode chunks are aligned then use simple maths to |
| 2363 | * find the location. Otherwise we have to do a btree |
| 2364 | * lookup to find the location. |
| 2365 | */ |
| 2366 | if (mp->m_inoalign_mask) { |
| 2367 | offset_agbno = agbno & mp->m_inoalign_mask; |
| 2368 | chunk_agbno = agbno - offset_agbno; |
| 2369 | } else { |
| 2370 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
| 2371 | &chunk_agbno, &offset_agbno, flags); |
| 2372 | if (error) |
| 2373 | return error; |
| 2374 | } |
| 2375 | |
| 2376 | out_map: |
| 2377 | ASSERT(agbno >= chunk_agbno); |
| 2378 | cluster_agbno = chunk_agbno + |
| 2379 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); |
| 2380 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + |
| 2381 | XFS_INO_TO_OFFSET(mp, ino); |
| 2382 | |
| 2383 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); |
| 2384 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); |
| 2385 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
| 2386 | |
| 2387 | /* |
| 2388 | * If the inode number maps to a block outside the bounds |
| 2389 | * of the file system then return NULL rather than calling |
| 2390 | * read_buf and panicing when we get an error from the |
| 2391 | * driver. |
| 2392 | */ |
| 2393 | if ((imap->im_blkno + imap->im_len) > |
| 2394 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { |
| 2395 | xfs_alert(mp, |
| 2396 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", |
| 2397 | __func__, (unsigned long long) imap->im_blkno, |
| 2398 | (unsigned long long) imap->im_len, |
| 2399 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); |
| 2400 | return -EINVAL; |
| 2401 | } |
| 2402 | return 0; |
| 2403 | } |
| 2404 | |
| 2405 | /* |
| 2406 | * Compute and fill in value of m_in_maxlevels. |
| 2407 | */ |
| 2408 | void |
| 2409 | xfs_ialloc_compute_maxlevels( |
| 2410 | xfs_mount_t *mp) /* file system mount structure */ |
| 2411 | { |
| 2412 | uint inodes; |
| 2413 | |
| 2414 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; |
| 2415 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, |
| 2416 | inodes); |
| 2417 | } |
| 2418 | |
| 2419 | /* |
| 2420 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
| 2421 | * structure over time requires that we interpret the buffer as two logical |
| 2422 | * regions delineated by the end of the unlinked list. This is due to the size |
| 2423 | * of the hash table and its location in the middle of the agi. |
| 2424 | * |
| 2425 | * For example, a request to log a field before agi_unlinked and a field after |
| 2426 | * agi_unlinked could cause us to log the entire hash table and use an excessive |
| 2427 | * amount of log space. To avoid this behavior, log the region up through |
| 2428 | * agi_unlinked in one call and the region after agi_unlinked through the end of |
| 2429 | * the structure in another. |
| 2430 | */ |
| 2431 | void |
| 2432 | xfs_ialloc_log_agi( |
| 2433 | xfs_trans_t *tp, /* transaction pointer */ |
| 2434 | xfs_buf_t *bp, /* allocation group header buffer */ |
| 2435 | int fields) /* bitmask of fields to log */ |
| 2436 | { |
| 2437 | int first; /* first byte number */ |
| 2438 | int last; /* last byte number */ |
| 2439 | static const short offsets[] = { /* field starting offsets */ |
| 2440 | /* keep in sync with bit definitions */ |
| 2441 | offsetof(xfs_agi_t, agi_magicnum), |
| 2442 | offsetof(xfs_agi_t, agi_versionnum), |
| 2443 | offsetof(xfs_agi_t, agi_seqno), |
| 2444 | offsetof(xfs_agi_t, agi_length), |
| 2445 | offsetof(xfs_agi_t, agi_count), |
| 2446 | offsetof(xfs_agi_t, agi_root), |
| 2447 | offsetof(xfs_agi_t, agi_level), |
| 2448 | offsetof(xfs_agi_t, agi_freecount), |
| 2449 | offsetof(xfs_agi_t, agi_newino), |
| 2450 | offsetof(xfs_agi_t, agi_dirino), |
| 2451 | offsetof(xfs_agi_t, agi_unlinked), |
| 2452 | offsetof(xfs_agi_t, agi_free_root), |
| 2453 | offsetof(xfs_agi_t, agi_free_level), |
| 2454 | sizeof(xfs_agi_t) |
| 2455 | }; |
| 2456 | #ifdef DEBUG |
| 2457 | xfs_agi_t *agi; /* allocation group header */ |
| 2458 | |
| 2459 | agi = XFS_BUF_TO_AGI(bp); |
| 2460 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
| 2461 | #endif |
| 2462 | |
| 2463 | /* |
| 2464 | * Compute byte offsets for the first and last fields in the first |
| 2465 | * region and log the agi buffer. This only logs up through |
| 2466 | * agi_unlinked. |
| 2467 | */ |
| 2468 | if (fields & XFS_AGI_ALL_BITS_R1) { |
| 2469 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, |
| 2470 | &first, &last); |
| 2471 | xfs_trans_log_buf(tp, bp, first, last); |
| 2472 | } |
| 2473 | |
| 2474 | /* |
| 2475 | * Mask off the bits in the first region and calculate the first and |
| 2476 | * last field offsets for any bits in the second region. |
| 2477 | */ |
| 2478 | fields &= ~XFS_AGI_ALL_BITS_R1; |
| 2479 | if (fields) { |
| 2480 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, |
| 2481 | &first, &last); |
| 2482 | xfs_trans_log_buf(tp, bp, first, last); |
| 2483 | } |
| 2484 | } |
| 2485 | |
| 2486 | #ifdef DEBUG |
| 2487 | STATIC void |
| 2488 | xfs_check_agi_unlinked( |
| 2489 | struct xfs_agi *agi) |
| 2490 | { |
| 2491 | int i; |
| 2492 | |
| 2493 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) |
| 2494 | ASSERT(agi->agi_unlinked[i]); |
| 2495 | } |
| 2496 | #else |
| 2497 | #define xfs_check_agi_unlinked(agi) |
| 2498 | #endif |
| 2499 | |
| 2500 | static bool |
| 2501 | xfs_agi_verify( |
| 2502 | struct xfs_buf *bp) |
| 2503 | { |
| 2504 | struct xfs_mount *mp = bp->b_target->bt_mount; |
| 2505 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); |
| 2506 | |
| 2507 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 2508 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) |
| 2509 | return false; |
| 2510 | if (!xfs_log_check_lsn(mp, |
| 2511 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) |
| 2512 | return false; |
| 2513 | } |
| 2514 | |
| 2515 | /* |
| 2516 | * Validate the magic number of the agi block. |
| 2517 | */ |
| 2518 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
| 2519 | return false; |
| 2520 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) |
| 2521 | return false; |
| 2522 | |
| 2523 | if (be32_to_cpu(agi->agi_level) < 1 || |
| 2524 | be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
| 2525 | return false; |
| 2526 | |
| 2527 | if (xfs_sb_version_hasfinobt(&mp->m_sb) && |
| 2528 | (be32_to_cpu(agi->agi_free_level) < 1 || |
| 2529 | be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS)) |
| 2530 | return false; |
| 2531 | |
| 2532 | /* |
| 2533 | * during growfs operations, the perag is not fully initialised, |
| 2534 | * so we can't use it for any useful checking. growfs ensures we can't |
| 2535 | * use it by using uncached buffers that don't have the perag attached |
| 2536 | * so we can detect and avoid this problem. |
| 2537 | */ |
| 2538 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
| 2539 | return false; |
| 2540 | |
| 2541 | xfs_check_agi_unlinked(agi); |
| 2542 | return true; |
| 2543 | } |
| 2544 | |
| 2545 | static void |
| 2546 | xfs_agi_read_verify( |
| 2547 | struct xfs_buf *bp) |
| 2548 | { |
| 2549 | struct xfs_mount *mp = bp->b_target->bt_mount; |
| 2550 | |
| 2551 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
| 2552 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) |
| 2553 | xfs_buf_ioerror(bp, -EFSBADCRC); |
| 2554 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
| 2555 | XFS_ERRTAG_IALLOC_READ_AGI)) |
| 2556 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
| 2557 | |
| 2558 | if (bp->b_error) |
| 2559 | xfs_verifier_error(bp); |
| 2560 | } |
| 2561 | |
| 2562 | static void |
| 2563 | xfs_agi_write_verify( |
| 2564 | struct xfs_buf *bp) |
| 2565 | { |
| 2566 | struct xfs_mount *mp = bp->b_target->bt_mount; |
| 2567 | struct xfs_buf_log_item *bip = bp->b_fspriv; |
| 2568 | |
| 2569 | if (!xfs_agi_verify(bp)) { |
| 2570 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
| 2571 | xfs_verifier_error(bp); |
| 2572 | return; |
| 2573 | } |
| 2574 | |
| 2575 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 2576 | return; |
| 2577 | |
| 2578 | if (bip) |
| 2579 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); |
| 2580 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
| 2581 | } |
| 2582 | |
| 2583 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
| 2584 | .name = "xfs_agi", |
| 2585 | .verify_read = xfs_agi_read_verify, |
| 2586 | .verify_write = xfs_agi_write_verify, |
| 2587 | }; |
| 2588 | |
| 2589 | /* |
| 2590 | * Read in the allocation group header (inode allocation section) |
| 2591 | */ |
| 2592 | int |
| 2593 | xfs_read_agi( |
| 2594 | struct xfs_mount *mp, /* file system mount structure */ |
| 2595 | struct xfs_trans *tp, /* transaction pointer */ |
| 2596 | xfs_agnumber_t agno, /* allocation group number */ |
| 2597 | struct xfs_buf **bpp) /* allocation group hdr buf */ |
| 2598 | { |
| 2599 | int error; |
| 2600 | |
| 2601 | trace_xfs_read_agi(mp, agno); |
| 2602 | |
| 2603 | ASSERT(agno != NULLAGNUMBER); |
| 2604 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
| 2605 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
| 2606 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
| 2607 | if (error) |
| 2608 | return error; |
| 2609 | if (tp) |
| 2610 | xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF); |
| 2611 | |
| 2612 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
| 2613 | return 0; |
| 2614 | } |
| 2615 | |
| 2616 | int |
| 2617 | xfs_ialloc_read_agi( |
| 2618 | struct xfs_mount *mp, /* file system mount structure */ |
| 2619 | struct xfs_trans *tp, /* transaction pointer */ |
| 2620 | xfs_agnumber_t agno, /* allocation group number */ |
| 2621 | struct xfs_buf **bpp) /* allocation group hdr buf */ |
| 2622 | { |
| 2623 | struct xfs_agi *agi; /* allocation group header */ |
| 2624 | struct xfs_perag *pag; /* per allocation group data */ |
| 2625 | int error; |
| 2626 | |
| 2627 | trace_xfs_ialloc_read_agi(mp, agno); |
| 2628 | |
| 2629 | error = xfs_read_agi(mp, tp, agno, bpp); |
| 2630 | if (error) |
| 2631 | return error; |
| 2632 | |
| 2633 | agi = XFS_BUF_TO_AGI(*bpp); |
| 2634 | pag = xfs_perag_get(mp, agno); |
| 2635 | if (!pag->pagi_init) { |
| 2636 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
| 2637 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
| 2638 | pag->pagi_init = 1; |
| 2639 | } |
| 2640 | |
| 2641 | /* |
| 2642 | * It's possible for these to be out of sync if |
| 2643 | * we are in the middle of a forced shutdown. |
| 2644 | */ |
| 2645 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || |
| 2646 | XFS_FORCED_SHUTDOWN(mp)); |
| 2647 | xfs_perag_put(pag); |
| 2648 | return 0; |
| 2649 | } |
| 2650 | |
| 2651 | /* |
| 2652 | * Read in the agi to initialise the per-ag data in the mount structure |
| 2653 | */ |
| 2654 | int |
| 2655 | xfs_ialloc_pagi_init( |
| 2656 | xfs_mount_t *mp, /* file system mount structure */ |
| 2657 | xfs_trans_t *tp, /* transaction pointer */ |
| 2658 | xfs_agnumber_t agno) /* allocation group number */ |
| 2659 | { |
| 2660 | xfs_buf_t *bp = NULL; |
| 2661 | int error; |
| 2662 | |
| 2663 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); |
| 2664 | if (error) |
| 2665 | return error; |
| 2666 | if (bp) |
| 2667 | xfs_trans_brelse(tp, bp); |
| 2668 | return 0; |
| 2669 | } |