lfs_inode.c revision 1.92
1 /* $NetBSD: lfs_inode.c,v 1.92 2005/04/14 00:02:46 perseant 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 1.92 2005/04/14 00:02:46 perseant 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/malloc.h> 85 #include <sys/trace.h> 86 #include <sys/resourcevar.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 proc *); 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(void *v) 133 { 134 struct vop_update_args /* { 135 struct vnode *a_vp; 136 struct timespec *a_access; 137 struct timespec *a_modify; 138 int a_flags; 139 } */ *ap = v; 140 struct inode *ip; 141 struct vnode *vp = ap->a_vp; 142 struct timespec ts; 143 struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs; 144 int s; 145 int flags; 146 147 ASSERT_NO_SEGLOCK(fs); 148 if (vp->v_mount->mnt_flag & MNT_RDONLY) 149 return (0); 150 ip = VTOI(vp); 151 152 /* 153 * If we are called from vinvalbuf, and the file's blocks have 154 * already been scheduled for writing, but the writes have not 155 * yet completed, lfs_vflush will not be called, and vinvalbuf 156 * will cause a panic. So, we must wait until any pending write 157 * for our inode completes, if we are called with UPDATE_WAIT set. 158 */ 159 s = splbio(); 160 simple_lock(&vp->v_interlock); 161 while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 162 WRITEINPROG(vp)) { 163 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 164 " (in progress)\n", ip->i_number)); 165 ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock); 166 } 167 simple_unlock(&vp->v_interlock); 168 splx(s); 169 TIMEVAL_TO_TIMESPEC(&time, &ts); 170 LFS_ITIMES(ip, 171 ap->a_access ? ap->a_access : &ts, 172 ap->a_modify ? ap->a_modify : &ts, &ts); 173 if (ap->a_flags & UPDATE_CLOSE) 174 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 175 else 176 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 177 if (flags == 0) 178 return (0); 179 180 /* If sync, push back the vnode and any dirty blocks it may have. */ 181 if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 182 /* Avoid flushing VDIROP. */ 183 simple_lock(&fs->lfs_interlock); 184 ++fs->lfs_diropwait; 185 while (vp->v_flag & VDIROP) { 186 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 187 " (dirops)\n", ip->i_number)); 188 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 189 " 0x%x\n", vp->v_flag, ip->i_flag)); 190 if (fs->lfs_dirops == 0) 191 lfs_flush_fs(fs, SEGM_SYNC); 192 else 193 ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 194 0, &fs->lfs_interlock); 195 /* XXX KS - by falling out here, are we writing the vn 196 twice? */ 197 } 198 --fs->lfs_diropwait; 199 simple_unlock(&fs->lfs_interlock); 200 return lfs_vflush(vp); 201 } 202 return 0; 203 } 204 205 #define SINGLE 0 /* index of single indirect block */ 206 #define DOUBLE 1 /* index of double indirect block */ 207 #define TRIPLE 2 /* index of triple indirect block */ 208 /* 209 * Truncate the inode oip to at most length size, freeing the 210 * disk blocks. 211 */ 212 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */ 213 214 int 215 lfs_truncate(void *v) 216 { 217 struct vop_truncate_args /* { 218 struct vnode *a_vp; 219 off_t a_length; 220 int a_flags; 221 struct ucred *a_cred; 222 struct proc *a_p; 223 } */ *ap = v; 224 struct vnode *ovp = ap->a_vp; 225 struct genfs_node *gp = VTOG(ovp); 226 daddr_t lastblock; 227 struct inode *oip = VTOI(ovp); 228 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; 229 /* XXX ondisk32 */ 230 int32_t newblks[NDADDR + NIADDR]; 231 off_t length = ap->a_length; 232 struct lfs *fs; 233 struct buf *bp; 234 int offset, size, level; 235 long count, rcount, blocksreleased = 0, real_released = 0; 236 int i, ioflag, nblocks; 237 int aflags, error, allerror = 0; 238 off_t osize; 239 long lastseg; 240 size_t bc; 241 int obufsize, odb; 242 int usepc; 243 struct ufsmount *ump = oip->i_ump; 244 245 if (length < 0) 246 return (EINVAL); 247 248 /* 249 * Just return and not update modification times. 250 */ 251 if (oip->i_size == length) 252 return (0); 253 254 if (ovp->v_type == VLNK && 255 (oip->i_size < ump->um_maxsymlinklen || 256 (ump->um_maxsymlinklen == 0 && 257 oip->i_ffs1_blocks == 0))) { 258 #ifdef DIAGNOSTIC 259 if (length != 0) 260 panic("lfs_truncate: partial truncate of symlink"); 261 #endif 262 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 263 oip->i_size = oip->i_ffs1_size = 0; 264 oip->i_flag |= IN_CHANGE | IN_UPDATE; 265 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 266 } 267 if (oip->i_size == length) { 268 oip->i_flag |= IN_CHANGE | IN_UPDATE; 269 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 270 } 271 #ifdef QUOTA 272 if ((error = getinoquota(oip)) != 0) 273 return (error); 274 #endif 275 fs = oip->i_lfs; 276 lfs_imtime(fs); 277 osize = oip->i_size; 278 ioflag = ap->a_flags; 279 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 280 281 ASSERT_NO_SEGLOCK(fs); 282 /* 283 * Lengthen the size of the file. We must ensure that the 284 * last byte of the file is allocated. Since the smallest 285 * value of osize is 0, length will be at least 1. 286 */ 287 if (osize < length) { 288 if (length > ump->um_maxfilesize) 289 return (EFBIG); 290 aflags = B_CLRBUF; 291 if (ioflag & IO_SYNC) 292 aflags |= B_SYNC; 293 if (usepc) { 294 if (lblkno(fs, osize) < NDADDR && 295 lblkno(fs, osize) != lblkno(fs, length) && 296 blkroundup(fs, osize) != osize) { 297 off_t eob; 298 299 eob = blkroundup(fs, osize); 300 error = ufs_balloc_range(ovp, osize, 301 eob - osize, ap->a_cred, aflags); 302 if (error) 303 return error; 304 if (ioflag & IO_SYNC) { 305 ovp->v_size = eob; 306 simple_lock(&ovp->v_interlock); 307 VOP_PUTPAGES(ovp, 308 trunc_page(osize & fs->lfs_bmask), 309 round_page(eob), 310 PGO_CLEANIT | PGO_SYNCIO); 311 } 312 } 313 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 314 aflags); 315 if (error) { 316 (void) VOP_TRUNCATE(ovp, osize, 317 ioflag & IO_SYNC, 318 ap->a_cred, ap->a_p); 319 return error; 320 } 321 uvm_vnp_setsize(ovp, length); 322 oip->i_flag |= IN_CHANGE | IN_UPDATE; 323 KASSERT(ovp->v_size == oip->i_size); 324 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 325 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 326 } else { 327 error = lfs_reserve(fs, ovp, NULL, 328 btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 329 if (error) 330 return (error); 331 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, 332 aflags, &bp); 333 lfs_reserve(fs, ovp, NULL, 334 -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift)); 335 if (error) 336 return (error); 337 oip->i_ffs1_size = oip->i_size = length; 338 uvm_vnp_setsize(ovp, length); 339 (void) VOP_BWRITE(bp); 340 oip->i_flag |= IN_CHANGE | IN_UPDATE; 341 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 342 return (VOP_UPDATE(ovp, NULL, NULL, 0)); 343 } 344 } 345 346 if ((error = lfs_reserve(fs, ovp, NULL, 347 btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0) 348 return (error); 349 350 /* 351 * Shorten the size of the file. If the file is not being 352 * truncated to a block boundary, the contents of the 353 * partial block following the end of the file must be 354 * zero'ed in case it ever becomes accessible again because 355 * of subsequent file growth. Directories however are not 356 * zero'ed as they should grow back initialized to empty. 357 */ 358 offset = blkoff(fs, length); 359 lastseg = -1; 360 bc = 0; 361 362 lfs_seglock(fs, SEGM_PROT); 363 if (offset == 0) { 364 oip->i_size = oip->i_ffs1_size = length; 365 } else if (!usepc) { 366 lbn = lblkno(fs, length); 367 aflags = B_CLRBUF; 368 if (ioflag & IO_SYNC) 369 aflags |= B_SYNC; 370 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 371 if (error) { 372 lfs_reserve(fs, ovp, NULL, 373 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 374 goto errout; 375 } 376 obufsize = bp->b_bufsize; 377 odb = btofsb(fs, bp->b_bcount); 378 oip->i_size = oip->i_ffs1_size = length; 379 size = blksize(fs, oip, lbn); 380 if (ovp->v_type != VDIR) 381 memset((char *)bp->b_data + offset, 0, 382 (u_int)(size - offset)); 383 allocbuf(bp, size, 1); 384 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 385 simple_lock(&lfs_subsys_lock); 386 locked_queue_bytes -= obufsize - bp->b_bufsize; 387 simple_unlock(&lfs_subsys_lock); 388 } 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 daddr_t lbn; 405 voff_t eoz; 406 407 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 408 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 409 aflags); 410 if (error) { 411 lfs_reserve(fs, ovp, NULL, 412 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 413 goto errout; 414 } 415 lbn = lblkno(fs, length); 416 size = blksize(fs, oip, lbn); 417 eoz = MIN(lblktosize(fs, lbn) + size, osize); 418 uvm_vnp_zerorange(ovp, length, eoz - length); 419 if (round_page(eoz) > round_page(length)) { 420 simple_lock(&ovp->v_interlock); 421 error = VOP_PUTPAGES(ovp, round_page(length), 422 round_page(eoz), 423 PGO_CLEANIT | PGO_DEACTIVATE | 424 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 425 if (error) { 426 lfs_reserve(fs, ovp, NULL, 427 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 428 goto errout; 429 } 430 } 431 } 432 433 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 434 435 oip->i_size = oip->i_ffs1_size = length; 436 uvm_vnp_setsize(ovp, length); 437 /* 438 * Calculate index into inode's block list of 439 * last direct and indirect blocks (if any) 440 * which we want to keep. Lastblock is -1 when 441 * the file is truncated to 0. 442 */ 443 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 444 lastiblock[SINGLE] = lastblock - NDADDR; 445 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 446 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 447 nblocks = btofsb(fs, fs->lfs_bsize); 448 /* 449 * Record changed file and block pointers before we start 450 * freeing blocks. lastiblock values are also normalized to -1 451 * for calls to lfs_indirtrunc below. 452 */ 453 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 454 for (level = TRIPLE; level >= SINGLE; level--) 455 if (lastiblock[level] < 0) { 456 newblks[NDADDR+level] = 0; 457 lastiblock[level] = -1; 458 } 459 for (i = NDADDR - 1; i > lastblock; i--) 460 newblks[i] = 0; 461 462 oip->i_size = oip->i_ffs1_size = osize; 463 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 464 if (error && !allerror) 465 allerror = error; 466 467 /* 468 * Indirect blocks first. 469 */ 470 indir_lbn[SINGLE] = -NDADDR; 471 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 472 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 473 for (level = TRIPLE; level >= SINGLE; level--) { 474 bn = oip->i_ffs1_ib[level]; 475 if (bn != 0) { 476 error = lfs_indirtrunc(oip, indir_lbn[level], 477 bn, lastiblock[level], 478 level, &count, &rcount, 479 &lastseg, &bc, ap->a_p); 480 if (error) 481 allerror = error; 482 real_released += rcount; 483 blocksreleased += count; 484 if (lastiblock[level] < 0) { 485 if (oip->i_ffs1_ib[level] > 0) 486 real_released += nblocks; 487 blocksreleased += nblocks; 488 oip->i_ffs1_ib[level] = 0; 489 lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc); 490 } 491 } 492 if (lastiblock[level] >= 0) 493 goto done; 494 } 495 496 /* 497 * All whole direct blocks or frags. 498 */ 499 for (i = NDADDR - 1; i > lastblock; i--) { 500 long bsize, obsize; 501 502 bn = oip->i_ffs1_db[i]; 503 if (bn == 0) 504 continue; 505 bsize = blksize(fs, oip, i); 506 if (oip->i_ffs1_db[i] > 0) { 507 /* Check for fragment size changes */ 508 obsize = oip->i_lfs_fragsize[i]; 509 real_released += btofsb(fs, obsize); 510 oip->i_lfs_fragsize[i] = 0; 511 } else 512 obsize = 0; 513 blocksreleased += btofsb(fs, bsize); 514 oip->i_ffs1_db[i] = 0; 515 lfs_blkfree(fs, bn, obsize, &lastseg, &bc); 516 } 517 if (lastblock < 0) 518 goto done; 519 520 /* 521 * Finally, look for a change in size of the 522 * last direct block; release any frags. 523 */ 524 bn = oip->i_ffs1_db[lastblock]; 525 if (bn != 0) { 526 long oldspace, newspace; 527 #if 0 528 long olddspace; 529 #endif 530 531 /* 532 * Calculate amount of space we're giving 533 * back as old block size minus new block size. 534 */ 535 oldspace = blksize(fs, oip, lastblock); 536 #if 0 537 olddspace = oip->i_lfs_fragsize[lastblock]; 538 #endif 539 540 oip->i_size = oip->i_ffs1_size = length; 541 newspace = blksize(fs, oip, lastblock); 542 if (newspace == 0) 543 panic("itrunc: newspace"); 544 if (oldspace - newspace > 0) { 545 blocksreleased += btofsb(fs, oldspace - newspace); 546 } 547 #if 0 548 if (bn > 0 && olddspace - newspace > 0) { 549 /* No segment accounting here, just vnode */ 550 real_released += btofsb(fs, olddspace - newspace); 551 } 552 #endif 553 } 554 555 done: 556 /* Finish segment accounting corrections */ 557 lfs_update_seguse(fs, lastseg, bc); 558 #ifdef DIAGNOSTIC 559 for (level = SINGLE; level <= TRIPLE; level++) 560 if ((newblks[NDADDR + level] == 0) != 561 (oip->i_ffs1_ib[level]) == 0) { 562 panic("lfs itrunc1"); 563 } 564 for (i = 0; i < NDADDR; i++) 565 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 566 panic("lfs itrunc2"); 567 } 568 if (length == 0 && 569 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 570 panic("lfs itrunc3"); 571 #endif /* DIAGNOSTIC */ 572 /* 573 * Put back the real size. 574 */ 575 oip->i_size = oip->i_ffs1_size = length; 576 oip->i_lfs_effnblks -= blocksreleased; 577 oip->i_ffs1_blocks -= real_released; 578 simple_lock(&fs->lfs_interlock); 579 fs->lfs_bfree += blocksreleased; 580 simple_unlock(&fs->lfs_interlock); 581 #ifdef DIAGNOSTIC 582 if (oip->i_size == 0 && 583 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 584 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 585 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 586 panic("lfs_truncate: persistent blocks"); 587 } 588 #endif 589 oip->i_flag |= IN_CHANGE; 590 #ifdef QUOTA 591 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 592 #endif 593 lfs_reserve(fs, ovp, NULL, 594 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 595 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 596 errout: 597 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 598 lfs_segunlock(fs); 599 return (allerror ? allerror : error); 600 } 601 602 /* Update segment 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 return error; 621 } 622 623 /* Finish the accounting updates for a segment. */ 624 static int 625 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num) 626 { 627 SEGUSE *sup; 628 struct buf *bp; 629 630 ASSERT_SEGLOCK(fs); 631 if (lastseg < 0 || num == 0) 632 return 0; 633 634 LFS_SEGENTRY(sup, fs, lastseg, bp); 635 if (num > sup->su_nbytes) { 636 printf("lfs_truncate: segment %ld short by %ld\n", 637 lastseg, (long)num - sup->su_nbytes); 638 panic("lfs_truncate: negative bytes"); 639 sup->su_nbytes = num; 640 } 641 sup->su_nbytes -= num; 642 LFS_WRITESEGENTRY(sup, fs, lastseg, bp); 643 644 return 0; 645 } 646 647 /* 648 * Release blocks associated with the inode ip and stored in the indirect 649 * block bn. Blocks are free'd in LIFO order up to (but not including) 650 * lastbn. If level is greater than SINGLE, the block is an indirect block 651 * and recursive calls to indirtrunc must be used to cleanse other indirect 652 * blocks. 653 * 654 * NB: triple indirect blocks are untested. 655 */ 656 static int 657 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 658 daddr_t lastbn, int level, long *countp, 659 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 660 { 661 int i; 662 struct buf *bp; 663 struct lfs *fs = ip->i_lfs; 664 int32_t *bap; /* XXX ondisk32 */ 665 struct vnode *vp; 666 daddr_t nb, nlbn, last; 667 int32_t *copy = NULL; /* XXX ondisk32 */ 668 long blkcount, rblkcount, factor; 669 int nblocks, blocksreleased = 0, real_released = 0; 670 int error = 0, allerror = 0; 671 672 ASSERT_SEGLOCK(fs); 673 /* 674 * Calculate index in current block of last 675 * block to be kept. -1 indicates the entire 676 * block so we need not calculate the index. 677 */ 678 factor = 1; 679 for (i = SINGLE; i < level; i++) 680 factor *= NINDIR(fs); 681 last = lastbn; 682 if (lastbn > 0) 683 last /= factor; 684 nblocks = btofsb(fs, fs->lfs_bsize); 685 /* 686 * Get buffer of block pointers, zero those entries corresponding 687 * to blocks to be free'd, and update on disk copy first. Since 688 * double(triple) indirect before single(double) indirect, calls 689 * to bmap on these blocks will fail. However, we already have 690 * the on disk address, so we have to set the b_blkno field 691 * explicitly instead of letting bread do everything for us. 692 */ 693 vp = ITOV(ip); 694 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 695 if (bp->b_flags & (B_DONE | B_DELWRI)) { 696 /* Braces must be here in case trace evaluates to nothing. */ 697 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 698 } else { 699 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 700 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 701 bp->b_flags |= B_READ; 702 if (bp->b_bcount > bp->b_bufsize) 703 panic("lfs_indirtrunc: bad buffer size"); 704 bp->b_blkno = fsbtodb(fs, dbn); 705 VOP_STRATEGY(vp, bp); 706 error = biowait(bp); 707 } 708 if (error) { 709 brelse(bp); 710 *countp = *rcountp = 0; 711 return (error); 712 } 713 714 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 715 if (lastbn >= 0) { 716 MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK); 717 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 718 memset((caddr_t)&bap[last + 1], 0, 719 /* XXX ondisk32 */ 720 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 721 error = VOP_BWRITE(bp); 722 if (error) 723 allerror = error; 724 bap = copy; 725 } 726 727 /* 728 * Recursively free totally unused blocks. 729 */ 730 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 731 i--, nlbn += factor) { 732 nb = bap[i]; 733 if (nb == 0) 734 continue; 735 if (level > SINGLE) { 736 error = lfs_indirtrunc(ip, nlbn, nb, 737 (daddr_t)-1, level - 1, 738 &blkcount, &rblkcount, 739 lastsegp, bcp, p); 740 if (error) 741 allerror = error; 742 blocksreleased += blkcount; 743 real_released += rblkcount; 744 } 745 lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp); 746 if (bap[i] > 0) 747 real_released += nblocks; 748 blocksreleased += nblocks; 749 } 750 751 /* 752 * Recursively free last partial block. 753 */ 754 if (level > SINGLE && lastbn >= 0) { 755 last = lastbn % factor; 756 nb = bap[i]; 757 if (nb != 0) { 758 error = lfs_indirtrunc(ip, nlbn, nb, 759 last, level - 1, &blkcount, 760 &rblkcount, lastsegp, bcp, p); 761 if (error) 762 allerror = error; 763 real_released += rblkcount; 764 blocksreleased += blkcount; 765 } 766 } 767 768 if (copy != NULL) { 769 FREE(copy, M_TEMP); 770 } else { 771 if (bp->b_flags & B_DELWRI) { 772 LFS_UNLOCK_BUF(bp); 773 fs->lfs_avail += btofsb(fs, bp->b_bcount); 774 wakeup(&fs->lfs_avail); 775 } 776 bp->b_flags |= B_INVAL; 777 brelse(bp); 778 } 779 780 *countp = blocksreleased; 781 *rcountp = real_released; 782 return (allerror); 783 } 784 785 /* 786 * Destroy any in core blocks past the truncation length. 787 * Inlined from vtruncbuf, so that lfs_avail could be updated. 788 * We take the seglock to prevent cleaning from occurring while we are 789 * invalidating blocks. 790 */ 791 static int 792 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 793 { 794 struct buf *bp, *nbp; 795 int s, error; 796 struct lfs *fs; 797 voff_t off; 798 799 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 800 simple_lock(&vp->v_interlock); 801 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 802 if (error) 803 return error; 804 805 fs = VTOI(vp)->i_lfs; 806 s = splbio(); 807 808 ASSERT_SEGLOCK(fs); 809 restart: 810 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 811 nbp = LIST_NEXT(bp, b_vnbufs); 812 if (bp->b_lblkno < lbn) 813 continue; 814 simple_lock(&bp->b_interlock); 815 if (bp->b_flags & B_BUSY) { 816 bp->b_flags |= B_WANTED; 817 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 818 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 819 if (error) { 820 splx(s); 821 return (error); 822 } 823 goto restart; 824 } 825 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 826 if (bp->b_flags & B_DELWRI) { 827 bp->b_flags &= ~B_DELWRI; 828 fs->lfs_avail += btofsb(fs, bp->b_bcount); 829 wakeup(&fs->lfs_avail); 830 } 831 LFS_UNLOCK_BUF(bp); 832 simple_unlock(&bp->b_interlock); 833 brelse(bp); 834 } 835 836 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 837 nbp = LIST_NEXT(bp, b_vnbufs); 838 if (bp->b_lblkno < lbn) 839 continue; 840 simple_lock(&bp->b_interlock); 841 if (bp->b_flags & B_BUSY) { 842 bp->b_flags |= B_WANTED; 843 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 844 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 845 if (error) { 846 splx(s); 847 return (error); 848 } 849 goto restart; 850 } 851 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 852 if (bp->b_flags & B_DELWRI) { 853 bp->b_flags &= ~B_DELWRI; 854 fs->lfs_avail += btofsb(fs, bp->b_bcount); 855 wakeup(&fs->lfs_avail); 856 } 857 LFS_UNLOCK_BUF(bp); 858 simple_unlock(&bp->b_interlock); 859 brelse(bp); 860 } 861 862 splx(s); 863 864 return (0); 865 } 866 867
Indexes created Fri Sep 26 20:09:58 GMT 2025