lfs_inode.c revision 1.112
1 /* $NetBSD: lfs_inode.c,v 1.112 2007/10/08 18:01:30 ad 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 1.112 2007/10/08 18:01:30 ad 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/malloc.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/trace.h> 86 #include <sys/resourcevar.h> 87 #include <sys/kauth.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 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t); 98 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t, 99 daddr_t, int, long *, long *, long *, size_t *, 100 struct lwp *); 101 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *); 102 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int); 103 104 /* Search a block for a specific dinode. */ 105 struct ufs1_dinode * 106 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 107 { 108 struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data; 109 struct ufs1_dinode *ldip, *fin; 110 111 ASSERT_NO_SEGLOCK(fs); 112 /* 113 * Read the inode block backwards, since later versions of the 114 * inode will supercede earlier ones. Though it is unlikely, it is 115 * possible that the same inode will appear in the same inode block. 116 */ 117 fin = dip + INOPB(fs); 118 for (ldip = fin - 1; ldip >= dip; --ldip) 119 if (ldip->di_inumber == ino) 120 return (ldip); 121 122 printf("searched %d entries\n", (int)(fin - dip)); 123 printf("offset is 0x%x (seg %d)\n", fs->lfs_offset, 124 dtosn(fs, fs->lfs_offset)); 125 printf("block is 0x%llx (seg %lld)\n", 126 (unsigned long long)dbtofsb(fs, bp->b_blkno), 127 (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno))); 128 129 return NULL; 130 } 131 132 int 133 lfs_update(struct vnode *vp, const struct timespec *acc, 134 const struct timespec *mod, int updflags) 135 { 136 struct inode *ip; 137 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 138 int s; 139 int flags; 140 141 ASSERT_NO_SEGLOCK(fs); 142 if (vp->v_mount->mnt_flag & MNT_RDONLY) 143 return (0); 144 ip = VTOI(vp); 145 146 /* 147 * If we are called from vinvalbuf, and the file's blocks have 148 * already been scheduled for writing, but the writes have not 149 * yet completed, lfs_vflush will not be called, and vinvalbuf 150 * will cause a panic. So, we must wait until any pending write 151 * for our inode completes, if we are called with UPDATE_WAIT set. 152 */ 153 s = splbio(); 154 simple_lock(&vp->v_interlock); 155 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 156 WRITEINPROG(vp)) { 157 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 158 " (in progress)\n", ip->i_number)); 159 ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 160 } 161 simple_unlock(&vp->v_interlock); 162 splx(s); 163 LFS_ITIMES(ip, acc, mod, NULL); 164 if (updflags & UPDATE_CLOSE) 165 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 166 else 167 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 168 if (flags == 0) 169 return (0); 170 171 /* If sync, push back the vnode and any dirty blocks it may have. */ 172 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 173 /* Avoid flushing VDIROP. */ 174 simple_lock(&fs->lfs_interlock); 175 ++fs->lfs_diropwait; 176 while (vp->v_flag & VDIROP) { 177 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 178 " (dirops)\n", ip->i_number)); 179 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 180 " 0x%x\n", vp->v_flag, ip->i_flag)); 181 if (fs->lfs_dirops == 0) 182 lfs_flush_fs(fs, SEGM_SYNC); 183 else 184 ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 185 0, &fs->lfs_interlock); 186 /* XXX KS - by falling out here, are we writing the vn 187 twice? */ 188 } 189 --fs->lfs_diropwait; 190 simple_unlock(&fs->lfs_interlock); 191 return lfs_vflush(vp); 192 } 193 return 0; 194 } 195 196 #define SINGLE 0 /* index of single indirect block */ 197 #define DOUBLE 1 /* index of double indirect block */ 198 #define TRIPLE 2 /* index of triple indirect block */ 199 /* 200 * Truncate the inode oip to at most length size, freeing the 201 * disk blocks. 202 */ 203 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */ 204 205 int 206 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, 207 kauth_cred_t cred, struct lwp *l) 208 { 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 fs = oip->i_lfs; 260 lfs_imtime(fs); 261 osize = oip->i_size; 262 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 263 264 ASSERT_NO_SEGLOCK(fs); 265 /* 266 * Lengthen the size of the file. We must ensure that the 267 * last byte of the file is allocated. Since the smallest 268 * value of osize is 0, length will be at least 1. 269 */ 270 if (osize < length) { 271 if (length > ump->um_maxfilesize) 272 return (EFBIG); 273 aflags = B_CLRBUF; 274 if (ioflag & IO_SYNC) 275 aflags |= B_SYNC; 276 if (usepc) { 277 if (lblkno(fs, osize) < NDADDR && 278 lblkno(fs, osize) != lblkno(fs, length) && 279 blkroundup(fs, osize) != osize) { 280 off_t eob; 281 282 eob = blkroundup(fs, osize); 283 uvm_vnp_setwritesize(ovp, eob); 284 error = ufs_balloc_range(ovp, osize, 285 eob - osize, cred, aflags); 286 if (error) 287 return error; 288 if (ioflag & IO_SYNC) { 289 simple_lock(&ovp->v_interlock); 290 VOP_PUTPAGES(ovp, 291 trunc_page(osize & fs->lfs_bmask), 292 round_page(eob), 293 PGO_CLEANIT | PGO_SYNCIO); 294 } 295 } 296 uvm_vnp_setwritesize(ovp, length); 297 error = ufs_balloc_range(ovp, length - 1, 1, cred, 298 aflags); 299 if (error) { 300 (void) lfs_truncate(ovp, osize, 301 ioflag & IO_SYNC, cred, l); 302 return error; 303 } 304 uvm_vnp_setsize(ovp, length); 305 oip->i_flag |= IN_CHANGE | IN_UPDATE; 306 KASSERT(ovp->v_size == oip->i_size); 307 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 308 return (lfs_update(ovp, NULL, NULL, 0)); 309 } else { 310 error = lfs_reserve(fs, ovp, NULL, 311 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 312 if (error) 313 return (error); 314 error = lfs_balloc(ovp, length - 1, 1, cred, 315 aflags, &bp); 316 lfs_reserve(fs, ovp, NULL, 317 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 318 if (error) 319 return (error); 320 oip->i_ffs1_size = oip->i_size = length; 321 uvm_vnp_setsize(ovp, length); 322 (void) VOP_BWRITE(bp); 323 oip->i_flag |= IN_CHANGE | IN_UPDATE; 324 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 325 return (lfs_update(ovp, NULL, NULL, 0)); 326 } 327 } 328 329 if ((error = lfs_reserve(fs, ovp, NULL, 330 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 331 return (error); 332 333 /* 334 * Shorten the size of the file. If the file is not being 335 * truncated to a block boundary, the contents of the 336 * partial block following the end of the file must be 337 * zero'ed in case it ever becomes accessible again because 338 * of subsequent file growth. Directories however are not 339 * zero'ed as they should grow back initialized to empty. 340 */ 341 offset = blkoff(fs, length); 342 lastseg = -1; 343 bc = 0; 344 345 if (ovp != fs->lfs_ivnode) 346 lfs_seglock(fs, SEGM_PROT); 347 if (offset == 0) { 348 oip->i_size = oip->i_ffs1_size = length; 349 } else if (!usepc) { 350 lbn = lblkno(fs, length); 351 aflags = B_CLRBUF; 352 if (ioflag & IO_SYNC) 353 aflags |= B_SYNC; 354 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 355 if (error) { 356 lfs_reserve(fs, ovp, NULL, 357 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 358 goto errout; 359 } 360 obufsize = bp->b_bufsize; 361 odb = btofsb(fs, bp->b_bcount); 362 oip->i_size = oip->i_ffs1_size = length; 363 size = blksize(fs, oip, lbn); 364 if (ovp->v_type != VDIR) 365 memset((char *)bp->b_data + offset, 0, 366 (u_int)(size - offset)); 367 allocbuf(bp, size, 1); 368 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 369 simple_lock(&lfs_subsys_lock); 370 locked_queue_bytes -= obufsize - bp->b_bufsize; 371 simple_unlock(&lfs_subsys_lock); 372 } 373 if (bp->b_flags & B_DELWRI) 374 fs->lfs_avail += odb - btofsb(fs, size); 375 (void) VOP_BWRITE(bp); 376 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 377 /* 378 * When truncating a regular file down to a non-block-aligned 379 * size, we must zero the part of last block which is past 380 * the new EOF. We must synchronously flush the zeroed pages 381 * to disk since the new pages will be invalidated as soon 382 * as we inform the VM system of the new, smaller size. 383 * We must do this before acquiring the GLOCK, since fetching 384 * the pages will acquire the GLOCK internally. 385 * So there is a window where another thread could see a whole 386 * zeroed page past EOF, but that's life. 387 */ 388 daddr_t xlbn; 389 voff_t eoz; 390 391 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 392 error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags); 393 if (error) { 394 lfs_reserve(fs, ovp, NULL, 395 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 396 goto errout; 397 } 398 xlbn = lblkno(fs, length); 399 size = blksize(fs, oip, xlbn); 400 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 401 uvm_vnp_zerorange(ovp, length, eoz - length); 402 if (round_page(eoz) > round_page(length)) { 403 simple_lock(&ovp->v_interlock); 404 error = VOP_PUTPAGES(ovp, round_page(length), 405 round_page(eoz), 406 PGO_CLEANIT | PGO_DEACTIVATE | 407 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 408 if (error) { 409 lfs_reserve(fs, ovp, NULL, 410 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 411 goto errout; 412 } 413 } 414 } 415 416 genfs_node_wrlock(ovp); 417 418 oip->i_size = oip->i_ffs1_size = length; 419 uvm_vnp_setsize(ovp, length); 420 /* 421 * Calculate index into inode's block list of 422 * last direct and indirect blocks (if any) 423 * which we want to keep. Lastblock is -1 when 424 * the file is truncated to 0. 425 */ 426 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 427 if (length > QUAD_MAX - fs->lfs_bsize) 428 lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize); 429 else 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((void *)newblks, (void *)&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, l); 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, oip, bn, fs->lfs_bsize, 477 &lastseg, &bc); 478 lfs_deregister_block(ovp, bn); 479 } 480 } 481 if (lastiblock[level] >= 0) 482 goto done; 483 } 484 485 /* 486 * All whole direct blocks or frags. 487 */ 488 for (i = NDADDR - 1; i > lastblock; i--) { 489 long bsize, obsize; 490 491 bn = oip->i_ffs1_db[i]; 492 if (bn == 0) 493 continue; 494 bsize = blksize(fs, oip, i); 495 if (oip->i_ffs1_db[i] > 0) { 496 /* Check for fragment size changes */ 497 obsize = oip->i_lfs_fragsize[i]; 498 real_released += btofsb(fs, obsize); 499 oip->i_lfs_fragsize[i] = 0; 500 } else 501 obsize = 0; 502 blocksreleased += btofsb(fs, bsize); 503 oip->i_ffs1_db[i] = 0; 504 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 505 lfs_deregister_block(ovp, bn); 506 } 507 if (lastblock < 0) 508 goto done; 509 510 /* 511 * Finally, look for a change in size of the 512 * last direct block; release any frags. 513 */ 514 bn = oip->i_ffs1_db[lastblock]; 515 if (bn != 0) { 516 long oldspace, newspace; 517 #if 0 518 long olddspace; 519 #endif 520 521 /* 522 * Calculate amount of space we're giving 523 * back as old block size minus new block size. 524 */ 525 oldspace = blksize(fs, oip, lastblock); 526 #if 0 527 olddspace = oip->i_lfs_fragsize[lastblock]; 528 #endif 529 530 oip->i_size = oip->i_ffs1_size = length; 531 newspace = blksize(fs, oip, lastblock); 532 if (newspace == 0) 533 panic("itrunc: newspace"); 534 if (oldspace - newspace > 0) { 535 blocksreleased += btofsb(fs, oldspace - newspace); 536 } 537 #if 0 538 if (bn > 0 && olddspace - newspace > 0) { 539 /* No segment accounting here, just vnode */ 540 real_released += btofsb(fs, olddspace - newspace); 541 } 542 #endif 543 } 544 545 done: 546 /* Finish segment accounting corrections */ 547 lfs_update_seguse(fs, oip, lastseg, bc); 548 #ifdef DIAGNOSTIC 549 for (level = SINGLE; level <= TRIPLE; level++) 550 if ((newblks[NDADDR + level] == 0) != 551 ((oip->i_ffs1_ib[level]) == 0)) { 552 panic("lfs itrunc1"); 553 } 554 for (i = 0; i < NDADDR; i++) 555 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 556 panic("lfs itrunc2"); 557 } 558 if (length == 0 && 559 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 560 panic("lfs itrunc3"); 561 #endif /* DIAGNOSTIC */ 562 /* 563 * Put back the real size. 564 */ 565 oip->i_size = oip->i_ffs1_size = length; 566 oip->i_lfs_effnblks -= blocksreleased; 567 oip->i_ffs1_blocks -= real_released; 568 simple_lock(&fs->lfs_interlock); 569 fs->lfs_bfree += blocksreleased; 570 simple_unlock(&fs->lfs_interlock); 571 #ifdef DIAGNOSTIC 572 if (oip->i_size == 0 && 573 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 574 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 575 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 576 panic("lfs_truncate: persistent blocks"); 577 } 578 #endif 579 580 /* 581 * If we truncated to zero, take us off the paging queue. 582 */ 583 simple_lock(&fs->lfs_interlock); 584 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 585 oip->i_flags &= ~IN_PAGING; 586 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 587 } 588 simple_unlock(&fs->lfs_interlock); 589 590 oip->i_flag |= IN_CHANGE; 591 #ifdef QUOTA 592 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 593 #endif 594 lfs_reserve(fs, ovp, NULL, 595 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 596 genfs_node_unlock(ovp); 597 errout: 598 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 599 if (ovp != fs->lfs_ivnode) 600 lfs_segunlock(fs); 601 return (allerror ? allerror : error); 602 } 603 604 /* Update segment and avail usage information when removing a block. */ 605 static int 606 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 607 size_t bsize, long *lastseg, size_t *num) 608 { 609 long seg; 610 int error = 0; 611 612 ASSERT_SEGLOCK(fs); 613 bsize = fragroundup(fs, bsize); 614 if (daddr > 0) { 615 if (*lastseg != (seg = dtosn(fs, daddr))) { 616 error = lfs_update_seguse(fs, ip, *lastseg, *num); 617 *num = bsize; 618 *lastseg = seg; 619 } else 620 *num += bsize; 621 } 622 623 return error; 624 } 625 626 /* Finish the accounting updates for a segment. */ 627 static int 628 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 629 { 630 struct segdelta *sd; 631 struct vnode *vp; 632 633 ASSERT_SEGLOCK(fs); 634 if (lastseg < 0 || num == 0) 635 return 0; 636 637 vp = ITOV(ip); 638 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list) 639 if (sd->segnum == lastseg) 640 break; 641 if (sd == NULL) { 642 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK); 643 sd->segnum = lastseg; 644 sd->num = 0; 645 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list); 646 } 647 sd->num += num; 648 649 return 0; 650 } 651 652 static void 653 lfs_finalize_seguse(struct lfs *fs, void *v) 654 { 655 SEGUSE *sup; 656 struct buf *bp; 657 struct segdelta *sd; 658 LIST_HEAD(, segdelta) *hd = v; 659 660 ASSERT_SEGLOCK(fs); 661 while((sd = LIST_FIRST(hd)) != NULL) { 662 LIST_REMOVE(sd, list); 663 LFS_SEGENTRY(sup, fs, sd->segnum, bp); 664 if (sd->num > sup->su_nbytes) { 665 printf("lfs_finalize_seguse: segment %ld short by %ld\n", 666 sd->segnum, (long)(sd->num - sup->su_nbytes)); 667 panic("lfs_finalize_seguse: negative bytes"); 668 sup->su_nbytes = sd->num; 669 } 670 sup->su_nbytes -= sd->num; 671 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp); 672 free(sd, M_SEGMENT); 673 } 674 } 675 676 /* Finish the accounting updates for a segment. */ 677 void 678 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip) 679 { 680 ASSERT_SEGLOCK(fs); 681 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd); 682 } 683 684 /* Finish the accounting updates for a segment. */ 685 void 686 lfs_finalize_fs_seguse(struct lfs *fs) 687 { 688 ASSERT_SEGLOCK(fs); 689 lfs_finalize_seguse(fs, &fs->lfs_segdhd); 690 } 691 692 /* 693 * Release blocks associated with the inode ip and stored in the indirect 694 * block bn. Blocks are free'd in LIFO order up to (but not including) 695 * lastbn. If level is greater than SINGLE, the block is an indirect block 696 * and recursive calls to indirtrunc must be used to cleanse other indirect 697 * blocks. 698 * 699 * NB: triple indirect blocks are untested. 700 */ 701 static int 702 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 703 daddr_t lastbn, int level, long *countp, 704 long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l) 705 { 706 int i; 707 struct buf *bp; 708 struct lfs *fs = ip->i_lfs; 709 int32_t *bap; /* XXX ondisk32 */ 710 struct vnode *vp; 711 daddr_t nb, nlbn, last; 712 int32_t *copy = NULL; /* XXX ondisk32 */ 713 long blkcount, rblkcount, factor; 714 int nblocks, blocksreleased = 0, real_released = 0; 715 int error = 0, allerror = 0; 716 717 ASSERT_SEGLOCK(fs); 718 /* 719 * Calculate index in current block of last 720 * block to be kept. -1 indicates the entire 721 * block so we need not calculate the index. 722 */ 723 factor = 1; 724 for (i = SINGLE; i < level; i++) 725 factor *= NINDIR(fs); 726 last = lastbn; 727 if (lastbn > 0) 728 last /= factor; 729 nblocks = btofsb(fs, fs->lfs_bsize); 730 /* 731 * Get buffer of block pointers, zero those entries corresponding 732 * to blocks to be free'd, and update on disk copy first. Since 733 * double(triple) indirect before single(double) indirect, calls 734 * to bmap on these blocks will fail. However, we already have 735 * the on disk address, so we have to set the b_blkno field 736 * explicitly instead of letting bread do everything for us. 737 */ 738 vp = ITOV(ip); 739 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 740 if (bp->b_flags & (B_DONE | B_DELWRI)) { 741 /* Braces must be here in case trace evaluates to nothing. */ 742 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 743 } else { 744 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 745 l->l_proc->p_stats->p_ru.ru_inblock++; /* pay for read */ 746 bp->b_flags |= B_READ; 747 if (bp->b_bcount > bp->b_bufsize) 748 panic("lfs_indirtrunc: bad buffer size"); 749 bp->b_blkno = fsbtodb(fs, dbn); 750 VOP_STRATEGY(vp, bp); 751 error = biowait(bp); 752 } 753 if (error) { 754 brelse(bp, 0); 755 *countp = *rcountp = 0; 756 return (error); 757 } 758 759 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 760 if (lastbn >= 0) { 761 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 762 memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize); 763 memset((void *)&bap[last + 1], 0, 764 /* XXX ondisk32 */ 765 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 766 error = VOP_BWRITE(bp); 767 if (error) 768 allerror = error; 769 bap = copy; 770 } 771 772 /* 773 * Recursively free totally unused blocks. 774 */ 775 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 776 i--, nlbn += factor) { 777 nb = bap[i]; 778 if (nb == 0) 779 continue; 780 if (level > SINGLE) { 781 error = lfs_indirtrunc(ip, nlbn, nb, 782 (daddr_t)-1, level - 1, 783 &blkcount, &rblkcount, 784 lastsegp, bcp, l); 785 if (error) 786 allerror = error; 787 blocksreleased += blkcount; 788 real_released += rblkcount; 789 } 790 lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp); 791 if (bap[i] > 0) 792 real_released += nblocks; 793 blocksreleased += nblocks; 794 } 795 796 /* 797 * Recursively free last partial block. 798 */ 799 if (level > SINGLE && lastbn >= 0) { 800 last = lastbn % factor; 801 nb = bap[i]; 802 if (nb != 0) { 803 error = lfs_indirtrunc(ip, nlbn, nb, 804 last, level - 1, &blkcount, 805 &rblkcount, lastsegp, bcp, l); 806 if (error) 807 allerror = error; 808 real_released += rblkcount; 809 blocksreleased += blkcount; 810 } 811 } 812 813 if (copy != NULL) { 814 lfs_free(fs, copy, LFS_NB_IBLOCK); 815 } else { 816 if (bp->b_flags & B_DELWRI) { 817 LFS_UNLOCK_BUF(bp); 818 fs->lfs_avail += btofsb(fs, bp->b_bcount); 819 wakeup(&fs->lfs_avail); 820 } 821 brelse(bp, BC_INVAL); 822 } 823 824 *countp = blocksreleased; 825 *rcountp = real_released; 826 return (allerror); 827 } 828 829 /* 830 * Destroy any in core blocks past the truncation length. 831 * Inlined from vtruncbuf, so that lfs_avail could be updated. 832 * We take the seglock to prevent cleaning from occurring while we are 833 * invalidating blocks. 834 */ 835 static int 836 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 837 { 838 struct buf *bp, *nbp; 839 int s, error; 840 struct lfs *fs; 841 voff_t off; 842 843 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 844 simple_lock(&vp->v_interlock); 845 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 846 if (error) 847 return error; 848 849 fs = VTOI(vp)->i_lfs; 850 s = splbio(); 851 852 ASSERT_SEGLOCK(fs); 853 restart: 854 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 855 nbp = LIST_NEXT(bp, b_vnbufs); 856 if (bp->b_lblkno < lbn) 857 continue; 858 simple_lock(&bp->b_interlock); 859 if (bp->b_flags & B_BUSY) { 860 bp->b_flags |= B_WANTED; 861 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 862 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 863 if (error) { 864 splx(s); 865 return (error); 866 } 867 goto restart; 868 } 869 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 870 if (bp->b_flags & B_DELWRI) { 871 bp->b_flags &= ~B_DELWRI; 872 fs->lfs_avail += btofsb(fs, bp->b_bcount); 873 wakeup(&fs->lfs_avail); 874 } 875 LFS_UNLOCK_BUF(bp); 876 simple_unlock(&bp->b_interlock); 877 brelse(bp, 0); 878 } 879 880 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 881 nbp = LIST_NEXT(bp, b_vnbufs); 882 if (bp->b_lblkno < lbn) 883 continue; 884 simple_lock(&bp->b_interlock); 885 if (bp->b_flags & B_BUSY) { 886 bp->b_flags |= B_WANTED; 887 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 888 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 889 if (error) { 890 splx(s); 891 return (error); 892 } 893 goto restart; 894 } 895 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 896 if (bp->b_flags & B_DELWRI) { 897 bp->b_flags &= ~B_DELWRI; 898 fs->lfs_avail += btofsb(fs, bp->b_bcount); 899 wakeup(&fs->lfs_avail); 900 } 901 LFS_UNLOCK_BUF(bp); 902 simple_unlock(&bp->b_interlock); 903 brelse(bp, 0); 904 } 905 906 splx(s); 907 908 return (0); 909 } 910 911
Indexes created Fri Sep 26 20:09:58 GMT 2025