lfs_vfsops.c revision 1.199
1 /* $NetBSD: lfs_vfsops.c,v 1.199 2006/04/08 00:16:56 perseant 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 * Copyright (c) 1989, 1991, 1993, 1994 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.199 2006/04/08 00:16:56 perseant Exp $"); 71 72 #if defined(_KERNEL_OPT) 73 #include "opt_quota.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/proc.h> 80 #include <sys/kernel.h> 81 #include <sys/vnode.h> 82 #include <sys/mount.h> 83 #include <sys/kthread.h> 84 #include <sys/buf.h> 85 #include <sys/device.h> 86 #include <sys/mbuf.h> 87 #include <sys/file.h> 88 #include <sys/disklabel.h> 89 #include <sys/ioctl.h> 90 #include <sys/errno.h> 91 #include <sys/malloc.h> 92 #include <sys/pool.h> 93 #include <sys/socket.h> 94 #include <sys/syslog.h> 95 #include <uvm/uvm_extern.h> 96 #include <sys/sysctl.h> 97 #include <sys/conf.h> 98 99 #include <miscfs/specfs/specdev.h> 100 101 #include <ufs/ufs/quota.h> 102 #include <ufs/ufs/inode.h> 103 #include <ufs/ufs/ufsmount.h> 104 #include <ufs/ufs/ufs_extern.h> 105 106 #include <uvm/uvm.h> 107 #include <uvm/uvm_stat.h> 108 #include <uvm/uvm_pager.h> 109 #include <uvm/uvm_pdaemon.h> 110 111 #include <ufs/lfs/lfs.h> 112 #include <ufs/lfs/lfs_extern.h> 113 114 #include <miscfs/genfs/genfs.h> 115 #include <miscfs/genfs/genfs_node.h> 116 117 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int); 118 static boolean_t lfs_issequential_hole(const struct ufsmount *, 119 daddr_t, daddr_t); 120 121 static int lfs_mountfs(struct vnode *, struct mount *, struct lwp *); 122 static void warn_ifile_size(struct lfs *); 123 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t, 124 struct ucred *, int, int *, struct lwp *); 125 126 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc; 127 extern const struct vnodeopv_desc lfs_specop_opv_desc; 128 extern const struct vnodeopv_desc lfs_fifoop_opv_desc; 129 130 pid_t lfs_writer_daemon = 0; 131 int lfs_do_flush = 0; 132 int lfs_do_rfw = 0; 133 134 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = { 135 &lfs_vnodeop_opv_desc, 136 &lfs_specop_opv_desc, 137 &lfs_fifoop_opv_desc, 138 NULL, 139 }; 140 141 struct vfsops lfs_vfsops = { 142 MOUNT_LFS, 143 lfs_mount, 144 ufs_start, 145 lfs_unmount, 146 ufs_root, 147 ufs_quotactl, 148 lfs_statvfs, 149 lfs_sync, 150 lfs_vget, 151 lfs_fhtovp, 152 lfs_vptofh, 153 lfs_init, 154 lfs_reinit, 155 lfs_done, 156 lfs_mountroot, 157 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 158 vfs_stdextattrctl, 159 lfs_vnodeopv_descs, 160 }; 161 VFS_ATTACH(lfs_vfsops); 162 163 const struct genfs_ops lfs_genfsops = { 164 .gop_size = lfs_gop_size, 165 .gop_alloc = ufs_gop_alloc, 166 .gop_write = lfs_gop_write, 167 .gop_markupdate = ufs_gop_markupdate, 168 }; 169 170 static const struct ufs_ops lfs_ufsops = { 171 .uo_itimes = NULL, 172 .uo_update = lfs_update, 173 .uo_truncate = lfs_truncate, 174 .uo_valloc = lfs_valloc, 175 .uo_vfree = lfs_vfree, 176 .uo_balloc = lfs_balloc, 177 }; 178 179 /* 180 * XXX Same structure as FFS inodes? Should we share a common pool? 181 */ 182 POOL_INIT(lfs_inode_pool, sizeof(struct inode), 0, 0, 0, "lfsinopl", 183 &pool_allocator_nointr); 184 POOL_INIT(lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, "lfsdinopl", 185 &pool_allocator_nointr); 186 POOL_INIT(lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, "lfsinoextpl", 187 &pool_allocator_nointr); 188 POOL_INIT(lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, "lfslbnpool", 189 &pool_allocator_nointr); 190 191 /* 192 * The writer daemon. UVM keeps track of how many dirty pages we are holding 193 * in lfs_subsys_pages; the daemon flushes the filesystem when this value 194 * crosses the (user-defined) threshhold LFS_MAX_PAGES. 195 */ 196 static void 197 lfs_writerd(void *arg) 198 { 199 struct mount *mp, *nmp; 200 struct lfs *fs; 201 int loopcount; 202 203 lfs_writer_daemon = curproc->p_pid; 204 205 simple_lock(&lfs_subsys_lock); 206 for (;;) { 207 ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", hz/10, 208 &lfs_subsys_lock); 209 210 /* 211 * Look through the list of LFSs to see if any of them 212 * have requested pageouts. 213 */ 214 simple_lock(&mountlist_slock); 215 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 216 mp = nmp) { 217 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 218 nmp = CIRCLEQ_NEXT(mp, mnt_list); 219 continue; 220 } 221 if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, 222 MFSNAMELEN) == 0) { 223 fs = VFSTOUFS(mp)->um_lfs; 224 simple_lock(&fs->lfs_interlock); 225 if (fs->lfs_pdflush || 226 !TAILQ_EMPTY(&fs->lfs_pchainhd)) { 227 DLOG((DLOG_FLUSH, "lfs_writerd: pdflush set\n")); 228 fs->lfs_pdflush = 0; 229 lfs_flush_fs(fs, 0); 230 } 231 simple_unlock(&fs->lfs_interlock); 232 } 233 234 simple_lock(&mountlist_slock); 235 nmp = CIRCLEQ_NEXT(mp, mnt_list); 236 vfs_unbusy(mp); 237 } 238 simple_unlock(&mountlist_slock); 239 240 /* 241 * If global state wants a flush, flush everything. 242 */ 243 simple_lock(&lfs_subsys_lock); 244 loopcount = 0; 245 if (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS || 246 locked_queue_bytes > LFS_MAX_BYTES || 247 lfs_subsys_pages > LFS_MAX_PAGES) { 248 249 if (lfs_do_flush) 250 DLOG((DLOG_FLUSH, "daemon: lfs_do_flush\n")); 251 if (locked_queue_count > LFS_MAX_BUFS) 252 DLOG((DLOG_FLUSH, "daemon: lqc = %d, max %d\n", 253 locked_queue_count, LFS_MAX_BUFS)); 254 if (locked_queue_bytes > LFS_MAX_BYTES) 255 DLOG((DLOG_FLUSH, "daemon: lqb = %ld, max %ld\n", 256 locked_queue_bytes, LFS_MAX_BYTES)); 257 if (lfs_subsys_pages > LFS_MAX_PAGES) 258 DLOG((DLOG_FLUSH, "daemon: lssp = %d, max %d\n", 259 lfs_subsys_pages, LFS_MAX_PAGES)); 260 261 lfs_flush(NULL, SEGM_WRITERD, 0); 262 lfs_do_flush = 0; 263 } 264 } 265 /* NOTREACHED */ 266 } 267 268 /* 269 * Initialize the filesystem, most work done by ufs_init. 270 */ 271 void 272 lfs_init() 273 { 274 #ifdef _LKM 275 malloc_type_attach(M_SEGMENT); 276 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0, 277 "lfsinopl", &pool_allocator_nointr); 278 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0, 279 "lfsdinopl", &pool_allocator_nointr); 280 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0, 281 "lfsinoextpl", &pool_allocator_nointr); 282 pool_init(&lfs_lbnentry_pool, sizeof(struct lbnentry), 0, 0, 0, 283 "lfslbnpool", &pool_allocator_nointr); 284 #endif 285 ufs_init(); 286 287 #ifdef DEBUG 288 memset(lfs_log, 0, sizeof(lfs_log)); 289 #endif 290 simple_lock_init(&lfs_subsys_lock); 291 } 292 293 void 294 lfs_reinit() 295 { 296 ufs_reinit(); 297 } 298 299 void 300 lfs_done() 301 { 302 ufs_done(); 303 #ifdef _LKM 304 pool_destroy(&lfs_inode_pool); 305 pool_destroy(&lfs_dinode_pool); 306 pool_destroy(&lfs_inoext_pool); 307 pool_destroy(&lfs_lbnentry_pool); 308 malloc_type_detach(M_SEGMENT); 309 #endif 310 } 311 312 /* 313 * Called by main() when ufs is going to be mounted as root. 314 */ 315 int 316 lfs_mountroot() 317 { 318 extern struct vnode *rootvp; 319 struct mount *mp; 320 struct lwp *l = curlwp; /* XXX */ 321 int error; 322 323 if (device_class(root_device) != DV_DISK) 324 return (ENODEV); 325 326 if (rootdev == NODEV) 327 return (ENODEV); 328 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) { 329 vrele(rootvp); 330 return (error); 331 } 332 if ((error = lfs_mountfs(rootvp, mp, l))) { 333 mp->mnt_op->vfs_refcount--; 334 vfs_unbusy(mp); 335 free(mp, M_MOUNT); 336 return (error); 337 } 338 simple_lock(&mountlist_slock); 339 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 340 simple_unlock(&mountlist_slock); 341 (void)lfs_statvfs(mp, &mp->mnt_stat, l); 342 vfs_unbusy(mp); 343 setrootfstime((time_t)(VFSTOUFS(mp)->um_lfs->lfs_tstamp)); 344 return (0); 345 } 346 347 /* 348 * VFS Operations. 349 * 350 * mount system call 351 */ 352 int 353 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct lwp *l) 354 { 355 struct vnode *devvp; 356 struct ufs_args args; 357 struct ufsmount *ump = NULL; 358 struct lfs *fs = NULL; /* LFS */ 359 struct proc *p; 360 int error, update; 361 mode_t accessmode; 362 363 p = l->l_proc; 364 if (mp->mnt_flag & MNT_GETARGS) { 365 ump = VFSTOUFS(mp); 366 if (ump == NULL) 367 return EIO; 368 args.fspec = NULL; 369 return copyout(&args, data, sizeof(args)); 370 } 371 error = copyin(data, &args, sizeof (struct ufs_args)); 372 if (error) 373 return (error); 374 375 update = mp->mnt_flag & MNT_UPDATE; 376 377 /* Check arguments */ 378 if (args.fspec != NULL) { 379 /* 380 * Look up the name and verify that it's sane. 381 */ 382 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, l); 383 if ((error = namei(ndp)) != 0) 384 return (error); 385 devvp = ndp->ni_vp; 386 387 if (!update) { 388 /* 389 * Be sure this is a valid block device 390 */ 391 if (devvp->v_type != VBLK) 392 error = ENOTBLK; 393 else if (bdevsw_lookup(devvp->v_rdev) == NULL) 394 error = ENXIO; 395 } else { 396 /* 397 * Be sure we're still naming the same device 398 * used for our initial mount 399 */ 400 ump = VFSTOUFS(mp); 401 if (devvp != ump->um_devvp) 402 error = EINVAL; 403 } 404 } else { 405 if (!update) { 406 /* New mounts must have a filename for the device */ 407 return (EINVAL); 408 } else { 409 /* Use the extant mount */ 410 ump = VFSTOUFS(mp); 411 devvp = ump->um_devvp; 412 vref(devvp); 413 } 414 } 415 416 417 /* 418 * If mount by non-root, then verify that user has necessary 419 * permissions on the device. 420 */ 421 if (error == 0 && p->p_ucred->cr_uid != 0) { 422 accessmode = VREAD; 423 if (update ? 424 (mp->mnt_iflag & IMNT_WANTRDWR) != 0 : 425 (mp->mnt_flag & MNT_RDONLY) == 0) 426 accessmode |= VWRITE; 427 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 428 error = VOP_ACCESS(devvp, accessmode, p->p_ucred, l); 429 VOP_UNLOCK(devvp, 0); 430 } 431 432 if (error) { 433 vrele(devvp); 434 return (error); 435 } 436 437 if (!update) { 438 int flags; 439 440 /* 441 * Disallow multiple mounts of the same device. 442 * Disallow mounting of a device that is currently in use 443 * (except for root, which might share swap device for 444 * miniroot). 445 */ 446 error = vfs_mountedon(devvp); 447 if (error) 448 goto fail; 449 if (vcount(devvp) > 1 && devvp != rootvp) { 450 error = EBUSY; 451 goto fail; 452 } 453 if (mp->mnt_flag & MNT_RDONLY) 454 flags = FREAD; 455 else 456 flags = FREAD|FWRITE; 457 error = VOP_OPEN(devvp, flags, FSCRED, l); 458 if (error) 459 goto fail; 460 error = lfs_mountfs(devvp, mp, l); /* LFS */ 461 if (error) { 462 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 463 (void)VOP_CLOSE(devvp, flags, NOCRED, l); 464 VOP_UNLOCK(devvp, 0); 465 goto fail; 466 } 467 468 ump = VFSTOUFS(mp); 469 fs = ump->um_lfs; 470 } else { 471 /* 472 * Update the mount. 473 */ 474 475 /* 476 * The initial mount got a reference on this 477 * device, so drop the one obtained via 478 * namei(), above. 479 */ 480 vrele(devvp); 481 482 ump = VFSTOUFS(mp); 483 fs = ump->um_lfs; 484 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) { 485 /* 486 * Changing from read-only to read/write. 487 * Note in the superblocks that we're writing. 488 */ 489 fs->lfs_ronly = 0; 490 if (fs->lfs_pflags & LFS_PF_CLEAN) { 491 fs->lfs_pflags &= ~LFS_PF_CLEAN; 492 lfs_writesuper(fs, fs->lfs_sboffs[0]); 493 lfs_writesuper(fs, fs->lfs_sboffs[1]); 494 } 495 } 496 if (args.fspec == NULL) 497 return EINVAL; 498 } 499 500 error = set_statvfs_info(path, UIO_USERSPACE, args.fspec, 501 UIO_USERSPACE, mp, l); 502 if (error == 0) 503 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 504 sizeof(fs->lfs_fsmnt)); 505 return error; 506 507 fail: 508 vrele(devvp); 509 return (error); 510 } 511 512 /* 513 * Roll-forward code. 514 */ 515 516 /* 517 * Load the appropriate indirect block, and change the appropriate pointer. 518 * Mark the block dirty. Do segment and avail accounting. 519 */ 520 static int 521 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 522 daddr_t ndaddr, size_t size, struct lwp *l) 523 { 524 int error; 525 struct vnode *vp; 526 struct inode *ip; 527 #ifdef DEBUG 528 daddr_t odaddr; 529 struct indir a[NIADDR]; 530 int num; 531 int i; 532 #endif /* DEBUG */ 533 struct buf *bp; 534 SEGUSE *sup; 535 536 KASSERT(lbn >= 0); /* no indirect blocks */ 537 538 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 539 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 540 " returned %d\n", ino, error)); 541 return error; 542 } 543 544 if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size, 545 NOCRED, 0, &bp)) != 0) { 546 vput(vp); 547 return (error); 548 } 549 /* No need to write, the block is already on disk */ 550 if (bp->b_flags & B_DELWRI) { 551 LFS_UNLOCK_BUF(bp); 552 fs->lfs_avail += btofsb(fs, bp->b_bcount); 553 } 554 bp->b_flags |= B_INVAL; 555 brelse(bp); 556 557 /* 558 * Extend the file, if it is not large enough already. 559 * XXX this is not exactly right, we don't know how much of the 560 * XXX last block is actually used. We hope that an inode will 561 * XXX appear later to give the correct size. 562 */ 563 ip = VTOI(vp); 564 if (ip->i_size <= (lbn << fs->lfs_bshift)) { 565 u_int64_t newsize; 566 567 if (lbn < NDADDR) 568 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 569 (size - fs->lfs_fsize) + 1; 570 else 571 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1; 572 573 if (ip->i_size < newsize) { 574 ip->i_size = newsize; 575 /* 576 * tell vm our new size for the case the inode won't 577 * appear later. 578 */ 579 uvm_vnp_setsize(vp, newsize); 580 } 581 } 582 583 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 584 585 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 586 sup->su_nbytes += size; 587 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 588 589 /* differences here should be due to UNWRITTEN indirect blocks. */ 590 KASSERT((lblkno(fs, ip->i_size) > NDADDR && 591 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 592 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 593 594 #ifdef DEBUG 595 /* Now look again to make sure it worked */ 596 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 597 for (i = num; i > 0; i--) { 598 if (!a[i].in_exists) 599 panic("update_meta: absent %d lv indirect block", i); 600 } 601 if (dbtofsb(fs, odaddr) != ndaddr) 602 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 603 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 604 #endif /* DEBUG */ 605 vput(vp); 606 return 0; 607 } 608 609 static int 610 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred, 611 struct lwp *l) 612 { 613 struct vnode *devvp, *vp; 614 struct inode *ip; 615 struct ufs1_dinode *dip; 616 struct buf *dbp, *ibp; 617 int error; 618 daddr_t daddr; 619 IFILE *ifp; 620 SEGUSE *sup; 621 622 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 623 624 /* 625 * Get the inode, update times and perms. 626 * DO NOT update disk blocks, we do that separately. 627 */ 628 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp); 629 if (error) { 630 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 631 return error; 632 } 633 dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs); 634 while (--dip >= (struct ufs1_dinode *)dbp->b_data) { 635 if (dip->di_inumber > LFS_IFILE_INUM) { 636 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 637 l, &vp); 638 if (error) { 639 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 640 " returned %d\n", error)); 641 continue; 642 } 643 ip = VTOI(vp); 644 if (dip->di_size != ip->i_size) 645 lfs_truncate(vp, dip->di_size, 0, NOCRED, l); 646 /* Get mode, link count, size, and times */ 647 memcpy(ip->i_din.ffs1_din, dip, 648 offsetof(struct ufs1_dinode, di_db[0])); 649 650 /* Then the rest, except di_blocks */ 651 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 652 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 653 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 654 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 655 656 ip->i_mode = ip->i_ffs1_mode; 657 ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink; 658 ip->i_size = ip->i_ffs1_size; 659 660 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 661 662 /* Re-initialize to get type right */ 663 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 664 &vp); 665 vput(vp); 666 667 /* Record change in location */ 668 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 669 daddr = ifp->if_daddr; 670 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno); 671 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 672 /* And do segment accounting */ 673 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) { 674 if (daddr > 0) { 675 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), 676 ibp); 677 sup->su_nbytes -= sizeof (struct ufs1_dinode); 678 LFS_WRITESEGENTRY(sup, fs, 679 dtosn(fs, daddr), 680 ibp); 681 } 682 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 683 ibp); 684 sup->su_nbytes += sizeof (struct ufs1_dinode); 685 LFS_WRITESEGENTRY(sup, fs, 686 dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 687 ibp); 688 } 689 } 690 } 691 dbp->b_flags |= B_AGE; 692 brelse(dbp); 693 694 return 0; 695 } 696 697 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 698 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 699 700 static daddr_t 701 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 702 struct ucred *cred, int flags, int *pseg_flags, struct lwp *l) 703 { 704 struct vnode *devvp; 705 struct buf *bp, *dbp; 706 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 707 SEGSUM *ssp; 708 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 709 daddr_t oldoffset; 710 int32_t *iaddr; /* XXX ondisk32 */ 711 FINFO *fip; 712 SEGUSE *sup; 713 size_t size; 714 715 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 716 /* 717 * If the segment has a superblock and we're at the top 718 * of the segment, skip the superblock. 719 */ 720 if (sntod(fs, dtosn(fs, offset)) == offset) { 721 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 722 if (sup->su_flags & SEGUSE_SUPERBLOCK) 723 offset += btofsb(fs, LFS_SBPAD); 724 brelse(bp); 725 } 726 727 /* Read in the segment summary */ 728 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp); 729 if (error) 730 return -1; 731 732 /* Check summary checksum */ 733 ssp = (SEGSUM *)bp->b_data; 734 if (flags & CHECK_CKSUM) { 735 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 736 fs->lfs_sumsize - 737 sizeof(ssp->ss_sumsum))) { 738 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 739 offset = -1; 740 goto err1; 741 } 742 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 743 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 744 offset = -1; 745 goto err1; 746 } 747 if (ssp->ss_create < fs->lfs_tstamp) { 748 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 749 offset = -1; 750 goto err1; 751 } 752 } 753 if (fs->lfs_version > 1) { 754 if (ssp->ss_serial != nextserial) { 755 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 756 "\n", offset)); 757 offset = -1; 758 goto err1; 759 } 760 if (ssp->ss_ident != fs->lfs_ident) { 761 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 762 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset)); 763 offset = -1; 764 goto err1; 765 } 766 } 767 if (pseg_flags) 768 *pseg_flags = ssp->ss_flags; 769 oldoffset = offset; 770 offset += btofsb(fs, fs->lfs_sumsize); 771 772 ninos = howmany(ssp->ss_ninos, INOPB(fs)); 773 /* XXX ondisk32 */ 774 iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t)); 775 if (flags & CHECK_CKSUM) { 776 /* Count blocks */ 777 nblocks = 0; 778 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 779 for (i = 0; i < ssp->ss_nfinfo; ++i) { 780 nblocks += fip->fi_nblocks; 781 if (fip->fi_nblocks <= 0) 782 break; 783 /* XXX ondisk32 */ 784 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 785 (fip->fi_nblocks * sizeof(int32_t))); 786 } 787 nblocks += ninos; 788 /* Create the sum array */ 789 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 790 M_SEGMENT, M_WAITOK); 791 } 792 793 /* Handle individual blocks */ 794 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 795 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 796 /* Inode block? */ 797 if (ninos && *iaddr == offset) { 798 if (flags & CHECK_CKSUM) { 799 /* Read in the head and add to the buffer */ 800 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize, 801 cred, &dbp); 802 if (error) { 803 offset = -1; 804 goto err2; 805 } 806 (*dp++) = ((u_long *)(dbp->b_data))[0]; 807 dbp->b_flags |= B_AGE; 808 brelse(dbp); 809 } 810 if (flags & CHECK_UPDATE) { 811 if ((error = update_inoblk(fs, offset, cred, l)) 812 != 0) { 813 offset = -1; 814 goto err2; 815 } 816 } 817 offset += btofsb(fs, fs->lfs_ibsize); 818 --iaddr; 819 --ninos; 820 --i; /* compensate */ 821 continue; 822 } 823 size = fs->lfs_bsize; 824 for (j = 0; j < fip->fi_nblocks; ++j) { 825 if (j == fip->fi_nblocks - 1) 826 size = fip->fi_lastlength; 827 if (flags & CHECK_CKSUM) { 828 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp); 829 if (error) { 830 offset = -1; 831 goto err2; 832 } 833 (*dp++) = ((u_long *)(dbp->b_data))[0]; 834 dbp->b_flags |= B_AGE; 835 brelse(dbp); 836 } 837 /* Account for and update any direct blocks */ 838 if ((flags & CHECK_UPDATE) && 839 fip->fi_ino > LFS_IFILE_INUM && 840 fip->fi_blocks[j] >= 0) { 841 update_meta(fs, fip->fi_ino, fip->fi_version, 842 fip->fi_blocks[j], offset, size, l); 843 } 844 offset += btofsb(fs, size); 845 } 846 /* XXX ondisk32 */ 847 fip = (FINFO *)(((char *)fip) + FINFOSIZE 848 + fip->fi_nblocks * sizeof(int32_t)); 849 } 850 /* Checksum the array, compare */ 851 if ((flags & CHECK_CKSUM) && 852 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 853 { 854 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 855 " (wanted %x got %x)\n", 856 offset, ssp->ss_datasum, cksum(datap, nblocks * 857 sizeof(u_long)))); 858 offset = -1; 859 goto err2; 860 } 861 862 /* If we're at the end of the segment, move to the next */ 863 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 864 dtosn(fs, offset)) { 865 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) { 866 offset = -1; 867 goto err2; 868 } 869 offset = ssp->ss_next; 870 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 871 " -> segment %d\n", offset, dtosn(fs,offset))); 872 } 873 874 if (flags & CHECK_UPDATE) { 875 fs->lfs_avail -= (offset - oldoffset); 876 /* Don't clog the buffer queue */ 877 simple_lock(&lfs_subsys_lock); 878 if (locked_queue_count > LFS_MAX_BUFS || 879 locked_queue_bytes > LFS_MAX_BYTES) { 880 lfs_flush(fs, SEGM_CKP, 0); 881 } 882 simple_unlock(&lfs_subsys_lock); 883 } 884 885 err2: 886 if (flags & CHECK_CKSUM) 887 free(datap, M_SEGMENT); 888 err1: 889 bp->b_flags |= B_AGE; 890 brelse(bp); 891 892 /* XXX should we update the serial number even for bad psegs? */ 893 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 894 fs->lfs_serial = nextserial; 895 return offset; 896 } 897 898 /* 899 * Common code for mount and mountroot 900 * LFS specific 901 */ 902 int 903 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 904 { 905 struct dlfs *tdfs, *dfs, *adfs; 906 struct lfs *fs; 907 struct ufsmount *ump; 908 struct vnode *vp; 909 struct buf *bp, *abp; 910 struct partinfo dpart; 911 struct proc *p; 912 dev_t dev; 913 int error, i, ronly, secsize, fsbsize; 914 struct ucred *cred; 915 CLEANERINFO *cip; 916 SEGUSE *sup; 917 int flags, dirty, do_rollforward; 918 daddr_t offset, oldoffset, lastgoodpseg, sb_addr; 919 int sn, curseg; 920 921 p = l ? l->l_proc : NULL; 922 cred = p ? p->p_ucred : NOCRED; 923 924 /* 925 * Flush out any old buffers remaining from a previous use. 926 */ 927 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 928 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 929 VOP_UNLOCK(devvp, 0); 930 if (error) 931 return (error); 932 933 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 934 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0) 935 secsize = DEV_BSIZE; 936 else 937 secsize = dpart.disklab->d_secsize; 938 939 /* Don't free random space on error. */ 940 bp = NULL; 941 abp = NULL; 942 ump = NULL; 943 944 sb_addr = LFS_LABELPAD / secsize; 945 while (1) { 946 /* Read in the superblock. */ 947 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 948 if (error) 949 goto out; 950 dfs = (struct dlfs *)bp->b_data; 951 952 /* Check the basics. */ 953 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 954 dfs->dlfs_version > LFS_VERSION || 955 dfs->dlfs_bsize < sizeof(struct dlfs)) { 956 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 957 error = EINVAL; /* XXX needs translation */ 958 goto out; 959 } 960 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 961 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 962 dfs->dlfs_inodefmt)); 963 error = EINVAL; 964 goto out; 965 } 966 967 if (dfs->dlfs_version == 1) 968 fsbsize = secsize; 969 else { 970 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 971 dfs->dlfs_fsbtodb); 972 /* 973 * Could be, if the frag size is large enough, that we 974 * don't have the "real" primary superblock. If that's 975 * the case, get the real one, and try again. 976 */ 977 if (sb_addr != dfs->dlfs_sboffs[0] << 978 dfs->dlfs_fsbtodb) { 979 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 980 " 0x%llx is not right, trying 0x%llx\n", 981 (long long)sb_addr, 982 (long long)(dfs->dlfs_sboffs[0] << 983 dfs->dlfs_fsbtodb))); 984 sb_addr = dfs->dlfs_sboffs[0] << 985 dfs->dlfs_fsbtodb; 986 brelse(bp); 987 continue; 988 } 989 } 990 break; 991 } 992 993 /* 994 * Check the second superblock to see which is newer; then mount 995 * using the older of the two. This is necessary to ensure that 996 * the filesystem is valid if it was not unmounted cleanly. 997 */ 998 999 if (dfs->dlfs_sboffs[1] && 1000 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 1001 { 1002 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 1003 LFS_SBPAD, cred, &abp); 1004 if (error) 1005 goto out; 1006 adfs = (struct dlfs *)abp->b_data; 1007 1008 if (dfs->dlfs_version == 1) { 1009 /* 1s resolution comparison */ 1010 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 1011 tdfs = adfs; 1012 else 1013 tdfs = dfs; 1014 } else { 1015 /* monotonic infinite-resolution comparison */ 1016 if (adfs->dlfs_serial < dfs->dlfs_serial) 1017 tdfs = adfs; 1018 else 1019 tdfs = dfs; 1020 } 1021 1022 /* Check the basics. */ 1023 if (tdfs->dlfs_magic != LFS_MAGIC || 1024 tdfs->dlfs_bsize > MAXBSIZE || 1025 tdfs->dlfs_version > LFS_VERSION || 1026 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 1027 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 1028 " sanity failed\n")); 1029 error = EINVAL; /* XXX needs translation */ 1030 goto out; 1031 } 1032 } else { 1033 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 1034 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 1035 error = EINVAL; 1036 goto out; 1037 } 1038 1039 /* Allocate the mount structure, copy the superblock into it. */ 1040 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 1041 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 1042 1043 /* Compatibility */ 1044 if (fs->lfs_version < 2) { 1045 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 1046 fs->lfs_ibsize = fs->lfs_bsize; 1047 fs->lfs_start = fs->lfs_sboffs[0]; 1048 fs->lfs_tstamp = fs->lfs_otstamp; 1049 fs->lfs_fsbtodb = 0; 1050 } 1051 1052 /* 1053 * If we aren't going to be able to write meaningfully to this 1054 * filesystem, and were not mounted readonly, bomb out now. 1055 */ 1056 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 1057 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 1058 " we need BUFPAGES >= %lld\n", 1059 (long long)((bufmem_hiwater / bufmem_lowater) * 1060 LFS_INVERSE_MAX_BYTES( 1061 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 1062 free(fs, M_UFSMNT); 1063 error = EFBIG; /* XXX needs translation */ 1064 goto out; 1065 } 1066 1067 /* Before rolling forward, lock so vget will sleep for other procs */ 1068 fs->lfs_flags = LFS_NOTYET; 1069 fs->lfs_rfpid = p->p_pid; 1070 1071 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1072 ump->um_lfs = fs; 1073 ump->um_ops = &lfs_ufsops; 1074 ump->um_fstype = UFS1; 1075 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 1076 bp->b_flags |= B_INVAL; 1077 abp->b_flags |= B_INVAL; 1078 } 1079 brelse(bp); 1080 bp = NULL; 1081 brelse(abp); 1082 abp = NULL; 1083 1084 /* Set up the I/O information */ 1085 fs->lfs_devbsize = secsize; 1086 fs->lfs_iocount = 0; 1087 fs->lfs_diropwait = 0; 1088 fs->lfs_activesb = 0; 1089 fs->lfs_uinodes = 0; 1090 fs->lfs_ravail = 0; 1091 fs->lfs_favail = 0; 1092 fs->lfs_sbactive = 0; 1093 1094 /* Set up the ifile and lock aflags */ 1095 fs->lfs_doifile = 0; 1096 fs->lfs_writer = 0; 1097 fs->lfs_dirops = 0; 1098 fs->lfs_nadirop = 0; 1099 fs->lfs_seglock = 0; 1100 fs->lfs_pdflush = 0; 1101 fs->lfs_sleepers = 0; 1102 fs->lfs_pages = 0; 1103 simple_lock_init(&fs->lfs_interlock); 1104 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0); 1105 lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0); 1106 1107 /* Set the file system readonly/modify bits. */ 1108 fs->lfs_ronly = ronly; 1109 if (ronly == 0) 1110 fs->lfs_fmod = 1; 1111 1112 /* Initialize the mount structure. */ 1113 dev = devvp->v_rdev; 1114 mp->mnt_data = ump; 1115 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1116 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1117 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1118 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1119 mp->mnt_stat.f_iosize = fs->lfs_bsize; 1120 mp->mnt_flag |= MNT_LOCAL; 1121 mp->mnt_fs_bshift = fs->lfs_bshift; 1122 ump->um_flags = 0; 1123 ump->um_mountp = mp; 1124 ump->um_dev = dev; 1125 ump->um_devvp = devvp; 1126 ump->um_bptrtodb = fs->lfs_fsbtodb; 1127 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 1128 ump->um_nindir = fs->lfs_nindir; 1129 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 1130 for (i = 0; i < MAXQUOTAS; i++) 1131 ump->um_quotas[i] = NULLVP; 1132 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 1133 ump->um_dirblksiz = DIRBLKSIZ; 1134 ump->um_maxfilesize = fs->lfs_maxfilesize; 1135 if (ump->um_maxsymlinklen > 0) 1136 mp->mnt_iflag |= IMNT_DTYPE; 1137 devvp->v_specmountpoint = mp; 1138 1139 /* Set up reserved memory for pageout */ 1140 lfs_setup_resblks(fs); 1141 /* Set up vdirop tailq */ 1142 TAILQ_INIT(&fs->lfs_dchainhd); 1143 /* and paging tailq */ 1144 TAILQ_INIT(&fs->lfs_pchainhd); 1145 1146 /* 1147 * We use the ifile vnode for almost every operation. Instead of 1148 * retrieving it from the hash table each time we retrieve it here, 1149 * artificially increment the reference count and keep a pointer 1150 * to it in the incore copy of the superblock. 1151 */ 1152 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 1153 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1154 goto out; 1155 } 1156 fs->lfs_ivnode = vp; 1157 VREF(vp); 1158 1159 /* Set up inode bitmap and order free list */ 1160 lfs_order_freelist(fs); 1161 1162 /* Set up segment usage flags for the autocleaner. */ 1163 fs->lfs_nactive = 0; 1164 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 1165 M_SEGMENT, M_WAITOK); 1166 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 1167 M_SEGMENT, M_WAITOK); 1168 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 1169 M_SEGMENT, M_WAITOK); 1170 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 1171 for (i = 0; i < fs->lfs_nseg; i++) { 1172 int changed; 1173 1174 LFS_SEGENTRY(sup, fs, i, bp); 1175 changed = 0; 1176 if (!ronly) { 1177 if (sup->su_nbytes == 0 && 1178 !(sup->su_flags & SEGUSE_EMPTY)) { 1179 sup->su_flags |= SEGUSE_EMPTY; 1180 ++changed; 1181 } else if (!(sup->su_nbytes == 0) && 1182 (sup->su_flags & SEGUSE_EMPTY)) { 1183 sup->su_flags &= ~SEGUSE_EMPTY; 1184 ++changed; 1185 } 1186 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1187 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1188 ++changed; 1189 } 1190 } 1191 fs->lfs_suflags[0][i] = sup->su_flags; 1192 if (changed) 1193 LFS_WRITESEGENTRY(sup, fs, i, bp); 1194 else 1195 brelse(bp); 1196 } 1197 1198 /* 1199 * Roll forward. 1200 * 1201 * We don't roll forward for v1 filesystems, because 1202 * of the danger that the clock was turned back between the last 1203 * checkpoint and crash. This would roll forward garbage. 1204 * 1205 * v2 filesystems don't have this problem because they use a 1206 * monotonically increasing serial number instead of a timestamp. 1207 */ 1208 do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) && 1209 lfs_do_rfw && fs->lfs_version > 1); 1210 if (do_rollforward) { 1211 u_int64_t nextserial; 1212 /* 1213 * Phase I: Find the address of the last good partial 1214 * segment that was written after the checkpoint. Mark 1215 * the segments in question dirty, so they won't be 1216 * reallocated. 1217 */ 1218 lastgoodpseg = oldoffset = offset = fs->lfs_offset; 1219 flags = 0x0; 1220 DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%" 1221 PRIx64 "\n", offset)); 1222 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 1223 if (!(sup->su_flags & SEGUSE_DIRTY)) 1224 --fs->lfs_nclean; 1225 sup->su_flags |= SEGUSE_DIRTY; 1226 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp); 1227 nextserial = fs->lfs_serial + 1; 1228 while ((offset = check_segsum(fs, offset, nextserial, 1229 cred, CHECK_CKSUM, &flags, l)) > 0) { 1230 nextserial++; 1231 if (sntod(fs, oldoffset) != sntod(fs, offset)) { 1232 LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset), 1233 bp); 1234 if (!(sup->su_flags & SEGUSE_DIRTY)) 1235 --fs->lfs_nclean; 1236 sup->su_flags |= SEGUSE_DIRTY; 1237 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset), 1238 bp); 1239 } 1240 1241 DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%" 1242 PRIx64 "\n", offset)); 1243 if (flags & SS_DIROP) { 1244 DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%" 1245 PRIx64 "\n", oldoffset)); 1246 if (!(flags & SS_CONT)) 1247 DLOG((DLOG_RF, "lfs_mountfs: dirops end " 1248 "at 0x%" PRIx64 "\n", oldoffset)); 1249 } 1250 if (!(flags & SS_CONT)) 1251 lastgoodpseg = offset; 1252 oldoffset = offset; 1253 } 1254 if (flags & SS_CONT) { 1255 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 1256 "dirops discarded\n")); 1257 } 1258 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 1259 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 1260 oldoffset = fs->lfs_offset; 1261 if (fs->lfs_offset != lastgoodpseg) { 1262 /* Don't overwrite what we're trying to preserve */ 1263 offset = fs->lfs_offset; 1264 fs->lfs_offset = lastgoodpseg; 1265 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset)); 1266 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) { 1267 sn = (sn + 1) % fs->lfs_nseg; 1268 if (sn == curseg) 1269 panic("lfs_mountfs: no clean segments"); 1270 LFS_SEGENTRY(sup, fs, sn, bp); 1271 dirty = (sup->su_flags & SEGUSE_DIRTY); 1272 brelse(bp); 1273 if (!dirty) 1274 break; 1275 } 1276 fs->lfs_nextseg = sntod(fs, sn); 1277 1278 /* 1279 * Phase II: Roll forward from the first superblock. 1280 */ 1281 while (offset != lastgoodpseg) { 1282 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 1283 PRIx64 "\n", offset)); 1284 offset = check_segsum(fs, offset, 1285 fs->lfs_serial + 1, cred, CHECK_UPDATE, 1286 NULL, l); 1287 } 1288 1289 /* 1290 * Finish: flush our changes to disk. 1291 */ 1292 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1293 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 1294 "recovered %lld blocks\n", 1295 (long long)(lastgoodpseg - oldoffset))); 1296 } 1297 DLOG((DLOG_RF, "LFS roll forward complete\n")); 1298 } 1299 /* If writing, sb is not clean; record in case of immediate crash */ 1300 if (!fs->lfs_ronly) { 1301 fs->lfs_pflags &= ~LFS_PF_CLEAN; 1302 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1303 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1304 } 1305 1306 /* Allow vget now that roll-forward is complete */ 1307 fs->lfs_flags &= ~(LFS_NOTYET); 1308 wakeup(&fs->lfs_flags); 1309 1310 /* 1311 * Initialize the ifile cleaner info with information from 1312 * the superblock. 1313 */ 1314 LFS_CLEANERINFO(cip, fs, bp); 1315 cip->clean = fs->lfs_nclean; 1316 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 1317 cip->avail = fs->lfs_avail; 1318 cip->bfree = fs->lfs_bfree; 1319 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1320 1321 /* 1322 * Mark the current segment as ACTIVE, since we're going to 1323 * be writing to it. 1324 */ 1325 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 1326 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1327 fs->lfs_nactive++; 1328 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 1329 1330 /* Now that roll-forward is done, unlock the Ifile */ 1331 vput(vp); 1332 1333 /* Comment on ifile size if it is too large */ 1334 warn_ifile_size(fs); 1335 1336 /* Start the pagedaemon-anticipating daemon */ 1337 if (lfs_writer_daemon == 0 && 1338 kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1339 panic("fork lfs_writer"); 1340 1341 return (0); 1342 1343 out: 1344 if (bp) 1345 brelse(bp); 1346 if (abp) 1347 brelse(abp); 1348 if (ump) { 1349 free(ump->um_lfs, M_UFSMNT); 1350 free(ump, M_UFSMNT); 1351 mp->mnt_data = NULL; 1352 } 1353 1354 return (error); 1355 } 1356 1357 /* 1358 * unmount system call 1359 */ 1360 int 1361 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l) 1362 { 1363 struct ufsmount *ump; 1364 struct lfs *fs; 1365 int error, flags, ronly; 1366 int s; 1367 1368 flags = 0; 1369 if (mntflags & MNT_FORCE) 1370 flags |= FORCECLOSE; 1371 1372 ump = VFSTOUFS(mp); 1373 fs = ump->um_lfs; 1374 1375 /* Two checkpoints */ 1376 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1377 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1378 1379 /* wake up the cleaner so it can die */ 1380 wakeup(&fs->lfs_nextseg); 1381 wakeup(&lfs_allclean_wakeup); 1382 simple_lock(&fs->lfs_interlock); 1383 while (fs->lfs_sleepers) 1384 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 1385 &fs->lfs_interlock); 1386 simple_unlock(&fs->lfs_interlock); 1387 1388 #ifdef QUOTA 1389 if (mp->mnt_flag & MNT_QUOTA) { 1390 int i; 1391 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 1392 if (error) 1393 return (error); 1394 for (i = 0; i < MAXQUOTAS; i++) { 1395 if (ump->um_quotas[i] == NULLVP) 1396 continue; 1397 quotaoff(l, mp, i); 1398 } 1399 /* 1400 * Here we fall through to vflush again to ensure 1401 * that we have gotten rid of all the system vnodes. 1402 */ 1403 } 1404 #endif 1405 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1406 return (error); 1407 if ((error = VFS_SYNC(mp, 1, l->l_proc->p_ucred, l)) != 0) 1408 return (error); 1409 s = splbio(); 1410 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 1411 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1412 splx(s); 1413 1414 /* Comment on ifile size if it has become too large */ 1415 if (!(fs->lfs_flags & LFS_WARNED)) 1416 warn_ifile_size(fs); 1417 1418 /* Explicitly write the superblock, to update serial and pflags */ 1419 fs->lfs_pflags |= LFS_PF_CLEAN; 1420 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1421 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1422 simple_lock(&fs->lfs_interlock); 1423 while (fs->lfs_iocount) 1424 ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1425 &fs->lfs_interlock); 1426 simple_unlock(&fs->lfs_interlock); 1427 1428 /* Finish with the Ifile, now that we're done with it */ 1429 vrele(fs->lfs_ivnode); 1430 vgone(fs->lfs_ivnode); 1431 1432 ronly = !fs->lfs_ronly; 1433 if (ump->um_devvp->v_type != VBAD) 1434 ump->um_devvp->v_specmountpoint = NULL; 1435 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1436 error = VOP_CLOSE(ump->um_devvp, 1437 ronly ? FREAD : FREAD|FWRITE, NOCRED, l); 1438 vput(ump->um_devvp); 1439 1440 /* Complain about page leakage */ 1441 if (fs->lfs_pages > 0) 1442 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1443 fs->lfs_pages, lfs_subsys_pages); 1444 1445 /* Free per-mount data structures */ 1446 free(fs->lfs_suflags[0], M_SEGMENT); 1447 free(fs->lfs_suflags[1], M_SEGMENT); 1448 free(fs->lfs_suflags, M_SEGMENT); 1449 lfs_free_resblks(fs); 1450 free(fs, M_UFSMNT); 1451 free(ump, M_UFSMNT); 1452 1453 mp->mnt_data = NULL; 1454 mp->mnt_flag &= ~MNT_LOCAL; 1455 return (error); 1456 } 1457 1458 /* 1459 * Get file system statistics. 1460 * 1461 * NB: We don't lock to access the superblock here, because it's not 1462 * really that important if we get it wrong. 1463 */ 1464 int 1465 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l) 1466 { 1467 struct lfs *fs; 1468 struct ufsmount *ump; 1469 1470 ump = VFSTOUFS(mp); 1471 fs = ump->um_lfs; 1472 if (fs->lfs_magic != LFS_MAGIC) 1473 panic("lfs_statvfs: magic"); 1474 1475 sbp->f_bsize = fs->lfs_bsize; 1476 sbp->f_frsize = fs->lfs_fsize; 1477 sbp->f_iosize = fs->lfs_bsize; 1478 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 1479 1480 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1481 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1482 if (sbp->f_bfree < 0) 1483 sbp->f_bfree = 0; 1484 1485 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1486 if (sbp->f_bfree > sbp->f_bresvd) 1487 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1488 else 1489 sbp->f_bavail = 0; 1490 1491 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1492 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1493 sbp->f_favail = sbp->f_ffree; 1494 sbp->f_fresvd = 0; 1495 copy_statvfs_info(sbp, mp); 1496 return (0); 1497 } 1498 1499 /* 1500 * Go through the disk queues to initiate sandbagged IO; 1501 * go through the inodes to write those that have been modified; 1502 * initiate the writing of the super block if it has been modified. 1503 * 1504 * Note: we are always called with the filesystem marked `MPBUSY'. 1505 */ 1506 int 1507 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct lwp *l) 1508 { 1509 int error; 1510 struct lfs *fs; 1511 1512 fs = VFSTOUFS(mp)->um_lfs; 1513 if (fs->lfs_ronly) 1514 return 0; 1515 lfs_writer_enter(fs, "lfs_dirops"); 1516 1517 /* All syncs must be checkpoints until roll-forward is implemented. */ 1518 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1519 lfs_writer_leave(fs); 1520 #ifdef QUOTA 1521 qsync(mp); 1522 #endif 1523 return (error); 1524 } 1525 1526 extern struct lock ufs_hashlock; 1527 1528 /* 1529 * Look up an LFS dinode number to find its incore vnode. If not already 1530 * in core, read it in from the specified device. Return the inode locked. 1531 * Detection and handling of mount points must be done by the calling routine. 1532 */ 1533 int 1534 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1535 { 1536 struct lfs *fs; 1537 struct ufs1_dinode *dip; 1538 struct inode *ip; 1539 struct buf *bp; 1540 struct ifile *ifp; 1541 struct vnode *vp; 1542 struct ufsmount *ump; 1543 daddr_t daddr; 1544 dev_t dev; 1545 int error, retries; 1546 struct timespec ts; 1547 1548 ump = VFSTOUFS(mp); 1549 dev = ump->um_dev; 1550 fs = ump->um_lfs; 1551 1552 /* 1553 * If the filesystem is not completely mounted yet, suspend 1554 * any access requests (wait for roll-forward to complete). 1555 */ 1556 simple_lock(&fs->lfs_interlock); 1557 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1558 ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1559 &fs->lfs_interlock); 1560 simple_unlock(&fs->lfs_interlock); 1561 1562 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1563 return (0); 1564 1565 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1566 *vpp = NULL; 1567 return (error); 1568 } 1569 1570 do { 1571 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1572 ungetnewvnode(vp); 1573 return (0); 1574 } 1575 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0)); 1576 1577 /* Translate the inode number to a disk address. */ 1578 if (ino == LFS_IFILE_INUM) 1579 daddr = fs->lfs_idaddr; 1580 else { 1581 /* XXX bounds-check this too */ 1582 LFS_IENTRY(ifp, fs, ino, bp); 1583 daddr = ifp->if_daddr; 1584 if (fs->lfs_version > 1) { 1585 ts.tv_sec = ifp->if_atime_sec; 1586 ts.tv_nsec = ifp->if_atime_nsec; 1587 } 1588 1589 brelse(bp); 1590 if (daddr == LFS_UNUSED_DADDR) { 1591 *vpp = NULLVP; 1592 ungetnewvnode(vp); 1593 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1594 return (ENOENT); 1595 } 1596 } 1597 1598 /* Allocate/init new vnode/inode. */ 1599 lfs_vcreate(mp, ino, vp); 1600 1601 /* 1602 * Put it onto its hash chain and lock it so that other requests for 1603 * this inode will block if they arrive while we are sleeping waiting 1604 * for old data structures to be purged or for the contents of the 1605 * disk portion of this inode to be read. 1606 */ 1607 ip = VTOI(vp); 1608 ufs_ihashins(ip); 1609 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1610 1611 /* 1612 * XXX 1613 * This may not need to be here, logically it should go down with 1614 * the i_devvp initialization. 1615 * Ask Kirk. 1616 */ 1617 ip->i_lfs = ump->um_lfs; 1618 1619 /* Read in the disk contents for the inode, copy into the inode. */ 1620 retries = 0; 1621 again: 1622 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1623 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1624 NOCRED, &bp); 1625 if (error) { 1626 /* 1627 * The inode does not contain anything useful, so it would 1628 * be misleading to leave it on its hash chain. With mode 1629 * still zero, it will be unlinked and returned to the free 1630 * list by vput(). 1631 */ 1632 vput(vp); 1633 brelse(bp); 1634 *vpp = NULL; 1635 return (error); 1636 } 1637 1638 dip = lfs_ifind(fs, ino, bp); 1639 if (dip == NULL) { 1640 /* Assume write has not completed yet; try again */ 1641 bp->b_flags |= B_INVAL; 1642 brelse(bp); 1643 ++retries; 1644 if (retries > LFS_IFIND_RETRIES) { 1645 #ifdef DEBUG 1646 /* If the seglock is held look at the bpp to see 1647 what is there anyway */ 1648 simple_lock(&fs->lfs_interlock); 1649 if (fs->lfs_seglock > 0) { 1650 struct buf **bpp; 1651 struct ufs1_dinode *dp; 1652 int i; 1653 1654 for (bpp = fs->lfs_sp->bpp; 1655 bpp != fs->lfs_sp->cbpp; ++bpp) { 1656 if ((*bpp)->b_vp == fs->lfs_ivnode && 1657 bpp != fs->lfs_sp->bpp) { 1658 /* Inode block */ 1659 printf("lfs_vget: block 0x%" PRIx64 ": ", 1660 (*bpp)->b_blkno); 1661 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1662 for (i = 0; i < INOPB(fs); i++) 1663 if (dp[i].di_u.inumber) 1664 printf("%d ", dp[i].di_u.inumber); 1665 printf("\n"); 1666 } 1667 } 1668 } 1669 simple_unlock(&fs->lfs_interlock); 1670 #endif /* DEBUG */ 1671 panic("lfs_vget: dinode not found"); 1672 } 1673 simple_lock(&fs->lfs_interlock); 1674 if (fs->lfs_iocount) { 1675 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1676 (void)ltsleep(&fs->lfs_iocount, PRIBIO + 1, 1677 "lfs ifind", 1, &fs->lfs_interlock); 1678 } else 1679 retries = LFS_IFIND_RETRIES; 1680 simple_unlock(&fs->lfs_interlock); 1681 goto again; 1682 } 1683 *ip->i_din.ffs1_din = *dip; 1684 brelse(bp); 1685 1686 if (fs->lfs_version > 1) { 1687 ip->i_ffs1_atime = ts.tv_sec; 1688 ip->i_ffs1_atimensec = ts.tv_nsec; 1689 } 1690 1691 lfs_vinit(mp, &vp); 1692 1693 *vpp = vp; 1694 1695 KASSERT(VOP_ISLOCKED(vp)); 1696 1697 return (0); 1698 } 1699 1700 /* 1701 * File handle to vnode 1702 */ 1703 int 1704 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1705 { 1706 struct lfid *lfhp; 1707 struct buf *bp; 1708 IFILE *ifp; 1709 int32_t daddr; 1710 struct lfs *fs; 1711 1712 lfhp = (struct lfid *)fhp; 1713 if (lfhp->lfid_ino < LFS_IFILE_INUM) 1714 return ESTALE; 1715 1716 fs = VFSTOUFS(mp)->um_lfs; 1717 if (lfhp->lfid_ident != fs->lfs_ident) 1718 return ESTALE; 1719 1720 if (lfhp->lfid_ino > 1721 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1722 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1723 return ESTALE; 1724 1725 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) { 1726 LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp); 1727 daddr = ifp->if_daddr; 1728 brelse(bp); 1729 if (daddr == LFS_UNUSED_DADDR) 1730 return ESTALE; 1731 } 1732 1733 return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp)); 1734 } 1735 1736 /* 1737 * Vnode pointer to File handle 1738 */ 1739 /* ARGSUSED */ 1740 int 1741 lfs_vptofh(struct vnode *vp, struct fid *fhp) 1742 { 1743 struct inode *ip; 1744 struct lfid *lfhp; 1745 1746 ip = VTOI(vp); 1747 lfhp = (struct lfid *)fhp; 1748 lfhp->lfid_len = sizeof(struct lfid); 1749 lfhp->lfid_ino = ip->i_number; 1750 lfhp->lfid_gen = ip->i_gen; 1751 lfhp->lfid_ident = ip->i_lfs->lfs_ident; 1752 return (0); 1753 } 1754 1755 static int 1756 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1757 { 1758 extern struct lfs_stats lfs_stats; 1759 extern int lfs_dostats; 1760 int error; 1761 1762 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1763 if (error || newp == NULL) 1764 return (error); 1765 1766 if (lfs_dostats == 0) 1767 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1768 1769 return (0); 1770 } 1771 1772 struct shortlong { 1773 const char *sname; 1774 const char *lname; 1775 }; 1776 1777 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1778 { 1779 int i; 1780 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1781 lfs_fs_pagetrip; 1782 #ifdef DEBUG 1783 extern int lfs_debug_log_subsys[DLOG_MAX]; 1784 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1785 { "rollforward", "Debug roll-forward code" }, 1786 { "alloc", "Debug inode allocation and free list" }, 1787 { "avail", "Debug space-available-now accounting" }, 1788 { "flush", "Debug flush triggers" }, 1789 { "lockedlist", "Debug locked list accounting" }, 1790 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1791 { "vnode", "Debug vnode use during segment write" }, 1792 { "segment", "Debug segment writing" }, 1793 { "seguse", "Debug segment used-bytes accounting" }, 1794 { "cleaner", "Debug cleaning routines" }, 1795 { "mount", "Debug mount/unmount routines" }, 1796 { "pagecache", "Debug UBC interactions" }, 1797 { "dirop", "Debug directory-operation accounting" }, 1798 { "malloc", "Debug private malloc accounting" }, 1799 }; 1800 #endif /* DEBUG */ 1801 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1802 { "segsused", "Number of new segments allocated" }, 1803 { "psegwrites", "Number of partial-segment writes" }, 1804 { "psyncwrites", "Number of synchronous partial-segment" 1805 " writes" }, 1806 { "pcleanwrites", "Number of partial-segment writes by the" 1807 " cleaner" }, 1808 { "blocktot", "Number of blocks written" }, 1809 { "cleanblocks", "Number of blocks written by the cleaner" }, 1810 { "ncheckpoints", "Number of checkpoints made" }, 1811 { "nwrites", "Number of whole writes" }, 1812 { "nsync_writes", "Number of synchronous writes" }, 1813 { "wait_exceeded", "Number of times writer waited for" 1814 " cleaner" }, 1815 { "write_exceeded", "Number of times writer invoked flush" }, 1816 { "flush_invoked", "Number of times flush was invoked" }, 1817 { "vflush_invoked", "Number of time vflush was called" }, 1818 { "clean_inlocked", "Number of vnodes skipped for VXLOCK" }, 1819 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1820 { "segs_reclaimed", "Number of segments reclaimed" }, 1821 }; 1822 1823 sysctl_createv(clog, 0, NULL, NULL, 1824 CTLFLAG_PERMANENT, 1825 CTLTYPE_NODE, "vfs", NULL, 1826 NULL, 0, NULL, 0, 1827 CTL_VFS, CTL_EOL); 1828 sysctl_createv(clog, 0, NULL, NULL, 1829 CTLFLAG_PERMANENT, 1830 CTLTYPE_NODE, "lfs", 1831 SYSCTL_DESCR("Log-structured file system"), 1832 NULL, 0, NULL, 0, 1833 CTL_VFS, 5, CTL_EOL); 1834 /* 1835 * XXX the "5" above could be dynamic, thereby eliminating one 1836 * more instance of the "number to vfs" mapping problem, but 1837 * "5" is the order as taken from sys/mount.h 1838 */ 1839 1840 sysctl_createv(clog, 0, NULL, NULL, 1841 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1842 CTLTYPE_INT, "flushindir", NULL, 1843 NULL, 0, &lfs_writeindir, 0, 1844 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1845 sysctl_createv(clog, 0, NULL, NULL, 1846 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1847 CTLTYPE_INT, "clean_vnhead", NULL, 1848 NULL, 0, &lfs_clean_vnhead, 0, 1849 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1850 sysctl_createv(clog, 0, NULL, NULL, 1851 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1852 CTLTYPE_INT, "dostats", 1853 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1854 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1855 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1856 sysctl_createv(clog, 0, NULL, NULL, 1857 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1858 CTLTYPE_INT, "pagetrip", 1859 SYSCTL_DESCR("How many dirty pages in fs triggers" 1860 " a flush"), 1861 NULL, 0, &lfs_fs_pagetrip, 0, 1862 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1863 sysctl_createv(clog, 0, NULL, NULL, 1864 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1865 CTLTYPE_INT, "rfw", 1866 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1867 NULL, 0, &lfs_do_rfw, 0, 1868 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1869 1870 sysctl_createv(clog, 0, NULL, NULL, 1871 CTLFLAG_PERMANENT, 1872 CTLTYPE_NODE, "stats", 1873 SYSCTL_DESCR("Debugging options"), 1874 NULL, 0, NULL, 0, 1875 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1876 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1877 sysctl_createv(clog, 0, NULL, NULL, 1878 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1879 CTLTYPE_INT, stat_names[i].sname, 1880 SYSCTL_DESCR(stat_names[i].lname), 1881 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1882 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1883 } 1884 1885 #ifdef DEBUG 1886 sysctl_createv(clog, 0, NULL, NULL, 1887 CTLFLAG_PERMANENT, 1888 CTLTYPE_NODE, "debug", 1889 SYSCTL_DESCR("Debugging options"), 1890 NULL, 0, NULL, 0, 1891 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1892 for (i = 0; i < DLOG_MAX; i++) { 1893 sysctl_createv(clog, 0, NULL, NULL, 1894 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1895 CTLTYPE_INT, dlog_names[i].sname, 1896 SYSCTL_DESCR(dlog_names[i].lname), 1897 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1898 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1899 } 1900 #endif 1901 } 1902 1903 /* 1904 * ufs_bmaparray callback function for writing. 1905 * 1906 * Since blocks will be written to the new segment anyway, 1907 * we don't care about current daddr of them. 1908 */ 1909 static boolean_t 1910 lfs_issequential_hole(const struct ufsmount *ump, 1911 daddr_t daddr0, daddr_t daddr1) 1912 { 1913 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1914 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1915 1916 KASSERT(daddr0 == UNWRITTEN || 1917 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1918 KASSERT(daddr1 == UNWRITTEN || 1919 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1920 1921 /* NOTE: all we want to know here is 'hole or not'. */ 1922 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1923 1924 /* 1925 * treat UNWRITTENs and all resident blocks as 'contiguous' 1926 */ 1927 if (daddr0 != 0 && daddr1 != 0) 1928 return TRUE; 1929 1930 /* 1931 * both are in hole? 1932 */ 1933 if (daddr0 == 0 && daddr1 == 0) 1934 return TRUE; /* all holes are 'contiguous' for us. */ 1935 1936 return FALSE; 1937 } 1938 1939 /* 1940 * lfs_gop_write functions exactly like genfs_gop_write, except that 1941 * (1) it requires the seglock to be held by its caller, and sp->fip 1942 * to be properly initialized (it will return without re-initializing 1943 * sp->fip, and without calling lfs_writeseg). 1944 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1945 * to determine how large a block it can write at once (though it does 1946 * still use VOP_BMAP to find holes in the file); 1947 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1948 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1949 * now have clusters of clusters, ick.) 1950 */ 1951 static int 1952 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags) 1953 { 1954 int i, s, error, run; 1955 int fs_bshift; 1956 vaddr_t kva; 1957 off_t eof, offset, startoffset = 0; 1958 size_t bytes, iobytes, skipbytes; 1959 daddr_t lbn, blkno; 1960 struct vm_page *pg; 1961 struct buf *mbp, *bp; 1962 struct vnode *devvp = VTOI(vp)->i_devvp; 1963 struct inode *ip = VTOI(vp); 1964 struct lfs *fs = ip->i_lfs; 1965 struct segment *sp = fs->lfs_sp; 1966 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1967 1968 ASSERT_SEGLOCK(fs); 1969 1970 /* The Ifile lives in the buffer cache */ 1971 KASSERT(vp != fs->lfs_ivnode); 1972 1973 /* 1974 * We don't want to fill the disk before the cleaner has a chance 1975 * to make room for us. If we're in danger of doing that, fail 1976 * with EAGAIN. The caller will have to notice this, unlock 1977 * so the cleaner can run, relock and try again. 1978 */ 1979 if (LFS_STARVED_FOR_SEGS(fs)) 1980 goto tryagain; 1981 1982 /* 1983 * Sometimes things slip past the filters in lfs_putpages, 1984 * and the pagedaemon tries to write pages---problem is 1985 * that the pagedaemon never acquires the segment lock. 1986 * 1987 * Alternatively, pages that were clean when we called 1988 * genfs_putpages may have become dirty in the meantime. In this 1989 * case the segment header is not properly set up for blocks 1990 * to be added to it. 1991 * 1992 * Unbusy and unclean the pages, and put them on the ACTIVE 1993 * queue under the hypothesis that they couldn't have got here 1994 * unless they were modified *quite* recently. 1995 * 1996 * XXXUBC that last statement is an oversimplification of course. 1997 */ 1998 if (!LFS_SEGLOCK_HELD(fs) || 1999 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 2000 (pgs[0]->offset & fs->lfs_bmask) != 0) { 2001 goto tryagain; 2002 } 2003 2004 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 2005 vp, pgs, npages, flags); 2006 2007 GOP_SIZE(vp, vp->v_size, &eof, 0); 2008 2009 if (vp->v_type == VREG) 2010 fs_bshift = vp->v_mount->mnt_fs_bshift; 2011 else 2012 fs_bshift = DEV_BSHIFT; 2013 error = 0; 2014 pg = pgs[0]; 2015 startoffset = pg->offset; 2016 KASSERT(eof >= 0); 2017 if (startoffset >= eof) { 2018 goto tryagain; 2019 } else 2020 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 2021 skipbytes = 0; 2022 2023 KASSERT(bytes != 0); 2024 2025 /* Swap PG_DELWRI for PG_PAGEOUT */ 2026 for (i = 0; i < npages; i++) 2027 if (pgs[i]->flags & PG_DELWRI) { 2028 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 2029 pgs[i]->flags &= ~PG_DELWRI; 2030 pgs[i]->flags |= PG_PAGEOUT; 2031 uvmexp.paging++; 2032 uvm_lock_pageq(); 2033 uvm_pageunwire(pgs[i]); 2034 uvm_unlock_pageq(); 2035 } 2036 2037 /* 2038 * Check to make sure we're starting on a block boundary. 2039 * We'll check later to make sure we always write entire 2040 * blocks (or fragments). 2041 */ 2042 if (startoffset & fs->lfs_bmask) 2043 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 2044 startoffset, fs->lfs_bmask, 2045 startoffset & fs->lfs_bmask); 2046 KASSERT((startoffset & fs->lfs_bmask) == 0); 2047 if (bytes & fs->lfs_ffmask) { 2048 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 2049 panic("lfs_gop_write: non-integer blocks"); 2050 } 2051 2052 /* 2053 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 2054 * If we would, write what we have and try again. If we don't 2055 * have anything to write, we'll have to sleep. 2056 */ 2057 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2058 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 2059 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 2060 int vers; 2061 2062 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 2063 #if 0 2064 " with nfinfo=%d at offset 0x%x\n", 2065 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 2066 (unsigned)fs->lfs_offset)); 2067 #endif 2068 if (sp->fip->fi_nblocks == 0) { 2069 /* Don't write zero-length finfos */ 2070 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2071 sp->sum_bytes_left += FINFOSIZE; 2072 } else 2073 lfs_updatemeta(sp); 2074 2075 vers = sp->fip->fi_version; 2076 (void) lfs_writeseg(fs, sp); 2077 2078 sp->fip->fi_version = vers; 2079 sp->fip->fi_ino = ip->i_number; 2080 /* Add the current file to the segment summary. */ 2081 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2082 sp->sum_bytes_left -= FINFOSIZE; 2083 2084 /* 2085 * Having given up all of the pager_map we were holding, 2086 * we can now wait for aiodoned to reclaim it for us 2087 * without fear of deadlock. 2088 */ 2089 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2090 UVMPAGER_MAPIN_WAITOK); 2091 } 2092 2093 s = splbio(); 2094 simple_lock(&global_v_numoutput_slock); 2095 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 2096 simple_unlock(&global_v_numoutput_slock); 2097 splx(s); 2098 2099 mbp = getiobuf(); 2100 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 2101 vp, mbp, vp->v_numoutput, bytes); 2102 mbp->b_bufsize = npages << PAGE_SHIFT; 2103 mbp->b_data = (void *)kva; 2104 mbp->b_resid = mbp->b_bcount = bytes; 2105 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL; 2106 mbp->b_iodone = uvm_aio_biodone; 2107 mbp->b_vp = vp; 2108 2109 bp = NULL; 2110 for (offset = startoffset; 2111 bytes > 0; 2112 offset += iobytes, bytes -= iobytes) { 2113 lbn = offset >> fs_bshift; 2114 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2115 lfs_issequential_hole); 2116 if (error) { 2117 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 2118 error,0,0,0); 2119 skipbytes += bytes; 2120 bytes = 0; 2121 break; 2122 } 2123 2124 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2125 bytes); 2126 if (blkno == (daddr_t)-1) { 2127 skipbytes += iobytes; 2128 continue; 2129 } 2130 2131 /* 2132 * Discover how much we can really pack into this buffer. 2133 */ 2134 /* If no room in the current segment, finish it up */ 2135 if (sp->sum_bytes_left < sizeof(int32_t) || 2136 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 2137 int vers; 2138 2139 lfs_updatemeta(sp); 2140 2141 vers = sp->fip->fi_version; 2142 (void) lfs_writeseg(fs, sp); 2143 2144 sp->fip->fi_version = vers; 2145 sp->fip->fi_ino = ip->i_number; 2146 /* Add the current file to the segment summary. */ 2147 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2148 sp->sum_bytes_left -= FINFOSIZE; 2149 } 2150 /* Check both for space in segment and space in segsum */ 2151 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2152 << fs_bshift); 2153 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2154 << fs_bshift); 2155 KASSERT(iobytes > 0); 2156 2157 /* if it's really one i/o, don't make a second buf */ 2158 if (offset == startoffset && iobytes == bytes) { 2159 bp = mbp; 2160 /* correct overcount if there is no second buffer */ 2161 s = splbio(); 2162 simple_lock(&global_v_numoutput_slock); 2163 --vp->v_numoutput; 2164 simple_unlock(&global_v_numoutput_slock); 2165 splx(s); 2166 } else { 2167 bp = getiobuf(); 2168 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 2169 vp, bp, vp->v_numoutput, 0); 2170 bp->b_data = (char *)kva + 2171 (vaddr_t)(offset - pg->offset); 2172 bp->b_resid = bp->b_bcount = iobytes; 2173 bp->b_flags = B_BUSY|B_WRITE|B_CALL; 2174 bp->b_iodone = uvm_aio_biodone1; 2175 } 2176 2177 /* XXX This is silly ... is this necessary? */ 2178 bp->b_vp = NULL; 2179 s = splbio(); 2180 bgetvp(vp, bp); 2181 splx(s); 2182 2183 bp->b_lblkno = lblkno(fs, offset); 2184 bp->b_private = mbp; 2185 if (devvp->v_type == VBLK) { 2186 bp->b_dev = devvp->v_rdev; 2187 } 2188 VOP_BWRITE(bp); 2189 while (lfs_gatherblock(sp, bp, NULL)) 2190 continue; 2191 } 2192 2193 if (skipbytes) { 2194 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 2195 s = splbio(); 2196 if (error) { 2197 mbp->b_flags |= B_ERROR; 2198 mbp->b_error = error; 2199 } 2200 mbp->b_resid -= skipbytes; 2201 if (mbp->b_resid == 0) { 2202 biodone(mbp); 2203 } 2204 splx(s); 2205 } 2206 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2207 return (0); 2208 2209 tryagain: 2210 /* 2211 * We can't write the pages, for whatever reason. 2212 * Clean up after ourselves, and make the caller try again. 2213 */ 2214 simple_lock(&vp->v_interlock); 2215 2216 /* Tell why we're here, if we know */ 2217 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) 2218 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 2219 else if ((pgs[0]->offset & fs->lfs_bmask) != 0) 2220 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 2221 else if (startoffset >= eof) 2222 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2223 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2224 pgs[0]->offset, eof, npages)); 2225 else if (LFS_STARVED_FOR_SEGS(fs)) 2226 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 2227 else 2228 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 2229 2230 uvm_lock_pageq(); 2231 for (i = 0; i < npages; i++) { 2232 pg = pgs[i]; 2233 2234 if (pg->flags & PG_PAGEOUT) 2235 uvmexp.paging--; 2236 if (pg->flags & PG_DELWRI) { 2237 uvm_pageunwire(pg); 2238 } 2239 uvm_pageactivate(pg); 2240 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2241 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2242 vp, pg->offset)); 2243 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2244 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2245 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2246 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2247 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2248 pg->wire_count)); 2249 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2250 pg->loan_count)); 2251 } 2252 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 2253 uvm_page_unbusy(pgs, npages); 2254 uvm_unlock_pageq(); 2255 simple_unlock(&vp->v_interlock); 2256 return EAGAIN; 2257 } 2258 2259 /* 2260 * finish vnode/inode initialization. 2261 * used by lfs_vget and lfs_fastvget. 2262 */ 2263 void 2264 lfs_vinit(struct mount *mp, struct vnode **vpp) 2265 { 2266 struct vnode *vp = *vpp; 2267 struct inode *ip = VTOI(vp); 2268 struct ufsmount *ump = VFSTOUFS(mp); 2269 int i; 2270 2271 ip->i_mode = ip->i_ffs1_mode; 2272 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 2273 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 2274 ip->i_flags = ip->i_ffs1_flags; 2275 ip->i_gen = ip->i_ffs1_gen; 2276 ip->i_uid = ip->i_ffs1_uid; 2277 ip->i_gid = ip->i_ffs1_gid; 2278 2279 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 2280 2281 /* 2282 * Initialize the vnode from the inode, check for aliases. In all 2283 * cases re-init ip, the underlying vnode/inode may have changed. 2284 */ 2285 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2286 ip = VTOI(vp); 2287 2288 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 2289 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 2290 struct lfs *fs = ump->um_lfs; 2291 #ifdef DEBUG 2292 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 2293 i < NDADDR; i++) { 2294 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2295 i == 0) 2296 continue; 2297 if (ip->i_ffs1_db[i] != 0) { 2298 inconsistent: 2299 lfs_dump_dinode(ip->i_din.ffs1_din); 2300 panic("inconsistent inode"); 2301 } 2302 } 2303 for ( ; i < NDADDR + NIADDR; i++) { 2304 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 2305 goto inconsistent; 2306 } 2307 } 2308 #endif /* DEBUG */ 2309 for (i = 0; i < NDADDR; i++) 2310 if (ip->i_ffs1_db[i] != 0) 2311 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 2312 } 2313 2314 #ifdef DIAGNOSTIC 2315 if (vp->v_type == VNON) { 2316 # ifdef DEBUG 2317 lfs_dump_dinode(ip->i_din.ffs1_din); 2318 # endif 2319 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2320 (unsigned long long)ip->i_number, 2321 (ip->i_mode & IFMT) >> 12); 2322 } 2323 #endif /* DIAGNOSTIC */ 2324 2325 /* 2326 * Finish inode initialization now that aliasing has been resolved. 2327 */ 2328 2329 ip->i_devvp = ump->um_devvp; 2330 VREF(ip->i_devvp); 2331 genfs_node_init(vp, &lfs_genfsops); 2332 uvm_vnp_setsize(vp, ip->i_size); 2333 2334 /* Initialize hiblk from file size */ 2335 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 2336 2337 *vpp = vp; 2338 } 2339 2340 /* 2341 * Warn if the inode portion of the Ifile is too large to be contained 2342 * in the buffer cache, according to LFS_MAX_BUFS / LFS_MAX_BYTES. 2343 * XXX the estimates don't take multiple LFSs into account. 2344 */ 2345 static void 2346 warn_ifile_size(struct lfs *fs) 2347 { 2348 KASSERT(LFS_MAX_BUFS > 0); 2349 KASSERT(LFS_MAX_BYTES > 0); 2350 if (((fs->lfs_ivnode->v_size >> fs->lfs_bshift) - fs->lfs_segtabsz) > 2351 LFS_MAX_BUFS) { 2352 simple_lock(&fs->lfs_interlock); 2353 fs->lfs_flags |= LFS_WARNED; 2354 simple_unlock(&fs->lfs_interlock); 2355 log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %" 2356 PRId64 " cannot fit in %d buffers\n", 2357 fs->lfs_ivnode->v_size - 2358 (fs->lfs_segtabsz << fs->lfs_bshift), 2359 LFS_MAX_BUFS); 2360 log(LOG_WARNING, "lfs_mountfs: please consider increasing NBUF" 2361 " to at least %" PRId64 "\n", 2362 LFS_INVERSE_MAX_BUFS((fs->lfs_ivnode->v_size >> 2363 fs->lfs_bshift) - 2364 fs->lfs_segtabsz)); 2365 } else if ((fs->lfs_ivnode->v_size >> fs->lfs_bshift) > LFS_MAX_BUFS) { 2366 /* Same thing but LOG_NOTICE */ 2367 simple_lock(&fs->lfs_interlock); 2368 fs->lfs_flags |= LFS_WARNED; 2369 simple_unlock(&fs->lfs_interlock); 2370 log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %" 2371 PRId64 " cannot fit in %d buffers\n", 2372 fs->lfs_ivnode->v_size, LFS_MAX_BUFS); 2373 log(LOG_NOTICE, "lfs_mountfs: please consider increasing NBUF" 2374 " to at least %" PRId64 "\n", 2375 LFS_INVERSE_MAX_BUFS(fs->lfs_ivnode->v_size >> 2376 fs->lfs_bshift)); 2377 } 2378 2379 if (fs->lfs_ivnode->v_size - (fs->lfs_segtabsz << fs->lfs_bshift) > 2380 LFS_MAX_BYTES) { 2381 simple_lock(&fs->lfs_interlock); 2382 fs->lfs_flags |= LFS_WARNED; 2383 simple_unlock(&fs->lfs_interlock); 2384 log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %" 2385 PRId64 " cannot fit in %lu bytes\n", 2386 fs->lfs_ivnode->v_size - (fs->lfs_segtabsz << 2387 fs->lfs_bshift), 2388 LFS_MAX_BYTES); 2389 log(LOG_WARNING, "lfs_mountfs: please consider increasing" 2390 " BUFPAGES to at least %" PRId64 "\n", 2391 LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size - 2392 (fs->lfs_segtabsz << 2393 fs->lfs_bshift)) >> 2394 PAGE_SHIFT); 2395 } else if(fs->lfs_ivnode->v_size > LFS_MAX_BYTES) { 2396 simple_lock(&fs->lfs_interlock); 2397 fs->lfs_flags |= LFS_WARNED; 2398 simple_unlock(&fs->lfs_interlock); 2399 log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %" PRId64 2400 " cannot fit in %lu buffer bytes\n", 2401 fs->lfs_ivnode->v_size, LFS_MAX_BYTES); 2402 log(LOG_NOTICE, "lfs_mountfs: please consider increasing" 2403 " BUFPAGES to at least %" PRId64 "\n", 2404 LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size - 2405 (fs->lfs_segtabsz << 2406 fs->lfs_bshift)) >> 2407 PAGE_SHIFT); 2408 } 2409 } 2410 2411 /* 2412 * Resize the filesystem to contain the specified number of segments. 2413 */ 2414 int 2415 lfs_resize_fs(struct lfs *fs, int newnsegs) 2416 { 2417 SEGUSE *sup; 2418 struct buf *bp, *obp; 2419 daddr_t olast, nlast, ilast, noff, start, end; 2420 struct vnode *ivp; 2421 struct inode *ip; 2422 int error, badnews, inc, oldnsegs; 2423 int sbbytes, csbbytes, gain, cgain; 2424 int i; 2425 2426 /* Only support v2 and up */ 2427 if (fs->lfs_version < 2) 2428 return EOPNOTSUPP; 2429 2430 /* If we're doing nothing, do it fast */ 2431 oldnsegs = fs->lfs_nseg; 2432 if (newnsegs == oldnsegs) 2433 return 0; 2434 2435 /* We always have to have two superblocks */ 2436 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 2437 return EFBIG; 2438 2439 ivp = fs->lfs_ivnode; 2440 ip = VTOI(ivp); 2441 error = 0; 2442 2443 /* Take the segment lock so no one else calls lfs_newseg() */ 2444 lfs_seglock(fs, SEGM_PROT); 2445 2446 /* 2447 * Make sure the segments we're going to be losing, if any, 2448 * are in fact empty. We hold the seglock, so their status 2449 * cannot change underneath us. Count the superblocks we lose, 2450 * while we're at it. 2451 */ 2452 sbbytes = csbbytes = 0; 2453 cgain = 0; 2454 for (i = newnsegs; i < oldnsegs; i++) { 2455 LFS_SEGENTRY(sup, fs, i, bp); 2456 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2457 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2458 sbbytes += LFS_SBPAD; 2459 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2460 ++cgain; 2461 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2462 csbbytes += LFS_SBPAD; 2463 } 2464 brelse(bp); 2465 if (badnews) { 2466 error = EBUSY; 2467 goto out; 2468 } 2469 } 2470 2471 /* Note old and new segment table endpoints, and old ifile size */ 2472 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 2473 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 2474 ilast = ivp->v_size >> fs->lfs_bshift; 2475 noff = nlast - olast; 2476 2477 /* 2478 * Make sure no one can use the Ifile while we change it around. 2479 * Even after taking the iflock we need to make sure no one still 2480 * is holding Ifile buffers, so we get each one, to drain them. 2481 * (XXX this could be done better.) 2482 */ 2483 simple_lock(&fs->lfs_interlock); 2484 lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock); 2485 simple_unlock(&fs->lfs_interlock); 2486 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 2487 for (i = 0; i < ilast; i++) { 2488 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 2489 brelse(bp); 2490 } 2491 2492 /* Allocate new Ifile blocks */ 2493 for (i = ilast; i < ilast + noff; i++) { 2494 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 2495 &bp) != 0) 2496 panic("balloc extending ifile"); 2497 memset(bp->b_data, 0, fs->lfs_bsize); 2498 VOP_BWRITE(bp); 2499 } 2500 2501 /* Register new ifile size */ 2502 ip->i_size += noff * fs->lfs_bsize; 2503 ip->i_ffs1_size = ip->i_size; 2504 uvm_vnp_setsize(ivp, ip->i_size); 2505 2506 /* Copy the inode table to its new position */ 2507 if (noff != 0) { 2508 if (noff < 0) { 2509 start = nlast; 2510 end = ilast + noff; 2511 inc = 1; 2512 } else { 2513 start = ilast + noff - 1; 2514 end = nlast - 1; 2515 inc = -1; 2516 } 2517 for (i = start; i != end; i += inc) { 2518 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 2519 panic("resize: bread dst blk failed"); 2520 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 2521 panic("resize: bread src blk failed"); 2522 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2523 VOP_BWRITE(bp); 2524 brelse(obp); 2525 } 2526 } 2527 2528 /* If we are expanding, write the new empty SEGUSE entries */ 2529 if (newnsegs > oldnsegs) { 2530 for (i = oldnsegs; i < newnsegs; i++) { 2531 if ((error = bread(ivp, i / fs->lfs_sepb + 2532 fs->lfs_cleansz, 2533 fs->lfs_bsize, NOCRED, &bp)) != 0) 2534 panic("lfs: ifile read: %d", error); 2535 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2536 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2537 memset(sup, 0, sizeof(*sup)); 2538 i++; 2539 } 2540 VOP_BWRITE(bp); 2541 } 2542 } 2543 2544 /* Zero out unused superblock offsets */ 2545 for (i = 2; i < LFS_MAXNUMSB; i++) 2546 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2547 fs->lfs_sboffs[i] = 0x0; 2548 2549 /* 2550 * Correct superblock entries that depend on fs size. 2551 * The computations of these are as follows: 2552 * 2553 * size = segtod(fs, nseg) 2554 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2555 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2556 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2557 * + (segtod(fs, 1) - (offset - curseg)) 2558 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2559 * 2560 * XXX - we should probably adjust minfreeseg as well. 2561 */ 2562 gain = (newnsegs - oldnsegs); 2563 fs->lfs_nseg = newnsegs; 2564 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2565 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2566 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2567 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2568 - gain * btofsb(fs, fs->lfs_bsize / 2); 2569 if (gain > 0) { 2570 fs->lfs_nclean += gain; 2571 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2572 } else { 2573 fs->lfs_nclean -= cgain; 2574 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2575 btofsb(fs, csbbytes); 2576 } 2577 2578 /* Resize segment flag cache */ 2579 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2580 fs->lfs_nseg * sizeof(u_int32_t), 2581 M_SEGMENT, M_WAITOK); 2582 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2583 fs->lfs_nseg * sizeof(u_int32_t), 2584 M_SEGMENT, M_WAITOK); 2585 for (i = oldnsegs; i < newnsegs; i++) 2586 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2587 2588 /* Truncate Ifile if necessary */ 2589 if (noff < 0) 2590 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2591 NOCRED, curlwp); 2592 2593 /* Update cleaner info so the cleaner can die */ 2594 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2595 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2596 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2597 VOP_BWRITE(bp); 2598 2599 /* Let Ifile accesses proceed */ 2600 VOP_UNLOCK(ivp, 0); 2601 simple_lock(&fs->lfs_interlock); 2602 lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock); 2603 simple_unlock(&fs->lfs_interlock); 2604 2605 out: 2606 lfs_segunlock(fs); 2607 return error; 2608 } 2609
Indexes created Mon Sep 29 03:10:08 GMT 2025