lfs_rfw.c revision 1.9
1 /* $NetBSD: lfs_rfw.c,v 1.9 2008/01/02 11:49:11 ad 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 #include <sys/cdefs.h> 40 __KERNEL_RCSID(0, "$NetBSD: lfs_rfw.c,v 1.9 2008/01/02 11:49:11 ad Exp $"); 41 42 #if defined(_KERNEL_OPT) 43 #include "opt_quota.h" 44 #endif 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/namei.h> 49 #include <sys/proc.h> 50 #include <sys/kernel.h> 51 #include <sys/vnode.h> 52 #include <sys/mount.h> 53 #include <sys/kthread.h> 54 #include <sys/buf.h> 55 #include <sys/device.h> 56 #include <sys/mbuf.h> 57 #include <sys/file.h> 58 #include <sys/disklabel.h> 59 #include <sys/ioctl.h> 60 #include <sys/errno.h> 61 #include <sys/malloc.h> 62 #include <sys/pool.h> 63 #include <sys/socket.h> 64 #include <sys/syslog.h> 65 #include <uvm/uvm_extern.h> 66 #include <sys/sysctl.h> 67 #include <sys/conf.h> 68 #include <sys/kauth.h> 69 70 #include <miscfs/specfs/specdev.h> 71 72 #include <ufs/ufs/quota.h> 73 #include <ufs/ufs/inode.h> 74 #include <ufs/ufs/ufsmount.h> 75 #include <ufs/ufs/ufs_extern.h> 76 77 #include <uvm/uvm.h> 78 #include <uvm/uvm_stat.h> 79 #include <uvm/uvm_pager.h> 80 #include <uvm/uvm_pdaemon.h> 81 82 #include <ufs/lfs/lfs.h> 83 #include <ufs/lfs/lfs_extern.h> 84 85 #include <miscfs/genfs/genfs.h> 86 #include <miscfs/genfs/genfs_node.h> 87 88 /* 89 * Roll-forward code. 90 */ 91 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t, 92 kauth_cred_t, int, int *, struct lwp *); 93 94 extern int lfs_do_rfw; 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); 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, 0); 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, 0); 174 } 175 if (ifp->if_nextfree == LFS_UNUSED_INUM) { 176 brelse(bp, 0); 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_uflag &= ~VU_DIROP; 202 mutex_enter(&lfs_lock); 203 --lfs_dirvcount; 204 --fs->lfs_dirvcount; 205 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 206 wakeup(&lfs_dirvcount); 207 wakeup(&fs->lfs_dirvcount); 208 mutex_exit(&lfs_lock); 209 } 210 *vpp = vp; 211 return error; 212 } 213 214 /* 215 * Load the appropriate indirect block, and change the appropriate pointer. 216 * Mark the block dirty. Do segment and avail accounting. 217 */ 218 static int 219 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 220 daddr_t ndaddr, size_t size, struct lwp *l) 221 { 222 int error; 223 struct vnode *vp; 224 struct inode *ip; 225 #ifdef DEBUG 226 daddr_t odaddr; 227 struct indir a[NIADDR]; 228 int num; 229 int i; 230 #endif /* DEBUG */ 231 struct buf *bp; 232 SEGUSE *sup; 233 234 KASSERT(lbn >= 0); /* no indirect blocks */ 235 236 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 237 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 238 " returned %d\n", ino, error)); 239 return error; 240 } 241 242 if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size, 243 NOCRED, 0, &bp)) != 0) { 244 vput(vp); 245 return (error); 246 } 247 /* No need to write, the block is already on disk */ 248 if (bp->b_oflags & BO_DELWRI) { 249 LFS_UNLOCK_BUF(bp); 250 fs->lfs_avail += btofsb(fs, bp->b_bcount); 251 } 252 brelse(bp, BC_INVAL); 253 254 /* 255 * Extend the file, if it is not large enough already. 256 * XXX this is not exactly right, we don't know how much of the 257 * XXX last block is actually used. We hope that an inode will 258 * XXX appear later to give the correct size. 259 */ 260 ip = VTOI(vp); 261 if (ip->i_size <= (lbn << fs->lfs_bshift)) { 262 u_int64_t newsize; 263 264 if (lbn < NDADDR) 265 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 266 (size - fs->lfs_fsize) + 1; 267 else 268 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1; 269 270 if (ip->i_size < newsize) { 271 ip->i_size = newsize; 272 /* 273 * tell vm our new size for the case the inode won't 274 * appear later. 275 */ 276 uvm_vnp_setsize(vp, newsize); 277 } 278 } 279 280 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 281 282 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 283 sup->su_nbytes += size; 284 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 285 286 /* differences here should be due to UNWRITTEN indirect blocks. */ 287 KASSERT((lblkno(fs, ip->i_size) > NDADDR && 288 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 289 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 290 291 #ifdef DEBUG 292 /* Now look again to make sure it worked */ 293 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 294 for (i = num; i > 0; i--) { 295 if (!a[i].in_exists) 296 panic("update_meta: absent %d lv indirect block", i); 297 } 298 if (dbtofsb(fs, odaddr) != ndaddr) 299 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 300 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 301 #endif /* DEBUG */ 302 vput(vp); 303 return 0; 304 } 305 306 static int 307 update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred, 308 struct lwp *l) 309 { 310 struct vnode *devvp, *vp; 311 struct inode *ip; 312 struct ufs1_dinode *dip; 313 struct buf *dbp, *ibp; 314 int error; 315 daddr_t daddr; 316 IFILE *ifp; 317 SEGUSE *sup; 318 319 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 320 321 /* 322 * Get the inode, update times and perms. 323 * DO NOT update disk blocks, we do that separately. 324 */ 325 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp); 326 if (error) { 327 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 328 return error; 329 } 330 dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs); 331 while (--dip >= (struct ufs1_dinode *)dbp->b_data) { 332 if (dip->di_inumber > LFS_IFILE_INUM) { 333 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 334 l, &vp); 335 if (error) { 336 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 337 " returned %d\n", error)); 338 continue; 339 } 340 ip = VTOI(vp); 341 if (dip->di_size != ip->i_size) 342 lfs_truncate(vp, dip->di_size, 0, NOCRED); 343 /* Get mode, link count, size, and times */ 344 memcpy(ip->i_din.ffs1_din, dip, 345 offsetof(struct ufs1_dinode, di_db[0])); 346 347 /* Then the rest, except di_blocks */ 348 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 349 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 350 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 351 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 352 353 ip->i_mode = ip->i_ffs1_mode; 354 ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink; 355 ip->i_size = ip->i_ffs1_size; 356 357 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 358 359 /* Re-initialize to get type right */ 360 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 361 &vp); 362 vput(vp); 363 364 /* Record change in location */ 365 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 366 daddr = ifp->if_daddr; 367 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno); 368 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 369 /* And do segment accounting */ 370 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) { 371 if (daddr > 0) { 372 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), 373 ibp); 374 sup->su_nbytes -= sizeof (struct ufs1_dinode); 375 LFS_WRITESEGENTRY(sup, fs, 376 dtosn(fs, daddr), 377 ibp); 378 } 379 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 380 ibp); 381 sup->su_nbytes += sizeof (struct ufs1_dinode); 382 LFS_WRITESEGENTRY(sup, fs, 383 dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 384 ibp); 385 } 386 } 387 } 388 brelse(dbp, BC_AGE); 389 390 return 0; 391 } 392 393 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 394 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 395 396 static daddr_t 397 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 398 kauth_cred_t cred, int flags, int *pseg_flags, struct lwp *l) 399 { 400 struct vnode *devvp; 401 struct buf *bp, *dbp; 402 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 403 SEGSUM *ssp; 404 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 405 daddr_t oldoffset; 406 int32_t *iaddr; /* XXX ondisk32 */ 407 FINFO *fip; 408 SEGUSE *sup; 409 size_t size; 410 411 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 412 /* 413 * If the segment has a superblock and we're at the top 414 * of the segment, skip the superblock. 415 */ 416 if (sntod(fs, dtosn(fs, offset)) == offset) { 417 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 418 if (sup->su_flags & SEGUSE_SUPERBLOCK) 419 offset += btofsb(fs, LFS_SBPAD); 420 brelse(bp, 0); 421 } 422 423 /* Read in the segment summary */ 424 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp); 425 if (error) 426 return -1; 427 428 /* Check summary checksum */ 429 ssp = (SEGSUM *)bp->b_data; 430 if (flags & CHECK_CKSUM) { 431 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 432 fs->lfs_sumsize - 433 sizeof(ssp->ss_sumsum))) { 434 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 435 offset = -1; 436 goto err1; 437 } 438 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 439 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 440 offset = -1; 441 goto err1; 442 } 443 if (ssp->ss_create < fs->lfs_tstamp) { 444 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 445 offset = -1; 446 goto err1; 447 } 448 } 449 if (fs->lfs_version > 1) { 450 if (ssp->ss_serial != nextserial) { 451 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 452 "\n", offset)); 453 offset = -1; 454 goto err1; 455 } 456 if (ssp->ss_ident != fs->lfs_ident) { 457 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 458 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset)); 459 offset = -1; 460 goto err1; 461 } 462 } 463 if (pseg_flags) 464 *pseg_flags = ssp->ss_flags; 465 oldoffset = offset; 466 offset += btofsb(fs, fs->lfs_sumsize); 467 468 ninos = howmany(ssp->ss_ninos, INOPB(fs)); 469 /* XXX ondisk32 */ 470 iaddr = (int32_t *)((char*)bp->b_data + fs->lfs_sumsize - sizeof(int32_t)); 471 if (flags & CHECK_CKSUM) { 472 /* Count blocks */ 473 nblocks = 0; 474 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 475 for (i = 0; i < ssp->ss_nfinfo; ++i) { 476 nblocks += fip->fi_nblocks; 477 if (fip->fi_nblocks <= 0) 478 break; 479 /* XXX ondisk32 */ 480 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 481 (fip->fi_nblocks * sizeof(int32_t))); 482 } 483 nblocks += ninos; 484 /* Create the sum array */ 485 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 486 M_SEGMENT, M_WAITOK); 487 } 488 489 /* Handle individual blocks */ 490 fip = (FINFO *)((char*)bp->b_data + SEGSUM_SIZE(fs)); 491 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 492 /* Inode block? */ 493 if (ninos && *iaddr == offset) { 494 if (flags & CHECK_CKSUM) { 495 /* Read in the head and add to the buffer */ 496 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize, 497 cred, &dbp); 498 if (error) { 499 offset = -1; 500 goto err2; 501 } 502 (*dp++) = ((u_long *)(dbp->b_data))[0]; 503 brelse(dbp, BC_AGE); 504 } 505 if (flags & CHECK_UPDATE) { 506 if ((error = update_inoblk(fs, offset, cred, l)) 507 != 0) { 508 offset = -1; 509 goto err2; 510 } 511 } 512 offset += btofsb(fs, fs->lfs_ibsize); 513 --iaddr; 514 --ninos; 515 --i; /* compensate */ 516 continue; 517 } 518 size = fs->lfs_bsize; 519 for (j = 0; j < fip->fi_nblocks; ++j) { 520 if (j == fip->fi_nblocks - 1) 521 size = fip->fi_lastlength; 522 if (flags & CHECK_CKSUM) { 523 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp); 524 if (error) { 525 offset = -1; 526 goto err2; 527 } 528 (*dp++) = ((u_long *)(dbp->b_data))[0]; 529 brelse(dbp, BC_AGE); 530 } 531 /* Account for and update any direct blocks */ 532 if ((flags & CHECK_UPDATE) && 533 fip->fi_ino > LFS_IFILE_INUM && 534 fip->fi_blocks[j] >= 0) { 535 update_meta(fs, fip->fi_ino, fip->fi_version, 536 fip->fi_blocks[j], offset, size, l); 537 } 538 offset += btofsb(fs, size); 539 } 540 /* XXX ondisk32 */ 541 fip = (FINFO *)(((char *)fip) + FINFOSIZE 542 + fip->fi_nblocks * sizeof(int32_t)); 543 } 544 /* Checksum the array, compare */ 545 if ((flags & CHECK_CKSUM) && 546 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 547 { 548 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 549 " (wanted %x got %x)\n", 550 offset, ssp->ss_datasum, cksum(datap, nblocks * 551 sizeof(u_long)))); 552 offset = -1; 553 goto err2; 554 } 555 556 /* If we're at the end of the segment, move to the next */ 557 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 558 dtosn(fs, offset)) { 559 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) { 560 offset = -1; 561 goto err2; 562 } 563 offset = ssp->ss_next; 564 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 565 " -> segment %d\n", offset, dtosn(fs,offset))); 566 } 567 568 if (flags & CHECK_UPDATE) { 569 fs->lfs_avail -= (offset - oldoffset); 570 /* Don't clog the buffer queue */ 571 mutex_enter(&lfs_lock); 572 if (locked_queue_count > LFS_MAX_BUFS || 573 locked_queue_bytes > LFS_MAX_BYTES) { 574 lfs_flush(fs, SEGM_CKP, 0); 575 } 576 mutex_exit(&lfs_lock); 577 } 578 579 err2: 580 if (flags & CHECK_CKSUM) 581 free(datap, M_SEGMENT); 582 err1: 583 brelse(bp, BC_AGE); 584 585 /* XXX should we update the serial number even for bad psegs? */ 586 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 587 fs->lfs_serial = nextserial; 588 return offset; 589 } 590 591 void 592 lfs_roll_forward(struct lfs *fs, struct mount *mp, struct lwp *l) 593 { 594 int flags, dirty; 595 daddr_t offset, oldoffset, lastgoodpseg; 596 int sn, curseg, do_rollforward; 597 struct proc *p; 598 kauth_cred_t cred; 599 SEGUSE *sup; 600 struct buf *bp; 601 602 p = l ? l->l_proc : NULL; 603 cred = p ? p->p_cred : NOCRED; 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 } 658 if (!(flags & SS_CONT)) 659 lastgoodpseg = offset; 660 oldoffset = offset; 661 } 662 if (flags & SS_CONT) { 663 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 664 "dirops discarded\n")); 665 } 666 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 667 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 668 oldoffset = fs->lfs_offset; 669 if (fs->lfs_offset != lastgoodpseg) { 670 /* Don't overwrite what we're trying to preserve */ 671 offset = fs->lfs_offset; 672 fs->lfs_offset = lastgoodpseg; 673 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset)); 674 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) { 675 sn = (sn + 1) % fs->lfs_nseg; 676 if (sn == curseg) 677 panic("lfs_mountfs: no clean segments"); 678 LFS_SEGENTRY(sup, fs, sn, bp); 679 dirty = (sup->su_flags & SEGUSE_DIRTY); 680 brelse(bp, 0); 681 if (!dirty) 682 break; 683 } 684 fs->lfs_nextseg = sntod(fs, sn); 685 686 /* 687 * Phase II: Roll forward from the first superblock. 688 */ 689 while (offset != lastgoodpseg) { 690 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 691 PRIx64 "\n", offset)); 692 offset = check_segsum(fs, offset, 693 fs->lfs_serial + 1, cred, CHECK_UPDATE, 694 NULL, l); 695 } 696 697 /* 698 * Finish: flush our changes to disk. 699 */ 700 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 701 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 702 "recovered %lld blocks\n", 703 (long long)(lastgoodpseg - oldoffset))); 704 } 705 DLOG((DLOG_RF, "LFS roll forward complete\n")); 706 } 707 } 708
Indexes created Mon Sep 29 21:09:56 GMT 2025