lfs_inode.c revision 1.51.2.1
1 /* $NetBSD: lfs_inode.c,v 1.51.2.1 2001/06/27 03:49:39 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 71 */ 72 73 #if defined(_KERNEL_OPT) 74 #include "opt_quota.h" 75 #endif 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/mount.h> 80 #include <sys/proc.h> 81 #include <sys/file.h> 82 #include <sys/buf.h> 83 #include <sys/vnode.h> 84 #include <sys/kernel.h> 85 #include <sys/malloc.h> 86 #include <sys/trace.h> 87 #include <sys/resourcevar.h> 88 89 #include <ufs/ufs/quota.h> 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/ufsmount.h> 92 #include <ufs/ufs/ufs_extern.h> 93 94 #include <ufs/lfs/lfs.h> 95 #include <ufs/lfs/lfs_extern.h> 96 97 extern int locked_queue_count; 98 extern long locked_queue_bytes; 99 100 static int lfs_update_seguse(struct lfs *, long, size_t); 101 static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t, 102 ufs_daddr_t, int, long *, long *, long *, size_t *, 103 struct proc *); 104 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 105 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 106 107 /* Search a block for a specific dinode. */ 108 struct dinode * 109 lfs_ifind(fs, ino, bp) 110 struct lfs *fs; 111 ino_t ino; 112 struct buf *bp; 113 { 114 struct dinode *dip = (struct dinode *)bp->b_data; 115 struct dinode *ldip, *fin; 116 117 if (fs->lfs_version == 1) 118 fin = dip + INOPB(fs); 119 else 120 fin = dip + INOPS(fs); 121 122 /* 123 * XXX we used to go from the top down here, presumably with the 124 * idea that the same inode could be written twice in the same 125 * block (which is not supposed to be true). 126 */ 127 for (ldip = dip; ldip < fin; ++ldip) 128 if (ldip->di_inumber == ino) 129 return (ldip); 130 131 printf("searched %d entries\n", (int)(fs->lfs_version == 1 ? 132 INOPB(fs) : 133 INOPS(fs))); 134 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 135 datosn(fs, fs->lfs_offset)); 136 printf("block is 0x%x (seg %d)\n", bp->b_blkno, 137 datosn(fs, bp->b_blkno)); 138 panic("lfs_ifind: dinode %u not found", ino); 139 /* NOTREACHED */ 140 } 141 142 int 143 lfs_update(v) 144 void *v; 145 { 146 struct vop_update_args /* { 147 struct vnode *a_vp; 148 struct timespec *a_access; 149 struct timespec *a_modify; 150 int a_flags; 151 } */ *ap = v; 152 struct inode *ip; 153 struct vnode *vp = ap->a_vp; 154 int oflag; 155 struct timespec ts; 156 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 157 158 if (vp->v_mount->mnt_flag & MNT_RDONLY) 159 return (0); 160 ip = VTOI(vp); 161 162 /* 163 * If we are called from vinvalbuf, and the file's blocks have 164 * already been scheduled for writing, but the writes have not 165 * yet completed, lfs_vflush will not be called, and vinvalbuf 166 * will cause a panic. So, we must wait until any pending write 167 * for our inode completes, if we are called with UPDATE_WAIT set. 168 */ 169 while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 170 WRITEINPROG(vp)) { 171 #ifdef DEBUG_LFS 172 printf("lfs_update: sleeping on inode %d (in-progress)\n", 173 ip->i_number); 174 #endif 175 tsleep(vp, (PRIBIO+1), "lfs_update", 0); 176 } 177 oflag = ip->i_flag; 178 TIMEVAL_TO_TIMESPEC(&time, &ts); 179 LFS_ITIMES(ip, 180 ap->a_access ? ap->a_access : &ts, 181 ap->a_modify ? ap->a_modify : &ts, &ts); 182 if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) { 183 return (0); 184 } 185 186 /* If sync, push back the vnode and any dirty blocks it may have. */ 187 if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) { 188 /* Avoid flushing VDIROP. */ 189 ++fs->lfs_diropwait; 190 while(vp->v_flag & VDIROP) { 191 #ifdef DEBUG_LFS 192 printf("lfs_update: sleeping on inode %d (dirops)\n", 193 ip->i_number); 194 printf("lfs_update: vflags 0x%lx, iflags 0x%x\n", 195 vp->v_flag, ip->i_flag); 196 #endif 197 if(fs->lfs_dirops == 0) 198 lfs_flush_fs(fs, SEGM_SYNC); 199 else 200 tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 201 0); 202 /* XXX KS - by falling out here, are we writing the vn 203 twice? */ 204 } 205 --fs->lfs_diropwait; 206 return lfs_vflush(vp); 207 } 208 return 0; 209 } 210 211 #define SINGLE 0 /* index of single indirect block */ 212 #define DOUBLE 1 /* index of double indirect block */ 213 #define TRIPLE 2 /* index of triple indirect block */ 214 /* 215 * Truncate the inode oip to at most length size, freeing the 216 * disk blocks. 217 */ 218 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 219 int 220 lfs_truncate(v) 221 void *v; 222 { 223 struct vop_truncate_args /* { 224 struct vnode *a_vp; 225 off_t a_length; 226 int a_flags; 227 struct ucred *a_cred; 228 struct proc *a_p; 229 } */ *ap = v; 230 struct vnode *ovp = ap->a_vp; 231 ufs_daddr_t lastblock; 232 struct inode *oip; 233 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 234 ufs_daddr_t newblks[NDADDR + NIADDR]; 235 off_t length = ap->a_length; 236 struct lfs *fs; 237 struct buf *bp; 238 int offset, size, level; 239 long count, rcount, nblocks, blocksreleased = 0, real_released = 0; 240 int i; 241 int aflags, error, allerror = 0; 242 off_t osize; 243 long lastseg; 244 size_t bc; 245 int obufsize, odb; 246 247 if (length < 0) 248 return (EINVAL); 249 oip = VTOI(ovp); 250 251 /* 252 * Just return and not update modification times. 253 */ 254 if (oip->i_ffs_size == length) 255 return (0); 256 257 if (ovp->v_type == VLNK && 258 (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen || 259 (ovp->v_mount->mnt_maxsymlinklen == 0 && 260 oip->i_din.ffs_din.di_blocks == 0))) { 261 #ifdef DIAGNOSTIC 262 if (length != 0) 263 panic("lfs_truncate: partial truncate of symlink"); 264 #endif 265 memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size); 266 oip->i_ffs_size = 0; 267 oip->i_flag |= IN_CHANGE | IN_UPDATE; 268 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 269 } 270 if (oip->i_ffs_size == length) { 271 oip->i_flag |= IN_CHANGE | IN_UPDATE; 272 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 273 } 274 #ifdef QUOTA 275 if ((error = getinoquota(oip)) != 0) 276 return (error); 277 #endif 278 fs = oip->i_lfs; 279 lfs_imtime(fs); 280 osize = oip->i_ffs_size; 281 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0; 282 283 /* 284 * Lengthen the size of the file. We must ensure that the 285 * last byte of the file is allocated. Since the smallest 286 * value of osize is 0, length will be at least 1. 287 */ 288 if (osize < length) { 289 if (length > fs->lfs_maxfilesize) 290 return (EFBIG); 291 aflags = B_CLRBUF; 292 if (ap->a_flags & IO_SYNC) 293 aflags |= B_SYNC; 294 error = lfs_reserve(fs, ovp, fsbtodb(fs, NIADDR + 2)); 295 if (error) 296 return (error); 297 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 298 lfs_reserve(fs, ovp, -fsbtodb(fs, NIADDR + 2)); 299 if (error) 300 return (error); 301 oip->i_ffs_size = length; 302 uvm_vnp_setsize(ovp, length); 303 (void) VOP_BWRITE(bp); 304 oip->i_flag |= IN_CHANGE | IN_UPDATE; 305 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 306 } 307 308 if ((error = lfs_reserve(fs, ovp, fsbtodb(fs, 2 * NIADDR + 3))) != 0) 309 return (error); 310 /* 311 * Make sure no writes to this inode can happen while we're 312 * truncating. Otherwise, blocks which are accounted for on the 313 * inode *and* which have been created for cleaning can coexist, 314 * and cause an overcounting. 315 * 316 * (We don't need to *hold* the seglock, though, because we already 317 * hold the inode lock; draining the seglock is sufficient.) 318 */ 319 if (ovp != fs->lfs_unlockvp) { 320 while(fs->lfs_seglock) { 321 tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0); 322 } 323 } 324 325 /* 326 * Shorten the size of the file. If the file is not being 327 * truncated to a block boundary, the contents of the 328 * partial block following the end of the file must be 329 * zero'ed in case it ever becomes accessible again because 330 * of subsequent file growth. Directories however are not 331 * zero'ed as they should grow back initialized to empty. 332 */ 333 offset = blkoff(fs, length); 334 lastseg = -1; 335 bc = 0; 336 if (offset == 0) { 337 oip->i_ffs_size = length; 338 } else { 339 lbn = lblkno(fs, length); 340 aflags = B_CLRBUF; 341 if (ap->a_flags & IO_SYNC) 342 aflags |= B_SYNC; 343 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 344 if (error) { 345 lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3)); 346 return (error); 347 } 348 obufsize = bp->b_bufsize; 349 odb = btodb(bp->b_bcount); 350 oip->i_ffs_size = length; 351 size = blksize(fs, oip, lbn); 352 if (ovp->v_type != VDIR) 353 memset((char *)bp->b_data + offset, 0, 354 (u_int)(size - offset)); 355 allocbuf(bp, size); 356 if (bp->b_flags & B_DELWRI) { 357 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 358 locked_queue_bytes -= obufsize - bp->b_bufsize; 359 fs->lfs_avail += odb - btodb(size); 360 } 361 (void) VOP_BWRITE(bp); 362 } 363 uvm_vnp_setsize(ovp, length); 364 /* 365 * Calculate index into inode's block list of 366 * last direct and indirect blocks (if any) 367 * which we want to keep. Lastblock is -1 when 368 * the file is truncated to 0. 369 */ 370 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 371 lastiblock[SINGLE] = lastblock - NDADDR; 372 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 373 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 374 nblocks = btodb(fs->lfs_bsize); 375 /* 376 * Record changed file and block pointers before we start 377 * freeing blocks. lastiblock values are also normalized to -1 378 * for calls to lfs_indirtrunc below. 379 */ 380 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks); 381 for (level = TRIPLE; level >= SINGLE; level--) 382 if (lastiblock[level] < 0) { 383 newblks[NDADDR+level] = 0; 384 lastiblock[level] = -1; 385 } 386 for (i = NDADDR - 1; i > lastblock; i--) 387 newblks[i] = 0; 388 389 oip->i_ffs_size = osize; 390 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 391 if (error && !allerror) 392 allerror = error; 393 394 /* 395 * Indirect blocks first. 396 */ 397 indir_lbn[SINGLE] = -NDADDR; 398 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 399 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 400 for (level = TRIPLE; level >= SINGLE; level--) { 401 bn = oip->i_ffs_ib[level]; 402 if (bn != 0) { 403 error = lfs_indirtrunc(oip, indir_lbn[level], 404 bn, lastiblock[level], 405 level, &count, &rcount, 406 &lastseg, &bc, ap->a_p); 407 if (error) 408 allerror = error; 409 real_released += rcount; 410 blocksreleased += count; 411 if (lastiblock[level] < 0) { 412 if (oip->i_ffs_ib[level] > 0) 413 real_released += nblocks; 414 blocksreleased += nblocks; 415 oip->i_ffs_ib[level] = 0; 416 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 417 } 418 } 419 if (lastiblock[level] >= 0) 420 goto done; 421 } 422 423 /* 424 * All whole direct blocks or frags. 425 */ 426 for (i = NDADDR - 1; i > lastblock; i--) { 427 long bsize; 428 429 bn = oip->i_ffs_db[i]; 430 if (bn == 0) 431 continue; 432 bsize = blksize(fs, oip, i); 433 if (oip->i_ffs_db[i] > 0) 434 real_released += btodb(bsize); 435 blocksreleased += btodb(bsize); 436 oip->i_ffs_db[i] = 0; 437 lfs_blkfree(fs, bn, bsize, &lastseg, &bc); 438 } 439 if (lastblock < 0) 440 goto done; 441 442 /* 443 * Finally, look for a change in size of the 444 * last direct block; release any frags. 445 */ 446 bn = oip->i_ffs_db[lastblock]; 447 if (bn != 0) { 448 long oldspace, newspace; 449 450 /* 451 * Calculate amount of space we're giving 452 * back as old block size minus new block size. 453 */ 454 oldspace = blksize(fs, oip, lastblock); 455 oip->i_ffs_size = length; 456 newspace = blksize(fs, oip, lastblock); 457 if (newspace == 0) 458 panic("itrunc: newspace"); 459 if (oldspace - newspace > 0) { 460 lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc); 461 if (bn > 0) 462 real_released += btodb(oldspace - newspace); 463 blocksreleased += btodb(oldspace - newspace); 464 } 465 } 466 467 done: 468 /* Finish segment accounting corrections */ 469 lfs_update_seguse(fs, lastseg, bc); 470 #ifdef DIAGNOSTIC 471 for (level = SINGLE; level <= TRIPLE; level++) 472 if (newblks[NDADDR + level] != oip->i_ffs_ib[level]) 473 panic("lfs itrunc1"); 474 for (i = 0; i < NDADDR; i++) 475 if (newblks[i] != oip->i_ffs_db[i]) 476 panic("lfs itrunc2"); 477 if (length == 0 && 478 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 479 panic("lfs itrunc3"); 480 #endif /* DIAGNOSTIC */ 481 /* 482 * Put back the real size. 483 */ 484 oip->i_ffs_size = length; 485 oip->i_lfs_effnblks -= blocksreleased; 486 oip->i_ffs_blocks -= real_released; 487 fs->lfs_bfree += blocksreleased; 488 #ifdef DIAGNOSTIC 489 if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) { 490 printf("lfs_truncate: truncate to 0 but %d blocks on inode\n", 491 oip->i_ffs_blocks); 492 panic("lfs_truncate: persistent blocks\n"); 493 } 494 #endif 495 oip->i_flag |= IN_CHANGE; 496 #ifdef QUOTA 497 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 498 #endif 499 lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3)); 500 return (allerror); 501 } 502 503 /* Update segment usage information when removing a block. */ 504 static int 505 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 506 size_t *num) 507 { 508 long seg; 509 int error = 0; 510 511 bsize = fragroundup(fs, bsize); 512 if (daddr > 0) { 513 if (*lastseg != (seg = datosn(fs, daddr))) { 514 error = lfs_update_seguse(fs, *lastseg, *num); 515 *num = bsize; 516 *lastseg = seg; 517 } else 518 *num += bsize; 519 } 520 return error; 521 } 522 523 /* Finish the accounting updates for a segment. */ 524 static int 525 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 526 { 527 SEGUSE *sup; 528 struct buf *bp; 529 530 if (lastseg < 0 || num == 0) 531 return 0; 532 533 534 LFS_SEGENTRY(sup, fs, lastseg, bp); 535 if (num > sup->su_nbytes) { 536 printf("lfs_truncate: segment %ld short by %ld\n", 537 lastseg, (long)num - sup->su_nbytes); 538 panic("lfs_truncate: negative bytes"); 539 sup->su_nbytes = num; 540 } 541 sup->su_nbytes -= num; 542 return (VOP_BWRITE(bp)); /* Ifile */ 543 } 544 545 /* 546 * Release blocks associated with the inode ip and stored in the indirect 547 * block bn. Blocks are free'd in LIFO order up to (but not including) 548 * lastbn. If level is greater than SINGLE, the block is an indirect block 549 * and recursive calls to indirtrunc must be used to cleanse other indirect 550 * blocks. 551 * 552 * NB: triple indirect blocks are untested. 553 */ 554 static int 555 lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn, 556 ufs_daddr_t lastbn, int level, long *countp, 557 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 558 { 559 int i; 560 struct buf *bp; 561 struct lfs *fs = ip->i_lfs; 562 ufs_daddr_t *bap; 563 struct vnode *vp; 564 ufs_daddr_t *copy = NULL, nb, nlbn, last; 565 long blkcount, rblkcount, factor; 566 int nblocks, blocksreleased = 0, real_released = 0; 567 int error = 0, allerror = 0; 568 569 /* 570 * Calculate index in current block of last 571 * block to be kept. -1 indicates the entire 572 * block so we need not calculate the index. 573 */ 574 factor = 1; 575 for (i = SINGLE; i < level; i++) 576 factor *= NINDIR(fs); 577 last = lastbn; 578 if (lastbn > 0) 579 last /= factor; 580 nblocks = btodb(fs->lfs_bsize); 581 /* 582 * Get buffer of block pointers, zero those entries corresponding 583 * to blocks to be free'd, and update on disk copy first. Since 584 * double(triple) indirect before single(double) indirect, calls 585 * to bmap on these blocks will fail. However, we already have 586 * the on disk address, so we have to set the b_blkno field 587 * explicitly instead of letting bread do everything for us. 588 */ 589 vp = ITOV(ip); 590 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 591 if (bp->b_flags & (B_DONE | B_DELWRI)) { 592 /* Braces must be here in case trace evaluates to nothing. */ 593 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 594 } else { 595 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 596 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 597 bp->b_flags |= B_READ; 598 if (bp->b_bcount > bp->b_bufsize) 599 panic("lfs_indirtrunc: bad buffer size"); 600 bp->b_blkno = dbn; 601 VOP_STRATEGY(bp); 602 error = biowait(bp); 603 } 604 if (error) { 605 brelse(bp); 606 *countp = *rcountp = 0; 607 return (error); 608 } 609 610 bap = (ufs_daddr_t *)bp->b_data; 611 if (lastbn >= 0) { 612 MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 613 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 614 memset((caddr_t)&bap[last + 1], 0, 615 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); 616 error = VOP_BWRITE(bp); 617 if (error) 618 allerror = error; 619 bap = copy; 620 } 621 622 /* 623 * Recursively free totally unused blocks. 624 */ 625 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 626 i--, nlbn += factor) { 627 nb = bap[i]; 628 if (nb == 0) 629 continue; 630 if (level > SINGLE) { 631 error = lfs_indirtrunc(ip, nlbn, nb, 632 (ufs_daddr_t)-1, level - 1, 633 &blkcount, &rblkcount, 634 lastsegp, bcp, p); 635 if (error) 636 allerror = error; 637 blocksreleased += blkcount; 638 real_released += rblkcount; 639 } 640 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 641 if (bap[i] > 0) 642 real_released += nblocks; 643 blocksreleased += nblocks; 644 } 645 646 /* 647 * Recursively free last partial block. 648 */ 649 if (level > SINGLE && lastbn >= 0) { 650 last = lastbn % factor; 651 nb = bap[i]; 652 if (nb != 0) { 653 error = lfs_indirtrunc(ip, nlbn, nb, 654 last, level - 1, &blkcount, 655 &rblkcount, lastsegp, bcp, p); 656 if (error) 657 allerror = error; 658 real_released += rblkcount; 659 blocksreleased += blkcount; 660 } 661 } 662 663 if (copy != NULL) { 664 FREE(copy, M_TEMP); 665 } else { 666 if (bp->b_flags & B_DELWRI) { 667 LFS_UNLOCK_BUF(bp); 668 fs->lfs_avail += btodb(bp->b_bcount); 669 wakeup(&fs->lfs_avail); 670 } 671 bp->b_flags |= B_INVAL; 672 brelse(bp); 673 } 674 675 *countp = blocksreleased; 676 *rcountp = real_released; 677 return (allerror); 678 } 679 680 /* 681 * Destroy any in core blocks past the truncation length. 682 * Inlined from vtruncbuf, so that lfs_avail could be updated. 683 */ 684 static int 685 lfs_vtruncbuf(vp, lbn, slpflag, slptimeo) 686 struct vnode *vp; 687 daddr_t lbn; 688 int slpflag, slptimeo; 689 { 690 struct buf *bp, *nbp; 691 int s, error; 692 struct lfs *fs; 693 694 fs = VTOI(vp)->i_lfs; 695 s = splbio(); 696 697 restart: 698 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 699 nbp = LIST_NEXT(bp, b_vnbufs); 700 if (bp->b_lblkno < lbn) 701 continue; 702 if (bp->b_flags & B_BUSY) { 703 bp->b_flags |= B_WANTED; 704 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 705 "lfs_vtruncbuf", slptimeo); 706 if (error) { 707 splx(s); 708 return (error); 709 } 710 goto restart; 711 } 712 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 713 if (bp->b_flags & B_DELWRI) { 714 bp->b_flags &= ~B_DELWRI; 715 fs->lfs_avail += btodb(bp->b_bcount); 716 wakeup(&fs->lfs_avail); 717 } 718 LFS_UNLOCK_BUF(bp); 719 brelse(bp); 720 } 721 722 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 723 nbp = LIST_NEXT(bp, b_vnbufs); 724 if (bp->b_lblkno < lbn) 725 continue; 726 if (bp->b_flags & B_BUSY) { 727 bp->b_flags |= B_WANTED; 728 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 729 "lfs_vtruncbuf", slptimeo); 730 if (error) { 731 splx(s); 732 return (error); 733 } 734 goto restart; 735 } 736 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 737 if (bp->b_flags & B_DELWRI) { 738 bp->b_flags &= ~B_DELWRI; 739 fs->lfs_avail += btodb(bp->b_bcount); 740 wakeup(&fs->lfs_avail); 741 } 742 LFS_UNLOCK_BUF(bp); 743 brelse(bp); 744 } 745 746 splx(s); 747 748 return (0); 749 } 750 751
Indexes created Thu Oct 02 14:10:14 GMT 2025