1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_quota.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/bio.h> 42 #include <sys/buf.h> 43 #include <sys/malloc.h> 44 #include <sys/mount.h> 45 #include <sys/proc.h> 46 #include <sys/racct.h> 47 #include <sys/random.h> 48 #include <sys/resourcevar.h> 49 #include <sys/rwlock.h> 50 #include <sys/stat.h> 51 #include <sys/vmmeter.h> 52 #include <sys/vnode.h> 53 54 #include <vm/vm.h> 55 #include <vm/vm_extern.h> 56 #include <vm/vm_object.h> 57 58 #include <ufs/ufs/extattr.h> 59 #include <ufs/ufs/quota.h> 60 #include <ufs/ufs/ufsmount.h> 61 #include <ufs/ufs/inode.h> 62 #include <ufs/ufs/ufs_extern.h> 63 64 #include <ufs/ffs/fs.h> 65 #include <ufs/ffs/ffs_extern.h> 66 67 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t, 68 ufs2_daddr_t, int, ufs2_daddr_t *); 69 70 /* 71 * Update the access, modified, and inode change times as specified by the 72 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode 73 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by 74 * the timestamp update). The IN_LAZYMOD flag is set to force a write 75 * later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs 76 * is currently being suspended (or is suspended) and vnode has been accessed. 77 * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to 78 * reflect the presumably successful write, and if waitfor is set, then wait 79 * for the write to complete. 80 */ 81 int 82 ffs_update(vp, waitfor) 83 struct vnode *vp; 84 int waitfor; 85 { 86 struct fs *fs; 87 struct buf *bp; 88 struct inode *ip; 89 int flags, error; 90 91 ASSERT_VOP_ELOCKED(vp, "ffs_update"); 92 ufs_itimes(vp); 93 ip = VTOI(vp); 94 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) 95 return (0); 96 ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED); 97 fs = ITOFS(ip); 98 if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0) 99 return (0); 100 /* 101 * If we are updating a snapshot and another process is currently 102 * writing the buffer containing the inode for this snapshot then 103 * a deadlock can occur when it tries to check the snapshot to see 104 * if that block needs to be copied. Thus when updating a snapshot 105 * we check to see if the buffer is already locked, and if it is 106 * we drop the snapshot lock until the buffer has been written 107 * and is available to us. We have to grab a reference to the 108 * snapshot vnode to prevent it from being removed while we are 109 * waiting for the buffer. 110 */ 111 flags = 0; 112 if (IS_SNAPSHOT(ip)) 113 flags = GB_LOCK_NOWAIT; 114 loop: 115 error = bread_gb(ITODEVVP(ip), 116 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 117 (int) fs->fs_bsize, NOCRED, flags, &bp); 118 if (error != 0) { 119 if (error != EBUSY) 120 return (error); 121 KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot")); 122 /* 123 * Wait for our inode block to become available. 124 * 125 * Hold a reference to the vnode to protect against 126 * ffs_snapgone(). Since we hold a reference, it can only 127 * get reclaimed (VI_DOOMED flag) in a forcible downgrade 128 * or unmount. For an unmount, the entire filesystem will be 129 * gone, so we cannot attempt to touch anything associated 130 * with it while the vnode is unlocked; all we can do is 131 * pause briefly and try again. If when we relock the vnode 132 * we discover that it has been reclaimed, updating it is no 133 * longer necessary and we can just return an error. 134 */ 135 vref(vp); 136 VOP_UNLOCK(vp, 0); 137 pause("ffsupd", 1); 138 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 139 vrele(vp); 140 if ((vp->v_iflag & VI_DOOMED) != 0) 141 return (ENOENT); 142 goto loop; 143 } 144 if (DOINGSOFTDEP(vp)) 145 softdep_update_inodeblock(ip, bp, waitfor); 146 else if (ip->i_effnlink != ip->i_nlink) 147 panic("ffs_update: bad link cnt"); 148 if (I_IS_UFS1(ip)) { 149 *((struct ufs1_dinode *)bp->b_data + 150 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1; 151 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */ 152 random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), 1, RANDOM_FS_ATIME); 153 } else { 154 *((struct ufs2_dinode *)bp->b_data + 155 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2; 156 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */ 157 random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), 1, RANDOM_FS_ATIME); 158 } 159 if (waitfor) 160 error = bwrite(bp); 161 else if (vm_page_count_severe() || buf_dirty_count_severe()) { 162 bawrite(bp); 163 error = 0; 164 } else { 165 if (bp->b_bufsize == fs->fs_bsize) 166 bp->b_flags |= B_CLUSTEROK; 167 bdwrite(bp); 168 error = 0; 169 } 170 return (error); 171 } 172 173 #define SINGLE 0 /* index of single indirect block */ 174 #define DOUBLE 1 /* index of double indirect block */ 175 #define TRIPLE 2 /* index of triple indirect block */ 176 /* 177 * Truncate the inode ip to at most length size, freeing the 178 * disk blocks. 179 */ 180 int 181 ffs_truncate(vp, length, flags, cred) 182 struct vnode *vp; 183 off_t length; 184 int flags; 185 struct ucred *cred; 186 { 187 struct inode *ip; 188 ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR]; 189 ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR]; 190 ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR]; 191 ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno; 192 struct bufobj *bo; 193 struct fs *fs; 194 struct buf *bp; 195 struct ufsmount *ump; 196 int softdeptrunc, journaltrunc; 197 int needextclean, extblocks; 198 int offset, size, level, nblocks; 199 int i, error, allerror, indiroff, waitforupdate; 200 off_t osize; 201 202 ip = VTOI(vp); 203 ump = VFSTOUFS(vp->v_mount); 204 fs = ump->um_fs; 205 bo = &vp->v_bufobj; 206 207 ASSERT_VOP_LOCKED(vp, "ffs_truncate"); 208 209 if (length < 0) 210 return (EINVAL); 211 if (length > fs->fs_maxfilesize) 212 return (EFBIG); 213 #ifdef QUOTA 214 error = getinoquota(ip); 215 if (error) 216 return (error); 217 #endif 218 /* 219 * Historically clients did not have to specify which data 220 * they were truncating. So, if not specified, we assume 221 * traditional behavior, e.g., just the normal data. 222 */ 223 if ((flags & (IO_EXT | IO_NORMAL)) == 0) 224 flags |= IO_NORMAL; 225 if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp)) 226 flags |= IO_SYNC; 227 waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp); 228 /* 229 * If we are truncating the extended-attributes, and cannot 230 * do it with soft updates, then do it slowly here. If we are 231 * truncating both the extended attributes and the file contents 232 * (e.g., the file is being unlinked), then pick it off with 233 * soft updates below. 234 */ 235 allerror = 0; 236 needextclean = 0; 237 softdeptrunc = 0; 238 journaltrunc = DOINGSUJ(vp); 239 if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0) 240 softdeptrunc = !softdep_slowdown(vp); 241 extblocks = 0; 242 datablocks = DIP(ip, i_blocks); 243 if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) { 244 extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); 245 datablocks -= extblocks; 246 } 247 if ((flags & IO_EXT) && extblocks > 0) { 248 if (length != 0) 249 panic("ffs_truncate: partial trunc of extdata"); 250 if (softdeptrunc || journaltrunc) { 251 if ((flags & IO_NORMAL) == 0) 252 goto extclean; 253 needextclean = 1; 254 } else { 255 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 256 return (error); 257 #ifdef QUOTA 258 (void) chkdq(ip, -extblocks, NOCRED, 0); 259 #endif 260 vinvalbuf(vp, V_ALT, 0, 0); 261 vn_pages_remove(vp, 262 OFF_TO_IDX(lblktosize(fs, -extblocks)), 0); 263 osize = ip->i_din2->di_extsize; 264 ip->i_din2->di_blocks -= extblocks; 265 ip->i_din2->di_extsize = 0; 266 for (i = 0; i < UFS_NXADDR; i++) { 267 oldblks[i] = ip->i_din2->di_extb[i]; 268 ip->i_din2->di_extb[i] = 0; 269 } 270 ip->i_flag |= IN_CHANGE; 271 if ((error = ffs_update(vp, waitforupdate))) 272 return (error); 273 for (i = 0; i < UFS_NXADDR; i++) { 274 if (oldblks[i] == 0) 275 continue; 276 ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i], 277 sblksize(fs, osize, i), ip->i_number, 278 vp->v_type, NULL); 279 } 280 } 281 } 282 if ((flags & IO_NORMAL) == 0) 283 return (0); 284 if (vp->v_type == VLNK && 285 (ip->i_size < vp->v_mount->mnt_maxsymlinklen || 286 datablocks == 0)) { 287 #ifdef INVARIANTS 288 if (length != 0) 289 panic("ffs_truncate: partial truncate of symlink"); 290 #endif 291 bzero(SHORTLINK(ip), (u_int)ip->i_size); 292 ip->i_size = 0; 293 DIP_SET(ip, i_size, 0); 294 ip->i_flag |= IN_CHANGE | IN_UPDATE; 295 if (needextclean) 296 goto extclean; 297 return (ffs_update(vp, waitforupdate)); 298 } 299 if (ip->i_size == length) { 300 ip->i_flag |= IN_CHANGE | IN_UPDATE; 301 if (needextclean) 302 goto extclean; 303 return (ffs_update(vp, 0)); 304 } 305 if (fs->fs_ronly) 306 panic("ffs_truncate: read-only filesystem"); 307 if (IS_SNAPSHOT(ip)) 308 ffs_snapremove(vp); 309 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; 310 osize = ip->i_size; 311 /* 312 * Lengthen the size of the file. We must ensure that the 313 * last byte of the file is allocated. Since the smallest 314 * value of osize is 0, length will be at least 1. 315 */ 316 if (osize < length) { 317 vnode_pager_setsize(vp, length); 318 flags |= BA_CLRBUF; 319 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 320 if (error) { 321 vnode_pager_setsize(vp, osize); 322 return (error); 323 } 324 ip->i_size = length; 325 DIP_SET(ip, i_size, length); 326 if (bp->b_bufsize == fs->fs_bsize) 327 bp->b_flags |= B_CLUSTEROK; 328 if (flags & IO_SYNC) 329 bwrite(bp); 330 else if (DOINGASYNC(vp)) 331 bdwrite(bp); 332 else 333 bawrite(bp); 334 ip->i_flag |= IN_CHANGE | IN_UPDATE; 335 return (ffs_update(vp, waitforupdate)); 336 } 337 /* 338 * Lookup block number for a given offset. Zero length files 339 * have no blocks, so return a blkno of -1. 340 */ 341 lbn = lblkno(fs, length - 1); 342 if (length == 0) { 343 blkno = -1; 344 } else if (lbn < UFS_NDADDR) { 345 blkno = DIP(ip, i_db[lbn]); 346 } else { 347 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, 348 cred, BA_METAONLY, &bp); 349 if (error) 350 return (error); 351 indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); 352 if (I_IS_UFS1(ip)) 353 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; 354 else 355 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; 356 /* 357 * If the block number is non-zero, then the indirect block 358 * must have been previously allocated and need not be written. 359 * If the block number is zero, then we may have allocated 360 * the indirect block and hence need to write it out. 361 */ 362 if (blkno != 0) 363 brelse(bp); 364 else if (flags & IO_SYNC) 365 bwrite(bp); 366 else 367 bdwrite(bp); 368 } 369 /* 370 * If the block number at the new end of the file is zero, 371 * then we must allocate it to ensure that the last block of 372 * the file is allocated. Soft updates does not handle this 373 * case, so here we have to clean up the soft updates data 374 * structures describing the allocation past the truncation 375 * point. Finding and deallocating those structures is a lot of 376 * work. Since partial truncation with a hole at the end occurs 377 * rarely, we solve the problem by syncing the file so that it 378 * will have no soft updates data structures left. 379 */ 380 if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 381 return (error); 382 if (blkno != 0 && DOINGSOFTDEP(vp)) { 383 if (softdeptrunc == 0 && journaltrunc == 0) { 384 /* 385 * If soft updates cannot handle this truncation, 386 * clean up soft dependency data structures and 387 * fall through to the synchronous truncation. 388 */ 389 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 390 return (error); 391 } else { 392 flags = IO_NORMAL | (needextclean ? IO_EXT: 0); 393 if (journaltrunc) 394 softdep_journal_freeblocks(ip, cred, length, 395 flags); 396 else 397 softdep_setup_freeblocks(ip, length, flags); 398 ASSERT_VOP_LOCKED(vp, "ffs_truncate1"); 399 if (journaltrunc == 0) { 400 ip->i_flag |= IN_CHANGE | IN_UPDATE; 401 error = ffs_update(vp, 0); 402 } 403 return (error); 404 } 405 } 406 /* 407 * Shorten the size of the file. If the last block of the 408 * shortened file is unallocated, we must allocate it. 409 * Additionally, if the file is not being truncated to a 410 * block boundary, the contents of the partial block 411 * following the end of the file must be zero'ed in 412 * case it ever becomes accessible again because of 413 * subsequent file growth. Directories however are not 414 * zero'ed as they should grow back initialized to empty. 415 */ 416 offset = blkoff(fs, length); 417 if (blkno != 0 && offset == 0) { 418 ip->i_size = length; 419 DIP_SET(ip, i_size, length); 420 } else { 421 lbn = lblkno(fs, length); 422 flags |= BA_CLRBUF; 423 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); 424 if (error) 425 return (error); 426 /* 427 * When we are doing soft updates and the UFS_BALLOC 428 * above fills in a direct block hole with a full sized 429 * block that will be truncated down to a fragment below, 430 * we must flush out the block dependency with an FSYNC 431 * so that we do not get a soft updates inconsistency 432 * when we create the fragment below. 433 */ 434 if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR && 435 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && 436 (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) 437 return (error); 438 ip->i_size = length; 439 DIP_SET(ip, i_size, length); 440 size = blksize(fs, ip, lbn); 441 if (vp->v_type != VDIR && offset != 0) 442 bzero((char *)bp->b_data + offset, 443 (u_int)(size - offset)); 444 /* Kirk's code has reallocbuf(bp, size, 1) here */ 445 allocbuf(bp, size); 446 if (bp->b_bufsize == fs->fs_bsize) 447 bp->b_flags |= B_CLUSTEROK; 448 if (flags & IO_SYNC) 449 bwrite(bp); 450 else if (DOINGASYNC(vp)) 451 bdwrite(bp); 452 else 453 bawrite(bp); 454 } 455 /* 456 * Calculate index into inode's block list of 457 * last direct and indirect blocks (if any) 458 * which we want to keep. Lastblock is -1 when 459 * the file is truncated to 0. 460 */ 461 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 462 lastiblock[SINGLE] = lastblock - UFS_NDADDR; 463 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 464 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 465 nblocks = btodb(fs->fs_bsize); 466 /* 467 * Update file and block pointers on disk before we start freeing 468 * blocks. If we crash before free'ing blocks below, the blocks 469 * will be returned to the free list. lastiblock values are also 470 * normalized to -1 for calls to ffs_indirtrunc below. 471 */ 472 for (level = TRIPLE; level >= SINGLE; level--) { 473 oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]); 474 if (lastiblock[level] < 0) { 475 DIP_SET(ip, i_ib[level], 0); 476 lastiblock[level] = -1; 477 } 478 } 479 for (i = 0; i < UFS_NDADDR; i++) { 480 oldblks[i] = DIP(ip, i_db[i]); 481 if (i > lastblock) 482 DIP_SET(ip, i_db[i], 0); 483 } 484 ip->i_flag |= IN_CHANGE | IN_UPDATE; 485 allerror = ffs_update(vp, waitforupdate); 486 487 /* 488 * Having written the new inode to disk, save its new configuration 489 * and put back the old block pointers long enough to process them. 490 * Note that we save the new block configuration so we can check it 491 * when we are done. 492 */ 493 for (i = 0; i < UFS_NDADDR; i++) { 494 newblks[i] = DIP(ip, i_db[i]); 495 DIP_SET(ip, i_db[i], oldblks[i]); 496 } 497 for (i = 0; i < UFS_NIADDR; i++) { 498 newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]); 499 DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]); 500 } 501 ip->i_size = osize; 502 DIP_SET(ip, i_size, osize); 503 504 error = vtruncbuf(vp, cred, length, fs->fs_bsize); 505 if (error && (allerror == 0)) 506 allerror = error; 507 508 /* 509 * Indirect blocks first. 510 */ 511 indir_lbn[SINGLE] = -UFS_NDADDR; 512 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 513 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 514 for (level = TRIPLE; level >= SINGLE; level--) { 515 bn = DIP(ip, i_ib[level]); 516 if (bn != 0) { 517 error = ffs_indirtrunc(ip, indir_lbn[level], 518 fsbtodb(fs, bn), lastiblock[level], level, &count); 519 if (error) 520 allerror = error; 521 blocksreleased += count; 522 if (lastiblock[level] < 0) { 523 DIP_SET(ip, i_ib[level], 0); 524 ffs_blkfree(ump, fs, ump->um_devvp, bn, 525 fs->fs_bsize, ip->i_number, 526 vp->v_type, NULL); 527 blocksreleased += nblocks; 528 } 529 } 530 if (lastiblock[level] >= 0) 531 goto done; 532 } 533 534 /* 535 * All whole direct blocks or frags. 536 */ 537 for (i = UFS_NDADDR - 1; i > lastblock; i--) { 538 long bsize; 539 540 bn = DIP(ip, i_db[i]); 541 if (bn == 0) 542 continue; 543 DIP_SET(ip, i_db[i], 0); 544 bsize = blksize(fs, ip, i); 545 ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number, 546 vp->v_type, NULL); 547 blocksreleased += btodb(bsize); 548 } 549 if (lastblock < 0) 550 goto done; 551 552 /* 553 * Finally, look for a change in size of the 554 * last direct block; release any frags. 555 */ 556 bn = DIP(ip, i_db[lastblock]); 557 if (bn != 0) { 558 long oldspace, newspace; 559 560 /* 561 * Calculate amount of space we're giving 562 * back as old block size minus new block size. 563 */ 564 oldspace = blksize(fs, ip, lastblock); 565 ip->i_size = length; 566 DIP_SET(ip, i_size, length); 567 newspace = blksize(fs, ip, lastblock); 568 if (newspace == 0) 569 panic("ffs_truncate: newspace"); 570 if (oldspace - newspace > 0) { 571 /* 572 * Block number of space to be free'd is 573 * the old block # plus the number of frags 574 * required for the storage we're keeping. 575 */ 576 bn += numfrags(fs, newspace); 577 ffs_blkfree(ump, fs, ump->um_devvp, bn, 578 oldspace - newspace, ip->i_number, vp->v_type, NULL); 579 blocksreleased += btodb(oldspace - newspace); 580 } 581 } 582 done: 583 #ifdef INVARIANTS 584 for (level = SINGLE; level <= TRIPLE; level++) 585 if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level])) 586 panic("ffs_truncate1"); 587 for (i = 0; i < UFS_NDADDR; i++) 588 if (newblks[i] != DIP(ip, i_db[i])) 589 panic("ffs_truncate2"); 590 BO_LOCK(bo); 591 if (length == 0 && 592 (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) && 593 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 594 panic("ffs_truncate3"); 595 BO_UNLOCK(bo); 596 #endif /* INVARIANTS */ 597 /* 598 * Put back the real size. 599 */ 600 ip->i_size = length; 601 DIP_SET(ip, i_size, length); 602 if (DIP(ip, i_blocks) >= blocksreleased) 603 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased); 604 else /* sanity */ 605 DIP_SET(ip, i_blocks, 0); 606 ip->i_flag |= IN_CHANGE; 607 #ifdef QUOTA 608 (void) chkdq(ip, -blocksreleased, NOCRED, 0); 609 #endif 610 return (allerror); 611 612 extclean: 613 if (journaltrunc) 614 softdep_journal_freeblocks(ip, cred, length, IO_EXT); 615 else 616 softdep_setup_freeblocks(ip, length, IO_EXT); 617 return (ffs_update(vp, waitforupdate)); 618 } 619 620 /* 621 * Release blocks associated with the inode ip and stored in the indirect 622 * block bn. Blocks are free'd in LIFO order up to (but not including) 623 * lastbn. If level is greater than SINGLE, the block is an indirect block 624 * and recursive calls to indirtrunc must be used to cleanse other indirect 625 * blocks. 626 */ 627 static int 628 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp) 629 struct inode *ip; 630 ufs2_daddr_t lbn, lastbn; 631 ufs2_daddr_t dbn; 632 int level; 633 ufs2_daddr_t *countp; 634 { 635 struct buf *bp; 636 struct fs *fs; 637 struct vnode *vp; 638 caddr_t copy = NULL; 639 int i, nblocks, error = 0, allerror = 0; 640 ufs2_daddr_t nb, nlbn, last; 641 ufs2_daddr_t blkcount, factor, blocksreleased = 0; 642 ufs1_daddr_t *bap1 = NULL; 643 ufs2_daddr_t *bap2 = NULL; 644 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i]) 645 646 fs = ITOFS(ip); 647 648 /* 649 * Calculate index in current block of last 650 * block to be kept. -1 indicates the entire 651 * block so we need not calculate the index. 652 */ 653 factor = lbn_offset(fs, level); 654 last = lastbn; 655 if (lastbn > 0) 656 last /= factor; 657 nblocks = btodb(fs->fs_bsize); 658 /* 659 * Get buffer of block pointers, zero those entries corresponding 660 * to blocks to be free'd, and update on disk copy first. Since 661 * double(triple) indirect before single(double) indirect, calls 662 * to bmap on these blocks will fail. However, we already have 663 * the on disk address, so we have to set the b_blkno field 664 * explicitly instead of letting bread do everything for us. 665 */ 666 vp = ITOV(ip); 667 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0); 668 if ((bp->b_flags & B_CACHE) == 0) { 669 #ifdef RACCT 670 if (racct_enable) { 671 PROC_LOCK(curproc); 672 racct_add_buf(curproc, bp, 0); 673 PROC_UNLOCK(curproc); 674 } 675 #endif /* RACCT */ 676 curthread->td_ru.ru_inblock++; /* pay for read */ 677 bp->b_iocmd = BIO_READ; 678 bp->b_flags &= ~B_INVAL; 679 bp->b_ioflags &= ~BIO_ERROR; 680 if (bp->b_bcount > bp->b_bufsize) 681 panic("ffs_indirtrunc: bad buffer size"); 682 bp->b_blkno = dbn; 683 vfs_busy_pages(bp, 0); 684 bp->b_iooffset = dbtob(bp->b_blkno); 685 bstrategy(bp); 686 error = bufwait(bp); 687 } 688 if (error) { 689 brelse(bp); 690 *countp = 0; 691 return (error); 692 } 693 694 if (I_IS_UFS1(ip)) 695 bap1 = (ufs1_daddr_t *)bp->b_data; 696 else 697 bap2 = (ufs2_daddr_t *)bp->b_data; 698 if (lastbn != -1) { 699 copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK); 700 bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize); 701 for (i = last + 1; i < NINDIR(fs); i++) 702 if (I_IS_UFS1(ip)) 703 bap1[i] = 0; 704 else 705 bap2[i] = 0; 706 if (DOINGASYNC(vp)) { 707 bdwrite(bp); 708 } else { 709 error = bwrite(bp); 710 if (error) 711 allerror = error; 712 } 713 if (I_IS_UFS1(ip)) 714 bap1 = (ufs1_daddr_t *)copy; 715 else 716 bap2 = (ufs2_daddr_t *)copy; 717 } 718 719 /* 720 * Recursively free totally unused blocks. 721 */ 722 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 723 i--, nlbn += factor) { 724 nb = BAP(ip, i); 725 if (nb == 0) 726 continue; 727 if (level > SINGLE) { 728 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 729 (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0) 730 allerror = error; 731 blocksreleased += blkcount; 732 } 733 ffs_blkfree(ITOUMP(ip), fs, ITODEVVP(ip), nb, fs->fs_bsize, 734 ip->i_number, vp->v_type, NULL); 735 blocksreleased += nblocks; 736 } 737 738 /* 739 * Recursively free last partial block. 740 */ 741 if (level > SINGLE && lastbn >= 0) { 742 last = lastbn % factor; 743 nb = BAP(ip, i); 744 if (nb != 0) { 745 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 746 last, level - 1, &blkcount); 747 if (error) 748 allerror = error; 749 blocksreleased += blkcount; 750 } 751 } 752 if (copy != NULL) { 753 free(copy, M_TEMP); 754 } else { 755 bp->b_flags |= B_INVAL | B_NOCACHE; 756 brelse(bp); 757 } 758 759 *countp = blocksreleased; 760 return (allerror); 761 } 762 763 int 764 ffs_rdonly(struct inode *ip) 765 { 766 767 return (ITOFS(ip)->fs_ronly != 0); 768 } 769 770