lfs_inode.c revision 1.100.10.3
1 /* $NetBSD: lfs_inode.c,v 1.100.10.3 2006/05/06 23:32:58 christos 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. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.100.10.3 2006/05/06 23:32:58 christos Exp $"); 71 72 #if defined(_KERNEL_OPT) 73 #include "opt_quota.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/mount.h> 79 #include <sys/proc.h> 80 #include <sys/file.h> 81 #include <sys/buf.h> 82 #include <sys/vnode.h> 83 #include <sys/kernel.h> 84 #include <sys/trace.h> 85 #include <sys/resourcevar.h> 86 #include <sys/kauth.h> 87 88 #include <ufs/ufs/quota.h> 89 #include <ufs/ufs/inode.h> 90 #include <ufs/ufs/ufsmount.h> 91 #include <ufs/ufs/ufs_extern.h> 92 93 #include <ufs/lfs/lfs.h> 94 #include <ufs/lfs/lfs_extern.h> 95 96 static int lfs_update_seguse(struct lfs *, long, size_t); 97 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 98 daddr_t, int, long *, long *, long *, size_t *, 99 struct lwp *); 100 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *); 101 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 102 103 /* Search a block for a specific dinode. */ 104 struct ufs1_dinode * 105 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 106 { 107 struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data; 108 struct ufs1_dinode *ldip, *fin; 109 110 ASSERT_NO_SEGLOCK(fs); 111 /* 112 * Read the inode block backwards, since later versions of the 113 * inode will supercede earlier ones. Though it is unlikely, it is 114 * possible that the same inode will appear in the same inode block. 115 */ 116 fin = dip + INOPB(fs); 117 for (ldip = fin - 1; ldip >= dip; --ldip) 118 if (ldip->di_inumber == ino) 119 return (ldip); 120 121 printf("searched %d entries\n", (int)(fin - dip)); 122 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 123 dtosn(fs, fs->lfs_offset)); 124 printf("block is 0x%llx (seg %lld)\n", 125 (unsigned long long)dbtofsb(fs, bp->b_blkno), 126 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 127 128 return NULL; 129 } 130 131 int 132 lfs_update(struct vnode *vp, const struct timespec *acc, 133 const struct timespec *mod, int updflags) 134 { 135 struct inode *ip; 136 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 137 int s; 138 int flags; 139 140 ASSERT_NO_SEGLOCK(fs); 141 if (vp->v_mount->mnt_flag & MNT_RDONLY) 142 return (0); 143 ip = VTOI(vp); 144 145 /* 146 * If we are called from vinvalbuf, and the file's blocks have 147 * already been scheduled for writing, but the writes have not 148 * yet completed, lfs_vflush will not be called, and vinvalbuf 149 * will cause a panic. So, we must wait until any pending write 150 * for our inode completes, if we are called with UPDATE_WAIT set. 151 */ 152 s = splbio(); 153 simple_lock(&vp->v_interlock); 154 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 155 WRITEINPROG(vp)) { 156 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 157 " (in progress)\n", ip->i_number)); 158 ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 159 } 160 simple_unlock(&vp->v_interlock); 161 splx(s); 162 LFS_ITIMES(ip, acc, mod, NULL); 163 if (updflags & UPDATE_CLOSE) 164 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 165 else 166 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 167 if (flags == 0) 168 return (0); 169 170 /* If sync, push back the vnode and any dirty blocks it may have. */ 171 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 172 /* Avoid flushing VDIROP. */ 173 simple_lock(&fs->lfs_interlock); 174 ++fs->lfs_diropwait; 175 while (vp->v_flag & VDIROP) { 176 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 177 " (dirops)\n", ip->i_number)); 178 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 179 " 0x%x\n", vp->v_flag, ip->i_flag)); 180 if (fs->lfs_dirops == 0) 181 lfs_flush_fs(fs, SEGM_SYNC); 182 else 183 ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 184 0, &fs->lfs_interlock); 185 /* XXX KS - by falling out here, are we writing the vn 186 twice? */ 187 } 188 --fs->lfs_diropwait; 189 simple_unlock(&fs->lfs_interlock); 190 return lfs_vflush(vp); 191 } 192 return 0; 193 } 194 195 #define SINGLE 0 /* index of single indirect block */ 196 #define DOUBLE 1 /* index of double indirect block */ 197 #define TRIPLE 2 /* index of triple indirect block */ 198 /* 199 * Truncate the inode oip to at most length size, freeing the 200 * disk blocks. 201 */ 202 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */ 203 204 int 205 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, 206 kauth_cred_t cred, struct lwp *l) 207 { 208 struct genfs_node *gp = VTOG(ovp); 209 daddr_t lastblock; 210 struct inode *oip = VTOI(ovp); 211 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 212 /* XXX ondisk32 */ 213 int32_t newblks[NDADDR + NIADDR]; 214 struct lfs *fs; 215 struct buf *bp; 216 int offset, size, level; 217 long count, rcount, blocksreleased = 0, real_released = 0; 218 int i, nblocks; 219 int aflags, error, allerror = 0; 220 off_t osize; 221 long lastseg; 222 size_t bc; 223 int obufsize, odb; 224 int usepc; 225 struct ufsmount *ump = oip->i_ump; 226 227 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 228 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 229 KASSERT(oip->i_size == 0); 230 return 0; 231 } 232 233 if (length < 0) 234 return (EINVAL); 235 236 /* 237 * Just return and not update modification times. 238 */ 239 if (oip->i_size == length) 240 return (0); 241 242 if (ovp->v_type == VLNK && 243 (oip->i_size < ump->um_maxsymlinklen || 244 (ump->um_maxsymlinklen == 0 && 245 oip->i_ffs1_blocks == 0))) { 246 #ifdef DIAGNOSTIC 247 if (length != 0) 248 panic("lfs_truncate: partial truncate of symlink"); 249 #endif 250 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 251 oip->i_size = oip->i_ffs1_size = 0; 252 oip->i_flag |= IN_CHANGE | IN_UPDATE; 253 return (lfs_update(ovp, NULL, NULL, 0)); 254 } 255 if (oip->i_size == length) { 256 oip->i_flag |= IN_CHANGE | IN_UPDATE; 257 return (lfs_update(ovp, NULL, NULL, 0)); 258 } 259 #ifdef QUOTA 260 if ((error = getinoquota(oip)) != 0) 261 return (error); 262 #endif 263 fs = oip->i_lfs; 264 lfs_imtime(fs); 265 osize = oip->i_size; 266 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 267 268 ASSERT_NO_SEGLOCK(fs); 269 /* 270 * Lengthen the size of the file. We must ensure that the 271 * last byte of the file is allocated. Since the smallest 272 * value of osize is 0, length will be at least 1. 273 */ 274 if (osize < length) { 275 if (length > ump->um_maxfilesize) 276 return (EFBIG); 277 aflags = B_CLRBUF; 278 if (ioflag & IO_SYNC) 279 aflags |= B_SYNC; 280 if (usepc) { 281 if (lblkno(fs, osize) < NDADDR && 282 lblkno(fs, osize) != lblkno(fs, length) && 283 blkroundup(fs, osize) != osize) { 284 off_t eob; 285 286 eob = blkroundup(fs, osize); 287 error = ufs_balloc_range(ovp, osize, 288 eob - osize, cred, aflags); 289 if (error) 290 return error; 291 if (ioflag & IO_SYNC) { 292 ovp->v_size = eob; 293 simple_lock(&ovp->v_interlock); 294 VOP_PUTPAGES(ovp, 295 trunc_page(osize & fs->lfs_bmask), 296 round_page(eob), 297 PGO_CLEANIT | PGO_SYNCIO); 298 } 299 } 300 error = ufs_balloc_range(ovp, length - 1, 1, cred, 301 aflags); 302 if (error) { 303 (void) lfs_truncate(ovp, osize, 304 ioflag & IO_SYNC, cred, l); 305 return error; 306 } 307 uvm_vnp_setsize(ovp, length); 308 oip->i_flag |= IN_CHANGE | IN_UPDATE; 309 KASSERT(ovp->v_size == oip->i_size); 310 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 311 return (lfs_update(ovp, NULL, NULL, 0)); 312 } else { 313 error = lfs_reserve(fs, ovp, NULL, 314 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 315 if (error) 316 return (error); 317 error = lfs_balloc(ovp, length - 1, 1, cred, 318 aflags, &bp); 319 lfs_reserve(fs, ovp, NULL, 320 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 321 if (error) 322 return (error); 323 oip->i_ffs1_size = oip->i_size = length; 324 uvm_vnp_setsize(ovp, length); 325 (void) VOP_BWRITE(bp); 326 oip->i_flag |= IN_CHANGE | IN_UPDATE; 327 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 328 return (lfs_update(ovp, NULL, NULL, 0)); 329 } 330 } 331 332 if ((error = lfs_reserve(fs, ovp, NULL, 333 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 334 return (error); 335 336 /* 337 * Shorten the size of the file. If the file is not being 338 * truncated to a block boundary, the contents of the 339 * partial block following the end of the file must be 340 * zero'ed in case it ever becomes accessible again because 341 * of subsequent file growth. Directories however are not 342 * zero'ed as they should grow back initialized to empty. 343 */ 344 offset = blkoff(fs, length); 345 lastseg = -1; 346 bc = 0; 347 348 if (ovp != fs->lfs_ivnode) 349 lfs_seglock(fs, SEGM_PROT); 350 if (offset == 0) { 351 oip->i_size = oip->i_ffs1_size = length; 352 } else if (!usepc) { 353 lbn = lblkno(fs, length); 354 aflags = B_CLRBUF; 355 if (ioflag & IO_SYNC) 356 aflags |= B_SYNC; 357 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 358 if (error) { 359 lfs_reserve(fs, ovp, NULL, 360 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 361 goto errout; 362 } 363 obufsize = bp->b_bufsize; 364 odb = btofsb(fs, bp->b_bcount); 365 oip->i_size = oip->i_ffs1_size = length; 366 size = blksize(fs, oip, lbn); 367 if (ovp->v_type != VDIR) 368 memset((char *)bp->b_data + offset, 0, 369 (u_int)(size - offset)); 370 allocbuf(bp, size, 1); 371 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 372 simple_lock(&lfs_subsys_lock); 373 locked_queue_bytes -= obufsize - bp->b_bufsize; 374 simple_unlock(&lfs_subsys_lock); 375 } 376 if (bp->b_flags & B_DELWRI) 377 fs->lfs_avail += odb - btofsb(fs, size); 378 (void) VOP_BWRITE(bp); 379 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 380 /* 381 * When truncating a regular file down to a non-block-aligned 382 * size, we must zero the part of last block which is past 383 * the new EOF. We must synchronously flush the zeroed pages 384 * to disk since the new pages will be invalidated as soon 385 * as we inform the VM system of the new, smaller size. 386 * We must do this before acquiring the GLOCK, since fetching 387 * the pages will acquire the GLOCK internally. 388 * So there is a window where another thread could see a whole 389 * zeroed page past EOF, but that's life. 390 */ 391 daddr_t xlbn; 392 voff_t eoz; 393 394 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 395 error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags); 396 if (error) { 397 lfs_reserve(fs, ovp, NULL, 398 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 399 goto errout; 400 } 401 xlbn = lblkno(fs, length); 402 size = blksize(fs, oip, xlbn); 403 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 404 uvm_vnp_zerorange(ovp, length, eoz - length); 405 if (round_page(eoz) > round_page(length)) { 406 simple_lock(&ovp->v_interlock); 407 error = VOP_PUTPAGES(ovp, round_page(length), 408 round_page(eoz), 409 PGO_CLEANIT | PGO_DEACTIVATE | 410 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 411 if (error) { 412 lfs_reserve(fs, ovp, NULL, 413 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 414 goto errout; 415 } 416 } 417 } 418 419 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 420 421 oip->i_size = oip->i_ffs1_size = length; 422 uvm_vnp_setsize(ovp, length); 423 /* 424 * Calculate index into inode's block list of 425 * last direct and indirect blocks (if any) 426 * which we want to keep. Lastblock is -1 when 427 * the file is truncated to 0. 428 */ 429 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 430 lastiblock[SINGLE] = lastblock - NDADDR; 431 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 432 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 433 nblocks = btofsb(fs, fs->lfs_bsize); 434 /* 435 * Record changed file and block pointers before we start 436 * freeing blocks. lastiblock values are also normalized to -1 437 * for calls to lfs_indirtrunc below. 438 */ 439 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 440 for (level = TRIPLE; level >= SINGLE; level--) 441 if (lastiblock[level] < 0) { 442 newblks[NDADDR+level] = 0; 443 lastiblock[level] = -1; 444 } 445 for (i = NDADDR - 1; i > lastblock; i--) 446 newblks[i] = 0; 447 448 oip->i_size = oip->i_ffs1_size = osize; 449 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 450 if (error && !allerror) 451 allerror = error; 452 453 /* 454 * Indirect blocks first. 455 */ 456 indir_lbn[SINGLE] = -NDADDR; 457 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 458 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 459 for (level = TRIPLE; level >= SINGLE; level--) { 460 bn = oip->i_ffs1_ib[level]; 461 if (bn != 0) { 462 error = lfs_indirtrunc(oip, indir_lbn[level], 463 bn, lastiblock[level], 464 level, &count, &rcount, 465 &lastseg, &bc, l); 466 if (error) 467 allerror = error; 468 real_released += rcount; 469 blocksreleased += count; 470 if (lastiblock[level] < 0) { 471 if (oip->i_ffs1_ib[level] > 0) 472 real_released += nblocks; 473 blocksreleased += nblocks; 474 oip->i_ffs1_ib[level] = 0; 475 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 476 lfs_deregister_block(ovp, bn); 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 lfs_deregister_block(ovp, bn); 504 } 505 if (lastblock < 0) 506 goto done; 507 508 /* 509 * Finally, look for a change in size of the 510 * last direct block; release any frags. 511 */ 512 bn = oip->i_ffs1_db[lastblock]; 513 if (bn != 0) { 514 long oldspace, newspace; 515 #if 0 516 long olddspace; 517 #endif 518 519 /* 520 * Calculate amount of space we're giving 521 * back as old block size minus new block size. 522 */ 523 oldspace = blksize(fs, oip, lastblock); 524 #if 0 525 olddspace = oip->i_lfs_fragsize[lastblock]; 526 #endif 527 528 oip->i_size = oip->i_ffs1_size = length; 529 newspace = blksize(fs, oip, lastblock); 530 if (newspace == 0) 531 panic("itrunc: newspace"); 532 if (oldspace - newspace > 0) { 533 blocksreleased += btofsb(fs, oldspace - newspace); 534 } 535 #if 0 536 if (bn > 0 && olddspace - newspace > 0) { 537 /* No segment accounting here, just vnode */ 538 real_released += btofsb(fs, olddspace - newspace); 539 } 540 #endif 541 } 542 543 done: 544 /* Finish segment accounting corrections */ 545 lfs_update_seguse(fs, lastseg, bc); 546 #ifdef DIAGNOSTIC 547 for (level = SINGLE; level <= TRIPLE; level++) 548 if ((newblks[NDADDR + level] == 0) != 549 (oip->i_ffs1_ib[level]) == 0) { 550 panic("lfs itrunc1"); 551 } 552 for (i = 0; i < NDADDR; i++) 553 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 554 panic("lfs itrunc2"); 555 } 556 if (length == 0 && 557 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 558 panic("lfs itrunc3"); 559 #endif /* DIAGNOSTIC */ 560 /* 561 * Put back the real size. 562 */ 563 oip->i_size = oip->i_ffs1_size = length; 564 oip->i_lfs_effnblks -= blocksreleased; 565 oip->i_ffs1_blocks -= real_released; 566 simple_lock(&fs->lfs_interlock); 567 fs->lfs_bfree += blocksreleased; 568 simple_unlock(&fs->lfs_interlock); 569 #ifdef DIAGNOSTIC 570 if (oip->i_size == 0 && 571 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 572 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 573 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 574 panic("lfs_truncate: persistent blocks"); 575 } 576 #endif 577 578 /* 579 * If we truncated to zero, take us off the paging queue. 580 */ 581 simple_lock(&fs->lfs_interlock); 582 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 583 oip->i_flags &= ~IN_PAGING; 584 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 585 } 586 simple_unlock(&fs->lfs_interlock); 587 588 oip->i_flag |= IN_CHANGE; 589 #ifdef QUOTA 590 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 591 #endif 592 lfs_reserve(fs, ovp, NULL, 593 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 594 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 595 errout: 596 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 597 if (ovp != fs->lfs_ivnode) 598 lfs_segunlock(fs); 599 return (allerror ? allerror : error); 600 } 601 602 /* Update segment and avail usage information when removing a block. */ 603 static int 604 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg, 605 size_t *num) 606 { 607 long seg; 608 int error = 0; 609 610 ASSERT_SEGLOCK(fs); 611 bsize = fragroundup(fs, bsize); 612 if (daddr > 0) { 613 if (*lastseg != (seg = dtosn(fs, daddr))) { 614 error = lfs_update_seguse(fs, *lastseg, *num); 615 *num = bsize; 616 *lastseg = seg; 617 } else 618 *num += bsize; 619 } 620 621 return error; 622 } 623 624 /* Finish the accounting updates for a segment. */ 625 static int 626 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 627 { 628 SEGUSE *sup; 629 struct buf *bp; 630 631 ASSERT_SEGLOCK(fs); 632 if (lastseg < 0 || num == 0) 633 return 0; 634 635 LFS_SEGENTRY(sup, fs, lastseg, bp); 636 if (num > sup->su_nbytes) { 637 printf("lfs_truncate: segment %ld short by %ld\n", 638 lastseg, (long)num - sup->su_nbytes); 639 panic("lfs_truncate: negative bytes"); 640 sup->su_nbytes = num; 641 } 642 sup->su_nbytes -= num; 643 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 644 645 return 0; 646 } 647 648 /* 649 * Release blocks associated with the inode ip and stored in the indirect 650 * block bn. Blocks are free'd in LIFO order up to (but not including) 651 * lastbn. If level is greater than SINGLE, the block is an indirect block 652 * and recursive calls to indirtrunc must be used to cleanse other indirect 653 * blocks. 654 * 655 * NB: triple indirect blocks are untested. 656 */ 657 static int 658 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 659 daddr_t lastbn, int level, long *countp, 660 long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l) 661 { 662 int i; 663 struct buf *bp; 664 struct lfs *fs = ip->i_lfs; 665 int32_t *bap; /* XXX ondisk32 */ 666 struct vnode *vp; 667 daddr_t nb, nlbn, last; 668 int32_t *copy = NULL; /* XXX ondisk32 */ 669 long blkcount, rblkcount, factor; 670 int nblocks, blocksreleased = 0, real_released = 0; 671 int error = 0, allerror = 0; 672 673 ASSERT_SEGLOCK(fs); 674 /* 675 * Calculate index in current block of last 676 * block to be kept. -1 indicates the entire 677 * block so we need not calculate the index. 678 */ 679 factor = 1; 680 for (i = SINGLE; i < level; i++) 681 factor *= NINDIR(fs); 682 last = lastbn; 683 if (lastbn > 0) 684 last /= factor; 685 nblocks = btofsb(fs, fs->lfs_bsize); 686 /* 687 * Get buffer of block pointers, zero those entries corresponding 688 * to blocks to be free'd, and update on disk copy first. Since 689 * double(triple) indirect before single(double) indirect, calls 690 * to bmap on these blocks will fail. However, we already have 691 * the on disk address, so we have to set the b_blkno field 692 * explicitly instead of letting bread do everything for us. 693 */ 694 vp = ITOV(ip); 695 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 696 if (bp->b_flags & (B_DONE | B_DELWRI)) { 697 /* Braces must be here in case trace evaluates to nothing. */ 698 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 699 } else { 700 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 701 l->l_proc->p_stats->p_ru.ru_inblock++; /* pay for read */ 702 bp->b_flags |= B_READ; 703 if (bp->b_bcount > bp->b_bufsize) 704 panic("lfs_indirtrunc: bad buffer size"); 705 bp->b_blkno = fsbtodb(fs, dbn); 706 VOP_STRATEGY(vp, bp); 707 error = biowait(bp); 708 } 709 if (error) { 710 brelse(bp); 711 *countp = *rcountp = 0; 712 return (error); 713 } 714 715 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 716 if (lastbn >= 0) { 717 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 718 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 719 memset((caddr_t)&bap[last + 1], 0, 720 /* XXX ondisk32 */ 721 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 722 error = VOP_BWRITE(bp); 723 if (error) 724 allerror = error; 725 bap = copy; 726 } 727 728 /* 729 * Recursively free totally unused blocks. 730 */ 731 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 732 i--, nlbn += factor) { 733 nb = bap[i]; 734 if (nb == 0) 735 continue; 736 if (level > SINGLE) { 737 error = lfs_indirtrunc(ip, nlbn, nb, 738 (daddr_t)-1, level - 1, 739 &blkcount, &rblkcount, 740 lastsegp, bcp, l); 741 if (error) 742 allerror = error; 743 blocksreleased += blkcount; 744 real_released += rblkcount; 745 } 746 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 747 if (bap[i] > 0) 748 real_released += nblocks; 749 blocksreleased += nblocks; 750 } 751 752 /* 753 * Recursively free last partial block. 754 */ 755 if (level > SINGLE && lastbn >= 0) { 756 last = lastbn % factor; 757 nb = bap[i]; 758 if (nb != 0) { 759 error = lfs_indirtrunc(ip, nlbn, nb, 760 last, level - 1, &blkcount, 761 &rblkcount, lastsegp, bcp, l); 762 if (error) 763 allerror = error; 764 real_released += rblkcount; 765 blocksreleased += blkcount; 766 } 767 } 768 769 if (copy != NULL) { 770 lfs_free(fs, copy, LFS_NB_IBLOCK); 771 } else { 772 if (bp->b_flags & B_DELWRI) { 773 LFS_UNLOCK_BUF(bp); 774 fs->lfs_avail += btofsb(fs, bp->b_bcount); 775 wakeup(&fs->lfs_avail); 776 } 777 bp->b_flags |= B_INVAL; 778 brelse(bp); 779 } 780 781 *countp = blocksreleased; 782 *rcountp = real_released; 783 return (allerror); 784 } 785 786 /* 787 * Destroy any in core blocks past the truncation length. 788 * Inlined from vtruncbuf, so that lfs_avail could be updated. 789 * We take the seglock to prevent cleaning from occurring while we are 790 * invalidating blocks. 791 */ 792 static int 793 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 794 { 795 struct buf *bp, *nbp; 796 int s, error; 797 struct lfs *fs; 798 voff_t off; 799 800 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 801 simple_lock(&vp->v_interlock); 802 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 803 if (error) 804 return error; 805 806 fs = VTOI(vp)->i_lfs; 807 s = splbio(); 808 809 ASSERT_SEGLOCK(fs); 810 restart: 811 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 812 nbp = LIST_NEXT(bp, b_vnbufs); 813 if (bp->b_lblkno < lbn) 814 continue; 815 simple_lock(&bp->b_interlock); 816 if (bp->b_flags & B_BUSY) { 817 bp->b_flags |= B_WANTED; 818 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 819 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 820 if (error) { 821 splx(s); 822 return (error); 823 } 824 goto restart; 825 } 826 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 827 if (bp->b_flags & B_DELWRI) { 828 bp->b_flags &= ~B_DELWRI; 829 fs->lfs_avail += btofsb(fs, bp->b_bcount); 830 wakeup(&fs->lfs_avail); 831 } 832 LFS_UNLOCK_BUF(bp); 833 simple_unlock(&bp->b_interlock); 834 brelse(bp); 835 } 836 837 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 838 nbp = LIST_NEXT(bp, b_vnbufs); 839 if (bp->b_lblkno < lbn) 840 continue; 841 simple_lock(&bp->b_interlock); 842 if (bp->b_flags & B_BUSY) { 843 bp->b_flags |= B_WANTED; 844 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 845 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 846 if (error) { 847 splx(s); 848 return (error); 849 } 850 goto restart; 851 } 852 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 853 if (bp->b_flags & B_DELWRI) { 854 bp->b_flags &= ~B_DELWRI; 855 fs->lfs_avail += btofsb(fs, bp->b_bcount); 856 wakeup(&fs->lfs_avail); 857 } 858 LFS_UNLOCK_BUF(bp); 859 simple_unlock(&bp->b_interlock); 860 brelse(bp); 861 } 862 863 splx(s); 864 865 return (0); 866 } 867 868
Indexes created Thu Oct 02 14:10:14 GMT 2025