1 /* $NetBSD: ffs_inode.c,v 1.51 2001/12/18 10:57:21 fvdl Exp $ */ 2 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. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: ffs_inode.c,v 1.51 2001/12/18 10:57:21 fvdl Exp $"); 40 41 #if defined(_KERNEL_OPT) 42 #include "opt_ffs.h" 43 #include "opt_quota.h" 44 #endif 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/mount.h> 49 #include <sys/proc.h> 50 #include <sys/file.h> 51 #include <sys/buf.h> 52 #include <sys/vnode.h> 53 #include <sys/kernel.h> 54 #include <sys/malloc.h> 55 #include <sys/trace.h> 56 #include <sys/resourcevar.h> 57 58 #include <ufs/ufs/quota.h> 59 #include <ufs/ufs/inode.h> 60 #include <ufs/ufs/ufsmount.h> 61 #include <ufs/ufs/ufs_extern.h> 62 #include <ufs/ufs/ufs_bswap.h> 63 64 #include <ufs/ffs/fs.h> 65 #include <ufs/ffs/ffs_extern.h> 66 67 static int ffs_indirtrunc __P((struct inode *, ufs_daddr_t, ufs_daddr_t, 68 ufs_daddr_t, int, long *)); 69 70 /* 71 * Update the access, modified, and inode change times as specified 72 * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. 73 * The IN_MODIFIED flag is used to specify that the inode needs to be 74 * updated but that the times have already been set. The access 75 * and modified times are taken from the second and third parameters; 76 * the inode change time is always taken from the current time. If 77 * UPDATE_WAIT flag is set, or UPDATE_DIROP is set and we are not doing 78 * softupdates, then wait for the disk write of the inode to complete. 79 */ 80 81 int 82 ffs_update(v) 83 void *v; 84 { 85 struct vop_update_args /* { 86 struct vnode *a_vp; 87 struct timespec *a_access; 88 struct timespec *a_modify; 89 int a_flags; 90 } */ *ap = v; 91 struct fs *fs; 92 struct buf *bp; 93 struct inode *ip; 94 int error; 95 struct timespec ts; 96 caddr_t cp; 97 int waitfor, flags; 98 99 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) 100 return (0); 101 ip = VTOI(ap->a_vp); 102 TIMEVAL_TO_TIMESPEC(&time, &ts); 103 FFS_ITIMES(ip, 104 ap->a_access ? ap->a_access : &ts, 105 ap->a_modify ? ap->a_modify : &ts, &ts); 106 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED); 107 if (flags == 0) 108 return (0); 109 fs = ip->i_fs; 110 111 if ((flags & IN_MODIFIED) != 0 && 112 (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0) { 113 waitfor = ap->a_flags & UPDATE_WAIT; 114 if ((ap->a_flags & UPDATE_DIROP) && !DOINGSOFTDEP(ap->a_vp)) 115 waitfor |= UPDATE_WAIT; 116 } else 117 waitfor = 0; 118 119 /* 120 * Ensure that uid and gid are correct. This is a temporary 121 * fix until fsck has been changed to do the update. 122 */ 123 if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */ 124 ip->i_din.ffs_din.di_ouid = ip->i_ffs_uid; /* XXX */ 125 ip->i_din.ffs_din.di_ogid = ip->i_ffs_gid; /* XXX */ 126 } /* XXX */ 127 error = bread(ip->i_devvp, 128 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 129 (int)fs->fs_bsize, NOCRED, &bp); 130 if (error) { 131 brelse(bp); 132 return (error); 133 } 134 ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED); 135 if (DOINGSOFTDEP(ap->a_vp)) 136 softdep_update_inodeblock(ip, bp, waitfor); 137 else if (ip->i_ffs_effnlink != ip->i_ffs_nlink) 138 panic("ffs_update: bad link cnt"); 139 cp = (caddr_t)bp->b_data + 140 (ino_to_fsbo(fs, ip->i_number) * DINODE_SIZE); 141 #ifdef FFS_EI 142 if (UFS_FSNEEDSWAP(fs)) 143 ffs_dinode_swap(&ip->i_din.ffs_din, (struct dinode *)cp); 144 else 145 #endif 146 memcpy(cp, &ip->i_din.ffs_din, DINODE_SIZE); 147 if (waitfor) { 148 return (bwrite(bp)); 149 } else { 150 bdwrite(bp); 151 return (0); 152 } 153 } 154 155 #define SINGLE 0 /* index of single indirect block */ 156 #define DOUBLE 1 /* index of double indirect block */ 157 #define TRIPLE 2 /* index of triple indirect block */ 158 /* 159 * Truncate the inode oip to at most length size, freeing the 160 * disk blocks. 161 */ 162 int 163 ffs_truncate(v) 164 void *v; 165 { 166 struct vop_truncate_args /* { 167 struct vnode *a_vp; 168 off_t a_length; 169 int a_flags; 170 struct ucred *a_cred; 171 struct proc *a_p; 172 } */ *ap = v; 173 struct vnode *ovp = ap->a_vp; 174 struct genfs_node *gp = VTOG(ovp); 175 ufs_daddr_t lastblock; 176 struct inode *oip; 177 ufs_daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR]; 178 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR]; 179 off_t length = ap->a_length; 180 struct fs *fs; 181 int offset, size, level; 182 long count, nblocks, blocksreleased = 0; 183 int i, ioflag, aflag; 184 int error, allerror = 0; 185 off_t osize; 186 187 if (length < 0) 188 return (EINVAL); 189 oip = VTOI(ovp); 190 if (ovp->v_type == VLNK && 191 (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen || 192 (ovp->v_mount->mnt_maxsymlinklen == 0 && 193 oip->i_din.ffs_din.di_blocks == 0))) { 194 KDASSERT(length == 0); 195 memset(&oip->i_ffs_shortlink, 0, (size_t)oip->i_ffs_size); 196 oip->i_ffs_size = 0; 197 oip->i_flag |= IN_CHANGE | IN_UPDATE; 198 return (VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT)); 199 } 200 if (oip->i_ffs_size == length) { 201 oip->i_flag |= IN_CHANGE | IN_UPDATE; 202 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 203 } 204 #ifdef QUOTA 205 if ((error = getinoquota(oip)) != 0) 206 return (error); 207 #endif 208 fs = oip->i_fs; 209 if (length > fs->fs_maxfilesize) 210 return (EFBIG); 211 212 osize = oip->i_ffs_size; 213 ioflag = ap->a_flags; 214 aflag = ioflag & IO_SYNC ? B_SYNC : 0; 215 216 /* 217 * Lengthen the size of the file. We must ensure that the 218 * last byte of the file is allocated. Since the smallest 219 * value of osize is 0, length will be at least 1. 220 */ 221 222 if (osize < length) { 223 if (lblkno(fs, osize) < NDADDR && 224 lblkno(fs, osize) != lblkno(fs, length) && 225 blkroundup(fs, osize) != osize) { 226 error = ufs_balloc_range(ovp, osize, 227 blkroundup(fs, osize) - osize, ap->a_cred, aflag); 228 if (error) { 229 return error; 230 } 231 if (ioflag & IO_SYNC) { 232 ovp->v_size = blkroundup(fs, osize); 233 simple_lock(&ovp->v_interlock); 234 VOP_PUTPAGES(ovp, 235 trunc_page(osize & ~(fs->fs_bsize - 1)), 236 round_page(ovp->v_size), 237 PGO_CLEANIT | PGO_SYNCIO); 238 } 239 } 240 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 241 aflag); 242 if (error) { 243 (void) VOP_TRUNCATE(ovp, osize, ioflag & IO_SYNC, 244 ap->a_cred, ap->a_p); 245 return error; 246 } 247 uvm_vnp_setsize(ovp, length); 248 oip->i_flag |= IN_CHANGE | IN_UPDATE; 249 KASSERT(ovp->v_size == oip->i_ffs_size); 250 return (VOP_UPDATE(ovp, NULL, NULL, 1)); 251 } 252 253 /* 254 * When truncating a regular file down to a non-block-aligned size, 255 * we must zero the part of last block which is past the new EOF. 256 * We must synchronously flush the zeroed pages to disk 257 * since the new pages will be invalidated as soon as we 258 * inform the VM system of the new, smaller size. 259 * We must do this before acquiring the GLOCK, since fetching 260 * the pages will acquire the GLOCK internally. 261 * So there is a window where another thread could see a whole 262 * zeroed page past EOF, but that's life. 263 */ 264 265 offset = blkoff(fs, length); 266 if (ovp->v_type == VREG && length < osize && offset != 0) { 267 voff_t eoz; 268 269 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 270 aflag); 271 if (error) { 272 return error; 273 } 274 size = blksize(fs, oip, lblkno(fs, length)); 275 eoz = MIN(lblktosize(fs, lblkno(fs, length)) + size, osize); 276 uvm_vnp_zerorange(ovp, length, eoz - length); 277 simple_lock(&ovp->v_interlock); 278 error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz), 279 PGO_CLEANIT | PGO_DEACTIVATE | PGO_SYNCIO); 280 if (error) { 281 return error; 282 } 283 } 284 285 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 286 287 if (DOINGSOFTDEP(ovp)) { 288 if (length > 0) { 289 /* 290 * If a file is only partially truncated, then 291 * we have to clean up the data structures 292 * describing the allocation past the truncation 293 * point. Finding and deallocating those structures 294 * is a lot of work. Since partial truncation occurs 295 * rarely, we solve the problem by syncing the file 296 * so that it will have no data structures left. 297 */ 298 if ((error = VOP_FSYNC(ovp, ap->a_cred, FSYNC_WAIT, 299 0, 0, ap->a_p)) != 0) { 300 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 301 return (error); 302 if (oip->i_flag & IN_SPACECOUNTED) 303 fs->fs_pendingblocks -= oip->i_ffs_blocks; 304 } 305 } else { 306 uvm_vnp_setsize(ovp, length); 307 #ifdef QUOTA 308 (void) chkdq(oip, -oip->i_ffs_blocks, NOCRED, 0); 309 #endif 310 softdep_setup_freeblocks(oip, length); 311 (void) vinvalbuf(ovp, 0, ap->a_cred, ap->a_p, 0, 0); 312 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 313 oip->i_flag |= IN_CHANGE | IN_UPDATE; 314 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 315 } 316 } 317 oip->i_ffs_size = length; 318 uvm_vnp_setsize(ovp, length); 319 /* 320 * Calculate index into inode's block list of 321 * last direct and indirect blocks (if any) 322 * which we want to keep. Lastblock is -1 when 323 * the file is truncated to 0. 324 */ 325 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; 326 lastiblock[SINGLE] = lastblock - NDADDR; 327 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 328 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 329 nblocks = btodb(fs->fs_bsize); 330 /* 331 * Update file and block pointers on disk before we start freeing 332 * blocks. If we crash before free'ing blocks below, the blocks 333 * will be returned to the free list. lastiblock values are also 334 * normalized to -1 for calls to ffs_indirtrunc below. 335 */ 336 memcpy((caddr_t)oldblks, (caddr_t)&oip->i_ffs_db[0], sizeof oldblks); 337 for (level = TRIPLE; level >= SINGLE; level--) 338 if (lastiblock[level] < 0) { 339 oip->i_ffs_ib[level] = 0; 340 lastiblock[level] = -1; 341 } 342 for (i = NDADDR - 1; i > lastblock; i--) 343 oip->i_ffs_db[i] = 0; 344 oip->i_flag |= IN_CHANGE | IN_UPDATE; 345 error = VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT); 346 if (error && !allerror) 347 allerror = error; 348 349 /* 350 * Having written the new inode to disk, save its new configuration 351 * and put back the old block pointers long enough to process them. 352 * Note that we save the new block configuration so we can check it 353 * when we are done. 354 */ 355 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks); 356 memcpy((caddr_t)&oip->i_ffs_db[0], (caddr_t)oldblks, sizeof oldblks); 357 oip->i_ffs_size = osize; 358 error = vtruncbuf(ovp, lastblock + 1, 0, 0); 359 if (error && !allerror) 360 allerror = error; 361 362 /* 363 * Indirect blocks first. 364 */ 365 indir_lbn[SINGLE] = -NDADDR; 366 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 367 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 368 for (level = TRIPLE; level >= SINGLE; level--) { 369 bn = ufs_rw32(oip->i_ffs_ib[level], UFS_FSNEEDSWAP(fs)); 370 if (bn != 0) { 371 error = ffs_indirtrunc(oip, indir_lbn[level], 372 fsbtodb(fs, bn), lastiblock[level], level, &count); 373 if (error) 374 allerror = error; 375 blocksreleased += count; 376 if (lastiblock[level] < 0) { 377 oip->i_ffs_ib[level] = 0; 378 ffs_blkfree(oip, bn, fs->fs_bsize); 379 blocksreleased += nblocks; 380 } 381 } 382 if (lastiblock[level] >= 0) 383 goto done; 384 } 385 386 /* 387 * All whole direct blocks or frags. 388 */ 389 for (i = NDADDR - 1; i > lastblock; i--) { 390 long bsize; 391 392 bn = ufs_rw32(oip->i_ffs_db[i], UFS_FSNEEDSWAP(fs)); 393 if (bn == 0) 394 continue; 395 oip->i_ffs_db[i] = 0; 396 bsize = blksize(fs, oip, i); 397 ffs_blkfree(oip, bn, bsize); 398 blocksreleased += btodb(bsize); 399 } 400 if (lastblock < 0) 401 goto done; 402 403 /* 404 * Finally, look for a change in size of the 405 * last direct block; release any frags. 406 */ 407 bn = ufs_rw32(oip->i_ffs_db[lastblock], UFS_FSNEEDSWAP(fs)); 408 if (bn != 0) { 409 long oldspace, newspace; 410 411 /* 412 * Calculate amount of space we're giving 413 * back as old block size minus new block size. 414 */ 415 oldspace = blksize(fs, oip, lastblock); 416 oip->i_ffs_size = length; 417 newspace = blksize(fs, oip, lastblock); 418 if (newspace == 0) 419 panic("itrunc: newspace"); 420 if (oldspace - newspace > 0) { 421 /* 422 * Block number of space to be free'd is 423 * the old block # plus the number of frags 424 * required for the storage we're keeping. 425 */ 426 bn += numfrags(fs, newspace); 427 ffs_blkfree(oip, bn, oldspace - newspace); 428 blocksreleased += btodb(oldspace - newspace); 429 } 430 } 431 432 done: 433 #ifdef DIAGNOSTIC 434 for (level = SINGLE; level <= TRIPLE; level++) 435 if (newblks[NDADDR + level] != oip->i_ffs_ib[level]) 436 panic("itrunc1"); 437 for (i = 0; i < NDADDR; i++) 438 if (newblks[i] != oip->i_ffs_db[i]) 439 panic("itrunc2"); 440 if (length == 0 && 441 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 442 panic("itrunc3"); 443 #endif /* DIAGNOSTIC */ 444 /* 445 * Put back the real size. 446 */ 447 oip->i_ffs_size = length; 448 oip->i_ffs_blocks -= blocksreleased; 449 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 450 oip->i_flag |= IN_CHANGE; 451 #ifdef QUOTA 452 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 453 #endif 454 KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_ffs_size); 455 return (allerror); 456 } 457 458 /* 459 * Release blocks associated with the inode ip and stored in the indirect 460 * block bn. Blocks are free'd in LIFO order up to (but not including) 461 * lastbn. If level is greater than SINGLE, the block is an indirect block 462 * and recursive calls to indirtrunc must be used to cleanse other indirect 463 * blocks. 464 * 465 * NB: triple indirect blocks are untested. 466 */ 467 static int 468 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp) 469 struct inode *ip; 470 ufs_daddr_t lbn, lastbn; 471 ufs_daddr_t dbn; 472 int level; 473 long *countp; 474 { 475 int i; 476 struct buf *bp; 477 struct fs *fs = ip->i_fs; 478 ufs_daddr_t *bap; 479 struct vnode *vp; 480 ufs_daddr_t *copy = NULL, nb, nlbn, last; 481 long blkcount, factor; 482 int nblocks, blocksreleased = 0; 483 int error = 0, allerror = 0; 484 485 /* 486 * Calculate index in current block of last 487 * block to be kept. -1 indicates the entire 488 * block so we need not calculate the index. 489 */ 490 factor = 1; 491 for (i = SINGLE; i < level; i++) 492 factor *= NINDIR(fs); 493 last = lastbn; 494 if (lastbn > 0) 495 last /= factor; 496 nblocks = btodb(fs->fs_bsize); 497 /* 498 * Get buffer of block pointers, zero those entries corresponding 499 * to blocks to be free'd, and update on disk copy first. Since 500 * double(triple) indirect before single(double) indirect, calls 501 * to bmap on these blocks will fail. However, we already have 502 * the on disk address, so we have to set the b_blkno field 503 * explicitly instead of letting bread do everything for us. 504 */ 505 vp = ITOV(ip); 506 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0); 507 if (bp->b_flags & (B_DONE | B_DELWRI)) { 508 /* Braces must be here in case trace evaluates to nothing. */ 509 trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn); 510 } else { 511 trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn); 512 curproc->p_stats->p_ru.ru_inblock++; /* pay for read */ 513 bp->b_flags |= B_READ; 514 if (bp->b_bcount > bp->b_bufsize) 515 panic("ffs_indirtrunc: bad buffer size"); 516 bp->b_blkno = dbn; 517 VOP_STRATEGY(bp); 518 error = biowait(bp); 519 } 520 if (error) { 521 brelse(bp); 522 *countp = 0; 523 return (error); 524 } 525 526 bap = (ufs_daddr_t *)bp->b_data; 527 if (lastbn >= 0) { 528 copy = (ufs_daddr_t *) malloc(fs->fs_bsize, M_TEMP, M_WAITOK); 529 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->fs_bsize); 530 memset((caddr_t)&bap[last + 1], 0, 531 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 532 error = bwrite(bp); 533 if (error) 534 allerror = error; 535 bap = copy; 536 } 537 538 /* 539 * Recursively free totally unused blocks. 540 */ 541 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 542 i--, nlbn += factor) { 543 nb = ufs_rw32(bap[i], UFS_FSNEEDSWAP(fs)); 544 if (nb == 0) 545 continue; 546 if (level > SINGLE) { 547 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 548 (ufs_daddr_t)-1, level - 1, 549 &blkcount); 550 if (error) 551 allerror = error; 552 blocksreleased += blkcount; 553 } 554 ffs_blkfree(ip, nb, fs->fs_bsize); 555 blocksreleased += nblocks; 556 } 557 558 /* 559 * Recursively free last partial block. 560 */ 561 if (level > SINGLE && lastbn >= 0) { 562 last = lastbn % factor; 563 nb = ufs_rw32(bap[i], UFS_FSNEEDSWAP(fs)); 564 if (nb != 0) { 565 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), 566 last, level - 1, &blkcount); 567 if (error) 568 allerror = error; 569 blocksreleased += blkcount; 570 } 571 } 572 573 if (copy != NULL) { 574 FREE(copy, M_TEMP); 575 } else { 576 bp->b_flags |= B_INVAL; 577 brelse(bp); 578 } 579 580 *countp = blocksreleased; 581 return (allerror); 582 } 583