lfs_inode.c revision 1.142
1 /* $NetBSD: lfs_inode.c,v 1.142 2015/08/02 18:12:41 dholland 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /* 32 * Copyright (c) 1986, 1989, 1991, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.142 2015/08/02 18:12:41 dholland Exp $"); 64 65 #if defined(_KERNEL_OPT) 66 #include "opt_quota.h" 67 #endif 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/mount.h> 72 #include <sys/malloc.h> 73 #include <sys/proc.h> 74 #include <sys/file.h> 75 #include <sys/buf.h> 76 #include <sys/vnode.h> 77 #include <sys/kernel.h> 78 #include <sys/trace.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kauth.h> 81 82 #include <ufs/lfs/ulfs_quotacommon.h> 83 #include <ufs/lfs/ulfs_inode.h> 84 #include <ufs/lfs/ulfsmount.h> 85 #include <ufs/lfs/ulfs_extern.h> 86 87 #include <ufs/lfs/lfs.h> 88 #include <ufs/lfs/lfs_accessors.h> 89 #include <ufs/lfs/lfs_extern.h> 90 #include <ufs/lfs/lfs_kernel.h> 91 92 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t); 93 static int lfs_indirtrunc(struct inode *, daddr_t, daddr_t, 94 daddr_t, int, daddr_t *, daddr_t *, 95 long *, size_t *); 96 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *); 97 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int); 98 99 /* Search a block for a specific dinode. */ 100 struct ulfs1_dinode * 101 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp) 102 { 103 struct ulfs1_dinode *dip = (struct ulfs1_dinode *)bp->b_data; 104 struct ulfs1_dinode *ldip, *fin; 105 106 ASSERT_NO_SEGLOCK(fs); 107 /* 108 * Read the inode block backwards, since later versions of the 109 * inode will supercede earlier ones. Though it is unlikely, it is 110 * possible that the same inode will appear in the same inode block. 111 */ 112 fin = dip + LFS_INOPB(fs); 113 for (ldip = fin - 1; ldip >= dip; --ldip) 114 if (ldip->di_inumber == ino) 115 return (ldip); 116 117 printf("searched %d entries\n", (int)(fin - dip)); 118 printf("offset is 0x%jx (seg %d)\n", (uintmax_t)lfs_sb_getoffset(fs), 119 lfs_dtosn(fs, lfs_sb_getoffset(fs))); 120 printf("block is 0x%jx (seg %d)\n", 121 (uintmax_t)LFS_DBTOFSB(fs, bp->b_blkno), 122 lfs_dtosn(fs, LFS_DBTOFSB(fs, bp->b_blkno))); 123 124 return NULL; 125 } 126 127 int 128 lfs_update(struct vnode *vp, const struct timespec *acc, 129 const struct timespec *mod, int updflags) 130 { 131 struct inode *ip; 132 struct lfs *fs = VFSTOULFS(vp->v_mount)->um_lfs; 133 int flags; 134 135 ASSERT_NO_SEGLOCK(fs); 136 if (vp->v_mount->mnt_flag & MNT_RDONLY) 137 return (0); 138 ip = VTOI(vp); 139 140 /* 141 * If we are called from vinvalbuf, and the file's blocks have 142 * already been scheduled for writing, but the writes have not 143 * yet completed, lfs_vflush will not be called, and vinvalbuf 144 * will cause a panic. So, we must wait until any pending write 145 * for our inode completes, if we are called with UPDATE_WAIT set. 146 */ 147 mutex_enter(vp->v_interlock); 148 while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT && 149 WRITEINPROG(vp)) { 150 DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d" 151 " (in progress)\n", ip->i_number)); 152 cv_wait(&vp->v_cv, vp->v_interlock); 153 } 154 mutex_exit(vp->v_interlock); 155 LFS_ITIMES(ip, acc, mod, NULL); 156 if (updflags & UPDATE_CLOSE) 157 flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING); 158 else 159 flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING); 160 if (flags == 0) 161 return (0); 162 163 /* If sync, push back the vnode and any dirty blocks it may have. */ 164 if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) { 165 /* Avoid flushing VU_DIROP. */ 166 mutex_enter(&lfs_lock); 167 ++fs->lfs_diropwait; 168 while (vp->v_uflag & VU_DIROP) { 169 DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d" 170 " (dirops)\n", ip->i_number)); 171 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags" 172 " 0x%x\n", 173 vp->v_iflag | vp->v_vflag | vp->v_uflag, 174 ip->i_flag)); 175 if (fs->lfs_dirops == 0) 176 lfs_flush_fs(fs, SEGM_SYNC); 177 else 178 mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync", 179 0, &lfs_lock); 180 /* XXX KS - by falling out here, are we writing the vn 181 twice? */ 182 } 183 --fs->lfs_diropwait; 184 mutex_exit(&lfs_lock); 185 return lfs_vflush(vp); 186 } 187 return 0; 188 } 189 190 #define SINGLE 0 /* index of single indirect block */ 191 #define DOUBLE 1 /* index of double indirect block */ 192 #define TRIPLE 2 /* index of triple indirect block */ 193 /* 194 * Truncate the inode oip to at most length size, freeing the 195 * disk blocks. 196 */ 197 /* VOP_BWRITE 1 + ULFS_NIADDR + lfs_balloc == 2 + 2*ULFS_NIADDR times */ 198 199 int 200 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred) 201 { 202 daddr_t lastblock; 203 struct inode *oip = VTOI(ovp); 204 daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR]; 205 /* XXX ondisk32 */ 206 int32_t newblks[ULFS_NDADDR + ULFS_NIADDR]; 207 struct lfs *fs; 208 struct buf *bp; 209 int offset, size, level; 210 long count, rcount; 211 daddr_t blocksreleased = 0, real_released = 0; 212 int i, nblocks; 213 int aflags, error, allerror = 0; 214 off_t osize; 215 long lastseg; 216 size_t bc; 217 int obufsize, odb; 218 int usepc; 219 220 if (ovp->v_type == VCHR || ovp->v_type == VBLK || 221 ovp->v_type == VFIFO || ovp->v_type == VSOCK) { 222 KASSERT(oip->i_size == 0); 223 return 0; 224 } 225 226 if (length < 0) 227 return (EINVAL); 228 229 /* 230 * Just return and not update modification times. 231 */ 232 if (oip->i_size == length) { 233 /* still do a uvm_vnp_setsize() as writesize may be larger */ 234 uvm_vnp_setsize(ovp, length); 235 return (0); 236 } 237 238 fs = oip->i_lfs; 239 240 if (ovp->v_type == VLNK && 241 (oip->i_size < fs->um_maxsymlinklen || 242 (fs->um_maxsymlinklen == 0 && 243 oip->i_ffs1_blocks == 0))) { 244 #ifdef DIAGNOSTIC 245 if (length != 0) 246 panic("lfs_truncate: partial truncate of symlink"); 247 #endif 248 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size); 249 oip->i_size = oip->i_ffs1_size = 0; 250 oip->i_flag |= IN_CHANGE | IN_UPDATE; 251 return (lfs_update(ovp, NULL, NULL, 0)); 252 } 253 if (oip->i_size == length) { 254 oip->i_flag |= IN_CHANGE | IN_UPDATE; 255 return (lfs_update(ovp, NULL, NULL, 0)); 256 } 257 lfs_imtime(fs); 258 osize = oip->i_size; 259 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode); 260 261 ASSERT_NO_SEGLOCK(fs); 262 /* 263 * Lengthen the size of the file. We must ensure that the 264 * last byte of the file is allocated. Since the smallest 265 * value of osize is 0, length will be at least 1. 266 */ 267 if (osize < length) { 268 if (length > fs->um_maxfilesize) 269 return (EFBIG); 270 aflags = B_CLRBUF; 271 if (ioflag & IO_SYNC) 272 aflags |= B_SYNC; 273 if (usepc) { 274 if (lfs_lblkno(fs, osize) < ULFS_NDADDR && 275 lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) && 276 lfs_blkroundup(fs, osize) != osize) { 277 off_t eob; 278 279 eob = lfs_blkroundup(fs, osize); 280 uvm_vnp_setwritesize(ovp, eob); 281 error = ulfs_balloc_range(ovp, osize, 282 eob - osize, cred, aflags); 283 if (error) { 284 (void) lfs_truncate(ovp, osize, 285 ioflag & IO_SYNC, cred); 286 return error; 287 } 288 if (ioflag & IO_SYNC) { 289 mutex_enter(ovp->v_interlock); 290 VOP_PUTPAGES(ovp, 291 trunc_page(osize & lfs_sb_getbmask(fs)), 292 round_page(eob), 293 PGO_CLEANIT | PGO_SYNCIO); 294 } 295 } 296 uvm_vnp_setwritesize(ovp, length); 297 error = ulfs_balloc_range(ovp, length - 1, 1, cred, 298 aflags); 299 if (error) { 300 (void) lfs_truncate(ovp, osize, 301 ioflag & IO_SYNC, cred); 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 = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 308 return (lfs_update(ovp, NULL, NULL, 0)); 309 } else { 310 error = lfs_reserve(fs, ovp, NULL, 311 lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 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 -lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs))); 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->b_vp, bp); 323 oip->i_flag |= IN_CHANGE | IN_UPDATE; 324 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 325 return (lfs_update(ovp, NULL, NULL, 0)); 326 } 327 } 328 329 if ((error = lfs_reserve(fs, ovp, NULL, 330 lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 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 = lfs_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 = lfs_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 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 358 goto errout; 359 } 360 obufsize = bp->b_bufsize; 361 odb = lfs_btofsb(fs, bp->b_bcount); 362 oip->i_size = oip->i_ffs1_size = length; 363 size = lfs_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) != 0 && bp->b_iodone == NULL) { 369 mutex_enter(&lfs_lock); 370 locked_queue_bytes -= obufsize - bp->b_bufsize; 371 mutex_exit(&lfs_lock); 372 } 373 if (bp->b_oflags & BO_DELWRI) { 374 lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size)); 375 /* XXX shouldn't this wake up on lfs_availsleep? */ 376 } 377 (void) VOP_BWRITE(bp->b_vp, 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 = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags); 395 if (error) { 396 lfs_reserve(fs, ovp, NULL, 397 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 398 goto errout; 399 } 400 xlbn = lfs_lblkno(fs, length); 401 size = lfs_blksize(fs, oip, xlbn); 402 eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize); 403 ubc_zerorange(&ovp->v_uobj, length, eoz - length, 404 UBC_UNMAP_FLAG(ovp)); 405 if (round_page(eoz) > round_page(length)) { 406 mutex_enter(ovp->v_interlock); 407 error = VOP_PUTPAGES(ovp, round_page(length), 408 round_page(eoz), 409 PGO_CLEANIT | PGO_DEACTIVATE | 410 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); 411 if (error) { 412 lfs_reserve(fs, ovp, NULL, 413 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 414 goto errout; 415 } 416 } 417 } 418 419 genfs_node_wrlock(ovp); 420 421 oip->i_size = oip->i_ffs1_size = length; 422 uvm_vnp_setsize(ovp, length); 423 424 /* 425 * Calculate index into inode's block list of 426 * last direct and indirect blocks (if any) 427 * which we want to keep. Lastblock is -1 when 428 * the file is truncated to 0. 429 */ 430 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */ 431 if (length > QUAD_MAX - lfs_sb_getbsize(fs)) 432 lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs)); 433 else 434 lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1; 435 lastiblock[SINGLE] = lastblock - ULFS_NDADDR; 436 lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs); 437 lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs); 438 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 439 /* 440 * Record changed file and block pointers before we start 441 * freeing blocks. lastiblock values are also normalized to -1 442 * for calls to lfs_indirtrunc below. 443 */ 444 memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks); 445 for (level = TRIPLE; level >= SINGLE; level--) 446 if (lastiblock[level] < 0) { 447 newblks[ULFS_NDADDR+level] = 0; 448 lastiblock[level] = -1; 449 } 450 for (i = ULFS_NDADDR - 1; i > lastblock; i--) 451 newblks[i] = 0; 452 453 oip->i_size = oip->i_ffs1_size = osize; 454 error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0); 455 if (error && !allerror) 456 allerror = error; 457 458 /* 459 * Indirect blocks first. 460 */ 461 indir_lbn[SINGLE] = -ULFS_NDADDR; 462 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1; 463 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1; 464 for (level = TRIPLE; level >= SINGLE; level--) { 465 bn = oip->i_ffs1_ib[level]; 466 if (bn != 0) { 467 error = lfs_indirtrunc(oip, indir_lbn[level], 468 bn, lastiblock[level], 469 level, &count, &rcount, 470 &lastseg, &bc); 471 if (error) 472 allerror = error; 473 real_released += rcount; 474 blocksreleased += count; 475 if (lastiblock[level] < 0) { 476 if (oip->i_ffs1_ib[level] > 0) 477 real_released += nblocks; 478 blocksreleased += nblocks; 479 oip->i_ffs1_ib[level] = 0; 480 lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs), 481 &lastseg, &bc); 482 lfs_deregister_block(ovp, bn); 483 } 484 } 485 if (lastiblock[level] >= 0) 486 goto done; 487 } 488 489 /* 490 * All whole direct blocks or frags. 491 */ 492 for (i = ULFS_NDADDR - 1; i > lastblock; i--) { 493 long bsize, obsize; 494 495 bn = oip->i_ffs1_db[i]; 496 if (bn == 0) 497 continue; 498 bsize = lfs_blksize(fs, oip, i); 499 if (oip->i_ffs1_db[i] > 0) { 500 /* Check for fragment size changes */ 501 obsize = oip->i_lfs_fragsize[i]; 502 real_released += lfs_btofsb(fs, obsize); 503 oip->i_lfs_fragsize[i] = 0; 504 } else 505 obsize = 0; 506 blocksreleased += lfs_btofsb(fs, bsize); 507 oip->i_ffs1_db[i] = 0; 508 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc); 509 lfs_deregister_block(ovp, bn); 510 } 511 if (lastblock < 0) 512 goto done; 513 514 /* 515 * Finally, look for a change in size of the 516 * last direct block; release any frags. 517 */ 518 bn = oip->i_ffs1_db[lastblock]; 519 if (bn != 0) { 520 long oldspace, newspace; 521 #if 0 522 long olddspace; 523 #endif 524 525 /* 526 * Calculate amount of space we're giving 527 * back as old block size minus new block size. 528 */ 529 oldspace = lfs_blksize(fs, oip, lastblock); 530 #if 0 531 olddspace = oip->i_lfs_fragsize[lastblock]; 532 #endif 533 534 oip->i_size = oip->i_ffs1_size = length; 535 newspace = lfs_blksize(fs, oip, lastblock); 536 if (newspace == 0) 537 panic("itrunc: newspace"); 538 if (oldspace - newspace > 0) { 539 blocksreleased += lfs_btofsb(fs, oldspace - newspace); 540 } 541 #if 0 542 if (bn > 0 && olddspace - newspace > 0) { 543 /* No segment accounting here, just vnode */ 544 real_released += lfs_btofsb(fs, olddspace - newspace); 545 } 546 #endif 547 } 548 549 done: 550 /* Finish segment accounting corrections */ 551 lfs_update_seguse(fs, oip, lastseg, bc); 552 #ifdef DIAGNOSTIC 553 for (level = SINGLE; level <= TRIPLE; level++) 554 if ((newblks[ULFS_NDADDR + level] == 0) != 555 ((oip->i_ffs1_ib[level]) == 0)) { 556 panic("lfs itrunc1"); 557 } 558 for (i = 0; i < ULFS_NDADDR; i++) 559 if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) { 560 panic("lfs itrunc2"); 561 } 562 if (length == 0 && 563 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) 564 panic("lfs itrunc3"); 565 #endif /* DIAGNOSTIC */ 566 /* 567 * Put back the real size. 568 */ 569 oip->i_size = oip->i_ffs1_size = length; 570 oip->i_lfs_effnblks -= blocksreleased; 571 oip->i_ffs1_blocks -= real_released; 572 mutex_enter(&lfs_lock); 573 lfs_sb_addbfree(fs, blocksreleased); 574 mutex_exit(&lfs_lock); 575 #ifdef DIAGNOSTIC 576 if (oip->i_size == 0 && 577 (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) { 578 printf("lfs_truncate: truncate to 0 but %d blks/%jd effblks\n", 579 oip->i_ffs1_blocks, (intmax_t)oip->i_lfs_effnblks); 580 panic("lfs_truncate: persistent blocks"); 581 } 582 #endif 583 584 /* 585 * If we truncated to zero, take us off the paging queue. 586 */ 587 mutex_enter(&lfs_lock); 588 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) { 589 oip->i_flags &= ~IN_PAGING; 590 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain); 591 } 592 mutex_exit(&lfs_lock); 593 594 oip->i_flag |= IN_CHANGE; 595 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2) 596 (void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0); 597 #endif 598 lfs_reserve(fs, ovp, NULL, 599 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs))); 600 genfs_node_unlock(ovp); 601 errout: 602 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1; 603 if (ovp != fs->lfs_ivnode) 604 lfs_segunlock(fs); 605 return (allerror ? allerror : error); 606 } 607 608 /* Update segment and avail usage information when removing a block. */ 609 static int 610 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr, 611 size_t bsize, long *lastseg, size_t *num) 612 { 613 long seg; 614 int error = 0; 615 616 ASSERT_SEGLOCK(fs); 617 bsize = lfs_fragroundup(fs, bsize); 618 if (daddr > 0) { 619 if (*lastseg != (seg = lfs_dtosn(fs, daddr))) { 620 error = lfs_update_seguse(fs, ip, *lastseg, *num); 621 *num = bsize; 622 *lastseg = seg; 623 } else 624 *num += bsize; 625 } 626 627 return error; 628 } 629 630 /* Finish the accounting updates for a segment. */ 631 static int 632 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num) 633 { 634 struct segdelta *sd; 635 636 ASSERT_SEGLOCK(fs); 637 if (lastseg < 0 || num == 0) 638 return 0; 639 640 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, daddr_t *countp, 706 daddr_t *rcountp, long *lastsegp, size_t *bcp) 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 daddr_t blkcount, rblkcount, factor; 716 int nblocks; 717 daddr_t blocksreleased = 0, real_released = 0; 718 int error = 0, allerror = 0; 719 720 ASSERT_SEGLOCK(fs); 721 /* 722 * Calculate index in current block of last 723 * block to be kept. -1 indicates the entire 724 * block so we need not calculate the index. 725 */ 726 factor = 1; 727 for (i = SINGLE; i < level; i++) 728 factor *= LFS_NINDIR(fs); 729 last = lastbn; 730 if (lastbn > 0) 731 last /= factor; 732 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs)); 733 /* 734 * Get buffer of block pointers, zero those entries corresponding 735 * to blocks to be free'd, and update on disk copy first. Since 736 * double(triple) indirect before single(double) indirect, calls 737 * to bmap on these blocks will fail. However, we already have 738 * the on disk address, so we have to set the b_blkno field 739 * explicitly instead of letting bread do everything for us. 740 */ 741 vp = ITOV(ip); 742 bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0); 743 if (bp->b_oflags & (BO_DONE | BO_DELWRI)) { 744 /* Braces must be here in case trace evaluates to nothing. */ 745 trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn); 746 } else { 747 trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn); 748 curlwp->l_ru.ru_inblock++; /* pay for read */ 749 bp->b_flags |= B_READ; 750 if (bp->b_bcount > bp->b_bufsize) 751 panic("lfs_indirtrunc: bad buffer size"); 752 bp->b_blkno = LFS_FSBTODB(fs, dbn); 753 VOP_STRATEGY(vp, bp); 754 error = biowait(bp); 755 } 756 if (error) { 757 brelse(bp, 0); 758 *countp = *rcountp = 0; 759 return (error); 760 } 761 762 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 763 if (lastbn >= 0) { 764 copy = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK); 765 memcpy((void *)copy, (void *)bap, lfs_sb_getbsize(fs)); 766 memset((void *)&bap[last + 1], 0, 767 /* XXX ondisk32 */ 768 (u_int)(LFS_NINDIR(fs) - (last + 1)) * sizeof (int32_t)); 769 error = VOP_BWRITE(bp->b_vp, bp); 770 if (error) 771 allerror = error; 772 bap = copy; 773 } 774 775 /* 776 * Recursively free totally unused blocks. 777 */ 778 for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; 779 i--, nlbn += factor) { 780 nb = bap[i]; 781 if (nb == 0) 782 continue; 783 if (level > SINGLE) { 784 error = lfs_indirtrunc(ip, nlbn, nb, 785 (daddr_t)-1, level - 1, 786 &blkcount, &rblkcount, 787 lastsegp, bcp); 788 if (error) 789 allerror = error; 790 blocksreleased += blkcount; 791 real_released += rblkcount; 792 } 793 lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp); 794 if (bap[i] > 0) 795 real_released += nblocks; 796 blocksreleased += nblocks; 797 } 798 799 /* 800 * Recursively free last partial block. 801 */ 802 if (level > SINGLE && lastbn >= 0) { 803 last = lastbn % factor; 804 nb = bap[i]; 805 if (nb != 0) { 806 error = lfs_indirtrunc(ip, nlbn, nb, 807 last, level - 1, &blkcount, 808 &rblkcount, lastsegp, bcp); 809 if (error) 810 allerror = error; 811 real_released += rblkcount; 812 blocksreleased += blkcount; 813 } 814 } 815 816 if (copy != NULL) { 817 lfs_free(fs, copy, LFS_NB_IBLOCK); 818 } else { 819 mutex_enter(&bufcache_lock); 820 if (bp->b_oflags & BO_DELWRI) { 821 LFS_UNLOCK_BUF(bp); 822 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 823 wakeup(&fs->lfs_availsleep); 824 } 825 brelsel(bp, BC_INVAL); 826 mutex_exit(&bufcache_lock); 827 } 828 829 *countp = blocksreleased; 830 *rcountp = real_released; 831 return (allerror); 832 } 833 834 /* 835 * Destroy any in core blocks past the truncation length. 836 * Inlined from vtruncbuf, so that lfs_avail could be updated. 837 * We take the seglock to prevent cleaning from occurring while we are 838 * invalidating blocks. 839 */ 840 static int 841 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 842 { 843 struct buf *bp, *nbp; 844 int error; 845 struct lfs *fs; 846 voff_t off; 847 848 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 849 mutex_enter(vp->v_interlock); 850 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 851 if (error) 852 return error; 853 854 fs = VTOI(vp)->i_lfs; 855 856 ASSERT_SEGLOCK(fs); 857 858 mutex_enter(&bufcache_lock); 859 restart: 860 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 861 nbp = LIST_NEXT(bp, b_vnbufs); 862 if (bp->b_lblkno < lbn) 863 continue; 864 error = bbusy(bp, catch, slptimeo, NULL); 865 if (error == EPASSTHROUGH) 866 goto restart; 867 if (error != 0) { 868 mutex_exit(&bufcache_lock); 869 return (error); 870 } 871 mutex_enter(bp->b_objlock); 872 if (bp->b_oflags & BO_DELWRI) { 873 bp->b_oflags &= ~BO_DELWRI; 874 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 875 wakeup(&fs->lfs_availsleep); 876 } 877 mutex_exit(bp->b_objlock); 878 LFS_UNLOCK_BUF(bp); 879 brelsel(bp, BC_INVAL | BC_VFLUSH); 880 } 881 882 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 883 nbp = LIST_NEXT(bp, b_vnbufs); 884 if (bp->b_lblkno < lbn) 885 continue; 886 error = bbusy(bp, catch, slptimeo, NULL); 887 if (error == EPASSTHROUGH) 888 goto restart; 889 if (error != 0) { 890 mutex_exit(&bufcache_lock); 891 return (error); 892 } 893 mutex_enter(bp->b_objlock); 894 if (bp->b_oflags & BO_DELWRI) { 895 bp->b_oflags &= ~BO_DELWRI; 896 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 897 wakeup(&fs->lfs_availsleep); 898 } 899 mutex_exit(bp->b_objlock); 900 LFS_UNLOCK_BUF(bp); 901 brelsel(bp, BC_INVAL | BC_VFLUSH); 902 } 903 mutex_exit(&bufcache_lock); 904 905 return (0); 906 } 907 908
Indexes created Fri Sep 26 20:09:58 GMT 2025