lfs_rfw.c revision 1.23
1 /* $NetBSD: lfs_rfw.c,v 1.23 2015/07/24 06:59:32 dholland Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.23 2015/07/24 06:59:32 dholland 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_kernel.h> 77 #include <ufs/lfs/lfs_extern.h> 78 79 #include <miscfs/genfs/genfs.h> 80 #include <miscfs/genfs/genfs_node.h> 81 82 /* 83 * Roll-forward code. 84 */ 85 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t, 86 kauth_cred_t, int, int *, struct lwp *); 87 88 extern int lfs_do_rfw; 89 90 /* 91 * Allocate a particular inode with a particular version number, freeing 92 * any previous versions of this inode that may have gone before. 93 * Used by the roll-forward code. 94 * 95 * XXX this function does not have appropriate locking to be used on a live fs; 96 * XXX but something similar could probably be used for an "undelete" call. 97 * 98 * Called with the Ifile inode locked. 99 */ 100 int 101 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l, 102 struct vnode **vpp) 103 { 104 struct vattr va; 105 struct vnode *vp; 106 struct inode *ip; 107 int error; 108 109 ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */ 110 111 /* 112 * First, just try a vget. If the version number is the one we want, 113 * we don't have to do anything else. If the version number is wrong, 114 * take appropriate action. 115 */ 116 error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp); 117 if (error == 0) { 118 DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp)); 119 120 *vpp = vp; 121 ip = VTOI(vp); 122 if (ip->i_gen == vers) 123 return 0; 124 else if (ip->i_gen < vers) { 125 lfs_truncate(vp, (off_t)0, 0, NOCRED); 126 ip->i_gen = ip->i_ffs1_gen = vers; 127 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 128 return 0; 129 } else { 130 DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n", 131 ino, vers, ip->i_ffs1_gen)); 132 vput(vp); 133 *vpp = NULLVP; 134 return EEXIST; 135 } 136 } 137 138 /* Not found, create as regular file. */ 139 vattr_null(&va); 140 va.va_type = VREG; 141 va.va_mode = 0; 142 va.va_fileid = ino; 143 va.va_gen = vers; 144 error = vcache_new(fs->lfs_ivnode->v_mount, NULL, &va, NOCRED, &vp); 145 if (error) 146 return error; 147 error = vn_lock(vp, LK_EXCLUSIVE); 148 if (error) { 149 vrele(vp); 150 *vpp = NULLVP; 151 return error; 152 } 153 ip = VTOI(vp); 154 ip->i_nlink = ip->i_ffs1_nlink = 1; 155 *vpp = vp; 156 return 0; 157 } 158 159 /* 160 * Load the appropriate indirect block, and change the appropriate pointer. 161 * Mark the block dirty. Do segment and avail accounting. 162 */ 163 static int 164 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 165 daddr_t ndaddr, size_t size, struct lwp *l) 166 { 167 int error; 168 struct vnode *vp; 169 struct inode *ip; 170 #ifdef DEBUG 171 daddr_t odaddr; 172 struct indir a[ULFS_NIADDR]; 173 int num; 174 int i; 175 #endif /* DEBUG */ 176 struct buf *bp; 177 SEGUSE *sup; 178 179 KASSERT(lbn >= 0); /* no indirect blocks */ 180 181 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 182 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 183 " returned %d\n", ino, error)); 184 return error; 185 } 186 187 if ((error = lfs_balloc(vp, (lbn << lfs_sb_getbshift(fs)), size, 188 NOCRED, 0, &bp)) != 0) { 189 vput(vp); 190 return (error); 191 } 192 /* No need to write, the block is already on disk */ 193 if (bp->b_oflags & BO_DELWRI) { 194 LFS_UNLOCK_BUF(bp); 195 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount)); 196 /* XXX should this wake up fs->lfs_availsleep? */ 197 } 198 brelse(bp, BC_INVAL); 199 200 /* 201 * Extend the file, if it is not large enough already. 202 * XXX this is not exactly right, we don't know how much of the 203 * XXX last block is actually used. We hope that an inode will 204 * XXX appear later to give the correct size. 205 */ 206 ip = VTOI(vp); 207 if (ip->i_size <= (lbn << lfs_sb_getbshift(fs))) { 208 u_int64_t newsize; 209 210 if (lbn < ULFS_NDADDR) 211 newsize = ip->i_ffs1_size = (lbn << lfs_sb_getbshift(fs)) + 212 (size - lfs_sb_getfsize(fs)) + 1; 213 else 214 newsize = ip->i_ffs1_size = (lbn << lfs_sb_getbshift(fs)) + 1; 215 216 if (ip->i_size < newsize) { 217 ip->i_size = newsize; 218 /* 219 * tell vm our new size for the case the inode won't 220 * appear later. 221 */ 222 uvm_vnp_setsize(vp, newsize); 223 } 224 } 225 226 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 227 228 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp); 229 sup->su_nbytes += size; 230 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, ndaddr), bp); 231 232 /* differences here should be due to UNWRITTEN indirect blocks. */ 233 KASSERT((lfs_lblkno(fs, ip->i_size) > ULFS_NDADDR && 234 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 235 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 236 237 #ifdef DEBUG 238 /* Now look again to make sure it worked */ 239 ulfs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 240 for (i = num; i > 0; i--) { 241 if (!a[i].in_exists) 242 panic("update_meta: absent %d lv indirect block", i); 243 } 244 if (LFS_DBTOFSB(fs, odaddr) != ndaddr) 245 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 246 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 247 #endif /* DEBUG */ 248 vput(vp); 249 return 0; 250 } 251 252 static int 253 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred, 254 struct lwp *l) 255 { 256 struct vnode *devvp, *vp; 257 struct inode *ip; 258 struct ulfs1_dinode *dip; 259 struct buf *dbp, *ibp; 260 int error; 261 daddr_t daddr; 262 IFILE *ifp; 263 SEGUSE *sup; 264 265 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 266 267 /* 268 * Get the inode, update times and perms. 269 * DO NOT update disk blocks, we do that separately. 270 */ 271 error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getibsize(fs), 272 0, &dbp); 273 if (error) { 274 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 275 return error; 276 } 277 dip = ((struct ulfs1_dinode *)(dbp->b_data)) + LFS_INOPB(fs); 278 while (--dip >= (struct ulfs1_dinode *)dbp->b_data) { 279 if (dip->di_inumber > LFS_IFILE_INUM) { 280 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 281 l, &vp); 282 if (error) { 283 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 284 " returned %d\n", error)); 285 continue; 286 } 287 ip = VTOI(vp); 288 if (dip->di_size != ip->i_size) 289 lfs_truncate(vp, dip->di_size, 0, NOCRED); 290 /* Get mode, link count, size, and times */ 291 memcpy(ip->i_din.ffs1_din, dip, 292 offsetof(struct ulfs1_dinode, di_db[0])); 293 294 /* Then the rest, except di_blocks */ 295 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 296 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 297 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 298 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 299 300 ip->i_mode = ip->i_ffs1_mode; 301 ip->i_nlink = ip->i_ffs1_nlink; 302 ip->i_size = ip->i_ffs1_size; 303 304 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 305 306 /* Re-initialize to get type right */ 307 ulfs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 308 &vp); 309 vput(vp); 310 311 /* Record change in location */ 312 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 313 daddr = ifp->if_daddr; 314 ifp->if_daddr = LFS_DBTOFSB(fs, dbp->b_blkno); 315 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 316 /* And do segment accounting */ 317 if (lfs_dtosn(fs, daddr) != lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno))) { 318 if (daddr > 0) { 319 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, daddr), 320 ibp); 321 sup->su_nbytes -= sizeof (struct ulfs1_dinode); 322 LFS_WRITESEGENTRY(sup, fs, 323 lfs_dtosn(fs, daddr), 324 ibp); 325 } 326 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)), 327 ibp); 328 sup->su_nbytes += sizeof (struct ulfs1_dinode); 329 LFS_WRITESEGENTRY(sup, fs, 330 lfs_dtosn(fs, LFS_DBTOFSB(fs, dbp->b_blkno)), 331 ibp); 332 } 333 } 334 } 335 brelse(dbp, BC_AGE); 336 337 return 0; 338 } 339 340 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 341 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 342 343 static daddr_t 344 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 345 kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l) 346 { 347 struct vnode *devvp; 348 struct buf *bp, *dbp; 349 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 350 SEGSUM *ssp; 351 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 352 daddr_t oldoffset; 353 int32_t *iaddr; /* XXX ondisk32 */ 354 FINFO *fip; 355 SEGUSE *sup; 356 size_t size; 357 358 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 359 /* 360 * If the segment has a superblock and we're at the top 361 * of the segment, skip the superblock. 362 */ 363 if (lfs_sntod(fs, lfs_dtosn(fs, offset)) == offset) { 364 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 365 if (sup->su_flags & SEGUSE_SUPERBLOCK) 366 offset += lfs_btofsb(fs, LFS_SBPAD); 367 brelse(bp, 0); 368 } 369 370 /* Read in the segment summary */ 371 error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getsumsize(fs), 372 0, &bp); 373 if (error) 374 return -1; 375 376 /* Check summary checksum */ 377 ssp = (SEGSUM *)bp->b_data; 378 if (flags & CHECK_CKSUM) { 379 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 380 lfs_sb_getsumsize(fs) - 381 sizeof(ssp->ss_sumsum))) { 382 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 383 offset = -1; 384 goto err1; 385 } 386 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 387 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 388 offset = -1; 389 goto err1; 390 } 391 if (ssp->ss_create < lfs_sb_gettstamp(fs)) { 392 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 393 offset = -1; 394 goto err1; 395 } 396 } 397 if (fs->lfs_version > 1) { 398 if (ssp->ss_serial != nextserial) { 399 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 400 "\n", offset)); 401 offset = -1; 402 goto err1; 403 } 404 if (ssp->ss_ident != lfs_sb_getident(fs)) { 405 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 406 PRIx64 "\n", ssp->ss_ident, lfs_sb_getident(fs), offset)); 407 offset = -1; 408 goto err1; 409 } 410 } 411 if (pseg_flags) 412 *pseg_flags = ssp->ss_flags; 413 oldoffset = offset; 414 offset += lfs_btofsb(fs, lfs_sb_getsumsize(fs)); 415 416 ninos = howmany(ssp->ss_ninos, LFS_INOPB(fs)); 417 /* XXX ondisk32 */ 418 iaddr = (int32_t *)((char*)bp->b_data + lfs_sb_getsumsize(fs) - sizeof(int32_t)); 419 if (flags & CHECK_CKSUM) { 420 /* Count blocks */ 421 nblocks = 0; 422 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 423 for (i = 0; i < ssp->ss_nfinfo; ++i) { 424 nblocks += fip->fi_nblocks; 425 if (fip->fi_nblocks <= 0) 426 break; 427 /* XXX ondisk32 */ 428 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 429 (fip->fi_nblocks * sizeof(int32_t))); 430 } 431 nblocks += ninos; 432 /* Create the sum array */ 433 datap = dp = malloc(nblocks * sizeof(u_long), 434 M_SEGMENT, M_WAITOK); 435 } 436 437 /* Handle individual blocks */ 438 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 439 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 440 /* Inode block? */ 441 if (ninos && *iaddr == offset) { 442 if (flags & CHECK_CKSUM) { 443 /* Read in the head and add to the buffer */ 444 error = bread(devvp, LFS_FSBTODB(fs, offset), lfs_sb_getbsize(fs), 445 0, &dbp); 446 if (error) { 447 offset = -1; 448 goto err2; 449 } 450 (*dp++) = ((u_long *)(dbp->b_data))[0]; 451 brelse(dbp, BC_AGE); 452 } 453 if (flags & CHECK_UPDATE) { 454 if ((error = update_inoblk(fs, offset, cred, l)) 455 != 0) { 456 offset = -1; 457 goto err2; 458 } 459 } 460 offset += lfs_btofsb(fs, lfs_sb_getibsize(fs)); 461 --iaddr; 462 --ninos; 463 --i; /* compensate */ 464 continue; 465 } 466 size = lfs_sb_getbsize(fs); 467 for (j = 0; j < fip->fi_nblocks; ++j) { 468 if (j == fip->fi_nblocks - 1) 469 size = fip->fi_lastlength; 470 if (flags & CHECK_CKSUM) { 471 error = bread(devvp, LFS_FSBTODB(fs, offset), size, 472 0, &dbp); 473 if (error) { 474 offset = -1; 475 goto err2; 476 } 477 (*dp++) = ((u_long *)(dbp->b_data))[0]; 478 brelse(dbp, BC_AGE); 479 } 480 /* Account for and update any direct blocks */ 481 if ((flags & CHECK_UPDATE) && 482 fip->fi_ino > LFS_IFILE_INUM && 483 fip->fi_blocks[j] >= 0) { 484 update_meta(fs, fip->fi_ino, fip->fi_version, 485 fip->fi_blocks[j], offset, size, l); 486 } 487 offset += lfs_btofsb(fs, size); 488 } 489 /* XXX ondisk32 */ 490 fip = (FINFO *)(((char *)fip) + FINFOSIZE 491 + fip->fi_nblocks * sizeof(int32_t)); 492 } 493 /* Checksum the array, compare */ 494 if ((flags & CHECK_CKSUM) && 495 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 496 { 497 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 498 " (wanted %x got %x)\n", 499 offset, ssp->ss_datasum, cksum(datap, nblocks * 500 sizeof(u_long)))); 501 offset = -1; 502 goto err2; 503 } 504 505 /* If we're at the end of the segment, move to the next */ 506 if (lfs_dtosn(fs, offset + lfs_btofsb(fs, lfs_sb_getsumsize(fs) + lfs_sb_getbsize(fs))) != 507 lfs_dtosn(fs, offset)) { 508 if (lfs_dtosn(fs, offset) == lfs_dtosn(fs, ssp->ss_next)) { 509 offset = -1; 510 goto err2; 511 } 512 offset = ssp->ss_next; 513 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 514 " -> segment %d\n", offset, lfs_dtosn(fs,offset))); 515 } 516 517 if (flags & CHECK_UPDATE) { 518 lfs_sb_subavail(fs, offset - oldoffset); 519 /* Don't clog the buffer queue */ 520 mutex_enter(&lfs_lock); 521 if (locked_queue_count > LFS_MAX_BUFS || 522 locked_queue_bytes > LFS_MAX_BYTES) { 523 lfs_flush(fs, SEGM_CKP, 0); 524 } 525 mutex_exit(&lfs_lock); 526 } 527 528 err2: 529 if (flags & CHECK_CKSUM) 530 free(datap, M_SEGMENT); 531 err1: 532 brelse(bp, BC_AGE); 533 534 /* XXX should we update the serial number even for bad psegs? */ 535 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 536 lfs_sb_setserial(fs, nextserial); 537 return offset; 538 } 539 540 void 541 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l) 542 { 543 int flags, dirty; 544 daddr_t offset, oldoffset, lastgoodpseg; 545 int sn, curseg, do_rollforward; 546 struct proc *p; 547 kauth_cred_t cred; 548 SEGUSE *sup; 549 struct buf *bp; 550 551 p = l ? l->l_proc : NULL; 552 cred = p ? p->p_cred : NOCRED; 553 554 /* 555 * Roll forward. 556 * 557 * We don't roll forward for v1 filesystems, because 558 * of the danger that the clock was turned back between the last 559 * checkpoint and crash. This would roll forward garbage. 560 * 561 * v2 filesystems don't have this problem because they use a 562 * monotonically increasing serial number instead of a timestamp. 563 */ 564 do_rollforward = (!(lfs_sb_getpflags(fs) & LFS_PF_CLEAN) && 565 lfs_do_rfw && fs->lfs_version > 1 && p != NULL); 566 if (do_rollforward) { 567 u_int64_t nextserial; 568 /* 569 * Phase I: Find the address of the last good partial 570 * segment that was written after the checkpoint. Mark 571 * the segments in question dirty, so they won't be 572 * reallocated. 573 */ 574 lastgoodpseg = oldoffset = offset = lfs_sb_getoffset(fs); 575 flags = 0x0; 576 DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%" 577 PRIx64 "\n", offset)); 578 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 579 if (!(sup->su_flags & SEGUSE_DIRTY)) 580 lfs_sb_subnclean(fs, 1); 581 sup->su_flags |= SEGUSE_DIRTY; 582 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, offset), bp); 583 nextserial = lfs_sb_getserial(fs) + 1; 584 while ((offset = check_segsum(fs, offset, nextserial, 585 cred, CHECK_CKSUM, &flags, l)) > 0) { 586 nextserial++; 587 if (lfs_sntod(fs, oldoffset) != lfs_sntod(fs, offset)) { 588 LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset), 589 bp); 590 if (!(sup->su_flags & SEGUSE_DIRTY)) 591 lfs_sb_subnclean(fs, 1); 592 sup->su_flags |= SEGUSE_DIRTY; 593 LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, oldoffset), 594 bp); 595 } 596 597 DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%" 598 PRIx64 "\n", offset)); 599 if (flags & SS_DIROP) { 600 DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%" 601 PRIx64 "\n", oldoffset)); 602 if (!(flags & SS_CONT)) { 603 DLOG((DLOG_RF, "lfs_mountfs: dirops end " 604 "at 0x%" PRIx64 "\n", oldoffset)); 605 } 606 } 607 if (!(flags & SS_CONT)) 608 lastgoodpseg = offset; 609 oldoffset = offset; 610 } 611 if (flags & SS_CONT) { 612 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 613 "dirops discarded\n")); 614 } 615 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 616 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 617 oldoffset = lfs_sb_getoffset(fs); 618 if (lfs_sb_getoffset(fs) != lastgoodpseg) { 619 /* Don't overwrite what we're trying to preserve */ 620 offset = lfs_sb_getoffset(fs); 621 lfs_sb_setoffset(fs, lastgoodpseg); 622 lfs_sb_setcurseg(fs, lfs_sntod(fs, lfs_dtosn(fs, lfs_sb_getoffset(fs)))); 623 for (sn = curseg = lfs_dtosn(fs, lfs_sb_getcurseg(fs));;) { 624 sn = (sn + 1) % lfs_sb_getnseg(fs); 625 if (sn == curseg) 626 panic("lfs_mountfs: no clean segments"); 627 LFS_SEGENTRY(sup, fs, sn, bp); 628 dirty = (sup->su_flags & SEGUSE_DIRTY); 629 brelse(bp, 0); 630 if (!dirty) 631 break; 632 } 633 lfs_sb_setnextseg(fs, lfs_sntod(fs, sn)); 634 635 /* 636 * Phase II: Roll forward from the first superblock. 637 */ 638 while (offset != lastgoodpseg) { 639 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 640 PRIx64 "\n", offset)); 641 offset = check_segsum(fs, offset, 642 lfs_sb_getserial(fs) + 1, cred, CHECK_UPDATE, 643 NULL, l); 644 } 645 646 /* 647 * Finish: flush our changes to disk. 648 */ 649 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 650 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 651 "recovered %jd blocks\n", 652 (intmax_t)(lastgoodpseg - oldoffset))); 653 } 654 DLOG((DLOG_RF, "LFS roll forward complete\n")); 655 } 656 } 657
Indexes created Mon Sep 29 21:09:56 GMT 2025