1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 34 * $FreeBSD: src/sys/ufs/ffs/ffs_inode.c,v 1.56.2.5 2002/02/05 18:35:03 dillon Exp $ 35 * $DragonFly: src/sys/vfs/ufs/ffs_inode.c,v 1.24 2007/06/14 02:55:25 dillon Exp $ 36 */ 37 38 #include "opt_quota.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/mount.h> 43 #include <sys/proc.h> 44 #include <sys/buf.h> 45 #include <sys/vnode.h> 46 #include <sys/kernel.h> 47 #include <sys/malloc.h> 48 #include <sys/resourcevar.h> 49 #include <sys/vmmeter.h> 50 51 #include <vm/vm.h> 52 #include <vm/vm_extern.h> 53 54 #include "quota.h" 55 #include "ufsmount.h" 56 #include "inode.h" 57 #include "ufs_extern.h" 58 59 #include "fs.h" 60 #include "ffs_extern.h" 61 62 #include <vm/vm_page2.h> 63 64 static int ffs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t, 65 ufs_daddr_t, int, long *); 66 67 /* 68 * Update the access, modified, and inode change times as specified by the 69 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode 70 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by 71 * the timestamp update). The IN_LAZYMOD flag is set to force a write 72 * later if not now. If we write now, then clear both IN_MODIFIED and 73 * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is 74 * set, then wait for the write to complete. 75 */ 76 int 77 ffs_update(struct vnode *vp, int waitfor) 78 { 79 struct fs *fs; 80 struct buf *bp; 81 struct inode *ip; 82 int error; 83 84 ufs_itimes(vp); 85 ip = VTOI(vp); 86 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) 87 return (0); 88 ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED); 89 fs = ip->i_fs; 90 if (fs->fs_ronly) 91 return (0); 92 93 /* 94 * The vnode type is usually set to VBAD if an unrecoverable I/O 95 * error has occured (such as when reading the inode). Clear the 96 * modified bits but do not write anything out in this case. 97 */ 98 if (vp->v_type == VBAD) 99 return (0); 100 /* 101 * Ensure that uid and gid are correct. This is a temporary 102 * fix until fsck has been changed to do the update. 103 */ 104 if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */ 105 ip->i_din.di_ouid = ip->i_uid; /* XXX */ 106 ip->i_din.di_ogid = ip->i_gid; /* XXX */ 107 } /* XXX */ 108 error = bread(ip->i_devvp, 109 fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)), 110 (int)fs->fs_bsize, &bp); 111 if (error) { 112 brelse(bp); 113 return (error); 114 } 115 if (DOINGSOFTDEP(vp)) 116 softdep_update_inodeblock(ip, bp, waitfor); 117 else if (ip->i_effnlink != ip->i_nlink) 118 panic("ffs_update: bad link cnt"); 119 *((struct ufs1_dinode *)bp->b_data + 120 ino_to_fsbo(fs, ip->i_number)) = ip->i_din; 121 if (waitfor && !DOINGASYNC(vp)) { 122 return (bwrite(bp)); 123 } else if (vm_page_count_severe() || buf_dirty_count_severe()) { 124 return (bwrite(bp)); 125 } else { 126 if (bp->b_bufsize == fs->fs_bsize) 127 bp->b_flags |= B_CLUSTEROK; 128 bdwrite(bp); 129 return (0); 130 } 131 } 132 133 #define SINGLE 0 /* index of single indirect block */ 134 #define DOUBLE 1 /* index of double indirect block */ 135 #define TRIPLE 2 /* index of triple indirect block */ 136 /* 137 * Truncate the inode oip to at most length size, freeing the 138 * disk blocks. 139 */ 140 int 141 ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred) 142 { 143 struct vnode *ovp = vp; 144 ufs_daddr_t lastblock; 145 struct inode *oip; 146 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 147 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR]; 148 struct fs *fs; 149 struct buf *bp; 150 int offset, size, level; 151 long count, nblocks, blocksreleased = 0; 152 int i; 153 int aflags, error, allerror; 154 off_t osize; 155 156 oip = VTOI(ovp); 157 fs = oip->i_fs; 158 if (length < 0) 159 return (EINVAL); 160 if (length > fs->fs_maxfilesize) 161 return (EFBIG); 162 if (ovp->v_type == VLNK && 163 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) { 164 #ifdef DIAGNOSTIC 165 if (length != 0) 166 panic("ffs_truncate: partial truncate of symlink"); 167 #endif /* DIAGNOSTIC */ 168 bzero((char *)&oip->i_shortlink, (uint)oip->i_size); 169 oip->i_size = 0; 170 oip->i_flag |= IN_CHANGE | IN_UPDATE; 171 return (ffs_update(ovp, 1)); 172 } 173 if (oip->i_size == length) { 174 oip->i_flag |= IN_CHANGE | IN_UPDATE; 175 return (ffs_update(ovp, 0)); 176 } 177 if (fs->fs_ronly) 178 panic("ffs_truncate: read-only filesystem"); 179 #ifdef QUOTA 180 error = ufs_getinoquota(oip); 181 if (error) 182 return (error); 183 #endif 184 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0; 185 if (DOINGSOFTDEP(ovp)) { 186 if (length > 0 || softdep_slowdown(ovp)) { 187 /* 188 * If a file is only partially truncated, then 189 * we have to clean up the data structures 190 * describing the allocation past the truncation 191 * point. Finding and deallocating those structures 192 * is a lot of work. Since partial truncation occurs 193 * rarely, we solve the problem by syncing the file 194 * so that it will have no data structures left. 195 */ 196 if ((error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0) 197 return (error); 198 } else { 199 #ifdef QUOTA 200 (void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0); 201 #endif 202 softdep_setup_freeblocks(oip, length); 203 vinvalbuf(ovp, 0, 0, 0); 204 vnode_pager_setsize(ovp, 0); 205 oip->i_flag |= IN_CHANGE | IN_UPDATE; 206 return (ffs_update(ovp, 0)); 207 } 208 } 209 osize = oip->i_size; 210 /* 211 * Lengthen the size of the file. We must ensure that the 212 * last byte of the file is allocated. Since the smallest 213 * value of osize is 0, length will be at least 1. 214 */ 215 if (osize < length) { 216 vnode_pager_setsize(ovp, length); 217 aflags = B_CLRBUF; 218 if (flags & IO_SYNC) 219 aflags |= B_SYNC; 220 error = VOP_BALLOC(ovp, length - 1, 1, 221 cred, aflags, &bp); 222 if (error) 223 return (error); 224 oip->i_size = length; 225 if (bp->b_bufsize == fs->fs_bsize) 226 bp->b_flags |= B_CLUSTEROK; 227 if (aflags & B_SYNC) 228 bwrite(bp); 229 else 230 bawrite(bp); 231 oip->i_flag |= IN_CHANGE | IN_UPDATE; 232 return (ffs_update(ovp, 1)); 233 } 234 /* 235 * Shorten the size of the file. If the file is not being 236 * truncated to a block boundary, the contents of the 237 * partial block following the end of the file must be 238 * zero'ed in case it ever becomes accessible again because 239 * of subsequent file growth. Directories however are not 240 * zero'ed as they should grow back initialized to empty. 241 */ 242 offset = blkoff(fs, length); 243 if (offset == 0) { 244 oip->i_size = length; 245 } else { 246 lbn = lblkno(fs, length); 247 aflags = B_CLRBUF; 248 if (flags & IO_SYNC) 249 aflags |= B_SYNC; 250 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp); 251 if (error) { 252 return (error); 253 } 254 /* 255 * When we are doing soft updates and the UFS_BALLOC 256 * above fills in a direct block hole with a full sized 257 * block that will be truncated down to a fragment below, 258 * we must flush out the block dependency with an FSYNC 259 * so that we do not get a soft updates inconsistency 260 * when we create the fragment below. 261 */ 262 if (DOINGSOFTDEP(ovp) && lbn < NDADDR && 263 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && 264 (error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0) { 265 return (error); 266 } 267 oip->i_size = length; 268 size = blksize(fs, oip, lbn); 269 if (ovp->v_type != VDIR) 270 bzero((char *)bp->b_data + offset, 271 (uint)(size - offset)); 272 /* Kirk's code has reallocbuf(bp, size, 1) here */ 273 allocbuf(bp, size); 274 if (bp->b_bufsize == fs->fs_bsize) 275 bp->b_flags |= B_CLUSTEROK; 276 if (aflags & B_SYNC) 277 bwrite(bp); 278 else 279 bawrite(bp); 280 } 281 /* 282 * Calculate index into inode's block list of 283 * last direct and indirect blocks (if any) 284 * which we want to keep. Lastblock is -1 when 285 * the file is truncated to 0. 286 */ 287 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 288 lastiblock[SINGLE] = lastblock - NDADDR; 289 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 290 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 291 nblocks = btodb(fs->fs_bsize); 292 293 /* 294 * Update file and block pointers on disk before we start freeing 295 * blocks. If we crash before free'ing blocks below, the blocks 296 * will be returned to the free list. lastiblock values are also 297 * normalized to -1 for calls to ffs_indirtrunc below. 298 */ 299 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks); 300 for (level = TRIPLE; level >= SINGLE; level--) 301 if (lastiblock[level] < 0) { 302 oip->i_ib[level] = 0; 303 lastiblock[level] = -1; 304 } 305 for (i = NDADDR - 1; i > lastblock; i--) 306 oip->i_db[i] = 0; 307 oip->i_flag |= IN_CHANGE | IN_UPDATE; 308 allerror = ffs_update(ovp, 1); 309 310 /* 311 * Having written the new inode to disk, save its new configuration 312 * and put back the old block pointers long enough to process them. 313 * Note that we save the new block configuration so we can check it 314 * when we are done. 315 */ 316 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks); 317 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks); 318 oip->i_size = osize; 319 320 error = vtruncbuf(ovp, length, fs->fs_bsize); 321 if (error && (allerror == 0)) 322 allerror = error; 323 324 /* 325 * Indirect blocks first. 326 */ 327 indir_lbn[SINGLE] = -NDADDR; 328 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 329 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 330 for (level = TRIPLE; level >= SINGLE; level--) { 331 bn = oip->i_ib[level]; 332 if (bn != 0) { 333 error = ffs_indirtrunc(oip, indir_lbn[level], 334 fsbtodb(fs, bn), lastiblock[level], level, &count); 335 if (error) 336 allerror = error; 337 blocksreleased += count; 338 if (lastiblock[level] < 0) { 339 oip->i_ib[level] = 0; 340 ffs_blkfree(oip, bn, fs->fs_bsize); 341 blocksreleased += nblocks; 342 } 343 } 344 if (lastiblock[level] >= 0) 345 goto done; 346 } 347 348 /* 349 * All whole direct blocks or frags. 350 */ 351 for (i = NDADDR - 1; i > lastblock; i--) { 352 long bsize; 353 354 bn = oip->i_db[i]; 355 if (bn == 0) 356 continue; 357 oip->i_db[i] = 0; 358 bsize = blksize(fs, oip, i); 359 ffs_blkfree(oip, bn, bsize); 360 blocksreleased += btodb(bsize); 361 } 362 if (lastblock < 0) 363 goto done; 364 365 /* 366 * Finally, look for a change in size of the 367 * last direct block; release any frags. 368 */ 369 bn = oip->i_db[lastblock]; 370 if (bn != 0) { 371 long oldspace, newspace; 372 373 /* 374 * Calculate amount of space we're giving 375 * back as old block size minus new block size. 376 */ 377 oldspace = blksize(fs, oip, lastblock); 378 oip->i_size = length; 379 newspace = blksize(fs, oip, lastblock); 380 if (newspace == 0) 381 panic("ffs_truncate: newspace"); 382 if (oldspace - newspace > 0) { 383 /* 384 * Block number of space to be free'd is 385 * the old block # plus the number of frags 386 * required for the storage we're keeping. 387 */ 388 bn += numfrags(fs, newspace); 389 ffs_blkfree(oip, bn, oldspace - newspace); 390 blocksreleased += btodb(oldspace - newspace); 391 } 392 } 393 done: 394 #ifdef DIAGNOSTIC 395 for (level = SINGLE; level <= TRIPLE; level++) 396 if (newblks[NDADDR + level] != oip->i_ib[level]) 397 panic("ffs_truncate1"); 398 for (i = 0; i < NDADDR; i++) 399 if (newblks[i] != oip->i_db[i]) 400 panic("ffs_truncate2"); 401 if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree)) 402 panic("ffs_truncate3"); 403 #endif /* DIAGNOSTIC */ 404 /* 405 * Put back the real size. 406 */ 407 oip->i_size = length; 408 oip->i_blocks -= blocksreleased; 409 410 if (oip->i_blocks < 0) /* sanity */ 411 oip->i_blocks = 0; 412 oip->i_flag |= IN_CHANGE; 413 #ifdef QUOTA 414 (void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0); 415 #endif 416 return (allerror); 417 } 418 419 /* 420 * Release blocks associated with the inode ip and stored in the indirect 421 * block bn. Blocks are free'd in LIFO order up to (but not including) 422 * lastbn. If level is greater than SINGLE, the block is an indirect block 423 * and recursive calls to indirtrunc must be used to cleanse other indirect 424 * blocks. 425 * 426 * NB: triple indirect blocks are untested. 427 */ 428 static int 429 ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn, 430 ufs_daddr_t lastbn, int level, long *countp) 431 { 432 int i; 433 struct buf *bp; 434 struct fs *fs = ip->i_fs; 435 ufs_daddr_t *bap; 436 struct vnode *vp; 437 ufs_daddr_t *copy = NULL, nb, nlbn, last; 438 long blkcount, factor; 439 int nblocks, blocksreleased = 0; 440 int error = 0, allerror = 0; 441 442 /* 443 * Calculate index in current block of last 444 * block to be kept. -1 indicates the entire 445 * block so we need not calculate the index. 446 */ 447 factor = 1; 448 for (i = SINGLE; i < level; i++) 449 factor *= NINDIR(fs); 450 last = lastbn; 451 if (lastbn > 0) 452 last /= factor; 453 nblocks = btodb(fs->fs_bsize); 454 /* 455 * Get buffer of block pointers, zero those entries corresponding 456 * to blocks to be free'd, and update on disk copy first. Since 457 * double(triple) indirect before single(double) indirect, calls 458 * to bmap on these blocks will fail. However, we already have 459 * the on disk address, so we have to set the bio_offset field 460 * explicitly instead of letting bread do everything for us. 461 */ 462 vp = ITOV(ip); 463 bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0); 464 if ((bp->b_flags & B_CACHE) == 0) { 465 bp->b_flags &= ~(B_ERROR|B_INVAL); 466 bp->b_cmd = BUF_CMD_READ; 467 if (bp->b_bcount > bp->b_bufsize) 468 panic("ffs_indirtrunc: bad buffer size"); 469 /* 470 * BIO is bio2 which chains back to bio1. We wait 471 * on bio1. 472 */ 473 bp->b_bio2.bio_offset = dbtodoff(fs, dbn); 474 bp->b_bio1.bio_done = biodone_sync; 475 bp->b_bio1.bio_flags |= BIO_SYNC; 476 vfs_busy_pages(vp, bp); 477 /* 478 * Access the block device layer using the device vnode 479 * and the translated block number (bio2) instead of the 480 * file vnode (vp) and logical block number (bio1). 481 * 482 * Even though we are bypassing the vnode layer, we still 483 * want the vnode state to indicate that an I/O on its behalf 484 * is in progress. 485 */ 486 bio_start_transaction(&bp->b_bio1, &vp->v_track_read); 487 vn_strategy(ip->i_devvp, &bp->b_bio2); 488 error = biowait(&bp->b_bio1, "biord"); 489 } 490 if (error) { 491 brelse(bp); 492 *countp = 0; 493 return (error); 494 } 495 496 bap = (ufs_daddr_t *)bp->b_data; 497 if (lastbn != -1) { 498 MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK); 499 bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize); 500 bzero((caddr_t)&bap[last + 1], 501 (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 502 if (DOINGASYNC(vp)) { 503 bawrite(bp); 504 } else { 505 error = bwrite(bp); 506 if (error) 507 allerror = error; 508 } 509 bap = copy; 510 } 511 512 /* 513 * Recursively free totally unused blocks. 514 */ 515 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 516 i--, nlbn += factor) { 517 nb = bap[i]; 518 if (nb == 0) 519 continue; 520 if (level > SINGLE) { 521 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 522 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0) 523 allerror = error; 524 blocksreleased += blkcount; 525 } 526 ffs_blkfree(ip, nb, fs->fs_bsize); 527 blocksreleased += nblocks; 528 } 529 530 /* 531 * Recursively free last partial block. 532 */ 533 if (level > SINGLE && lastbn >= 0) { 534 last = lastbn % factor; 535 nb = bap[i]; 536 if (nb != 0) { 537 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 538 last, level - 1, &blkcount); 539 if (error) 540 allerror = error; 541 blocksreleased += blkcount; 542 } 543 } 544 if (copy != NULL) { 545 FREE(copy, M_TEMP); 546 } else { 547 bp->b_flags |= B_INVAL | B_NOCACHE; 548 brelse(bp); 549 } 550 551 *countp = blocksreleased; 552 return (allerror); 553 } 554