1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18 #include "xfs.h" 19 #include "xfs_fs.h" 20 #include "xfs_types.h" 21 #include "xfs_bit.h" 22 #include "xfs_log.h" 23 #include "xfs_inum.h" 24 #include "xfs_imap.h" 25 #include "xfs_trans.h" 26 #include "xfs_trans_priv.h" 27 #include "xfs_sb.h" 28 #include "xfs_ag.h" 29 #include "xfs_dir2.h" 30 #include "xfs_dmapi.h" 31 #include "xfs_mount.h" 32 #include "xfs_bmap_btree.h" 33 #include "xfs_alloc_btree.h" 34 #include "xfs_ialloc_btree.h" 35 #include "xfs_dir2_sf.h" 36 #include "xfs_attr_sf.h" 37 #include "xfs_dinode.h" 38 #include "xfs_inode.h" 39 #include "xfs_buf_item.h" 40 #include "xfs_inode_item.h" 41 #include "xfs_btree.h" 42 #include "xfs_alloc.h" 43 #include "xfs_ialloc.h" 44 #include "xfs_bmap.h" 45 #include "xfs_rw.h" 46 #include "xfs_error.h" 47 #include "xfs_utils.h" 48 #include "xfs_dir2_trace.h" 49 #include "xfs_quota.h" 50 #include "xfs_acl.h" 51 52 53 kmem_zone_t *xfs_ifork_zone; 54 kmem_zone_t *xfs_inode_zone; 55 kmem_zone_t *xfs_chashlist_zone; 56 57 /* 58 * Used in xfs_itruncate(). This is the maximum number of extents 59 * freed from a file in a single transaction. 60 */ 61 #define XFS_ITRUNC_MAX_EXTENTS 2 62 63 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); 64 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); 65 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); 66 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); 67 68 69 #ifdef DEBUG 70 /* 71 * Make sure that the extents in the given memory buffer 72 * are valid. 73 */ 74 STATIC void 75 xfs_validate_extents( 76 xfs_ifork_t *ifp, 77 int nrecs, 78 int disk, 79 xfs_exntfmt_t fmt) 80 { 81 xfs_bmbt_rec_t *ep; 82 xfs_bmbt_irec_t irec; 83 xfs_bmbt_rec_t rec; 84 int i; 85 86 for (i = 0; i < nrecs; i++) { 87 ep = xfs_iext_get_ext(ifp, i); 88 rec.l0 = get_unaligned((__uint64_t*)&ep->l0); 89 rec.l1 = get_unaligned((__uint64_t*)&ep->l1); 90 if (disk) 91 xfs_bmbt_disk_get_all(&rec, &irec); 92 else 93 xfs_bmbt_get_all(&rec, &irec); 94 if (fmt == XFS_EXTFMT_NOSTATE) 95 ASSERT(irec.br_state == XFS_EXT_NORM); 96 } 97 } 98 #else /* DEBUG */ 99 #define xfs_validate_extents(ifp, nrecs, disk, fmt) 100 #endif /* DEBUG */ 101 102 /* 103 * Check that none of the inode's in the buffer have a next 104 * unlinked field of 0. 105 */ 106 #if defined(DEBUG) 107 void 108 xfs_inobp_check( 109 xfs_mount_t *mp, 110 xfs_buf_t *bp) 111 { 112 int i; 113 int j; 114 xfs_dinode_t *dip; 115 116 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 117 118 for (i = 0; i < j; i++) { 119 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 120 i * mp->m_sb.sb_inodesize); 121 if (!dip->di_next_unlinked) { 122 xfs_fs_cmn_err(CE_ALERT, mp, 123 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", 124 bp); 125 ASSERT(dip->di_next_unlinked); 126 } 127 } 128 } 129 #endif 130 131 /* 132 * This routine is called to map an inode number within a file 133 * system to the buffer containing the on-disk version of the 134 * inode. It returns a pointer to the buffer containing the 135 * on-disk inode in the bpp parameter, and in the dip parameter 136 * it returns a pointer to the on-disk inode within that buffer. 137 * 138 * If a non-zero error is returned, then the contents of bpp and 139 * dipp are undefined. 140 * 141 * Use xfs_imap() to determine the size and location of the 142 * buffer to read from disk. 143 */ 144 STATIC int 145 xfs_inotobp( 146 xfs_mount_t *mp, 147 xfs_trans_t *tp, 148 xfs_ino_t ino, 149 xfs_dinode_t **dipp, 150 xfs_buf_t **bpp, 151 int *offset) 152 { 153 int di_ok; 154 xfs_imap_t imap; 155 xfs_buf_t *bp; 156 int error; 157 xfs_dinode_t *dip; 158 159 /* 160 * Call the space management code to find the location of the 161 * inode on disk. 162 */ 163 imap.im_blkno = 0; 164 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP); 165 if (error != 0) { 166 cmn_err(CE_WARN, 167 "xfs_inotobp: xfs_imap() returned an " 168 "error %d on %s. Returning error.", error, mp->m_fsname); 169 return error; 170 } 171 172 /* 173 * If the inode number maps to a block outside the bounds of the 174 * file system then return NULL rather than calling read_buf 175 * and panicing when we get an error from the driver. 176 */ 177 if ((imap.im_blkno + imap.im_len) > 178 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { 179 cmn_err(CE_WARN, 180 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds " 181 "of the file system %s. Returning EINVAL.", 182 (unsigned long long)imap.im_blkno, 183 imap.im_len, mp->m_fsname); 184 return XFS_ERROR(EINVAL); 185 } 186 187 /* 188 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will 189 * default to just a read_buf() call. 190 */ 191 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, 192 (int)imap.im_len, XFS_BUF_LOCK, &bp); 193 194 if (error) { 195 cmn_err(CE_WARN, 196 "xfs_inotobp: xfs_trans_read_buf() returned an " 197 "error %d on %s. Returning error.", error, mp->m_fsname); 198 return error; 199 } 200 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0); 201 di_ok = 202 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && 203 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); 204 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP, 205 XFS_RANDOM_ITOBP_INOTOBP))) { 206 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip); 207 xfs_trans_brelse(tp, bp); 208 cmn_err(CE_WARN, 209 "xfs_inotobp: XFS_TEST_ERROR() returned an " 210 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname); 211 return XFS_ERROR(EFSCORRUPTED); 212 } 213 214 xfs_inobp_check(mp, bp); 215 216 /* 217 * Set *dipp to point to the on-disk inode in the buffer. 218 */ 219 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); 220 *bpp = bp; 221 *offset = imap.im_boffset; 222 return 0; 223 } 224 225 226 /* 227 * This routine is called to map an inode to the buffer containing 228 * the on-disk version of the inode. It returns a pointer to the 229 * buffer containing the on-disk inode in the bpp parameter, and in 230 * the dip parameter it returns a pointer to the on-disk inode within 231 * that buffer. 232 * 233 * If a non-zero error is returned, then the contents of bpp and 234 * dipp are undefined. 235 * 236 * If the inode is new and has not yet been initialized, use xfs_imap() 237 * to determine the size and location of the buffer to read from disk. 238 * If the inode has already been mapped to its buffer and read in once, 239 * then use the mapping information stored in the inode rather than 240 * calling xfs_imap(). This allows us to avoid the overhead of looking 241 * at the inode btree for small block file systems (see xfs_dilocate()). 242 * We can tell whether the inode has been mapped in before by comparing 243 * its disk block address to 0. Only uninitialized inodes will have 244 * 0 for the disk block address. 245 */ 246 int 247 xfs_itobp( 248 xfs_mount_t *mp, 249 xfs_trans_t *tp, 250 xfs_inode_t *ip, 251 xfs_dinode_t **dipp, 252 xfs_buf_t **bpp, 253 xfs_daddr_t bno, 254 uint imap_flags) 255 { 256 xfs_imap_t imap; 257 xfs_buf_t *bp; 258 int error; 259 int i; 260 int ni; 261 262 if (ip->i_blkno == (xfs_daddr_t)0) { 263 /* 264 * Call the space management code to find the location of the 265 * inode on disk. 266 */ 267 imap.im_blkno = bno; 268 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap, 269 XFS_IMAP_LOOKUP | imap_flags))) 270 return error; 271 272 /* 273 * If the inode number maps to a block outside the bounds 274 * of the file system then return NULL rather than calling 275 * read_buf and panicing when we get an error from the 276 * driver. 277 */ 278 if ((imap.im_blkno + imap.im_len) > 279 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { 280 #ifdef DEBUG 281 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " 282 "(imap.im_blkno (0x%llx) " 283 "+ imap.im_len (0x%llx)) > " 284 " XFS_FSB_TO_BB(mp, " 285 "mp->m_sb.sb_dblocks) (0x%llx)", 286 (unsigned long long) imap.im_blkno, 287 (unsigned long long) imap.im_len, 288 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); 289 #endif /* DEBUG */ 290 return XFS_ERROR(EINVAL); 291 } 292 293 /* 294 * Fill in the fields in the inode that will be used to 295 * map the inode to its buffer from now on. 296 */ 297 ip->i_blkno = imap.im_blkno; 298 ip->i_len = imap.im_len; 299 ip->i_boffset = imap.im_boffset; 300 } else { 301 /* 302 * We've already mapped the inode once, so just use the 303 * mapping that we saved the first time. 304 */ 305 imap.im_blkno = ip->i_blkno; 306 imap.im_len = ip->i_len; 307 imap.im_boffset = ip->i_boffset; 308 } 309 ASSERT(bno == 0 || bno == imap.im_blkno); 310 311 /* 312 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will 313 * default to just a read_buf() call. 314 */ 315 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, 316 (int)imap.im_len, XFS_BUF_LOCK, &bp); 317 if (error) { 318 #ifdef DEBUG 319 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " 320 "xfs_trans_read_buf() returned error %d, " 321 "imap.im_blkno 0x%llx, imap.im_len 0x%llx", 322 error, (unsigned long long) imap.im_blkno, 323 (unsigned long long) imap.im_len); 324 #endif /* DEBUG */ 325 return error; 326 } 327 328 /* 329 * Validate the magic number and version of every inode in the buffer 330 * (if DEBUG kernel) or the first inode in the buffer, otherwise. 331 * No validation is done here in userspace (xfs_repair). 332 */ 333 #if !defined(__KERNEL__) 334 ni = 0; 335 #elif defined(DEBUG) 336 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog; 337 #else /* usual case */ 338 ni = 1; 339 #endif 340 341 for (i = 0; i < ni; i++) { 342 int di_ok; 343 xfs_dinode_t *dip; 344 345 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 346 (i << mp->m_sb.sb_inodelog)); 347 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && 348 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); 349 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, 350 XFS_ERRTAG_ITOBP_INOTOBP, 351 XFS_RANDOM_ITOBP_INOTOBP))) { 352 if (imap_flags & XFS_IMAP_BULKSTAT) { 353 xfs_trans_brelse(tp, bp); 354 return XFS_ERROR(EINVAL); 355 } 356 #ifdef DEBUG 357 cmn_err(CE_ALERT, 358 "Device %s - bad inode magic/vsn " 359 "daddr %lld #%d (magic=%x)", 360 XFS_BUFTARG_NAME(mp->m_ddev_targp), 361 (unsigned long long)imap.im_blkno, i, 362 INT_GET(dip->di_core.di_magic, ARCH_CONVERT)); 363 #endif 364 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH, 365 mp, dip); 366 xfs_trans_brelse(tp, bp); 367 return XFS_ERROR(EFSCORRUPTED); 368 } 369 } 370 371 xfs_inobp_check(mp, bp); 372 373 /* 374 * Mark the buffer as an inode buffer now that it looks good 375 */ 376 XFS_BUF_SET_VTYPE(bp, B_FS_INO); 377 378 /* 379 * Set *dipp to point to the on-disk inode in the buffer. 380 */ 381 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); 382 *bpp = bp; 383 return 0; 384 } 385 386 /* 387 * Move inode type and inode format specific information from the 388 * on-disk inode to the in-core inode. For fifos, devs, and sockets 389 * this means set if_rdev to the proper value. For files, directories, 390 * and symlinks this means to bring in the in-line data or extent 391 * pointers. For a file in B-tree format, only the root is immediately 392 * brought in-core. The rest will be in-lined in if_extents when it 393 * is first referenced (see xfs_iread_extents()). 394 */ 395 STATIC int 396 xfs_iformat( 397 xfs_inode_t *ip, 398 xfs_dinode_t *dip) 399 { 400 xfs_attr_shortform_t *atp; 401 int size; 402 int error; 403 xfs_fsize_t di_size; 404 ip->i_df.if_ext_max = 405 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); 406 error = 0; 407 408 if (unlikely( 409 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) + 410 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) > 411 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) { 412 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 413 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", 414 (unsigned long long)ip->i_ino, 415 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) 416 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)), 417 (unsigned long long) 418 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT)); 419 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, 420 ip->i_mount, dip); 421 return XFS_ERROR(EFSCORRUPTED); 422 } 423 424 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) { 425 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 426 "corrupt dinode %Lu, forkoff = 0x%x.", 427 (unsigned long long)ip->i_ino, 428 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT))); 429 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, 430 ip->i_mount, dip); 431 return XFS_ERROR(EFSCORRUPTED); 432 } 433 434 switch (ip->i_d.di_mode & S_IFMT) { 435 case S_IFIFO: 436 case S_IFCHR: 437 case S_IFBLK: 438 case S_IFSOCK: 439 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) { 440 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, 441 ip->i_mount, dip); 442 return XFS_ERROR(EFSCORRUPTED); 443 } 444 ip->i_d.di_size = 0; 445 ip->i_size = 0; 446 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT); 447 break; 448 449 case S_IFREG: 450 case S_IFLNK: 451 case S_IFDIR: 452 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) { 453 case XFS_DINODE_FMT_LOCAL: 454 /* 455 * no local regular files yet 456 */ 457 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) { 458 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 459 "corrupt inode %Lu " 460 "(local format for regular file).", 461 (unsigned long long) ip->i_ino); 462 XFS_CORRUPTION_ERROR("xfs_iformat(4)", 463 XFS_ERRLEVEL_LOW, 464 ip->i_mount, dip); 465 return XFS_ERROR(EFSCORRUPTED); 466 } 467 468 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT); 469 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { 470 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 471 "corrupt inode %Lu " 472 "(bad size %Ld for local inode).", 473 (unsigned long long) ip->i_ino, 474 (long long) di_size); 475 XFS_CORRUPTION_ERROR("xfs_iformat(5)", 476 XFS_ERRLEVEL_LOW, 477 ip->i_mount, dip); 478 return XFS_ERROR(EFSCORRUPTED); 479 } 480 481 size = (int)di_size; 482 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); 483 break; 484 case XFS_DINODE_FMT_EXTENTS: 485 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); 486 break; 487 case XFS_DINODE_FMT_BTREE: 488 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); 489 break; 490 default: 491 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, 492 ip->i_mount); 493 return XFS_ERROR(EFSCORRUPTED); 494 } 495 break; 496 497 default: 498 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); 499 return XFS_ERROR(EFSCORRUPTED); 500 } 501 if (error) { 502 return error; 503 } 504 if (!XFS_DFORK_Q(dip)) 505 return 0; 506 ASSERT(ip->i_afp == NULL); 507 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); 508 ip->i_afp->if_ext_max = 509 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); 510 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) { 511 case XFS_DINODE_FMT_LOCAL: 512 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); 513 size = be16_to_cpu(atp->hdr.totsize); 514 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); 515 break; 516 case XFS_DINODE_FMT_EXTENTS: 517 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); 518 break; 519 case XFS_DINODE_FMT_BTREE: 520 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); 521 break; 522 default: 523 error = XFS_ERROR(EFSCORRUPTED); 524 break; 525 } 526 if (error) { 527 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 528 ip->i_afp = NULL; 529 xfs_idestroy_fork(ip, XFS_DATA_FORK); 530 } 531 return error; 532 } 533 534 /* 535 * The file is in-lined in the on-disk inode. 536 * If it fits into if_inline_data, then copy 537 * it there, otherwise allocate a buffer for it 538 * and copy the data there. Either way, set 539 * if_data to point at the data. 540 * If we allocate a buffer for the data, make 541 * sure that its size is a multiple of 4 and 542 * record the real size in i_real_bytes. 543 */ 544 STATIC int 545 xfs_iformat_local( 546 xfs_inode_t *ip, 547 xfs_dinode_t *dip, 548 int whichfork, 549 int size) 550 { 551 xfs_ifork_t *ifp; 552 int real_size; 553 554 /* 555 * If the size is unreasonable, then something 556 * is wrong and we just bail out rather than crash in 557 * kmem_alloc() or memcpy() below. 558 */ 559 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 560 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 561 "corrupt inode %Lu " 562 "(bad size %d for local fork, size = %d).", 563 (unsigned long long) ip->i_ino, size, 564 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); 565 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, 566 ip->i_mount, dip); 567 return XFS_ERROR(EFSCORRUPTED); 568 } 569 ifp = XFS_IFORK_PTR(ip, whichfork); 570 real_size = 0; 571 if (size == 0) 572 ifp->if_u1.if_data = NULL; 573 else if (size <= sizeof(ifp->if_u2.if_inline_data)) 574 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 575 else { 576 real_size = roundup(size, 4); 577 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); 578 } 579 ifp->if_bytes = size; 580 ifp->if_real_bytes = real_size; 581 if (size) 582 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); 583 ifp->if_flags &= ~XFS_IFEXTENTS; 584 ifp->if_flags |= XFS_IFINLINE; 585 return 0; 586 } 587 588 /* 589 * The file consists of a set of extents all 590 * of which fit into the on-disk inode. 591 * If there are few enough extents to fit into 592 * the if_inline_ext, then copy them there. 593 * Otherwise allocate a buffer for them and copy 594 * them into it. Either way, set if_extents 595 * to point at the extents. 596 */ 597 STATIC int 598 xfs_iformat_extents( 599 xfs_inode_t *ip, 600 xfs_dinode_t *dip, 601 int whichfork) 602 { 603 xfs_bmbt_rec_t *ep, *dp; 604 xfs_ifork_t *ifp; 605 int nex; 606 int size; 607 int i; 608 609 ifp = XFS_IFORK_PTR(ip, whichfork); 610 nex = XFS_DFORK_NEXTENTS(dip, whichfork); 611 size = nex * (uint)sizeof(xfs_bmbt_rec_t); 612 613 /* 614 * If the number of extents is unreasonable, then something 615 * is wrong and we just bail out rather than crash in 616 * kmem_alloc() or memcpy() below. 617 */ 618 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 619 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 620 "corrupt inode %Lu ((a)extents = %d).", 621 (unsigned long long) ip->i_ino, nex); 622 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, 623 ip->i_mount, dip); 624 return XFS_ERROR(EFSCORRUPTED); 625 } 626 627 ifp->if_real_bytes = 0; 628 if (nex == 0) 629 ifp->if_u1.if_extents = NULL; 630 else if (nex <= XFS_INLINE_EXTS) 631 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 632 else 633 xfs_iext_add(ifp, 0, nex); 634 635 ifp->if_bytes = size; 636 if (size) { 637 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); 638 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip)); 639 for (i = 0; i < nex; i++, dp++) { 640 ep = xfs_iext_get_ext(ifp, i); 641 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0), 642 ARCH_CONVERT); 643 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1), 644 ARCH_CONVERT); 645 } 646 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex, 647 whichfork); 648 if (whichfork != XFS_DATA_FORK || 649 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) 650 if (unlikely(xfs_check_nostate_extents( 651 ifp, 0, nex))) { 652 XFS_ERROR_REPORT("xfs_iformat_extents(2)", 653 XFS_ERRLEVEL_LOW, 654 ip->i_mount); 655 return XFS_ERROR(EFSCORRUPTED); 656 } 657 } 658 ifp->if_flags |= XFS_IFEXTENTS; 659 return 0; 660 } 661 662 /* 663 * The file has too many extents to fit into 664 * the inode, so they are in B-tree format. 665 * Allocate a buffer for the root of the B-tree 666 * and copy the root into it. The i_extents 667 * field will remain NULL until all of the 668 * extents are read in (when they are needed). 669 */ 670 STATIC int 671 xfs_iformat_btree( 672 xfs_inode_t *ip, 673 xfs_dinode_t *dip, 674 int whichfork) 675 { 676 xfs_bmdr_block_t *dfp; 677 xfs_ifork_t *ifp; 678 /* REFERENCED */ 679 int nrecs; 680 int size; 681 682 ifp = XFS_IFORK_PTR(ip, whichfork); 683 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); 684 size = XFS_BMAP_BROOT_SPACE(dfp); 685 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp); 686 687 /* 688 * blow out if -- fork has less extents than can fit in 689 * fork (fork shouldn't be a btree format), root btree 690 * block has more records than can fit into the fork, 691 * or the number of extents is greater than the number of 692 * blocks. 693 */ 694 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max 695 || XFS_BMDR_SPACE_CALC(nrecs) > 696 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) 697 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { 698 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, 699 "corrupt inode %Lu (btree).", 700 (unsigned long long) ip->i_ino); 701 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, 702 ip->i_mount); 703 return XFS_ERROR(EFSCORRUPTED); 704 } 705 706 ifp->if_broot_bytes = size; 707 ifp->if_broot = kmem_alloc(size, KM_SLEEP); 708 ASSERT(ifp->if_broot != NULL); 709 /* 710 * Copy and convert from the on-disk structure 711 * to the in-memory structure. 712 */ 713 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), 714 ifp->if_broot, size); 715 ifp->if_flags &= ~XFS_IFEXTENTS; 716 ifp->if_flags |= XFS_IFBROOT; 717 718 return 0; 719 } 720 721 /* 722 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk 723 * and native format 724 * 725 * buf = on-disk representation 726 * dip = native representation 727 * dir = direction - +ve -> disk to native 728 * -ve -> native to disk 729 */ 730 void 731 xfs_xlate_dinode_core( 732 xfs_caddr_t buf, 733 xfs_dinode_core_t *dip, 734 int dir) 735 { 736 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf; 737 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip; 738 xfs_arch_t arch = ARCH_CONVERT; 739 740 ASSERT(dir); 741 742 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch); 743 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch); 744 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch); 745 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch); 746 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch); 747 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch); 748 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch); 749 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch); 750 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch); 751 752 if (dir > 0) { 753 memcpy(mem_core->di_pad, buf_core->di_pad, 754 sizeof(buf_core->di_pad)); 755 } else { 756 memcpy(buf_core->di_pad, mem_core->di_pad, 757 sizeof(buf_core->di_pad)); 758 } 759 760 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch); 761 762 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec, 763 dir, arch); 764 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec, 765 dir, arch); 766 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec, 767 dir, arch); 768 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec, 769 dir, arch); 770 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec, 771 dir, arch); 772 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec, 773 dir, arch); 774 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch); 775 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch); 776 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch); 777 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch); 778 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch); 779 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch); 780 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch); 781 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch); 782 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch); 783 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch); 784 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch); 785 } 786 787 STATIC uint 788 _xfs_dic2xflags( 789 __uint16_t di_flags) 790 { 791 uint flags = 0; 792 793 if (di_flags & XFS_DIFLAG_ANY) { 794 if (di_flags & XFS_DIFLAG_REALTIME) 795 flags |= XFS_XFLAG_REALTIME; 796 if (di_flags & XFS_DIFLAG_PREALLOC) 797 flags |= XFS_XFLAG_PREALLOC; 798 if (di_flags & XFS_DIFLAG_IMMUTABLE) 799 flags |= XFS_XFLAG_IMMUTABLE; 800 if (di_flags & XFS_DIFLAG_APPEND) 801 flags |= XFS_XFLAG_APPEND; 802 if (di_flags & XFS_DIFLAG_SYNC) 803 flags |= XFS_XFLAG_SYNC; 804 if (di_flags & XFS_DIFLAG_NOATIME) 805 flags |= XFS_XFLAG_NOATIME; 806 if (di_flags & XFS_DIFLAG_NODUMP) 807 flags |= XFS_XFLAG_NODUMP; 808 if (di_flags & XFS_DIFLAG_RTINHERIT) 809 flags |= XFS_XFLAG_RTINHERIT; 810 if (di_flags & XFS_DIFLAG_PROJINHERIT) 811 flags |= XFS_XFLAG_PROJINHERIT; 812 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 813 flags |= XFS_XFLAG_NOSYMLINKS; 814 if (di_flags & XFS_DIFLAG_EXTSIZE) 815 flags |= XFS_XFLAG_EXTSIZE; 816 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 817 flags |= XFS_XFLAG_EXTSZINHERIT; 818 if (di_flags & XFS_DIFLAG_NODEFRAG) 819 flags |= XFS_XFLAG_NODEFRAG; 820 } 821 822 return flags; 823 } 824 825 uint 826 xfs_ip2xflags( 827 xfs_inode_t *ip) 828 { 829 xfs_dinode_core_t *dic = &ip->i_d; 830 831 return _xfs_dic2xflags(dic->di_flags) | 832 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0); 833 } 834 835 uint 836 xfs_dic2xflags( 837 xfs_dinode_core_t *dic) 838 { 839 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) | 840 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0); 841 } 842 843 /* 844 * Given a mount structure and an inode number, return a pointer 845 * to a newly allocated in-core inode corresponding to the given 846 * inode number. 847 * 848 * Initialize the inode's attributes and extent pointers if it 849 * already has them (it will not if the inode has no links). 850 */ 851 int 852 xfs_iread( 853 xfs_mount_t *mp, 854 xfs_trans_t *tp, 855 xfs_ino_t ino, 856 xfs_inode_t **ipp, 857 xfs_daddr_t bno, 858 uint imap_flags) 859 { 860 xfs_buf_t *bp; 861 xfs_dinode_t *dip; 862 xfs_inode_t *ip; 863 int error; 864 865 ASSERT(xfs_inode_zone != NULL); 866 867 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP); 868 ip->i_ino = ino; 869 ip->i_mount = mp; 870 spin_lock_init(&ip->i_flags_lock); 871 872 /* 873 * Get pointer's to the on-disk inode and the buffer containing it. 874 * If the inode number refers to a block outside the file system 875 * then xfs_itobp() will return NULL. In this case we should 876 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will 877 * know that this is a new incore inode. 878 */ 879 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags); 880 if (error) { 881 kmem_zone_free(xfs_inode_zone, ip); 882 return error; 883 } 884 885 /* 886 * Initialize inode's trace buffers. 887 * Do this before xfs_iformat in case it adds entries. 888 */ 889 #ifdef XFS_BMAP_TRACE 890 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP); 891 #endif 892 #ifdef XFS_BMBT_TRACE 893 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP); 894 #endif 895 #ifdef XFS_RW_TRACE 896 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP); 897 #endif 898 #ifdef XFS_ILOCK_TRACE 899 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP); 900 #endif 901 #ifdef XFS_DIR2_TRACE 902 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP); 903 #endif 904 905 /* 906 * If we got something that isn't an inode it means someone 907 * (nfs or dmi) has a stale handle. 908 */ 909 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) { 910 kmem_zone_free(xfs_inode_zone, ip); 911 xfs_trans_brelse(tp, bp); 912 #ifdef DEBUG 913 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " 914 "dip->di_core.di_magic (0x%x) != " 915 "XFS_DINODE_MAGIC (0x%x)", 916 INT_GET(dip->di_core.di_magic, ARCH_CONVERT), 917 XFS_DINODE_MAGIC); 918 #endif /* DEBUG */ 919 return XFS_ERROR(EINVAL); 920 } 921 922 /* 923 * If the on-disk inode is already linked to a directory 924 * entry, copy all of the inode into the in-core inode. 925 * xfs_iformat() handles copying in the inode format 926 * specific information. 927 * Otherwise, just get the truly permanent information. 928 */ 929 if (dip->di_core.di_mode) { 930 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core, 931 &(ip->i_d), 1); 932 error = xfs_iformat(ip, dip); 933 if (error) { 934 kmem_zone_free(xfs_inode_zone, ip); 935 xfs_trans_brelse(tp, bp); 936 #ifdef DEBUG 937 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " 938 "xfs_iformat() returned error %d", 939 error); 940 #endif /* DEBUG */ 941 return error; 942 } 943 } else { 944 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT); 945 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT); 946 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT); 947 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT); 948 /* 949 * Make sure to pull in the mode here as well in 950 * case the inode is released without being used. 951 * This ensures that xfs_inactive() will see that 952 * the inode is already free and not try to mess 953 * with the uninitialized part of it. 954 */ 955 ip->i_d.di_mode = 0; 956 /* 957 * Initialize the per-fork minima and maxima for a new 958 * inode here. xfs_iformat will do it for old inodes. 959 */ 960 ip->i_df.if_ext_max = 961 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); 962 } 963 964 INIT_LIST_HEAD(&ip->i_reclaim); 965 966 /* 967 * The inode format changed when we moved the link count and 968 * made it 32 bits long. If this is an old format inode, 969 * convert it in memory to look like a new one. If it gets 970 * flushed to disk we will convert back before flushing or 971 * logging it. We zero out the new projid field and the old link 972 * count field. We'll handle clearing the pad field (the remains 973 * of the old uuid field) when we actually convert the inode to 974 * the new format. We don't change the version number so that we 975 * can distinguish this from a real new format inode. 976 */ 977 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { 978 ip->i_d.di_nlink = ip->i_d.di_onlink; 979 ip->i_d.di_onlink = 0; 980 ip->i_d.di_projid = 0; 981 } 982 983 ip->i_delayed_blks = 0; 984 ip->i_size = ip->i_d.di_size; 985 986 /* 987 * Mark the buffer containing the inode as something to keep 988 * around for a while. This helps to keep recently accessed 989 * meta-data in-core longer. 990 */ 991 XFS_BUF_SET_REF(bp, XFS_INO_REF); 992 993 /* 994 * Use xfs_trans_brelse() to release the buffer containing the 995 * on-disk inode, because it was acquired with xfs_trans_read_buf() 996 * in xfs_itobp() above. If tp is NULL, this is just a normal 997 * brelse(). If we're within a transaction, then xfs_trans_brelse() 998 * will only release the buffer if it is not dirty within the 999 * transaction. It will be OK to release the buffer in this case, 1000 * because inodes on disk are never destroyed and we will be 1001 * locking the new in-core inode before putting it in the hash 1002 * table where other processes can find it. Thus we don't have 1003 * to worry about the inode being changed just because we released 1004 * the buffer. 1005 */ 1006 xfs_trans_brelse(tp, bp); 1007 *ipp = ip; 1008 return 0; 1009 } 1010 1011 /* 1012 * Read in extents from a btree-format inode. 1013 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. 1014 */ 1015 int 1016 xfs_iread_extents( 1017 xfs_trans_t *tp, 1018 xfs_inode_t *ip, 1019 int whichfork) 1020 { 1021 int error; 1022 xfs_ifork_t *ifp; 1023 xfs_extnum_t nextents; 1024 size_t size; 1025 1026 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { 1027 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, 1028 ip->i_mount); 1029 return XFS_ERROR(EFSCORRUPTED); 1030 } 1031 nextents = XFS_IFORK_NEXTENTS(ip, whichfork); 1032 size = nextents * sizeof(xfs_bmbt_rec_t); 1033 ifp = XFS_IFORK_PTR(ip, whichfork); 1034 1035 /* 1036 * We know that the size is valid (it's checked in iformat_btree) 1037 */ 1038 ifp->if_lastex = NULLEXTNUM; 1039 ifp->if_bytes = ifp->if_real_bytes = 0; 1040 ifp->if_flags |= XFS_IFEXTENTS; 1041 xfs_iext_add(ifp, 0, nextents); 1042 error = xfs_bmap_read_extents(tp, ip, whichfork); 1043 if (error) { 1044 xfs_iext_destroy(ifp); 1045 ifp->if_flags &= ~XFS_IFEXTENTS; 1046 return error; 1047 } 1048 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip)); 1049 return 0; 1050 } 1051 1052 /* 1053 * Allocate an inode on disk and return a copy of its in-core version. 1054 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 1055 * appropriately within the inode. The uid and gid for the inode are 1056 * set according to the contents of the given cred structure. 1057 * 1058 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 1059 * has a free inode available, call xfs_iget() 1060 * to obtain the in-core version of the allocated inode. Finally, 1061 * fill in the inode and log its initial contents. In this case, 1062 * ialloc_context would be set to NULL and call_again set to false. 1063 * 1064 * If xfs_dialloc() does not have an available inode, 1065 * it will replenish its supply by doing an allocation. Since we can 1066 * only do one allocation within a transaction without deadlocks, we 1067 * must commit the current transaction before returning the inode itself. 1068 * In this case, therefore, we will set call_again to true and return. 1069 * The caller should then commit the current transaction, start a new 1070 * transaction, and call xfs_ialloc() again to actually get the inode. 1071 * 1072 * To ensure that some other process does not grab the inode that 1073 * was allocated during the first call to xfs_ialloc(), this routine 1074 * also returns the [locked] bp pointing to the head of the freelist 1075 * as ialloc_context. The caller should hold this buffer across 1076 * the commit and pass it back into this routine on the second call. 1077 */ 1078 int 1079 xfs_ialloc( 1080 xfs_trans_t *tp, 1081 xfs_inode_t *pip, 1082 mode_t mode, 1083 xfs_nlink_t nlink, 1084 xfs_dev_t rdev, 1085 cred_t *cr, 1086 xfs_prid_t prid, 1087 int okalloc, 1088 xfs_buf_t **ialloc_context, 1089 boolean_t *call_again, 1090 xfs_inode_t **ipp) 1091 { 1092 xfs_ino_t ino; 1093 xfs_inode_t *ip; 1094 bhv_vnode_t *vp; 1095 uint flags; 1096 int error; 1097 1098 /* 1099 * Call the space management code to pick 1100 * the on-disk inode to be allocated. 1101 */ 1102 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc, 1103 ialloc_context, call_again, &ino); 1104 if (error != 0) { 1105 return error; 1106 } 1107 if (*call_again || ino == NULLFSINO) { 1108 *ipp = NULL; 1109 return 0; 1110 } 1111 ASSERT(*ialloc_context == NULL); 1112 1113 /* 1114 * Get the in-core inode with the lock held exclusively. 1115 * This is because we're setting fields here we need 1116 * to prevent others from looking at until we're done. 1117 */ 1118 error = xfs_trans_iget(tp->t_mountp, tp, ino, 1119 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip); 1120 if (error != 0) { 1121 return error; 1122 } 1123 ASSERT(ip != NULL); 1124 1125 vp = XFS_ITOV(ip); 1126 ip->i_d.di_mode = (__uint16_t)mode; 1127 ip->i_d.di_onlink = 0; 1128 ip->i_d.di_nlink = nlink; 1129 ASSERT(ip->i_d.di_nlink == nlink); 1130 ip->i_d.di_uid = current_fsuid(cr); 1131 ip->i_d.di_gid = current_fsgid(cr); 1132 ip->i_d.di_projid = prid; 1133 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 1134 1135 /* 1136 * If the superblock version is up to where we support new format 1137 * inodes and this is currently an old format inode, then change 1138 * the inode version number now. This way we only do the conversion 1139 * here rather than here and in the flush/logging code. 1140 */ 1141 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) && 1142 ip->i_d.di_version == XFS_DINODE_VERSION_1) { 1143 ip->i_d.di_version = XFS_DINODE_VERSION_2; 1144 /* 1145 * We've already zeroed the old link count, the projid field, 1146 * and the pad field. 1147 */ 1148 } 1149 1150 /* 1151 * Project ids won't be stored on disk if we are using a version 1 inode. 1152 */ 1153 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) 1154 xfs_bump_ino_vers2(tp, ip); 1155 1156 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) { 1157 ip->i_d.di_gid = pip->i_d.di_gid; 1158 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) { 1159 ip->i_d.di_mode |= S_ISGID; 1160 } 1161 } 1162 1163 /* 1164 * If the group ID of the new file does not match the effective group 1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 1166 * (and only if the irix_sgid_inherit compatibility variable is set). 1167 */ 1168 if ((irix_sgid_inherit) && 1169 (ip->i_d.di_mode & S_ISGID) && 1170 (!in_group_p((gid_t)ip->i_d.di_gid))) { 1171 ip->i_d.di_mode &= ~S_ISGID; 1172 } 1173 1174 ip->i_d.di_size = 0; 1175 ip->i_size = 0; 1176 ip->i_d.di_nextents = 0; 1177 ASSERT(ip->i_d.di_nblocks == 0); 1178 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD); 1179 /* 1180 * di_gen will have been taken care of in xfs_iread. 1181 */ 1182 ip->i_d.di_extsize = 0; 1183 ip->i_d.di_dmevmask = 0; 1184 ip->i_d.di_dmstate = 0; 1185 ip->i_d.di_flags = 0; 1186 flags = XFS_ILOG_CORE; 1187 switch (mode & S_IFMT) { 1188 case S_IFIFO: 1189 case S_IFCHR: 1190 case S_IFBLK: 1191 case S_IFSOCK: 1192 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 1193 ip->i_df.if_u2.if_rdev = rdev; 1194 ip->i_df.if_flags = 0; 1195 flags |= XFS_ILOG_DEV; 1196 break; 1197 case S_IFREG: 1198 case S_IFDIR: 1199 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 1200 uint di_flags = 0; 1201 1202 if ((mode & S_IFMT) == S_IFDIR) { 1203 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 1204 di_flags |= XFS_DIFLAG_RTINHERIT; 1205 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1206 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 1207 ip->i_d.di_extsize = pip->i_d.di_extsize; 1208 } 1209 } else if ((mode & S_IFMT) == S_IFREG) { 1210 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) { 1211 di_flags |= XFS_DIFLAG_REALTIME; 1212 ip->i_iocore.io_flags |= XFS_IOCORE_RT; 1213 } 1214 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1215 di_flags |= XFS_DIFLAG_EXTSIZE; 1216 ip->i_d.di_extsize = pip->i_d.di_extsize; 1217 } 1218 } 1219 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 1220 xfs_inherit_noatime) 1221 di_flags |= XFS_DIFLAG_NOATIME; 1222 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 1223 xfs_inherit_nodump) 1224 di_flags |= XFS_DIFLAG_NODUMP; 1225 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 1226 xfs_inherit_sync) 1227 di_flags |= XFS_DIFLAG_SYNC; 1228 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 1229 xfs_inherit_nosymlinks) 1230 di_flags |= XFS_DIFLAG_NOSYMLINKS; 1231 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 1232 di_flags |= XFS_DIFLAG_PROJINHERIT; 1233 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 1234 xfs_inherit_nodefrag) 1235 di_flags |= XFS_DIFLAG_NODEFRAG; 1236 ip->i_d.di_flags |= di_flags; 1237 } 1238 /* FALLTHROUGH */ 1239 case S_IFLNK: 1240 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 1241 ip->i_df.if_flags = XFS_IFEXTENTS; 1242 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 1243 ip->i_df.if_u1.if_extents = NULL; 1244 break; 1245 default: 1246 ASSERT(0); 1247 } 1248 /* 1249 * Attribute fork settings for new inode. 1250 */ 1251 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 1252 ip->i_d.di_anextents = 0; 1253 1254 /* 1255 * Log the new values stuffed into the inode. 1256 */ 1257 xfs_trans_log_inode(tp, ip, flags); 1258 1259 /* now that we have an i_mode we can setup inode ops and unlock */ 1260 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1); 1261 1262 *ipp = ip; 1263 return 0; 1264 } 1265 1266 /* 1267 * Check to make sure that there are no blocks allocated to the 1268 * file beyond the size of the file. We don't check this for 1269 * files with fixed size extents or real time extents, but we 1270 * at least do it for regular files. 1271 */ 1272 #ifdef DEBUG 1273 void 1274 xfs_isize_check( 1275 xfs_mount_t *mp, 1276 xfs_inode_t *ip, 1277 xfs_fsize_t isize) 1278 { 1279 xfs_fileoff_t map_first; 1280 int nimaps; 1281 xfs_bmbt_irec_t imaps[2]; 1282 1283 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG) 1284 return; 1285 1286 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE)) 1287 return; 1288 1289 nimaps = 2; 1290 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); 1291 /* 1292 * The filesystem could be shutting down, so bmapi may return 1293 * an error. 1294 */ 1295 if (xfs_bmapi(NULL, ip, map_first, 1296 (XFS_B_TO_FSB(mp, 1297 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) - 1298 map_first), 1299 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps, 1300 NULL, NULL)) 1301 return; 1302 ASSERT(nimaps == 1); 1303 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK); 1304 } 1305 #endif /* DEBUG */ 1306 1307 /* 1308 * Calculate the last possible buffered byte in a file. This must 1309 * include data that was buffered beyond the EOF by the write code. 1310 * This also needs to deal with overflowing the xfs_fsize_t type 1311 * which can happen for sizes near the limit. 1312 * 1313 * We also need to take into account any blocks beyond the EOF. It 1314 * may be the case that they were buffered by a write which failed. 1315 * In that case the pages will still be in memory, but the inode size 1316 * will never have been updated. 1317 */ 1318 xfs_fsize_t 1319 xfs_file_last_byte( 1320 xfs_inode_t *ip) 1321 { 1322 xfs_mount_t *mp; 1323 xfs_fsize_t last_byte; 1324 xfs_fileoff_t last_block; 1325 xfs_fileoff_t size_last_block; 1326 int error; 1327 1328 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS)); 1329 1330 mp = ip->i_mount; 1331 /* 1332 * Only check for blocks beyond the EOF if the extents have 1333 * been read in. This eliminates the need for the inode lock, 1334 * and it also saves us from looking when it really isn't 1335 * necessary. 1336 */ 1337 if (ip->i_df.if_flags & XFS_IFEXTENTS) { 1338 error = xfs_bmap_last_offset(NULL, ip, &last_block, 1339 XFS_DATA_FORK); 1340 if (error) { 1341 last_block = 0; 1342 } 1343 } else { 1344 last_block = 0; 1345 } 1346 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size); 1347 last_block = XFS_FILEOFF_MAX(last_block, size_last_block); 1348 1349 last_byte = XFS_FSB_TO_B(mp, last_block); 1350 if (last_byte < 0) { 1351 return XFS_MAXIOFFSET(mp); 1352 } 1353 last_byte += (1 << mp->m_writeio_log); 1354 if (last_byte < 0) { 1355 return XFS_MAXIOFFSET(mp); 1356 } 1357 return last_byte; 1358 } 1359 1360 #if defined(XFS_RW_TRACE) 1361 STATIC void 1362 xfs_itrunc_trace( 1363 int tag, 1364 xfs_inode_t *ip, 1365 int flag, 1366 xfs_fsize_t new_size, 1367 xfs_off_t toss_start, 1368 xfs_off_t toss_finish) 1369 { 1370 if (ip->i_rwtrace == NULL) { 1371 return; 1372 } 1373 1374 ktrace_enter(ip->i_rwtrace, 1375 (void*)((long)tag), 1376 (void*)ip, 1377 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff), 1378 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff), 1379 (void*)((long)flag), 1380 (void*)(unsigned long)((new_size >> 32) & 0xffffffff), 1381 (void*)(unsigned long)(new_size & 0xffffffff), 1382 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff), 1383 (void*)(unsigned long)(toss_start & 0xffffffff), 1384 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff), 1385 (void*)(unsigned long)(toss_finish & 0xffffffff), 1386 (void*)(unsigned long)current_cpu(), 1387 (void*)(unsigned long)current_pid(), 1388 (void*)NULL, 1389 (void*)NULL, 1390 (void*)NULL); 1391 } 1392 #else 1393 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish) 1394 #endif 1395 1396 /* 1397 * Start the truncation of the file to new_size. The new size 1398 * must be smaller than the current size. This routine will 1399 * clear the buffer and page caches of file data in the removed 1400 * range, and xfs_itruncate_finish() will remove the underlying 1401 * disk blocks. 1402 * 1403 * The inode must have its I/O lock locked EXCLUSIVELY, and it 1404 * must NOT have the inode lock held at all. This is because we're 1405 * calling into the buffer/page cache code and we can't hold the 1406 * inode lock when we do so. 1407 * 1408 * We need to wait for any direct I/Os in flight to complete before we 1409 * proceed with the truncate. This is needed to prevent the extents 1410 * being read or written by the direct I/Os from being removed while the 1411 * I/O is in flight as there is no other method of synchronising 1412 * direct I/O with the truncate operation. Also, because we hold 1413 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being 1414 * started until the truncate completes and drops the lock. Essentially, 1415 * the vn_iowait() call forms an I/O barrier that provides strict ordering 1416 * between direct I/Os and the truncate operation. 1417 * 1418 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE 1419 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used 1420 * in the case that the caller is locking things out of order and 1421 * may not be able to call xfs_itruncate_finish() with the inode lock 1422 * held without dropping the I/O lock. If the caller must drop the 1423 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start() 1424 * must be called again with all the same restrictions as the initial 1425 * call. 1426 */ 1427 int 1428 xfs_itruncate_start( 1429 xfs_inode_t *ip, 1430 uint flags, 1431 xfs_fsize_t new_size) 1432 { 1433 xfs_fsize_t last_byte; 1434 xfs_off_t toss_start; 1435 xfs_mount_t *mp; 1436 bhv_vnode_t *vp; 1437 int error = 0; 1438 1439 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); 1440 ASSERT((new_size == 0) || (new_size <= ip->i_size)); 1441 ASSERT((flags == XFS_ITRUNC_DEFINITE) || 1442 (flags == XFS_ITRUNC_MAYBE)); 1443 1444 mp = ip->i_mount; 1445 vp = XFS_ITOV(ip); 1446 1447 vn_iowait(vp); /* wait for the completion of any pending DIOs */ 1448 1449 /* 1450 * Call toss_pages or flushinval_pages to get rid of pages 1451 * overlapping the region being removed. We have to use 1452 * the less efficient flushinval_pages in the case that the 1453 * caller may not be able to finish the truncate without 1454 * dropping the inode's I/O lock. Make sure 1455 * to catch any pages brought in by buffers overlapping 1456 * the EOF by searching out beyond the isize by our 1457 * block size. We round new_size up to a block boundary 1458 * so that we don't toss things on the same block as 1459 * new_size but before it. 1460 * 1461 * Before calling toss_page or flushinval_pages, make sure to 1462 * call remapf() over the same region if the file is mapped. 1463 * This frees up mapped file references to the pages in the 1464 * given range and for the flushinval_pages case it ensures 1465 * that we get the latest mapped changes flushed out. 1466 */ 1467 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1468 toss_start = XFS_FSB_TO_B(mp, toss_start); 1469 if (toss_start < 0) { 1470 /* 1471 * The place to start tossing is beyond our maximum 1472 * file size, so there is no way that the data extended 1473 * out there. 1474 */ 1475 return 0; 1476 } 1477 last_byte = xfs_file_last_byte(ip); 1478 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start, 1479 last_byte); 1480 if (last_byte > toss_start) { 1481 if (flags & XFS_ITRUNC_DEFINITE) { 1482 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED); 1483 } else { 1484 error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED); 1485 } 1486 } 1487 1488 #ifdef DEBUG 1489 if (new_size == 0) { 1490 ASSERT(VN_CACHED(vp) == 0); 1491 } 1492 #endif 1493 return error; 1494 } 1495 1496 /* 1497 * Shrink the file to the given new_size. The new 1498 * size must be smaller than the current size. 1499 * This will free up the underlying blocks 1500 * in the removed range after a call to xfs_itruncate_start() 1501 * or xfs_atruncate_start(). 1502 * 1503 * The transaction passed to this routine must have made 1504 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES. 1505 * This routine may commit the given transaction and 1506 * start new ones, so make sure everything involved in 1507 * the transaction is tidy before calling here. 1508 * Some transaction will be returned to the caller to be 1509 * committed. The incoming transaction must already include 1510 * the inode, and both inode locks must be held exclusively. 1511 * The inode must also be "held" within the transaction. On 1512 * return the inode will be "held" within the returned transaction. 1513 * This routine does NOT require any disk space to be reserved 1514 * for it within the transaction. 1515 * 1516 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, 1517 * and it indicates the fork which is to be truncated. For the 1518 * attribute fork we only support truncation to size 0. 1519 * 1520 * We use the sync parameter to indicate whether or not the first 1521 * transaction we perform might have to be synchronous. For the attr fork, 1522 * it needs to be so if the unlink of the inode is not yet known to be 1523 * permanent in the log. This keeps us from freeing and reusing the 1524 * blocks of the attribute fork before the unlink of the inode becomes 1525 * permanent. 1526 * 1527 * For the data fork, we normally have to run synchronously if we're 1528 * being called out of the inactive path or we're being called 1529 * out of the create path where we're truncating an existing file. 1530 * Either way, the truncate needs to be sync so blocks don't reappear 1531 * in the file with altered data in case of a crash. wsync filesystems 1532 * can run the first case async because anything that shrinks the inode 1533 * has to run sync so by the time we're called here from inactive, the 1534 * inode size is permanently set to 0. 1535 * 1536 * Calls from the truncate path always need to be sync unless we're 1537 * in a wsync filesystem and the file has already been unlinked. 1538 * 1539 * The caller is responsible for correctly setting the sync parameter. 1540 * It gets too hard for us to guess here which path we're being called 1541 * out of just based on inode state. 1542 */ 1543 int 1544 xfs_itruncate_finish( 1545 xfs_trans_t **tp, 1546 xfs_inode_t *ip, 1547 xfs_fsize_t new_size, 1548 int fork, 1549 int sync) 1550 { 1551 xfs_fsblock_t first_block; 1552 xfs_fileoff_t first_unmap_block; 1553 xfs_fileoff_t last_block; 1554 xfs_filblks_t unmap_len=0; 1555 xfs_mount_t *mp; 1556 xfs_trans_t *ntp; 1557 int done; 1558 int committed; 1559 xfs_bmap_free_t free_list; 1560 int error; 1561 1562 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); 1563 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0); 1564 ASSERT((new_size == 0) || (new_size <= ip->i_size)); 1565 ASSERT(*tp != NULL); 1566 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); 1567 ASSERT(ip->i_transp == *tp); 1568 ASSERT(ip->i_itemp != NULL); 1569 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD); 1570 1571 1572 ntp = *tp; 1573 mp = (ntp)->t_mountp; 1574 ASSERT(! XFS_NOT_DQATTACHED(mp, ip)); 1575 1576 /* 1577 * We only support truncating the entire attribute fork. 1578 */ 1579 if (fork == XFS_ATTR_FORK) { 1580 new_size = 0LL; 1581 } 1582 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1583 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0); 1584 /* 1585 * The first thing we do is set the size to new_size permanently 1586 * on disk. This way we don't have to worry about anyone ever 1587 * being able to look at the data being freed even in the face 1588 * of a crash. What we're getting around here is the case where 1589 * we free a block, it is allocated to another file, it is written 1590 * to, and then we crash. If the new data gets written to the 1591 * file but the log buffers containing the free and reallocation 1592 * don't, then we'd end up with garbage in the blocks being freed. 1593 * As long as we make the new_size permanent before actually 1594 * freeing any blocks it doesn't matter if they get writtten to. 1595 * 1596 * The callers must signal into us whether or not the size 1597 * setting here must be synchronous. There are a few cases 1598 * where it doesn't have to be synchronous. Those cases 1599 * occur if the file is unlinked and we know the unlink is 1600 * permanent or if the blocks being truncated are guaranteed 1601 * to be beyond the inode eof (regardless of the link count) 1602 * and the eof value is permanent. Both of these cases occur 1603 * only on wsync-mounted filesystems. In those cases, we're 1604 * guaranteed that no user will ever see the data in the blocks 1605 * that are being truncated so the truncate can run async. 1606 * In the free beyond eof case, the file may wind up with 1607 * more blocks allocated to it than it needs if we crash 1608 * and that won't get fixed until the next time the file 1609 * is re-opened and closed but that's ok as that shouldn't 1610 * be too many blocks. 1611 * 1612 * However, we can't just make all wsync xactions run async 1613 * because there's one call out of the create path that needs 1614 * to run sync where it's truncating an existing file to size 1615 * 0 whose size is > 0. 1616 * 1617 * It's probably possible to come up with a test in this 1618 * routine that would correctly distinguish all the above 1619 * cases from the values of the function parameters and the 1620 * inode state but for sanity's sake, I've decided to let the 1621 * layers above just tell us. It's simpler to correctly figure 1622 * out in the layer above exactly under what conditions we 1623 * can run async and I think it's easier for others read and 1624 * follow the logic in case something has to be changed. 1625 * cscope is your friend -- rcc. 1626 * 1627 * The attribute fork is much simpler. 1628 * 1629 * For the attribute fork we allow the caller to tell us whether 1630 * the unlink of the inode that led to this call is yet permanent 1631 * in the on disk log. If it is not and we will be freeing extents 1632 * in this inode then we make the first transaction synchronous 1633 * to make sure that the unlink is permanent by the time we free 1634 * the blocks. 1635 */ 1636 if (fork == XFS_DATA_FORK) { 1637 if (ip->i_d.di_nextents > 0) { 1638 /* 1639 * If we are not changing the file size then do 1640 * not update the on-disk file size - we may be 1641 * called from xfs_inactive_free_eofblocks(). If we 1642 * update the on-disk file size and then the system 1643 * crashes before the contents of the file are 1644 * flushed to disk then the files may be full of 1645 * holes (ie NULL files bug). 1646 */ 1647 if (ip->i_size != new_size) { 1648 ip->i_d.di_size = new_size; 1649 ip->i_size = new_size; 1650 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); 1651 } 1652 } 1653 } else if (sync) { 1654 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC)); 1655 if (ip->i_d.di_anextents > 0) 1656 xfs_trans_set_sync(ntp); 1657 } 1658 ASSERT(fork == XFS_DATA_FORK || 1659 (fork == XFS_ATTR_FORK && 1660 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) || 1661 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC))))); 1662 1663 /* 1664 * Since it is possible for space to become allocated beyond 1665 * the end of the file (in a crash where the space is allocated 1666 * but the inode size is not yet updated), simply remove any 1667 * blocks which show up between the new EOF and the maximum 1668 * possible file size. If the first block to be removed is 1669 * beyond the maximum file size (ie it is the same as last_block), 1670 * then there is nothing to do. 1671 */ 1672 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); 1673 ASSERT(first_unmap_block <= last_block); 1674 done = 0; 1675 if (last_block == first_unmap_block) { 1676 done = 1; 1677 } else { 1678 unmap_len = last_block - first_unmap_block + 1; 1679 } 1680 while (!done) { 1681 /* 1682 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi() 1683 * will tell us whether it freed the entire range or 1684 * not. If this is a synchronous mount (wsync), 1685 * then we can tell bunmapi to keep all the 1686 * transactions asynchronous since the unlink 1687 * transaction that made this inode inactive has 1688 * already hit the disk. There's no danger of 1689 * the freed blocks being reused, there being a 1690 * crash, and the reused blocks suddenly reappearing 1691 * in this file with garbage in them once recovery 1692 * runs. 1693 */ 1694 XFS_BMAP_INIT(&free_list, &first_block); 1695 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore, 1696 first_unmap_block, unmap_len, 1697 XFS_BMAPI_AFLAG(fork) | 1698 (sync ? 0 : XFS_BMAPI_ASYNC), 1699 XFS_ITRUNC_MAX_EXTENTS, 1700 &first_block, &free_list, 1701 NULL, &done); 1702 if (error) { 1703 /* 1704 * If the bunmapi call encounters an error, 1705 * return to the caller where the transaction 1706 * can be properly aborted. We just need to 1707 * make sure we're not holding any resources 1708 * that we were not when we came in. 1709 */ 1710 xfs_bmap_cancel(&free_list); 1711 return error; 1712 } 1713 1714 /* 1715 * Duplicate the transaction that has the permanent 1716 * reservation and commit the old transaction. 1717 */ 1718 error = xfs_bmap_finish(tp, &free_list, &committed); 1719 ntp = *tp; 1720 if (error) { 1721 /* 1722 * If the bmap finish call encounters an error, 1723 * return to the caller where the transaction 1724 * can be properly aborted. We just need to 1725 * make sure we're not holding any resources 1726 * that we were not when we came in. 1727 * 1728 * Aborting from this point might lose some 1729 * blocks in the file system, but oh well. 1730 */ 1731 xfs_bmap_cancel(&free_list); 1732 if (committed) { 1733 /* 1734 * If the passed in transaction committed 1735 * in xfs_bmap_finish(), then we want to 1736 * add the inode to this one before returning. 1737 * This keeps things simple for the higher 1738 * level code, because it always knows that 1739 * the inode is locked and held in the 1740 * transaction that returns to it whether 1741 * errors occur or not. We don't mark the 1742 * inode dirty so that this transaction can 1743 * be easily aborted if possible. 1744 */ 1745 xfs_trans_ijoin(ntp, ip, 1746 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); 1747 xfs_trans_ihold(ntp, ip); 1748 } 1749 return error; 1750 } 1751 1752 if (committed) { 1753 /* 1754 * The first xact was committed, 1755 * so add the inode to the new one. 1756 * Mark it dirty so it will be logged 1757 * and moved forward in the log as 1758 * part of every commit. 1759 */ 1760 xfs_trans_ijoin(ntp, ip, 1761 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); 1762 xfs_trans_ihold(ntp, ip); 1763 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); 1764 } 1765 ntp = xfs_trans_dup(ntp); 1766 (void) xfs_trans_commit(*tp, 0); 1767 *tp = ntp; 1768 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 1769 XFS_TRANS_PERM_LOG_RES, 1770 XFS_ITRUNCATE_LOG_COUNT); 1771 /* 1772 * Add the inode being truncated to the next chained 1773 * transaction. 1774 */ 1775 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); 1776 xfs_trans_ihold(ntp, ip); 1777 if (error) 1778 return (error); 1779 } 1780 /* 1781 * Only update the size in the case of the data fork, but 1782 * always re-log the inode so that our permanent transaction 1783 * can keep on rolling it forward in the log. 1784 */ 1785 if (fork == XFS_DATA_FORK) { 1786 xfs_isize_check(mp, ip, new_size); 1787 /* 1788 * If we are not changing the file size then do 1789 * not update the on-disk file size - we may be 1790 * called from xfs_inactive_free_eofblocks(). If we 1791 * update the on-disk file size and then the system 1792 * crashes before the contents of the file are 1793 * flushed to disk then the files may be full of 1794 * holes (ie NULL files bug). 1795 */ 1796 if (ip->i_size != new_size) { 1797 ip->i_d.di_size = new_size; 1798 ip->i_size = new_size; 1799 } 1800 } 1801 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); 1802 ASSERT((new_size != 0) || 1803 (fork == XFS_ATTR_FORK) || 1804 (ip->i_delayed_blks == 0)); 1805 ASSERT((new_size != 0) || 1806 (fork == XFS_ATTR_FORK) || 1807 (ip->i_d.di_nextents == 0)); 1808 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0); 1809 return 0; 1810 } 1811 1812 1813 /* 1814 * xfs_igrow_start 1815 * 1816 * Do the first part of growing a file: zero any data in the last 1817 * block that is beyond the old EOF. We need to do this before 1818 * the inode is joined to the transaction to modify the i_size. 1819 * That way we can drop the inode lock and call into the buffer 1820 * cache to get the buffer mapping the EOF. 1821 */ 1822 int 1823 xfs_igrow_start( 1824 xfs_inode_t *ip, 1825 xfs_fsize_t new_size, 1826 cred_t *credp) 1827 { 1828 int error; 1829 1830 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); 1831 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); 1832 ASSERT(new_size > ip->i_size); 1833 1834 /* 1835 * Zero any pages that may have been created by 1836 * xfs_write_file() beyond the end of the file 1837 * and any blocks between the old and new file sizes. 1838 */ 1839 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, 1840 ip->i_size); 1841 return error; 1842 } 1843 1844 /* 1845 * xfs_igrow_finish 1846 * 1847 * This routine is called to extend the size of a file. 1848 * The inode must have both the iolock and the ilock locked 1849 * for update and it must be a part of the current transaction. 1850 * The xfs_igrow_start() function must have been called previously. 1851 * If the change_flag is not zero, the inode change timestamp will 1852 * be updated. 1853 */ 1854 void 1855 xfs_igrow_finish( 1856 xfs_trans_t *tp, 1857 xfs_inode_t *ip, 1858 xfs_fsize_t new_size, 1859 int change_flag) 1860 { 1861 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); 1862 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); 1863 ASSERT(ip->i_transp == tp); 1864 ASSERT(new_size > ip->i_size); 1865 1866 /* 1867 * Update the file size. Update the inode change timestamp 1868 * if change_flag set. 1869 */ 1870 ip->i_d.di_size = new_size; 1871 ip->i_size = new_size; 1872 if (change_flag) 1873 xfs_ichgtime(ip, XFS_ICHGTIME_CHG); 1874 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1875 1876 } 1877 1878 1879 /* 1880 * This is called when the inode's link count goes to 0. 1881 * We place the on-disk inode on a list in the AGI. It 1882 * will be pulled from this list when the inode is freed. 1883 */ 1884 int 1885 xfs_iunlink( 1886 xfs_trans_t *tp, 1887 xfs_inode_t *ip) 1888 { 1889 xfs_mount_t *mp; 1890 xfs_agi_t *agi; 1891 xfs_dinode_t *dip; 1892 xfs_buf_t *agibp; 1893 xfs_buf_t *ibp; 1894 xfs_agnumber_t agno; 1895 xfs_daddr_t agdaddr; 1896 xfs_agino_t agino; 1897 short bucket_index; 1898 int offset; 1899 int error; 1900 int agi_ok; 1901 1902 ASSERT(ip->i_d.di_nlink == 0); 1903 ASSERT(ip->i_d.di_mode != 0); 1904 ASSERT(ip->i_transp == tp); 1905 1906 mp = tp->t_mountp; 1907 1908 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 1909 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); 1910 1911 /* 1912 * Get the agi buffer first. It ensures lock ordering 1913 * on the list. 1914 */ 1915 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, 1916 XFS_FSS_TO_BB(mp, 1), 0, &agibp); 1917 if (error) { 1918 return error; 1919 } 1920 /* 1921 * Validate the magic number of the agi block. 1922 */ 1923 agi = XFS_BUF_TO_AGI(agibp); 1924 agi_ok = 1925 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && 1926 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); 1927 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK, 1928 XFS_RANDOM_IUNLINK))) { 1929 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi); 1930 xfs_trans_brelse(tp, agibp); 1931 return XFS_ERROR(EFSCORRUPTED); 1932 } 1933 /* 1934 * Get the index into the agi hash table for the 1935 * list this inode will go on. 1936 */ 1937 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1938 ASSERT(agino != 0); 1939 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1940 ASSERT(agi->agi_unlinked[bucket_index]); 1941 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1942 1943 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) { 1944 /* 1945 * There is already another inode in the bucket we need 1946 * to add ourselves to. Add us at the front of the list. 1947 * Here we put the head pointer into our next pointer, 1948 * and then we fall through to point the head at us. 1949 */ 1950 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); 1951 if (error) { 1952 return error; 1953 } 1954 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO); 1955 ASSERT(dip->di_next_unlinked); 1956 /* both on-disk, don't endian flip twice */ 1957 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1958 offset = ip->i_boffset + 1959 offsetof(xfs_dinode_t, di_next_unlinked); 1960 xfs_trans_inode_buf(tp, ibp); 1961 xfs_trans_log_buf(tp, ibp, offset, 1962 (offset + sizeof(xfs_agino_t) - 1)); 1963 xfs_inobp_check(mp, ibp); 1964 } 1965 1966 /* 1967 * Point the bucket head pointer at the inode being inserted. 1968 */ 1969 ASSERT(agino != 0); 1970 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1971 offset = offsetof(xfs_agi_t, agi_unlinked) + 1972 (sizeof(xfs_agino_t) * bucket_index); 1973 xfs_trans_log_buf(tp, agibp, offset, 1974 (offset + sizeof(xfs_agino_t) - 1)); 1975 return 0; 1976 } 1977 1978 /* 1979 * Pull the on-disk inode from the AGI unlinked list. 1980 */ 1981 STATIC int 1982 xfs_iunlink_remove( 1983 xfs_trans_t *tp, 1984 xfs_inode_t *ip) 1985 { 1986 xfs_ino_t next_ino; 1987 xfs_mount_t *mp; 1988 xfs_agi_t *agi; 1989 xfs_dinode_t *dip; 1990 xfs_buf_t *agibp; 1991 xfs_buf_t *ibp; 1992 xfs_agnumber_t agno; 1993 xfs_daddr_t agdaddr; 1994 xfs_agino_t agino; 1995 xfs_agino_t next_agino; 1996 xfs_buf_t *last_ibp; 1997 xfs_dinode_t *last_dip = NULL; 1998 short bucket_index; 1999 int offset, last_offset = 0; 2000 int error; 2001 int agi_ok; 2002 2003 /* 2004 * First pull the on-disk inode from the AGI unlinked list. 2005 */ 2006 mp = tp->t_mountp; 2007 2008 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2009 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); 2010 2011 /* 2012 * Get the agi buffer first. It ensures lock ordering 2013 * on the list. 2014 */ 2015 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, 2016 XFS_FSS_TO_BB(mp, 1), 0, &agibp); 2017 if (error) { 2018 cmn_err(CE_WARN, 2019 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.", 2020 error, mp->m_fsname); 2021 return error; 2022 } 2023 /* 2024 * Validate the magic number of the agi block. 2025 */ 2026 agi = XFS_BUF_TO_AGI(agibp); 2027 agi_ok = 2028 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && 2029 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); 2030 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE, 2031 XFS_RANDOM_IUNLINK_REMOVE))) { 2032 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW, 2033 mp, agi); 2034 xfs_trans_brelse(tp, agibp); 2035 cmn_err(CE_WARN, 2036 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.", 2037 mp->m_fsname); 2038 return XFS_ERROR(EFSCORRUPTED); 2039 } 2040 /* 2041 * Get the index into the agi hash table for the 2042 * list this inode will go on. 2043 */ 2044 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2045 ASSERT(agino != 0); 2046 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2047 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO); 2048 ASSERT(agi->agi_unlinked[bucket_index]); 2049 2050 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2051 /* 2052 * We're at the head of the list. Get the inode's 2053 * on-disk buffer to see if there is anyone after us 2054 * on the list. Only modify our next pointer if it 2055 * is not already NULLAGINO. This saves us the overhead 2056 * of dealing with the buffer when there is no need to 2057 * change it. 2058 */ 2059 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); 2060 if (error) { 2061 cmn_err(CE_WARN, 2062 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", 2063 error, mp->m_fsname); 2064 return error; 2065 } 2066 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); 2067 ASSERT(next_agino != 0); 2068 if (next_agino != NULLAGINO) { 2069 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); 2070 offset = ip->i_boffset + 2071 offsetof(xfs_dinode_t, di_next_unlinked); 2072 xfs_trans_inode_buf(tp, ibp); 2073 xfs_trans_log_buf(tp, ibp, offset, 2074 (offset + sizeof(xfs_agino_t) - 1)); 2075 xfs_inobp_check(mp, ibp); 2076 } else { 2077 xfs_trans_brelse(tp, ibp); 2078 } 2079 /* 2080 * Point the bucket head pointer at the next inode. 2081 */ 2082 ASSERT(next_agino != 0); 2083 ASSERT(next_agino != agino); 2084 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2085 offset = offsetof(xfs_agi_t, agi_unlinked) + 2086 (sizeof(xfs_agino_t) * bucket_index); 2087 xfs_trans_log_buf(tp, agibp, offset, 2088 (offset + sizeof(xfs_agino_t) - 1)); 2089 } else { 2090 /* 2091 * We need to search the list for the inode being freed. 2092 */ 2093 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2094 last_ibp = NULL; 2095 while (next_agino != agino) { 2096 /* 2097 * If the last inode wasn't the one pointing to 2098 * us, then release its buffer since we're not 2099 * going to do anything with it. 2100 */ 2101 if (last_ibp != NULL) { 2102 xfs_trans_brelse(tp, last_ibp); 2103 } 2104 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2105 error = xfs_inotobp(mp, tp, next_ino, &last_dip, 2106 &last_ibp, &last_offset); 2107 if (error) { 2108 cmn_err(CE_WARN, 2109 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.", 2110 error, mp->m_fsname); 2111 return error; 2112 } 2113 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT); 2114 ASSERT(next_agino != NULLAGINO); 2115 ASSERT(next_agino != 0); 2116 } 2117 /* 2118 * Now last_ibp points to the buffer previous to us on 2119 * the unlinked list. Pull us from the list. 2120 */ 2121 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); 2122 if (error) { 2123 cmn_err(CE_WARN, 2124 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", 2125 error, mp->m_fsname); 2126 return error; 2127 } 2128 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); 2129 ASSERT(next_agino != 0); 2130 ASSERT(next_agino != agino); 2131 if (next_agino != NULLAGINO) { 2132 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); 2133 offset = ip->i_boffset + 2134 offsetof(xfs_dinode_t, di_next_unlinked); 2135 xfs_trans_inode_buf(tp, ibp); 2136 xfs_trans_log_buf(tp, ibp, offset, 2137 (offset + sizeof(xfs_agino_t) - 1)); 2138 xfs_inobp_check(mp, ibp); 2139 } else { 2140 xfs_trans_brelse(tp, ibp); 2141 } 2142 /* 2143 * Point the previous inode on the list to the next inode. 2144 */ 2145 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino); 2146 ASSERT(next_agino != 0); 2147 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2148 xfs_trans_inode_buf(tp, last_ibp); 2149 xfs_trans_log_buf(tp, last_ibp, offset, 2150 (offset + sizeof(xfs_agino_t) - 1)); 2151 xfs_inobp_check(mp, last_ibp); 2152 } 2153 return 0; 2154 } 2155 2156 STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip) 2157 { 2158 return (((ip->i_itemp == NULL) || 2159 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) && 2160 (ip->i_update_core == 0)); 2161 } 2162 2163 STATIC void 2164 xfs_ifree_cluster( 2165 xfs_inode_t *free_ip, 2166 xfs_trans_t *tp, 2167 xfs_ino_t inum) 2168 { 2169 xfs_mount_t *mp = free_ip->i_mount; 2170 int blks_per_cluster; 2171 int nbufs; 2172 int ninodes; 2173 int i, j, found, pre_flushed; 2174 xfs_daddr_t blkno; 2175 xfs_buf_t *bp; 2176 xfs_ihash_t *ih; 2177 xfs_inode_t *ip, **ip_found; 2178 xfs_inode_log_item_t *iip; 2179 xfs_log_item_t *lip; 2180 SPLDECL(s); 2181 2182 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { 2183 blks_per_cluster = 1; 2184 ninodes = mp->m_sb.sb_inopblock; 2185 nbufs = XFS_IALLOC_BLOCKS(mp); 2186 } else { 2187 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / 2188 mp->m_sb.sb_blocksize; 2189 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; 2190 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; 2191 } 2192 2193 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS); 2194 2195 for (j = 0; j < nbufs; j++, inum += ninodes) { 2196 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2197 XFS_INO_TO_AGBNO(mp, inum)); 2198 2199 2200 /* 2201 * Look for each inode in memory and attempt to lock it, 2202 * we can be racing with flush and tail pushing here. 2203 * any inode we get the locks on, add to an array of 2204 * inode items to process later. 2205 * 2206 * The get the buffer lock, we could beat a flush 2207 * or tail pushing thread to the lock here, in which 2208 * case they will go looking for the inode buffer 2209 * and fail, we need some other form of interlock 2210 * here. 2211 */ 2212 found = 0; 2213 for (i = 0; i < ninodes; i++) { 2214 ih = XFS_IHASH(mp, inum + i); 2215 read_lock(&ih->ih_lock); 2216 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { 2217 if (ip->i_ino == inum + i) 2218 break; 2219 } 2220 2221 /* Inode not in memory or we found it already, 2222 * nothing to do 2223 */ 2224 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) { 2225 read_unlock(&ih->ih_lock); 2226 continue; 2227 } 2228 2229 if (xfs_inode_clean(ip)) { 2230 read_unlock(&ih->ih_lock); 2231 continue; 2232 } 2233 2234 /* If we can get the locks then add it to the 2235 * list, otherwise by the time we get the bp lock 2236 * below it will already be attached to the 2237 * inode buffer. 2238 */ 2239 2240 /* This inode will already be locked - by us, lets 2241 * keep it that way. 2242 */ 2243 2244 if (ip == free_ip) { 2245 if (xfs_iflock_nowait(ip)) { 2246 xfs_iflags_set(ip, XFS_ISTALE); 2247 if (xfs_inode_clean(ip)) { 2248 xfs_ifunlock(ip); 2249 } else { 2250 ip_found[found++] = ip; 2251 } 2252 } 2253 read_unlock(&ih->ih_lock); 2254 continue; 2255 } 2256 2257 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2258 if (xfs_iflock_nowait(ip)) { 2259 xfs_iflags_set(ip, XFS_ISTALE); 2260 2261 if (xfs_inode_clean(ip)) { 2262 xfs_ifunlock(ip); 2263 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2264 } else { 2265 ip_found[found++] = ip; 2266 } 2267 } else { 2268 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2269 } 2270 } 2271 2272 read_unlock(&ih->ih_lock); 2273 } 2274 2275 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2276 mp->m_bsize * blks_per_cluster, 2277 XFS_BUF_LOCK); 2278 2279 pre_flushed = 0; 2280 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); 2281 while (lip) { 2282 if (lip->li_type == XFS_LI_INODE) { 2283 iip = (xfs_inode_log_item_t *)lip; 2284 ASSERT(iip->ili_logged == 1); 2285 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done; 2286 AIL_LOCK(mp,s); 2287 iip->ili_flush_lsn = iip->ili_item.li_lsn; 2288 AIL_UNLOCK(mp, s); 2289 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2290 pre_flushed++; 2291 } 2292 lip = lip->li_bio_list; 2293 } 2294 2295 for (i = 0; i < found; i++) { 2296 ip = ip_found[i]; 2297 iip = ip->i_itemp; 2298 2299 if (!iip) { 2300 ip->i_update_core = 0; 2301 xfs_ifunlock(ip); 2302 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2303 continue; 2304 } 2305 2306 iip->ili_last_fields = iip->ili_format.ilf_fields; 2307 iip->ili_format.ilf_fields = 0; 2308 iip->ili_logged = 1; 2309 AIL_LOCK(mp,s); 2310 iip->ili_flush_lsn = iip->ili_item.li_lsn; 2311 AIL_UNLOCK(mp, s); 2312 2313 xfs_buf_attach_iodone(bp, 2314 (void(*)(xfs_buf_t*,xfs_log_item_t*)) 2315 xfs_istale_done, (xfs_log_item_t *)iip); 2316 if (ip != free_ip) { 2317 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2318 } 2319 } 2320 2321 if (found || pre_flushed) 2322 xfs_trans_stale_inode_buf(tp, bp); 2323 xfs_trans_binval(tp, bp); 2324 } 2325 2326 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *)); 2327 } 2328 2329 /* 2330 * This is called to return an inode to the inode free list. 2331 * The inode should already be truncated to 0 length and have 2332 * no pages associated with it. This routine also assumes that 2333 * the inode is already a part of the transaction. 2334 * 2335 * The on-disk copy of the inode will have been added to the list 2336 * of unlinked inodes in the AGI. We need to remove the inode from 2337 * that list atomically with respect to freeing it here. 2338 */ 2339 int 2340 xfs_ifree( 2341 xfs_trans_t *tp, 2342 xfs_inode_t *ip, 2343 xfs_bmap_free_t *flist) 2344 { 2345 int error; 2346 int delete; 2347 xfs_ino_t first_ino; 2348 2349 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); 2350 ASSERT(ip->i_transp == tp); 2351 ASSERT(ip->i_d.di_nlink == 0); 2352 ASSERT(ip->i_d.di_nextents == 0); 2353 ASSERT(ip->i_d.di_anextents == 0); 2354 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) || 2355 ((ip->i_d.di_mode & S_IFMT) != S_IFREG)); 2356 ASSERT(ip->i_d.di_nblocks == 0); 2357 2358 /* 2359 * Pull the on-disk inode from the AGI unlinked list. 2360 */ 2361 error = xfs_iunlink_remove(tp, ip); 2362 if (error != 0) { 2363 return error; 2364 } 2365 2366 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); 2367 if (error != 0) { 2368 return error; 2369 } 2370 ip->i_d.di_mode = 0; /* mark incore inode as free */ 2371 ip->i_d.di_flags = 0; 2372 ip->i_d.di_dmevmask = 0; 2373 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2374 ip->i_df.if_ext_max = 2375 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); 2376 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2377 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2378 /* 2379 * Bump the generation count so no one will be confused 2380 * by reincarnations of this inode. 2381 */ 2382 ip->i_d.di_gen++; 2383 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2384 2385 if (delete) { 2386 xfs_ifree_cluster(ip, tp, first_ino); 2387 } 2388 2389 return 0; 2390 } 2391 2392 /* 2393 * Reallocate the space for if_broot based on the number of records 2394 * being added or deleted as indicated in rec_diff. Move the records 2395 * and pointers in if_broot to fit the new size. When shrinking this 2396 * will eliminate holes between the records and pointers created by 2397 * the caller. When growing this will create holes to be filled in 2398 * by the caller. 2399 * 2400 * The caller must not request to add more records than would fit in 2401 * the on-disk inode root. If the if_broot is currently NULL, then 2402 * if we adding records one will be allocated. The caller must also 2403 * not request that the number of records go below zero, although 2404 * it can go to zero. 2405 * 2406 * ip -- the inode whose if_broot area is changing 2407 * ext_diff -- the change in the number of records, positive or negative, 2408 * requested for the if_broot array. 2409 */ 2410 void 2411 xfs_iroot_realloc( 2412 xfs_inode_t *ip, 2413 int rec_diff, 2414 int whichfork) 2415 { 2416 int cur_max; 2417 xfs_ifork_t *ifp; 2418 xfs_bmbt_block_t *new_broot; 2419 int new_max; 2420 size_t new_size; 2421 char *np; 2422 char *op; 2423 2424 /* 2425 * Handle the degenerate case quietly. 2426 */ 2427 if (rec_diff == 0) { 2428 return; 2429 } 2430 2431 ifp = XFS_IFORK_PTR(ip, whichfork); 2432 if (rec_diff > 0) { 2433 /* 2434 * If there wasn't any memory allocated before, just 2435 * allocate it now and get out. 2436 */ 2437 if (ifp->if_broot_bytes == 0) { 2438 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); 2439 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size, 2440 KM_SLEEP); 2441 ifp->if_broot_bytes = (int)new_size; 2442 return; 2443 } 2444 2445 /* 2446 * If there is already an existing if_broot, then we need 2447 * to realloc() it and shift the pointers to their new 2448 * location. The records don't change location because 2449 * they are kept butted up against the btree block header. 2450 */ 2451 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); 2452 new_max = cur_max + rec_diff; 2453 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); 2454 ifp->if_broot = (xfs_bmbt_block_t *) 2455 kmem_realloc(ifp->if_broot, 2456 new_size, 2457 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ 2458 KM_SLEEP); 2459 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, 2460 ifp->if_broot_bytes); 2461 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, 2462 (int)new_size); 2463 ifp->if_broot_bytes = (int)new_size; 2464 ASSERT(ifp->if_broot_bytes <= 2465 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); 2466 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); 2467 return; 2468 } 2469 2470 /* 2471 * rec_diff is less than 0. In this case, we are shrinking the 2472 * if_broot buffer. It must already exist. If we go to zero 2473 * records, just get rid of the root and clear the status bit. 2474 */ 2475 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); 2476 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); 2477 new_max = cur_max + rec_diff; 2478 ASSERT(new_max >= 0); 2479 if (new_max > 0) 2480 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); 2481 else 2482 new_size = 0; 2483 if (new_size > 0) { 2484 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP); 2485 /* 2486 * First copy over the btree block header. 2487 */ 2488 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t)); 2489 } else { 2490 new_broot = NULL; 2491 ifp->if_flags &= ~XFS_IFBROOT; 2492 } 2493 2494 /* 2495 * Only copy the records and pointers if there are any. 2496 */ 2497 if (new_max > 0) { 2498 /* 2499 * First copy the records. 2500 */ 2501 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1, 2502 ifp->if_broot_bytes); 2503 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1, 2504 (int)new_size); 2505 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); 2506 2507 /* 2508 * Then copy the pointers. 2509 */ 2510 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, 2511 ifp->if_broot_bytes); 2512 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1, 2513 (int)new_size); 2514 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); 2515 } 2516 kmem_free(ifp->if_broot, ifp->if_broot_bytes); 2517 ifp->if_broot = new_broot; 2518 ifp->if_broot_bytes = (int)new_size; 2519 ASSERT(ifp->if_broot_bytes <= 2520 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); 2521 return; 2522 } 2523 2524 2525 /* 2526 * This is called when the amount of space needed for if_data 2527 * is increased or decreased. The change in size is indicated by 2528 * the number of bytes that need to be added or deleted in the 2529 * byte_diff parameter. 2530 * 2531 * If the amount of space needed has decreased below the size of the 2532 * inline buffer, then switch to using the inline buffer. Otherwise, 2533 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer 2534 * to what is needed. 2535 * 2536 * ip -- the inode whose if_data area is changing 2537 * byte_diff -- the change in the number of bytes, positive or negative, 2538 * requested for the if_data array. 2539 */ 2540 void 2541 xfs_idata_realloc( 2542 xfs_inode_t *ip, 2543 int byte_diff, 2544 int whichfork) 2545 { 2546 xfs_ifork_t *ifp; 2547 int new_size; 2548 int real_size; 2549 2550 if (byte_diff == 0) { 2551 return; 2552 } 2553 2554 ifp = XFS_IFORK_PTR(ip, whichfork); 2555 new_size = (int)ifp->if_bytes + byte_diff; 2556 ASSERT(new_size >= 0); 2557 2558 if (new_size == 0) { 2559 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2560 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); 2561 } 2562 ifp->if_u1.if_data = NULL; 2563 real_size = 0; 2564 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { 2565 /* 2566 * If the valid extents/data can fit in if_inline_ext/data, 2567 * copy them from the malloc'd vector and free it. 2568 */ 2569 if (ifp->if_u1.if_data == NULL) { 2570 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2571 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2572 ASSERT(ifp->if_real_bytes != 0); 2573 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, 2574 new_size); 2575 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); 2576 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2577 } 2578 real_size = 0; 2579 } else { 2580 /* 2581 * Stuck with malloc/realloc. 2582 * For inline data, the underlying buffer must be 2583 * a multiple of 4 bytes in size so that it can be 2584 * logged and stay on word boundaries. We enforce 2585 * that here. 2586 */ 2587 real_size = roundup(new_size, 4); 2588 if (ifp->if_u1.if_data == NULL) { 2589 ASSERT(ifp->if_real_bytes == 0); 2590 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); 2591 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2592 /* 2593 * Only do the realloc if the underlying size 2594 * is really changing. 2595 */ 2596 if (ifp->if_real_bytes != real_size) { 2597 ifp->if_u1.if_data = 2598 kmem_realloc(ifp->if_u1.if_data, 2599 real_size, 2600 ifp->if_real_bytes, 2601 KM_SLEEP); 2602 } 2603 } else { 2604 ASSERT(ifp->if_real_bytes == 0); 2605 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); 2606 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, 2607 ifp->if_bytes); 2608 } 2609 } 2610 ifp->if_real_bytes = real_size; 2611 ifp->if_bytes = new_size; 2612 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2613 } 2614 2615 2616 2617 2618 /* 2619 * Map inode to disk block and offset. 2620 * 2621 * mp -- the mount point structure for the current file system 2622 * tp -- the current transaction 2623 * ino -- the inode number of the inode to be located 2624 * imap -- this structure is filled in with the information necessary 2625 * to retrieve the given inode from disk 2626 * flags -- flags to pass to xfs_dilocate indicating whether or not 2627 * lookups in the inode btree were OK or not 2628 */ 2629 int 2630 xfs_imap( 2631 xfs_mount_t *mp, 2632 xfs_trans_t *tp, 2633 xfs_ino_t ino, 2634 xfs_imap_t *imap, 2635 uint flags) 2636 { 2637 xfs_fsblock_t fsbno; 2638 int len; 2639 int off; 2640 int error; 2641 2642 fsbno = imap->im_blkno ? 2643 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; 2644 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); 2645 if (error != 0) { 2646 return error; 2647 } 2648 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); 2649 imap->im_len = XFS_FSB_TO_BB(mp, len); 2650 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); 2651 imap->im_ioffset = (ushort)off; 2652 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); 2653 return 0; 2654 } 2655 2656 void 2657 xfs_idestroy_fork( 2658 xfs_inode_t *ip, 2659 int whichfork) 2660 { 2661 xfs_ifork_t *ifp; 2662 2663 ifp = XFS_IFORK_PTR(ip, whichfork); 2664 if (ifp->if_broot != NULL) { 2665 kmem_free(ifp->if_broot, ifp->if_broot_bytes); 2666 ifp->if_broot = NULL; 2667 } 2668 2669 /* 2670 * If the format is local, then we can't have an extents 2671 * array so just look for an inline data array. If we're 2672 * not local then we may or may not have an extents list, 2673 * so check and free it up if we do. 2674 */ 2675 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { 2676 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && 2677 (ifp->if_u1.if_data != NULL)) { 2678 ASSERT(ifp->if_real_bytes != 0); 2679 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); 2680 ifp->if_u1.if_data = NULL; 2681 ifp->if_real_bytes = 0; 2682 } 2683 } else if ((ifp->if_flags & XFS_IFEXTENTS) && 2684 ((ifp->if_flags & XFS_IFEXTIREC) || 2685 ((ifp->if_u1.if_extents != NULL) && 2686 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { 2687 ASSERT(ifp->if_real_bytes != 0); 2688 xfs_iext_destroy(ifp); 2689 } 2690 ASSERT(ifp->if_u1.if_extents == NULL || 2691 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); 2692 ASSERT(ifp->if_real_bytes == 0); 2693 if (whichfork == XFS_ATTR_FORK) { 2694 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 2695 ip->i_afp = NULL; 2696 } 2697 } 2698 2699 /* 2700 * This is called free all the memory associated with an inode. 2701 * It must free the inode itself and any buffers allocated for 2702 * if_extents/if_data and if_broot. It must also free the lock 2703 * associated with the inode. 2704 */ 2705 void 2706 xfs_idestroy( 2707 xfs_inode_t *ip) 2708 { 2709 2710 switch (ip->i_d.di_mode & S_IFMT) { 2711 case S_IFREG: 2712 case S_IFDIR: 2713 case S_IFLNK: 2714 xfs_idestroy_fork(ip, XFS_DATA_FORK); 2715 break; 2716 } 2717 if (ip->i_afp) 2718 xfs_idestroy_fork(ip, XFS_ATTR_FORK); 2719 mrfree(&ip->i_lock); 2720 mrfree(&ip->i_iolock); 2721 freesema(&ip->i_flock); 2722 #ifdef XFS_BMAP_TRACE 2723 ktrace_free(ip->i_xtrace); 2724 #endif 2725 #ifdef XFS_BMBT_TRACE 2726 ktrace_free(ip->i_btrace); 2727 #endif 2728 #ifdef XFS_RW_TRACE 2729 ktrace_free(ip->i_rwtrace); 2730 #endif 2731 #ifdef XFS_ILOCK_TRACE 2732 ktrace_free(ip->i_lock_trace); 2733 #endif 2734 #ifdef XFS_DIR2_TRACE 2735 ktrace_free(ip->i_dir_trace); 2736 #endif 2737 if (ip->i_itemp) { 2738 /* 2739 * Only if we are shutting down the fs will we see an 2740 * inode still in the AIL. If it is there, we should remove 2741 * it to prevent a use-after-free from occurring. 2742 */ 2743 xfs_mount_t *mp = ip->i_mount; 2744 xfs_log_item_t *lip = &ip->i_itemp->ili_item; 2745 int s; 2746 2747 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || 2748 XFS_FORCED_SHUTDOWN(ip->i_mount)); 2749 if (lip->li_flags & XFS_LI_IN_AIL) { 2750 AIL_LOCK(mp, s); 2751 if (lip->li_flags & XFS_LI_IN_AIL) 2752 xfs_trans_delete_ail(mp, lip, s); 2753 else 2754 AIL_UNLOCK(mp, s); 2755 } 2756 xfs_inode_item_destroy(ip); 2757 } 2758 kmem_zone_free(xfs_inode_zone, ip); 2759 } 2760 2761 2762 /* 2763 * Increment the pin count of the given buffer. 2764 * This value is protected by ipinlock spinlock in the mount structure. 2765 */ 2766 void 2767 xfs_ipin( 2768 xfs_inode_t *ip) 2769 { 2770 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); 2771 2772 atomic_inc(&ip->i_pincount); 2773 } 2774 2775 /* 2776 * Decrement the pin count of the given inode, and wake up 2777 * anyone in xfs_iwait_unpin() if the count goes to 0. The 2778 * inode must have been previously pinned with a call to xfs_ipin(). 2779 */ 2780 void 2781 xfs_iunpin( 2782 xfs_inode_t *ip) 2783 { 2784 ASSERT(atomic_read(&ip->i_pincount) > 0); 2785 2786 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) { 2787 2788 /* 2789 * If the inode is currently being reclaimed, the link between 2790 * the bhv_vnode and the xfs_inode will be broken after the 2791 * XFS_IRECLAIM* flag is set. Hence, if these flags are not 2792 * set, then we can move forward and mark the linux inode dirty 2793 * knowing that it is still valid as it won't freed until after 2794 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The 2795 * i_flags_lock is used to synchronise the setting of the 2796 * XFS_IRECLAIM* flags and the breaking of the link, and so we 2797 * can execute atomically w.r.t to reclaim by holding this lock 2798 * here. 2799 * 2800 * However, we still need to issue the unpin wakeup call as the 2801 * inode reclaim may be blocked waiting for the inode to become 2802 * unpinned. 2803 */ 2804 2805 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) { 2806 bhv_vnode_t *vp = XFS_ITOV_NULL(ip); 2807 struct inode *inode = NULL; 2808 2809 BUG_ON(vp == NULL); 2810 inode = vn_to_inode(vp); 2811 BUG_ON(inode->i_state & I_CLEAR); 2812 2813 /* make sync come back and flush this inode */ 2814 if (!(inode->i_state & (I_NEW|I_FREEING))) 2815 mark_inode_dirty_sync(inode); 2816 } 2817 spin_unlock(&ip->i_flags_lock); 2818 wake_up(&ip->i_ipin_wait); 2819 } 2820 } 2821 2822 /* 2823 * This is called to wait for the given inode to be unpinned. 2824 * It will sleep until this happens. The caller must have the 2825 * inode locked in at least shared mode so that the buffer cannot 2826 * be subsequently pinned once someone is waiting for it to be 2827 * unpinned. 2828 */ 2829 STATIC void 2830 xfs_iunpin_wait( 2831 xfs_inode_t *ip) 2832 { 2833 xfs_inode_log_item_t *iip; 2834 xfs_lsn_t lsn; 2835 2836 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS)); 2837 2838 if (atomic_read(&ip->i_pincount) == 0) { 2839 return; 2840 } 2841 2842 iip = ip->i_itemp; 2843 if (iip && iip->ili_last_lsn) { 2844 lsn = iip->ili_last_lsn; 2845 } else { 2846 lsn = (xfs_lsn_t)0; 2847 } 2848 2849 /* 2850 * Give the log a push so we don't wait here too long. 2851 */ 2852 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE); 2853 2854 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0)); 2855 } 2856 2857 2858 /* 2859 * xfs_iextents_copy() 2860 * 2861 * This is called to copy the REAL extents (as opposed to the delayed 2862 * allocation extents) from the inode into the given buffer. It 2863 * returns the number of bytes copied into the buffer. 2864 * 2865 * If there are no delayed allocation extents, then we can just 2866 * memcpy() the extents into the buffer. Otherwise, we need to 2867 * examine each extent in turn and skip those which are delayed. 2868 */ 2869 int 2870 xfs_iextents_copy( 2871 xfs_inode_t *ip, 2872 xfs_bmbt_rec_t *buffer, 2873 int whichfork) 2874 { 2875 int copied; 2876 xfs_bmbt_rec_t *dest_ep; 2877 xfs_bmbt_rec_t *ep; 2878 #ifdef XFS_BMAP_TRACE 2879 static char fname[] = "xfs_iextents_copy"; 2880 #endif 2881 int i; 2882 xfs_ifork_t *ifp; 2883 int nrecs; 2884 xfs_fsblock_t start_block; 2885 2886 ifp = XFS_IFORK_PTR(ip, whichfork); 2887 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); 2888 ASSERT(ifp->if_bytes > 0); 2889 2890 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 2891 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork); 2892 ASSERT(nrecs > 0); 2893 2894 /* 2895 * There are some delayed allocation extents in the 2896 * inode, so copy the extents one at a time and skip 2897 * the delayed ones. There must be at least one 2898 * non-delayed extent. 2899 */ 2900 dest_ep = buffer; 2901 copied = 0; 2902 for (i = 0; i < nrecs; i++) { 2903 ep = xfs_iext_get_ext(ifp, i); 2904 start_block = xfs_bmbt_get_startblock(ep); 2905 if (ISNULLSTARTBLOCK(start_block)) { 2906 /* 2907 * It's a delayed allocation extent, so skip it. 2908 */ 2909 continue; 2910 } 2911 2912 /* Translate to on disk format */ 2913 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT), 2914 (__uint64_t*)&dest_ep->l0); 2915 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT), 2916 (__uint64_t*)&dest_ep->l1); 2917 dest_ep++; 2918 copied++; 2919 } 2920 ASSERT(copied != 0); 2921 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip)); 2922 2923 return (copied * (uint)sizeof(xfs_bmbt_rec_t)); 2924 } 2925 2926 /* 2927 * Each of the following cases stores data into the same region 2928 * of the on-disk inode, so only one of them can be valid at 2929 * any given time. While it is possible to have conflicting formats 2930 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is 2931 * in EXTENTS format, this can only happen when the fork has 2932 * changed formats after being modified but before being flushed. 2933 * In these cases, the format always takes precedence, because the 2934 * format indicates the current state of the fork. 2935 */ 2936 /*ARGSUSED*/ 2937 STATIC int 2938 xfs_iflush_fork( 2939 xfs_inode_t *ip, 2940 xfs_dinode_t *dip, 2941 xfs_inode_log_item_t *iip, 2942 int whichfork, 2943 xfs_buf_t *bp) 2944 { 2945 char *cp; 2946 xfs_ifork_t *ifp; 2947 xfs_mount_t *mp; 2948 #ifdef XFS_TRANS_DEBUG 2949 int first; 2950 #endif 2951 static const short brootflag[2] = 2952 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; 2953 static const short dataflag[2] = 2954 { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; 2955 static const short extflag[2] = 2956 { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; 2957 2958 if (iip == NULL) 2959 return 0; 2960 ifp = XFS_IFORK_PTR(ip, whichfork); 2961 /* 2962 * This can happen if we gave up in iformat in an error path, 2963 * for the attribute fork. 2964 */ 2965 if (ifp == NULL) { 2966 ASSERT(whichfork == XFS_ATTR_FORK); 2967 return 0; 2968 } 2969 cp = XFS_DFORK_PTR(dip, whichfork); 2970 mp = ip->i_mount; 2971 switch (XFS_IFORK_FORMAT(ip, whichfork)) { 2972 case XFS_DINODE_FMT_LOCAL: 2973 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && 2974 (ifp->if_bytes > 0)) { 2975 ASSERT(ifp->if_u1.if_data != NULL); 2976 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2977 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); 2978 } 2979 break; 2980 2981 case XFS_DINODE_FMT_EXTENTS: 2982 ASSERT((ifp->if_flags & XFS_IFEXTENTS) || 2983 !(iip->ili_format.ilf_fields & extflag[whichfork])); 2984 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) || 2985 (ifp->if_bytes == 0)); 2986 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) || 2987 (ifp->if_bytes > 0)); 2988 if ((iip->ili_format.ilf_fields & extflag[whichfork]) && 2989 (ifp->if_bytes > 0)) { 2990 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); 2991 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, 2992 whichfork); 2993 } 2994 break; 2995 2996 case XFS_DINODE_FMT_BTREE: 2997 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && 2998 (ifp->if_broot_bytes > 0)) { 2999 ASSERT(ifp->if_broot != NULL); 3000 ASSERT(ifp->if_broot_bytes <= 3001 (XFS_IFORK_SIZE(ip, whichfork) + 3002 XFS_BROOT_SIZE_ADJ)); 3003 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes, 3004 (xfs_bmdr_block_t *)cp, 3005 XFS_DFORK_SIZE(dip, mp, whichfork)); 3006 } 3007 break; 3008 3009 case XFS_DINODE_FMT_DEV: 3010 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { 3011 ASSERT(whichfork == XFS_DATA_FORK); 3012 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev); 3013 } 3014 break; 3015 3016 case XFS_DINODE_FMT_UUID: 3017 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { 3018 ASSERT(whichfork == XFS_DATA_FORK); 3019 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, 3020 sizeof(uuid_t)); 3021 } 3022 break; 3023 3024 default: 3025 ASSERT(0); 3026 break; 3027 } 3028 3029 return 0; 3030 } 3031 3032 /* 3033 * xfs_iflush() will write a modified inode's changes out to the 3034 * inode's on disk home. The caller must have the inode lock held 3035 * in at least shared mode and the inode flush semaphore must be 3036 * held as well. The inode lock will still be held upon return from 3037 * the call and the caller is free to unlock it. 3038 * The inode flush lock will be unlocked when the inode reaches the disk. 3039 * The flags indicate how the inode's buffer should be written out. 3040 */ 3041 int 3042 xfs_iflush( 3043 xfs_inode_t *ip, 3044 uint flags) 3045 { 3046 xfs_inode_log_item_t *iip; 3047 xfs_buf_t *bp; 3048 xfs_dinode_t *dip; 3049 xfs_mount_t *mp; 3050 int error; 3051 /* REFERENCED */ 3052 xfs_chash_t *ch; 3053 xfs_inode_t *iq; 3054 int clcount; /* count of inodes clustered */ 3055 int bufwasdelwri; 3056 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) }; 3057 SPLDECL(s); 3058 3059 XFS_STATS_INC(xs_iflush_count); 3060 3061 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); 3062 ASSERT(issemalocked(&(ip->i_flock))); 3063 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3064 ip->i_d.di_nextents > ip->i_df.if_ext_max); 3065 3066 iip = ip->i_itemp; 3067 mp = ip->i_mount; 3068 3069 /* 3070 * If the inode isn't dirty, then just release the inode 3071 * flush lock and do nothing. 3072 */ 3073 if ((ip->i_update_core == 0) && 3074 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { 3075 ASSERT((iip != NULL) ? 3076 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1); 3077 xfs_ifunlock(ip); 3078 return 0; 3079 } 3080 3081 /* 3082 * We can't flush the inode until it is unpinned, so 3083 * wait for it. We know noone new can pin it, because 3084 * we are holding the inode lock shared and you need 3085 * to hold it exclusively to pin the inode. 3086 */ 3087 xfs_iunpin_wait(ip); 3088 3089 /* 3090 * This may have been unpinned because the filesystem is shutting 3091 * down forcibly. If that's the case we must not write this inode 3092 * to disk, because the log record didn't make it to disk! 3093 */ 3094 if (XFS_FORCED_SHUTDOWN(mp)) { 3095 ip->i_update_core = 0; 3096 if (iip) 3097 iip->ili_format.ilf_fields = 0; 3098 xfs_ifunlock(ip); 3099 return XFS_ERROR(EIO); 3100 } 3101 3102 /* 3103 * Get the buffer containing the on-disk inode. 3104 */ 3105 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0); 3106 if (error) { 3107 xfs_ifunlock(ip); 3108 return error; 3109 } 3110 3111 /* 3112 * Decide how buffer will be flushed out. This is done before 3113 * the call to xfs_iflush_int because this field is zeroed by it. 3114 */ 3115 if (iip != NULL && iip->ili_format.ilf_fields != 0) { 3116 /* 3117 * Flush out the inode buffer according to the directions 3118 * of the caller. In the cases where the caller has given 3119 * us a choice choose the non-delwri case. This is because 3120 * the inode is in the AIL and we need to get it out soon. 3121 */ 3122 switch (flags) { 3123 case XFS_IFLUSH_SYNC: 3124 case XFS_IFLUSH_DELWRI_ELSE_SYNC: 3125 flags = 0; 3126 break; 3127 case XFS_IFLUSH_ASYNC: 3128 case XFS_IFLUSH_DELWRI_ELSE_ASYNC: 3129 flags = INT_ASYNC; 3130 break; 3131 case XFS_IFLUSH_DELWRI: 3132 flags = INT_DELWRI; 3133 break; 3134 default: 3135 ASSERT(0); 3136 flags = 0; 3137 break; 3138 } 3139 } else { 3140 switch (flags) { 3141 case XFS_IFLUSH_DELWRI_ELSE_SYNC: 3142 case XFS_IFLUSH_DELWRI_ELSE_ASYNC: 3143 case XFS_IFLUSH_DELWRI: 3144 flags = INT_DELWRI; 3145 break; 3146 case XFS_IFLUSH_ASYNC: 3147 flags = INT_ASYNC; 3148 break; 3149 case XFS_IFLUSH_SYNC: 3150 flags = 0; 3151 break; 3152 default: 3153 ASSERT(0); 3154 flags = 0; 3155 break; 3156 } 3157 } 3158 3159 /* 3160 * First flush out the inode that xfs_iflush was called with. 3161 */ 3162 error = xfs_iflush_int(ip, bp); 3163 if (error) { 3164 goto corrupt_out; 3165 } 3166 3167 /* 3168 * inode clustering: 3169 * see if other inodes can be gathered into this write 3170 */ 3171 3172 ip->i_chash->chl_buf = bp; 3173 3174 ch = XFS_CHASH(mp, ip->i_blkno); 3175 s = mutex_spinlock(&ch->ch_lock); 3176 3177 clcount = 0; 3178 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) { 3179 /* 3180 * Do an un-protected check to see if the inode is dirty and 3181 * is a candidate for flushing. These checks will be repeated 3182 * later after the appropriate locks are acquired. 3183 */ 3184 iip = iq->i_itemp; 3185 if ((iq->i_update_core == 0) && 3186 ((iip == NULL) || 3187 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) && 3188 xfs_ipincount(iq) == 0) { 3189 continue; 3190 } 3191 3192 /* 3193 * Try to get locks. If any are unavailable, 3194 * then this inode cannot be flushed and is skipped. 3195 */ 3196 3197 /* get inode locks (just i_lock) */ 3198 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) { 3199 /* get inode flush lock */ 3200 if (xfs_iflock_nowait(iq)) { 3201 /* check if pinned */ 3202 if (xfs_ipincount(iq) == 0) { 3203 /* arriving here means that 3204 * this inode can be flushed. 3205 * first re-check that it's 3206 * dirty 3207 */ 3208 iip = iq->i_itemp; 3209 if ((iq->i_update_core != 0)|| 3210 ((iip != NULL) && 3211 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { 3212 clcount++; 3213 error = xfs_iflush_int(iq, bp); 3214 if (error) { 3215 xfs_iunlock(iq, 3216 XFS_ILOCK_SHARED); 3217 goto cluster_corrupt_out; 3218 } 3219 } else { 3220 xfs_ifunlock(iq); 3221 } 3222 } else { 3223 xfs_ifunlock(iq); 3224 } 3225 } 3226 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3227 } 3228 } 3229 mutex_spinunlock(&ch->ch_lock, s); 3230 3231 if (clcount) { 3232 XFS_STATS_INC(xs_icluster_flushcnt); 3233 XFS_STATS_ADD(xs_icluster_flushinode, clcount); 3234 } 3235 3236 /* 3237 * If the buffer is pinned then push on the log so we won't 3238 * get stuck waiting in the write for too long. 3239 */ 3240 if (XFS_BUF_ISPINNED(bp)){ 3241 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); 3242 } 3243 3244 if (flags & INT_DELWRI) { 3245 xfs_bdwrite(mp, bp); 3246 } else if (flags & INT_ASYNC) { 3247 xfs_bawrite(mp, bp); 3248 } else { 3249 error = xfs_bwrite(mp, bp); 3250 } 3251 return error; 3252 3253 corrupt_out: 3254 xfs_buf_relse(bp); 3255 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3256 xfs_iflush_abort(ip); 3257 /* 3258 * Unlocks the flush lock 3259 */ 3260 return XFS_ERROR(EFSCORRUPTED); 3261 3262 cluster_corrupt_out: 3263 /* Corruption detected in the clustering loop. Invalidate the 3264 * inode buffer and shut down the filesystem. 3265 */ 3266 mutex_spinunlock(&ch->ch_lock, s); 3267 3268 /* 3269 * Clean up the buffer. If it was B_DELWRI, just release it -- 3270 * brelse can handle it with no problems. If not, shut down the 3271 * filesystem before releasing the buffer. 3272 */ 3273 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) { 3274 xfs_buf_relse(bp); 3275 } 3276 3277 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3278 3279 if(!bufwasdelwri) { 3280 /* 3281 * Just like incore_relse: if we have b_iodone functions, 3282 * mark the buffer as an error and call them. Otherwise 3283 * mark it as stale and brelse. 3284 */ 3285 if (XFS_BUF_IODONE_FUNC(bp)) { 3286 XFS_BUF_CLR_BDSTRAT_FUNC(bp); 3287 XFS_BUF_UNDONE(bp); 3288 XFS_BUF_STALE(bp); 3289 XFS_BUF_SHUT(bp); 3290 XFS_BUF_ERROR(bp,EIO); 3291 xfs_biodone(bp); 3292 } else { 3293 XFS_BUF_STALE(bp); 3294 xfs_buf_relse(bp); 3295 } 3296 } 3297 3298 xfs_iflush_abort(iq); 3299 /* 3300 * Unlocks the flush lock 3301 */ 3302 return XFS_ERROR(EFSCORRUPTED); 3303 } 3304 3305 3306 STATIC int 3307 xfs_iflush_int( 3308 xfs_inode_t *ip, 3309 xfs_buf_t *bp) 3310 { 3311 xfs_inode_log_item_t *iip; 3312 xfs_dinode_t *dip; 3313 xfs_mount_t *mp; 3314 #ifdef XFS_TRANS_DEBUG 3315 int first; 3316 #endif 3317 SPLDECL(s); 3318 3319 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); 3320 ASSERT(issemalocked(&(ip->i_flock))); 3321 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3322 ip->i_d.di_nextents > ip->i_df.if_ext_max); 3323 3324 iip = ip->i_itemp; 3325 mp = ip->i_mount; 3326 3327 3328 /* 3329 * If the inode isn't dirty, then just release the inode 3330 * flush lock and do nothing. 3331 */ 3332 if ((ip->i_update_core == 0) && 3333 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { 3334 xfs_ifunlock(ip); 3335 return 0; 3336 } 3337 3338 /* set *dip = inode's place in the buffer */ 3339 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset); 3340 3341 /* 3342 * Clear i_update_core before copying out the data. 3343 * This is for coordination with our timestamp updates 3344 * that don't hold the inode lock. They will always 3345 * update the timestamps BEFORE setting i_update_core, 3346 * so if we clear i_update_core after they set it we 3347 * are guaranteed to see their updates to the timestamps. 3348 * I believe that this depends on strongly ordered memory 3349 * semantics, but we have that. We use the SYNCHRONIZE 3350 * macro to make sure that the compiler does not reorder 3351 * the i_update_core access below the data copy below. 3352 */ 3353 ip->i_update_core = 0; 3354 SYNCHRONIZE(); 3355 3356 /* 3357 * Make sure to get the latest atime from the Linux inode. 3358 */ 3359 xfs_synchronize_atime(ip); 3360 3361 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC, 3362 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 3363 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3364 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3365 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip); 3366 goto corrupt_out; 3367 } 3368 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, 3369 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { 3370 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3371 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x", 3372 ip->i_ino, ip, ip->i_d.di_magic); 3373 goto corrupt_out; 3374 } 3375 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { 3376 if (XFS_TEST_ERROR( 3377 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3378 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3379 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 3380 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3381 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p", 3382 ip->i_ino, ip); 3383 goto corrupt_out; 3384 } 3385 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { 3386 if (XFS_TEST_ERROR( 3387 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3388 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3389 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3390 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 3391 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3392 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p", 3393 ip->i_ino, ip); 3394 goto corrupt_out; 3395 } 3396 } 3397 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3398 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 3399 XFS_RANDOM_IFLUSH_5)) { 3400 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3401 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p", 3402 ip->i_ino, 3403 ip->i_d.di_nextents + ip->i_d.di_anextents, 3404 ip->i_d.di_nblocks, 3405 ip); 3406 goto corrupt_out; 3407 } 3408 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3409 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 3410 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, 3411 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3412 ip->i_ino, ip->i_d.di_forkoff, ip); 3413 goto corrupt_out; 3414 } 3415 /* 3416 * bump the flush iteration count, used to detect flushes which 3417 * postdate a log record during recovery. 3418 */ 3419 3420 ip->i_d.di_flushiter++; 3421 3422 /* 3423 * Copy the dirty parts of the inode into the on-disk 3424 * inode. We always copy out the core of the inode, 3425 * because if the inode is dirty at all the core must 3426 * be. 3427 */ 3428 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1); 3429 3430 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3431 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3432 ip->i_d.di_flushiter = 0; 3433 3434 /* 3435 * If this is really an old format inode and the superblock version 3436 * has not been updated to support only new format inodes, then 3437 * convert back to the old inode format. If the superblock version 3438 * has been updated, then make the conversion permanent. 3439 */ 3440 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 || 3441 XFS_SB_VERSION_HASNLINK(&mp->m_sb)); 3442 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { 3443 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { 3444 /* 3445 * Convert it back. 3446 */ 3447 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); 3448 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink); 3449 } else { 3450 /* 3451 * The superblock version has already been bumped, 3452 * so just make the conversion to the new inode 3453 * format permanent. 3454 */ 3455 ip->i_d.di_version = XFS_DINODE_VERSION_2; 3456 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2); 3457 ip->i_d.di_onlink = 0; 3458 dip->di_core.di_onlink = 0; 3459 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 3460 memset(&(dip->di_core.di_pad[0]), 0, 3461 sizeof(dip->di_core.di_pad)); 3462 ASSERT(ip->i_d.di_projid == 0); 3463 } 3464 } 3465 3466 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) { 3467 goto corrupt_out; 3468 } 3469 3470 if (XFS_IFORK_Q(ip)) { 3471 /* 3472 * The only error from xfs_iflush_fork is on the data fork. 3473 */ 3474 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); 3475 } 3476 xfs_inobp_check(mp, bp); 3477 3478 /* 3479 * We've recorded everything logged in the inode, so we'd 3480 * like to clear the ilf_fields bits so we don't log and 3481 * flush things unnecessarily. However, we can't stop 3482 * logging all this information until the data we've copied 3483 * into the disk buffer is written to disk. If we did we might 3484 * overwrite the copy of the inode in the log with all the 3485 * data after re-logging only part of it, and in the face of 3486 * a crash we wouldn't have all the data we need to recover. 3487 * 3488 * What we do is move the bits to the ili_last_fields field. 3489 * When logging the inode, these bits are moved back to the 3490 * ilf_fields field. In the xfs_iflush_done() routine we 3491 * clear ili_last_fields, since we know that the information 3492 * those bits represent is permanently on disk. As long as 3493 * the flush completes before the inode is logged again, then 3494 * both ilf_fields and ili_last_fields will be cleared. 3495 * 3496 * We can play with the ilf_fields bits here, because the inode 3497 * lock must be held exclusively in order to set bits there 3498 * and the flush lock protects the ili_last_fields bits. 3499 * Set ili_logged so the flush done 3500 * routine can tell whether or not to look in the AIL. 3501 * Also, store the current LSN of the inode so that we can tell 3502 * whether the item has moved in the AIL from xfs_iflush_done(). 3503 * In order to read the lsn we need the AIL lock, because 3504 * it is a 64 bit value that cannot be read atomically. 3505 */ 3506 if (iip != NULL && iip->ili_format.ilf_fields != 0) { 3507 iip->ili_last_fields = iip->ili_format.ilf_fields; 3508 iip->ili_format.ilf_fields = 0; 3509 iip->ili_logged = 1; 3510 3511 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */ 3512 AIL_LOCK(mp,s); 3513 iip->ili_flush_lsn = iip->ili_item.li_lsn; 3514 AIL_UNLOCK(mp, s); 3515 3516 /* 3517 * Attach the function xfs_iflush_done to the inode's 3518 * buffer. This will remove the inode from the AIL 3519 * and unlock the inode's flush lock when the inode is 3520 * completely written to disk. 3521 */ 3522 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*)) 3523 xfs_iflush_done, (xfs_log_item_t *)iip); 3524 3525 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); 3526 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL); 3527 } else { 3528 /* 3529 * We're flushing an inode which is not in the AIL and has 3530 * not been logged but has i_update_core set. For this 3531 * case we can use a B_DELWRI flush and immediately drop 3532 * the inode flush lock because we can avoid the whole 3533 * AIL state thing. It's OK to drop the flush lock now, 3534 * because we've already locked the buffer and to do anything 3535 * you really need both. 3536 */ 3537 if (iip != NULL) { 3538 ASSERT(iip->ili_logged == 0); 3539 ASSERT(iip->ili_last_fields == 0); 3540 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); 3541 } 3542 xfs_ifunlock(ip); 3543 } 3544 3545 return 0; 3546 3547 corrupt_out: 3548 return XFS_ERROR(EFSCORRUPTED); 3549 } 3550 3551 3552 /* 3553 * Flush all inactive inodes in mp. 3554 */ 3555 void 3556 xfs_iflush_all( 3557 xfs_mount_t *mp) 3558 { 3559 xfs_inode_t *ip; 3560 bhv_vnode_t *vp; 3561 3562 again: 3563 XFS_MOUNT_ILOCK(mp); 3564 ip = mp->m_inodes; 3565 if (ip == NULL) 3566 goto out; 3567 3568 do { 3569 /* Make sure we skip markers inserted by sync */ 3570 if (ip->i_mount == NULL) { 3571 ip = ip->i_mnext; 3572 continue; 3573 } 3574 3575 vp = XFS_ITOV_NULL(ip); 3576 if (!vp) { 3577 XFS_MOUNT_IUNLOCK(mp); 3578 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC); 3579 goto again; 3580 } 3581 3582 ASSERT(vn_count(vp) == 0); 3583 3584 ip = ip->i_mnext; 3585 } while (ip != mp->m_inodes); 3586 out: 3587 XFS_MOUNT_IUNLOCK(mp); 3588 } 3589 3590 /* 3591 * xfs_iaccess: check accessibility of inode for mode. 3592 */ 3593 int 3594 xfs_iaccess( 3595 xfs_inode_t *ip, 3596 mode_t mode, 3597 cred_t *cr) 3598 { 3599 int error; 3600 mode_t orgmode = mode; 3601 struct inode *inode = vn_to_inode(XFS_ITOV(ip)); 3602 3603 if (mode & S_IWUSR) { 3604 umode_t imode = inode->i_mode; 3605 3606 if (IS_RDONLY(inode) && 3607 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode))) 3608 return XFS_ERROR(EROFS); 3609 3610 if (IS_IMMUTABLE(inode)) 3611 return XFS_ERROR(EACCES); 3612 } 3613 3614 /* 3615 * If there's an Access Control List it's used instead of 3616 * the mode bits. 3617 */ 3618 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1) 3619 return error ? XFS_ERROR(error) : 0; 3620 3621 if (current_fsuid(cr) != ip->i_d.di_uid) { 3622 mode >>= 3; 3623 if (!in_group_p((gid_t)ip->i_d.di_gid)) 3624 mode >>= 3; 3625 } 3626 3627 /* 3628 * If the DACs are ok we don't need any capability check. 3629 */ 3630 if ((ip->i_d.di_mode & mode) == mode) 3631 return 0; 3632 /* 3633 * Read/write DACs are always overridable. 3634 * Executable DACs are overridable if at least one exec bit is set. 3635 */ 3636 if (!(orgmode & S_IXUSR) || 3637 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) 3638 if (capable_cred(cr, CAP_DAC_OVERRIDE)) 3639 return 0; 3640 3641 if ((orgmode == S_IRUSR) || 3642 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) { 3643 if (capable_cred(cr, CAP_DAC_READ_SEARCH)) 3644 return 0; 3645 #ifdef NOISE 3646 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode); 3647 #endif /* NOISE */ 3648 return XFS_ERROR(EACCES); 3649 } 3650 return XFS_ERROR(EACCES); 3651 } 3652 3653 /* 3654 * xfs_iroundup: round up argument to next power of two 3655 */ 3656 uint 3657 xfs_iroundup( 3658 uint v) 3659 { 3660 int i; 3661 uint m; 3662 3663 if ((v & (v - 1)) == 0) 3664 return v; 3665 ASSERT((v & 0x80000000) == 0); 3666 if ((v & (v + 1)) == 0) 3667 return v + 1; 3668 for (i = 0, m = 1; i < 31; i++, m <<= 1) { 3669 if (v & m) 3670 continue; 3671 v |= m; 3672 if ((v & (v + 1)) == 0) 3673 return v + 1; 3674 } 3675 ASSERT(0); 3676 return( 0 ); 3677 } 3678 3679 #ifdef XFS_ILOCK_TRACE 3680 ktrace_t *xfs_ilock_trace_buf; 3681 3682 void 3683 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra) 3684 { 3685 ktrace_enter(ip->i_lock_trace, 3686 (void *)ip, 3687 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */ 3688 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */ 3689 (void *)ra, /* caller of ilock */ 3690 (void *)(unsigned long)current_cpu(), 3691 (void *)(unsigned long)current_pid(), 3692 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL); 3693 } 3694 #endif 3695 3696 /* 3697 * Return a pointer to the extent record at file index idx. 3698 */ 3699 xfs_bmbt_rec_t * 3700 xfs_iext_get_ext( 3701 xfs_ifork_t *ifp, /* inode fork pointer */ 3702 xfs_extnum_t idx) /* index of target extent */ 3703 { 3704 ASSERT(idx >= 0); 3705 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { 3706 return ifp->if_u1.if_ext_irec->er_extbuf; 3707 } else if (ifp->if_flags & XFS_IFEXTIREC) { 3708 xfs_ext_irec_t *erp; /* irec pointer */ 3709 int erp_idx = 0; /* irec index */ 3710 xfs_extnum_t page_idx = idx; /* ext index in target list */ 3711 3712 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 3713 return &erp->er_extbuf[page_idx]; 3714 } else if (ifp->if_bytes) { 3715 return &ifp->if_u1.if_extents[idx]; 3716 } else { 3717 return NULL; 3718 } 3719 } 3720 3721 /* 3722 * Insert new item(s) into the extent records for incore inode 3723 * fork 'ifp'. 'count' new items are inserted at index 'idx'. 3724 */ 3725 void 3726 xfs_iext_insert( 3727 xfs_ifork_t *ifp, /* inode fork pointer */ 3728 xfs_extnum_t idx, /* starting index of new items */ 3729 xfs_extnum_t count, /* number of inserted items */ 3730 xfs_bmbt_irec_t *new) /* items to insert */ 3731 { 3732 xfs_bmbt_rec_t *ep; /* extent record pointer */ 3733 xfs_extnum_t i; /* extent record index */ 3734 3735 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 3736 xfs_iext_add(ifp, idx, count); 3737 for (i = idx; i < idx + count; i++, new++) { 3738 ep = xfs_iext_get_ext(ifp, i); 3739 xfs_bmbt_set_all(ep, new); 3740 } 3741 } 3742 3743 /* 3744 * This is called when the amount of space required for incore file 3745 * extents needs to be increased. The ext_diff parameter stores the 3746 * number of new extents being added and the idx parameter contains 3747 * the extent index where the new extents will be added. If the new 3748 * extents are being appended, then we just need to (re)allocate and 3749 * initialize the space. Otherwise, if the new extents are being 3750 * inserted into the middle of the existing entries, a bit more work 3751 * is required to make room for the new extents to be inserted. The 3752 * caller is responsible for filling in the new extent entries upon 3753 * return. 3754 */ 3755 void 3756 xfs_iext_add( 3757 xfs_ifork_t *ifp, /* inode fork pointer */ 3758 xfs_extnum_t idx, /* index to begin adding exts */ 3759 int ext_diff) /* number of extents to add */ 3760 { 3761 int byte_diff; /* new bytes being added */ 3762 int new_size; /* size of extents after adding */ 3763 xfs_extnum_t nextents; /* number of extents in file */ 3764 3765 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3766 ASSERT((idx >= 0) && (idx <= nextents)); 3767 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); 3768 new_size = ifp->if_bytes + byte_diff; 3769 /* 3770 * If the new number of extents (nextents + ext_diff) 3771 * fits inside the inode, then continue to use the inline 3772 * extent buffer. 3773 */ 3774 if (nextents + ext_diff <= XFS_INLINE_EXTS) { 3775 if (idx < nextents) { 3776 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], 3777 &ifp->if_u2.if_inline_ext[idx], 3778 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 3779 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); 3780 } 3781 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 3782 ifp->if_real_bytes = 0; 3783 ifp->if_lastex = nextents + ext_diff; 3784 } 3785 /* 3786 * Otherwise use a linear (direct) extent list. 3787 * If the extents are currently inside the inode, 3788 * xfs_iext_realloc_direct will switch us from 3789 * inline to direct extent allocation mode. 3790 */ 3791 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { 3792 xfs_iext_realloc_direct(ifp, new_size); 3793 if (idx < nextents) { 3794 memmove(&ifp->if_u1.if_extents[idx + ext_diff], 3795 &ifp->if_u1.if_extents[idx], 3796 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 3797 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); 3798 } 3799 } 3800 /* Indirection array */ 3801 else { 3802 xfs_ext_irec_t *erp; 3803 int erp_idx = 0; 3804 int page_idx = idx; 3805 3806 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); 3807 if (ifp->if_flags & XFS_IFEXTIREC) { 3808 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); 3809 } else { 3810 xfs_iext_irec_init(ifp); 3811 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3812 erp = ifp->if_u1.if_ext_irec; 3813 } 3814 /* Extents fit in target extent page */ 3815 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { 3816 if (page_idx < erp->er_extcount) { 3817 memmove(&erp->er_extbuf[page_idx + ext_diff], 3818 &erp->er_extbuf[page_idx], 3819 (erp->er_extcount - page_idx) * 3820 sizeof(xfs_bmbt_rec_t)); 3821 memset(&erp->er_extbuf[page_idx], 0, byte_diff); 3822 } 3823 erp->er_extcount += ext_diff; 3824 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3825 } 3826 /* Insert a new extent page */ 3827 else if (erp) { 3828 xfs_iext_add_indirect_multi(ifp, 3829 erp_idx, page_idx, ext_diff); 3830 } 3831 /* 3832 * If extent(s) are being appended to the last page in 3833 * the indirection array and the new extent(s) don't fit 3834 * in the page, then erp is NULL and erp_idx is set to 3835 * the next index needed in the indirection array. 3836 */ 3837 else { 3838 int count = ext_diff; 3839 3840 while (count) { 3841 erp = xfs_iext_irec_new(ifp, erp_idx); 3842 erp->er_extcount = count; 3843 count -= MIN(count, (int)XFS_LINEAR_EXTS); 3844 if (count) { 3845 erp_idx++; 3846 } 3847 } 3848 } 3849 } 3850 ifp->if_bytes = new_size; 3851 } 3852 3853 /* 3854 * This is called when incore extents are being added to the indirection 3855 * array and the new extents do not fit in the target extent list. The 3856 * erp_idx parameter contains the irec index for the target extent list 3857 * in the indirection array, and the idx parameter contains the extent 3858 * index within the list. The number of extents being added is stored 3859 * in the count parameter. 3860 * 3861 * |-------| |-------| 3862 * | | | | idx - number of extents before idx 3863 * | idx | | count | 3864 * | | | | count - number of extents being inserted at idx 3865 * |-------| |-------| 3866 * | count | | nex2 | nex2 - number of extents after idx + count 3867 * |-------| |-------| 3868 */ 3869 void 3870 xfs_iext_add_indirect_multi( 3871 xfs_ifork_t *ifp, /* inode fork pointer */ 3872 int erp_idx, /* target extent irec index */ 3873 xfs_extnum_t idx, /* index within target list */ 3874 int count) /* new extents being added */ 3875 { 3876 int byte_diff; /* new bytes being added */ 3877 xfs_ext_irec_t *erp; /* pointer to irec entry */ 3878 xfs_extnum_t ext_diff; /* number of extents to add */ 3879 xfs_extnum_t ext_cnt; /* new extents still needed */ 3880 xfs_extnum_t nex2; /* extents after idx + count */ 3881 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ 3882 int nlists; /* number of irec's (lists) */ 3883 3884 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3885 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3886 nex2 = erp->er_extcount - idx; 3887 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3888 3889 /* 3890 * Save second part of target extent list 3891 * (all extents past */ 3892 if (nex2) { 3893 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3894 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP); 3895 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); 3896 erp->er_extcount -= nex2; 3897 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); 3898 memset(&erp->er_extbuf[idx], 0, byte_diff); 3899 } 3900 3901 /* 3902 * Add the new extents to the end of the target 3903 * list, then allocate new irec record(s) and 3904 * extent buffer(s) as needed to store the rest 3905 * of the new extents. 3906 */ 3907 ext_cnt = count; 3908 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); 3909 if (ext_diff) { 3910 erp->er_extcount += ext_diff; 3911 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3912 ext_cnt -= ext_diff; 3913 } 3914 while (ext_cnt) { 3915 erp_idx++; 3916 erp = xfs_iext_irec_new(ifp, erp_idx); 3917 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); 3918 erp->er_extcount = ext_diff; 3919 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3920 ext_cnt -= ext_diff; 3921 } 3922 3923 /* Add nex2 extents back to indirection array */ 3924 if (nex2) { 3925 xfs_extnum_t ext_avail; 3926 int i; 3927 3928 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3929 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; 3930 i = 0; 3931 /* 3932 * If nex2 extents fit in the current page, append 3933 * nex2_ep after the new extents. 3934 */ 3935 if (nex2 <= ext_avail) { 3936 i = erp->er_extcount; 3937 } 3938 /* 3939 * Otherwise, check if space is available in the 3940 * next page. 3941 */ 3942 else if ((erp_idx < nlists - 1) && 3943 (nex2 <= (ext_avail = XFS_LINEAR_EXTS - 3944 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { 3945 erp_idx++; 3946 erp++; 3947 /* Create a hole for nex2 extents */ 3948 memmove(&erp->er_extbuf[nex2], erp->er_extbuf, 3949 erp->er_extcount * sizeof(xfs_bmbt_rec_t)); 3950 } 3951 /* 3952 * Final choice, create a new extent page for 3953 * nex2 extents. 3954 */ 3955 else { 3956 erp_idx++; 3957 erp = xfs_iext_irec_new(ifp, erp_idx); 3958 } 3959 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); 3960 kmem_free(nex2_ep, byte_diff); 3961 erp->er_extcount += nex2; 3962 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); 3963 } 3964 } 3965 3966 /* 3967 * This is called when the amount of space required for incore file 3968 * extents needs to be decreased. The ext_diff parameter stores the 3969 * number of extents to be removed and the idx parameter contains 3970 * the extent index where the extents will be removed from. 3971 * 3972 * If the amount of space needed has decreased below the linear 3973 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous 3974 * extent array. Otherwise, use kmem_realloc() to adjust the 3975 * size to what is needed. 3976 */ 3977 void 3978 xfs_iext_remove( 3979 xfs_ifork_t *ifp, /* inode fork pointer */ 3980 xfs_extnum_t idx, /* index to begin removing exts */ 3981 int ext_diff) /* number of extents to remove */ 3982 { 3983 xfs_extnum_t nextents; /* number of extents in file */ 3984 int new_size; /* size of extents after removal */ 3985 3986 ASSERT(ext_diff > 0); 3987 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3988 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); 3989 3990 if (new_size == 0) { 3991 xfs_iext_destroy(ifp); 3992 } else if (ifp->if_flags & XFS_IFEXTIREC) { 3993 xfs_iext_remove_indirect(ifp, idx, ext_diff); 3994 } else if (ifp->if_real_bytes) { 3995 xfs_iext_remove_direct(ifp, idx, ext_diff); 3996 } else { 3997 xfs_iext_remove_inline(ifp, idx, ext_diff); 3998 } 3999 ifp->if_bytes = new_size; 4000 } 4001 4002 /* 4003 * This removes ext_diff extents from the inline buffer, beginning 4004 * at extent index idx. 4005 */ 4006 void 4007 xfs_iext_remove_inline( 4008 xfs_ifork_t *ifp, /* inode fork pointer */ 4009 xfs_extnum_t idx, /* index to begin removing exts */ 4010 int ext_diff) /* number of extents to remove */ 4011 { 4012 int nextents; /* number of extents in file */ 4013 4014 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 4015 ASSERT(idx < XFS_INLINE_EXTS); 4016 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4017 ASSERT(((nextents - ext_diff) > 0) && 4018 (nextents - ext_diff) < XFS_INLINE_EXTS); 4019 4020 if (idx + ext_diff < nextents) { 4021 memmove(&ifp->if_u2.if_inline_ext[idx], 4022 &ifp->if_u2.if_inline_ext[idx + ext_diff], 4023 (nextents - (idx + ext_diff)) * 4024 sizeof(xfs_bmbt_rec_t)); 4025 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], 4026 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 4027 } else { 4028 memset(&ifp->if_u2.if_inline_ext[idx], 0, 4029 ext_diff * sizeof(xfs_bmbt_rec_t)); 4030 } 4031 } 4032 4033 /* 4034 * This removes ext_diff extents from a linear (direct) extent list, 4035 * beginning at extent index idx. If the extents are being removed 4036 * from the end of the list (ie. truncate) then we just need to re- 4037 * allocate the list to remove the extra space. Otherwise, if the 4038 * extents are being removed from the middle of the existing extent 4039 * entries, then we first need to move the extent records beginning 4040 * at idx + ext_diff up in the list to overwrite the records being 4041 * removed, then remove the extra space via kmem_realloc. 4042 */ 4043 void 4044 xfs_iext_remove_direct( 4045 xfs_ifork_t *ifp, /* inode fork pointer */ 4046 xfs_extnum_t idx, /* index to begin removing exts */ 4047 int ext_diff) /* number of extents to remove */ 4048 { 4049 xfs_extnum_t nextents; /* number of extents in file */ 4050 int new_size; /* size of extents after removal */ 4051 4052 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 4053 new_size = ifp->if_bytes - 4054 (ext_diff * sizeof(xfs_bmbt_rec_t)); 4055 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4056 4057 if (new_size == 0) { 4058 xfs_iext_destroy(ifp); 4059 return; 4060 } 4061 /* Move extents up in the list (if needed) */ 4062 if (idx + ext_diff < nextents) { 4063 memmove(&ifp->if_u1.if_extents[idx], 4064 &ifp->if_u1.if_extents[idx + ext_diff], 4065 (nextents - (idx + ext_diff)) * 4066 sizeof(xfs_bmbt_rec_t)); 4067 } 4068 memset(&ifp->if_u1.if_extents[nextents - ext_diff], 4069 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 4070 /* 4071 * Reallocate the direct extent list. If the extents 4072 * will fit inside the inode then xfs_iext_realloc_direct 4073 * will switch from direct to inline extent allocation 4074 * mode for us. 4075 */ 4076 xfs_iext_realloc_direct(ifp, new_size); 4077 ifp->if_bytes = new_size; 4078 } 4079 4080 /* 4081 * This is called when incore extents are being removed from the 4082 * indirection array and the extents being removed span multiple extent 4083 * buffers. The idx parameter contains the file extent index where we 4084 * want to begin removing extents, and the count parameter contains 4085 * how many extents need to be removed. 4086 * 4087 * |-------| |-------| 4088 * | nex1 | | | nex1 - number of extents before idx 4089 * |-------| | count | 4090 * | | | | count - number of extents being removed at idx 4091 * | count | |-------| 4092 * | | | nex2 | nex2 - number of extents after idx + count 4093 * |-------| |-------| 4094 */ 4095 void 4096 xfs_iext_remove_indirect( 4097 xfs_ifork_t *ifp, /* inode fork pointer */ 4098 xfs_extnum_t idx, /* index to begin removing extents */ 4099 int count) /* number of extents to remove */ 4100 { 4101 xfs_ext_irec_t *erp; /* indirection array pointer */ 4102 int erp_idx = 0; /* indirection array index */ 4103 xfs_extnum_t ext_cnt; /* extents left to remove */ 4104 xfs_extnum_t ext_diff; /* extents to remove in current list */ 4105 xfs_extnum_t nex1; /* number of extents before idx */ 4106 xfs_extnum_t nex2; /* extents after idx + count */ 4107 int nlists; /* entries in indirection array */ 4108 int page_idx = idx; /* index in target extent list */ 4109 4110 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4111 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 4112 ASSERT(erp != NULL); 4113 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4114 nex1 = page_idx; 4115 ext_cnt = count; 4116 while (ext_cnt) { 4117 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); 4118 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); 4119 /* 4120 * Check for deletion of entire list; 4121 * xfs_iext_irec_remove() updates extent offsets. 4122 */ 4123 if (ext_diff == erp->er_extcount) { 4124 xfs_iext_irec_remove(ifp, erp_idx); 4125 ext_cnt -= ext_diff; 4126 nex1 = 0; 4127 if (ext_cnt) { 4128 ASSERT(erp_idx < ifp->if_real_bytes / 4129 XFS_IEXT_BUFSZ); 4130 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4131 nex1 = 0; 4132 continue; 4133 } else { 4134 break; 4135 } 4136 } 4137 /* Move extents up (if needed) */ 4138 if (nex2) { 4139 memmove(&erp->er_extbuf[nex1], 4140 &erp->er_extbuf[nex1 + ext_diff], 4141 nex2 * sizeof(xfs_bmbt_rec_t)); 4142 } 4143 /* Zero out rest of page */ 4144 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - 4145 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); 4146 /* Update remaining counters */ 4147 erp->er_extcount -= ext_diff; 4148 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); 4149 ext_cnt -= ext_diff; 4150 nex1 = 0; 4151 erp_idx++; 4152 erp++; 4153 } 4154 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); 4155 xfs_iext_irec_compact(ifp); 4156 } 4157 4158 /* 4159 * Create, destroy, or resize a linear (direct) block of extents. 4160 */ 4161 void 4162 xfs_iext_realloc_direct( 4163 xfs_ifork_t *ifp, /* inode fork pointer */ 4164 int new_size) /* new size of extents */ 4165 { 4166 int rnew_size; /* real new size of extents */ 4167 4168 rnew_size = new_size; 4169 4170 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || 4171 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && 4172 (new_size != ifp->if_real_bytes))); 4173 4174 /* Free extent records */ 4175 if (new_size == 0) { 4176 xfs_iext_destroy(ifp); 4177 } 4178 /* Resize direct extent list and zero any new bytes */ 4179 else if (ifp->if_real_bytes) { 4180 /* Check if extents will fit inside the inode */ 4181 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { 4182 xfs_iext_direct_to_inline(ifp, new_size / 4183 (uint)sizeof(xfs_bmbt_rec_t)); 4184 ifp->if_bytes = new_size; 4185 return; 4186 } 4187 if ((new_size & (new_size - 1)) != 0) { 4188 rnew_size = xfs_iroundup(new_size); 4189 } 4190 if (rnew_size != ifp->if_real_bytes) { 4191 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) 4192 kmem_realloc(ifp->if_u1.if_extents, 4193 rnew_size, 4194 ifp->if_real_bytes, 4195 KM_SLEEP); 4196 } 4197 if (rnew_size > ifp->if_real_bytes) { 4198 memset(&ifp->if_u1.if_extents[ifp->if_bytes / 4199 (uint)sizeof(xfs_bmbt_rec_t)], 0, 4200 rnew_size - ifp->if_real_bytes); 4201 } 4202 } 4203 /* 4204 * Switch from the inline extent buffer to a direct 4205 * extent list. Be sure to include the inline extent 4206 * bytes in new_size. 4207 */ 4208 else { 4209 new_size += ifp->if_bytes; 4210 if ((new_size & (new_size - 1)) != 0) { 4211 rnew_size = xfs_iroundup(new_size); 4212 } 4213 xfs_iext_inline_to_direct(ifp, rnew_size); 4214 } 4215 ifp->if_real_bytes = rnew_size; 4216 ifp->if_bytes = new_size; 4217 } 4218 4219 /* 4220 * Switch from linear (direct) extent records to inline buffer. 4221 */ 4222 void 4223 xfs_iext_direct_to_inline( 4224 xfs_ifork_t *ifp, /* inode fork pointer */ 4225 xfs_extnum_t nextents) /* number of extents in file */ 4226 { 4227 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 4228 ASSERT(nextents <= XFS_INLINE_EXTS); 4229 /* 4230 * The inline buffer was zeroed when we switched 4231 * from inline to direct extent allocation mode, 4232 * so we don't need to clear it here. 4233 */ 4234 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, 4235 nextents * sizeof(xfs_bmbt_rec_t)); 4236 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); 4237 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 4238 ifp->if_real_bytes = 0; 4239 } 4240 4241 /* 4242 * Switch from inline buffer to linear (direct) extent records. 4243 * new_size should already be rounded up to the next power of 2 4244 * by the caller (when appropriate), so use new_size as it is. 4245 * However, since new_size may be rounded up, we can't update 4246 * if_bytes here. It is the caller's responsibility to update 4247 * if_bytes upon return. 4248 */ 4249 void 4250 xfs_iext_inline_to_direct( 4251 xfs_ifork_t *ifp, /* inode fork pointer */ 4252 int new_size) /* number of extents in file */ 4253 { 4254 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) 4255 kmem_alloc(new_size, KM_SLEEP); 4256 memset(ifp->if_u1.if_extents, 0, new_size); 4257 if (ifp->if_bytes) { 4258 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, 4259 ifp->if_bytes); 4260 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 4261 sizeof(xfs_bmbt_rec_t)); 4262 } 4263 ifp->if_real_bytes = new_size; 4264 } 4265 4266 /* 4267 * Resize an extent indirection array to new_size bytes. 4268 */ 4269 void 4270 xfs_iext_realloc_indirect( 4271 xfs_ifork_t *ifp, /* inode fork pointer */ 4272 int new_size) /* new indirection array size */ 4273 { 4274 int nlists; /* number of irec's (ex lists) */ 4275 int size; /* current indirection array size */ 4276 4277 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4278 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4279 size = nlists * sizeof(xfs_ext_irec_t); 4280 ASSERT(ifp->if_real_bytes); 4281 ASSERT((new_size >= 0) && (new_size != size)); 4282 if (new_size == 0) { 4283 xfs_iext_destroy(ifp); 4284 } else { 4285 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) 4286 kmem_realloc(ifp->if_u1.if_ext_irec, 4287 new_size, size, KM_SLEEP); 4288 } 4289 } 4290 4291 /* 4292 * Switch from indirection array to linear (direct) extent allocations. 4293 */ 4294 void 4295 xfs_iext_indirect_to_direct( 4296 xfs_ifork_t *ifp) /* inode fork pointer */ 4297 { 4298 xfs_bmbt_rec_t *ep; /* extent record pointer */ 4299 xfs_extnum_t nextents; /* number of extents in file */ 4300 int size; /* size of file extents */ 4301 4302 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4303 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4304 ASSERT(nextents <= XFS_LINEAR_EXTS); 4305 size = nextents * sizeof(xfs_bmbt_rec_t); 4306 4307 xfs_iext_irec_compact_full(ifp); 4308 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); 4309 4310 ep = ifp->if_u1.if_ext_irec->er_extbuf; 4311 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t)); 4312 ifp->if_flags &= ~XFS_IFEXTIREC; 4313 ifp->if_u1.if_extents = ep; 4314 ifp->if_bytes = size; 4315 if (nextents < XFS_LINEAR_EXTS) { 4316 xfs_iext_realloc_direct(ifp, size); 4317 } 4318 } 4319 4320 /* 4321 * Free incore file extents. 4322 */ 4323 void 4324 xfs_iext_destroy( 4325 xfs_ifork_t *ifp) /* inode fork pointer */ 4326 { 4327 if (ifp->if_flags & XFS_IFEXTIREC) { 4328 int erp_idx; 4329 int nlists; 4330 4331 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4332 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { 4333 xfs_iext_irec_remove(ifp, erp_idx); 4334 } 4335 ifp->if_flags &= ~XFS_IFEXTIREC; 4336 } else if (ifp->if_real_bytes) { 4337 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); 4338 } else if (ifp->if_bytes) { 4339 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 4340 sizeof(xfs_bmbt_rec_t)); 4341 } 4342 ifp->if_u1.if_extents = NULL; 4343 ifp->if_real_bytes = 0; 4344 ifp->if_bytes = 0; 4345 } 4346 4347 /* 4348 * Return a pointer to the extent record for file system block bno. 4349 */ 4350 xfs_bmbt_rec_t * /* pointer to found extent record */ 4351 xfs_iext_bno_to_ext( 4352 xfs_ifork_t *ifp, /* inode fork pointer */ 4353 xfs_fileoff_t bno, /* block number to search for */ 4354 xfs_extnum_t *idxp) /* index of target extent */ 4355 { 4356 xfs_bmbt_rec_t *base; /* pointer to first extent */ 4357 xfs_filblks_t blockcount = 0; /* number of blocks in extent */ 4358 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */ 4359 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 4360 int high; /* upper boundary in search */ 4361 xfs_extnum_t idx = 0; /* index of target extent */ 4362 int low; /* lower boundary in search */ 4363 xfs_extnum_t nextents; /* number of file extents */ 4364 xfs_fileoff_t startoff = 0; /* start offset of extent */ 4365 4366 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4367 if (nextents == 0) { 4368 *idxp = 0; 4369 return NULL; 4370 } 4371 low = 0; 4372 if (ifp->if_flags & XFS_IFEXTIREC) { 4373 /* Find target extent list */ 4374 int erp_idx = 0; 4375 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); 4376 base = erp->er_extbuf; 4377 high = erp->er_extcount - 1; 4378 } else { 4379 base = ifp->if_u1.if_extents; 4380 high = nextents - 1; 4381 } 4382 /* Binary search extent records */ 4383 while (low <= high) { 4384 idx = (low + high) >> 1; 4385 ep = base + idx; 4386 startoff = xfs_bmbt_get_startoff(ep); 4387 blockcount = xfs_bmbt_get_blockcount(ep); 4388 if (bno < startoff) { 4389 high = idx - 1; 4390 } else if (bno >= startoff + blockcount) { 4391 low = idx + 1; 4392 } else { 4393 /* Convert back to file-based extent index */ 4394 if (ifp->if_flags & XFS_IFEXTIREC) { 4395 idx += erp->er_extoff; 4396 } 4397 *idxp = idx; 4398 return ep; 4399 } 4400 } 4401 /* Convert back to file-based extent index */ 4402 if (ifp->if_flags & XFS_IFEXTIREC) { 4403 idx += erp->er_extoff; 4404 } 4405 if (bno >= startoff + blockcount) { 4406 if (++idx == nextents) { 4407 ep = NULL; 4408 } else { 4409 ep = xfs_iext_get_ext(ifp, idx); 4410 } 4411 } 4412 *idxp = idx; 4413 return ep; 4414 } 4415 4416 /* 4417 * Return a pointer to the indirection array entry containing the 4418 * extent record for filesystem block bno. Store the index of the 4419 * target irec in *erp_idxp. 4420 */ 4421 xfs_ext_irec_t * /* pointer to found extent record */ 4422 xfs_iext_bno_to_irec( 4423 xfs_ifork_t *ifp, /* inode fork pointer */ 4424 xfs_fileoff_t bno, /* block number to search for */ 4425 int *erp_idxp) /* irec index of target ext list */ 4426 { 4427 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 4428 xfs_ext_irec_t *erp_next; /* next indirection array entry */ 4429 int erp_idx; /* indirection array index */ 4430 int nlists; /* number of extent irec's (lists) */ 4431 int high; /* binary search upper limit */ 4432 int low; /* binary search lower limit */ 4433 4434 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4435 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4436 erp_idx = 0; 4437 low = 0; 4438 high = nlists - 1; 4439 while (low <= high) { 4440 erp_idx = (low + high) >> 1; 4441 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4442 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; 4443 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { 4444 high = erp_idx - 1; 4445 } else if (erp_next && bno >= 4446 xfs_bmbt_get_startoff(erp_next->er_extbuf)) { 4447 low = erp_idx + 1; 4448 } else { 4449 break; 4450 } 4451 } 4452 *erp_idxp = erp_idx; 4453 return erp; 4454 } 4455 4456 /* 4457 * Return a pointer to the indirection array entry containing the 4458 * extent record at file extent index *idxp. Store the index of the 4459 * target irec in *erp_idxp and store the page index of the target 4460 * extent record in *idxp. 4461 */ 4462 xfs_ext_irec_t * 4463 xfs_iext_idx_to_irec( 4464 xfs_ifork_t *ifp, /* inode fork pointer */ 4465 xfs_extnum_t *idxp, /* extent index (file -> page) */ 4466 int *erp_idxp, /* pointer to target irec */ 4467 int realloc) /* new bytes were just added */ 4468 { 4469 xfs_ext_irec_t *prev; /* pointer to previous irec */ 4470 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ 4471 int erp_idx; /* indirection array index */ 4472 int nlists; /* number of irec's (ex lists) */ 4473 int high; /* binary search upper limit */ 4474 int low; /* binary search lower limit */ 4475 xfs_extnum_t page_idx = *idxp; /* extent index in target list */ 4476 4477 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4478 ASSERT(page_idx >= 0 && page_idx <= 4479 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)); 4480 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4481 erp_idx = 0; 4482 low = 0; 4483 high = nlists - 1; 4484 4485 /* Binary search extent irec's */ 4486 while (low <= high) { 4487 erp_idx = (low + high) >> 1; 4488 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4489 prev = erp_idx > 0 ? erp - 1 : NULL; 4490 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && 4491 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { 4492 high = erp_idx - 1; 4493 } else if (page_idx > erp->er_extoff + erp->er_extcount || 4494 (page_idx == erp->er_extoff + erp->er_extcount && 4495 !realloc)) { 4496 low = erp_idx + 1; 4497 } else if (page_idx == erp->er_extoff + erp->er_extcount && 4498 erp->er_extcount == XFS_LINEAR_EXTS) { 4499 ASSERT(realloc); 4500 page_idx = 0; 4501 erp_idx++; 4502 erp = erp_idx < nlists ? erp + 1 : NULL; 4503 break; 4504 } else { 4505 page_idx -= erp->er_extoff; 4506 break; 4507 } 4508 } 4509 *idxp = page_idx; 4510 *erp_idxp = erp_idx; 4511 return(erp); 4512 } 4513 4514 /* 4515 * Allocate and initialize an indirection array once the space needed 4516 * for incore extents increases above XFS_IEXT_BUFSZ. 4517 */ 4518 void 4519 xfs_iext_irec_init( 4520 xfs_ifork_t *ifp) /* inode fork pointer */ 4521 { 4522 xfs_ext_irec_t *erp; /* indirection array pointer */ 4523 xfs_extnum_t nextents; /* number of extents in file */ 4524 4525 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 4526 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4527 ASSERT(nextents <= XFS_LINEAR_EXTS); 4528 4529 erp = (xfs_ext_irec_t *) 4530 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP); 4531 4532 if (nextents == 0) { 4533 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) 4534 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); 4535 } else if (!ifp->if_real_bytes) { 4536 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); 4537 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { 4538 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); 4539 } 4540 erp->er_extbuf = ifp->if_u1.if_extents; 4541 erp->er_extcount = nextents; 4542 erp->er_extoff = 0; 4543 4544 ifp->if_flags |= XFS_IFEXTIREC; 4545 ifp->if_real_bytes = XFS_IEXT_BUFSZ; 4546 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); 4547 ifp->if_u1.if_ext_irec = erp; 4548 4549 return; 4550 } 4551 4552 /* 4553 * Allocate and initialize a new entry in the indirection array. 4554 */ 4555 xfs_ext_irec_t * 4556 xfs_iext_irec_new( 4557 xfs_ifork_t *ifp, /* inode fork pointer */ 4558 int erp_idx) /* index for new irec */ 4559 { 4560 xfs_ext_irec_t *erp; /* indirection array pointer */ 4561 int i; /* loop counter */ 4562 int nlists; /* number of irec's (ex lists) */ 4563 4564 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4565 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4566 4567 /* Resize indirection array */ 4568 xfs_iext_realloc_indirect(ifp, ++nlists * 4569 sizeof(xfs_ext_irec_t)); 4570 /* 4571 * Move records down in the array so the 4572 * new page can use erp_idx. 4573 */ 4574 erp = ifp->if_u1.if_ext_irec; 4575 for (i = nlists - 1; i > erp_idx; i--) { 4576 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); 4577 } 4578 ASSERT(i == erp_idx); 4579 4580 /* Initialize new extent record */ 4581 erp = ifp->if_u1.if_ext_irec; 4582 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *) 4583 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); 4584 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 4585 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); 4586 erp[erp_idx].er_extcount = 0; 4587 erp[erp_idx].er_extoff = erp_idx > 0 ? 4588 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; 4589 return (&erp[erp_idx]); 4590 } 4591 4592 /* 4593 * Remove a record from the indirection array. 4594 */ 4595 void 4596 xfs_iext_irec_remove( 4597 xfs_ifork_t *ifp, /* inode fork pointer */ 4598 int erp_idx) /* irec index to remove */ 4599 { 4600 xfs_ext_irec_t *erp; /* indirection array pointer */ 4601 int i; /* loop counter */ 4602 int nlists; /* number of irec's (ex lists) */ 4603 4604 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4605 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4606 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4607 if (erp->er_extbuf) { 4608 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, 4609 -erp->er_extcount); 4610 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ); 4611 } 4612 /* Compact extent records */ 4613 erp = ifp->if_u1.if_ext_irec; 4614 for (i = erp_idx; i < nlists - 1; i++) { 4615 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); 4616 } 4617 /* 4618 * Manually free the last extent record from the indirection 4619 * array. A call to xfs_iext_realloc_indirect() with a size 4620 * of zero would result in a call to xfs_iext_destroy() which 4621 * would in turn call this function again, creating a nasty 4622 * infinite loop. 4623 */ 4624 if (--nlists) { 4625 xfs_iext_realloc_indirect(ifp, 4626 nlists * sizeof(xfs_ext_irec_t)); 4627 } else { 4628 kmem_free(ifp->if_u1.if_ext_irec, 4629 sizeof(xfs_ext_irec_t)); 4630 } 4631 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 4632 } 4633 4634 /* 4635 * This is called to clean up large amounts of unused memory allocated 4636 * by the indirection array. Before compacting anything though, verify 4637 * that the indirection array is still needed and switch back to the 4638 * linear extent list (or even the inline buffer) if possible. The 4639 * compaction policy is as follows: 4640 * 4641 * Full Compaction: Extents fit into a single page (or inline buffer) 4642 * Full Compaction: Extents occupy less than 10% of allocated space 4643 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space 4644 * No Compaction: Extents occupy at least 50% of allocated space 4645 */ 4646 void 4647 xfs_iext_irec_compact( 4648 xfs_ifork_t *ifp) /* inode fork pointer */ 4649 { 4650 xfs_extnum_t nextents; /* number of extents in file */ 4651 int nlists; /* number of irec's (ex lists) */ 4652 4653 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4654 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4655 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 4656 4657 if (nextents == 0) { 4658 xfs_iext_destroy(ifp); 4659 } else if (nextents <= XFS_INLINE_EXTS) { 4660 xfs_iext_indirect_to_direct(ifp); 4661 xfs_iext_direct_to_inline(ifp, nextents); 4662 } else if (nextents <= XFS_LINEAR_EXTS) { 4663 xfs_iext_indirect_to_direct(ifp); 4664 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) { 4665 xfs_iext_irec_compact_full(ifp); 4666 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { 4667 xfs_iext_irec_compact_pages(ifp); 4668 } 4669 } 4670 4671 /* 4672 * Combine extents from neighboring extent pages. 4673 */ 4674 void 4675 xfs_iext_irec_compact_pages( 4676 xfs_ifork_t *ifp) /* inode fork pointer */ 4677 { 4678 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ 4679 int erp_idx = 0; /* indirection array index */ 4680 int nlists; /* number of irec's (ex lists) */ 4681 4682 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4683 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4684 while (erp_idx < nlists - 1) { 4685 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4686 erp_next = erp + 1; 4687 if (erp_next->er_extcount <= 4688 (XFS_LINEAR_EXTS - erp->er_extcount)) { 4689 memmove(&erp->er_extbuf[erp->er_extcount], 4690 erp_next->er_extbuf, erp_next->er_extcount * 4691 sizeof(xfs_bmbt_rec_t)); 4692 erp->er_extcount += erp_next->er_extcount; 4693 /* 4694 * Free page before removing extent record 4695 * so er_extoffs don't get modified in 4696 * xfs_iext_irec_remove. 4697 */ 4698 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ); 4699 erp_next->er_extbuf = NULL; 4700 xfs_iext_irec_remove(ifp, erp_idx + 1); 4701 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4702 } else { 4703 erp_idx++; 4704 } 4705 } 4706 } 4707 4708 /* 4709 * Fully compact the extent records managed by the indirection array. 4710 */ 4711 void 4712 xfs_iext_irec_compact_full( 4713 xfs_ifork_t *ifp) /* inode fork pointer */ 4714 { 4715 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */ 4716 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */ 4717 int erp_idx = 0; /* extent irec index */ 4718 int ext_avail; /* empty entries in ex list */ 4719 int ext_diff; /* number of exts to add */ 4720 int nlists; /* number of irec's (ex lists) */ 4721 4722 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4723 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4724 erp = ifp->if_u1.if_ext_irec; 4725 ep = &erp->er_extbuf[erp->er_extcount]; 4726 erp_next = erp + 1; 4727 ep_next = erp_next->er_extbuf; 4728 while (erp_idx < nlists - 1) { 4729 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; 4730 ext_diff = MIN(ext_avail, erp_next->er_extcount); 4731 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t)); 4732 erp->er_extcount += ext_diff; 4733 erp_next->er_extcount -= ext_diff; 4734 /* Remove next page */ 4735 if (erp_next->er_extcount == 0) { 4736 /* 4737 * Free page before removing extent record 4738 * so er_extoffs don't get modified in 4739 * xfs_iext_irec_remove. 4740 */ 4741 kmem_free(erp_next->er_extbuf, 4742 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); 4743 erp_next->er_extbuf = NULL; 4744 xfs_iext_irec_remove(ifp, erp_idx + 1); 4745 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4746 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4747 /* Update next page */ 4748 } else { 4749 /* Move rest of page up to become next new page */ 4750 memmove(erp_next->er_extbuf, ep_next, 4751 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); 4752 ep_next = erp_next->er_extbuf; 4753 memset(&ep_next[erp_next->er_extcount], 0, 4754 (XFS_LINEAR_EXTS - erp_next->er_extcount) * 4755 sizeof(xfs_bmbt_rec_t)); 4756 } 4757 if (erp->er_extcount == XFS_LINEAR_EXTS) { 4758 erp_idx++; 4759 if (erp_idx < nlists) 4760 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 4761 else 4762 break; 4763 } 4764 ep = &erp->er_extbuf[erp->er_extcount]; 4765 erp_next = erp + 1; 4766 ep_next = erp_next->er_extbuf; 4767 } 4768 } 4769 4770 /* 4771 * This is called to update the er_extoff field in the indirection 4772 * array when extents have been added or removed from one of the 4773 * extent lists. erp_idx contains the irec index to begin updating 4774 * at and ext_diff contains the number of extents that were added 4775 * or removed. 4776 */ 4777 void 4778 xfs_iext_irec_update_extoffs( 4779 xfs_ifork_t *ifp, /* inode fork pointer */ 4780 int erp_idx, /* irec index to update */ 4781 int ext_diff) /* number of new extents */ 4782 { 4783 int i; /* loop counter */ 4784 int nlists; /* number of irec's (ex lists */ 4785 4786 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4787 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4788 for (i = erp_idx; i < nlists; i++) { 4789 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; 4790 } 4791 } 4792