lfs_rfw.c revision 1.17
1 /* $NetBSD: lfs_rfw.c,v 1.17 2013/06/18 18:18:58 christos 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 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.17 2013/06/18 18:18:58 christos Exp $"); 34 35 #if defined(_KERNEL_OPT) 36 #include "opt_quota.h" 37 #endif 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/namei.h> 42 #include <sys/proc.h> 43 #include <sys/kernel.h> 44 #include <sys/vnode.h> 45 #include <sys/mount.h> 46 #include <sys/kthread.h> 47 #include <sys/buf.h> 48 #include <sys/device.h> 49 #include <sys/mbuf.h> 50 #include <sys/file.h> 51 #include <sys/disklabel.h> 52 #include <sys/ioctl.h> 53 #include <sys/errno.h> 54 #include <sys/malloc.h> 55 #include <sys/pool.h> 56 #include <sys/socket.h> 57 #include <sys/syslog.h> 58 #include <uvm/uvm_extern.h> 59 #include <sys/sysctl.h> 60 #include <sys/conf.h> 61 #include <sys/kauth.h> 62 63 #include <miscfs/specfs/specdev.h> 64 65 #include <ufs/lfs/ulfs_quotacommon.h> 66 #include <ufs/lfs/ulfs_inode.h> 67 #include <ufs/lfs/ulfsmount.h> 68 #include <ufs/lfs/ulfs_extern.h> 69 70 #include <uvm/uvm.h> 71 #include <uvm/uvm_stat.h> 72 #include <uvm/uvm_pager.h> 73 #include <uvm/uvm_pdaemon.h> 74 75 #include <ufs/lfs/lfs.h> 76 #include <ufs/lfs/lfs_extern.h> 77 78 #include <miscfs/genfs/genfs.h> 79 #include <miscfs/genfs/genfs_node.h> 80 81 /* 82 * Roll-forward code. 83 */ 84 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t, 85 kauth_cred_t, int, int *, struct lwp *); 86 87 extern int lfs_do_rfw; 88 89 /* 90 * Allocate a particular inode with a particular version number, freeing 91 * any previous versions of this inode that may have gone before. 92 * Used by the roll-forward code. 93 * 94 * XXX this function does not have appropriate locking to be used on a live fs; 95 * XXX but something similar could probably be used for an "undelete" call. 96 * 97 * Called with the Ifile inode locked. 98 */ 99 int 100 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l, 101 struct vnode **vpp) 102 { 103 IFILE *ifp; 104 struct buf *bp, *cbp; 105 struct vnode *vp; 106 struct inode *ip; 107 ino_t tino, oldnext; 108 int error; 109 CLEANERINFO *cip; 110 111 ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */ 112 113 /* 114 * First, just try a vget. If the version number is the one we want, 115 * we don't have to do anything else. If the version number is wrong, 116 * take appropriate action. 117 */ 118 error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp); 119 if (error == 0) { 120 DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp)); 121 122 *vpp = vp; 123 ip = VTOI(vp); 124 if (ip->i_gen == vers) 125 return 0; 126 else if (ip->i_gen < vers) { 127 lfs_truncate(vp, (off_t)0, 0, NOCRED); 128 ip->i_gen = ip->i_ffs1_gen = vers; 129 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 130 return 0; 131 } else { 132 DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n", 133 ino, vers, ip->i_ffs1_gen)); 134 vput(vp); 135 *vpp = NULLVP; 136 return EEXIST; 137 } 138 } 139 140 /* 141 * The inode is not in use. Find it on the free list. 142 */ 143 /* If the Ifile is too short to contain this inum, extend it */ 144 while (VTOI(fs->lfs_ivnode)->i_size <= (ino / 145 fs->lfs_ifpb + fs->lfs_cleansz + fs->lfs_segtabsz) 146 << fs->lfs_bshift) { 147 lfs_extend_ifile(fs, NOCRED); 148 } 149 150 LFS_IENTRY(ifp, fs, ino, bp); 151 oldnext = ifp->if_nextfree; 152 ifp->if_version = vers; 153 brelse(bp, 0); 154 155 LFS_GET_HEADFREE(fs, cip, cbp, &ino); 156 if (ino) { 157 LFS_PUT_HEADFREE(fs, cip, cbp, oldnext); 158 } else { 159 tino = ino; 160 while (1) { 161 LFS_IENTRY(ifp, fs, tino, bp); 162 if (ifp->if_nextfree == ino || 163 ifp->if_nextfree == LFS_UNUSED_INUM) 164 break; 165 tino = ifp->if_nextfree; 166 brelse(bp, 0); 167 } 168 if (ifp->if_nextfree == LFS_UNUSED_INUM) { 169 brelse(bp, 0); 170 return ENOENT; 171 } 172 ifp->if_nextfree = oldnext; 173 LFS_BWRITE_LOG(bp); 174 } 175 176 error = lfs_ialloc(fs, fs->lfs_ivnode, ino, vers, &vp); 177 if (error == 0) { 178 /* 179 * Make it VREG so we can put blocks on it. We will change 180 * this later if it turns out to be some other kind of file. 181 */ 182 ip = VTOI(vp); 183 ip->i_mode = ip->i_ffs1_mode = LFS_IFREG; 184 ip->i_nlink = ip->i_ffs1_nlink = 1; 185 ulfs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, &vp); 186 ip = VTOI(vp); 187 188 DLOG((DLOG_RF, "lfs_rf_valloc: ino %d vp %p\n", ino, vp)); 189 190 /* The dirop-nature of this vnode is past */ 191 lfs_unmark_vnode(vp); 192 (void)lfs_vunref(vp); 193 vp->v_uflag &= ~VU_DIROP; 194 mutex_enter(&lfs_lock); 195 --lfs_dirvcount; 196 --fs->lfs_dirvcount; 197 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 198 wakeup(&lfs_dirvcount); 199 wakeup(&fs->lfs_dirvcount); 200 mutex_exit(&lfs_lock); 201 } 202 *vpp = vp; 203 return error; 204 } 205 206 /* 207 * Load the appropriate indirect block, and change the appropriate pointer. 208 * Mark the block dirty. Do segment and avail accounting. 209 */ 210 static int 211 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 212 daddr_t ndaddr, size_t size, struct lwp *l) 213 { 214 int error; 215 struct vnode *vp; 216 struct inode *ip; 217 #ifdef DEBUG 218 daddr_t odaddr; 219 struct indir a[ULFS_NIADDR]; 220 int num; 221 int i; 222 #endif /* DEBUG */ 223 struct buf *bp; 224 SEGUSE *sup; 225 226 KASSERT(lbn >= 0); /* no indirect blocks */ 227 228 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 229 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 230 " returned %d\n", ino, error)); 231 return error; 232 } 233 234 if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size, 235 NOCRED, 0, &bp)) != 0) { 236 vput(vp); 237 return (error); 238 } 239 /* No need to write, the block is already on disk */ 240 if (bp->b_oflags & BO_DELWRI) { 241 LFS_UNLOCK_BUF(bp); 242 fs->lfs_avail += lfs_btofsb(fs, bp->b_bcount); 243 } 244 brelse(bp, BC_INVAL); 245 246 /* 247 * Extend the file, if it is not large enough already. 248 * XXX this is not exactly right, we don't know how much of the 249 * XXX last block is actually used. We hope that an inode will 250 * XXX appear later to give the correct size. 251 */ 252 ip = VTOI(vp); 253 if (ip->i_size <= (lbn << fs->lfs_bshift)) { 254 u_int64_t newsize; 255 256 if (lbn < ULFS_NDADDR) 257 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 258 (size - fs->lfs_fsize) + 1; 259 else 260 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1; 261 262 if (ip->i_size < newsize) { 263 ip->i_size = newsize; 264 /* 265 * tell vm our new size for the case the inode won't 266 * appear later. 267 */ 268 uvm_vnp_setsize(vp, newsize); 269 } 270 } 271 272 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 273 274 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp); 275 sup->su_nbytes += size; 276 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp); 277 278 /* differences here should be due to UNWRITTEN indirect blocks. */ 279 KASSERT((lfs_lblkno(fs, ip->i_size) > ULFS_NDADDR && 280 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 281 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 282 283 #ifdef DEBUG 284 /* Now look again to make sure it worked */ 285 ulfs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 286 for (i = num; i > 0; i--) { 287 if (!a[i].in_exists) 288 panic("update_meta: absent %d lv indirect block", i); 289 } 290 if (LFS_DBTOFSB(fs, odaddr) != ndaddr) 291 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 292 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 293 #endif /* DEBUG */ 294 vput(vp); 295 return 0; 296 } 297 298 static int 299 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred, 300 struct lwp *l) 301 { 302 struct vnode *devvp, *vp; 303 struct inode *ip; 304 struct ulfs1_dinode *dip; 305 struct buf *dbp, *ibp; 306 int error; 307 daddr_t daddr; 308 IFILE *ifp; 309 SEGUSE *sup; 310 311 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 312 313 /* 314 * Get the inode, update times and perms. 315 * DO NOT update disk blocks, we do that separately. 316 */ 317 error = bread(devvp, LFS_FSBTODB(fs, offset), fs->lfs_ibsize, 318 cred, 0, &dbp); 319 if (error) { 320 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 321 return error; 322 } 323 dip = ((struct ulfs1_dinode *)(dbp->b_data)) + LFS_INOPB(fs); 324 while (--dip >= (struct ulfs1_dinode *)dbp->b_data) { 325 if (dip->di_inumber > LFS_IFILE_INUM) { 326 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 327 l, &vp); 328 if (error) { 329 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 330 " returned %d\n", error)); 331 continue; 332 } 333 ip = VTOI(vp); 334 if (dip->di_size != ip->i_size) 335 lfs_truncate(vp, dip->di_size, 0, NOCRED); 336 /* Get mode, link count, size, and times */ 337 memcpy(ip->i_din.ffs1_din, dip, 338 offsetof(struct ulfs1_dinode, di_db[0])); 339 340 /* Then the rest, except di_blocks */ 341 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 342 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 343 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 344 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 345 346 ip->i_mode = ip->i_ffs1_mode; 347 ip->i_nlink = ip->i_ffs1_nlink; 348 ip->i_size = ip->i_ffs1_size; 349 350 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 351 352 /* Re-initialize to get type right */ 353 ulfs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 354 &vp); 355 vput(vp); 356 357 /* Record change in location */ 358 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 359 daddr = ifp->if_daddr; 360 ifp->if_daddr = LFS_DBTOFSB(fs, dbp->b_blkno); 361 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 362 /* And do segment accounting */ 363 if (lfs_dtosn(fs, daddr) != lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno))) { 364 if (daddr > 0) { 365 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, daddr), 366 ibp); 367 sup->su_nbytes -= sizeof (struct ulfs1_dinode); 368 LFS_WRITESEGENTRY(sup, fs, 369 lfs_dtosn(fs, daddr), 370 ibp); 371 } 372 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)), 373 ibp); 374 sup->su_nbytes += sizeof (struct ulfs1_dinode); 375 LFS_WRITESEGENTRY(sup, fs, 376 lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)), 377 ibp); 378 } 379 } 380 } 381 brelse(dbp, BC_AGE); 382 383 return 0; 384 } 385 386 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 387 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 388 389 static daddr_t 390 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 391 kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l) 392 { 393 struct vnode *devvp; 394 struct buf *bp, *dbp; 395 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 396 SEGSUM *ssp; 397 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 398 daddr_t oldoffset; 399 int32_t *iaddr; /* XXX ondisk32 */ 400 FINFO *fip; 401 SEGUSE *sup; 402 size_t size; 403 404 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 405 /* 406 * If the segment has a superblock and we're at the top 407 * of the segment, skip the superblock. 408 */ 409 if (lfs_sntod(fs, lfs_dtosn(fs, offset)) == offset) { 410 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 411 if (sup->su_flags & SEGUSE_SUPERBLOCK) 412 offset += lfs_btofsb(fs, LFS_SBPAD); 413 brelse(bp, 0); 414 } 415 416 /* Read in the segment summary */ 417 error = bread(devvp, LFS_FSBTODB(fs, offset), fs->lfs_sumsize, 418 cred, 0, &bp); 419 if (error) 420 return -1; 421 422 /* Check summary checksum */ 423 ssp = (SEGSUM *)bp->b_data; 424 if (flags & CHECK_CKSUM) { 425 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 426 fs->lfs_sumsize - 427 sizeof(ssp->ss_sumsum))) { 428 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 429 offset = -1; 430 goto err1; 431 } 432 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 433 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 434 offset = -1; 435 goto err1; 436 } 437 if (ssp->ss_create < fs->lfs_tstamp) { 438 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 439 offset = -1; 440 goto err1; 441 } 442 } 443 if (fs->lfs_version > 1) { 444 if (ssp->ss_serial != nextserial) { 445 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 446 "\n", offset)); 447 offset = -1; 448 goto err1; 449 } 450 if (ssp->ss_ident != fs->lfs_ident) { 451 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 452 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset)); 453 offset = -1; 454 goto err1; 455 } 456 } 457 if (pseg_flags) 458 *pseg_flags = ssp->ss_flags; 459 oldoffset = offset; 460 offset += lfs_btofsb(fs, fs->lfs_sumsize); 461 462 ninos = howmany(ssp->ss_ninos, LFS_INOPB(fs)); 463 /* XXX ondisk32 */ 464 iaddr = (int32_t *)((char*)bp->b_data + fs->lfs_sumsize - sizeof(int32_t)); 465 if (flags & CHECK_CKSUM) { 466 /* Count blocks */ 467 nblocks = 0; 468 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 469 for (i = 0; i < ssp->ss_nfinfo; ++i) { 470 nblocks += fip->fi_nblocks; 471 if (fip->fi_nblocks <= 0) 472 break; 473 /* XXX ondisk32 */ 474 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 475 (fip->fi_nblocks * sizeof(int32_t))); 476 } 477 nblocks += ninos; 478 /* Create the sum array */ 479 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 480 M_SEGMENT, M_WAITOK); 481 } 482 483 /* Handle individual blocks */ 484 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 485 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 486 /* Inode block? */ 487 if (ninos && *iaddr == offset) { 488 if (flags & CHECK_CKSUM) { 489 /* Read in the head and add to the buffer */ 490 error = bread(devvp, LFS_FSBTODB(fs, offset), fs->lfs_bsize, 491 cred, 0, &dbp); 492 if (error) { 493 offset = -1; 494 goto err2; 495 } 496 (*dp++) = ((u_long *)(dbp->b_data))[0]; 497 brelse(dbp, BC_AGE); 498 } 499 if (flags & CHECK_UPDATE) { 500 if ((error = update_inoblk(fs, offset, cred, l)) 501 != 0) { 502 offset = -1; 503 goto err2; 504 } 505 } 506 offset += lfs_btofsb(fs, fs->lfs_ibsize); 507 --iaddr; 508 --ninos; 509 --i; /* compensate */ 510 continue; 511 } 512 size = fs->lfs_bsize; 513 for (j = 0; j < fip->fi_nblocks; ++j) { 514 if (j == fip->fi_nblocks - 1) 515 size = fip->fi_lastlength; 516 if (flags & CHECK_CKSUM) { 517 error = bread(devvp, LFS_FSBTODB(fs, offset), size, 518 cred, 0, &dbp); 519 if (error) { 520 offset = -1; 521 goto err2; 522 } 523 (*dp++) = ((u_long *)(dbp->b_data))[0]; 524 brelse(dbp, BC_AGE); 525 } 526 /* Account for and update any direct blocks */ 527 if ((flags & CHECK_UPDATE) && 528 fip->fi_ino > LFS_IFILE_INUM && 529 fip->fi_blocks[j] >= 0) { 530 update_meta(fs, fip->fi_ino, fip->fi_version, 531 fip->fi_blocks[j], offset, size, l); 532 } 533 offset += lfs_btofsb(fs, size); 534 } 535 /* XXX ondisk32 */ 536 fip = (FINFO *)(((char *)fip) + FINFOSIZE 537 + fip->fi_nblocks * sizeof(int32_t)); 538 } 539 /* Checksum the array, compare */ 540 if ((flags & CHECK_CKSUM) && 541 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 542 { 543 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 544 " (wanted %x got %x)\n", 545 offset, ssp->ss_datasum, cksum(datap, nblocks * 546 sizeof(u_long)))); 547 offset = -1; 548 goto err2; 549 } 550 551 /* If we're at the end of the segment, move to the next */ 552 if (lfs_dtosn(fs, offset + lfs_btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 553 lfs_dtosn(fs, offset)) { 554 if (lfs_dtosn(fs, offset) == lfs_dtosn(fs, ssp->ss_next)) { 555 offset = -1; 556 goto err2; 557 } 558 offset = ssp->ss_next; 559 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 560 " -> segment %d\n", offset, lfs_dtosn(fs,offset))); 561 } 562 563 if (flags & CHECK_UPDATE) { 564 fs->lfs_avail -= (offset - oldoffset); 565 /* Don't clog the buffer queue */ 566 mutex_enter(&lfs_lock); 567 if (locked_queue_count > LFS_MAX_BUFS || 568 locked_queue_bytes > LFS_MAX_BYTES) { 569 lfs_flush(fs, SEGM_CKP, 0); 570 } 571 mutex_exit(&lfs_lock); 572 } 573 574 err2: 575 if (flags & CHECK_CKSUM) 576 free(datap, M_SEGMENT); 577 err1: 578 brelse(bp, BC_AGE); 579 580 /* XXX should we update the serial number even for bad psegs? */ 581 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 582 fs->lfs_serial = nextserial; 583 return offset; 584 } 585 586 void 587 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l) 588 { 589 int flags, dirty; 590 daddr_t offset, oldoffset, lastgoodpseg; 591 int sn, curseg, do_rollforward; 592 struct proc *p; 593 kauth_cred_t cred; 594 SEGUSE *sup; 595 struct buf *bp; 596 597 p = l ? l->l_proc : NULL; 598 cred = p ? p->p_cred : NOCRED; 599 600 /* 601 * Roll forward. 602 * 603 * We don't roll forward for v1 filesystems, because 604 * of the danger that the clock was turned back between the last 605 * checkpoint and crash. This would roll forward garbage. 606 * 607 * v2 filesystems don't have this problem because they use a 608 * monotonically increasing serial number instead of a timestamp. 609 */ 610 do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) && 611 lfs_do_rfw && fs->lfs_version > 1 && p != NULL); 612 if (do_rollforward) { 613 u_int64_t nextserial; 614 /* 615 * Phase I: Find the address of the last good partial 616 * segment that was written after the checkpoint. Mark 617 * the segments in question dirty, so they won't be 618 * reallocated. 619 */ 620 lastgoodpseg = oldoffset = offset = fs->lfs_offset; 621 flags = 0x0; 622 DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%" 623 PRIx64 "\n", offset)); 624 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 625 if (!(sup->su_flags & SEGUSE_DIRTY)) 626 --fs->lfs_nclean; 627 sup->su_flags |= SEGUSE_DIRTY; 628 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 629 nextserial = fs->lfs_serial + 1; 630 while ((offset = check_segsum(fs, offset, nextserial, 631 cred, CHECK_CKSUM, &flags, l)) > 0) { 632 nextserial++; 633 if (lfs_sntod(fs, oldoffset) != lfs_sntod(fs, offset)) { 634 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset), 635 bp); 636 if (!(sup->su_flags & SEGUSE_DIRTY)) 637 --fs->lfs_nclean; 638 sup->su_flags |= SEGUSE_DIRTY; 639 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset), 640 bp); 641 } 642 643 DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%" 644 PRIx64 "\n", offset)); 645 if (flags & SS_DIROP) { 646 DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%" 647 PRIx64 "\n", oldoffset)); 648 if (!(flags & SS_CONT)) { 649 DLOG((DLOG_RF, "lfs_mountfs: dirops end " 650 "at 0x%" PRIx64 "\n", oldoffset)); 651 } 652 } 653 if (!(flags & SS_CONT)) 654 lastgoodpseg = offset; 655 oldoffset = offset; 656 } 657 if (flags & SS_CONT) { 658 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 659 "dirops discarded\n")); 660 } 661 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 662 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 663 oldoffset = fs->lfs_offset; 664 if (fs->lfs_offset != lastgoodpseg) { 665 /* Don't overwrite what we're trying to preserve */ 666 offset = fs->lfs_offset; 667 fs->lfs_offset = lastgoodpseg; 668 fs->lfs_curseg = lfs_sntod(fs, lfs_dtosn(fs, fs->lfs_offset)); 669 for (sn = curseg = lfs_dtosn(fs, fs->lfs_curseg);;) { 670 sn = (sn + 1) % fs->lfs_nseg; 671 if (sn == curseg) 672 panic("lfs_mountfs: no clean segments"); 673 LFS_SEGENTRY(sup, fs, sn, bp); 674 dirty = (sup->su_flags & SEGUSE_DIRTY); 675 brelse(bp, 0); 676 if (!dirty) 677 break; 678 } 679 fs->lfs_nextseg = lfs_sntod(fs, sn); 680 681 /* 682 * Phase II: Roll forward from the first superblock. 683 */ 684 while (offset != lastgoodpseg) { 685 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 686 PRIx64 "\n", offset)); 687 offset = check_segsum(fs, offset, 688 fs->lfs_serial + 1, cred, CHECK_UPDATE, 689 NULL, l); 690 } 691 692 /* 693 * Finish: flush our changes to disk. 694 */ 695 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 696 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 697 "recovered %lld blocks\n", 698 (long long)(lastgoodpseg - oldoffset))); 699 } 700 DLOG((DLOG_RF, "LFS roll forward complete\n")); 701 } 702 } 703
Indexes created Mon Sep 29 21:09:56 GMT 2025