lfs_rfw.c revision 1.2.4.2
1 /* $NetBSD: lfs_rfw.c,v 1.2.4.2 2006/08/11 15:47:37 yamt 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 #ifdef LFS_KERNEL_RFW 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.2.4.2 2006/08/11 15:47:37 yamt Exp $"); 43 44 #if defined(_KERNEL_OPT) 45 #include "opt_quota.h" 46 #endif 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/namei.h> 51 #include <sys/proc.h> 52 #include <sys/kernel.h> 53 #include <sys/vnode.h> 54 #include <sys/mount.h> 55 #include <sys/kthread.h> 56 #include <sys/buf.h> 57 #include <sys/device.h> 58 #include <sys/mbuf.h> 59 #include <sys/file.h> 60 #include <sys/disklabel.h> 61 #include <sys/ioctl.h> 62 #include <sys/errno.h> 63 #include <sys/malloc.h> 64 #include <sys/pool.h> 65 #include <sys/socket.h> 66 #include <sys/syslog.h> 67 #include <uvm/uvm_extern.h> 68 #include <sys/sysctl.h> 69 #include <sys/conf.h> 70 #include <sys/kauth.h> 71 72 #include <miscfs/specfs/specdev.h> 73 74 #include <ufs/ufs/quota.h> 75 #include <ufs/ufs/inode.h> 76 #include <ufs/ufs/ufsmount.h> 77 #include <ufs/ufs/ufs_extern.h> 78 79 #include <uvm/uvm.h> 80 #include <uvm/uvm_stat.h> 81 #include <uvm/uvm_pager.h> 82 #include <uvm/uvm_pdaemon.h> 83 84 #include <ufs/lfs/lfs.h> 85 #include <ufs/lfs/lfs_extern.h> 86 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/genfs/genfs_node.h> 89 90 /* 91 * Roll-forward code. 92 */ 93 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t, 94 kauth_cred_t, int, int *, struct lwp *); 95 96 /* 97 * Allocate a particular inode with a particular version number, freeing 98 * any previous versions of this inode that may have gone before. 99 * Used by the roll-forward code. 100 * 101 * XXX this function does not have appropriate locking to be used on a live fs; 102 * XXX but something similar could probably be used for an "undelete" call. 103 * 104 * Called with the Ifile inode locked. 105 */ 106 int 107 lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l, 108 struct vnode **vpp) 109 { 110 IFILE *ifp; 111 struct buf *bp, *cbp; 112 struct vnode *vp; 113 struct inode *ip; 114 ino_t tino, oldnext; 115 int error; 116 CLEANERINFO *cip; 117 118 ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */ 119 120 /* 121 * First, just try a vget. If the version number is the one we want, 122 * we don't have to do anything else. If the version number is wrong, 123 * take appropriate action. 124 */ 125 error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp); 126 if (error == 0) { 127 DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp)); 128 129 *vpp = vp; 130 ip = VTOI(vp); 131 if (ip->i_gen == vers) 132 return 0; 133 else if (ip->i_gen < vers) { 134 lfs_truncate(vp, (off_t)0, 0, NOCRED, l); 135 ip->i_gen = ip->i_ffs1_gen = vers; 136 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 137 return 0; 138 } else { 139 DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n", 140 ino, vers, ip->i_ffs1_gen)); 141 vput(vp); 142 *vpp = NULLVP; 143 return EEXIST; 144 } 145 } 146 147 /* 148 * The inode is not in use. Find it on the free list. 149 */ 150 /* If the Ifile is too short to contain this inum, extend it */ 151 while (VTOI(fs->lfs_ivnode)->i_size <= (ino / 152 fs->lfs_ifpb + fs->lfs_cleansz + fs->lfs_segtabsz) 153 << fs->lfs_bshift) { 154 lfs_extend_ifile(fs, NOCRED); 155 } 156 157 LFS_IENTRY(ifp, fs, ino, bp); 158 oldnext = ifp->if_nextfree; 159 ifp->if_version = vers; 160 brelse(bp); 161 162 LFS_GET_HEADFREE(fs, cip, cbp, &ino); 163 if (ino) { 164 LFS_PUT_HEADFREE(fs, cip, cbp, oldnext); 165 } else { 166 tino = ino; 167 while (1) { 168 LFS_IENTRY(ifp, fs, tino, bp); 169 if (ifp->if_nextfree == ino || 170 ifp->if_nextfree == LFS_UNUSED_INUM) 171 break; 172 tino = ifp->if_nextfree; 173 brelse(bp); 174 } 175 if (ifp->if_nextfree == LFS_UNUSED_INUM) { 176 brelse(bp); 177 return ENOENT; 178 } 179 ifp->if_nextfree = oldnext; 180 LFS_BWRITE_LOG(bp); 181 } 182 183 error = lfs_ialloc(fs, fs->lfs_ivnode, ino, vers, &vp); 184 if (error == 0) { 185 /* 186 * Make it VREG so we can put blocks on it. We will change 187 * this later if it turns out to be some other kind of file. 188 */ 189 ip = VTOI(vp); 190 ip->i_mode = ip->i_ffs1_mode = IFREG; 191 ip->i_nlink = ip->i_ffs1_nlink = 1; 192 ip->i_ffs_effnlink = 1; 193 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, &vp); 194 ip = VTOI(vp); 195 196 DLOG((DLOG_RF, "lfs_rf_valloc: ino %d vp %p\n", ino, vp)); 197 198 /* The dirop-nature of this vnode is past */ 199 lfs_unmark_vnode(vp); 200 (void)lfs_vunref(vp); 201 vp->v_flag &= ~VDIROP; 202 simple_lock(&fs->lfs_interlock); 203 simple_lock(&lfs_subsys_lock); 204 --lfs_dirvcount; 205 simple_unlock(&lfs_subsys_lock); 206 --fs->lfs_dirvcount; 207 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 208 wakeup(&lfs_dirvcount); 209 wakeup(&fs->lfs_dirvcount); 210 simple_unlock(&fs->lfs_interlock); 211 } 212 *vpp = vp; 213 return error; 214 } 215 216 /* 217 * Load the appropriate indirect block, and change the appropriate pointer. 218 * Mark the block dirty. Do segment and avail accounting. 219 */ 220 static int 221 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 222 daddr_t ndaddr, size_t size, struct lwp *l) 223 { 224 int error; 225 struct vnode *vp; 226 struct inode *ip; 227 #ifdef DEBUG 228 daddr_t odaddr; 229 struct indir a[NIADDR]; 230 int num; 231 int i; 232 #endif /* DEBUG */ 233 struct buf *bp; 234 SEGUSE *sup; 235 236 KASSERT(lbn >= 0); /* no indirect blocks */ 237 238 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 239 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 240 " returned %d\n", ino, error)); 241 return error; 242 } 243 244 if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size, 245 NOCRED, 0, &bp)) != 0) { 246 vput(vp); 247 return (error); 248 } 249 /* No need to write, the block is already on disk */ 250 if (bp->b_flags & B_DELWRI) { 251 LFS_UNLOCK_BUF(bp); 252 fs->lfs_avail += btofsb(fs, bp->b_bcount); 253 } 254 bp->b_flags |= B_INVAL; 255 brelse(bp); 256 257 /* 258 * Extend the file, if it is not large enough already. 259 * XXX this is not exactly right, we don't know how much of the 260 * XXX last block is actually used. We hope that an inode will 261 * XXX appear later to give the correct size. 262 */ 263 ip = VTOI(vp); 264 if (ip->i_size <= (lbn << fs->lfs_bshift)) { 265 u_int64_t newsize; 266 267 if (lbn < NDADDR) 268 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 269 (size - fs->lfs_fsize) + 1; 270 else 271 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1; 272 273 if (ip->i_size < newsize) { 274 ip->i_size = newsize; 275 /* 276 * tell vm our new size for the case the inode won't 277 * appear later. 278 */ 279 uvm_vnp_setsize(vp, newsize); 280 } 281 } 282 283 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 284 285 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 286 sup->su_nbytes += size; 287 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 288 289 /* differences here should be due to UNWRITTEN indirect blocks. */ 290 KASSERT((lblkno(fs, ip->i_size) > NDADDR && 291 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 292 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 293 294 #ifdef DEBUG 295 /* Now look again to make sure it worked */ 296 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 297 for (i = num; i > 0; i--) { 298 if (!a[i].in_exists) 299 panic("update_meta: absent %d lv indirect block", i); 300 } 301 if (dbtofsb(fs, odaddr) != ndaddr) 302 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 303 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 304 #endif /* DEBUG */ 305 vput(vp); 306 return 0; 307 } 308 309 static int 310 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred, 311 struct lwp *l) 312 { 313 struct vnode *devvp, *vp; 314 struct inode *ip; 315 struct ufs1_dinode *dip; 316 struct buf *dbp, *ibp; 317 int error; 318 daddr_t daddr; 319 IFILE *ifp; 320 SEGUSE *sup; 321 322 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 323 324 /* 325 * Get the inode, update times and perms. 326 * DO NOT update disk blocks, we do that separately. 327 */ 328 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp); 329 if (error) { 330 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 331 return error; 332 } 333 dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs); 334 while (--dip >= (struct ufs1_dinode *)dbp->b_data) { 335 if (dip->di_inumber > LFS_IFILE_INUM) { 336 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 337 l, &vp); 338 if (error) { 339 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 340 " returned %d\n", error)); 341 continue; 342 } 343 ip = VTOI(vp); 344 if (dip->di_size != ip->i_size) 345 lfs_truncate(vp, dip->di_size, 0, NOCRED, l); 346 /* Get mode, link count, size, and times */ 347 memcpy(ip->i_din.ffs1_din, dip, 348 offsetof(struct ufs1_dinode, di_db[0])); 349 350 /* Then the rest, except di_blocks */ 351 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 352 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 353 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 354 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 355 356 ip->i_mode = ip->i_ffs1_mode; 357 ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink; 358 ip->i_size = ip->i_ffs1_size; 359 360 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 361 362 /* Re-initialize to get type right */ 363 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 364 &vp); 365 vput(vp); 366 367 /* Record change in location */ 368 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 369 daddr = ifp->if_daddr; 370 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno); 371 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 372 /* And do segment accounting */ 373 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) { 374 if (daddr > 0) { 375 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), 376 ibp); 377 sup->su_nbytes -= sizeof (struct ufs1_dinode); 378 LFS_WRITESEGENTRY(sup, fs, 379 dtosn(fs, daddr), 380 ibp); 381 } 382 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 383 ibp); 384 sup->su_nbytes += sizeof (struct ufs1_dinode); 385 LFS_WRITESEGENTRY(sup, fs, 386 dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 387 ibp); 388 } 389 } 390 } 391 dbp->b_flags |= B_AGE; 392 brelse(dbp); 393 394 return 0; 395 } 396 397 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 398 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 399 400 static daddr_t 401 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 402 kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l) 403 { 404 struct vnode *devvp; 405 struct buf *bp, *dbp; 406 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 407 SEGSUM *ssp; 408 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 409 daddr_t oldoffset; 410 int32_t *iaddr; /* XXX ondisk32 */ 411 FINFO *fip; 412 SEGUSE *sup; 413 size_t size; 414 415 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 416 /* 417 * If the segment has a superblock and we're at the top 418 * of the segment, skip the superblock. 419 */ 420 if (sntod(fs, dtosn(fs, offset)) == offset) { 421 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 422 if (sup->su_flags & SEGUSE_SUPERBLOCK) 423 offset += btofsb(fs, LFS_SBPAD); 424 brelse(bp); 425 } 426 427 /* Read in the segment summary */ 428 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp); 429 if (error) 430 return -1; 431 432 /* Check summary checksum */ 433 ssp = (SEGSUM *)bp->b_data; 434 if (flags & CHECK_CKSUM) { 435 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 436 fs->lfs_sumsize - 437 sizeof(ssp->ss_sumsum))) { 438 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 439 offset = -1; 440 goto err1; 441 } 442 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 443 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 444 offset = -1; 445 goto err1; 446 } 447 if (ssp->ss_create < fs->lfs_tstamp) { 448 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 449 offset = -1; 450 goto err1; 451 } 452 } 453 if (fs->lfs_version > 1) { 454 if (ssp->ss_serial != nextserial) { 455 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 456 "\n", offset)); 457 offset = -1; 458 goto err1; 459 } 460 if (ssp->ss_ident != fs->lfs_ident) { 461 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 462 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset)); 463 offset = -1; 464 goto err1; 465 } 466 } 467 if (pseg_flags) 468 *pseg_flags = ssp->ss_flags; 469 oldoffset = offset; 470 offset += btofsb(fs, fs->lfs_sumsize); 471 472 ninos = howmany(ssp->ss_ninos, INOPB(fs)); 473 /* XXX ondisk32 */ 474 iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t)); 475 if (flags & CHECK_CKSUM) { 476 /* Count blocks */ 477 nblocks = 0; 478 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 479 for (i = 0; i < ssp->ss_nfinfo; ++i) { 480 nblocks += fip->fi_nblocks; 481 if (fip->fi_nblocks <= 0) 482 break; 483 /* XXX ondisk32 */ 484 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 485 (fip->fi_nblocks * sizeof(int32_t))); 486 } 487 nblocks += ninos; 488 /* Create the sum array */ 489 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 490 M_SEGMENT, M_WAITOK); 491 } 492 493 /* Handle individual blocks */ 494 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 495 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 496 /* Inode block? */ 497 if (ninos && *iaddr == offset) { 498 if (flags & CHECK_CKSUM) { 499 /* Read in the head and add to the buffer */ 500 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize, 501 cred, &dbp); 502 if (error) { 503 offset = -1; 504 goto err2; 505 } 506 (*dp++) = ((u_long *)(dbp->b_data))[0]; 507 dbp->b_flags |= B_AGE; 508 brelse(dbp); 509 } 510 if (flags & CHECK_UPDATE) { 511 if ((error = update_inoblk(fs, offset, cred, l)) 512 != 0) { 513 offset = -1; 514 goto err2; 515 } 516 } 517 offset += btofsb(fs, fs->lfs_ibsize); 518 --iaddr; 519 --ninos; 520 --i; /* compensate */ 521 continue; 522 } 523 size = fs->lfs_bsize; 524 for (j = 0; j < fip->fi_nblocks; ++j) { 525 if (j == fip->fi_nblocks - 1) 526 size = fip->fi_lastlength; 527 if (flags & CHECK_CKSUM) { 528 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp); 529 if (error) { 530 offset = -1; 531 goto err2; 532 } 533 (*dp++) = ((u_long *)(dbp->b_data))[0]; 534 dbp->b_flags |= B_AGE; 535 brelse(dbp); 536 } 537 /* Account for and update any direct blocks */ 538 if ((flags & CHECK_UPDATE) && 539 fip->fi_ino > LFS_IFILE_INUM && 540 fip->fi_blocks[j] >= 0) { 541 update_meta(fs, fip->fi_ino, fip->fi_version, 542 fip->fi_blocks[j], offset, size, l); 543 } 544 offset += btofsb(fs, size); 545 } 546 /* XXX ondisk32 */ 547 fip = (FINFO *)(((char *)fip) + FINFOSIZE 548 + fip->fi_nblocks * sizeof(int32_t)); 549 } 550 /* Checksum the array, compare */ 551 if ((flags & CHECK_CKSUM) && 552 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 553 { 554 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 555 " (wanted %x got %x)\n", 556 offset, ssp->ss_datasum, cksum(datap, nblocks * 557 sizeof(u_long)))); 558 offset = -1; 559 goto err2; 560 } 561 562 /* If we're at the end of the segment, move to the next */ 563 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 564 dtosn(fs, offset)) { 565 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) { 566 offset = -1; 567 goto err2; 568 } 569 offset = ssp->ss_next; 570 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 571 " -> segment %d\n", offset, dtosn(fs,offset))); 572 } 573 574 if (flags & CHECK_UPDATE) { 575 fs->lfs_avail -= (offset - oldoffset); 576 /* Don't clog the buffer queue */ 577 simple_lock(&lfs_subsys_lock); 578 if (locked_queue_count > LFS_MAX_BUFS || 579 locked_queue_bytes > LFS_MAX_BYTES) { 580 lfs_flush(fs, SEGM_CKP, 0); 581 } 582 simple_unlock(&lfs_subsys_lock); 583 } 584 585 err2: 586 if (flags & CHECK_CKSUM) 587 free(datap, M_SEGMENT); 588 err1: 589 bp->b_flags |= B_AGE; 590 brelse(bp); 591 592 /* XXX should we update the serial number even for bad psegs? */ 593 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 594 fs->lfs_serial = nextserial; 595 return offset; 596 } 597 598 void 599 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l) 600 { 601 int flags, dirty; 602 daddr_t offset, oldoffset, lastgoodpseg; 603 int sn, curseg, do_rollforward; 604 605 /* 606 * Roll forward. 607 * 608 * We don't roll forward for v1 filesystems, because 609 * of the danger that the clock was turned back between the last 610 * checkpoint and crash. This would roll forward garbage. 611 * 612 * v2 filesystems don't have this problem because they use a 613 * monotonically increasing serial number instead of a timestamp. 614 */ 615 do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) && 616 lfs_do_rfw && fs->lfs_version > 1 && p != NULL); 617 if (do_rollforward) { 618 u_int64_t nextserial; 619 /* 620 * Phase I: Find the address of the last good partial 621 * segment that was written after the checkpoint. Mark 622 * the segments in question dirty, so they won't be 623 * reallocated. 624 */ 625 lastgoodpseg = oldoffset = offset = fs->lfs_offset; 626 flags = 0x0; 627 DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%" 628 PRIx64 "\n", offset)); 629 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 630 if (!(sup->su_flags & SEGUSE_DIRTY)) 631 --fs->lfs_nclean; 632 sup->su_flags |= SEGUSE_DIRTY; 633 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp); 634 nextserial = fs->lfs_serial + 1; 635 while ((offset = check_segsum(fs, offset, nextserial, 636 cred, CHECK_CKSUM, &flags, l)) > 0) { 637 nextserial++; 638 if (sntod(fs, oldoffset) != sntod(fs, offset)) { 639 LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset), 640 bp); 641 if (!(sup->su_flags & SEGUSE_DIRTY)) 642 --fs->lfs_nclean; 643 sup->su_flags |= SEGUSE_DIRTY; 644 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset), 645 bp); 646 } 647 648 DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%" 649 PRIx64 "\n", offset)); 650 if (flags & SS_DIROP) { 651 DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%" 652 PRIx64 "\n", oldoffset)); 653 if (!(flags & SS_CONT)) 654 DLOG((DLOG_RF, "lfs_mountfs: dirops end " 655 "at 0x%" PRIx64 "\n", oldoffset)); 656 } 657 if (!(flags & SS_CONT)) 658 lastgoodpseg = offset; 659 oldoffset = offset; 660 } 661 if (flags & SS_CONT) { 662 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 663 "dirops discarded\n")); 664 } 665 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 666 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 667 oldoffset = fs->lfs_offset; 668 if (fs->lfs_offset != lastgoodpseg) { 669 /* Don't overwrite what we're trying to preserve */ 670 offset = fs->lfs_offset; 671 fs->lfs_offset = lastgoodpseg; 672 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset)); 673 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) { 674 sn = (sn + 1) % fs->lfs_nseg; 675 if (sn == curseg) 676 panic("lfs_mountfs: no clean segments"); 677 LFS_SEGENTRY(sup, fs, sn, bp); 678 dirty = (sup->su_flags & SEGUSE_DIRTY); 679 brelse(bp); 680 if (!dirty) 681 break; 682 } 683 fs->lfs_nextseg = sntod(fs, sn); 684 685 /* 686 * Phase II: Roll forward from the first superblock. 687 */ 688 while (offset != lastgoodpseg) { 689 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 690 PRIx64 "\n", offset)); 691 offset = check_segsum(fs, offset, 692 fs->lfs_serial + 1, cred, CHECK_UPDATE, 693 NULL, l); 694 } 695 696 /* 697 * Finish: flush our changes to disk. 698 */ 699 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 700 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 701 "recovered %lld blocks\n", 702 (long long)(lastgoodpseg - oldoffset))); 703 } 704 DLOG((DLOG_RF, "LFS roll forward complete\n")); 705 } 706 } 707 #endif /* LFS_KERNEL_RFW */ 708
Indexes created Sun Oct 19 02:09:48 GMT 2025