lfs_inode.c revision 1.90.2.5
1 /* $NetBSD: lfs_inode.c,v 1.90.2.5 2006/05/20 22:38:57 riz 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 1.90.2.5 2006/05/20 22:38:57 riz 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 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 *, struct inode *ip, 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 *, struct inode *, 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 if (ovp != fs->lfs_ivnode) 363 lfs_seglock(fs, SEGM_PROT); 364 if (offset == 0) { 365 oip->i_size = oip->i_ffs1_size = length; 366 } else if (!usepc) { 367 lbn = lblkno(fs, length); 368 aflags = B_CLRBUF; 369 if (ioflag & IO_SYNC) 370 aflags |= B_SYNC; 371 error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp); 372 if (error) { 373 lfs_reserve(fs, ovp, NULL, 374 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 375 goto errout; 376 } 377 obufsize = bp->b_bufsize; 378 odb = btofsb(fs, bp->b_bcount); 379 oip->i_size = oip->i_ffs1_size = length; 380 size = blksize(fs, oip, lbn); 381 if (ovp->v_type != VDIR) 382 memset((char *)bp->b_data + offset, 0, 383 (u_int)(size - offset)); 384 allocbuf(bp, size, 1); 385 if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) { 386 simple_lock(&lfs_subsys_lock); 387 locked_queue_bytes -= obufsize - bp->b_bufsize; 388 simple_unlock(&lfs_subsys_lock); 389 } 390 if (bp->b_flags & B_DELWRI) 391 fs->lfs_avail += odb - btofsb(fs, size); 392 (void) VOP_BWRITE(bp); 393 } else { /* vp->v_type == VREG && length < osize && offset != 0 */ 394 /* 395 * When truncating a regular file down to a non-block-aligned 396 * size, we must zero the part of last block which is past 397 * the new EOF. We must synchronously flush the zeroed pages 398 * to disk since the new pages will be invalidated as soon 399 * as we inform the VM system of the new, smaller size. 400 * We must do this before acquiring the GLOCK, since fetching 401 * the pages will acquire the GLOCK internally. 402 * So there is a window where another thread could see a whole 403 * zeroed page past EOF, but that's life. 404 */ 405 daddr_t xlbn; 406 voff_t eoz; 407 408 aflags = ioflag & IO_SYNC ? B_SYNC : 0; 409 error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred, 410 aflags); 411 if (error) { 412 lfs_reserve(fs, ovp, NULL, 413 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 414 goto errout; 415 } 416 xlbn = lblkno(fs, length); 417 size = blksize(fs, oip, xlbn); 418 eoz = MIN(lblktosize(fs, xlbn) + size, osize); 419 uvm_vnp_zerorange(ovp, length, eoz - length); 420 if (round_page(eoz) > round_page(length)) { 421 simple_lock(&ovp->v_interlock); 422 error = VOP_PUTPAGES(ovp, round_page(length), 423 round_page(eoz), 424 PGO_CLEANIT | PGO_DEACTIVATE | 425 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 426 if (error) { 427 lfs_reserve(fs, ovp, NULL, 428 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 429 goto errout; 430 } 431 } 432 } 433 434 lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL); 435 436 oip->i_size = oip->i_ffs1_size = length; 437 uvm_vnp_setsize(ovp, length); 438 /* 439 * Calculate index into inode's block list of 440 * last direct and indirect blocks (if any) 441 * which we want to keep. Lastblock is -1 when 442 * the file is truncated to 0. 443 */ 444 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 445 if (length > QUAD_MAX - fs->lfs_bsize) 446 lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize); 447 else 448 lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1; 449 lastiblock[SINGLE] = lastblock - NDADDR; 450 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); 451 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); 452 nblocks = btofsb(fs, fs->lfs_bsize); 453 /* 454 * Record changed file and block pointers before we start 455 * freeing blocks. lastiblock values are also normalized to -1 456 * for calls to lfs_indirtrunc below. 457 */ 458 memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks); 459 for (level = TRIPLE; level >= SINGLE; level--) 460 if (lastiblock[level] < 0) { 461 newblks[NDADDR+level] = 0; 462 lastiblock[level] = -1; 463 } 464 for (i = NDADDR - 1; i > lastblock; i--) 465 newblks[i] = 0; 466 467 oip->i_size = oip->i_ffs1_size = osize; 468 error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0); 469 if (error && !allerror) 470 allerror = error; 471 472 /* 473 * Indirect blocks first. 474 */ 475 indir_lbn[SINGLE] = -NDADDR; 476 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; 477 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; 478 for (level = TRIPLE; level >= SINGLE; level--) { 479 bn = oip->i_ffs1_ib[level]; 480 if (bn != 0) { 481 error = lfs_indirtrunc(oip, indir_lbn[level], 482 bn, lastiblock[level], 483 level, &count, &rcount, 484 &lastseg, &bc, ap->a_p); 485 if (error) 486 allerror = error; 487 real_released += rcount; 488 blocksreleased += count; 489 if (lastiblock[level] < 0) { 490 if (oip->i_ffs1_ib[level] > 0) 491 real_released += nblocks; 492 blocksreleased += nblocks; 493 oip->i_ffs1_ib[level] = 0; 494 lfs_blkfree(fs, oip, bn, fs->lfs_bsize, 495 &lastseg, &bc); 496 lfs_deregister_block(ovp, bn); 497 } 498 } 499 if (lastiblock[level] >= 0) 500 goto done; 501 } 502 503 /* 504 * All whole direct blocks or frags. 505 */ 506 for (i = NDADDR - 1; i > lastblock; i--) { 507 long bsize, obsize; 508 509 bn = oip->i_ffs1_db[i]; 510 if (bn == 0) 511 continue; 512 bsize = blksize(fs, oip, i); 513 if (oip->i_ffs1_db[i] > 0) { 514 /* Check for fragment size changes */ 515 obsize = oip->i_lfs_fragsize[i]; 516 real_released += btofsb(fs, obsize); 517 oip->i_lfs_fragsize[i] = 0; 518 } else 519 obsize = 0; 520 blocksreleased += btofsb(fs, bsize); 521 oip->i_ffs1_db[i] = 0; 522 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 523 lfs_deregister_block(ovp, bn); 524 } 525 if (lastblock < 0) 526 goto done; 527 528 /* 529 * Finally, look for a change in size of the 530 * last direct block; release any frags. 531 */ 532 bn = oip->i_ffs1_db[lastblock]; 533 if (bn != 0) { 534 long oldspace, newspace; 535 #if 0 536 long olddspace; 537 #endif 538 539 /* 540 * Calculate amount of space we're giving 541 * back as old block size minus new block size. 542 */ 543 oldspace = blksize(fs, oip, lastblock); 544 #if 0 545 olddspace = oip->i_lfs_fragsize[lastblock]; 546 #endif 547 548 oip->i_size = oip->i_ffs1_size = length; 549 newspace = blksize(fs, oip, lastblock); 550 if (newspace == 0) 551 panic("itrunc: newspace"); 552 if (oldspace - newspace > 0) { 553 blocksreleased += btofsb(fs, oldspace - newspace); 554 } 555 #if 0 556 if (bn > 0 && olddspace - newspace > 0) { 557 /* No segment accounting here, just vnode */ 558 real_released += btofsb(fs, olddspace - newspace); 559 } 560 #endif 561 } 562 563 done: 564 /* Finish segment accounting corrections */ 565 lfs_update_seguse(fs, oip, lastseg, bc); 566 #ifdef DIAGNOSTIC 567 for (level = SINGLE; level <= TRIPLE; level++) 568 if ((newblks[NDADDR + level] == 0) != 569 (oip->i_ffs1_ib[level]) == 0) { 570 panic("lfs itrunc1"); 571 } 572 for (i = 0; i < NDADDR; i++) 573 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 574 panic("lfs itrunc2"); 575 } 576 if (length == 0 && 577 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 578 panic("lfs itrunc3"); 579 #endif /* DIAGNOSTIC */ 580 /* 581 * Put back the real size. 582 */ 583 oip->i_size = oip->i_ffs1_size = length; 584 oip->i_lfs_effnblks -= blocksreleased; 585 oip->i_ffs1_blocks -= real_released; 586 simple_lock(&fs->lfs_interlock); 587 fs->lfs_bfree += blocksreleased; 588 simple_unlock(&fs->lfs_interlock); 589 #ifdef DIAGNOSTIC 590 if (oip->i_size == 0 && 591 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 592 printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n", 593 oip->i_ffs1_blocks, oip->i_lfs_effnblks); 594 panic("lfs_truncate: persistent blocks"); 595 } 596 #endif 597 598 /* 599 * If we truncated to zero, take us off the paging queue. 600 */ 601 simple_lock(&fs->lfs_interlock); 602 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 603 oip->i_flags &= ~IN_PAGING; 604 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 605 } 606 simple_unlock(&fs->lfs_interlock); 607 608 oip->i_flag |= IN_CHANGE; 609 #ifdef QUOTA 610 (void) chkdq(oip, -blocksreleased, NOCRED, 0); 611 #endif 612 lfs_reserve(fs, ovp, NULL, 613 -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift)); 614 lockmgr(&gp->g_glock, LK_RELEASE, NULL); 615 errout: 616 oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1; 617 if (ovp != fs->lfs_ivnode) 618 lfs_segunlock(fs); 619 return (allerror ? allerror : error); 620 } 621 622 /* Update segment and avail usage information when removing a block. */ 623 static int 624 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 625 size_t bsize, long *lastseg, size_t *num) 626 { 627 long seg; 628 int error = 0; 629 630 ASSERT_SEGLOCK(fs); 631 bsize = fragroundup(fs, bsize); 632 if (daddr > 0) { 633 if (*lastseg != (seg = dtosn(fs, daddr))) { 634 error = lfs_update_seguse(fs, ip, *lastseg, *num); 635 *num = bsize; 636 *lastseg = seg; 637 } else 638 *num += bsize; 639 } 640 641 return error; 642 } 643 644 /* Finish the accounting updates for a segment. */ 645 static int 646 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 647 { 648 struct segdelta *sd; 649 struct vnode *vp; 650 651 ASSERT_SEGLOCK(fs); 652 if (lastseg < 0 || num == 0) 653 return 0; 654 655 vp = ITOV(ip); 656 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list) 657 if (sd->segnum == lastseg) 658 break; 659 if (sd == NULL) { 660 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK); 661 sd->segnum = lastseg; 662 sd->num = 0; 663 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list); 664 } 665 sd->num += num; 666 667 return 0; 668 } 669 670 static void 671 lfs_finalize_seguse(struct lfs *fs, void *v) 672 { 673 SEGUSE *sup; 674 struct buf *bp; 675 struct segdelta *sd; 676 LIST_HEAD(, segdelta) *hd = v; 677 678 ASSERT_SEGLOCK(fs); 679 while((sd = LIST_FIRST(hd)) != NULL) { 680 LIST_REMOVE(sd, list); 681 LFS_SEGENTRY(sup, fs, sd->segnum, bp); 682 if (sd->num > sup->su_nbytes) { 683 printf("lfs_finalize_seguse: segment %ld short by %ld\n", 684 sd->segnum, (long)(sd->num - sup->su_nbytes)); 685 panic("lfs_finalize_seguse: negative bytes"); 686 sup->su_nbytes = sd->num; 687 } 688 sup->su_nbytes -= sd->num; 689 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp); 690 free(sd, M_SEGMENT); 691 } 692 } 693 694 /* Finish the accounting updates for a segment. */ 695 void 696 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip) 697 { 698 ASSERT_SEGLOCK(fs); 699 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd); 700 } 701 702 /* Finish the accounting updates for a segment. */ 703 void 704 lfs_finalize_fs_seguse(struct lfs *fs) 705 { 706 ASSERT_SEGLOCK(fs); 707 lfs_finalize_seguse(fs, &fs->lfs_segdhd); 708 } 709 710 /* 711 * Release blocks associated with the inode ip and stored in the indirect 712 * block bn. Blocks are free'd in LIFO order up to (but not including) 713 * lastbn. If level is greater than SINGLE, the block is an indirect block 714 * and recursive calls to indirtrunc must be used to cleanse other indirect 715 * blocks. 716 * 717 * NB: triple indirect blocks are untested. 718 */ 719 static int 720 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, 721 daddr_t lastbn, int level, long *countp, 722 long *rcountp, long *lastsegp, size_t *bcp, struct proc *p) 723 { 724 int i; 725 struct buf *bp; 726 struct lfs *fs = ip->i_lfs; 727 int32_t *bap; /* XXX ondisk32 */ 728 struct vnode *vp; 729 daddr_t nb, nlbn, last; 730 int32_t *copy = NULL; /* XXX ondisk32 */ 731 long blkcount, rblkcount, factor; 732 int nblocks, blocksreleased = 0, real_released = 0; 733 int error = 0, allerror = 0; 734 735 ASSERT_SEGLOCK(fs); 736 /* 737 * Calculate index in current block of last 738 * block to be kept. -1 indicates the entire 739 * block so we need not calculate the index. 740 */ 741 factor = 1; 742 for (i = SINGLE; i < level; i++) 743 factor *= NINDIR(fs); 744 last = lastbn; 745 if (lastbn > 0) 746 last /= factor; 747 nblocks = btofsb(fs, fs->lfs_bsize); 748 /* 749 * Get buffer of block pointers, zero those entries corresponding 750 * to blocks to be free'd, and update on disk copy first. Since 751 * double(triple) indirect before single(double) indirect, calls 752 * to bmap on these blocks will fail. However, we already have 753 * the on disk address, so we have to set the b_blkno field 754 * explicitly instead of letting bread do everything for us. 755 */ 756 vp = ITOV(ip); 757 bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0); 758 if (bp->b_flags & (B_DONE | B_DELWRI)) { 759 /* Braces must be here in case trace evaluates to nothing. */ 760 trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn); 761 } else { 762 trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn); 763 p->p_stats->p_ru.ru_inblock++; /* pay for read */ 764 bp->b_flags |= B_READ; 765 if (bp->b_bcount > bp->b_bufsize) 766 panic("lfs_indirtrunc: bad buffer size"); 767 bp->b_blkno = fsbtodb(fs, dbn); 768 VOP_STRATEGY(vp, bp); 769 error = biowait(bp); 770 } 771 if (error) { 772 brelse(bp); 773 *countp = *rcountp = 0; 774 return (error); 775 } 776 777 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 778 if (lastbn >= 0) { 779 copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK); 780 memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize); 781 memset((caddr_t)&bap[last + 1], 0, 782 /* XXX ondisk32 */ 783 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 784 error = VOP_BWRITE(bp); 785 if (error) 786 allerror = error; 787 bap = copy; 788 } 789 790 /* 791 * Recursively free totally unused blocks. 792 */ 793 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 794 i--, nlbn += factor) { 795 nb = bap[i]; 796 if (nb == 0) 797 continue; 798 if (level > SINGLE) { 799 error = lfs_indirtrunc(ip, nlbn, nb, 800 (daddr_t)-1, level - 1, 801 &blkcount, &rblkcount, 802 lastsegp, bcp, p); 803 if (error) 804 allerror = error; 805 blocksreleased += blkcount; 806 real_released += rblkcount; 807 } 808 lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp); 809 if (bap[i] > 0) 810 real_released += nblocks; 811 blocksreleased += nblocks; 812 } 813 814 /* 815 * Recursively free last partial block. 816 */ 817 if (level > SINGLE && lastbn >= 0) { 818 last = lastbn % factor; 819 nb = bap[i]; 820 if (nb != 0) { 821 error = lfs_indirtrunc(ip, nlbn, nb, 822 last, level - 1, &blkcount, 823 &rblkcount, lastsegp, bcp, p); 824 if (error) 825 allerror = error; 826 real_released += rblkcount; 827 blocksreleased += blkcount; 828 } 829 } 830 831 if (copy != NULL) { 832 lfs_free(fs, copy, LFS_NB_IBLOCK); 833 } else { 834 if (bp->b_flags & B_DELWRI) { 835 LFS_UNLOCK_BUF(bp); 836 fs->lfs_avail += btofsb(fs, bp->b_bcount); 837 wakeup(&fs->lfs_avail); 838 } 839 bp->b_flags |= B_INVAL; 840 brelse(bp); 841 } 842 843 *countp = blocksreleased; 844 *rcountp = real_released; 845 return (allerror); 846 } 847 848 /* 849 * Destroy any in core blocks past the truncation length. 850 * Inlined from vtruncbuf, so that lfs_avail could be updated. 851 * We take the seglock to prevent cleaning from occurring while we are 852 * invalidating blocks. 853 */ 854 static int 855 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo) 856 { 857 struct buf *bp, *nbp; 858 int s, error; 859 struct lfs *fs; 860 voff_t off; 861 862 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 863 simple_lock(&vp->v_interlock); 864 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 865 if (error) 866 return error; 867 868 fs = VTOI(vp)->i_lfs; 869 s = splbio(); 870 871 ASSERT_SEGLOCK(fs); 872 restart: 873 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 874 nbp = LIST_NEXT(bp, b_vnbufs); 875 if (bp->b_lblkno < lbn) 876 continue; 877 simple_lock(&bp->b_interlock); 878 if (bp->b_flags & B_BUSY) { 879 bp->b_flags |= B_WANTED; 880 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 881 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 882 if (error) { 883 splx(s); 884 return (error); 885 } 886 goto restart; 887 } 888 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 889 if (bp->b_flags & B_DELWRI) { 890 bp->b_flags &= ~B_DELWRI; 891 fs->lfs_avail += btofsb(fs, bp->b_bcount); 892 wakeup(&fs->lfs_avail); 893 } 894 LFS_UNLOCK_BUF(bp); 895 simple_unlock(&bp->b_interlock); 896 brelse(bp); 897 } 898 899 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 900 nbp = LIST_NEXT(bp, b_vnbufs); 901 if (bp->b_lblkno < lbn) 902 continue; 903 simple_lock(&bp->b_interlock); 904 if (bp->b_flags & B_BUSY) { 905 bp->b_flags |= B_WANTED; 906 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 907 "lfs_vtruncbuf", slptimeo, &bp->b_interlock); 908 if (error) { 909 splx(s); 910 return (error); 911 } 912 goto restart; 913 } 914 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 915 if (bp->b_flags & B_DELWRI) { 916 bp->b_flags &= ~B_DELWRI; 917 fs->lfs_avail += btofsb(fs, bp->b_bcount); 918 wakeup(&fs->lfs_avail); 919 } 920 LFS_UNLOCK_BUF(bp); 921 simple_unlock(&bp->b_interlock); 922 brelse(bp); 923 } 924 925 splx(s); 926 927 return (0); 928 } 929 930
Indexes created Fri Oct 03 02:09:57 GMT 2025