lfs.c revision 1.24
1/* $NetBSD: lfs.c,v 1.24 2006/07/18 23:37:13 perseant 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 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37/* 38 * Copyright (c) 1989, 1991, 1993 39 * The Regents of the University of California. All rights reserved. 40 * (c) UNIX System Laboratories, Inc. 41 * All or some portions of this file are derived from material licensed 42 * to the University of California by American Telephone and Telegraph 43 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 44 * the permission of UNIX System Laboratories, Inc. 45 * 46 * Redistribution and use in source and binary forms, with or without 47 * modification, are permitted provided that the following conditions 48 * are met: 49 * 1. Redistributions of source code must retain the above copyright 50 * notice, this list of conditions and the following disclaimer. 51 * 2. Redistributions in binary form must reproduce the above copyright 52 * notice, this list of conditions and the following disclaimer in the 53 * documentation and/or other materials provided with the distribution. 54 * 3. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)ufs_bmap.c 8.8 (Berkeley) 8/11/95 71 */ 72 73 74#include <sys/types.h> 75#include <sys/param.h> 76#include <sys/time.h> 77#include <sys/buf.h> 78#include <sys/mount.h> 79 80#include <ufs/ufs/inode.h> 81#include <ufs/ufs/ufsmount.h> 82#define vnode uvnode 83#include <ufs/lfs/lfs.h> 84#undef vnode 85 86#include <assert.h> 87#include <err.h> 88#include <errno.h> 89#include <stdarg.h> 90#include <stdio.h> 91#include <stdlib.h> 92#include <string.h> 93#include <unistd.h> 94 95#include "bufcache.h" 96#include "vnode.h" 97#include "lfs_user.h" 98#include "segwrite.h" 99 100#define panic call_panic 101 102extern u_int32_t cksum(void *, size_t); 103extern u_int32_t lfs_sb_cksum(struct dlfs *); 104extern void pwarn(const char *, ...); 105 106extern struct uvnodelst vnodelist; 107extern struct uvnodelst getvnodelist[VNODE_HASH_MAX]; 108extern int nvnodes; 109 110static int 111lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **); 112 113int fsdirty = 0; 114void (*panic_func)(int, const char *, va_list) = my_vpanic; 115 116/* 117 * LFS buffer and uvnode operations 118 */ 119 120int 121lfs_vop_strategy(struct ubuf * bp) 122{ 123 int count; 124 125 if (bp->b_flags & B_READ) { 126 count = pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 127 dbtob(bp->b_blkno)); 128 if (count == bp->b_bcount) 129 bp->b_flags |= B_DONE; 130 } else { 131 count = pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 132 dbtob(bp->b_blkno)); 133 if (count == 0) { 134 perror("pwrite"); 135 return -1; 136 } 137 bp->b_flags &= ~B_DELWRI; 138 reassignbuf(bp, bp->b_vp); 139 } 140 return 0; 141} 142 143int 144lfs_vop_bwrite(struct ubuf * bp) 145{ 146 struct lfs *fs; 147 148 fs = bp->b_vp->v_fs; 149 if (!(bp->b_flags & B_DELWRI)) { 150 fs->lfs_avail -= btofsb(fs, bp->b_bcount); 151 } 152 bp->b_flags |= B_DELWRI | B_LOCKED; 153 reassignbuf(bp, bp->b_vp); 154 brelse(bp); 155 return 0; 156} 157 158/* 159 * ufs_bmaparray does the bmap conversion, and if requested returns the 160 * array of logical blocks which must be traversed to get to a block. 161 * Each entry contains the offset into that block that gets you to the 162 * next block and the disk address of the block (if it is assigned). 163 */ 164int 165ufs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump) 166{ 167 struct inode *ip; 168 struct ubuf *bp; 169 struct indir a[NIADDR + 1], *xap; 170 daddr_t daddr; 171 daddr_t metalbn; 172 int error, num; 173 174 ip = VTOI(vp); 175 176 if (bn >= 0 && bn < NDADDR) { 177 if (nump != NULL) 178 *nump = 0; 179 *bnp = fsbtodb(fs, ip->i_ffs1_db[bn]); 180 if (*bnp == 0) 181 *bnp = -1; 182 return (0); 183 } 184 xap = ap == NULL ? a : ap; 185 if (!nump) 186 nump = # 187 if ((error = ufs_getlbns(fs, vp, bn, xap, nump)) != 0) 188 return (error); 189 190 num = *nump; 191 192 /* Get disk address out of indirect block array */ 193 daddr = ip->i_ffs1_ib[xap->in_off]; 194 195 for (bp = NULL, ++xap; --num; ++xap) { 196 /* Exit the loop if there is no disk address assigned yet and 197 * the indirect block isn't in the cache, or if we were 198 * looking for an indirect block and we've found it. */ 199 200 metalbn = xap->in_lbn; 201 if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn) 202 break; 203 /* 204 * If we get here, we've either got the block in the cache 205 * or we have a disk address for it, go fetch it. 206 */ 207 if (bp) 208 brelse(bp); 209 210 xap->in_exists = 1; 211 bp = getblk(vp, metalbn, fs->lfs_bsize); 212 213 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 214 bp->b_blkno = fsbtodb(fs, daddr); 215 bp->b_flags |= B_READ; 216 VOP_STRATEGY(bp); 217 } 218 daddr = ((ufs_daddr_t *) bp->b_data)[xap->in_off]; 219 } 220 if (bp) 221 brelse(bp); 222 223 daddr = fsbtodb(fs, (ufs_daddr_t) daddr); 224 *bnp = daddr == 0 ? -1 : daddr; 225 return (0); 226} 227 228/* 229 * Create an array of logical block number/offset pairs which represent the 230 * path of indirect blocks required to access a data block. The first "pair" 231 * contains the logical block number of the appropriate single, double or 232 * triple indirect block and the offset into the inode indirect block array. 233 * Note, the logical block number of the inode single/double/triple indirect 234 * block appears twice in the array, once with the offset into the i_ffs1_ib and 235 * once with the offset into the page itself. 236 */ 237int 238ufs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump) 239{ 240 daddr_t metalbn, realbn; 241 int64_t blockcnt; 242 int lbc; 243 int i, numlevels, off; 244 int lognindir, indir; 245 246 metalbn = 0; /* XXXGCC -Wuninitialized [sh3] */ 247 248 if (nump) 249 *nump = 0; 250 numlevels = 0; 251 realbn = bn; 252 if (bn < 0) 253 bn = -bn; 254 255 lognindir = -1; 256 for (indir = fs->lfs_nindir; indir; indir >>= 1) 257 ++lognindir; 258 259 /* Determine the number of levels of indirection. After this loop is 260 * done, blockcnt indicates the number of data blocks possible at the 261 * given level of indirection, and NIADDR - i is the number of levels 262 * of indirection needed to locate the requested block. */ 263 264 bn -= NDADDR; 265 for (lbc = 0, i = NIADDR;; i--, bn -= blockcnt) { 266 if (i == 0) 267 return (EFBIG); 268 269 lbc += lognindir; 270 blockcnt = (int64_t) 1 << lbc; 271 272 if (bn < blockcnt) 273 break; 274 } 275 276 /* Calculate the address of the first meta-block. */ 277 metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + NIADDR - i); 278 279 /* At each iteration, off is the offset into the bap array which is an 280 * array of disk addresses at the current level of indirection. The 281 * logical block number and the offset in that block are stored into 282 * the argument array. */ 283 ap->in_lbn = metalbn; 284 ap->in_off = off = NIADDR - i; 285 ap->in_exists = 0; 286 ap++; 287 for (++numlevels; i <= NIADDR; i++) { 288 /* If searching for a meta-data block, quit when found. */ 289 if (metalbn == realbn) 290 break; 291 292 lbc -= lognindir; 293 blockcnt = (int64_t) 1 << lbc; 294 off = (bn >> lbc) & (fs->lfs_nindir - 1); 295 296 ++numlevels; 297 ap->in_lbn = metalbn; 298 ap->in_off = off; 299 ap->in_exists = 0; 300 ++ap; 301 302 metalbn -= -1 + (off << lbc); 303 } 304 if (nump) 305 *nump = numlevels; 306 return (0); 307} 308 309int 310lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp) 311{ 312 return ufs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 313} 314 315/* Search a block for a specific dinode. */ 316struct ufs1_dinode * 317lfs_ifind(struct lfs * fs, ino_t ino, struct ubuf * bp) 318{ 319 struct ufs1_dinode *dip = (struct ufs1_dinode *) bp->b_data; 320 struct ufs1_dinode *ldip, *fin; 321 322 fin = dip + INOPB(fs); 323 324 /* 325 * Read the inode block backwards, since later versions of the 326 * inode will supercede earlier ones. Though it is unlikely, it is 327 * possible that the same inode will appear in the same inode block. 328 */ 329 for (ldip = fin - 1; ldip >= dip; --ldip) 330 if (ldip->di_inumber == ino) 331 return (ldip); 332 return NULL; 333} 334 335/* 336 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 337 * XXX it currently loses atime information. 338 */ 339struct uvnode * 340lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, ufs_daddr_t daddr) 341{ 342 struct uvnode *vp; 343 struct inode *ip; 344 struct ufs1_dinode *dip; 345 struct ubuf *bp; 346 int i, hash; 347 348 vp = (struct uvnode *) malloc(sizeof(*vp)); 349 if (vp == NULL) 350 err(1, NULL); 351 memset(vp, 0, sizeof(*vp)); 352 vp->v_fd = fd; 353 vp->v_fs = fs; 354 vp->v_usecount = 0; 355 vp->v_strategy_op = lfs_vop_strategy; 356 vp->v_bwrite_op = lfs_vop_bwrite; 357 vp->v_bmap_op = lfs_vop_bmap; 358 LIST_INIT(&vp->v_cleanblkhd); 359 LIST_INIT(&vp->v_dirtyblkhd); 360 361 ip = (struct inode *) malloc(sizeof(*ip)); 362 if (ip == NULL) 363 err(1, NULL); 364 memset(ip, 0, sizeof(*ip)); 365 366 ip->i_din.ffs1_din = (struct ufs1_dinode *) 367 malloc(sizeof(struct ufs1_dinode)); 368 if (ip->i_din.ffs1_din == NULL) 369 err(1, NULL); 370 memset(ip->i_din.ffs1_din, 0, sizeof (struct ufs1_dinode)); 371 372 /* Initialize the inode -- from lfs_vcreate. */ 373 ip->inode_ext.lfs = malloc(sizeof(struct lfs_inode_ext)); 374 if (ip->inode_ext.lfs == NULL) 375 err(1, NULL); 376 memset(ip->inode_ext.lfs, 0, sizeof(struct lfs_inode_ext)); 377 vp->v_data = ip; 378 /* ip->i_vnode = vp; */ 379 ip->i_number = ino; 380 ip->i_lockf = 0; 381 ip->i_diroff = 0; 382 ip->i_lfs_effnblks = 0; 383 ip->i_flag = 0; 384 385 /* Load inode block and find inode */ 386 if (daddr > 0) { 387 bread(fs->lfs_devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, NULL, &bp); 388 bp->b_flags |= B_AGE; 389 dip = lfs_ifind(fs, ino, bp); 390 if (dip == NULL) { 391 brelse(bp); 392 free(ip); 393 free(vp); 394 return NULL; 395 } 396 memcpy(ip->i_din.ffs1_din, dip, sizeof(*dip)); 397 brelse(bp); 398 } 399 ip->i_number = ino; 400 /* ip->i_devvp = fs->lfs_devvp; */ 401 ip->i_lfs = fs; 402 403 ip->i_ffs_effnlink = ip->i_ffs1_nlink; 404 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 405 ip->i_lfs_osize = ip->i_ffs1_size; 406#if 0 407 if (fs->lfs_version > 1) { 408 ip->i_ffs1_atime = ts.tv_sec; 409 ip->i_ffs1_atimensec = ts.tv_nsec; 410 } 411#endif 412 413 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 414 for (i = 0; i < NDADDR; i++) 415 if (ip->i_ffs1_db[i] != 0) 416 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 417 418 ++nvnodes; 419 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 420 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 421 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 422 423 return vp; 424} 425 426static struct uvnode * 427lfs_vget(void *vfs, ino_t ino) 428{ 429 struct lfs *fs = (struct lfs *)vfs; 430 ufs_daddr_t daddr; 431 struct ubuf *bp; 432 IFILE *ifp; 433 434 LFS_IENTRY(ifp, fs, ino, bp); 435 daddr = ifp->if_daddr; 436 brelse(bp); 437 if (daddr <= 0 || dtosn(fs, daddr) >= fs->lfs_nseg) 438 return NULL; 439 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 440} 441 442/* Check superblock magic number and checksum */ 443static int 444check_sb(struct lfs *fs) 445{ 446 u_int32_t checksum; 447 448 if (fs->lfs_magic != LFS_MAGIC) { 449 printf("Superblock magic number (0x%lx) does not match " 450 "expected 0x%lx\n", (unsigned long) fs->lfs_magic, 451 (unsigned long) LFS_MAGIC); 452 return 1; 453 } 454 /* checksum */ 455 checksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 456 if (fs->lfs_cksum != checksum) { 457 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 458 (unsigned long) fs->lfs_cksum, (unsigned long) checksum); 459 return 1; 460 } 461 return 0; 462} 463 464/* Initialize LFS library; load superblocks and choose which to use. */ 465struct lfs * 466lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 467{ 468 struct uvnode *devvp; 469 struct ubuf *bp; 470 int tryalt; 471 struct lfs *fs, *altfs; 472 int error; 473 474 vfs_init(); 475 476 devvp = (struct uvnode *) malloc(sizeof(*devvp)); 477 if (devvp == NULL) 478 err(1, NULL); 479 memset(devvp, 0, sizeof(*devvp)); 480 devvp->v_fs = NULL; 481 devvp->v_fd = devfd; 482 devvp->v_strategy_op = raw_vop_strategy; 483 devvp->v_bwrite_op = raw_vop_bwrite; 484 devvp->v_bmap_op = raw_vop_bmap; 485 LIST_INIT(&devvp->v_cleanblkhd); 486 LIST_INIT(&devvp->v_dirtyblkhd); 487 488 tryalt = 0; 489 if (dummy_read) { 490 if (sblkno == 0) 491 sblkno = btodb(LFS_LABELPAD); 492 fs = (struct lfs *) malloc(sizeof(*fs)); 493 if (fs == NULL) 494 err(1, NULL); 495 memset(fs, 0, sizeof(*fs)); 496 fs->lfs_devvp = devvp; 497 } else { 498 if (sblkno == 0) { 499 sblkno = btodb(LFS_LABELPAD); 500 tryalt = 1; 501 } else if (debug) { 502 printf("No -b flag given, not attempting to verify checkpoint\n"); 503 } 504 error = bread(devvp, sblkno, LFS_SBPAD, NOCRED, &bp); 505 fs = (struct lfs *) malloc(sizeof(*fs)); 506 if (fs == NULL) 507 err(1, NULL); 508 memset(fs, 0, sizeof(*fs)); 509 fs->lfs_dlfs = *((struct dlfs *) bp->b_data); 510 fs->lfs_devvp = devvp; 511 bp->b_flags |= B_INVAL; 512 brelse(bp); 513 514 if (tryalt) { 515 error = bread(devvp, fsbtodb(fs, fs->lfs_sboffs[1]), 516 LFS_SBPAD, NOCRED, &bp); 517 altfs = (struct lfs *) malloc(sizeof(*altfs)); 518 if (altfs == NULL) 519 err(1, NULL); 520 memset(altfs, 0, sizeof(*altfs)); 521 altfs->lfs_dlfs = *((struct dlfs *) bp->b_data); 522 altfs->lfs_devvp = devvp; 523 bp->b_flags |= B_INVAL; 524 brelse(bp); 525 526 if (check_sb(fs) || fs->lfs_idaddr <= 0) { 527 if (debug) 528 printf("Primary superblock is no good, using first alternate\n"); 529 free(fs); 530 fs = altfs; 531 } else { 532 /* If both superblocks check out, try verification */ 533 if (check_sb(altfs)) { 534 if (debug) 535 printf("First alternate superblock is no good, using primary\n"); 536 free(altfs); 537 } else { 538 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 539 free(altfs); 540 } else { 541 free(fs); 542 fs = altfs; 543 } 544 } 545 } 546 } 547 if (check_sb(fs)) { 548 free(fs); 549 return NULL; 550 } 551 } 552 553 /* Compatibility */ 554 if (fs->lfs_version < 2) { 555 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 556 fs->lfs_ibsize = fs->lfs_bsize; 557 fs->lfs_start = fs->lfs_sboffs[0]; 558 fs->lfs_tstamp = fs->lfs_otstamp; 559 fs->lfs_fsbtodb = 0; 560 } 561 562 if (!dummy_read) { 563 fs->lfs_suflags = (u_int32_t **) malloc(2 * sizeof(u_int32_t *)); 564 if (fs->lfs_suflags == NULL) 565 err(1, NULL); 566 fs->lfs_suflags[0] = (u_int32_t *) malloc(fs->lfs_nseg * sizeof(u_int32_t)); 567 if (fs->lfs_suflags[0] == NULL) 568 err(1, NULL); 569 fs->lfs_suflags[1] = (u_int32_t *) malloc(fs->lfs_nseg * sizeof(u_int32_t)); 570 if (fs->lfs_suflags[1] == NULL) 571 err(1, NULL); 572 } 573 574 if (idaddr == 0) 575 idaddr = fs->lfs_idaddr; 576 else 577 fs->lfs_idaddr = idaddr; 578 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 579 fs->lfs_ivnode = lfs_raw_vget(fs, 580 (dummy_read ? LFS_IFILE_INUM : fs->lfs_ifile), devvp->v_fd, 581 idaddr); 582 if (fs->lfs_ivnode == NULL) 583 return NULL; 584 585 register_vget((void *)fs, lfs_vget); 586 587 return fs; 588} 589 590/* 591 * Check partial segment validity between fs->lfs_offset and the given goal. 592 * 593 * If goal == 0, just keep on going until the segments stop making sense, 594 * and return the address of the last valid partial segment. 595 * 596 * If goal != 0, return the address of the first partial segment that failed, 597 * or "goal" if we reached it without failure (the partial segment *at* goal 598 * need not be valid). 599 */ 600ufs_daddr_t 601try_verify(struct lfs *osb, struct uvnode *devvp, ufs_daddr_t goal, int debug) 602{ 603 ufs_daddr_t daddr, odaddr; 604 SEGSUM *sp; 605 int i, bc, dc; 606 struct ubuf *bp; 607 ufs_daddr_t nodirop_daddr; 608 u_int64_t serial; 609 610 bc = dc = 0; 611 odaddr = -1; 612 daddr = osb->lfs_offset; 613 nodirop_daddr = daddr; 614 serial = osb->lfs_serial; 615 while (daddr != goal) { 616 /* 617 * Don't mistakenly read a superblock, if there is one here. 618 */ 619 if (sntod(osb, dtosn(osb, daddr)) == daddr) { 620 for (i = 0; i < LFS_MAXNUMSB; i++) { 621 if (osb->lfs_sboffs[i] < daddr) 622 break; 623 if (osb->lfs_sboffs[i] == daddr) 624 daddr += btofsb(osb, LFS_SBPAD); 625 } 626 } 627 628 /* Read in summary block */ 629 bread(devvp, fsbtodb(osb, daddr), osb->lfs_sumsize, NULL, &bp); 630 sp = (SEGSUM *)bp->b_data; 631 632 /* 633 * Check for a valid segment summary belonging to our fs. 634 */ 635 if (sp->ss_magic != SS_MAGIC || 636 sp->ss_ident != osb->lfs_ident || 637 sp->ss_serial < serial || /* XXX strengthen this */ 638 sp->ss_sumsum != cksum(&sp->ss_datasum, osb->lfs_sumsize - 639 sizeof(sp->ss_sumsum))) { 640 brelse(bp); 641 if (debug) { 642 if (sp->ss_magic != SS_MAGIC) 643 pwarn("pseg at 0x%x: " 644 "wrong magic number\n", 645 (int)daddr); 646 else if (sp->ss_ident != osb->lfs_ident) 647 pwarn("pseg at 0x%x: " 648 "expected ident %llx, got %llx\n", 649 (int)daddr, 650 (long long)sp->ss_ident, 651 (long long)osb->lfs_ident); 652 else if (sp->ss_serial >= serial) 653 pwarn("pseg at 0x%x: " 654 "serial %d < %d\n", (int)daddr, 655 (int)sp->ss_serial, (int)serial); 656 else 657 pwarn("pseg at 0x%x: " 658 "summary checksum wrong\n", 659 (int)daddr); 660 } 661 break; 662 } 663 if (debug && sp->ss_serial != serial) 664 pwarn("warning, serial=%d ss_serial=%d", 665 (int)serial, (int)sp->ss_serial); 666 ++serial; 667 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 668 if (bc == 0) { 669 brelse(bp); 670 break; 671 } 672 if (debug) 673 pwarn("summary good: 0x%x/%d\n", (int)daddr, 674 (int)sp->ss_serial); 675 assert (bc > 0); 676 odaddr = daddr; 677 daddr += btofsb(osb, osb->lfs_sumsize + bc); 678 if (dtosn(osb, odaddr) != dtosn(osb, daddr) || 679 dtosn(osb, daddr) != dtosn(osb, daddr + 680 btofsb(osb, osb->lfs_sumsize + osb->lfs_bsize))) { 681 daddr = sp->ss_next; 682 } 683 684 /* 685 * Check for the beginning and ending of a sequence of 686 * dirops. We have to do the check this way, rather than 687 * simply checking for the lack of SS_CONT, because the 688 * cleaner sometimes injects SS_DIROP|SS_CONT partial-segments 689 * without actually completing the dirop. 690 */ 691 if (sp->ss_flags & SS_CONT) 692 dc = 1; 693 if ((sp->ss_flags & (SS_DIROP | SS_CONT)) == SS_DIROP) 694 dc = 0; 695 if (dc == 0) 696 nodirop_daddr = daddr; 697 698 brelse(bp); 699 } 700 701 if (goal == 0) 702 return nodirop_daddr; 703 else 704 return daddr; 705} 706 707/* Use try_verify to check whether the newer superblock is valid. */ 708struct lfs * 709lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 710{ 711 ufs_daddr_t daddr; 712 struct lfs *osb, *nsb; 713 714 /* 715 * Verify the checkpoint of the newer superblock, 716 * if the timestamp/serial number of the two superblocks is 717 * different. 718 */ 719 720 osb = NULL; 721 if (debug) 722 pwarn("sb0 %lld, sb1 %lld", 723 (long long) sb0->lfs_serial, 724 (long long) sb1->lfs_serial); 725 726 if ((sb0->lfs_version == 1 && 727 sb0->lfs_otstamp != sb1->lfs_otstamp) || 728 (sb0->lfs_version > 1 && 729 sb0->lfs_serial != sb1->lfs_serial)) { 730 if (sb0->lfs_version == 1) { 731 if (sb0->lfs_otstamp > sb1->lfs_otstamp) { 732 osb = sb1; 733 nsb = sb0; 734 } else { 735 osb = sb0; 736 nsb = sb1; 737 } 738 } else { 739 if (sb0->lfs_serial > sb1->lfs_serial) { 740 osb = sb1; 741 nsb = sb0; 742 } else { 743 osb = sb0; 744 nsb = sb1; 745 } 746 } 747 if (debug) { 748 printf("Attempting to verify newer checkpoint..."); 749 fflush(stdout); 750 } 751 daddr = try_verify(osb, devvp, nsb->lfs_offset, debug); 752 753 if (debug) 754 printf("done.\n"); 755 if (daddr == nsb->lfs_offset) { 756 pwarn("** Newer checkpoint verified, recovered %lld seconds of data\n", 757 (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 758 sbdirty(); 759 } else { 760 pwarn("** Newer checkpoint invalid, lost %lld seconds of data\n", (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 761 } 762 return (daddr == nsb->lfs_offset ? nsb : osb); 763 } 764 /* Nothing to check */ 765 return osb; 766} 767 768/* Verify a partial-segment summary; return the number of bytes on disk. */ 769int 770check_summary(struct lfs *fs, SEGSUM *sp, ufs_daddr_t pseg_addr, int debug, 771 struct uvnode *devvp, void (func(ufs_daddr_t, FINFO *))) 772{ 773 FINFO *fp; 774 int bc; /* Bytes in partial segment */ 775 int nblocks; 776 ufs_daddr_t seg_addr, daddr; 777 ufs_daddr_t *dp, *idp; 778 struct ubuf *bp; 779 int i, j, k, datac, len; 780 long sn; 781 u_int32_t *datap; 782 u_int32_t ccksum; 783 784 sn = dtosn(fs, pseg_addr); 785 seg_addr = sntod(fs, sn); 786 787 /* We've already checked the sumsum, just do the data bounds and sum */ 788 789 /* Count the blocks. */ 790 nblocks = howmany(sp->ss_ninos, INOPB(fs)); 791 bc = nblocks << (fs->lfs_version > 1 ? fs->lfs_ffshift : fs->lfs_bshift); 792 assert(bc >= 0); 793 794 fp = (FINFO *) (sp + 1); 795 for (i = 0; i < sp->ss_nfinfo; i++) { 796 nblocks += fp->fi_nblocks; 797 bc += fp->fi_lastlength + ((fp->fi_nblocks - 1) 798 << fs->lfs_bshift); 799 assert(bc >= 0); 800 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 801 if (((char *)fp) - (char *)sp > fs->lfs_sumsize) 802 return 0; 803 } 804 datap = (u_int32_t *) malloc(nblocks * sizeof(*datap)); 805 if (datap == NULL) 806 err(1, NULL); 807 datac = 0; 808 809 dp = (ufs_daddr_t *) sp; 810 dp += fs->lfs_sumsize / sizeof(ufs_daddr_t); 811 dp--; 812 813 idp = dp; 814 daddr = pseg_addr + btofsb(fs, fs->lfs_sumsize); 815 fp = (FINFO *) (sp + 1); 816 for (i = 0, j = 0; 817 i < sp->ss_nfinfo || j < howmany(sp->ss_ninos, INOPB(fs)); i++) { 818 if (i >= sp->ss_nfinfo && *idp != daddr) { 819 pwarn("Not enough inode blocks in pseg at 0x%" PRIx32 820 ": found %d, wanted %d\n", 821 pseg_addr, j, howmany(sp->ss_ninos, INOPB(fs))); 822 if (debug) 823 pwarn("*idp=%x, daddr=%" PRIx32 "\n", *idp, 824 daddr); 825 break; 826 } 827 while (j < howmany(sp->ss_ninos, INOPB(fs)) && *idp == daddr) { 828 bread(devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, NOCRED, &bp); 829 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 830 brelse(bp); 831 832 ++j; 833 daddr += btofsb(fs, fs->lfs_ibsize); 834 --idp; 835 } 836 if (i < sp->ss_nfinfo) { 837 if (func) 838 func(daddr, fp); 839 for (k = 0; k < fp->fi_nblocks; k++) { 840 len = (k == fp->fi_nblocks - 1 ? 841 fp->fi_lastlength 842 : fs->lfs_bsize); 843 bread(devvp, fsbtodb(fs, daddr), len, NOCRED, &bp); 844 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 845 brelse(bp); 846 daddr += btofsb(fs, len); 847 } 848 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 849 } 850 } 851 852 if (datac != nblocks) { 853 pwarn("Partial segment at 0x%llx expected %d blocks counted %d\n", 854 (long long) pseg_addr, nblocks, datac); 855 } 856 ccksum = cksum(datap, nblocks * sizeof(u_int32_t)); 857 /* Check the data checksum */ 858 if (ccksum != sp->ss_datasum) { 859 pwarn("Partial segment at 0x%" PRIx32 " data checksum" 860 " mismatch: given 0x%x, computed 0x%x\n", 861 pseg_addr, sp->ss_datasum, ccksum); 862 free(datap); 863 return 0; 864 } 865 free(datap); 866 assert(bc >= 0); 867 return bc; 868} 869 870/* print message and exit */ 871void 872my_vpanic(int fatal, const char *fmt, va_list ap) 873{ 874 (void) vprintf(fmt, ap); 875 exit(8); 876} 877 878void 879call_panic(const char *fmt, ...) 880{ 881 va_list ap; 882 883 va_start(ap, fmt); 884 panic_func(1, fmt, ap); 885 va_end(ap); 886} 887 888/* Allocate a new inode. */ 889struct uvnode * 890lfs_valloc(struct lfs *fs, ino_t ino) 891{ 892 struct ubuf *bp, *cbp; 893 struct ifile *ifp; 894 ino_t new_ino; 895 int error; 896 int new_gen; 897 CLEANERINFO *cip; 898 899 /* Get the head of the freelist. */ 900 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 901 902 /* 903 * Remove the inode from the free list and write the new start 904 * of the free list into the superblock. 905 */ 906 LFS_IENTRY(ifp, fs, new_ino, bp); 907 if (ifp->if_daddr != LFS_UNUSED_DADDR) 908 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 909 LFS_PUT_HEADFREE(fs, cip, cbp, ifp->if_nextfree); 910 911 new_gen = ifp->if_version; /* version was updated by vfree */ 912 brelse(bp); 913 914 /* Extend IFILE so that the next lfs_valloc will succeed. */ 915 if (fs->lfs_freehd == LFS_UNUSED_INUM) { 916 if ((error = extend_ifile(fs)) != 0) { 917 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 918 return NULL; 919 } 920 } 921 922 /* Set superblock modified bit and increment file count. */ 923 sbdirty(); 924 ++fs->lfs_nfiles; 925 926 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 927} 928 929#ifdef IN_FSCK_LFS 930void reset_maxino(ino_t); 931#endif 932 933/* 934 * Add a new block to the Ifile, to accommodate future file creations. 935 */ 936int 937extend_ifile(struct lfs *fs) 938{ 939 struct uvnode *vp; 940 struct inode *ip; 941 IFILE *ifp; 942 IFILE_V1 *ifp_v1; 943 struct ubuf *bp, *cbp; 944 daddr_t i, blkno, max; 945 ino_t oldlast; 946 CLEANERINFO *cip; 947 948 vp = fs->lfs_ivnode; 949 ip = VTOI(vp); 950 blkno = lblkno(fs, ip->i_ffs1_size); 951 952 lfs_balloc(vp, ip->i_ffs1_size, fs->lfs_bsize, &bp); 953 ip->i_ffs1_size += fs->lfs_bsize; 954 ip->i_flag |= IN_MODIFIED; 955 956 i = (blkno - fs->lfs_segtabsz - fs->lfs_cleansz) * 957 fs->lfs_ifpb; 958 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 959 LFS_PUT_HEADFREE(fs, cip, cbp, i); 960 max = i + fs->lfs_ifpb; 961 fs->lfs_bfree -= btofsb(fs, fs->lfs_bsize); 962 963 if (fs->lfs_version == 1) { 964 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 965 ifp_v1->if_version = 1; 966 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 967 ifp_v1->if_nextfree = ++i; 968 } 969 ifp_v1--; 970 ifp_v1->if_nextfree = oldlast; 971 } else { 972 for (ifp = (IFILE *)bp->b_data; i < max; ++ifp) { 973 ifp->if_version = 1; 974 ifp->if_daddr = LFS_UNUSED_DADDR; 975 ifp->if_nextfree = ++i; 976 } 977 ifp--; 978 ifp->if_nextfree = oldlast; 979 } 980 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 981 982 LFS_BWRITE_LOG(bp); 983 984#ifdef IN_FSCK_LFS 985 reset_maxino(((ip->i_ffs1_size >> fs->lfs_bshift) - fs->lfs_segtabsz - 986 fs->lfs_cleansz) * fs->lfs_ifpb); 987#endif 988 return 0; 989} 990 991/* 992 * Allocate a block, and to inode and filesystem block accounting for it 993 * and for any indirect blocks the may need to be created in order for 994 * this block to be created. 995 * 996 * Blocks which have never been accounted for (i.e., which "do not exist") 997 * have disk address 0, which is translated by ufs_bmap to the special value 998 * UNASSIGNED == -1, as in the historical UFS. 999 * 1000 * Blocks which have been accounted for but which have not yet been written 1001 * to disk are given the new special disk address UNWRITTEN == -2, so that 1002 * they can be differentiated from completely new blocks. 1003 */ 1004int 1005lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 1006{ 1007 int offset; 1008 daddr_t daddr, idaddr; 1009 struct ubuf *ibp, *bp; 1010 struct inode *ip; 1011 struct lfs *fs; 1012 struct indir indirs[NIADDR+2], *idp; 1013 daddr_t lbn, lastblock; 1014 int bb, bcount; 1015 int error, frags, i, nsize, osize, num; 1016 1017 ip = VTOI(vp); 1018 fs = ip->i_lfs; 1019 offset = blkoff(fs, startoffset); 1020 lbn = lblkno(fs, startoffset); 1021 1022 /* 1023 * Three cases: it's a block beyond the end of file, it's a block in 1024 * the file that may or may not have been assigned a disk address or 1025 * we're writing an entire block. 1026 * 1027 * Note, if the daddr is UNWRITTEN, the block already exists in 1028 * the cache (it was read or written earlier). If so, make sure 1029 * we don't count it as a new block or zero out its contents. If 1030 * it did not, make sure we allocate any necessary indirect 1031 * blocks. 1032 * 1033 * If we are writing a block beyond the end of the file, we need to 1034 * check if the old last block was a fragment. If it was, we need 1035 * to rewrite it. 1036 */ 1037 1038 if (bpp) 1039 *bpp = NULL; 1040 1041 /* Check for block beyond end of file and fragment extension needed. */ 1042 lastblock = lblkno(fs, ip->i_ffs1_size); 1043 if (lastblock < NDADDR && lastblock < lbn) { 1044 osize = blksize(fs, ip, lastblock); 1045 if (osize < fs->lfs_bsize && osize > 0) { 1046 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 1047 lastblock, 1048 (bpp ? &bp : NULL)))) 1049 return (error); 1050 ip->i_ffs1_size = ip->i_ffs1_size = 1051 (lastblock + 1) * fs->lfs_bsize; 1052 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1053 if (bpp) 1054 (void) VOP_BWRITE(bp); 1055 } 1056 } 1057 1058 /* 1059 * If the block we are writing is a direct block, it's the last 1060 * block in the file, and offset + iosize is less than a full 1061 * block, we can write one or more fragments. There are two cases: 1062 * the block is brand new and we should allocate it the correct 1063 * size or it already exists and contains some fragments and 1064 * may need to extend it. 1065 */ 1066 if (lbn < NDADDR && lblkno(fs, ip->i_ffs1_size) <= lbn) { 1067 osize = blksize(fs, ip, lbn); 1068 nsize = fragroundup(fs, offset + iosize); 1069 if (lblktosize(fs, lbn) >= ip->i_ffs1_size) { 1070 /* Brand new block or fragment */ 1071 frags = numfrags(fs, nsize); 1072 bb = fragstofsb(fs, frags); 1073 if (bpp) { 1074 *bpp = bp = getblk(vp, lbn, nsize); 1075 bp->b_blkno = UNWRITTEN; 1076 } 1077 ip->i_lfs_effnblks += bb; 1078 fs->lfs_bfree -= bb; 1079 ip->i_ffs1_db[lbn] = UNWRITTEN; 1080 } else { 1081 if (nsize <= osize) { 1082 /* No need to extend */ 1083 if (bpp && (error = bread(vp, lbn, osize, NOCRED, &bp))) 1084 return error; 1085 } else { 1086 /* Extend existing block */ 1087 if ((error = 1088 lfs_fragextend(vp, osize, nsize, lbn, 1089 (bpp ? &bp : NULL)))) 1090 return error; 1091 } 1092 if (bpp) 1093 *bpp = bp; 1094 } 1095 return 0; 1096 } 1097 1098 error = ufs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1099 if (error) 1100 return (error); 1101 1102 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 1103 1104 /* 1105 * Do byte accounting all at once, so we can gracefully fail *before* 1106 * we start assigning blocks. 1107 */ 1108 bb = fsbtodb(fs, 1); /* bb = VFSTOUFS(vp->v_mount)->um_seqinc; */ 1109 bcount = 0; 1110 if (daddr == UNASSIGNED) { 1111 bcount = bb; 1112 } 1113 for (i = 1; i < num; ++i) { 1114 if (!indirs[i].in_exists) { 1115 bcount += bb; 1116 } 1117 } 1118 fs->lfs_bfree -= bcount; 1119 ip->i_lfs_effnblks += bcount; 1120 1121 if (daddr == UNASSIGNED) { 1122 if (num > 0 && ip->i_ffs1_ib[indirs[0].in_off] == 0) { 1123 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1124 } 1125 1126 /* 1127 * Create new indirect blocks if necessary 1128 */ 1129 if (num > 1) { 1130 idaddr = ip->i_ffs1_ib[indirs[0].in_off]; 1131 for (i = 1; i < num; ++i) { 1132 ibp = getblk(vp, indirs[i].in_lbn, 1133 fs->lfs_bsize); 1134 if (!indirs[i].in_exists) { 1135 memset(ibp->b_data, 0, ibp->b_bufsize); 1136 ibp->b_blkno = UNWRITTEN; 1137 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1138 ibp->b_blkno = fsbtodb(fs, idaddr); 1139 ibp->b_flags |= B_READ; 1140 VOP_STRATEGY(ibp); 1141 } 1142 /* 1143 * This block exists, but the next one may not. 1144 * If that is the case mark it UNWRITTEN to 1145 * keep the accounting straight. 1146 */ 1147 /* XXX ondisk32 */ 1148 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 1149 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 1150 UNWRITTEN; 1151 /* XXX ondisk32 */ 1152 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 1153 if ((error = VOP_BWRITE(ibp))) 1154 return error; 1155 } 1156 } 1157 } 1158 1159 1160 /* 1161 * Get the existing block from the cache, if requested. 1162 */ 1163 frags = fsbtofrags(fs, bb); 1164 if (bpp) 1165 *bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn)); 1166 1167 /* 1168 * The block we are writing may be a brand new block 1169 * in which case we need to do accounting. 1170 * 1171 * We can tell a truly new block because ufs_bmaparray will say 1172 * it is UNASSIGNED. Once we allocate it we will assign it the 1173 * disk address UNWRITTEN. 1174 */ 1175 if (daddr == UNASSIGNED) { 1176 if (bpp) { 1177 /* Note the new address */ 1178 bp->b_blkno = UNWRITTEN; 1179 } 1180 1181 switch (num) { 1182 case 0: 1183 ip->i_ffs1_db[lbn] = UNWRITTEN; 1184 break; 1185 case 1: 1186 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1187 break; 1188 default: 1189 idp = &indirs[num - 1]; 1190 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 1191 &ibp)) 1192 panic("lfs_balloc: bread bno %lld", 1193 (long long)idp->in_lbn); 1194 /* XXX ondisk32 */ 1195 ((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 1196 VOP_BWRITE(ibp); 1197 } 1198 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1199 /* 1200 * Not a brand new block, also not in the cache; 1201 * read it in from disk. 1202 */ 1203 if (iosize == fs->lfs_bsize) 1204 /* Optimization: I/O is unnecessary. */ 1205 bp->b_blkno = daddr; 1206 else { 1207 /* 1208 * We need to read the block to preserve the 1209 * existing bytes. 1210 */ 1211 bp->b_blkno = daddr; 1212 bp->b_flags |= B_READ; 1213 VOP_STRATEGY(bp); 1214 return 0; 1215 } 1216 } 1217 1218 return (0); 1219} 1220 1221int 1222lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1223 struct ubuf **bpp) 1224{ 1225 struct inode *ip; 1226 struct lfs *fs; 1227 long bb; 1228 int error; 1229 size_t obufsize; 1230 1231 ip = VTOI(vp); 1232 fs = ip->i_lfs; 1233 bb = (long)fragstofsb(fs, numfrags(fs, nsize - osize)); 1234 error = 0; 1235 1236 /* 1237 * If we are not asked to actually return the block, all we need 1238 * to do is allocate space for it. UBC will handle dirtying the 1239 * appropriate things and making sure it all goes to disk. 1240 * Don't bother to read in that case. 1241 */ 1242 if (bpp && (error = bread(vp, lbn, osize, NOCRED, bpp))) { 1243 brelse(*bpp); 1244 goto out; 1245 } 1246 1247 fs->lfs_bfree -= bb; 1248 ip->i_lfs_effnblks += bb; 1249 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1250 1251 if (bpp) { 1252 obufsize = (*bpp)->b_bufsize; 1253 (*bpp)->b_data = realloc((*bpp)->b_data, nsize); 1254 bzero((char *)((*bpp)->b_data) + osize, (u_int)(nsize - osize)); 1255 } 1256 1257 out: 1258 return (error); 1259} 1260