lfs_inode.c revision 1.78
1 /* $NetBSD: lfs_inode.c,v 1.78 2003/07/12 16:17:07 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 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 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.78 2003/07/12 16:17:07 yamt Exp $"); 75 76 #if defined(_KERNEL_OPT) 77 #include "opt_quota.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/mount.h> 83 #include <sys/proc.h> 84 #include <sys/file.h> 85 #include <sys/buf.h> 86 #include <sys/vnode.h> 87 #include <sys/kernel.h> 88 #include <sys/malloc.h> 89 #include <sys/trace.h> 90 #include <sys/resourcevar.h> 91 92 #include <ufs/ufs/quota.h> 93 #include <ufs/ufs/inode.h> 94 #include <ufs/ufs/ufsmount.h> 95 #include <ufs/ufs/ufs_extern.h> 96 97 #include <ufs/lfs/lfs.h> 98 #include <ufs/lfs/lfs_extern.h> 99 100 static int lfs_update_seguse(struct lfs *, long, size_t); 101 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 102 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 ufs1_dinode * 109 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 110 { 111 struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data; 112 struct ufs1_dinode *ldip, *fin; 113 114 #ifdef LFS_IFILE_FRAG_ADDRESSING 115 if (fs->lfs_version == 1) 116 fin = dip + INOPB(fs); 117 else 118 fin = dip + INOPF(fs); 119 #else 120 fin = dip + INOPB(fs); 121 #endif 122 123 /* 124 * Read the inode block backwards, since later versions of the 125 * inode will supercede earlier ones. Though it is unlikely, it is 126 * possible that the same inode will appear in the same inode block. 127 */ 128 for (ldip = fin - 1; ldip >= dip; --ldip) 129 if (ldip->di_inumber == ino) 130 return (ldip); 131 132 printf("searched %d entries\n", (int)(fin - dip)); 133 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 134 dtosn(fs, fs->lfs_offset)); 135 printf("block is 0x%llx (seg %lld)\n", 136 (unsigned long long)dbtofsb(fs, bp->b_blkno), 137 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 138 139 return NULL; 140 } 141 142 int 143 lfs_update(void *v) 144 { 145 struct vop_update_args /* { 146 struct vnode *a_vp; 147 struct timespec *a_access; 148 struct timespec *a_modify; 149 int a_flags; 150 } */ *ap = v; 151 struct inode *ip; 152 struct vnode *vp = ap->a_vp; 153 struct timespec ts; 154 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 155 int s; 156 157 if (vp->v_mount->mnt_flag & MNT_RDONLY) 158 return (0); 159 ip = VTOI(vp); 160 161 /* 162 * If we are called from vinvalbuf, and the file's blocks have 163 * already been scheduled for writing, but the writes have not 164 * yet completed, lfs_vflush will not be called, and vinvalbuf 165 * will cause a panic. So, we must wait until any pending write 166 * for our inode completes, if we are called with UPDATE_WAIT set. 167 */ 168 s = splbio(); 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 splx(s); 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%x, 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 */ 219 220 int 221 lfs_truncate(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 struct genfs_node *gp = VTOG(ovp); 232 daddr_t lastblock; 233 struct inode *oip; 234 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 235 /* XXX ondisk32 */ 236 int32_t newblks[NDADDR + NIADDR]; 237 off_t length = ap->a_length; 238 struct lfs *fs; 239 struct buf *bp; 240 int offset, size, level; 241 long count, rcount, nblocks, blocksreleased = 0, real_released = 0; 242 int i; 243 int aflags, error, allerror = 0; 244 off_t osize; 245 voff_t eoz; 246 long lastseg; 247 size_t bc; 248 int obufsize, odb; 249 int usepc; 250 251 if (length < 0) 252 return (EINVAL); 253 oip = VTOI(ovp); 254 255 /* 256 * Just return and not update modification times. 257 */ 258 if (oip->i_size == length) 259 return (0); 260 261 if (ovp->v_type == VLNK && 262 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || 263 (ovp->v_mount->mnt_maxsymlinklen == 0 && 264 oip->i_ffs1_blocks == 0))) { 265 #ifdef DIAGNOSTIC 266 if (length != 0) 267 panic("lfs_truncate: partial truncate of symlink"); 268 #endif 269 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 270 oip->i_size = oip->i_ffs1_size = 0; 271 oip->i_flag |= IN_CHANGE | IN_UPDATE; 272 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 273 } 274 if (oip->i_size == length) { 275 oip->i_flag |= IN_CHANGE | IN_UPDATE; 276 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 277 } 278 #ifdef QUOTA 279 if ((error = getinoquota(oip)) != 0) 280 return (error); 281 #endif 282 fs = oip->i_lfs; 283 lfs_imtime(fs); 284 osize = oip->i_size; 285 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 286 287 /* 288 * Lengthen the size of the file. We must ensure that the 289 * last byte of the file is allocated. Since the smallest 290 * value of osize is 0, length will be at least 1. 291 */ 292 if (osize < length) { 293 if (length > fs->lfs_maxfilesize) 294 return (EFBIG); 295 aflags = B_CLRBUF; 296 if (ap->a_flags & IO_SYNC) 297 aflags |= B_SYNC; 298 if (usepc) { 299 if (lblkno(fs, osize) < NDADDR && 300 lblkno(fs, osize) != lblkno(fs, length) && 301 blkroundup(fs, osize) != osize) { 302 error = ufs_balloc_range(ovp, osize, 303 blkroundup(fs, osize) - 304 osize, ap->a_cred, 305 aflags); 306 if (error) { 307 return error; 308 } 309 if (ap->a_flags & IO_SYNC) { 310 ovp->v_size = blkroundup(fs, osize); 311 simple_lock(&ovp->v_interlock); 312 VOP_PUTPAGES(ovp, 313 trunc_page(osize & fs->lfs_bmask), 314 round_page(ovp->v_size), 315 PGO_CLEANIT | PGO_SYNCIO); 316 } 317 } 318 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 319 aflags); 320 if (error) { 321 (void) VOP_TRUNCATE(ovp, osize, 322 ap->a_flags & IO_SYNC, 323 ap->a_cred, ap->a_p); 324 return error; 325 } 326 uvm_vnp_setsize(ovp, length); 327 oip->i_flag |= IN_CHANGE | IN_UPDATE; 328 KASSERT(ovp->v_size == oip->i_size); 329 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 330 } else { 331 error = lfs_reserve(fs, ovp, NULL, 332 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 333 if (error) 334 return (error); 335 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, 336 aflags, &bp); 337 lfs_reserve(fs, ovp, NULL, 338 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 339 if (error) 340 return (error); 341 oip->i_ffs1_size = oip->i_size = length; 342 uvm_vnp_setsize(ovp, length); 343 (void) VOP_BWRITE(bp); 344 oip->i_flag |= IN_CHANGE | IN_UPDATE; 345 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 346 } 347 } 348 349 if ((error = lfs_reserve(fs, ovp, NULL, 350 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 351 return (error); 352 353 /* 354 * Shorten the size of the file. If the file is not being 355 * truncated to a block boundary, the contents of the 356 * partial block following the end of the file must be 357 * zero'ed in case it ever becomes accessible again because 358 * of subsequent file growth. Directories however are not 359 * zero'ed as they should grow back initialized to empty. 360 */ 361 offset = blkoff(fs, length); 362 lastseg = -1; 363 bc = 0; 364 365 lfs_seglock(fs, SEGM_PROT); 366 if (offset == 0) { 367 oip->i_size = oip->i_ffs1_size = length; 368 } else if (!usepc) { 369 lbn = lblkno(fs, length); 370 aflags = B_CLRBUF; 371 if (ap->a_flags & IO_SYNC) 372 aflags |= B_SYNC; 373 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 374 if (error) { 375 lfs_reserve(fs, ovp, NULL, 376 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 377 goto errout; 378 } 379 obufsize = bp->b_bufsize; 380 odb = btofsb(fs, bp->b_bcount); 381 oip->i_size = oip->i_ffs1_size = length; 382 size = blksize(fs, oip, lbn); 383 if (ovp->v_type != VDIR) 384 memset((char *)bp->b_data + offset, 0, 385 (u_int)(size - offset)); 386 allocbuf(bp, size); 387 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 388 locked_queue_bytes -= obufsize - bp->b_bufsize; 389 if (bp->b_flags & B_DELWRI) 390 fs->lfs_avail += odb - btofsb(fs, size); 391 (void) VOP_BWRITE(bp); 392 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 393 /* 394 * When truncating a regular file down to a non-block-aligned 395 * size, we must zero the part of last block which is past 396 * the new EOF. We must synchronously flush the zeroed pages 397 * to disk since the new pages will be invalidated as soon 398 * as we inform the VM system of the new, smaller size. 399 * We must do this before acquiring the GLOCK, since fetching 400 * the pages will acquire the GLOCK internally. 401 * So there is a window where another thread could see a whole 402 * zeroed page past EOF, but that's life. 403 */ 404 aflags = ap->a_flags & IO_SYNC ? B_SYNC : 0; 405 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 406 aflags); 407 if (error) { 408 goto errout; 409 } 410 eoz = blkroundup(fs, length); 411 uvm_vnp_zerorange(ovp, length, eoz - length); 412 simple_lock(&ovp->v_interlock); 413 error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz), 414 PGO_CLEANIT | PGO_DEACTIVATE | (aflags ? PGO_SYNCIO : 0)); 415 if (error) { 416 goto errout; 417 } 418 } 419 420 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 421 422 oip->i_size = oip->i_ffs1_size = length; 423 uvm_vnp_setsize(ovp, length); 424 /* 425 * Calculate index into inode's block list of 426 * last direct and indirect blocks (if any) 427 * which we want to keep. Lastblock is -1 when 428 * the file is truncated to 0. 429 */ 430 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 431 lastiblock[SINGLE] = lastblock - NDADDR; 432 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 433 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 434 nblocks = btofsb(fs, fs->lfs_bsize); 435 /* 436 * Record changed file and block pointers before we start 437 * freeing blocks. lastiblock values are also normalized to -1 438 * for calls to lfs_indirtrunc below. 439 */ 440 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 441 for (level = TRIPLE; level >= SINGLE; level--) 442 if (lastiblock[level] < 0) { 443 newblks[NDADDR+level] = 0; 444 lastiblock[level] = -1; 445 } 446 for (i = NDADDR - 1; i > lastblock; i--) 447 newblks[i] = 0; 448 449 oip->i_size = oip->i_ffs1_size = osize; 450 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 451 if (error && !allerror) 452 allerror = error; 453 454 /* 455 * Indirect blocks first. 456 */ 457 indir_lbn[SINGLE] = -NDADDR; 458 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 459 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 460 for (level = TRIPLE; level >= SINGLE; level--) { 461 bn = oip->i_ffs1_ib[level]; 462 if (bn != 0) { 463 error = lfs_indirtrunc(oip, indir_lbn[level], 464 bn, lastiblock[level], 465 level, &count, &rcount, 466 &lastseg, &bc, ap->a_p); 467 if (error) 468 allerror = error; 469 real_released += rcount; 470 blocksreleased += count; 471 if (lastiblock[level] < 0) { 472 if (oip->i_ffs1_ib[level] > 0) 473 real_released += nblocks; 474 blocksreleased += nblocks; 475 oip->i_ffs1_ib[level] = 0; 476 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 477 } 478 } 479 if (lastiblock[level] >= 0) 480 goto done; 481 } 482 483 /* 484 * All whole direct blocks or frags. 485 */ 486 for (i = NDADDR - 1; i > lastblock; i--) { 487 long bsize, obsize; 488 489 bn = oip->i_ffs1_db[i]; 490 if (bn == 0) 491 continue; 492 bsize = blksize(fs, oip, i); 493 if (oip->i_ffs1_db[i] > 0) { 494 /* Check for fragment size changes */ 495 obsize = oip->i_lfs_fragsize[i]; 496 real_released += btofsb(fs, obsize); 497 oip->i_lfs_fragsize[i] = 0; 498 } else 499 obsize = 0; 500 blocksreleased += btofsb(fs, bsize); 501 oip->i_ffs1_db[i] = 0; 502 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 503 } 504 if (lastblock < 0) 505 goto done; 506 507 /* 508 * Finally, look for a change in size of the 509 * last direct block; release any frags. 510 */ 511 bn = oip->i_ffs1_db[lastblock]; 512 if (bn != 0) { 513 long oldspace, newspace; 514 #if 0 515 long olddspace; 516 #endif 517 518 /* 519 * Calculate amount of space we're giving 520 * back as old block size minus new block size. 521 */ 522 oldspace = blksize(fs, oip, lastblock); 523 #if 0 524 olddspace = oip->i_lfs_fragsize[lastblock]; 525 #endif 526 527 oip->i_size = oip->i_ffs1_size = length; 528 newspace = blksize(fs, oip, lastblock); 529 if (newspace == 0) 530 panic("itrunc: newspace"); 531 if (oldspace - newspace > 0) { 532 blocksreleased += btofsb(fs, oldspace - newspace); 533 } 534 #if 0 535 if (bn > 0 && olddspace - newspace > 0) { 536 /* No segment accounting here, just vnode */ 537 real_released += btofsb(fs, olddspace - newspace); 538 } 539 #endif 540 } 541 542 done: 543 /* Finish segment accounting corrections */ 544 lfs_update_seguse(fs, lastseg, bc); 545 #ifdef DIAGNOSTIC 546 for (level = SINGLE; level <= TRIPLE; level++) 547 if ((newblks[NDADDR + level] == 0) != 548 (oip->i_ffs1_ib[level]) == 0) { 549 panic("lfs itrunc1"); 550 } 551 for (i = 0; i < NDADDR; i++) 552 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 553 panic("lfs itrunc2"); 554 } 555 if (length == 0 && 556 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 557 panic("lfs itrunc3"); 558 #endif /* DIAGNOSTIC */ 559 /* 560 * Put back the real size. 561 */ 562 oip->i_size = oip->i_ffs1_size = length; 563 oip->i_lfs_effnblks -= blocksreleased; 564 oip->i_ffs1_blocks -= real_released; 565 fs->lfs_bfree += blocksreleased; 566 #ifdef DIAGNOSTIC 567 if (oip->i_size == 0 && oip->i_ffs1_blocks != 0) { 568 printf("lfs_truncate: truncate to 0 but %d blocks on inode\n", 569 oip->i_ffs1_blocks); 570 panic("lfs_truncate: persistent blocks"); 571 } 572 #endif 573 oip->i_flag |= IN_CHANGE; 574 #ifdef QUOTA 575 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 576 #endif 577 lfs_reserve(fs, ovp, NULL, 578 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 579 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 580 errout: 581 lfs_segunlock(fs); 582 return (allerror ? allerror : error); 583 } 584 585 /* Update segment usage information when removing a block. */ 586 static int 587 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 588 size_t *num) 589 { 590 long seg; 591 int error = 0; 592 593 bsize = fragroundup(fs, bsize); 594 if (daddr > 0) { 595 if (*lastseg != (seg = dtosn(fs, daddr))) { 596 error = lfs_update_seguse(fs, *lastseg, *num); 597 *num = bsize; 598 *lastseg = seg; 599 } else 600 *num += bsize; 601 } 602 return error; 603 } 604 605 /* Finish the accounting updates for a segment. */ 606 static int 607 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 608 { 609 SEGUSE *sup; 610 struct buf *bp; 611 612 if (lastseg < 0 || num == 0) 613 return 0; 614 615 LFS_SEGENTRY(sup, fs, lastseg, bp); 616 if (num > sup->su_nbytes) { 617 printf("lfs_truncate: segment %ld short by %ld\n", 618 lastseg, (long)num - sup->su_nbytes); 619 panic("lfs_truncate: negative bytes"); 620 sup->su_nbytes = num; 621 } 622 sup->su_nbytes -= num; 623 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 624 625 return 0; 626 } 627 628 /* 629 * Release blocks associated with the inode ip and stored in the indirect 630 * block bn. Blocks are free'd in LIFO order up to (but not including) 631 * lastbn. If level is greater than SINGLE, the block is an indirect block 632 * and recursive calls to indirtrunc must be used to cleanse other indirect 633 * blocks. 634 * 635 * NB: triple indirect blocks are untested. 636 */ 637 static int 638 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 639 daddr_t lastbn, int level, long *countp, 640 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 641 { 642 int i; 643 struct buf *bp; 644 struct lfs *fs = ip->i_lfs; 645 int32_t *bap; /* XXX ondisk32 */ 646 struct vnode *vp; 647 daddr_t nb, nlbn, last; 648 int32_t *copy = NULL; /* XXX ondisk32 */ 649 long blkcount, rblkcount, factor; 650 int nblocks, blocksreleased = 0, real_released = 0; 651 int error = 0, allerror = 0; 652 653 /* 654 * Calculate index in current block of last 655 * block to be kept. -1 indicates the entire 656 * block so we need not calculate the index. 657 */ 658 factor = 1; 659 for (i = SINGLE; i < level; i++) 660 factor *= NINDIR(fs); 661 last = lastbn; 662 if (lastbn > 0) 663 last /= factor; 664 nblocks = btofsb(fs, fs->lfs_bsize); 665 /* 666 * Get buffer of block pointers, zero those entries corresponding 667 * to blocks to be free'd, and update on disk copy first. Since 668 * double(triple) indirect before single(double) indirect, calls 669 * to bmap on these blocks will fail. However, we already have 670 * the on disk address, so we have to set the b_blkno field 671 * explicitly instead of letting bread do everything for us. 672 */ 673 vp = ITOV(ip); 674 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 675 if (bp->b_flags & (B_DONE | B_DELWRI)) { 676 /* Braces must be here in case trace evaluates to nothing. */ 677 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 678 } else { 679 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 680 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 681 bp->b_flags |= B_READ; 682 if (bp->b_bcount > bp->b_bufsize) 683 panic("lfs_indirtrunc: bad buffer size"); 684 bp->b_blkno = fsbtodb(fs, dbn); 685 VOP_STRATEGY(bp); 686 error = biowait(bp); 687 } 688 if (error) { 689 brelse(bp); 690 *countp = *rcountp = 0; 691 return (error); 692 } 693 694 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 695 if (lastbn >= 0) { 696 MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 697 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 698 memset((caddr_t)&bap[last + 1], 0, 699 /* XXX ondisk32 */ 700 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 701 error = VOP_BWRITE(bp); 702 if (error) 703 allerror = error; 704 bap = copy; 705 } 706 707 /* 708 * Recursively free totally unused blocks. 709 */ 710 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 711 i--, nlbn += factor) { 712 nb = bap[i]; 713 if (nb == 0) 714 continue; 715 if (level > SINGLE) { 716 error = lfs_indirtrunc(ip, nlbn, nb, 717 (daddr_t)-1, level - 1, 718 &blkcount, &rblkcount, 719 lastsegp, bcp, p); 720 if (error) 721 allerror = error; 722 blocksreleased += blkcount; 723 real_released += rblkcount; 724 } 725 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 726 if (bap[i] > 0) 727 real_released += nblocks; 728 blocksreleased += nblocks; 729 } 730 731 /* 732 * Recursively free last partial block. 733 */ 734 if (level > SINGLE && lastbn >= 0) { 735 last = lastbn % factor; 736 nb = bap[i]; 737 if (nb != 0) { 738 error = lfs_indirtrunc(ip, nlbn, nb, 739 last, level - 1, &blkcount, 740 &rblkcount, lastsegp, bcp, p); 741 if (error) 742 allerror = error; 743 real_released += rblkcount; 744 blocksreleased += blkcount; 745 } 746 } 747 748 if (copy != NULL) { 749 FREE(copy, M_TEMP); 750 } else { 751 if (bp->b_flags & B_DELWRI) { 752 LFS_UNLOCK_BUF(bp); 753 fs->lfs_avail += btofsb(fs, bp->b_bcount); 754 wakeup(&fs->lfs_avail); 755 } 756 bp->b_flags |= B_INVAL; 757 brelse(bp); 758 } 759 760 *countp = blocksreleased; 761 *rcountp = real_released; 762 return (allerror); 763 } 764 765 /* 766 * Destroy any in core blocks past the truncation length. 767 * Inlined from vtruncbuf, so that lfs_avail could be updated. 768 * We take the seglock to prevent cleaning from occurring while we are 769 * invalidating blocks. 770 */ 771 static int 772 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 773 { 774 struct buf *bp, *nbp; 775 int s, error; 776 struct lfs *fs; 777 voff_t off; 778 779 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 780 simple_lock(&vp->v_interlock); 781 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 782 if (error) { 783 return error; 784 } 785 786 fs = VTOI(vp)->i_lfs; 787 s = splbio(); 788 789 restart: 790 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 791 nbp = LIST_NEXT(bp, b_vnbufs); 792 if (bp->b_lblkno < lbn) 793 continue; 794 simple_lock(&bp->b_interlock); 795 if (bp->b_flags & B_BUSY) { 796 bp->b_flags |= B_WANTED; 797 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 798 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 799 if (error) { 800 splx(s); 801 return (error); 802 } 803 goto restart; 804 } 805 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 806 if (bp->b_flags & B_DELWRI) { 807 bp->b_flags &= ~B_DELWRI; 808 fs->lfs_avail += btofsb(fs, bp->b_bcount); 809 wakeup(&fs->lfs_avail); 810 } 811 LFS_UNLOCK_BUF(bp); 812 simple_unlock(&bp->b_interlock); 813 brelse(bp); 814 } 815 816 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 817 nbp = LIST_NEXT(bp, b_vnbufs); 818 if (bp->b_lblkno < lbn) 819 continue; 820 simple_lock(&bp->b_interlock); 821 if (bp->b_flags & B_BUSY) { 822 bp->b_flags |= B_WANTED; 823 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 824 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 825 if (error) { 826 splx(s); 827 return (error); 828 } 829 goto restart; 830 } 831 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 832 if (bp->b_flags & B_DELWRI) { 833 bp->b_flags &= ~B_DELWRI; 834 fs->lfs_avail += btofsb(fs, bp->b_bcount); 835 wakeup(&fs->lfs_avail); 836 } 837 LFS_UNLOCK_BUF(bp); 838 simple_unlock(&bp->b_interlock); 839 brelse(bp); 840 } 841 842 splx(s); 843 844 return (0); 845 } 846 847
Indexes created Thu Oct 02 14:10:14 GMT 2025