lfs_inode.c revision 1.114.6.1
1 /* $NetBSD: lfs_inode.c,v 1.114.6.1 2007/12/04 13:03:50 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 1.114.6.1 2007/12/04 13:03:50 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, bool, 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 mutex_enter(&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 mtsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 160 } 161 mutex_exit(&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 mutex_enter(&fs->lfs_interlock); 175 ++fs->lfs_diropwait; 176 while (vp->v_uflag & VU_DIROP) { 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", 181 vp->v_iflag | vp->v_vflag | vp->v_uflag, 182 ip->i_flag)); 183 if (fs->lfs_dirops == 0) 184 lfs_flush_fs(fs, SEGM_SYNC); 185 else 186 mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 187 0, &fs->lfs_interlock); 188 /* XXX KS - by falling out here, are we writing the vn 189 twice? */ 190 } 191 --fs->lfs_diropwait; 192 mutex_exit(&fs->lfs_interlock); 193 return lfs_vflush(vp); 194 } 195 return 0; 196 } 197 198 #define SINGLE 0 /* index of single indirect block */ 199 #define DOUBLE 1 /* index of double indirect block */ 200 #define TRIPLE 2 /* index of triple indirect block */ 201 /* 202 * Truncate the inode oip to at most length size, freeing the 203 * disk blocks. 204 */ 205 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */ 206 207 int 208 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, 209 kauth_cred_t cred, struct lwp *l) 210 { 211 daddr_t lastblock; 212 struct inode *oip = VTOI(ovp); 213 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 214 /* XXX ondisk32 */ 215 int32_t newblks[NDADDR + NIADDR]; 216 struct lfs *fs; 217 struct buf *bp; 218 int offset, size, level; 219 long count, rcount, blocksreleased = 0, real_released = 0; 220 int i, nblocks; 221 int aflags, error, allerror = 0; 222 off_t osize; 223 long lastseg; 224 size_t bc; 225 int obufsize, odb; 226 int usepc; 227 struct ufsmount *ump = oip->i_ump; 228 229 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 230 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 231 KASSERT(oip->i_size == 0); 232 return 0; 233 } 234 235 if (length < 0) 236 return (EINVAL); 237 238 /* 239 * Just return and not update modification times. 240 */ 241 if (oip->i_size == length) 242 return (0); 243 244 if (ovp->v_type == VLNK && 245 (oip->i_size < ump->um_maxsymlinklen || 246 (ump->um_maxsymlinklen == 0 && 247 oip->i_ffs1_blocks == 0))) { 248 #ifdef DIAGNOSTIC 249 if (length != 0) 250 panic("lfs_truncate: partial truncate of symlink"); 251 #endif 252 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 253 oip->i_size = oip->i_ffs1_size = 0; 254 oip->i_flag |= IN_CHANGE | IN_UPDATE; 255 return (lfs_update(ovp, NULL, NULL, 0)); 256 } 257 if (oip->i_size == length) { 258 oip->i_flag |= IN_CHANGE | IN_UPDATE; 259 return (lfs_update(ovp, NULL, NULL, 0)); 260 } 261 fs = oip->i_lfs; 262 lfs_imtime(fs); 263 osize = oip->i_size; 264 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 265 266 ASSERT_NO_SEGLOCK(fs); 267 /* 268 * Lengthen the size of the file. We must ensure that the 269 * last byte of the file is allocated. Since the smallest 270 * value of osize is 0, length will be at least 1. 271 */ 272 if (osize < length) { 273 if (length > ump->um_maxfilesize) 274 return (EFBIG); 275 aflags = B_CLRBUF; 276 if (ioflag & IO_SYNC) 277 aflags |= B_SYNC; 278 if (usepc) { 279 if (lblkno(fs, osize) < NDADDR && 280 lblkno(fs, osize) != lblkno(fs, length) && 281 blkroundup(fs, osize) != osize) { 282 off_t eob; 283 284 eob = blkroundup(fs, osize); 285 uvm_vnp_setwritesize(ovp, eob); 286 error = ufs_balloc_range(ovp, osize, 287 eob - osize, cred, aflags); 288 if (error) 289 return error; 290 if (ioflag & IO_SYNC) { 291 mutex_enter(&ovp->v_interlock); 292 VOP_PUTPAGES(ovp, 293 trunc_page(osize & fs->lfs_bmask), 294 round_page(eob), 295 PGO_CLEANIT | PGO_SYNCIO); 296 } 297 } 298 uvm_vnp_setwritesize(ovp, length); 299 error = ufs_balloc_range(ovp, length - 1, 1, cred, 300 aflags); 301 if (error) { 302 (void) lfs_truncate(ovp, osize, 303 ioflag & IO_SYNC, cred, l); 304 return error; 305 } 306 uvm_vnp_setsize(ovp, length); 307 oip->i_flag |= IN_CHANGE | IN_UPDATE; 308 KASSERT(ovp->v_size == oip->i_size); 309 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 310 return (lfs_update(ovp, NULL, NULL, 0)); 311 } else { 312 error = lfs_reserve(fs, ovp, NULL, 313 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 314 if (error) 315 return (error); 316 error = lfs_balloc(ovp, length - 1, 1, cred, 317 aflags, &bp); 318 lfs_reserve(fs, ovp, NULL, 319 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 320 if (error) 321 return (error); 322 oip->i_ffs1_size = oip->i_size = length; 323 uvm_vnp_setsize(ovp, length); 324 (void) VOP_BWRITE(bp); 325 oip->i_flag |= IN_CHANGE | IN_UPDATE; 326 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 327 return (lfs_update(ovp, NULL, NULL, 0)); 328 } 329 } 330 331 if ((error = lfs_reserve(fs, ovp, NULL, 332 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 333 return (error); 334 335 /* 336 * Shorten the size of the file. If the file is not being 337 * truncated to a block boundary, the contents of the 338 * partial block following the end of the file must be 339 * zero'ed in case it ever becomes accessible again because 340 * of subsequent file growth. Directories however are not 341 * zero'ed as they should grow back initialized to empty. 342 */ 343 offset = blkoff(fs, length); 344 lastseg = -1; 345 bc = 0; 346 347 if (ovp != fs->lfs_ivnode) 348 lfs_seglock(fs, SEGM_PROT); 349 if (offset == 0) { 350 oip->i_size = oip->i_ffs1_size = length; 351 } else if (!usepc) { 352 lbn = lblkno(fs, length); 353 aflags = B_CLRBUF; 354 if (ioflag & IO_SYNC) 355 aflags |= B_SYNC; 356 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp); 357 if (error) { 358 lfs_reserve(fs, ovp, NULL, 359 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 360 goto errout; 361 } 362 obufsize = bp->b_bufsize; 363 odb = btofsb(fs, bp->b_bcount); 364 oip->i_size = oip->i_ffs1_size = length; 365 size = blksize(fs, oip, lbn); 366 if (ovp->v_type != VDIR) 367 memset((char *)bp->b_data + offset, 0, 368 (u_int)(size - offset)); 369 allocbuf(bp, size, 1); 370 if ((bp->b_cflags & BC_LOCKED) != 0 && bp->b_iodone == NULL) { 371 mutex_enter(&lfs_subsys_lock); 372 locked_queue_bytes -= obufsize - bp->b_bufsize; 373 mutex_exit(&lfs_subsys_lock); 374 } 375 if (bp->b_oflags & BO_DELWRI) 376 fs->lfs_avail += odb - btofsb(fs, size); 377 (void) VOP_BWRITE(bp); 378 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 379 /* 380 * When truncating a regular file down to a non-block-aligned 381 * size, we must zero the part of last block which is past 382 * the new EOF. We must synchronously flush the zeroed pages 383 * to disk since the new pages will be invalidated as soon 384 * as we inform the VM system of the new, smaller size. 385 * We must do this before acquiring the GLOCK, since fetching 386 * the pages will acquire the GLOCK internally. 387 * So there is a window where another thread could see a whole 388 * zeroed page past EOF, but that's life. 389 */ 390 daddr_t xlbn; 391 voff_t eoz; 392 393 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 394 error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags); 395 if (error) { 396 lfs_reserve(fs, ovp, NULL, 397 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 398 goto errout; 399 } 400 xlbn = lblkno(fs, length); 401 size = blksize(fs, oip, xlbn); 402 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 403 uvm_vnp_zerorange(ovp, length, eoz - length); 404 if (round_page(eoz) > round_page(length)) { 405 mutex_enter(&ovp->v_interlock); 406 error = VOP_PUTPAGES(ovp, round_page(length), 407 round_page(eoz), 408 PGO_CLEANIT | PGO_DEACTIVATE | 409 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 410 if (error) { 411 lfs_reserve(fs, ovp, NULL, 412 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 413 goto errout; 414 } 415 } 416 } 417 418 genfs_node_wrlock(ovp); 419 420 oip->i_size = oip->i_ffs1_size = length; 421 uvm_vnp_setsize(ovp, length); 422 /* 423 * Calculate index into inode's block list of 424 * last direct and indirect blocks (if any) 425 * which we want to keep. Lastblock is -1 when 426 * the file is truncated to 0. 427 */ 428 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 429 if (length > QUAD_MAX - fs->lfs_bsize) 430 lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize); 431 else 432 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 433 lastiblock[SINGLE] = lastblock - NDADDR; 434 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 435 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 436 nblocks = btofsb(fs, fs->lfs_bsize); 437 /* 438 * Record changed file and block pointers before we start 439 * freeing blocks. lastiblock values are also normalized to -1 440 * for calls to lfs_indirtrunc below. 441 */ 442 memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks); 443 for (level = TRIPLE; level >= SINGLE; level--) 444 if (lastiblock[level] < 0) { 445 newblks[NDADDR+level] = 0; 446 lastiblock[level] = -1; 447 } 448 for (i = NDADDR - 1; i > lastblock; i--) 449 newblks[i] = 0; 450 451 oip->i_size = oip->i_ffs1_size = osize; 452 error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0); 453 if (error && !allerror) 454 allerror = error; 455 456 /* 457 * Indirect blocks first. 458 */ 459 indir_lbn[SINGLE] = -NDADDR; 460 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 461 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 462 for (level = TRIPLE; level >= SINGLE; level--) { 463 bn = oip->i_ffs1_ib[level]; 464 if (bn != 0) { 465 error = lfs_indirtrunc(oip, indir_lbn[level], 466 bn, lastiblock[level], 467 level, &count, &rcount, 468 &lastseg, &bc, l); 469 if (error) 470 allerror = error; 471 real_released += rcount; 472 blocksreleased += count; 473 if (lastiblock[level] < 0) { 474 if (oip->i_ffs1_ib[level] > 0) 475 real_released += nblocks; 476 blocksreleased += nblocks; 477 oip->i_ffs1_ib[level] = 0; 478 lfs_blkfree(fs, oip, bn, fs->lfs_bsize, 479 &lastseg, &bc); 480 lfs_deregister_block(ovp, bn); 481 } 482 } 483 if (lastiblock[level] >= 0) 484 goto done; 485 } 486 487 /* 488 * All whole direct blocks or frags. 489 */ 490 for (i = NDADDR - 1; i > lastblock; i--) { 491 long bsize, obsize; 492 493 bn = oip->i_ffs1_db[i]; 494 if (bn == 0) 495 continue; 496 bsize = blksize(fs, oip, i); 497 if (oip->i_ffs1_db[i] > 0) { 498 /* Check for fragment size changes */ 499 obsize = oip->i_lfs_fragsize[i]; 500 real_released += btofsb(fs, obsize); 501 oip->i_lfs_fragsize[i] = 0; 502 } else 503 obsize = 0; 504 blocksreleased += btofsb(fs, bsize); 505 oip->i_ffs1_db[i] = 0; 506 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 507 lfs_deregister_block(ovp, bn); 508 } 509 if (lastblock < 0) 510 goto done; 511 512 /* 513 * Finally, look for a change in size of the 514 * last direct block; release any frags. 515 */ 516 bn = oip->i_ffs1_db[lastblock]; 517 if (bn != 0) { 518 long oldspace, newspace; 519 #if 0 520 long olddspace; 521 #endif 522 523 /* 524 * Calculate amount of space we're giving 525 * back as old block size minus new block size. 526 */ 527 oldspace = blksize(fs, oip, lastblock); 528 #if 0 529 olddspace = oip->i_lfs_fragsize[lastblock]; 530 #endif 531 532 oip->i_size = oip->i_ffs1_size = length; 533 newspace = blksize(fs, oip, lastblock); 534 if (newspace == 0) 535 panic("itrunc: newspace"); 536 if (oldspace - newspace > 0) { 537 blocksreleased += btofsb(fs, oldspace - newspace); 538 } 539 #if 0 540 if (bn > 0 && olddspace - newspace > 0) { 541 /* No segment accounting here, just vnode */ 542 real_released += btofsb(fs, olddspace - newspace); 543 } 544 #endif 545 } 546 547 done: 548 /* Finish segment accounting corrections */ 549 lfs_update_seguse(fs, oip, lastseg, bc); 550 #ifdef DIAGNOSTIC 551 for (level = SINGLE; level <= TRIPLE; level++) 552 if ((newblks[NDADDR + level] == 0) != 553 ((oip->i_ffs1_ib[level]) == 0)) { 554 panic("lfs itrunc1"); 555 } 556 for (i = 0; i < NDADDR; i++) 557 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 558 panic("lfs itrunc2"); 559 } 560 if (length == 0 && 561 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 562 panic("lfs itrunc3"); 563 #endif /* DIAGNOSTIC */ 564 /* 565 * Put back the real size. 566 */ 567 oip->i_size = oip->i_ffs1_size = length; 568 oip->i_lfs_effnblks -= blocksreleased; 569 oip->i_ffs1_blocks -= real_released; 570 mutex_enter(&fs->lfs_interlock); 571 fs->lfs_bfree += blocksreleased; 572 mutex_exit(&fs->lfs_interlock); 573 #ifdef DIAGNOSTIC 574 if (oip->i_size == 0 && 575 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 576 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 577 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 578 panic("lfs_truncate: persistent blocks"); 579 } 580 #endif 581 582 /* 583 * If we truncated to zero, take us off the paging queue. 584 */ 585 mutex_enter(&fs->lfs_interlock); 586 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 587 oip->i_flags &= ~IN_PAGING; 588 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 589 } 590 mutex_exit(&fs->lfs_interlock); 591 592 oip->i_flag |= IN_CHANGE; 593 #ifdef QUOTA 594 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 595 #endif 596 lfs_reserve(fs, ovp, NULL, 597 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 598 genfs_node_unlock(ovp); 599 errout: 600 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 601 if (ovp != fs->lfs_ivnode) 602 lfs_segunlock(fs); 603 return (allerror ? allerror : error); 604 } 605 606 /* Update segment and avail usage information when removing a block. */ 607 static int 608 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 609 size_t bsize, long *lastseg, size_t *num) 610 { 611 long seg; 612 int error = 0; 613 614 ASSERT_SEGLOCK(fs); 615 bsize = fragroundup(fs, bsize); 616 if (daddr > 0) { 617 if (*lastseg != (seg = dtosn(fs, daddr))) { 618 error = lfs_update_seguse(fs, ip, *lastseg, *num); 619 *num = bsize; 620 *lastseg = seg; 621 } else 622 *num += bsize; 623 } 624 625 return error; 626 } 627 628 /* Finish the accounting updates for a segment. */ 629 static int 630 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 631 { 632 struct segdelta *sd; 633 struct vnode *vp; 634 635 ASSERT_SEGLOCK(fs); 636 if (lastseg < 0 || num == 0) 637 return 0; 638 639 vp = ITOV(ip); 640 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list) 641 if (sd->segnum == lastseg) 642 break; 643 if (sd == NULL) { 644 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK); 645 sd->segnum = lastseg; 646 sd->num = 0; 647 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list); 648 } 649 sd->num += num; 650 651 return 0; 652 } 653 654 static void 655 lfs_finalize_seguse(struct lfs *fs, void *v) 656 { 657 SEGUSE *sup; 658 struct buf *bp; 659 struct segdelta *sd; 660 LIST_HEAD(, segdelta) *hd = v; 661 662 ASSERT_SEGLOCK(fs); 663 while((sd = LIST_FIRST(hd)) != NULL) { 664 LIST_REMOVE(sd, list); 665 LFS_SEGENTRY(sup, fs, sd->segnum, bp); 666 if (sd->num > sup->su_nbytes) { 667 printf("lfs_finalize_seguse: segment %ld short by %ld\n", 668 sd->segnum, (long)(sd->num - sup->su_nbytes)); 669 panic("lfs_finalize_seguse: negative bytes"); 670 sup->su_nbytes = sd->num; 671 } 672 sup->su_nbytes -= sd->num; 673 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp); 674 free(sd, M_SEGMENT); 675 } 676 } 677 678 /* Finish the accounting updates for a segment. */ 679 void 680 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip) 681 { 682 ASSERT_SEGLOCK(fs); 683 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd); 684 } 685 686 /* Finish the accounting updates for a segment. */ 687 void 688 lfs_finalize_fs_seguse(struct lfs *fs) 689 { 690 ASSERT_SEGLOCK(fs); 691 lfs_finalize_seguse(fs, &fs->lfs_segdhd); 692 } 693 694 /* 695 * Release blocks associated with the inode ip and stored in the indirect 696 * block bn. Blocks are free'd in LIFO order up to (but not including) 697 * lastbn. If level is greater than SINGLE, the block is an indirect block 698 * and recursive calls to indirtrunc must be used to cleanse other indirect 699 * blocks. 700 * 701 * NB: triple indirect blocks are untested. 702 */ 703 static int 704 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 705 daddr_t lastbn, int level, long *countp, 706 long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l) 707 { 708 int i; 709 struct buf *bp; 710 struct lfs *fs = ip->i_lfs; 711 int32_t *bap; /* XXX ondisk32 */ 712 struct vnode *vp; 713 daddr_t nb, nlbn, last; 714 int32_t *copy = NULL; /* XXX ondisk32 */ 715 long blkcount, rblkcount, factor; 716 int nblocks, blocksreleased = 0, real_released = 0; 717 int error = 0, allerror = 0; 718 719 ASSERT_SEGLOCK(fs); 720 /* 721 * Calculate index in current block of last 722 * block to be kept. -1 indicates the entire 723 * block so we need not calculate the index. 724 */ 725 factor = 1; 726 for (i = SINGLE; i < level; i++) 727 factor *= NINDIR(fs); 728 last = lastbn; 729 if (lastbn > 0) 730 last /= factor; 731 nblocks = btofsb(fs, fs->lfs_bsize); 732 /* 733 * Get buffer of block pointers, zero those entries corresponding 734 * to blocks to be free'd, and update on disk copy first. Since 735 * double(triple) indirect before single(double) indirect, calls 736 * to bmap on these blocks will fail. However, we already have 737 * the on disk address, so we have to set the b_blkno field 738 * explicitly instead of letting bread do everything for us. 739 */ 740 vp = ITOV(ip); 741 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 742 if (bp->b_oflags & (BO_DONE | BO_DELWRI)) { 743 /* Braces must be here in case trace evaluates to nothing. */ 744 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 745 } else { 746 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 747 l->l_proc->p_stats->p_ru.ru_inblock++; /* pay for read */ 748 bp->b_flags |= B_READ; 749 if (bp->b_bcount > bp->b_bufsize) 750 panic("lfs_indirtrunc: bad buffer size"); 751 bp->b_blkno = fsbtodb(fs, dbn); 752 VOP_STRATEGY(vp, bp); 753 error = biowait(bp); 754 } 755 if (error) { 756 brelse(bp, 0); 757 *countp = *rcountp = 0; 758 return (error); 759 } 760 761 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 762 if (lastbn >= 0) { 763 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 764 memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize); 765 memset((void *)&bap[last + 1], 0, 766 /* XXX ondisk32 */ 767 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 768 error = VOP_BWRITE(bp); 769 if (error) 770 allerror = error; 771 bap = copy; 772 } 773 774 /* 775 * Recursively free totally unused blocks. 776 */ 777 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 778 i--, nlbn += factor) { 779 nb = bap[i]; 780 if (nb == 0) 781 continue; 782 if (level > SINGLE) { 783 error = lfs_indirtrunc(ip, nlbn, nb, 784 (daddr_t)-1, level - 1, 785 &blkcount, &rblkcount, 786 lastsegp, bcp, l); 787 if (error) 788 allerror = error; 789 blocksreleased += blkcount; 790 real_released += rblkcount; 791 } 792 lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp); 793 if (bap[i] > 0) 794 real_released += nblocks; 795 blocksreleased += nblocks; 796 } 797 798 /* 799 * Recursively free last partial block. 800 */ 801 if (level > SINGLE && lastbn >= 0) { 802 last = lastbn % factor; 803 nb = bap[i]; 804 if (nb != 0) { 805 error = lfs_indirtrunc(ip, nlbn, nb, 806 last, level - 1, &blkcount, 807 &rblkcount, lastsegp, bcp, l); 808 if (error) 809 allerror = error; 810 real_released += rblkcount; 811 blocksreleased += blkcount; 812 } 813 } 814 815 if (copy != NULL) { 816 lfs_free(fs, copy, LFS_NB_IBLOCK); 817 } else { 818 mutex_enter(&bufcache_lock); 819 if (bp->b_oflags & BO_DELWRI) { 820 LFS_UNLOCK_BUF(bp); 821 fs->lfs_avail += btofsb(fs, bp->b_bcount); 822 wakeup(&fs->lfs_avail); 823 } 824 brelsel(bp, BC_INVAL); 825 mutex_exit(&bufcache_lock); 826 } 827 828 *countp = blocksreleased; 829 *rcountp = real_released; 830 return (allerror); 831 } 832 833 /* 834 * Destroy any in core blocks past the truncation length. 835 * Inlined from vtruncbuf, so that lfs_avail could be updated. 836 * We take the seglock to prevent cleaning from occurring while we are 837 * invalidating blocks. 838 */ 839 static int 840 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 841 { 842 struct buf *bp, *nbp; 843 int error; 844 struct lfs *fs; 845 voff_t off; 846 847 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 848 mutex_enter(&vp->v_interlock); 849 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 850 if (error) 851 return error; 852 853 fs = VTOI(vp)->i_lfs; 854 855 ASSERT_SEGLOCK(fs); 856 857 mutex_enter(&bufcache_lock); 858 restart: 859 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 860 nbp = LIST_NEXT(bp, b_vnbufs); 861 if (bp->b_lblkno < lbn) 862 continue; 863 error = bbusy(bp, catch, slptimeo); 864 if (error == EPASSTHROUGH) 865 goto restart; 866 if (error != 0) { 867 mutex_exit(&bufcache_lock); 868 return (error); 869 } 870 mutex_enter(bp->b_objlock); 871 if (bp->b_oflags & BO_DELWRI) { 872 bp->b_oflags &= ~BO_DELWRI; 873 fs->lfs_avail += btofsb(fs, bp->b_bcount); 874 wakeup(&fs->lfs_avail); 875 } 876 mutex_exit(bp->b_objlock); 877 LFS_UNLOCK_BUF(bp); 878 brelsel(bp, BC_INVAL | BC_VFLUSH); 879 } 880 881 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 882 nbp = LIST_NEXT(bp, b_vnbufs); 883 if (bp->b_lblkno < lbn) 884 continue; 885 error = bbusy(bp, catch, slptimeo); 886 if (error == EPASSTHROUGH) 887 goto restart; 888 if (error != 0) { 889 mutex_exit(&bufcache_lock); 890 return (error); 891 } 892 mutex_enter(bp->b_objlock); 893 if (bp->b_oflags & BO_DELWRI) { 894 bp->b_oflags &= ~BO_DELWRI; 895 fs->lfs_avail += btofsb(fs, bp->b_bcount); 896 wakeup(&fs->lfs_avail); 897 } 898 mutex_exit(bp->b_objlock); 899 LFS_UNLOCK_BUF(bp); 900 brelsel(bp, BC_INVAL | BC_VFLUSH); 901 mutex_enter(&vp->v_interlock); 902 } 903 mutex_exit(&bufcache_lock); 904 905 return (0); 906 } 907 908
Indexes created Thu Oct 02 14:10:14 GMT 2025