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 <linux/log2.h> 19 20 #include "xfs.h" 21 #include "xfs_fs.h" 22 #include "xfs_types.h" 23 #include "xfs_log.h" 24 #include "xfs_inum.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_mount.h" 30 #include "xfs_bmap_btree.h" 31 #include "xfs_alloc_btree.h" 32 #include "xfs_ialloc_btree.h" 33 #include "xfs_attr_sf.h" 34 #include "xfs_dinode.h" 35 #include "xfs_inode.h" 36 #include "xfs_buf_item.h" 37 #include "xfs_inode_item.h" 38 #include "xfs_btree.h" 39 #include "xfs_alloc.h" 40 #include "xfs_ialloc.h" 41 #include "xfs_bmap.h" 42 #include "xfs_error.h" 43 #include "xfs_utils.h" 44 #include "xfs_quota.h" 45 #include "xfs_filestream.h" 46 #include "xfs_vnodeops.h" 47 #include "xfs_cksum.h" 48 #include "xfs_trace.h" 49 #include "xfs_icache.h" 50 51 kmem_zone_t *xfs_ifork_zone; 52 kmem_zone_t *xfs_inode_zone; 53 54 /* 55 * Used in xfs_itruncate_extents(). This is the maximum number of extents 56 * freed from a file in a single transaction. 57 */ 58 #define XFS_ITRUNC_MAX_EXTENTS 2 59 60 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); 61 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); 62 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); 63 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); 64 65 /* 66 * helper function to extract extent size hint from inode 67 */ 68 xfs_extlen_t 69 xfs_get_extsz_hint( 70 struct xfs_inode *ip) 71 { 72 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 73 return ip->i_d.di_extsize; 74 if (XFS_IS_REALTIME_INODE(ip)) 75 return ip->i_mount->m_sb.sb_rextsize; 76 return 0; 77 } 78 79 /* 80 * This is a wrapper routine around the xfs_ilock() routine used to centralize 81 * some grungy code. It is used in places that wish to lock the inode solely 82 * for reading the extents. The reason these places can't just call 83 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the 84 * extents from disk for a file in b-tree format. If the inode is in b-tree 85 * format, then we need to lock the inode exclusively until the extents are read 86 * in. Locking it exclusively all the time would limit our parallelism 87 * unnecessarily, though. What we do instead is check to see if the extents 88 * have been read in yet, and only lock the inode exclusively if they have not. 89 * 90 * The function returns a value which should be given to the corresponding 91 * xfs_iunlock_map_shared(). This value is the mode in which the lock was 92 * actually taken. 93 */ 94 uint 95 xfs_ilock_map_shared( 96 xfs_inode_t *ip) 97 { 98 uint lock_mode; 99 100 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && 101 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { 102 lock_mode = XFS_ILOCK_EXCL; 103 } else { 104 lock_mode = XFS_ILOCK_SHARED; 105 } 106 107 xfs_ilock(ip, lock_mode); 108 109 return lock_mode; 110 } 111 112 /* 113 * This is simply the unlock routine to go with xfs_ilock_map_shared(). 114 * All it does is call xfs_iunlock() with the given lock_mode. 115 */ 116 void 117 xfs_iunlock_map_shared( 118 xfs_inode_t *ip, 119 unsigned int lock_mode) 120 { 121 xfs_iunlock(ip, lock_mode); 122 } 123 124 /* 125 * The xfs inode contains 2 locks: a multi-reader lock called the 126 * i_iolock and a multi-reader lock called the i_lock. This routine 127 * allows either or both of the locks to be obtained. 128 * 129 * The 2 locks should always be ordered so that the IO lock is 130 * obtained first in order to prevent deadlock. 131 * 132 * ip -- the inode being locked 133 * lock_flags -- this parameter indicates the inode's locks 134 * to be locked. It can be: 135 * XFS_IOLOCK_SHARED, 136 * XFS_IOLOCK_EXCL, 137 * XFS_ILOCK_SHARED, 138 * XFS_ILOCK_EXCL, 139 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, 140 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, 141 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, 142 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL 143 */ 144 void 145 xfs_ilock( 146 xfs_inode_t *ip, 147 uint lock_flags) 148 { 149 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 150 151 /* 152 * You can't set both SHARED and EXCL for the same lock, 153 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 154 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 155 */ 156 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 157 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 158 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 159 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 160 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 161 162 if (lock_flags & XFS_IOLOCK_EXCL) 163 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 164 else if (lock_flags & XFS_IOLOCK_SHARED) 165 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 166 167 if (lock_flags & XFS_ILOCK_EXCL) 168 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 169 else if (lock_flags & XFS_ILOCK_SHARED) 170 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 171 } 172 173 /* 174 * This is just like xfs_ilock(), except that the caller 175 * is guaranteed not to sleep. It returns 1 if it gets 176 * the requested locks and 0 otherwise. If the IO lock is 177 * obtained but the inode lock cannot be, then the IO lock 178 * is dropped before returning. 179 * 180 * ip -- the inode being locked 181 * lock_flags -- this parameter indicates the inode's locks to be 182 * to be locked. See the comment for xfs_ilock() for a list 183 * of valid values. 184 */ 185 int 186 xfs_ilock_nowait( 187 xfs_inode_t *ip, 188 uint lock_flags) 189 { 190 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 191 192 /* 193 * You can't set both SHARED and EXCL for the same lock, 194 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 195 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 196 */ 197 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 198 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 199 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 200 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 201 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 202 203 if (lock_flags & XFS_IOLOCK_EXCL) { 204 if (!mrtryupdate(&ip->i_iolock)) 205 goto out; 206 } else if (lock_flags & XFS_IOLOCK_SHARED) { 207 if (!mrtryaccess(&ip->i_iolock)) 208 goto out; 209 } 210 if (lock_flags & XFS_ILOCK_EXCL) { 211 if (!mrtryupdate(&ip->i_lock)) 212 goto out_undo_iolock; 213 } else if (lock_flags & XFS_ILOCK_SHARED) { 214 if (!mrtryaccess(&ip->i_lock)) 215 goto out_undo_iolock; 216 } 217 return 1; 218 219 out_undo_iolock: 220 if (lock_flags & XFS_IOLOCK_EXCL) 221 mrunlock_excl(&ip->i_iolock); 222 else if (lock_flags & XFS_IOLOCK_SHARED) 223 mrunlock_shared(&ip->i_iolock); 224 out: 225 return 0; 226 } 227 228 /* 229 * xfs_iunlock() is used to drop the inode locks acquired with 230 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 231 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 232 * that we know which locks to drop. 233 * 234 * ip -- the inode being unlocked 235 * lock_flags -- this parameter indicates the inode's locks to be 236 * to be unlocked. See the comment for xfs_ilock() for a list 237 * of valid values for this parameter. 238 * 239 */ 240 void 241 xfs_iunlock( 242 xfs_inode_t *ip, 243 uint lock_flags) 244 { 245 /* 246 * You can't set both SHARED and EXCL for the same lock, 247 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 248 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 249 */ 250 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 251 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 252 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 253 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 254 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 255 ASSERT(lock_flags != 0); 256 257 if (lock_flags & XFS_IOLOCK_EXCL) 258 mrunlock_excl(&ip->i_iolock); 259 else if (lock_flags & XFS_IOLOCK_SHARED) 260 mrunlock_shared(&ip->i_iolock); 261 262 if (lock_flags & XFS_ILOCK_EXCL) 263 mrunlock_excl(&ip->i_lock); 264 else if (lock_flags & XFS_ILOCK_SHARED) 265 mrunlock_shared(&ip->i_lock); 266 267 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 268 } 269 270 /* 271 * give up write locks. the i/o lock cannot be held nested 272 * if it is being demoted. 273 */ 274 void 275 xfs_ilock_demote( 276 xfs_inode_t *ip, 277 uint lock_flags) 278 { 279 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); 280 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 281 282 if (lock_flags & XFS_ILOCK_EXCL) 283 mrdemote(&ip->i_lock); 284 if (lock_flags & XFS_IOLOCK_EXCL) 285 mrdemote(&ip->i_iolock); 286 287 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 288 } 289 290 #if defined(DEBUG) || defined(XFS_WARN) 291 int 292 xfs_isilocked( 293 xfs_inode_t *ip, 294 uint lock_flags) 295 { 296 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 297 if (!(lock_flags & XFS_ILOCK_SHARED)) 298 return !!ip->i_lock.mr_writer; 299 return rwsem_is_locked(&ip->i_lock.mr_lock); 300 } 301 302 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 303 if (!(lock_flags & XFS_IOLOCK_SHARED)) 304 return !!ip->i_iolock.mr_writer; 305 return rwsem_is_locked(&ip->i_iolock.mr_lock); 306 } 307 308 ASSERT(0); 309 return 0; 310 } 311 #endif 312 313 void 314 __xfs_iflock( 315 struct xfs_inode *ip) 316 { 317 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 318 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 319 320 do { 321 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 322 if (xfs_isiflocked(ip)) 323 io_schedule(); 324 } while (!xfs_iflock_nowait(ip)); 325 326 finish_wait(wq, &wait.wait); 327 } 328 329 #ifdef DEBUG 330 /* 331 * Make sure that the extents in the given memory buffer 332 * are valid. 333 */ 334 STATIC void 335 xfs_validate_extents( 336 xfs_ifork_t *ifp, 337 int nrecs, 338 xfs_exntfmt_t fmt) 339 { 340 xfs_bmbt_irec_t irec; 341 xfs_bmbt_rec_host_t rec; 342 int i; 343 344 for (i = 0; i < nrecs; i++) { 345 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 346 rec.l0 = get_unaligned(&ep->l0); 347 rec.l1 = get_unaligned(&ep->l1); 348 xfs_bmbt_get_all(&rec, &irec); 349 if (fmt == XFS_EXTFMT_NOSTATE) 350 ASSERT(irec.br_state == XFS_EXT_NORM); 351 } 352 } 353 #else /* DEBUG */ 354 #define xfs_validate_extents(ifp, nrecs, fmt) 355 #endif /* DEBUG */ 356 357 /* 358 * Check that none of the inode's in the buffer have a next 359 * unlinked field of 0. 360 */ 361 #if defined(DEBUG) 362 void 363 xfs_inobp_check( 364 xfs_mount_t *mp, 365 xfs_buf_t *bp) 366 { 367 int i; 368 int j; 369 xfs_dinode_t *dip; 370 371 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 372 373 for (i = 0; i < j; i++) { 374 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 375 i * mp->m_sb.sb_inodesize); 376 if (!dip->di_next_unlinked) { 377 xfs_alert(mp, 378 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.", 379 bp); 380 ASSERT(dip->di_next_unlinked); 381 } 382 } 383 } 384 #endif 385 386 static void 387 xfs_inode_buf_verify( 388 struct xfs_buf *bp) 389 { 390 struct xfs_mount *mp = bp->b_target->bt_mount; 391 int i; 392 int ni; 393 394 /* 395 * Validate the magic number and version of every inode in the buffer 396 */ 397 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock; 398 for (i = 0; i < ni; i++) { 399 int di_ok; 400 xfs_dinode_t *dip; 401 402 dip = (struct xfs_dinode *)xfs_buf_offset(bp, 403 (i << mp->m_sb.sb_inodelog)); 404 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) && 405 XFS_DINODE_GOOD_VERSION(dip->di_version); 406 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, 407 XFS_ERRTAG_ITOBP_INOTOBP, 408 XFS_RANDOM_ITOBP_INOTOBP))) { 409 xfs_buf_ioerror(bp, EFSCORRUPTED); 410 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH, 411 mp, dip); 412 #ifdef DEBUG 413 xfs_emerg(mp, 414 "bad inode magic/vsn daddr %lld #%d (magic=%x)", 415 (unsigned long long)bp->b_bn, i, 416 be16_to_cpu(dip->di_magic)); 417 ASSERT(0); 418 #endif 419 } 420 } 421 xfs_inobp_check(mp, bp); 422 } 423 424 425 static void 426 xfs_inode_buf_read_verify( 427 struct xfs_buf *bp) 428 { 429 xfs_inode_buf_verify(bp); 430 } 431 432 static void 433 xfs_inode_buf_write_verify( 434 struct xfs_buf *bp) 435 { 436 xfs_inode_buf_verify(bp); 437 } 438 439 const struct xfs_buf_ops xfs_inode_buf_ops = { 440 .verify_read = xfs_inode_buf_read_verify, 441 .verify_write = xfs_inode_buf_write_verify, 442 }; 443 444 445 /* 446 * This routine is called to map an inode to the buffer containing the on-disk 447 * version of the inode. It returns a pointer to the buffer containing the 448 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a 449 * pointer to the on-disk inode within that buffer. 450 * 451 * If a non-zero error is returned, then the contents of bpp and dipp are 452 * undefined. 453 */ 454 int 455 xfs_imap_to_bp( 456 struct xfs_mount *mp, 457 struct xfs_trans *tp, 458 struct xfs_imap *imap, 459 struct xfs_dinode **dipp, 460 struct xfs_buf **bpp, 461 uint buf_flags, 462 uint iget_flags) 463 { 464 struct xfs_buf *bp; 465 int error; 466 467 buf_flags |= XBF_UNMAPPED; 468 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, 469 (int)imap->im_len, buf_flags, &bp, 470 &xfs_inode_buf_ops); 471 if (error) { 472 if (error == EAGAIN) { 473 ASSERT(buf_flags & XBF_TRYLOCK); 474 return error; 475 } 476 477 if (error == EFSCORRUPTED && 478 (iget_flags & XFS_IGET_UNTRUSTED)) 479 return XFS_ERROR(EINVAL); 480 481 xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.", 482 __func__, error); 483 return error; 484 } 485 486 *bpp = bp; 487 *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset); 488 return 0; 489 } 490 491 /* 492 * Move inode type and inode format specific information from the 493 * on-disk inode to the in-core inode. For fifos, devs, and sockets 494 * this means set if_rdev to the proper value. For files, directories, 495 * and symlinks this means to bring in the in-line data or extent 496 * pointers. For a file in B-tree format, only the root is immediately 497 * brought in-core. The rest will be in-lined in if_extents when it 498 * is first referenced (see xfs_iread_extents()). 499 */ 500 STATIC int 501 xfs_iformat( 502 xfs_inode_t *ip, 503 xfs_dinode_t *dip) 504 { 505 xfs_attr_shortform_t *atp; 506 int size; 507 int error = 0; 508 xfs_fsize_t di_size; 509 510 if (unlikely(be32_to_cpu(dip->di_nextents) + 511 be16_to_cpu(dip->di_anextents) > 512 be64_to_cpu(dip->di_nblocks))) { 513 xfs_warn(ip->i_mount, 514 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", 515 (unsigned long long)ip->i_ino, 516 (int)(be32_to_cpu(dip->di_nextents) + 517 be16_to_cpu(dip->di_anextents)), 518 (unsigned long long) 519 be64_to_cpu(dip->di_nblocks)); 520 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, 521 ip->i_mount, dip); 522 return XFS_ERROR(EFSCORRUPTED); 523 } 524 525 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { 526 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.", 527 (unsigned long long)ip->i_ino, 528 dip->di_forkoff); 529 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, 530 ip->i_mount, dip); 531 return XFS_ERROR(EFSCORRUPTED); 532 } 533 534 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && 535 !ip->i_mount->m_rtdev_targp)) { 536 xfs_warn(ip->i_mount, 537 "corrupt dinode %Lu, has realtime flag set.", 538 ip->i_ino); 539 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)", 540 XFS_ERRLEVEL_LOW, ip->i_mount, dip); 541 return XFS_ERROR(EFSCORRUPTED); 542 } 543 544 switch (ip->i_d.di_mode & S_IFMT) { 545 case S_IFIFO: 546 case S_IFCHR: 547 case S_IFBLK: 548 case S_IFSOCK: 549 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) { 550 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, 551 ip->i_mount, dip); 552 return XFS_ERROR(EFSCORRUPTED); 553 } 554 ip->i_d.di_size = 0; 555 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip); 556 break; 557 558 case S_IFREG: 559 case S_IFLNK: 560 case S_IFDIR: 561 switch (dip->di_format) { 562 case XFS_DINODE_FMT_LOCAL: 563 /* 564 * no local regular files yet 565 */ 566 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) { 567 xfs_warn(ip->i_mount, 568 "corrupt inode %Lu (local format for regular file).", 569 (unsigned long long) ip->i_ino); 570 XFS_CORRUPTION_ERROR("xfs_iformat(4)", 571 XFS_ERRLEVEL_LOW, 572 ip->i_mount, dip); 573 return XFS_ERROR(EFSCORRUPTED); 574 } 575 576 di_size = be64_to_cpu(dip->di_size); 577 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { 578 xfs_warn(ip->i_mount, 579 "corrupt inode %Lu (bad size %Ld for local inode).", 580 (unsigned long long) ip->i_ino, 581 (long long) di_size); 582 XFS_CORRUPTION_ERROR("xfs_iformat(5)", 583 XFS_ERRLEVEL_LOW, 584 ip->i_mount, dip); 585 return XFS_ERROR(EFSCORRUPTED); 586 } 587 588 size = (int)di_size; 589 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); 590 break; 591 case XFS_DINODE_FMT_EXTENTS: 592 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); 593 break; 594 case XFS_DINODE_FMT_BTREE: 595 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); 596 break; 597 default: 598 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, 599 ip->i_mount); 600 return XFS_ERROR(EFSCORRUPTED); 601 } 602 break; 603 604 default: 605 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); 606 return XFS_ERROR(EFSCORRUPTED); 607 } 608 if (error) { 609 return error; 610 } 611 if (!XFS_DFORK_Q(dip)) 612 return 0; 613 614 ASSERT(ip->i_afp == NULL); 615 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS); 616 617 switch (dip->di_aformat) { 618 case XFS_DINODE_FMT_LOCAL: 619 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); 620 size = be16_to_cpu(atp->hdr.totsize); 621 622 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) { 623 xfs_warn(ip->i_mount, 624 "corrupt inode %Lu (bad attr fork size %Ld).", 625 (unsigned long long) ip->i_ino, 626 (long long) size); 627 XFS_CORRUPTION_ERROR("xfs_iformat(8)", 628 XFS_ERRLEVEL_LOW, 629 ip->i_mount, dip); 630 return XFS_ERROR(EFSCORRUPTED); 631 } 632 633 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); 634 break; 635 case XFS_DINODE_FMT_EXTENTS: 636 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); 637 break; 638 case XFS_DINODE_FMT_BTREE: 639 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); 640 break; 641 default: 642 error = XFS_ERROR(EFSCORRUPTED); 643 break; 644 } 645 if (error) { 646 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 647 ip->i_afp = NULL; 648 xfs_idestroy_fork(ip, XFS_DATA_FORK); 649 } 650 return error; 651 } 652 653 /* 654 * The file is in-lined in the on-disk inode. 655 * If it fits into if_inline_data, then copy 656 * it there, otherwise allocate a buffer for it 657 * and copy the data there. Either way, set 658 * if_data to point at the data. 659 * If we allocate a buffer for the data, make 660 * sure that its size is a multiple of 4 and 661 * record the real size in i_real_bytes. 662 */ 663 STATIC int 664 xfs_iformat_local( 665 xfs_inode_t *ip, 666 xfs_dinode_t *dip, 667 int whichfork, 668 int size) 669 { 670 xfs_ifork_t *ifp; 671 int real_size; 672 673 /* 674 * If the size is unreasonable, then something 675 * is wrong and we just bail out rather than crash in 676 * kmem_alloc() or memcpy() below. 677 */ 678 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 679 xfs_warn(ip->i_mount, 680 "corrupt inode %Lu (bad size %d for local fork, size = %d).", 681 (unsigned long long) ip->i_ino, size, 682 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); 683 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, 684 ip->i_mount, dip); 685 return XFS_ERROR(EFSCORRUPTED); 686 } 687 ifp = XFS_IFORK_PTR(ip, whichfork); 688 real_size = 0; 689 if (size == 0) 690 ifp->if_u1.if_data = NULL; 691 else if (size <= sizeof(ifp->if_u2.if_inline_data)) 692 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 693 else { 694 real_size = roundup(size, 4); 695 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS); 696 } 697 ifp->if_bytes = size; 698 ifp->if_real_bytes = real_size; 699 if (size) 700 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); 701 ifp->if_flags &= ~XFS_IFEXTENTS; 702 ifp->if_flags |= XFS_IFINLINE; 703 return 0; 704 } 705 706 /* 707 * The file consists of a set of extents all 708 * of which fit into the on-disk inode. 709 * If there are few enough extents to fit into 710 * the if_inline_ext, then copy them there. 711 * Otherwise allocate a buffer for them and copy 712 * them into it. Either way, set if_extents 713 * to point at the extents. 714 */ 715 STATIC int 716 xfs_iformat_extents( 717 xfs_inode_t *ip, 718 xfs_dinode_t *dip, 719 int whichfork) 720 { 721 xfs_bmbt_rec_t *dp; 722 xfs_ifork_t *ifp; 723 int nex; 724 int size; 725 int i; 726 727 ifp = XFS_IFORK_PTR(ip, whichfork); 728 nex = XFS_DFORK_NEXTENTS(dip, whichfork); 729 size = nex * (uint)sizeof(xfs_bmbt_rec_t); 730 731 /* 732 * If the number of extents is unreasonable, then something 733 * is wrong and we just bail out rather than crash in 734 * kmem_alloc() or memcpy() below. 735 */ 736 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 737 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).", 738 (unsigned long long) ip->i_ino, nex); 739 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, 740 ip->i_mount, dip); 741 return XFS_ERROR(EFSCORRUPTED); 742 } 743 744 ifp->if_real_bytes = 0; 745 if (nex == 0) 746 ifp->if_u1.if_extents = NULL; 747 else if (nex <= XFS_INLINE_EXTS) 748 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 749 else 750 xfs_iext_add(ifp, 0, nex); 751 752 ifp->if_bytes = size; 753 if (size) { 754 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); 755 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); 756 for (i = 0; i < nex; i++, dp++) { 757 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 758 ep->l0 = get_unaligned_be64(&dp->l0); 759 ep->l1 = get_unaligned_be64(&dp->l1); 760 } 761 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); 762 if (whichfork != XFS_DATA_FORK || 763 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) 764 if (unlikely(xfs_check_nostate_extents( 765 ifp, 0, nex))) { 766 XFS_ERROR_REPORT("xfs_iformat_extents(2)", 767 XFS_ERRLEVEL_LOW, 768 ip->i_mount); 769 return XFS_ERROR(EFSCORRUPTED); 770 } 771 } 772 ifp->if_flags |= XFS_IFEXTENTS; 773 return 0; 774 } 775 776 /* 777 * The file has too many extents to fit into 778 * the inode, so they are in B-tree format. 779 * Allocate a buffer for the root of the B-tree 780 * and copy the root into it. The i_extents 781 * field will remain NULL until all of the 782 * extents are read in (when they are needed). 783 */ 784 STATIC int 785 xfs_iformat_btree( 786 xfs_inode_t *ip, 787 xfs_dinode_t *dip, 788 int whichfork) 789 { 790 struct xfs_mount *mp = ip->i_mount; 791 xfs_bmdr_block_t *dfp; 792 xfs_ifork_t *ifp; 793 /* REFERENCED */ 794 int nrecs; 795 int size; 796 797 ifp = XFS_IFORK_PTR(ip, whichfork); 798 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); 799 size = XFS_BMAP_BROOT_SPACE(mp, dfp); 800 nrecs = be16_to_cpu(dfp->bb_numrecs); 801 802 /* 803 * blow out if -- fork has less extents than can fit in 804 * fork (fork shouldn't be a btree format), root btree 805 * block has more records than can fit into the fork, 806 * or the number of extents is greater than the number of 807 * blocks. 808 */ 809 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= 810 XFS_IFORK_MAXEXT(ip, whichfork) || 811 XFS_BMDR_SPACE_CALC(nrecs) > 812 XFS_DFORK_SIZE(dip, mp, whichfork) || 813 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { 814 xfs_warn(mp, "corrupt inode %Lu (btree).", 815 (unsigned long long) ip->i_ino); 816 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW, 817 mp, dip); 818 return XFS_ERROR(EFSCORRUPTED); 819 } 820 821 ifp->if_broot_bytes = size; 822 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS); 823 ASSERT(ifp->if_broot != NULL); 824 /* 825 * Copy and convert from the on-disk structure 826 * to the in-memory structure. 827 */ 828 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), 829 ifp->if_broot, size); 830 ifp->if_flags &= ~XFS_IFEXTENTS; 831 ifp->if_flags |= XFS_IFBROOT; 832 833 return 0; 834 } 835 836 STATIC void 837 xfs_dinode_from_disk( 838 xfs_icdinode_t *to, 839 xfs_dinode_t *from) 840 { 841 to->di_magic = be16_to_cpu(from->di_magic); 842 to->di_mode = be16_to_cpu(from->di_mode); 843 to->di_version = from ->di_version; 844 to->di_format = from->di_format; 845 to->di_onlink = be16_to_cpu(from->di_onlink); 846 to->di_uid = be32_to_cpu(from->di_uid); 847 to->di_gid = be32_to_cpu(from->di_gid); 848 to->di_nlink = be32_to_cpu(from->di_nlink); 849 to->di_projid_lo = be16_to_cpu(from->di_projid_lo); 850 to->di_projid_hi = be16_to_cpu(from->di_projid_hi); 851 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); 852 to->di_flushiter = be16_to_cpu(from->di_flushiter); 853 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); 854 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); 855 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); 856 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); 857 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); 858 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); 859 to->di_size = be64_to_cpu(from->di_size); 860 to->di_nblocks = be64_to_cpu(from->di_nblocks); 861 to->di_extsize = be32_to_cpu(from->di_extsize); 862 to->di_nextents = be32_to_cpu(from->di_nextents); 863 to->di_anextents = be16_to_cpu(from->di_anextents); 864 to->di_forkoff = from->di_forkoff; 865 to->di_aformat = from->di_aformat; 866 to->di_dmevmask = be32_to_cpu(from->di_dmevmask); 867 to->di_dmstate = be16_to_cpu(from->di_dmstate); 868 to->di_flags = be16_to_cpu(from->di_flags); 869 to->di_gen = be32_to_cpu(from->di_gen); 870 871 if (to->di_version == 3) { 872 to->di_changecount = be64_to_cpu(from->di_changecount); 873 to->di_crtime.t_sec = be32_to_cpu(from->di_crtime.t_sec); 874 to->di_crtime.t_nsec = be32_to_cpu(from->di_crtime.t_nsec); 875 to->di_flags2 = be64_to_cpu(from->di_flags2); 876 to->di_ino = be64_to_cpu(from->di_ino); 877 to->di_lsn = be64_to_cpu(from->di_lsn); 878 memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2)); 879 uuid_copy(&to->di_uuid, &from->di_uuid); 880 } 881 } 882 883 void 884 xfs_dinode_to_disk( 885 xfs_dinode_t *to, 886 xfs_icdinode_t *from) 887 { 888 to->di_magic = cpu_to_be16(from->di_magic); 889 to->di_mode = cpu_to_be16(from->di_mode); 890 to->di_version = from ->di_version; 891 to->di_format = from->di_format; 892 to->di_onlink = cpu_to_be16(from->di_onlink); 893 to->di_uid = cpu_to_be32(from->di_uid); 894 to->di_gid = cpu_to_be32(from->di_gid); 895 to->di_nlink = cpu_to_be32(from->di_nlink); 896 to->di_projid_lo = cpu_to_be16(from->di_projid_lo); 897 to->di_projid_hi = cpu_to_be16(from->di_projid_hi); 898 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); 899 to->di_flushiter = cpu_to_be16(from->di_flushiter); 900 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); 901 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); 902 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); 903 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); 904 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); 905 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); 906 to->di_size = cpu_to_be64(from->di_size); 907 to->di_nblocks = cpu_to_be64(from->di_nblocks); 908 to->di_extsize = cpu_to_be32(from->di_extsize); 909 to->di_nextents = cpu_to_be32(from->di_nextents); 910 to->di_anextents = cpu_to_be16(from->di_anextents); 911 to->di_forkoff = from->di_forkoff; 912 to->di_aformat = from->di_aformat; 913 to->di_dmevmask = cpu_to_be32(from->di_dmevmask); 914 to->di_dmstate = cpu_to_be16(from->di_dmstate); 915 to->di_flags = cpu_to_be16(from->di_flags); 916 to->di_gen = cpu_to_be32(from->di_gen); 917 918 if (from->di_version == 3) { 919 to->di_changecount = cpu_to_be64(from->di_changecount); 920 to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec); 921 to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec); 922 to->di_flags2 = cpu_to_be64(from->di_flags2); 923 to->di_ino = cpu_to_be64(from->di_ino); 924 to->di_lsn = cpu_to_be64(from->di_lsn); 925 memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2)); 926 uuid_copy(&to->di_uuid, &from->di_uuid); 927 } 928 } 929 930 STATIC uint 931 _xfs_dic2xflags( 932 __uint16_t di_flags) 933 { 934 uint flags = 0; 935 936 if (di_flags & XFS_DIFLAG_ANY) { 937 if (di_flags & XFS_DIFLAG_REALTIME) 938 flags |= XFS_XFLAG_REALTIME; 939 if (di_flags & XFS_DIFLAG_PREALLOC) 940 flags |= XFS_XFLAG_PREALLOC; 941 if (di_flags & XFS_DIFLAG_IMMUTABLE) 942 flags |= XFS_XFLAG_IMMUTABLE; 943 if (di_flags & XFS_DIFLAG_APPEND) 944 flags |= XFS_XFLAG_APPEND; 945 if (di_flags & XFS_DIFLAG_SYNC) 946 flags |= XFS_XFLAG_SYNC; 947 if (di_flags & XFS_DIFLAG_NOATIME) 948 flags |= XFS_XFLAG_NOATIME; 949 if (di_flags & XFS_DIFLAG_NODUMP) 950 flags |= XFS_XFLAG_NODUMP; 951 if (di_flags & XFS_DIFLAG_RTINHERIT) 952 flags |= XFS_XFLAG_RTINHERIT; 953 if (di_flags & XFS_DIFLAG_PROJINHERIT) 954 flags |= XFS_XFLAG_PROJINHERIT; 955 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 956 flags |= XFS_XFLAG_NOSYMLINKS; 957 if (di_flags & XFS_DIFLAG_EXTSIZE) 958 flags |= XFS_XFLAG_EXTSIZE; 959 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 960 flags |= XFS_XFLAG_EXTSZINHERIT; 961 if (di_flags & XFS_DIFLAG_NODEFRAG) 962 flags |= XFS_XFLAG_NODEFRAG; 963 if (di_flags & XFS_DIFLAG_FILESTREAM) 964 flags |= XFS_XFLAG_FILESTREAM; 965 } 966 967 return flags; 968 } 969 970 uint 971 xfs_ip2xflags( 972 xfs_inode_t *ip) 973 { 974 xfs_icdinode_t *dic = &ip->i_d; 975 976 return _xfs_dic2xflags(dic->di_flags) | 977 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); 978 } 979 980 uint 981 xfs_dic2xflags( 982 xfs_dinode_t *dip) 983 { 984 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | 985 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); 986 } 987 988 static bool 989 xfs_dinode_verify( 990 struct xfs_mount *mp, 991 struct xfs_inode *ip, 992 struct xfs_dinode *dip) 993 { 994 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) 995 return false; 996 997 /* only version 3 or greater inodes are extensively verified here */ 998 if (dip->di_version < 3) 999 return true; 1000 1001 if (!xfs_sb_version_hascrc(&mp->m_sb)) 1002 return false; 1003 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize, 1004 offsetof(struct xfs_dinode, di_crc))) 1005 return false; 1006 if (be64_to_cpu(dip->di_ino) != ip->i_ino) 1007 return false; 1008 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_uuid)) 1009 return false; 1010 return true; 1011 } 1012 1013 void 1014 xfs_dinode_calc_crc( 1015 struct xfs_mount *mp, 1016 struct xfs_dinode *dip) 1017 { 1018 __uint32_t crc; 1019 1020 if (dip->di_version < 3) 1021 return; 1022 1023 ASSERT(xfs_sb_version_hascrc(&mp->m_sb)); 1024 crc = xfs_start_cksum((char *)dip, mp->m_sb.sb_inodesize, 1025 offsetof(struct xfs_dinode, di_crc)); 1026 dip->di_crc = xfs_end_cksum(crc); 1027 } 1028 1029 /* 1030 * Read the disk inode attributes into the in-core inode structure. 1031 */ 1032 int 1033 xfs_iread( 1034 xfs_mount_t *mp, 1035 xfs_trans_t *tp, 1036 xfs_inode_t *ip, 1037 uint iget_flags) 1038 { 1039 xfs_buf_t *bp; 1040 xfs_dinode_t *dip; 1041 int error; 1042 1043 /* 1044 * Fill in the location information in the in-core inode. 1045 */ 1046 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags); 1047 if (error) 1048 return error; 1049 1050 /* 1051 * Get pointers to the on-disk inode and the buffer containing it. 1052 */ 1053 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags); 1054 if (error) 1055 return error; 1056 1057 /* even unallocated inodes are verified */ 1058 if (!xfs_dinode_verify(mp, ip, dip)) { 1059 xfs_alert(mp, "%s: validation failed for inode %lld failed", 1060 __func__, ip->i_ino); 1061 1062 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, dip); 1063 error = XFS_ERROR(EFSCORRUPTED); 1064 goto out_brelse; 1065 } 1066 1067 /* 1068 * If the on-disk inode is already linked to a directory 1069 * entry, copy all of the inode into the in-core inode. 1070 * xfs_iformat() handles copying in the inode format 1071 * specific information. 1072 * Otherwise, just get the truly permanent information. 1073 */ 1074 if (dip->di_mode) { 1075 xfs_dinode_from_disk(&ip->i_d, dip); 1076 error = xfs_iformat(ip, dip); 1077 if (error) { 1078 #ifdef DEBUG 1079 xfs_alert(mp, "%s: xfs_iformat() returned error %d", 1080 __func__, error); 1081 #endif /* DEBUG */ 1082 goto out_brelse; 1083 } 1084 } else { 1085 /* 1086 * Partial initialisation of the in-core inode. Just the bits 1087 * that xfs_ialloc won't overwrite or relies on being correct. 1088 */ 1089 ip->i_d.di_magic = be16_to_cpu(dip->di_magic); 1090 ip->i_d.di_version = dip->di_version; 1091 ip->i_d.di_gen = be32_to_cpu(dip->di_gen); 1092 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter); 1093 1094 if (dip->di_version == 3) { 1095 ip->i_d.di_ino = be64_to_cpu(dip->di_ino); 1096 uuid_copy(&ip->i_d.di_uuid, &dip->di_uuid); 1097 } 1098 1099 /* 1100 * Make sure to pull in the mode here as well in 1101 * case the inode is released without being used. 1102 * This ensures that xfs_inactive() will see that 1103 * the inode is already free and not try to mess 1104 * with the uninitialized part of it. 1105 */ 1106 ip->i_d.di_mode = 0; 1107 } 1108 1109 /* 1110 * The inode format changed when we moved the link count and 1111 * made it 32 bits long. If this is an old format inode, 1112 * convert it in memory to look like a new one. If it gets 1113 * flushed to disk we will convert back before flushing or 1114 * logging it. We zero out the new projid field and the old link 1115 * count field. We'll handle clearing the pad field (the remains 1116 * of the old uuid field) when we actually convert the inode to 1117 * the new format. We don't change the version number so that we 1118 * can distinguish this from a real new format inode. 1119 */ 1120 if (ip->i_d.di_version == 1) { 1121 ip->i_d.di_nlink = ip->i_d.di_onlink; 1122 ip->i_d.di_onlink = 0; 1123 xfs_set_projid(ip, 0); 1124 } 1125 1126 ip->i_delayed_blks = 0; 1127 1128 /* 1129 * Mark the buffer containing the inode as something to keep 1130 * around for a while. This helps to keep recently accessed 1131 * meta-data in-core longer. 1132 */ 1133 xfs_buf_set_ref(bp, XFS_INO_REF); 1134 1135 /* 1136 * Use xfs_trans_brelse() to release the buffer containing the 1137 * on-disk inode, because it was acquired with xfs_trans_read_buf() 1138 * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal 1139 * brelse(). If we're within a transaction, then xfs_trans_brelse() 1140 * will only release the buffer if it is not dirty within the 1141 * transaction. It will be OK to release the buffer in this case, 1142 * because inodes on disk are never destroyed and we will be 1143 * locking the new in-core inode before putting it in the hash 1144 * table where other processes can find it. Thus we don't have 1145 * to worry about the inode being changed just because we released 1146 * the buffer. 1147 */ 1148 out_brelse: 1149 xfs_trans_brelse(tp, bp); 1150 return error; 1151 } 1152 1153 /* 1154 * Read in extents from a btree-format inode. 1155 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. 1156 */ 1157 int 1158 xfs_iread_extents( 1159 xfs_trans_t *tp, 1160 xfs_inode_t *ip, 1161 int whichfork) 1162 { 1163 int error; 1164 xfs_ifork_t *ifp; 1165 xfs_extnum_t nextents; 1166 1167 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { 1168 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, 1169 ip->i_mount); 1170 return XFS_ERROR(EFSCORRUPTED); 1171 } 1172 nextents = XFS_IFORK_NEXTENTS(ip, whichfork); 1173 ifp = XFS_IFORK_PTR(ip, whichfork); 1174 1175 /* 1176 * We know that the size is valid (it's checked in iformat_btree) 1177 */ 1178 ifp->if_bytes = ifp->if_real_bytes = 0; 1179 ifp->if_flags |= XFS_IFEXTENTS; 1180 xfs_iext_add(ifp, 0, nextents); 1181 error = xfs_bmap_read_extents(tp, ip, whichfork); 1182 if (error) { 1183 xfs_iext_destroy(ifp); 1184 ifp->if_flags &= ~XFS_IFEXTENTS; 1185 return error; 1186 } 1187 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); 1188 return 0; 1189 } 1190 1191 /* 1192 * Allocate an inode on disk and return a copy of its in-core version. 1193 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 1194 * appropriately within the inode. The uid and gid for the inode are 1195 * set according to the contents of the given cred structure. 1196 * 1197 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 1198 * has a free inode available, call xfs_iget() to obtain the in-core 1199 * version of the allocated inode. Finally, fill in the inode and 1200 * log its initial contents. In this case, ialloc_context would be 1201 * set to NULL. 1202 * 1203 * If xfs_dialloc() does not have an available inode, it will replenish 1204 * its supply by doing an allocation. Since we can only do one 1205 * allocation within a transaction without deadlocks, we must commit 1206 * the current transaction before returning the inode itself. 1207 * In this case, therefore, we will set ialloc_context and return. 1208 * The caller should then commit the current transaction, start a new 1209 * transaction, and call xfs_ialloc() again to actually get the inode. 1210 * 1211 * To ensure that some other process does not grab the inode that 1212 * was allocated during the first call to xfs_ialloc(), this routine 1213 * also returns the [locked] bp pointing to the head of the freelist 1214 * as ialloc_context. The caller should hold this buffer across 1215 * the commit and pass it back into this routine on the second call. 1216 * 1217 * If we are allocating quota inodes, we do not have a parent inode 1218 * to attach to or associate with (i.e. pip == NULL) because they 1219 * are not linked into the directory structure - they are attached 1220 * directly to the superblock - and so have no parent. 1221 */ 1222 int 1223 xfs_ialloc( 1224 xfs_trans_t *tp, 1225 xfs_inode_t *pip, 1226 umode_t mode, 1227 xfs_nlink_t nlink, 1228 xfs_dev_t rdev, 1229 prid_t prid, 1230 int okalloc, 1231 xfs_buf_t **ialloc_context, 1232 xfs_inode_t **ipp) 1233 { 1234 struct xfs_mount *mp = tp->t_mountp; 1235 xfs_ino_t ino; 1236 xfs_inode_t *ip; 1237 uint flags; 1238 int error; 1239 timespec_t tv; 1240 int filestreams = 0; 1241 1242 /* 1243 * Call the space management code to pick 1244 * the on-disk inode to be allocated. 1245 */ 1246 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 1247 ialloc_context, &ino); 1248 if (error) 1249 return error; 1250 if (*ialloc_context || ino == NULLFSINO) { 1251 *ipp = NULL; 1252 return 0; 1253 } 1254 ASSERT(*ialloc_context == NULL); 1255 1256 /* 1257 * Get the in-core inode with the lock held exclusively. 1258 * This is because we're setting fields here we need 1259 * to prevent others from looking at until we're done. 1260 */ 1261 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 1262 XFS_ILOCK_EXCL, &ip); 1263 if (error) 1264 return error; 1265 ASSERT(ip != NULL); 1266 1267 ip->i_d.di_mode = mode; 1268 ip->i_d.di_onlink = 0; 1269 ip->i_d.di_nlink = nlink; 1270 ASSERT(ip->i_d.di_nlink == nlink); 1271 ip->i_d.di_uid = current_fsuid(); 1272 ip->i_d.di_gid = current_fsgid(); 1273 xfs_set_projid(ip, prid); 1274 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 1275 1276 /* 1277 * If the superblock version is up to where we support new format 1278 * inodes and this is currently an old format inode, then change 1279 * the inode version number now. This way we only do the conversion 1280 * here rather than here and in the flush/logging code. 1281 */ 1282 if (xfs_sb_version_hasnlink(&mp->m_sb) && 1283 ip->i_d.di_version == 1) { 1284 ip->i_d.di_version = 2; 1285 /* 1286 * We've already zeroed the old link count, the projid field, 1287 * and the pad field. 1288 */ 1289 } 1290 1291 /* 1292 * Project ids won't be stored on disk if we are using a version 1 inode. 1293 */ 1294 if ((prid != 0) && (ip->i_d.di_version == 1)) 1295 xfs_bump_ino_vers2(tp, ip); 1296 1297 if (pip && XFS_INHERIT_GID(pip)) { 1298 ip->i_d.di_gid = pip->i_d.di_gid; 1299 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { 1300 ip->i_d.di_mode |= S_ISGID; 1301 } 1302 } 1303 1304 /* 1305 * If the group ID of the new file does not match the effective group 1306 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 1307 * (and only if the irix_sgid_inherit compatibility variable is set). 1308 */ 1309 if ((irix_sgid_inherit) && 1310 (ip->i_d.di_mode & S_ISGID) && 1311 (!in_group_p((gid_t)ip->i_d.di_gid))) { 1312 ip->i_d.di_mode &= ~S_ISGID; 1313 } 1314 1315 ip->i_d.di_size = 0; 1316 ip->i_d.di_nextents = 0; 1317 ASSERT(ip->i_d.di_nblocks == 0); 1318 1319 nanotime(&tv); 1320 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; 1321 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; 1322 ip->i_d.di_atime = ip->i_d.di_mtime; 1323 ip->i_d.di_ctime = ip->i_d.di_mtime; 1324 1325 /* 1326 * di_gen will have been taken care of in xfs_iread. 1327 */ 1328 ip->i_d.di_extsize = 0; 1329 ip->i_d.di_dmevmask = 0; 1330 ip->i_d.di_dmstate = 0; 1331 ip->i_d.di_flags = 0; 1332 1333 if (ip->i_d.di_version == 3) { 1334 ASSERT(ip->i_d.di_ino == ino); 1335 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); 1336 ip->i_d.di_crc = 0; 1337 ip->i_d.di_changecount = 1; 1338 ip->i_d.di_lsn = 0; 1339 ip->i_d.di_flags2 = 0; 1340 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); 1341 ip->i_d.di_crtime = ip->i_d.di_mtime; 1342 } 1343 1344 1345 flags = XFS_ILOG_CORE; 1346 switch (mode & S_IFMT) { 1347 case S_IFIFO: 1348 case S_IFCHR: 1349 case S_IFBLK: 1350 case S_IFSOCK: 1351 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 1352 ip->i_df.if_u2.if_rdev = rdev; 1353 ip->i_df.if_flags = 0; 1354 flags |= XFS_ILOG_DEV; 1355 break; 1356 case S_IFREG: 1357 /* 1358 * we can't set up filestreams until after the VFS inode 1359 * is set up properly. 1360 */ 1361 if (pip && xfs_inode_is_filestream(pip)) 1362 filestreams = 1; 1363 /* fall through */ 1364 case S_IFDIR: 1365 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 1366 uint di_flags = 0; 1367 1368 if (S_ISDIR(mode)) { 1369 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 1370 di_flags |= XFS_DIFLAG_RTINHERIT; 1371 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1372 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 1373 ip->i_d.di_extsize = pip->i_d.di_extsize; 1374 } 1375 } else if (S_ISREG(mode)) { 1376 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 1377 di_flags |= XFS_DIFLAG_REALTIME; 1378 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1379 di_flags |= XFS_DIFLAG_EXTSIZE; 1380 ip->i_d.di_extsize = pip->i_d.di_extsize; 1381 } 1382 } 1383 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 1384 xfs_inherit_noatime) 1385 di_flags |= XFS_DIFLAG_NOATIME; 1386 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 1387 xfs_inherit_nodump) 1388 di_flags |= XFS_DIFLAG_NODUMP; 1389 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 1390 xfs_inherit_sync) 1391 di_flags |= XFS_DIFLAG_SYNC; 1392 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 1393 xfs_inherit_nosymlinks) 1394 di_flags |= XFS_DIFLAG_NOSYMLINKS; 1395 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 1396 di_flags |= XFS_DIFLAG_PROJINHERIT; 1397 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 1398 xfs_inherit_nodefrag) 1399 di_flags |= XFS_DIFLAG_NODEFRAG; 1400 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 1401 di_flags |= XFS_DIFLAG_FILESTREAM; 1402 ip->i_d.di_flags |= di_flags; 1403 } 1404 /* FALLTHROUGH */ 1405 case S_IFLNK: 1406 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 1407 ip->i_df.if_flags = XFS_IFEXTENTS; 1408 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 1409 ip->i_df.if_u1.if_extents = NULL; 1410 break; 1411 default: 1412 ASSERT(0); 1413 } 1414 /* 1415 * Attribute fork settings for new inode. 1416 */ 1417 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 1418 ip->i_d.di_anextents = 0; 1419 1420 /* 1421 * Log the new values stuffed into the inode. 1422 */ 1423 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1424 xfs_trans_log_inode(tp, ip, flags); 1425 1426 /* now that we have an i_mode we can setup inode ops and unlock */ 1427 xfs_setup_inode(ip); 1428 1429 /* now we have set up the vfs inode we can associate the filestream */ 1430 if (filestreams) { 1431 error = xfs_filestream_associate(pip, ip); 1432 if (error < 0) 1433 return -error; 1434 if (!error) 1435 xfs_iflags_set(ip, XFS_IFILESTREAM); 1436 } 1437 1438 *ipp = ip; 1439 return 0; 1440 } 1441 1442 /* 1443 * Free up the underlying blocks past new_size. The new size must be smaller 1444 * than the current size. This routine can be used both for the attribute and 1445 * data fork, and does not modify the inode size, which is left to the caller. 1446 * 1447 * The transaction passed to this routine must have made a permanent log 1448 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1449 * given transaction and start new ones, so make sure everything involved in 1450 * the transaction is tidy before calling here. Some transaction will be 1451 * returned to the caller to be committed. The incoming transaction must 1452 * already include the inode, and both inode locks must be held exclusively. 1453 * The inode must also be "held" within the transaction. On return the inode 1454 * will be "held" within the returned transaction. This routine does NOT 1455 * require any disk space to be reserved for it within the transaction. 1456 * 1457 * If we get an error, we must return with the inode locked and linked into the 1458 * current transaction. This keeps things simple for the higher level code, 1459 * because it always knows that the inode is locked and held in the transaction 1460 * that returns to it whether errors occur or not. We don't mark the inode 1461 * dirty on error so that transactions can be easily aborted if possible. 1462 */ 1463 int 1464 xfs_itruncate_extents( 1465 struct xfs_trans **tpp, 1466 struct xfs_inode *ip, 1467 int whichfork, 1468 xfs_fsize_t new_size) 1469 { 1470 struct xfs_mount *mp = ip->i_mount; 1471 struct xfs_trans *tp = *tpp; 1472 struct xfs_trans *ntp; 1473 xfs_bmap_free_t free_list; 1474 xfs_fsblock_t first_block; 1475 xfs_fileoff_t first_unmap_block; 1476 xfs_fileoff_t last_block; 1477 xfs_filblks_t unmap_len; 1478 int committed; 1479 int error = 0; 1480 int done = 0; 1481 1482 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1483 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1484 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1485 ASSERT(new_size <= XFS_ISIZE(ip)); 1486 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1487 ASSERT(ip->i_itemp != NULL); 1488 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1489 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1490 1491 trace_xfs_itruncate_extents_start(ip, new_size); 1492 1493 /* 1494 * Since it is possible for space to become allocated beyond 1495 * the end of the file (in a crash where the space is allocated 1496 * but the inode size is not yet updated), simply remove any 1497 * blocks which show up between the new EOF and the maximum 1498 * possible file size. If the first block to be removed is 1499 * beyond the maximum file size (ie it is the same as last_block), 1500 * then there is nothing to do. 1501 */ 1502 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1503 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1504 if (first_unmap_block == last_block) 1505 return 0; 1506 1507 ASSERT(first_unmap_block < last_block); 1508 unmap_len = last_block - first_unmap_block + 1; 1509 while (!done) { 1510 xfs_bmap_init(&free_list, &first_block); 1511 error = xfs_bunmapi(tp, ip, 1512 first_unmap_block, unmap_len, 1513 xfs_bmapi_aflag(whichfork), 1514 XFS_ITRUNC_MAX_EXTENTS, 1515 &first_block, &free_list, 1516 &done); 1517 if (error) 1518 goto out_bmap_cancel; 1519 1520 /* 1521 * Duplicate the transaction that has the permanent 1522 * reservation and commit the old transaction. 1523 */ 1524 error = xfs_bmap_finish(&tp, &free_list, &committed); 1525 if (committed) 1526 xfs_trans_ijoin(tp, ip, 0); 1527 if (error) 1528 goto out_bmap_cancel; 1529 1530 if (committed) { 1531 /* 1532 * Mark the inode dirty so it will be logged and 1533 * moved forward in the log as part of every commit. 1534 */ 1535 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1536 } 1537 1538 ntp = xfs_trans_dup(tp); 1539 error = xfs_trans_commit(tp, 0); 1540 tp = ntp; 1541 1542 xfs_trans_ijoin(tp, ip, 0); 1543 1544 if (error) 1545 goto out; 1546 1547 /* 1548 * Transaction commit worked ok so we can drop the extra ticket 1549 * reference that we gained in xfs_trans_dup() 1550 */ 1551 xfs_log_ticket_put(tp->t_ticket); 1552 error = xfs_trans_reserve(tp, 0, 1553 XFS_ITRUNCATE_LOG_RES(mp), 0, 1554 XFS_TRANS_PERM_LOG_RES, 1555 XFS_ITRUNCATE_LOG_COUNT); 1556 if (error) 1557 goto out; 1558 } 1559 1560 /* 1561 * Always re-log the inode so that our permanent transaction can keep 1562 * on rolling it forward in the log. 1563 */ 1564 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1565 1566 trace_xfs_itruncate_extents_end(ip, new_size); 1567 1568 out: 1569 *tpp = tp; 1570 return error; 1571 out_bmap_cancel: 1572 /* 1573 * If the bunmapi call encounters an error, return to the caller where 1574 * the transaction can be properly aborted. We just need to make sure 1575 * we're not holding any resources that we were not when we came in. 1576 */ 1577 xfs_bmap_cancel(&free_list); 1578 goto out; 1579 } 1580 1581 /* 1582 * This is called when the inode's link count goes to 0. 1583 * We place the on-disk inode on a list in the AGI. It 1584 * will be pulled from this list when the inode is freed. 1585 */ 1586 int 1587 xfs_iunlink( 1588 xfs_trans_t *tp, 1589 xfs_inode_t *ip) 1590 { 1591 xfs_mount_t *mp; 1592 xfs_agi_t *agi; 1593 xfs_dinode_t *dip; 1594 xfs_buf_t *agibp; 1595 xfs_buf_t *ibp; 1596 xfs_agino_t agino; 1597 short bucket_index; 1598 int offset; 1599 int error; 1600 1601 ASSERT(ip->i_d.di_nlink == 0); 1602 ASSERT(ip->i_d.di_mode != 0); 1603 1604 mp = tp->t_mountp; 1605 1606 /* 1607 * Get the agi buffer first. It ensures lock ordering 1608 * on the list. 1609 */ 1610 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1611 if (error) 1612 return error; 1613 agi = XFS_BUF_TO_AGI(agibp); 1614 1615 /* 1616 * Get the index into the agi hash table for the 1617 * list this inode will go on. 1618 */ 1619 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1620 ASSERT(agino != 0); 1621 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1622 ASSERT(agi->agi_unlinked[bucket_index]); 1623 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1624 1625 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1626 /* 1627 * There is already another inode in the bucket we need 1628 * to add ourselves to. Add us at the front of the list. 1629 * Here we put the head pointer into our next pointer, 1630 * and then we fall through to point the head at us. 1631 */ 1632 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1633 0, 0); 1634 if (error) 1635 return error; 1636 1637 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1638 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1639 offset = ip->i_imap.im_boffset + 1640 offsetof(xfs_dinode_t, di_next_unlinked); 1641 1642 /* need to recalc the inode CRC if appropriate */ 1643 xfs_dinode_calc_crc(mp, dip); 1644 1645 xfs_trans_inode_buf(tp, ibp); 1646 xfs_trans_log_buf(tp, ibp, offset, 1647 (offset + sizeof(xfs_agino_t) - 1)); 1648 xfs_inobp_check(mp, ibp); 1649 } 1650 1651 /* 1652 * Point the bucket head pointer at the inode being inserted. 1653 */ 1654 ASSERT(agino != 0); 1655 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1656 offset = offsetof(xfs_agi_t, agi_unlinked) + 1657 (sizeof(xfs_agino_t) * bucket_index); 1658 xfs_trans_log_buf(tp, agibp, offset, 1659 (offset + sizeof(xfs_agino_t) - 1)); 1660 return 0; 1661 } 1662 1663 /* 1664 * Pull the on-disk inode from the AGI unlinked list. 1665 */ 1666 STATIC int 1667 xfs_iunlink_remove( 1668 xfs_trans_t *tp, 1669 xfs_inode_t *ip) 1670 { 1671 xfs_ino_t next_ino; 1672 xfs_mount_t *mp; 1673 xfs_agi_t *agi; 1674 xfs_dinode_t *dip; 1675 xfs_buf_t *agibp; 1676 xfs_buf_t *ibp; 1677 xfs_agnumber_t agno; 1678 xfs_agino_t agino; 1679 xfs_agino_t next_agino; 1680 xfs_buf_t *last_ibp; 1681 xfs_dinode_t *last_dip = NULL; 1682 short bucket_index; 1683 int offset, last_offset = 0; 1684 int error; 1685 1686 mp = tp->t_mountp; 1687 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 1688 1689 /* 1690 * Get the agi buffer first. It ensures lock ordering 1691 * on the list. 1692 */ 1693 error = xfs_read_agi(mp, tp, agno, &agibp); 1694 if (error) 1695 return error; 1696 1697 agi = XFS_BUF_TO_AGI(agibp); 1698 1699 /* 1700 * Get the index into the agi hash table for the 1701 * list this inode will go on. 1702 */ 1703 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1704 ASSERT(agino != 0); 1705 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1706 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 1707 ASSERT(agi->agi_unlinked[bucket_index]); 1708 1709 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 1710 /* 1711 * We're at the head of the list. Get the inode's on-disk 1712 * buffer to see if there is anyone after us on the list. 1713 * Only modify our next pointer if it is not already NULLAGINO. 1714 * This saves us the overhead of dealing with the buffer when 1715 * there is no need to change it. 1716 */ 1717 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1718 0, 0); 1719 if (error) { 1720 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 1721 __func__, error); 1722 return error; 1723 } 1724 next_agino = be32_to_cpu(dip->di_next_unlinked); 1725 ASSERT(next_agino != 0); 1726 if (next_agino != NULLAGINO) { 1727 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 1728 offset = ip->i_imap.im_boffset + 1729 offsetof(xfs_dinode_t, di_next_unlinked); 1730 1731 /* need to recalc the inode CRC if appropriate */ 1732 xfs_dinode_calc_crc(mp, dip); 1733 1734 xfs_trans_inode_buf(tp, ibp); 1735 xfs_trans_log_buf(tp, ibp, offset, 1736 (offset + sizeof(xfs_agino_t) - 1)); 1737 xfs_inobp_check(mp, ibp); 1738 } else { 1739 xfs_trans_brelse(tp, ibp); 1740 } 1741 /* 1742 * Point the bucket head pointer at the next inode. 1743 */ 1744 ASSERT(next_agino != 0); 1745 ASSERT(next_agino != agino); 1746 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 1747 offset = offsetof(xfs_agi_t, agi_unlinked) + 1748 (sizeof(xfs_agino_t) * bucket_index); 1749 xfs_trans_log_buf(tp, agibp, offset, 1750 (offset + sizeof(xfs_agino_t) - 1)); 1751 } else { 1752 /* 1753 * We need to search the list for the inode being freed. 1754 */ 1755 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 1756 last_ibp = NULL; 1757 while (next_agino != agino) { 1758 struct xfs_imap imap; 1759 1760 if (last_ibp) 1761 xfs_trans_brelse(tp, last_ibp); 1762 1763 imap.im_blkno = 0; 1764 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 1765 1766 error = xfs_imap(mp, tp, next_ino, &imap, 0); 1767 if (error) { 1768 xfs_warn(mp, 1769 "%s: xfs_imap returned error %d.", 1770 __func__, error); 1771 return error; 1772 } 1773 1774 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 1775 &last_ibp, 0, 0); 1776 if (error) { 1777 xfs_warn(mp, 1778 "%s: xfs_imap_to_bp returned error %d.", 1779 __func__, error); 1780 return error; 1781 } 1782 1783 last_offset = imap.im_boffset; 1784 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 1785 ASSERT(next_agino != NULLAGINO); 1786 ASSERT(next_agino != 0); 1787 } 1788 1789 /* 1790 * Now last_ibp points to the buffer previous to us on the 1791 * unlinked list. Pull us from the list. 1792 */ 1793 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1794 0, 0); 1795 if (error) { 1796 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 1797 __func__, error); 1798 return error; 1799 } 1800 next_agino = be32_to_cpu(dip->di_next_unlinked); 1801 ASSERT(next_agino != 0); 1802 ASSERT(next_agino != agino); 1803 if (next_agino != NULLAGINO) { 1804 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 1805 offset = ip->i_imap.im_boffset + 1806 offsetof(xfs_dinode_t, di_next_unlinked); 1807 1808 /* need to recalc the inode CRC if appropriate */ 1809 xfs_dinode_calc_crc(mp, dip); 1810 1811 xfs_trans_inode_buf(tp, ibp); 1812 xfs_trans_log_buf(tp, ibp, offset, 1813 (offset + sizeof(xfs_agino_t) - 1)); 1814 xfs_inobp_check(mp, ibp); 1815 } else { 1816 xfs_trans_brelse(tp, ibp); 1817 } 1818 /* 1819 * Point the previous inode on the list to the next inode. 1820 */ 1821 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 1822 ASSERT(next_agino != 0); 1823 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 1824 1825 /* need to recalc the inode CRC if appropriate */ 1826 xfs_dinode_calc_crc(mp, last_dip); 1827 1828 xfs_trans_inode_buf(tp, last_ibp); 1829 xfs_trans_log_buf(tp, last_ibp, offset, 1830 (offset + sizeof(xfs_agino_t) - 1)); 1831 xfs_inobp_check(mp, last_ibp); 1832 } 1833 return 0; 1834 } 1835 1836 /* 1837 * A big issue when freeing the inode cluster is is that we _cannot_ skip any 1838 * inodes that are in memory - they all must be marked stale and attached to 1839 * the cluster buffer. 1840 */ 1841 STATIC int 1842 xfs_ifree_cluster( 1843 xfs_inode_t *free_ip, 1844 xfs_trans_t *tp, 1845 xfs_ino_t inum) 1846 { 1847 xfs_mount_t *mp = free_ip->i_mount; 1848 int blks_per_cluster; 1849 int nbufs; 1850 int ninodes; 1851 int i, j; 1852 xfs_daddr_t blkno; 1853 xfs_buf_t *bp; 1854 xfs_inode_t *ip; 1855 xfs_inode_log_item_t *iip; 1856 xfs_log_item_t *lip; 1857 struct xfs_perag *pag; 1858 1859 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 1860 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { 1861 blks_per_cluster = 1; 1862 ninodes = mp->m_sb.sb_inopblock; 1863 nbufs = XFS_IALLOC_BLOCKS(mp); 1864 } else { 1865 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / 1866 mp->m_sb.sb_blocksize; 1867 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; 1868 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; 1869 } 1870 1871 for (j = 0; j < nbufs; j++, inum += ninodes) { 1872 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 1873 XFS_INO_TO_AGBNO(mp, inum)); 1874 1875 /* 1876 * We obtain and lock the backing buffer first in the process 1877 * here, as we have to ensure that any dirty inode that we 1878 * can't get the flush lock on is attached to the buffer. 1879 * If we scan the in-memory inodes first, then buffer IO can 1880 * complete before we get a lock on it, and hence we may fail 1881 * to mark all the active inodes on the buffer stale. 1882 */ 1883 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 1884 mp->m_bsize * blks_per_cluster, 1885 XBF_UNMAPPED); 1886 1887 if (!bp) 1888 return ENOMEM; 1889 1890 /* 1891 * This buffer may not have been correctly initialised as we 1892 * didn't read it from disk. That's not important because we are 1893 * only using to mark the buffer as stale in the log, and to 1894 * attach stale cached inodes on it. That means it will never be 1895 * dispatched for IO. If it is, we want to know about it, and we 1896 * want it to fail. We can acheive this by adding a write 1897 * verifier to the buffer. 1898 */ 1899 bp->b_ops = &xfs_inode_buf_ops; 1900 1901 /* 1902 * Walk the inodes already attached to the buffer and mark them 1903 * stale. These will all have the flush locks held, so an 1904 * in-memory inode walk can't lock them. By marking them all 1905 * stale first, we will not attempt to lock them in the loop 1906 * below as the XFS_ISTALE flag will be set. 1907 */ 1908 lip = bp->b_fspriv; 1909 while (lip) { 1910 if (lip->li_type == XFS_LI_INODE) { 1911 iip = (xfs_inode_log_item_t *)lip; 1912 ASSERT(iip->ili_logged == 1); 1913 lip->li_cb = xfs_istale_done; 1914 xfs_trans_ail_copy_lsn(mp->m_ail, 1915 &iip->ili_flush_lsn, 1916 &iip->ili_item.li_lsn); 1917 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 1918 } 1919 lip = lip->li_bio_list; 1920 } 1921 1922 1923 /* 1924 * For each inode in memory attempt to add it to the inode 1925 * buffer and set it up for being staled on buffer IO 1926 * completion. This is safe as we've locked out tail pushing 1927 * and flushing by locking the buffer. 1928 * 1929 * We have already marked every inode that was part of a 1930 * transaction stale above, which means there is no point in 1931 * even trying to lock them. 1932 */ 1933 for (i = 0; i < ninodes; i++) { 1934 retry: 1935 rcu_read_lock(); 1936 ip = radix_tree_lookup(&pag->pag_ici_root, 1937 XFS_INO_TO_AGINO(mp, (inum + i))); 1938 1939 /* Inode not in memory, nothing to do */ 1940 if (!ip) { 1941 rcu_read_unlock(); 1942 continue; 1943 } 1944 1945 /* 1946 * because this is an RCU protected lookup, we could 1947 * find a recently freed or even reallocated inode 1948 * during the lookup. We need to check under the 1949 * i_flags_lock for a valid inode here. Skip it if it 1950 * is not valid, the wrong inode or stale. 1951 */ 1952 spin_lock(&ip->i_flags_lock); 1953 if (ip->i_ino != inum + i || 1954 __xfs_iflags_test(ip, XFS_ISTALE)) { 1955 spin_unlock(&ip->i_flags_lock); 1956 rcu_read_unlock(); 1957 continue; 1958 } 1959 spin_unlock(&ip->i_flags_lock); 1960 1961 /* 1962 * Don't try to lock/unlock the current inode, but we 1963 * _cannot_ skip the other inodes that we did not find 1964 * in the list attached to the buffer and are not 1965 * already marked stale. If we can't lock it, back off 1966 * and retry. 1967 */ 1968 if (ip != free_ip && 1969 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 1970 rcu_read_unlock(); 1971 delay(1); 1972 goto retry; 1973 } 1974 rcu_read_unlock(); 1975 1976 xfs_iflock(ip); 1977 xfs_iflags_set(ip, XFS_ISTALE); 1978 1979 /* 1980 * we don't need to attach clean inodes or those only 1981 * with unlogged changes (which we throw away, anyway). 1982 */ 1983 iip = ip->i_itemp; 1984 if (!iip || xfs_inode_clean(ip)) { 1985 ASSERT(ip != free_ip); 1986 xfs_ifunlock(ip); 1987 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1988 continue; 1989 } 1990 1991 iip->ili_last_fields = iip->ili_fields; 1992 iip->ili_fields = 0; 1993 iip->ili_logged = 1; 1994 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 1995 &iip->ili_item.li_lsn); 1996 1997 xfs_buf_attach_iodone(bp, xfs_istale_done, 1998 &iip->ili_item); 1999 2000 if (ip != free_ip) 2001 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2002 } 2003 2004 xfs_trans_stale_inode_buf(tp, bp); 2005 xfs_trans_binval(tp, bp); 2006 } 2007 2008 xfs_perag_put(pag); 2009 return 0; 2010 } 2011 2012 /* 2013 * This is called to return an inode to the inode free list. 2014 * The inode should already be truncated to 0 length and have 2015 * no pages associated with it. This routine also assumes that 2016 * the inode is already a part of the transaction. 2017 * 2018 * The on-disk copy of the inode will have been added to the list 2019 * of unlinked inodes in the AGI. We need to remove the inode from 2020 * that list atomically with respect to freeing it here. 2021 */ 2022 int 2023 xfs_ifree( 2024 xfs_trans_t *tp, 2025 xfs_inode_t *ip, 2026 xfs_bmap_free_t *flist) 2027 { 2028 int error; 2029 int delete; 2030 xfs_ino_t first_ino; 2031 xfs_dinode_t *dip; 2032 xfs_buf_t *ibp; 2033 2034 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2035 ASSERT(ip->i_d.di_nlink == 0); 2036 ASSERT(ip->i_d.di_nextents == 0); 2037 ASSERT(ip->i_d.di_anextents == 0); 2038 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); 2039 ASSERT(ip->i_d.di_nblocks == 0); 2040 2041 /* 2042 * Pull the on-disk inode from the AGI unlinked list. 2043 */ 2044 error = xfs_iunlink_remove(tp, ip); 2045 if (error != 0) { 2046 return error; 2047 } 2048 2049 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); 2050 if (error != 0) { 2051 return error; 2052 } 2053 ip->i_d.di_mode = 0; /* mark incore inode as free */ 2054 ip->i_d.di_flags = 0; 2055 ip->i_d.di_dmevmask = 0; 2056 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2057 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2058 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2059 /* 2060 * Bump the generation count so no one will be confused 2061 * by reincarnations of this inode. 2062 */ 2063 ip->i_d.di_gen++; 2064 2065 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2066 2067 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp, 2068 0, 0); 2069 if (error) 2070 return error; 2071 2072 /* 2073 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat 2074 * from picking up this inode when it is reclaimed (its incore state 2075 * initialzed but not flushed to disk yet). The in-core di_mode is 2076 * already cleared and a corresponding transaction logged. 2077 * The hack here just synchronizes the in-core to on-disk 2078 * di_mode value in advance before the actual inode sync to disk. 2079 * This is OK because the inode is already unlinked and would never 2080 * change its di_mode again for this inode generation. 2081 * This is a temporary hack that would require a proper fix 2082 * in the future. 2083 */ 2084 dip->di_mode = 0; 2085 2086 if (delete) { 2087 error = xfs_ifree_cluster(ip, tp, first_ino); 2088 } 2089 2090 return error; 2091 } 2092 2093 /* 2094 * Reallocate the space for if_broot based on the number of records 2095 * being added or deleted as indicated in rec_diff. Move the records 2096 * and pointers in if_broot to fit the new size. When shrinking this 2097 * will eliminate holes between the records and pointers created by 2098 * the caller. When growing this will create holes to be filled in 2099 * by the caller. 2100 * 2101 * The caller must not request to add more records than would fit in 2102 * the on-disk inode root. If the if_broot is currently NULL, then 2103 * if we adding records one will be allocated. The caller must also 2104 * not request that the number of records go below zero, although 2105 * it can go to zero. 2106 * 2107 * ip -- the inode whose if_broot area is changing 2108 * ext_diff -- the change in the number of records, positive or negative, 2109 * requested for the if_broot array. 2110 */ 2111 void 2112 xfs_iroot_realloc( 2113 xfs_inode_t *ip, 2114 int rec_diff, 2115 int whichfork) 2116 { 2117 struct xfs_mount *mp = ip->i_mount; 2118 int cur_max; 2119 xfs_ifork_t *ifp; 2120 struct xfs_btree_block *new_broot; 2121 int new_max; 2122 size_t new_size; 2123 char *np; 2124 char *op; 2125 2126 /* 2127 * Handle the degenerate case quietly. 2128 */ 2129 if (rec_diff == 0) { 2130 return; 2131 } 2132 2133 ifp = XFS_IFORK_PTR(ip, whichfork); 2134 if (rec_diff > 0) { 2135 /* 2136 * If there wasn't any memory allocated before, just 2137 * allocate it now and get out. 2138 */ 2139 if (ifp->if_broot_bytes == 0) { 2140 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff); 2141 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); 2142 ifp->if_broot_bytes = (int)new_size; 2143 return; 2144 } 2145 2146 /* 2147 * If there is already an existing if_broot, then we need 2148 * to realloc() it and shift the pointers to their new 2149 * location. The records don't change location because 2150 * they are kept butted up against the btree block header. 2151 */ 2152 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); 2153 new_max = cur_max + rec_diff; 2154 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max); 2155 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, 2156 XFS_BMAP_BROOT_SPACE_CALC(mp, cur_max), 2157 KM_SLEEP | KM_NOFS); 2158 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2159 ifp->if_broot_bytes); 2160 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2161 (int)new_size); 2162 ifp->if_broot_bytes = (int)new_size; 2163 ASSERT(ifp->if_broot_bytes <= 2164 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ(ip)); 2165 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); 2166 return; 2167 } 2168 2169 /* 2170 * rec_diff is less than 0. In this case, we are shrinking the 2171 * if_broot buffer. It must already exist. If we go to zero 2172 * records, just get rid of the root and clear the status bit. 2173 */ 2174 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); 2175 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); 2176 new_max = cur_max + rec_diff; 2177 ASSERT(new_max >= 0); 2178 if (new_max > 0) 2179 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max); 2180 else 2181 new_size = 0; 2182 if (new_size > 0) { 2183 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); 2184 /* 2185 * First copy over the btree block header. 2186 */ 2187 memcpy(new_broot, ifp->if_broot, 2188 XFS_BMBT_BLOCK_LEN(ip->i_mount)); 2189 } else { 2190 new_broot = NULL; 2191 ifp->if_flags &= ~XFS_IFBROOT; 2192 } 2193 2194 /* 2195 * Only copy the records and pointers if there are any. 2196 */ 2197 if (new_max > 0) { 2198 /* 2199 * First copy the records. 2200 */ 2201 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); 2202 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); 2203 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); 2204 2205 /* 2206 * Then copy the pointers. 2207 */ 2208 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2209 ifp->if_broot_bytes); 2210 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, 2211 (int)new_size); 2212 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); 2213 } 2214 kmem_free(ifp->if_broot); 2215 ifp->if_broot = new_broot; 2216 ifp->if_broot_bytes = (int)new_size; 2217 ASSERT(ifp->if_broot_bytes <= 2218 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ(ip)); 2219 return; 2220 } 2221 2222 2223 /* 2224 * This is called when the amount of space needed for if_data 2225 * is increased or decreased. The change in size is indicated by 2226 * the number of bytes that need to be added or deleted in the 2227 * byte_diff parameter. 2228 * 2229 * If the amount of space needed has decreased below the size of the 2230 * inline buffer, then switch to using the inline buffer. Otherwise, 2231 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer 2232 * to what is needed. 2233 * 2234 * ip -- the inode whose if_data area is changing 2235 * byte_diff -- the change in the number of bytes, positive or negative, 2236 * requested for the if_data array. 2237 */ 2238 void 2239 xfs_idata_realloc( 2240 xfs_inode_t *ip, 2241 int byte_diff, 2242 int whichfork) 2243 { 2244 xfs_ifork_t *ifp; 2245 int new_size; 2246 int real_size; 2247 2248 if (byte_diff == 0) { 2249 return; 2250 } 2251 2252 ifp = XFS_IFORK_PTR(ip, whichfork); 2253 new_size = (int)ifp->if_bytes + byte_diff; 2254 ASSERT(new_size >= 0); 2255 2256 if (new_size == 0) { 2257 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2258 kmem_free(ifp->if_u1.if_data); 2259 } 2260 ifp->if_u1.if_data = NULL; 2261 real_size = 0; 2262 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { 2263 /* 2264 * If the valid extents/data can fit in if_inline_ext/data, 2265 * copy them from the malloc'd vector and free it. 2266 */ 2267 if (ifp->if_u1.if_data == NULL) { 2268 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2269 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2270 ASSERT(ifp->if_real_bytes != 0); 2271 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, 2272 new_size); 2273 kmem_free(ifp->if_u1.if_data); 2274 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2275 } 2276 real_size = 0; 2277 } else { 2278 /* 2279 * Stuck with malloc/realloc. 2280 * For inline data, the underlying buffer must be 2281 * a multiple of 4 bytes in size so that it can be 2282 * logged and stay on word boundaries. We enforce 2283 * that here. 2284 */ 2285 real_size = roundup(new_size, 4); 2286 if (ifp->if_u1.if_data == NULL) { 2287 ASSERT(ifp->if_real_bytes == 0); 2288 ifp->if_u1.if_data = kmem_alloc(real_size, 2289 KM_SLEEP | KM_NOFS); 2290 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2291 /* 2292 * Only do the realloc if the underlying size 2293 * is really changing. 2294 */ 2295 if (ifp->if_real_bytes != real_size) { 2296 ifp->if_u1.if_data = 2297 kmem_realloc(ifp->if_u1.if_data, 2298 real_size, 2299 ifp->if_real_bytes, 2300 KM_SLEEP | KM_NOFS); 2301 } 2302 } else { 2303 ASSERT(ifp->if_real_bytes == 0); 2304 ifp->if_u1.if_data = kmem_alloc(real_size, 2305 KM_SLEEP | KM_NOFS); 2306 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, 2307 ifp->if_bytes); 2308 } 2309 } 2310 ifp->if_real_bytes = real_size; 2311 ifp->if_bytes = new_size; 2312 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2313 } 2314 2315 void 2316 xfs_idestroy_fork( 2317 xfs_inode_t *ip, 2318 int whichfork) 2319 { 2320 xfs_ifork_t *ifp; 2321 2322 ifp = XFS_IFORK_PTR(ip, whichfork); 2323 if (ifp->if_broot != NULL) { 2324 kmem_free(ifp->if_broot); 2325 ifp->if_broot = NULL; 2326 } 2327 2328 /* 2329 * If the format is local, then we can't have an extents 2330 * array so just look for an inline data array. If we're 2331 * not local then we may or may not have an extents list, 2332 * so check and free it up if we do. 2333 */ 2334 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { 2335 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && 2336 (ifp->if_u1.if_data != NULL)) { 2337 ASSERT(ifp->if_real_bytes != 0); 2338 kmem_free(ifp->if_u1.if_data); 2339 ifp->if_u1.if_data = NULL; 2340 ifp->if_real_bytes = 0; 2341 } 2342 } else if ((ifp->if_flags & XFS_IFEXTENTS) && 2343 ((ifp->if_flags & XFS_IFEXTIREC) || 2344 ((ifp->if_u1.if_extents != NULL) && 2345 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { 2346 ASSERT(ifp->if_real_bytes != 0); 2347 xfs_iext_destroy(ifp); 2348 } 2349 ASSERT(ifp->if_u1.if_extents == NULL || 2350 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); 2351 ASSERT(ifp->if_real_bytes == 0); 2352 if (whichfork == XFS_ATTR_FORK) { 2353 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 2354 ip->i_afp = NULL; 2355 } 2356 } 2357 2358 /* 2359 * This is called to unpin an inode. The caller must have the inode locked 2360 * in at least shared mode so that the buffer cannot be subsequently pinned 2361 * once someone is waiting for it to be unpinned. 2362 */ 2363 static void 2364 xfs_iunpin( 2365 struct xfs_inode *ip) 2366 { 2367 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2368 2369 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2370 2371 /* Give the log a push to start the unpinning I/O */ 2372 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2373 2374 } 2375 2376 static void 2377 __xfs_iunpin_wait( 2378 struct xfs_inode *ip) 2379 { 2380 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2381 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2382 2383 xfs_iunpin(ip); 2384 2385 do { 2386 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2387 if (xfs_ipincount(ip)) 2388 io_schedule(); 2389 } while (xfs_ipincount(ip)); 2390 finish_wait(wq, &wait.wait); 2391 } 2392 2393 void 2394 xfs_iunpin_wait( 2395 struct xfs_inode *ip) 2396 { 2397 if (xfs_ipincount(ip)) 2398 __xfs_iunpin_wait(ip); 2399 } 2400 2401 /* 2402 * xfs_iextents_copy() 2403 * 2404 * This is called to copy the REAL extents (as opposed to the delayed 2405 * allocation extents) from the inode into the given buffer. It 2406 * returns the number of bytes copied into the buffer. 2407 * 2408 * If there are no delayed allocation extents, then we can just 2409 * memcpy() the extents into the buffer. Otherwise, we need to 2410 * examine each extent in turn and skip those which are delayed. 2411 */ 2412 int 2413 xfs_iextents_copy( 2414 xfs_inode_t *ip, 2415 xfs_bmbt_rec_t *dp, 2416 int whichfork) 2417 { 2418 int copied; 2419 int i; 2420 xfs_ifork_t *ifp; 2421 int nrecs; 2422 xfs_fsblock_t start_block; 2423 2424 ifp = XFS_IFORK_PTR(ip, whichfork); 2425 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2426 ASSERT(ifp->if_bytes > 0); 2427 2428 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 2429 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); 2430 ASSERT(nrecs > 0); 2431 2432 /* 2433 * There are some delayed allocation extents in the 2434 * inode, so copy the extents one at a time and skip 2435 * the delayed ones. There must be at least one 2436 * non-delayed extent. 2437 */ 2438 copied = 0; 2439 for (i = 0; i < nrecs; i++) { 2440 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 2441 start_block = xfs_bmbt_get_startblock(ep); 2442 if (isnullstartblock(start_block)) { 2443 /* 2444 * It's a delayed allocation extent, so skip it. 2445 */ 2446 continue; 2447 } 2448 2449 /* Translate to on disk format */ 2450 put_unaligned(cpu_to_be64(ep->l0), &dp->l0); 2451 put_unaligned(cpu_to_be64(ep->l1), &dp->l1); 2452 dp++; 2453 copied++; 2454 } 2455 ASSERT(copied != 0); 2456 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); 2457 2458 return (copied * (uint)sizeof(xfs_bmbt_rec_t)); 2459 } 2460 2461 /* 2462 * Each of the following cases stores data into the same region 2463 * of the on-disk inode, so only one of them can be valid at 2464 * any given time. While it is possible to have conflicting formats 2465 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is 2466 * in EXTENTS format, this can only happen when the fork has 2467 * changed formats after being modified but before being flushed. 2468 * In these cases, the format always takes precedence, because the 2469 * format indicates the current state of the fork. 2470 */ 2471 /*ARGSUSED*/ 2472 STATIC void 2473 xfs_iflush_fork( 2474 xfs_inode_t *ip, 2475 xfs_dinode_t *dip, 2476 xfs_inode_log_item_t *iip, 2477 int whichfork, 2478 xfs_buf_t *bp) 2479 { 2480 char *cp; 2481 xfs_ifork_t *ifp; 2482 xfs_mount_t *mp; 2483 static const short brootflag[2] = 2484 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; 2485 static const short dataflag[2] = 2486 { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; 2487 static const short extflag[2] = 2488 { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; 2489 2490 if (!iip) 2491 return; 2492 ifp = XFS_IFORK_PTR(ip, whichfork); 2493 /* 2494 * This can happen if we gave up in iformat in an error path, 2495 * for the attribute fork. 2496 */ 2497 if (!ifp) { 2498 ASSERT(whichfork == XFS_ATTR_FORK); 2499 return; 2500 } 2501 cp = XFS_DFORK_PTR(dip, whichfork); 2502 mp = ip->i_mount; 2503 switch (XFS_IFORK_FORMAT(ip, whichfork)) { 2504 case XFS_DINODE_FMT_LOCAL: 2505 if ((iip->ili_fields & dataflag[whichfork]) && 2506 (ifp->if_bytes > 0)) { 2507 ASSERT(ifp->if_u1.if_data != NULL); 2508 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2509 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); 2510 } 2511 break; 2512 2513 case XFS_DINODE_FMT_EXTENTS: 2514 ASSERT((ifp->if_flags & XFS_IFEXTENTS) || 2515 !(iip->ili_fields & extflag[whichfork])); 2516 if ((iip->ili_fields & extflag[whichfork]) && 2517 (ifp->if_bytes > 0)) { 2518 ASSERT(xfs_iext_get_ext(ifp, 0)); 2519 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); 2520 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, 2521 whichfork); 2522 } 2523 break; 2524 2525 case XFS_DINODE_FMT_BTREE: 2526 if ((iip->ili_fields & brootflag[whichfork]) && 2527 (ifp->if_broot_bytes > 0)) { 2528 ASSERT(ifp->if_broot != NULL); 2529 ASSERT(ifp->if_broot_bytes <= 2530 (XFS_IFORK_SIZE(ip, whichfork) + 2531 XFS_BROOT_SIZE_ADJ(ip))); 2532 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, 2533 (xfs_bmdr_block_t *)cp, 2534 XFS_DFORK_SIZE(dip, mp, whichfork)); 2535 } 2536 break; 2537 2538 case XFS_DINODE_FMT_DEV: 2539 if (iip->ili_fields & XFS_ILOG_DEV) { 2540 ASSERT(whichfork == XFS_DATA_FORK); 2541 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev); 2542 } 2543 break; 2544 2545 case XFS_DINODE_FMT_UUID: 2546 if (iip->ili_fields & XFS_ILOG_UUID) { 2547 ASSERT(whichfork == XFS_DATA_FORK); 2548 memcpy(XFS_DFORK_DPTR(dip), 2549 &ip->i_df.if_u2.if_uuid, 2550 sizeof(uuid_t)); 2551 } 2552 break; 2553 2554 default: 2555 ASSERT(0); 2556 break; 2557 } 2558 } 2559 2560 STATIC int 2561 xfs_iflush_cluster( 2562 xfs_inode_t *ip, 2563 xfs_buf_t *bp) 2564 { 2565 xfs_mount_t *mp = ip->i_mount; 2566 struct xfs_perag *pag; 2567 unsigned long first_index, mask; 2568 unsigned long inodes_per_cluster; 2569 int ilist_size; 2570 xfs_inode_t **ilist; 2571 xfs_inode_t *iq; 2572 int nr_found; 2573 int clcount = 0; 2574 int bufwasdelwri; 2575 int i; 2576 2577 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 2578 2579 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; 2580 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 2581 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); 2582 if (!ilist) 2583 goto out_put; 2584 2585 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); 2586 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 2587 rcu_read_lock(); 2588 /* really need a gang lookup range call here */ 2589 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, 2590 first_index, inodes_per_cluster); 2591 if (nr_found == 0) 2592 goto out_free; 2593 2594 for (i = 0; i < nr_found; i++) { 2595 iq = ilist[i]; 2596 if (iq == ip) 2597 continue; 2598 2599 /* 2600 * because this is an RCU protected lookup, we could find a 2601 * recently freed or even reallocated inode during the lookup. 2602 * We need to check under the i_flags_lock for a valid inode 2603 * here. Skip it if it is not valid or the wrong inode. 2604 */ 2605 spin_lock(&ip->i_flags_lock); 2606 if (!ip->i_ino || 2607 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { 2608 spin_unlock(&ip->i_flags_lock); 2609 continue; 2610 } 2611 spin_unlock(&ip->i_flags_lock); 2612 2613 /* 2614 * Do an un-protected check to see if the inode is dirty and 2615 * is a candidate for flushing. These checks will be repeated 2616 * later after the appropriate locks are acquired. 2617 */ 2618 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) 2619 continue; 2620 2621 /* 2622 * Try to get locks. If any are unavailable or it is pinned, 2623 * then this inode cannot be flushed and is skipped. 2624 */ 2625 2626 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) 2627 continue; 2628 if (!xfs_iflock_nowait(iq)) { 2629 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2630 continue; 2631 } 2632 if (xfs_ipincount(iq)) { 2633 xfs_ifunlock(iq); 2634 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2635 continue; 2636 } 2637 2638 /* 2639 * arriving here means that this inode can be flushed. First 2640 * re-check that it's dirty before flushing. 2641 */ 2642 if (!xfs_inode_clean(iq)) { 2643 int error; 2644 error = xfs_iflush_int(iq, bp); 2645 if (error) { 2646 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2647 goto cluster_corrupt_out; 2648 } 2649 clcount++; 2650 } else { 2651 xfs_ifunlock(iq); 2652 } 2653 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2654 } 2655 2656 if (clcount) { 2657 XFS_STATS_INC(xs_icluster_flushcnt); 2658 XFS_STATS_ADD(xs_icluster_flushinode, clcount); 2659 } 2660 2661 out_free: 2662 rcu_read_unlock(); 2663 kmem_free(ilist); 2664 out_put: 2665 xfs_perag_put(pag); 2666 return 0; 2667 2668 2669 cluster_corrupt_out: 2670 /* 2671 * Corruption detected in the clustering loop. Invalidate the 2672 * inode buffer and shut down the filesystem. 2673 */ 2674 rcu_read_unlock(); 2675 /* 2676 * Clean up the buffer. If it was delwri, just release it -- 2677 * brelse can handle it with no problems. If not, shut down the 2678 * filesystem before releasing the buffer. 2679 */ 2680 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 2681 if (bufwasdelwri) 2682 xfs_buf_relse(bp); 2683 2684 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 2685 2686 if (!bufwasdelwri) { 2687 /* 2688 * Just like incore_relse: if we have b_iodone functions, 2689 * mark the buffer as an error and call them. Otherwise 2690 * mark it as stale and brelse. 2691 */ 2692 if (bp->b_iodone) { 2693 XFS_BUF_UNDONE(bp); 2694 xfs_buf_stale(bp); 2695 xfs_buf_ioerror(bp, EIO); 2696 xfs_buf_ioend(bp, 0); 2697 } else { 2698 xfs_buf_stale(bp); 2699 xfs_buf_relse(bp); 2700 } 2701 } 2702 2703 /* 2704 * Unlocks the flush lock 2705 */ 2706 xfs_iflush_abort(iq, false); 2707 kmem_free(ilist); 2708 xfs_perag_put(pag); 2709 return XFS_ERROR(EFSCORRUPTED); 2710 } 2711 2712 /* 2713 * Flush dirty inode metadata into the backing buffer. 2714 * 2715 * The caller must have the inode lock and the inode flush lock held. The 2716 * inode lock will still be held upon return to the caller, and the inode 2717 * flush lock will be released after the inode has reached the disk. 2718 * 2719 * The caller must write out the buffer returned in *bpp and release it. 2720 */ 2721 int 2722 xfs_iflush( 2723 struct xfs_inode *ip, 2724 struct xfs_buf **bpp) 2725 { 2726 struct xfs_mount *mp = ip->i_mount; 2727 struct xfs_buf *bp; 2728 struct xfs_dinode *dip; 2729 int error; 2730 2731 XFS_STATS_INC(xs_iflush_count); 2732 2733 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2734 ASSERT(xfs_isiflocked(ip)); 2735 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 2736 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 2737 2738 *bpp = NULL; 2739 2740 xfs_iunpin_wait(ip); 2741 2742 /* 2743 * For stale inodes we cannot rely on the backing buffer remaining 2744 * stale in cache for the remaining life of the stale inode and so 2745 * xfs_imap_to_bp() below may give us a buffer that no longer contains 2746 * inodes below. We have to check this after ensuring the inode is 2747 * unpinned so that it is safe to reclaim the stale inode after the 2748 * flush call. 2749 */ 2750 if (xfs_iflags_test(ip, XFS_ISTALE)) { 2751 xfs_ifunlock(ip); 2752 return 0; 2753 } 2754 2755 /* 2756 * This may have been unpinned because the filesystem is shutting 2757 * down forcibly. If that's the case we must not write this inode 2758 * to disk, because the log record didn't make it to disk. 2759 * 2760 * We also have to remove the log item from the AIL in this case, 2761 * as we wait for an empty AIL as part of the unmount process. 2762 */ 2763 if (XFS_FORCED_SHUTDOWN(mp)) { 2764 error = XFS_ERROR(EIO); 2765 goto abort_out; 2766 } 2767 2768 /* 2769 * Get the buffer containing the on-disk inode. 2770 */ 2771 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 2772 0); 2773 if (error || !bp) { 2774 xfs_ifunlock(ip); 2775 return error; 2776 } 2777 2778 /* 2779 * First flush out the inode that xfs_iflush was called with. 2780 */ 2781 error = xfs_iflush_int(ip, bp); 2782 if (error) 2783 goto corrupt_out; 2784 2785 /* 2786 * If the buffer is pinned then push on the log now so we won't 2787 * get stuck waiting in the write for too long. 2788 */ 2789 if (xfs_buf_ispinned(bp)) 2790 xfs_log_force(mp, 0); 2791 2792 /* 2793 * inode clustering: 2794 * see if other inodes can be gathered into this write 2795 */ 2796 error = xfs_iflush_cluster(ip, bp); 2797 if (error) 2798 goto cluster_corrupt_out; 2799 2800 *bpp = bp; 2801 return 0; 2802 2803 corrupt_out: 2804 xfs_buf_relse(bp); 2805 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 2806 cluster_corrupt_out: 2807 error = XFS_ERROR(EFSCORRUPTED); 2808 abort_out: 2809 /* 2810 * Unlocks the flush lock 2811 */ 2812 xfs_iflush_abort(ip, false); 2813 return error; 2814 } 2815 2816 2817 STATIC int 2818 xfs_iflush_int( 2819 struct xfs_inode *ip, 2820 struct xfs_buf *bp) 2821 { 2822 struct xfs_inode_log_item *iip = ip->i_itemp; 2823 struct xfs_dinode *dip; 2824 struct xfs_mount *mp = ip->i_mount; 2825 2826 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2827 ASSERT(xfs_isiflocked(ip)); 2828 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 2829 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 2830 ASSERT(iip != NULL && iip->ili_fields != 0); 2831 2832 /* set *dip = inode's place in the buffer */ 2833 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); 2834 2835 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 2836 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 2837 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2838 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 2839 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 2840 goto corrupt_out; 2841 } 2842 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, 2843 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { 2844 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2845 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", 2846 __func__, ip->i_ino, ip, ip->i_d.di_magic); 2847 goto corrupt_out; 2848 } 2849 if (S_ISREG(ip->i_d.di_mode)) { 2850 if (XFS_TEST_ERROR( 2851 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 2852 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 2853 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 2854 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2855 "%s: Bad regular inode %Lu, ptr 0x%p", 2856 __func__, ip->i_ino, ip); 2857 goto corrupt_out; 2858 } 2859 } else if (S_ISDIR(ip->i_d.di_mode)) { 2860 if (XFS_TEST_ERROR( 2861 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 2862 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 2863 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 2864 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 2865 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2866 "%s: Bad directory inode %Lu, ptr 0x%p", 2867 __func__, ip->i_ino, ip); 2868 goto corrupt_out; 2869 } 2870 } 2871 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 2872 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 2873 XFS_RANDOM_IFLUSH_5)) { 2874 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2875 "%s: detected corrupt incore inode %Lu, " 2876 "total extents = %d, nblocks = %Ld, ptr 0x%p", 2877 __func__, ip->i_ino, 2878 ip->i_d.di_nextents + ip->i_d.di_anextents, 2879 ip->i_d.di_nblocks, ip); 2880 goto corrupt_out; 2881 } 2882 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 2883 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 2884 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2885 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 2886 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 2887 goto corrupt_out; 2888 } 2889 /* 2890 * bump the flush iteration count, used to detect flushes which 2891 * postdate a log record during recovery. This is redundant as we now 2892 * log every change and hence this can't happen. Still, it doesn't hurt. 2893 */ 2894 ip->i_d.di_flushiter++; 2895 2896 /* 2897 * Copy the dirty parts of the inode into the on-disk 2898 * inode. We always copy out the core of the inode, 2899 * because if the inode is dirty at all the core must 2900 * be. 2901 */ 2902 xfs_dinode_to_disk(dip, &ip->i_d); 2903 2904 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 2905 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 2906 ip->i_d.di_flushiter = 0; 2907 2908 /* 2909 * If this is really an old format inode and the superblock version 2910 * has not been updated to support only new format inodes, then 2911 * convert back to the old inode format. If the superblock version 2912 * has been updated, then make the conversion permanent. 2913 */ 2914 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); 2915 if (ip->i_d.di_version == 1) { 2916 if (!xfs_sb_version_hasnlink(&mp->m_sb)) { 2917 /* 2918 * Convert it back. 2919 */ 2920 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); 2921 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); 2922 } else { 2923 /* 2924 * The superblock version has already been bumped, 2925 * so just make the conversion to the new inode 2926 * format permanent. 2927 */ 2928 ip->i_d.di_version = 2; 2929 dip->di_version = 2; 2930 ip->i_d.di_onlink = 0; 2931 dip->di_onlink = 0; 2932 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 2933 memset(&(dip->di_pad[0]), 0, 2934 sizeof(dip->di_pad)); 2935 ASSERT(xfs_get_projid(ip) == 0); 2936 } 2937 } 2938 2939 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); 2940 if (XFS_IFORK_Q(ip)) 2941 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); 2942 xfs_inobp_check(mp, bp); 2943 2944 /* 2945 * We've recorded everything logged in the inode, so we'd like to clear 2946 * the ili_fields bits so we don't log and flush things unnecessarily. 2947 * However, we can't stop logging all this information until the data 2948 * we've copied into the disk buffer is written to disk. If we did we 2949 * might overwrite the copy of the inode in the log with all the data 2950 * after re-logging only part of it, and in the face of a crash we 2951 * wouldn't have all the data we need to recover. 2952 * 2953 * What we do is move the bits to the ili_last_fields field. When 2954 * logging the inode, these bits are moved back to the ili_fields field. 2955 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 2956 * know that the information those bits represent is permanently on 2957 * disk. As long as the flush completes before the inode is logged 2958 * again, then both ili_fields and ili_last_fields will be cleared. 2959 * 2960 * We can play with the ili_fields bits here, because the inode lock 2961 * must be held exclusively in order to set bits there and the flush 2962 * lock protects the ili_last_fields bits. Set ili_logged so the flush 2963 * done routine can tell whether or not to look in the AIL. Also, store 2964 * the current LSN of the inode so that we can tell whether the item has 2965 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 2966 * need the AIL lock, because it is a 64 bit value that cannot be read 2967 * atomically. 2968 */ 2969 iip->ili_last_fields = iip->ili_fields; 2970 iip->ili_fields = 0; 2971 iip->ili_logged = 1; 2972 2973 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2974 &iip->ili_item.li_lsn); 2975 2976 /* 2977 * Attach the function xfs_iflush_done to the inode's 2978 * buffer. This will remove the inode from the AIL 2979 * and unlock the inode's flush lock when the inode is 2980 * completely written to disk. 2981 */ 2982 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 2983 2984 /* update the lsn in the on disk inode if required */ 2985 if (ip->i_d.di_version == 3) 2986 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); 2987 2988 /* generate the checksum. */ 2989 xfs_dinode_calc_crc(mp, dip); 2990 2991 ASSERT(bp->b_fspriv != NULL); 2992 ASSERT(bp->b_iodone != NULL); 2993 return 0; 2994 2995 corrupt_out: 2996 return XFS_ERROR(EFSCORRUPTED); 2997 } 2998 2999 /* 3000 * Return a pointer to the extent record at file index idx. 3001 */ 3002 xfs_bmbt_rec_host_t * 3003 xfs_iext_get_ext( 3004 xfs_ifork_t *ifp, /* inode fork pointer */ 3005 xfs_extnum_t idx) /* index of target extent */ 3006 { 3007 ASSERT(idx >= 0); 3008 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); 3009 3010 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { 3011 return ifp->if_u1.if_ext_irec->er_extbuf; 3012 } else if (ifp->if_flags & XFS_IFEXTIREC) { 3013 xfs_ext_irec_t *erp; /* irec pointer */ 3014 int erp_idx = 0; /* irec index */ 3015 xfs_extnum_t page_idx = idx; /* ext index in target list */ 3016 3017 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 3018 return &erp->er_extbuf[page_idx]; 3019 } else if (ifp->if_bytes) { 3020 return &ifp->if_u1.if_extents[idx]; 3021 } else { 3022 return NULL; 3023 } 3024 } 3025 3026 /* 3027 * Insert new item(s) into the extent records for incore inode 3028 * fork 'ifp'. 'count' new items are inserted at index 'idx'. 3029 */ 3030 void 3031 xfs_iext_insert( 3032 xfs_inode_t *ip, /* incore inode pointer */ 3033 xfs_extnum_t idx, /* starting index of new items */ 3034 xfs_extnum_t count, /* number of inserted items */ 3035 xfs_bmbt_irec_t *new, /* items to insert */ 3036 int state) /* type of extent conversion */ 3037 { 3038 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; 3039 xfs_extnum_t i; /* extent record index */ 3040 3041 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_); 3042 3043 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 3044 xfs_iext_add(ifp, idx, count); 3045 for (i = idx; i < idx + count; i++, new++) 3046 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); 3047 } 3048 3049 /* 3050 * This is called when the amount of space required for incore file 3051 * extents needs to be increased. The ext_diff parameter stores the 3052 * number of new extents being added and the idx parameter contains 3053 * the extent index where the new extents will be added. If the new 3054 * extents are being appended, then we just need to (re)allocate and 3055 * initialize the space. Otherwise, if the new extents are being 3056 * inserted into the middle of the existing entries, a bit more work 3057 * is required to make room for the new extents to be inserted. The 3058 * caller is responsible for filling in the new extent entries upon 3059 * return. 3060 */ 3061 void 3062 xfs_iext_add( 3063 xfs_ifork_t *ifp, /* inode fork pointer */ 3064 xfs_extnum_t idx, /* index to begin adding exts */ 3065 int ext_diff) /* number of extents to add */ 3066 { 3067 int byte_diff; /* new bytes being added */ 3068 int new_size; /* size of extents after adding */ 3069 xfs_extnum_t nextents; /* number of extents in file */ 3070 3071 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3072 ASSERT((idx >= 0) && (idx <= nextents)); 3073 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); 3074 new_size = ifp->if_bytes + byte_diff; 3075 /* 3076 * If the new number of extents (nextents + ext_diff) 3077 * fits inside the inode, then continue to use the inline 3078 * extent buffer. 3079 */ 3080 if (nextents + ext_diff <= XFS_INLINE_EXTS) { 3081 if (idx < nextents) { 3082 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], 3083 &ifp->if_u2.if_inline_ext[idx], 3084 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 3085 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); 3086 } 3087 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 3088 ifp->if_real_bytes = 0; 3089 } 3090 /* 3091 * Otherwise use a linear (direct) extent list. 3092 * If the extents are currently inside the inode, 3093 * xfs_iext_realloc_direct will switch us from 3094 * inline to direct extent allocation mode. 3095 */ 3096 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { 3097 xfs_iext_realloc_direct(ifp, new_size); 3098 if (idx < nextents) { 3099 memmove(&ifp->if_u1.if_extents[idx + ext_diff], 3100 &ifp->if_u1.if_extents[idx], 3101 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 3102 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); 3103 } 3104 } 3105 /* Indirection array */ 3106 else { 3107 xfs_ext_irec_t *erp; 3108 int erp_idx = 0; 3109 int page_idx = idx; 3110 3111 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); 3112 if (ifp->if_flags & XFS_IFEXTIREC) { 3113 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); 3114 } else { 3115 xfs_iext_irec_init(ifp); 3116 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3117 erp = ifp->if_u1.if_ext_irec; 3118 } 3119 /* Extents fit in target extent page */ 3120 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { 3121 if (page_idx < erp->er_extcount) { 3122 memmove(&erp->er_extbuf[page_idx + ext_diff], 3123 &erp->er_extbuf[page_idx], 3124 (erp->er_extcount - page_idx) * 3125 sizeof(xfs_bmbt_rec_t)); 3126 memset(&erp->er_extbuf[page_idx], 0, byte_diff); 3127 } 3128 erp->er_extcount += ext_diff; 3129 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3130 } 3131 /* Insert a new extent page */ 3132 else if (erp) { 3133 xfs_iext_add_indirect_multi(ifp, 3134 erp_idx, page_idx, ext_diff); 3135 } 3136 /* 3137 * If extent(s) are being appended to the last page in 3138 * the indirection array and the new extent(s) don't fit 3139 * in the page, then erp is NULL and erp_idx is set to 3140 * the next index needed in the indirection array. 3141 */ 3142 else { 3143 int count = ext_diff; 3144 3145 while (count) { 3146 erp = xfs_iext_irec_new(ifp, erp_idx); 3147 erp->er_extcount = count; 3148 count -= MIN(count, (int)XFS_LINEAR_EXTS); 3149 if (count) { 3150 erp_idx++; 3151 } 3152 } 3153 } 3154 } 3155 ifp->if_bytes = new_size; 3156 } 3157 3158 /* 3159 * This is called when incore extents are being added to the indirection 3160 * array and the new extents do not fit in the target extent list. The 3161 * erp_idx parameter contains the irec index for the target extent list 3162 * in the indirection array, and the idx parameter contains the extent 3163 * index within the list. The number of extents being added is stored 3164 * in the count parameter. 3165 * 3166 * |-------| |-------| 3167 * | | | | idx - number of extents before idx 3168 * | idx | | count | 3169 * | | | | count - number of extents being inserted at idx 3170 * |-------| |-------| 3171 * | count | | nex2 | nex2 - number of extents after idx + count 3172 * |-------| |-------| 3173 */ 3174 void 3175 xfs_iext_add_indirect_multi( 3176 xfs_ifork_t *ifp, /* inode fork pointer */ 3177 int erp_idx, /* target extent irec index */ 3178 xfs_extnum_t idx, /* index within target list */ 3179 int count) /* new extents being added */ 3180 { 3181 int byte_diff; /* new bytes being added */ 3182 xfs_ext_irec_t *erp; /* pointer to irec entry */ 3183 xfs_extnum_t ext_diff; /* number of extents to add */ 3184 xfs_extnum_t ext_cnt; /* new extents still needed */ 3185 xfs_extnum_t nex2; /* extents after idx + count */ 3186 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ 3187 int nlists; /* number of irec's (lists) */ 3188 3189 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3190 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3191 nex2 = erp->er_extcount - idx; 3192 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3193 3194 /* 3195 * Save second part of target extent list 3196 * (all extents past */ 3197 if (nex2) { 3198 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3199 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); 3200 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); 3201 erp->er_extcount -= nex2; 3202 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); 3203 memset(&erp->er_extbuf[idx], 0, byte_diff); 3204 } 3205 3206 /* 3207 * Add the new extents to the end of the target 3208 * list, then allocate new irec record(s) and 3209 * extent buffer(s) as needed to store the rest 3210 * of the new extents. 3211 */ 3212 ext_cnt = count; 3213 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); 3214 if (ext_diff) { 3215 erp->er_extcount += ext_diff; 3216 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3217 ext_cnt -= ext_diff; 3218 } 3219 while (ext_cnt) { 3220 erp_idx++; 3221 erp = xfs_iext_irec_new(ifp, erp_idx); 3222 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); 3223 erp->er_extcount = ext_diff; 3224 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3225 ext_cnt -= ext_diff; 3226 } 3227 3228 /* Add nex2 extents back to indirection array */ 3229 if (nex2) { 3230 xfs_extnum_t ext_avail; 3231 int i; 3232 3233 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3234 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; 3235 i = 0; 3236 /* 3237 * If nex2 extents fit in the current page, append 3238 * nex2_ep after the new extents. 3239 */ 3240 if (nex2 <= ext_avail) { 3241 i = erp->er_extcount; 3242 } 3243 /* 3244 * Otherwise, check if space is available in the 3245 * next page. 3246 */ 3247 else if ((erp_idx < nlists - 1) && 3248 (nex2 <= (ext_avail = XFS_LINEAR_EXTS - 3249 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { 3250 erp_idx++; 3251 erp++; 3252 /* Create a hole for nex2 extents */ 3253 memmove(&erp->er_extbuf[nex2], erp->er_extbuf, 3254 erp->er_extcount * sizeof(xfs_bmbt_rec_t)); 3255 } 3256 /* 3257 * Final choice, create a new extent page for 3258 * nex2 extents. 3259 */ 3260 else { 3261 erp_idx++; 3262 erp = xfs_iext_irec_new(ifp, erp_idx); 3263 } 3264 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); 3265 kmem_free(nex2_ep); 3266 erp->er_extcount += nex2; 3267 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); 3268 } 3269 } 3270 3271 /* 3272 * This is called when the amount of space required for incore file 3273 * extents needs to be decreased. The ext_diff parameter stores the 3274 * number of extents to be removed and the idx parameter contains 3275 * the extent index where the extents will be removed from. 3276 * 3277 * If the amount of space needed has decreased below the linear 3278 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous 3279 * extent array. Otherwise, use kmem_realloc() to adjust the 3280 * size to what is needed. 3281 */ 3282 void 3283 xfs_iext_remove( 3284 xfs_inode_t *ip, /* incore inode pointer */ 3285 xfs_extnum_t idx, /* index to begin removing exts */ 3286 int ext_diff, /* number of extents to remove */ 3287 int state) /* type of extent conversion */ 3288 { 3289 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; 3290 xfs_extnum_t nextents; /* number of extents in file */ 3291 int new_size; /* size of extents after removal */ 3292 3293 trace_xfs_iext_remove(ip, idx, state, _RET_IP_); 3294 3295 ASSERT(ext_diff > 0); 3296 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3297 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); 3298 3299 if (new_size == 0) { 3300 xfs_iext_destroy(ifp); 3301 } else if (ifp->if_flags & XFS_IFEXTIREC) { 3302 xfs_iext_remove_indirect(ifp, idx, ext_diff); 3303 } else if (ifp->if_real_bytes) { 3304 xfs_iext_remove_direct(ifp, idx, ext_diff); 3305 } else { 3306 xfs_iext_remove_inline(ifp, idx, ext_diff); 3307 } 3308 ifp->if_bytes = new_size; 3309 } 3310 3311 /* 3312 * This removes ext_diff extents from the inline buffer, beginning 3313 * at extent index idx. 3314 */ 3315 void 3316 xfs_iext_remove_inline( 3317 xfs_ifork_t *ifp, /* inode fork pointer */ 3318 xfs_extnum_t idx, /* index to begin removing exts */ 3319 int ext_diff) /* number of extents to remove */ 3320 { 3321 int nextents; /* number of extents in file */ 3322 3323 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3324 ASSERT(idx < XFS_INLINE_EXTS); 3325 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3326 ASSERT(((nextents - ext_diff) > 0) && 3327 (nextents - ext_diff) < XFS_INLINE_EXTS); 3328 3329 if (idx + ext_diff < nextents) { 3330 memmove(&ifp->if_u2.if_inline_ext[idx], 3331 &ifp->if_u2.if_inline_ext[idx + ext_diff], 3332 (nextents - (idx + ext_diff)) * 3333 sizeof(xfs_bmbt_rec_t)); 3334 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], 3335 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 3336 } else { 3337 memset(&ifp->if_u2.if_inline_ext[idx], 0, 3338 ext_diff * sizeof(xfs_bmbt_rec_t)); 3339 } 3340 } 3341 3342 /* 3343 * This removes ext_diff extents from a linear (direct) extent list, 3344 * beginning at extent index idx. If the extents are being removed 3345 * from the end of the list (ie. truncate) then we just need to re- 3346 * allocate the list to remove the extra space. Otherwise, if the 3347 * extents are being removed from the middle of the existing extent 3348 * entries, then we first need to move the extent records beginning 3349 * at idx + ext_diff up in the list to overwrite the records being 3350 * removed, then remove the extra space via kmem_realloc. 3351 */ 3352 void 3353 xfs_iext_remove_direct( 3354 xfs_ifork_t *ifp, /* inode fork pointer */ 3355 xfs_extnum_t idx, /* index to begin removing exts */ 3356 int ext_diff) /* number of extents to remove */ 3357 { 3358 xfs_extnum_t nextents; /* number of extents in file */ 3359 int new_size; /* size of extents after removal */ 3360 3361 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3362 new_size = ifp->if_bytes - 3363 (ext_diff * sizeof(xfs_bmbt_rec_t)); 3364 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3365 3366 if (new_size == 0) { 3367 xfs_iext_destroy(ifp); 3368 return; 3369 } 3370 /* Move extents up in the list (if needed) */ 3371 if (idx + ext_diff < nextents) { 3372 memmove(&ifp->if_u1.if_extents[idx], 3373 &ifp->if_u1.if_extents[idx + ext_diff], 3374 (nextents - (idx + ext_diff)) * 3375 sizeof(xfs_bmbt_rec_t)); 3376 } 3377 memset(&ifp->if_u1.if_extents[nextents - ext_diff], 3378 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 3379 /* 3380 * Reallocate the direct extent list. If the extents 3381 * will fit inside the inode then xfs_iext_realloc_direct 3382 * will switch from direct to inline extent allocation 3383 * mode for us. 3384 */ 3385 xfs_iext_realloc_direct(ifp, new_size); 3386 ifp->if_bytes = new_size; 3387 } 3388 3389 /* 3390 * This is called when incore extents are being removed from the 3391 * indirection array and the extents being removed span multiple extent 3392 * buffers. The idx parameter contains the file extent index where we 3393 * want to begin removing extents, and the count parameter contains 3394 * how many extents need to be removed. 3395 * 3396 * |-------| |-------| 3397 * | nex1 | | | nex1 - number of extents before idx 3398 * |-------| | count | 3399 * | | | | count - number of extents being removed at idx 3400 * | count | |-------| 3401 * | | | nex2 | nex2 - number of extents after idx + count 3402 * |-------| |-------| 3403 */ 3404 void 3405 xfs_iext_remove_indirect( 3406 xfs_ifork_t *ifp, /* inode fork pointer */ 3407 xfs_extnum_t idx, /* index to begin removing extents */ 3408 int count) /* number of extents to remove */ 3409 { 3410 xfs_ext_irec_t *erp; /* indirection array pointer */ 3411 int erp_idx = 0; /* indirection array index */ 3412 xfs_extnum_t ext_cnt; /* extents left to remove */ 3413 xfs_extnum_t ext_diff; /* extents to remove in current list */ 3414 xfs_extnum_t nex1; /* number of extents before idx */ 3415 xfs_extnum_t nex2; /* extents after idx + count */ 3416 int page_idx = idx; /* index in target extent list */ 3417 3418 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3419 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 3420 ASSERT(erp != NULL); 3421 nex1 = page_idx; 3422 ext_cnt = count; 3423 while (ext_cnt) { 3424 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); 3425 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); 3426 /* 3427 * Check for deletion of entire list; 3428 * xfs_iext_irec_remove() updates extent offsets. 3429 */ 3430 if (ext_diff == erp->er_extcount) { 3431 xfs_iext_irec_remove(ifp, erp_idx); 3432 ext_cnt -= ext_diff; 3433 nex1 = 0; 3434 if (ext_cnt) { 3435 ASSERT(erp_idx < ifp->if_real_bytes / 3436 XFS_IEXT_BUFSZ); 3437 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3438 nex1 = 0; 3439 continue; 3440 } else { 3441 break; 3442 } 3443 } 3444 /* Move extents up (if needed) */ 3445 if (nex2) { 3446 memmove(&erp->er_extbuf[nex1], 3447 &erp->er_extbuf[nex1 + ext_diff], 3448 nex2 * sizeof(xfs_bmbt_rec_t)); 3449 } 3450 /* Zero out rest of page */ 3451 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - 3452 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); 3453 /* Update remaining counters */ 3454 erp->er_extcount -= ext_diff; 3455 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); 3456 ext_cnt -= ext_diff; 3457 nex1 = 0; 3458 erp_idx++; 3459 erp++; 3460 } 3461 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); 3462 xfs_iext_irec_compact(ifp); 3463 } 3464 3465 /* 3466 * Create, destroy, or resize a linear (direct) block of extents. 3467 */ 3468 void 3469 xfs_iext_realloc_direct( 3470 xfs_ifork_t *ifp, /* inode fork pointer */ 3471 int new_size) /* new size of extents */ 3472 { 3473 int rnew_size; /* real new size of extents */ 3474 3475 rnew_size = new_size; 3476 3477 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || 3478 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && 3479 (new_size != ifp->if_real_bytes))); 3480 3481 /* Free extent records */ 3482 if (new_size == 0) { 3483 xfs_iext_destroy(ifp); 3484 } 3485 /* Resize direct extent list and zero any new bytes */ 3486 else if (ifp->if_real_bytes) { 3487 /* Check if extents will fit inside the inode */ 3488 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { 3489 xfs_iext_direct_to_inline(ifp, new_size / 3490 (uint)sizeof(xfs_bmbt_rec_t)); 3491 ifp->if_bytes = new_size; 3492 return; 3493 } 3494 if (!is_power_of_2(new_size)){ 3495 rnew_size = roundup_pow_of_two(new_size); 3496 } 3497 if (rnew_size != ifp->if_real_bytes) { 3498 ifp->if_u1.if_extents = 3499 kmem_realloc(ifp->if_u1.if_extents, 3500 rnew_size, 3501 ifp->if_real_bytes, KM_NOFS); 3502 } 3503 if (rnew_size > ifp->if_real_bytes) { 3504 memset(&ifp->if_u1.if_extents[ifp->if_bytes / 3505 (uint)sizeof(xfs_bmbt_rec_t)], 0, 3506 rnew_size - ifp->if_real_bytes); 3507 } 3508 } 3509 /* 3510 * Switch from the inline extent buffer to a direct 3511 * extent list. Be sure to include the inline extent 3512 * bytes in new_size. 3513 */ 3514 else { 3515 new_size += ifp->if_bytes; 3516 if (!is_power_of_2(new_size)) { 3517 rnew_size = roundup_pow_of_two(new_size); 3518 } 3519 xfs_iext_inline_to_direct(ifp, rnew_size); 3520 } 3521 ifp->if_real_bytes = rnew_size; 3522 ifp->if_bytes = new_size; 3523 } 3524 3525 /* 3526 * Switch from linear (direct) extent records to inline buffer. 3527 */ 3528 void 3529 xfs_iext_direct_to_inline( 3530 xfs_ifork_t *ifp, /* inode fork pointer */ 3531 xfs_extnum_t nextents) /* number of extents in file */ 3532 { 3533 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 3534 ASSERT(nextents <= XFS_INLINE_EXTS); 3535 /* 3536 * The inline buffer was zeroed when we switched 3537 * from inline to direct extent allocation mode, 3538 * so we don't need to clear it here. 3539 */ 3540 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, 3541 nextents * sizeof(xfs_bmbt_rec_t)); 3542 kmem_free(ifp->if_u1.if_extents); 3543 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 3544 ifp->if_real_bytes = 0; 3545 } 3546 3547 /* 3548 * Switch from inline buffer to linear (direct) extent records. 3549 * new_size should already be rounded up to the next power of 2 3550 * by the caller (when appropriate), so use new_size as it is. 3551 * However, since new_size may be rounded up, we can't update 3552 * if_bytes here. It is the caller's responsibility to update 3553 * if_bytes upon return. 3554 */ 3555 void 3556 xfs_iext_inline_to_direct( 3557 xfs_ifork_t *ifp, /* inode fork pointer */ 3558 int new_size) /* number of extents in file */ 3559 { 3560 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); 3561 memset(ifp->if_u1.if_extents, 0, new_size); 3562 if (ifp->if_bytes) { 3563 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, 3564 ifp->if_bytes); 3565 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 3566 sizeof(xfs_bmbt_rec_t)); 3567 } 3568 ifp->if_real_bytes = new_size; 3569 } 3570 3571 /* 3572 * Resize an extent indirection array to new_size bytes. 3573 */ 3574 STATIC void 3575 xfs_iext_realloc_indirect( 3576 xfs_ifork_t *ifp, /* inode fork pointer */ 3577 int new_size) /* new indirection array size */ 3578 { 3579 int nlists; /* number of irec's (ex lists) */ 3580 int size; /* current indirection array size */ 3581 3582 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3583 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3584 size = nlists * sizeof(xfs_ext_irec_t); 3585 ASSERT(ifp->if_real_bytes); 3586 ASSERT((new_size >= 0) && (new_size != size)); 3587 if (new_size == 0) { 3588 xfs_iext_destroy(ifp); 3589 } else { 3590 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) 3591 kmem_realloc(ifp->if_u1.if_ext_irec, 3592 new_size, size, KM_NOFS); 3593 } 3594 } 3595 3596 /* 3597 * Switch from indirection array to linear (direct) extent allocations. 3598 */ 3599 STATIC void 3600 xfs_iext_indirect_to_direct( 3601 xfs_ifork_t *ifp) /* inode fork pointer */ 3602 { 3603 xfs_bmbt_rec_host_t *ep; /* extent record pointer */ 3604 xfs_extnum_t nextents; /* number of extents in file */ 3605 int size; /* size of file extents */ 3606 3607 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3608 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3609 ASSERT(nextents <= XFS_LINEAR_EXTS); 3610 size = nextents * sizeof(xfs_bmbt_rec_t); 3611 3612 xfs_iext_irec_compact_pages(ifp); 3613 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); 3614 3615 ep = ifp->if_u1.if_ext_irec->er_extbuf; 3616 kmem_free(ifp->if_u1.if_ext_irec); 3617 ifp->if_flags &= ~XFS_IFEXTIREC; 3618 ifp->if_u1.if_extents = ep; 3619 ifp->if_bytes = size; 3620 if (nextents < XFS_LINEAR_EXTS) { 3621 xfs_iext_realloc_direct(ifp, size); 3622 } 3623 } 3624 3625 /* 3626 * Free incore file extents. 3627 */ 3628 void 3629 xfs_iext_destroy( 3630 xfs_ifork_t *ifp) /* inode fork pointer */ 3631 { 3632 if (ifp->if_flags & XFS_IFEXTIREC) { 3633 int erp_idx; 3634 int nlists; 3635 3636 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3637 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { 3638 xfs_iext_irec_remove(ifp, erp_idx); 3639 } 3640 ifp->if_flags &= ~XFS_IFEXTIREC; 3641 } else if (ifp->if_real_bytes) { 3642 kmem_free(ifp->if_u1.if_extents); 3643 } else if (ifp->if_bytes) { 3644 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 3645 sizeof(xfs_bmbt_rec_t)); 3646 } 3647 ifp->if_u1.if_extents = NULL; 3648 ifp->if_real_bytes = 0; 3649 ifp->if_bytes = 0; 3650 } 3651 3652 /* 3653 * Return a pointer to the extent record for file system block bno. 3654 */ 3655 xfs_bmbt_rec_host_t * /* pointer to found extent record */ 3656 xfs_iext_bno_to_ext( 3657 xfs_ifork_t *ifp, /* inode fork pointer */ 3658 xfs_fileoff_t bno, /* block number to search for */ 3659 xfs_extnum_t *idxp) /* index of target extent */ 3660 { 3661 xfs_bmbt_rec_host_t *base; /* pointer to first extent */ 3662 xfs_filblks_t blockcount = 0; /* number of blocks in extent */ 3663 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ 3664 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 3665 int high; /* upper boundary in search */ 3666 xfs_extnum_t idx = 0; /* index of target extent */ 3667 int low; /* lower boundary in search */ 3668 xfs_extnum_t nextents; /* number of file extents */ 3669 xfs_fileoff_t startoff = 0; /* start offset of extent */ 3670 3671 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3672 if (nextents == 0) { 3673 *idxp = 0; 3674 return NULL; 3675 } 3676 low = 0; 3677 if (ifp->if_flags & XFS_IFEXTIREC) { 3678 /* Find target extent list */ 3679 int erp_idx = 0; 3680 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); 3681 base = erp->er_extbuf; 3682 high = erp->er_extcount - 1; 3683 } else { 3684 base = ifp->if_u1.if_extents; 3685 high = nextents - 1; 3686 } 3687 /* Binary search extent records */ 3688 while (low <= high) { 3689 idx = (low + high) >> 1; 3690 ep = base + idx; 3691 startoff = xfs_bmbt_get_startoff(ep); 3692 blockcount = xfs_bmbt_get_blockcount(ep); 3693 if (bno < startoff) { 3694 high = idx - 1; 3695 } else if (bno >= startoff + blockcount) { 3696 low = idx + 1; 3697 } else { 3698 /* Convert back to file-based extent index */ 3699 if (ifp->if_flags & XFS_IFEXTIREC) { 3700 idx += erp->er_extoff; 3701 } 3702 *idxp = idx; 3703 return ep; 3704 } 3705 } 3706 /* Convert back to file-based extent index */ 3707 if (ifp->if_flags & XFS_IFEXTIREC) { 3708 idx += erp->er_extoff; 3709 } 3710 if (bno >= startoff + blockcount) { 3711 if (++idx == nextents) { 3712 ep = NULL; 3713 } else { 3714 ep = xfs_iext_get_ext(ifp, idx); 3715 } 3716 } 3717 *idxp = idx; 3718 return ep; 3719 } 3720 3721 /* 3722 * Return a pointer to the indirection array entry containing the 3723 * extent record for filesystem block bno. Store the index of the 3724 * target irec in *erp_idxp. 3725 */ 3726 xfs_ext_irec_t * /* pointer to found extent record */ 3727 xfs_iext_bno_to_irec( 3728 xfs_ifork_t *ifp, /* inode fork pointer */ 3729 xfs_fileoff_t bno, /* block number to search for */ 3730 int *erp_idxp) /* irec index of target ext list */ 3731 { 3732 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 3733 xfs_ext_irec_t *erp_next; /* next indirection array entry */ 3734 int erp_idx; /* indirection array index */ 3735 int nlists; /* number of extent irec's (lists) */ 3736 int high; /* binary search upper limit */ 3737 int low; /* binary search lower limit */ 3738 3739 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3740 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3741 erp_idx = 0; 3742 low = 0; 3743 high = nlists - 1; 3744 while (low <= high) { 3745 erp_idx = (low + high) >> 1; 3746 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3747 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; 3748 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { 3749 high = erp_idx - 1; 3750 } else if (erp_next && bno >= 3751 xfs_bmbt_get_startoff(erp_next->er_extbuf)) { 3752 low = erp_idx + 1; 3753 } else { 3754 break; 3755 } 3756 } 3757 *erp_idxp = erp_idx; 3758 return erp; 3759 } 3760 3761 /* 3762 * Return a pointer to the indirection array entry containing the 3763 * extent record at file extent index *idxp. Store the index of the 3764 * target irec in *erp_idxp and store the page index of the target 3765 * extent record in *idxp. 3766 */ 3767 xfs_ext_irec_t * 3768 xfs_iext_idx_to_irec( 3769 xfs_ifork_t *ifp, /* inode fork pointer */ 3770 xfs_extnum_t *idxp, /* extent index (file -> page) */ 3771 int *erp_idxp, /* pointer to target irec */ 3772 int realloc) /* new bytes were just added */ 3773 { 3774 xfs_ext_irec_t *prev; /* pointer to previous irec */ 3775 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ 3776 int erp_idx; /* indirection array index */ 3777 int nlists; /* number of irec's (ex lists) */ 3778 int high; /* binary search upper limit */ 3779 int low; /* binary search lower limit */ 3780 xfs_extnum_t page_idx = *idxp; /* extent index in target list */ 3781 3782 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3783 ASSERT(page_idx >= 0); 3784 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); 3785 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc); 3786 3787 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3788 erp_idx = 0; 3789 low = 0; 3790 high = nlists - 1; 3791 3792 /* Binary search extent irec's */ 3793 while (low <= high) { 3794 erp_idx = (low + high) >> 1; 3795 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3796 prev = erp_idx > 0 ? erp - 1 : NULL; 3797 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && 3798 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { 3799 high = erp_idx - 1; 3800 } else if (page_idx > erp->er_extoff + erp->er_extcount || 3801 (page_idx == erp->er_extoff + erp->er_extcount && 3802 !realloc)) { 3803 low = erp_idx + 1; 3804 } else if (page_idx == erp->er_extoff + erp->er_extcount && 3805 erp->er_extcount == XFS_LINEAR_EXTS) { 3806 ASSERT(realloc); 3807 page_idx = 0; 3808 erp_idx++; 3809 erp = erp_idx < nlists ? erp + 1 : NULL; 3810 break; 3811 } else { 3812 page_idx -= erp->er_extoff; 3813 break; 3814 } 3815 } 3816 *idxp = page_idx; 3817 *erp_idxp = erp_idx; 3818 return(erp); 3819 } 3820 3821 /* 3822 * Allocate and initialize an indirection array once the space needed 3823 * for incore extents increases above XFS_IEXT_BUFSZ. 3824 */ 3825 void 3826 xfs_iext_irec_init( 3827 xfs_ifork_t *ifp) /* inode fork pointer */ 3828 { 3829 xfs_ext_irec_t *erp; /* indirection array pointer */ 3830 xfs_extnum_t nextents; /* number of extents in file */ 3831 3832 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3833 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3834 ASSERT(nextents <= XFS_LINEAR_EXTS); 3835 3836 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); 3837 3838 if (nextents == 0) { 3839 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); 3840 } else if (!ifp->if_real_bytes) { 3841 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); 3842 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { 3843 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); 3844 } 3845 erp->er_extbuf = ifp->if_u1.if_extents; 3846 erp->er_extcount = nextents; 3847 erp->er_extoff = 0; 3848 3849 ifp->if_flags |= XFS_IFEXTIREC; 3850 ifp->if_real_bytes = XFS_IEXT_BUFSZ; 3851 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); 3852 ifp->if_u1.if_ext_irec = erp; 3853 3854 return; 3855 } 3856 3857 /* 3858 * Allocate and initialize a new entry in the indirection array. 3859 */ 3860 xfs_ext_irec_t * 3861 xfs_iext_irec_new( 3862 xfs_ifork_t *ifp, /* inode fork pointer */ 3863 int erp_idx) /* index for new irec */ 3864 { 3865 xfs_ext_irec_t *erp; /* indirection array pointer */ 3866 int i; /* loop counter */ 3867 int nlists; /* number of irec's (ex lists) */ 3868 3869 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3870 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3871 3872 /* Resize indirection array */ 3873 xfs_iext_realloc_indirect(ifp, ++nlists * 3874 sizeof(xfs_ext_irec_t)); 3875 /* 3876 * Move records down in the array so the 3877 * new page can use erp_idx. 3878 */ 3879 erp = ifp->if_u1.if_ext_irec; 3880 for (i = nlists - 1; i > erp_idx; i--) { 3881 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); 3882 } 3883 ASSERT(i == erp_idx); 3884 3885 /* Initialize new extent record */ 3886 erp = ifp->if_u1.if_ext_irec; 3887 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); 3888 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 3889 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); 3890 erp[erp_idx].er_extcount = 0; 3891 erp[erp_idx].er_extoff = erp_idx > 0 ? 3892 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; 3893 return (&erp[erp_idx]); 3894 } 3895 3896 /* 3897 * Remove a record from the indirection array. 3898 */ 3899 void 3900 xfs_iext_irec_remove( 3901 xfs_ifork_t *ifp, /* inode fork pointer */ 3902 int erp_idx) /* irec index to remove */ 3903 { 3904 xfs_ext_irec_t *erp; /* indirection array pointer */ 3905 int i; /* loop counter */ 3906 int nlists; /* number of irec's (ex lists) */ 3907 3908 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3909 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3910 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3911 if (erp->er_extbuf) { 3912 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, 3913 -erp->er_extcount); 3914 kmem_free(erp->er_extbuf); 3915 } 3916 /* Compact extent records */ 3917 erp = ifp->if_u1.if_ext_irec; 3918 for (i = erp_idx; i < nlists - 1; i++) { 3919 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); 3920 } 3921 /* 3922 * Manually free the last extent record from the indirection 3923 * array. A call to xfs_iext_realloc_indirect() with a size 3924 * of zero would result in a call to xfs_iext_destroy() which 3925 * would in turn call this function again, creating a nasty 3926 * infinite loop. 3927 */ 3928 if (--nlists) { 3929 xfs_iext_realloc_indirect(ifp, 3930 nlists * sizeof(xfs_ext_irec_t)); 3931 } else { 3932 kmem_free(ifp->if_u1.if_ext_irec); 3933 } 3934 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 3935 } 3936 3937 /* 3938 * This is called to clean up large amounts of unused memory allocated 3939 * by the indirection array. Before compacting anything though, verify 3940 * that the indirection array is still needed and switch back to the 3941 * linear extent list (or even the inline buffer) if possible. The 3942 * compaction policy is as follows: 3943 * 3944 * Full Compaction: Extents fit into a single page (or inline buffer) 3945 * Partial Compaction: Extents occupy less than 50% of allocated space 3946 * No Compaction: Extents occupy at least 50% of allocated space 3947 */ 3948 void 3949 xfs_iext_irec_compact( 3950 xfs_ifork_t *ifp) /* inode fork pointer */ 3951 { 3952 xfs_extnum_t nextents; /* number of extents in file */ 3953 int nlists; /* number of irec's (ex lists) */ 3954 3955 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3956 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3957 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3958 3959 if (nextents == 0) { 3960 xfs_iext_destroy(ifp); 3961 } else if (nextents <= XFS_INLINE_EXTS) { 3962 xfs_iext_indirect_to_direct(ifp); 3963 xfs_iext_direct_to_inline(ifp, nextents); 3964 } else if (nextents <= XFS_LINEAR_EXTS) { 3965 xfs_iext_indirect_to_direct(ifp); 3966 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { 3967 xfs_iext_irec_compact_pages(ifp); 3968 } 3969 } 3970 3971 /* 3972 * Combine extents from neighboring extent pages. 3973 */ 3974 void 3975 xfs_iext_irec_compact_pages( 3976 xfs_ifork_t *ifp) /* inode fork pointer */ 3977 { 3978 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ 3979 int erp_idx = 0; /* indirection array index */ 3980 int nlists; /* number of irec's (ex lists) */ 3981 3982 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3983 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3984 while (erp_idx < nlists - 1) { 3985 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3986 erp_next = erp + 1; 3987 if (erp_next->er_extcount <= 3988 (XFS_LINEAR_EXTS - erp->er_extcount)) { 3989 memcpy(&erp->er_extbuf[erp->er_extcount], 3990 erp_next->er_extbuf, erp_next->er_extcount * 3991 sizeof(xfs_bmbt_rec_t)); 3992 erp->er_extcount += erp_next->er_extcount; 3993 /* 3994 * Free page before removing extent record 3995 * so er_extoffs don't get modified in 3996 * xfs_iext_irec_remove. 3997 */ 3998 kmem_free(erp_next->er_extbuf); 3999 erp_next->er_extbuf = NULL; 4000 xfs_iext_irec_remove(ifp, erp_idx + 1); 4001 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4002 } else { 4003 erp_idx++; 4004 } 4005 } 4006 } 4007 4008 /* 4009 * This is called to update the er_extoff field in the indirection 4010 * array when extents have been added or removed from one of the 4011 * extent lists. erp_idx contains the irec index to begin updating 4012 * at and ext_diff contains the number of extents that were added 4013 * or removed. 4014 */ 4015 void 4016 xfs_iext_irec_update_extoffs( 4017 xfs_ifork_t *ifp, /* inode fork pointer */ 4018 int erp_idx, /* irec index to update */ 4019 int ext_diff) /* number of new extents */ 4020 { 4021 int i; /* loop counter */ 4022 int nlists; /* number of irec's (ex lists */ 4023 4024 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 4025 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 4026 for (i = erp_idx; i < nlists; i++) { 4027 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; 4028 } 4029 } 4030 4031 /* 4032 * Test whether it is appropriate to check an inode for and free post EOF 4033 * blocks. The 'force' parameter determines whether we should also consider 4034 * regular files that are marked preallocated or append-only. 4035 */ 4036 bool 4037 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) 4038 { 4039 /* prealloc/delalloc exists only on regular files */ 4040 if (!S_ISREG(ip->i_d.di_mode)) 4041 return false; 4042 4043 /* 4044 * Zero sized files with no cached pages and delalloc blocks will not 4045 * have speculative prealloc/delalloc blocks to remove. 4046 */ 4047 if (VFS_I(ip)->i_size == 0 && 4048 VN_CACHED(VFS_I(ip)) == 0 && 4049 ip->i_delayed_blks == 0) 4050 return false; 4051 4052 /* If we haven't read in the extent list, then don't do it now. */ 4053 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) 4054 return false; 4055 4056 /* 4057 * Do not free real preallocated or append-only files unless the file 4058 * has delalloc blocks and we are forced to remove them. 4059 */ 4060 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 4061 if (!force || ip->i_delayed_blks == 0) 4062 return false; 4063 4064 return true; 4065 } 4066 4067