1 /* $NetBSD: lfs.c,v 1.40 2013/06/18 18:18:58 christos Exp $ */ 2 /*- 3 * Copyright (c) 2003 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Konrad E. Schroder <perseant@hhhh.org>. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30 /* 31 * Copyright (c) 1989, 1991, 1993 32 * The Regents of the University of California. All rights reserved. 33 * (c) UNIX System Laboratories, Inc. 34 * All or some portions of this file are derived from material licensed 35 * to the University of California by American Telephone and Telegraph 36 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 37 * the permission of UNIX System Laboratories, Inc. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * @(#)ufs_bmap.c 8.8 (Berkeley) 8/11/95 64 */ 65 66 67 #include <sys/types.h> 68 #include <sys/param.h> 69 #include <sys/time.h> 70 #include <sys/buf.h> 71 #include <sys/mount.h> 72 73 #define vnode uvnode 74 #include <ufs/lfs/lfs.h> 75 #include <ufs/lfs/lfs_inode.h> 76 #undef vnode 77 78 #include <assert.h> 79 #include <err.h> 80 #include <errno.h> 81 #include <stdarg.h> 82 #include <stdio.h> 83 #include <stdlib.h> 84 #include <string.h> 85 #include <unistd.h> 86 #include <util.h> 87 88 #include "bufcache.h" 89 #include "vnode.h" 90 #include "lfs_user.h" 91 #include "segwrite.h" 92 #include "kernelops.h" 93 94 #define panic call_panic 95 96 extern u_int32_t cksum(void *, size_t); 97 extern u_int32_t lfs_sb_cksum(struct dlfs *); 98 extern void pwarn(const char *, ...); 99 100 extern struct uvnodelst vnodelist; 101 extern struct uvnodelst getvnodelist[VNODE_HASH_MAX]; 102 extern int nvnodes; 103 104 long dev_bsize = DEV_BSIZE; 105 106 static int 107 lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **); 108 109 int fsdirty = 0; 110 void (*panic_func)(int, const char *, va_list) = my_vpanic; 111 112 /* 113 * LFS buffer and uvnode operations 114 */ 115 116 int 117 lfs_vop_strategy(struct ubuf * bp) 118 { 119 int count; 120 121 if (bp->b_flags & B_READ) { 122 count = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 123 bp->b_blkno * dev_bsize); 124 if (count == bp->b_bcount) 125 bp->b_flags |= B_DONE; 126 } else { 127 count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 128 bp->b_blkno * dev_bsize); 129 if (count == 0) { 130 perror("pwrite"); 131 return -1; 132 } 133 bp->b_flags &= ~B_DELWRI; 134 reassignbuf(bp, bp->b_vp); 135 } 136 return 0; 137 } 138 139 int 140 lfs_vop_bwrite(struct ubuf * bp) 141 { 142 struct lfs *fs; 143 144 fs = bp->b_vp->v_fs; 145 if (!(bp->b_flags & B_DELWRI)) { 146 fs->lfs_avail -= lfs_btofsb(fs, bp->b_bcount); 147 } 148 bp->b_flags |= B_DELWRI | B_LOCKED; 149 reassignbuf(bp, bp->b_vp); 150 brelse(bp, 0); 151 return 0; 152 } 153 154 /* 155 * ulfs_bmaparray does the bmap conversion, and if requested returns the 156 * array of logical blocks which must be traversed to get to a block. 157 * Each entry contains the offset into that block that gets you to the 158 * next block and the disk address of the block (if it is assigned). 159 */ 160 int 161 ulfs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump) 162 { 163 struct inode *ip; 164 struct ubuf *bp; 165 struct indir a[ULFS_NIADDR + 1], *xap; 166 daddr_t daddr; 167 daddr_t metalbn; 168 int error, num; 169 170 ip = VTOI(vp); 171 172 if (bn >= 0 && bn < ULFS_NDADDR) { 173 if (nump != NULL) 174 *nump = 0; 175 *bnp = LFS_FSBTODB(fs, ip->i_ffs1_db[bn]); 176 if (*bnp == 0) 177 *bnp = -1; 178 return (0); 179 } 180 xap = ap == NULL ? a : ap; 181 if (!nump) 182 nump = # 183 if ((error = ulfs_getlbns(fs, vp, bn, xap, nump)) != 0) 184 return (error); 185 186 num = *nump; 187 188 /* Get disk address out of indirect block array */ 189 daddr = ip->i_ffs1_ib[xap->in_off]; 190 191 for (bp = NULL, ++xap; --num; ++xap) { 192 /* Exit the loop if there is no disk address assigned yet and 193 * the indirect block isn't in the cache, or if we were 194 * looking for an indirect block and we've found it. */ 195 196 metalbn = xap->in_lbn; 197 if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn) 198 break; 199 /* 200 * If we get here, we've either got the block in the cache 201 * or we have a disk address for it, go fetch it. 202 */ 203 if (bp) 204 brelse(bp, 0); 205 206 xap->in_exists = 1; 207 bp = getblk(vp, metalbn, fs->lfs_bsize); 208 209 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 210 bp->b_blkno = LFS_FSBTODB(fs, daddr); 211 bp->b_flags |= B_READ; 212 VOP_STRATEGY(bp); 213 } 214 daddr = ((ulfs_daddr_t *) bp->b_data)[xap->in_off]; 215 } 216 if (bp) 217 brelse(bp, 0); 218 219 daddr = LFS_FSBTODB(fs, (ulfs_daddr_t) daddr); 220 *bnp = daddr == 0 ? -1 : daddr; 221 return (0); 222 } 223 224 /* 225 * Create an array of logical block number/offset pairs which represent the 226 * path of indirect blocks required to access a data block. The first "pair" 227 * contains the logical block number of the appropriate single, double or 228 * triple indirect block and the offset into the inode indirect block array. 229 * Note, the logical block number of the inode single/double/triple indirect 230 * block appears twice in the array, once with the offset into the i_ffs1_ib and 231 * once with the offset into the page itself. 232 */ 233 int 234 ulfs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump) 235 { 236 daddr_t metalbn, realbn; 237 int64_t blockcnt; 238 int lbc; 239 int i, numlevels, off; 240 int lognindir, indir; 241 242 metalbn = 0; /* XXXGCC -Wuninitialized [sh3] */ 243 244 if (nump) 245 *nump = 0; 246 numlevels = 0; 247 realbn = bn; 248 if (bn < 0) 249 bn = -bn; 250 251 lognindir = -1; 252 for (indir = fs->lfs_nindir; indir; indir >>= 1) 253 ++lognindir; 254 255 /* Determine the number of levels of indirection. After this loop is 256 * done, blockcnt indicates the number of data blocks possible at the 257 * given level of indirection, and ULFS_NIADDR - i is the number of levels 258 * of indirection needed to locate the requested block. */ 259 260 bn -= ULFS_NDADDR; 261 for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) { 262 if (i == 0) 263 return (EFBIG); 264 265 lbc += lognindir; 266 blockcnt = (int64_t) 1 << lbc; 267 268 if (bn < blockcnt) 269 break; 270 } 271 272 /* Calculate the address of the first meta-block. */ 273 metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i); 274 275 /* At each iteration, off is the offset into the bap array which is an 276 * array of disk addresses at the current level of indirection. The 277 * logical block number and the offset in that block are stored into 278 * the argument array. */ 279 ap->in_lbn = metalbn; 280 ap->in_off = off = ULFS_NIADDR - i; 281 ap->in_exists = 0; 282 ap++; 283 for (++numlevels; i <= ULFS_NIADDR; i++) { 284 /* If searching for a meta-data block, quit when found. */ 285 if (metalbn == realbn) 286 break; 287 288 lbc -= lognindir; 289 blockcnt = (int64_t) 1 << lbc; 290 off = (bn >> lbc) & (fs->lfs_nindir - 1); 291 292 ++numlevels; 293 ap->in_lbn = metalbn; 294 ap->in_off = off; 295 ap->in_exists = 0; 296 ++ap; 297 298 metalbn -= -1 + (off << lbc); 299 } 300 if (nump) 301 *nump = numlevels; 302 return (0); 303 } 304 305 int 306 lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp) 307 { 308 return ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 309 } 310 311 /* Search a block for a specific dinode. */ 312 struct ulfs1_dinode * 313 lfs_ifind(struct lfs * fs, ino_t ino, struct ubuf * bp) 314 { 315 struct ulfs1_dinode *dip = (struct ulfs1_dinode *) bp->b_data; 316 struct ulfs1_dinode *ldip, *fin; 317 318 fin = dip + LFS_INOPB(fs); 319 320 /* 321 * Read the inode block backwards, since later versions of the 322 * inode will supercede earlier ones. Though it is unlikely, it is 323 * possible that the same inode will appear in the same inode block. 324 */ 325 for (ldip = fin - 1; ldip >= dip; --ldip) 326 if (ldip->di_inumber == ino) 327 return (ldip); 328 return NULL; 329 } 330 331 /* 332 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 333 * XXX it currently loses atime information. 334 */ 335 struct uvnode * 336 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, ulfs_daddr_t daddr) 337 { 338 struct uvnode *vp; 339 struct inode *ip; 340 struct ulfs1_dinode *dip; 341 struct ubuf *bp; 342 int i, hash; 343 344 vp = ecalloc(1, sizeof(*vp)); 345 vp->v_fd = fd; 346 vp->v_fs = fs; 347 vp->v_usecount = 0; 348 vp->v_strategy_op = lfs_vop_strategy; 349 vp->v_bwrite_op = lfs_vop_bwrite; 350 vp->v_bmap_op = lfs_vop_bmap; 351 LIST_INIT(&vp->v_cleanblkhd); 352 LIST_INIT(&vp->v_dirtyblkhd); 353 354 ip = ecalloc(1, sizeof(*ip)); 355 356 ip->i_din.ffs1_din = ecalloc(1, sizeof(*ip->i_din.ffs1_din)); 357 358 /* Initialize the inode -- from lfs_vcreate. */ 359 ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs)); 360 vp->v_data = ip; 361 /* ip->i_vnode = vp; */ 362 ip->i_number = ino; 363 ip->i_lockf = 0; 364 ip->i_lfs_effnblks = 0; 365 ip->i_flag = 0; 366 367 /* Load inode block and find inode */ 368 if (daddr > 0) { 369 bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), fs->lfs_ibsize, 370 NULL, 0, &bp); 371 bp->b_flags |= B_AGE; 372 dip = lfs_ifind(fs, ino, bp); 373 if (dip == NULL) { 374 brelse(bp, 0); 375 free(ip); 376 free(vp); 377 return NULL; 378 } 379 memcpy(ip->i_din.ffs1_din, dip, sizeof(*dip)); 380 brelse(bp, 0); 381 } 382 ip->i_number = ino; 383 /* ip->i_devvp = fs->lfs_devvp; */ 384 ip->i_lfs = fs; 385 386 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 387 ip->i_lfs_osize = ip->i_ffs1_size; 388 #if 0 389 if (fs->lfs_version > 1) { 390 ip->i_ffs1_atime = ts.tv_sec; 391 ip->i_ffs1_atimensec = ts.tv_nsec; 392 } 393 #endif 394 395 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 396 for (i = 0; i < ULFS_NDADDR; i++) 397 if (ip->i_ffs1_db[i] != 0) 398 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 399 400 ++nvnodes; 401 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 402 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 403 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 404 405 return vp; 406 } 407 408 static struct uvnode * 409 lfs_vget(void *vfs, ino_t ino) 410 { 411 struct lfs *fs = (struct lfs *)vfs; 412 ulfs_daddr_t daddr; 413 struct ubuf *bp; 414 IFILE *ifp; 415 416 LFS_IENTRY(ifp, fs, ino, bp); 417 daddr = ifp->if_daddr; 418 brelse(bp, 0); 419 if (daddr <= 0 || lfs_dtosn(fs, daddr) >= fs->lfs_nseg) 420 return NULL; 421 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 422 } 423 424 /* Check superblock magic number and checksum */ 425 static int 426 check_sb(struct lfs *fs) 427 { 428 u_int32_t checksum; 429 430 if (fs->lfs_magic != LFS_MAGIC) { 431 printf("Superblock magic number (0x%lx) does not match " 432 "expected 0x%lx\n", (unsigned long) fs->lfs_magic, 433 (unsigned long) LFS_MAGIC); 434 return 1; 435 } 436 /* checksum */ 437 checksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 438 if (fs->lfs_cksum != checksum) { 439 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 440 (unsigned long) fs->lfs_cksum, (unsigned long) checksum); 441 return 1; 442 } 443 return 0; 444 } 445 446 /* Initialize LFS library; load superblocks and choose which to use. */ 447 struct lfs * 448 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 449 { 450 struct uvnode *devvp; 451 struct ubuf *bp; 452 int tryalt; 453 struct lfs *fs, *altfs; 454 int error; 455 456 vfs_init(); 457 458 devvp = ecalloc(1, sizeof(*devvp)); 459 devvp->v_fs = NULL; 460 devvp->v_fd = devfd; 461 devvp->v_strategy_op = raw_vop_strategy; 462 devvp->v_bwrite_op = raw_vop_bwrite; 463 devvp->v_bmap_op = raw_vop_bmap; 464 LIST_INIT(&devvp->v_cleanblkhd); 465 LIST_INIT(&devvp->v_dirtyblkhd); 466 467 tryalt = 0; 468 if (dummy_read) { 469 if (sblkno == 0) 470 sblkno = LFS_LABELPAD / dev_bsize; 471 fs = ecalloc(1, sizeof(*fs)); 472 fs->lfs_devvp = devvp; 473 } else { 474 if (sblkno == 0) { 475 sblkno = LFS_LABELPAD / dev_bsize; 476 tryalt = 1; 477 } else if (debug) { 478 printf("No -b flag given, not attempting to verify checkpoint\n"); 479 } 480 481 dev_bsize = DEV_BSIZE; 482 483 error = bread(devvp, sblkno, LFS_SBPAD, NOCRED, 0, &bp); 484 fs = ecalloc(1, sizeof(*fs)); 485 fs->lfs_dlfs = *((struct dlfs *) bp->b_data); 486 fs->lfs_devvp = devvp; 487 bp->b_flags |= B_INVAL; 488 brelse(bp, 0); 489 490 dev_bsize = fs->lfs_fsize >> fs->lfs_fsbtodb; 491 492 if (tryalt) { 493 error = bread(devvp, LFS_FSBTODB(fs, fs->lfs_sboffs[1]), 494 LFS_SBPAD, NOCRED, 0, &bp); 495 altfs = ecalloc(1, sizeof(*altfs)); 496 altfs->lfs_dlfs = *((struct dlfs *) bp->b_data); 497 altfs->lfs_devvp = devvp; 498 bp->b_flags |= B_INVAL; 499 brelse(bp, 0); 500 501 if (check_sb(fs) || fs->lfs_idaddr <= 0) { 502 if (debug) 503 printf("Primary superblock is no good, using first alternate\n"); 504 free(fs); 505 fs = altfs; 506 } else { 507 /* If both superblocks check out, try verification */ 508 if (check_sb(altfs)) { 509 if (debug) 510 printf("First alternate superblock is no good, using primary\n"); 511 free(altfs); 512 } else { 513 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 514 free(altfs); 515 } else { 516 free(fs); 517 fs = altfs; 518 } 519 } 520 } 521 } 522 if (check_sb(fs)) { 523 free(fs); 524 return NULL; 525 } 526 } 527 528 /* Compatibility */ 529 if (fs->lfs_version < 2) { 530 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 531 fs->lfs_ibsize = fs->lfs_bsize; 532 fs->lfs_start = fs->lfs_sboffs[0]; 533 fs->lfs_tstamp = fs->lfs_otstamp; 534 fs->lfs_fsbtodb = 0; 535 } 536 537 if (!dummy_read) { 538 fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *)); 539 fs->lfs_suflags[0] = emalloc(fs->lfs_nseg * sizeof(u_int32_t)); 540 fs->lfs_suflags[1] = emalloc(fs->lfs_nseg * sizeof(u_int32_t)); 541 } 542 543 if (idaddr == 0) 544 idaddr = fs->lfs_idaddr; 545 else 546 fs->lfs_idaddr = idaddr; 547 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 548 fs->lfs_ivnode = lfs_raw_vget(fs, 549 (dummy_read ? LFS_IFILE_INUM : fs->lfs_ifile), devvp->v_fd, 550 idaddr); 551 if (fs->lfs_ivnode == NULL) 552 return NULL; 553 554 register_vget((void *)fs, lfs_vget); 555 556 return fs; 557 } 558 559 /* 560 * Check partial segment validity between fs->lfs_offset and the given goal. 561 * 562 * If goal == 0, just keep on going until the segments stop making sense, 563 * and return the address of the last valid partial segment. 564 * 565 * If goal != 0, return the address of the first partial segment that failed, 566 * or "goal" if we reached it without failure (the partial segment *at* goal 567 * need not be valid). 568 */ 569 ulfs_daddr_t 570 try_verify(struct lfs *osb, struct uvnode *devvp, ulfs_daddr_t goal, int debug) 571 { 572 ulfs_daddr_t daddr, odaddr; 573 SEGSUM *sp; 574 int i, bc, hitclean; 575 struct ubuf *bp; 576 ulfs_daddr_t nodirop_daddr; 577 u_int64_t serial; 578 579 bc = 0; 580 hitclean = 0; 581 odaddr = -1; 582 daddr = osb->lfs_offset; 583 nodirop_daddr = daddr; 584 serial = osb->lfs_serial; 585 while (daddr != goal) { 586 /* 587 * Don't mistakenly read a superblock, if there is one here. 588 */ 589 if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) { 590 if (daddr == osb->lfs_start) 591 daddr += lfs_btofsb(osb, LFS_LABELPAD); 592 for (i = 0; i < LFS_MAXNUMSB; i++) { 593 if (osb->lfs_sboffs[i] < daddr) 594 break; 595 if (osb->lfs_sboffs[i] == daddr) 596 daddr += lfs_btofsb(osb, LFS_SBPAD); 597 } 598 } 599 600 /* Read in summary block */ 601 bread(devvp, LFS_FSBTODB(osb, daddr), osb->lfs_sumsize, 602 NULL, 0, &bp); 603 sp = (SEGSUM *)bp->b_data; 604 605 /* 606 * Check for a valid segment summary belonging to our fs. 607 */ 608 if (sp->ss_magic != SS_MAGIC || 609 sp->ss_ident != osb->lfs_ident || 610 sp->ss_serial < serial || /* XXX strengthen this */ 611 sp->ss_sumsum != cksum(&sp->ss_datasum, osb->lfs_sumsize - 612 sizeof(sp->ss_sumsum))) { 613 brelse(bp, 0); 614 if (debug) { 615 if (sp->ss_magic != SS_MAGIC) 616 pwarn("pseg at 0x%x: " 617 "wrong magic number\n", 618 (int)daddr); 619 else if (sp->ss_ident != osb->lfs_ident) 620 pwarn("pseg at 0x%x: " 621 "expected ident %llx, got %llx\n", 622 (int)daddr, 623 (long long)sp->ss_ident, 624 (long long)osb->lfs_ident); 625 else if (sp->ss_serial >= serial) 626 pwarn("pseg at 0x%x: " 627 "serial %d < %d\n", (int)daddr, 628 (int)sp->ss_serial, (int)serial); 629 else 630 pwarn("pseg at 0x%x: " 631 "summary checksum wrong\n", 632 (int)daddr); 633 } 634 break; 635 } 636 if (debug && sp->ss_serial != serial) 637 pwarn("warning, serial=%d ss_serial=%d\n", 638 (int)serial, (int)sp->ss_serial); 639 ++serial; 640 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 641 if (bc == 0) { 642 brelse(bp, 0); 643 break; 644 } 645 if (debug) 646 pwarn("summary good: 0x%x/%d\n", (int)daddr, 647 (int)sp->ss_serial); 648 assert (bc > 0); 649 odaddr = daddr; 650 daddr += lfs_btofsb(osb, osb->lfs_sumsize + bc); 651 if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) || 652 lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr + 653 lfs_btofsb(osb, osb->lfs_sumsize + osb->lfs_bsize) - 1)) { 654 daddr = sp->ss_next; 655 } 656 657 /* 658 * Check for the beginning and ending of a sequence of 659 * dirops. Writes from the cleaner never involve new 660 * information, and are always checkpoints; so don't try 661 * to roll forward through them. Likewise, psegs written 662 * by a previous roll-forward attempt are not interesting. 663 */ 664 if (sp->ss_flags & (SS_CLEAN | SS_RFW)) 665 hitclean = 1; 666 if (hitclean == 0 && (sp->ss_flags & SS_CONT) == 0) 667 nodirop_daddr = daddr; 668 669 brelse(bp, 0); 670 } 671 672 if (goal == 0) 673 return nodirop_daddr; 674 else 675 return daddr; 676 } 677 678 /* Use try_verify to check whether the newer superblock is valid. */ 679 struct lfs * 680 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 681 { 682 ulfs_daddr_t daddr; 683 struct lfs *osb, *nsb; 684 685 /* 686 * Verify the checkpoint of the newer superblock, 687 * if the timestamp/serial number of the two superblocks is 688 * different. 689 */ 690 691 osb = NULL; 692 if (debug) 693 pwarn("sb0 %lld, sb1 %lld", 694 (long long) sb0->lfs_serial, 695 (long long) sb1->lfs_serial); 696 697 if ((sb0->lfs_version == 1 && 698 sb0->lfs_otstamp != sb1->lfs_otstamp) || 699 (sb0->lfs_version > 1 && 700 sb0->lfs_serial != sb1->lfs_serial)) { 701 if (sb0->lfs_version == 1) { 702 if (sb0->lfs_otstamp > sb1->lfs_otstamp) { 703 osb = sb1; 704 nsb = sb0; 705 } else { 706 osb = sb0; 707 nsb = sb1; 708 } 709 } else { 710 if (sb0->lfs_serial > sb1->lfs_serial) { 711 osb = sb1; 712 nsb = sb0; 713 } else { 714 osb = sb0; 715 nsb = sb1; 716 } 717 } 718 if (debug) { 719 printf("Attempting to verify newer checkpoint..."); 720 fflush(stdout); 721 } 722 daddr = try_verify(osb, devvp, nsb->lfs_offset, debug); 723 724 if (debug) 725 printf("done.\n"); 726 if (daddr == nsb->lfs_offset) { 727 pwarn("** Newer checkpoint verified, recovered %lld seconds of data\n", 728 (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 729 sbdirty(); 730 } else { 731 pwarn("** Newer checkpoint invalid, lost %lld seconds of data\n", (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 732 } 733 return (daddr == nsb->lfs_offset ? nsb : osb); 734 } 735 /* Nothing to check */ 736 return osb; 737 } 738 739 /* Verify a partial-segment summary; return the number of bytes on disk. */ 740 int 741 check_summary(struct lfs *fs, SEGSUM *sp, ulfs_daddr_t pseg_addr, int debug, 742 struct uvnode *devvp, void (func(ulfs_daddr_t, FINFO *))) 743 { 744 FINFO *fp; 745 int bc; /* Bytes in partial segment */ 746 int nblocks; 747 ulfs_daddr_t seg_addr, daddr; 748 ulfs_daddr_t *dp, *idp; 749 struct ubuf *bp; 750 int i, j, k, datac, len; 751 long sn; 752 u_int32_t *datap; 753 u_int32_t ccksum; 754 755 sn = lfs_dtosn(fs, pseg_addr); 756 seg_addr = lfs_sntod(fs, sn); 757 758 /* We've already checked the sumsum, just do the data bounds and sum */ 759 760 /* Count the blocks. */ 761 nblocks = howmany(sp->ss_ninos, LFS_INOPB(fs)); 762 bc = nblocks << (fs->lfs_version > 1 ? fs->lfs_ffshift : fs->lfs_bshift); 763 assert(bc >= 0); 764 765 fp = (FINFO *) (sp + 1); 766 for (i = 0; i < sp->ss_nfinfo; i++) { 767 nblocks += fp->fi_nblocks; 768 bc += fp->fi_lastlength + ((fp->fi_nblocks - 1) 769 << fs->lfs_bshift); 770 assert(bc >= 0); 771 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 772 if (((char *)fp) - (char *)sp > fs->lfs_sumsize) 773 return 0; 774 } 775 datap = emalloc(nblocks * sizeof(*datap)); 776 datac = 0; 777 778 dp = (ulfs_daddr_t *) sp; 779 dp += fs->lfs_sumsize / sizeof(ulfs_daddr_t); 780 dp--; 781 782 idp = dp; 783 daddr = pseg_addr + lfs_btofsb(fs, fs->lfs_sumsize); 784 fp = (FINFO *) (sp + 1); 785 for (i = 0, j = 0; 786 i < sp->ss_nfinfo || j < howmany(sp->ss_ninos, LFS_INOPB(fs)); i++) { 787 if (i >= sp->ss_nfinfo && *idp != daddr) { 788 pwarn("Not enough inode blocks in pseg at 0x%" PRIx32 789 ": found %d, wanted %d\n", 790 pseg_addr, j, howmany(sp->ss_ninos, LFS_INOPB(fs))); 791 if (debug) 792 pwarn("*idp=%x, daddr=%" PRIx32 "\n", *idp, 793 daddr); 794 break; 795 } 796 while (j < howmany(sp->ss_ninos, LFS_INOPB(fs)) && *idp == daddr) { 797 bread(devvp, LFS_FSBTODB(fs, daddr), fs->lfs_ibsize, 798 NOCRED, 0, &bp); 799 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 800 brelse(bp, 0); 801 802 ++j; 803 daddr += lfs_btofsb(fs, fs->lfs_ibsize); 804 --idp; 805 } 806 if (i < sp->ss_nfinfo) { 807 if (func) 808 func(daddr, fp); 809 for (k = 0; k < fp->fi_nblocks; k++) { 810 len = (k == fp->fi_nblocks - 1 ? 811 fp->fi_lastlength 812 : fs->lfs_bsize); 813 bread(devvp, LFS_FSBTODB(fs, daddr), len, 814 NOCRED, 0, &bp); 815 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 816 brelse(bp, 0); 817 daddr += lfs_btofsb(fs, len); 818 } 819 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 820 } 821 } 822 823 if (datac != nblocks) { 824 pwarn("Partial segment at 0x%llx expected %d blocks counted %d\n", 825 (long long) pseg_addr, nblocks, datac); 826 } 827 ccksum = cksum(datap, nblocks * sizeof(u_int32_t)); 828 /* Check the data checksum */ 829 if (ccksum != sp->ss_datasum) { 830 pwarn("Partial segment at 0x%" PRIx32 " data checksum" 831 " mismatch: given 0x%x, computed 0x%x\n", 832 pseg_addr, sp->ss_datasum, ccksum); 833 free(datap); 834 return 0; 835 } 836 free(datap); 837 assert(bc >= 0); 838 return bc; 839 } 840 841 /* print message and exit */ 842 void 843 my_vpanic(int fatal, const char *fmt, va_list ap) 844 { 845 (void) vprintf(fmt, ap); 846 exit(8); 847 } 848 849 void 850 call_panic(const char *fmt, ...) 851 { 852 va_list ap; 853 854 va_start(ap, fmt); 855 panic_func(1, fmt, ap); 856 va_end(ap); 857 } 858 859 /* Allocate a new inode. */ 860 struct uvnode * 861 lfs_valloc(struct lfs *fs, ino_t ino) 862 { 863 struct ubuf *bp, *cbp; 864 struct ifile *ifp; 865 ino_t new_ino; 866 int error; 867 int new_gen; 868 CLEANERINFO *cip; 869 870 /* Get the head of the freelist. */ 871 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 872 873 /* 874 * Remove the inode from the free list and write the new start 875 * of the free list into the superblock. 876 */ 877 LFS_IENTRY(ifp, fs, new_ino, bp); 878 if (ifp->if_daddr != LFS_UNUSED_DADDR) 879 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 880 LFS_PUT_HEADFREE(fs, cip, cbp, ifp->if_nextfree); 881 882 new_gen = ifp->if_version; /* version was updated by vfree */ 883 brelse(bp, 0); 884 885 /* Extend IFILE so that the next lfs_valloc will succeed. */ 886 if (fs->lfs_freehd == LFS_UNUSED_INUM) { 887 if ((error = extend_ifile(fs)) != 0) { 888 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 889 return NULL; 890 } 891 } 892 893 /* Set superblock modified bit and increment file count. */ 894 sbdirty(); 895 ++fs->lfs_nfiles; 896 897 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 898 } 899 900 #ifdef IN_FSCK_LFS 901 void reset_maxino(ino_t); 902 #endif 903 904 /* 905 * Add a new block to the Ifile, to accommodate future file creations. 906 */ 907 int 908 extend_ifile(struct lfs *fs) 909 { 910 struct uvnode *vp; 911 struct inode *ip; 912 IFILE *ifp; 913 IFILE_V1 *ifp_v1; 914 struct ubuf *bp, *cbp; 915 daddr_t i, blkno, max; 916 ino_t oldlast; 917 CLEANERINFO *cip; 918 919 vp = fs->lfs_ivnode; 920 ip = VTOI(vp); 921 blkno = lfs_lblkno(fs, ip->i_ffs1_size); 922 923 lfs_balloc(vp, ip->i_ffs1_size, fs->lfs_bsize, &bp); 924 ip->i_ffs1_size += fs->lfs_bsize; 925 ip->i_flag |= IN_MODIFIED; 926 927 i = (blkno - fs->lfs_segtabsz - fs->lfs_cleansz) * 928 fs->lfs_ifpb; 929 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 930 LFS_PUT_HEADFREE(fs, cip, cbp, i); 931 max = i + fs->lfs_ifpb; 932 fs->lfs_bfree -= lfs_btofsb(fs, fs->lfs_bsize); 933 934 if (fs->lfs_version == 1) { 935 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 936 ifp_v1->if_version = 1; 937 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 938 ifp_v1->if_nextfree = ++i; 939 } 940 ifp_v1--; 941 ifp_v1->if_nextfree = oldlast; 942 } else { 943 for (ifp = (IFILE *)bp->b_data; i < max; ++ifp) { 944 ifp->if_version = 1; 945 ifp->if_daddr = LFS_UNUSED_DADDR; 946 ifp->if_nextfree = ++i; 947 } 948 ifp--; 949 ifp->if_nextfree = oldlast; 950 } 951 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 952 953 LFS_BWRITE_LOG(bp); 954 955 #ifdef IN_FSCK_LFS 956 reset_maxino(((ip->i_ffs1_size >> fs->lfs_bshift) - fs->lfs_segtabsz - 957 fs->lfs_cleansz) * fs->lfs_ifpb); 958 #endif 959 return 0; 960 } 961 962 /* 963 * Allocate a block, and to inode and filesystem block accounting for it 964 * and for any indirect blocks the may need to be created in order for 965 * this block to be created. 966 * 967 * Blocks which have never been accounted for (i.e., which "do not exist") 968 * have disk address 0, which is translated by ulfs_bmap to the special value 969 * UNASSIGNED == -1, as in the historical ULFS. 970 * 971 * Blocks which have been accounted for but which have not yet been written 972 * to disk are given the new special disk address UNWRITTEN == -2, so that 973 * they can be differentiated from completely new blocks. 974 */ 975 int 976 lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 977 { 978 int offset; 979 daddr_t daddr, idaddr; 980 struct ubuf *ibp, *bp; 981 struct inode *ip; 982 struct lfs *fs; 983 struct indir indirs[ULFS_NIADDR+2], *idp; 984 daddr_t lbn, lastblock; 985 int bcount; 986 int error, frags, i, nsize, osize, num; 987 988 ip = VTOI(vp); 989 fs = ip->i_lfs; 990 offset = lfs_blkoff(fs, startoffset); 991 lbn = lfs_lblkno(fs, startoffset); 992 993 /* 994 * Three cases: it's a block beyond the end of file, it's a block in 995 * the file that may or may not have been assigned a disk address or 996 * we're writing an entire block. 997 * 998 * Note, if the daddr is UNWRITTEN, the block already exists in 999 * the cache (it was read or written earlier). If so, make sure 1000 * we don't count it as a new block or zero out its contents. If 1001 * it did not, make sure we allocate any necessary indirect 1002 * blocks. 1003 * 1004 * If we are writing a block beyond the end of the file, we need to 1005 * check if the old last block was a fragment. If it was, we need 1006 * to rewrite it. 1007 */ 1008 1009 if (bpp) 1010 *bpp = NULL; 1011 1012 /* Check for block beyond end of file and fragment extension needed. */ 1013 lastblock = lfs_lblkno(fs, ip->i_ffs1_size); 1014 if (lastblock < ULFS_NDADDR && lastblock < lbn) { 1015 osize = lfs_blksize(fs, ip, lastblock); 1016 if (osize < fs->lfs_bsize && osize > 0) { 1017 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 1018 lastblock, 1019 (bpp ? &bp : NULL)))) 1020 return (error); 1021 ip->i_ffs1_size = (lastblock + 1) * fs->lfs_bsize; 1022 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1023 if (bpp) 1024 (void) VOP_BWRITE(bp); 1025 } 1026 } 1027 1028 /* 1029 * If the block we are writing is a direct block, it's the last 1030 * block in the file, and offset + iosize is less than a full 1031 * block, we can write one or more fragments. There are two cases: 1032 * the block is brand new and we should allocate it the correct 1033 * size or it already exists and contains some fragments and 1034 * may need to extend it. 1035 */ 1036 if (lbn < ULFS_NDADDR && lfs_lblkno(fs, ip->i_ffs1_size) <= lbn) { 1037 osize = lfs_blksize(fs, ip, lbn); 1038 nsize = lfs_fragroundup(fs, offset + iosize); 1039 if (lfs_lblktosize(fs, lbn) >= ip->i_ffs1_size) { 1040 /* Brand new block or fragment */ 1041 frags = lfs_numfrags(fs, nsize); 1042 if (bpp) { 1043 *bpp = bp = getblk(vp, lbn, nsize); 1044 bp->b_blkno = UNWRITTEN; 1045 } 1046 ip->i_lfs_effnblks += frags; 1047 fs->lfs_bfree -= frags; 1048 ip->i_ffs1_db[lbn] = UNWRITTEN; 1049 } else { 1050 if (nsize <= osize) { 1051 /* No need to extend */ 1052 if (bpp && (error = bread(vp, lbn, osize, 1053 NOCRED, 0, &bp))) 1054 return error; 1055 } else { 1056 /* Extend existing block */ 1057 if ((error = 1058 lfs_fragextend(vp, osize, nsize, lbn, 1059 (bpp ? &bp : NULL)))) 1060 return error; 1061 } 1062 if (bpp) 1063 *bpp = bp; 1064 } 1065 return 0; 1066 } 1067 1068 error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1069 if (error) 1070 return (error); 1071 1072 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 1073 1074 /* 1075 * Do byte accounting all at once, so we can gracefully fail *before* 1076 * we start assigning blocks. 1077 */ 1078 frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */ 1079 bcount = 0; 1080 if (daddr == UNASSIGNED) { 1081 bcount = frags; 1082 } 1083 for (i = 1; i < num; ++i) { 1084 if (!indirs[i].in_exists) { 1085 bcount += frags; 1086 } 1087 } 1088 fs->lfs_bfree -= bcount; 1089 ip->i_lfs_effnblks += bcount; 1090 1091 if (daddr == UNASSIGNED) { 1092 if (num > 0 && ip->i_ffs1_ib[indirs[0].in_off] == 0) { 1093 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1094 } 1095 1096 /* 1097 * Create new indirect blocks if necessary 1098 */ 1099 if (num > 1) { 1100 idaddr = ip->i_ffs1_ib[indirs[0].in_off]; 1101 for (i = 1; i < num; ++i) { 1102 ibp = getblk(vp, indirs[i].in_lbn, 1103 fs->lfs_bsize); 1104 if (!indirs[i].in_exists) { 1105 memset(ibp->b_data, 0, ibp->b_bufsize); 1106 ibp->b_blkno = UNWRITTEN; 1107 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1108 ibp->b_blkno = LFS_FSBTODB(fs, idaddr); 1109 ibp->b_flags |= B_READ; 1110 VOP_STRATEGY(ibp); 1111 } 1112 /* 1113 * This block exists, but the next one may not. 1114 * If that is the case mark it UNWRITTEN to 1115 * keep the accounting straight. 1116 */ 1117 /* XXX ondisk32 */ 1118 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 1119 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 1120 UNWRITTEN; 1121 /* XXX ondisk32 */ 1122 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 1123 if ((error = VOP_BWRITE(ibp))) 1124 return error; 1125 } 1126 } 1127 } 1128 1129 1130 /* 1131 * Get the existing block from the cache, if requested. 1132 */ 1133 if (bpp) 1134 *bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn)); 1135 1136 /* 1137 * The block we are writing may be a brand new block 1138 * in which case we need to do accounting. 1139 * 1140 * We can tell a truly new block because ulfs_bmaparray will say 1141 * it is UNASSIGNED. Once we allocate it we will assign it the 1142 * disk address UNWRITTEN. 1143 */ 1144 if (daddr == UNASSIGNED) { 1145 if (bpp) { 1146 /* Note the new address */ 1147 bp->b_blkno = UNWRITTEN; 1148 } 1149 1150 switch (num) { 1151 case 0: 1152 ip->i_ffs1_db[lbn] = UNWRITTEN; 1153 break; 1154 case 1: 1155 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1156 break; 1157 default: 1158 idp = &indirs[num - 1]; 1159 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 1160 0, &ibp)) 1161 panic("lfs_balloc: bread bno %lld", 1162 (long long)idp->in_lbn); 1163 /* XXX ondisk32 */ 1164 ((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 1165 VOP_BWRITE(ibp); 1166 } 1167 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1168 /* 1169 * Not a brand new block, also not in the cache; 1170 * read it in from disk. 1171 */ 1172 if (iosize == fs->lfs_bsize) 1173 /* Optimization: I/O is unnecessary. */ 1174 bp->b_blkno = daddr; 1175 else { 1176 /* 1177 * We need to read the block to preserve the 1178 * existing bytes. 1179 */ 1180 bp->b_blkno = daddr; 1181 bp->b_flags |= B_READ; 1182 VOP_STRATEGY(bp); 1183 return 0; 1184 } 1185 } 1186 1187 return (0); 1188 } 1189 1190 int 1191 lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1192 struct ubuf **bpp) 1193 { 1194 struct inode *ip; 1195 struct lfs *fs; 1196 int frags; 1197 int error; 1198 size_t obufsize; 1199 1200 ip = VTOI(vp); 1201 fs = ip->i_lfs; 1202 frags = (long)lfs_numfrags(fs, nsize - osize); 1203 error = 0; 1204 1205 /* 1206 * If we are not asked to actually return the block, all we need 1207 * to do is allocate space for it. UBC will handle dirtying the 1208 * appropriate things and making sure it all goes to disk. 1209 * Don't bother to read in that case. 1210 */ 1211 if (bpp && (error = bread(vp, lbn, osize, NOCRED, 0, bpp))) { 1212 brelse(*bpp, 0); 1213 goto out; 1214 } 1215 1216 fs->lfs_bfree -= frags; 1217 ip->i_lfs_effnblks += frags; 1218 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1219 1220 if (bpp) { 1221 obufsize = (*bpp)->b_bufsize; 1222 (*bpp)->b_data = erealloc((*bpp)->b_data, nsize); 1223 (void)memset((*bpp)->b_data + osize, 0, nsize - osize); 1224 } 1225 1226 out: 1227 return (error); 1228 } 1229