lfs_vfsops.c revision 1.195
1 /* $NetBSD: lfs_vfsops.c,v 1.195 2006/03/24 20:05:32 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 1.195 2006/03/24 20:05:32 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 */ 488 fs->lfs_ronly = 0; 489 } 490 if (args.fspec == NULL) 491 return EINVAL; 492 } 493 494 error = set_statvfs_info(path, UIO_USERSPACE, args.fspec, 495 UIO_USERSPACE, mp, l); 496 if (error == 0) 497 (void)strncpy(fs->lfs_fsmnt, mp->mnt_stat.f_mntonname, 498 sizeof(fs->lfs_fsmnt)); 499 return error; 500 501 fail: 502 vrele(devvp); 503 return (error); 504 } 505 506 /* 507 * Roll-forward code. 508 */ 509 510 /* 511 * Load the appropriate indirect block, and change the appropriate pointer. 512 * Mark the block dirty. Do segment and avail accounting. 513 */ 514 static int 515 update_meta(struct lfs *fs, ino_t ino, int vers, daddr_t lbn, 516 daddr_t ndaddr, size_t size, struct lwp *l) 517 { 518 int error; 519 struct vnode *vp; 520 struct inode *ip; 521 #ifdef DEBUG 522 daddr_t odaddr; 523 struct indir a[NIADDR]; 524 int num; 525 int i; 526 #endif /* DEBUG */ 527 struct buf *bp; 528 SEGUSE *sup; 529 530 KASSERT(lbn >= 0); /* no indirect blocks */ 531 532 if ((error = lfs_rf_valloc(fs, ino, vers, l, &vp)) != 0) { 533 DLOG((DLOG_RF, "update_meta: ino %d: lfs_rf_valloc" 534 " returned %d\n", ino, error)); 535 return error; 536 } 537 538 if ((error = lfs_balloc(vp, (lbn << fs->lfs_bshift), size, 539 NOCRED, 0, &bp)) != 0) { 540 vput(vp); 541 return (error); 542 } 543 /* No need to write, the block is already on disk */ 544 if (bp->b_flags & B_DELWRI) { 545 LFS_UNLOCK_BUF(bp); 546 fs->lfs_avail += btofsb(fs, bp->b_bcount); 547 } 548 bp->b_flags |= B_INVAL; 549 brelse(bp); 550 551 /* 552 * Extend the file, if it is not large enough already. 553 * XXX this is not exactly right, we don't know how much of the 554 * XXX last block is actually used. We hope that an inode will 555 * XXX appear later to give the correct size. 556 */ 557 ip = VTOI(vp); 558 if (ip->i_size <= (lbn << fs->lfs_bshift)) { 559 u_int64_t newsize; 560 561 if (lbn < NDADDR) 562 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 563 (size - fs->lfs_fsize) + 1; 564 else 565 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1; 566 567 if (ip->i_size < newsize) { 568 ip->i_size = newsize; 569 /* 570 * tell vm our new size for the case the inode won't 571 * appear later. 572 */ 573 uvm_vnp_setsize(vp, newsize); 574 } 575 } 576 577 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size); 578 579 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 580 sup->su_nbytes += size; 581 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp); 582 583 /* differences here should be due to UNWRITTEN indirect blocks. */ 584 KASSERT((lblkno(fs, ip->i_size) > NDADDR && 585 ip->i_lfs_effnblks == ip->i_ffs1_blocks) || 586 ip->i_lfs_effnblks >= ip->i_ffs1_blocks); 587 588 #ifdef DEBUG 589 /* Now look again to make sure it worked */ 590 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL); 591 for (i = num; i > 0; i--) { 592 if (!a[i].in_exists) 593 panic("update_meta: absent %d lv indirect block", i); 594 } 595 if (dbtofsb(fs, odaddr) != ndaddr) 596 DLOG((DLOG_RF, "update_meta: failed setting ino %d lbn %" 597 PRId64 " to %" PRId64 "\n", ino, lbn, ndaddr)); 598 #endif /* DEBUG */ 599 vput(vp); 600 return 0; 601 } 602 603 static int 604 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred, 605 struct lwp *l) 606 { 607 struct vnode *devvp, *vp; 608 struct inode *ip; 609 struct ufs1_dinode *dip; 610 struct buf *dbp, *ibp; 611 int error; 612 daddr_t daddr; 613 IFILE *ifp; 614 SEGUSE *sup; 615 616 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 617 618 /* 619 * Get the inode, update times and perms. 620 * DO NOT update disk blocks, we do that separately. 621 */ 622 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp); 623 if (error) { 624 DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error)); 625 return error; 626 } 627 dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs); 628 while (--dip >= (struct ufs1_dinode *)dbp->b_data) { 629 if (dip->di_inumber > LFS_IFILE_INUM) { 630 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen, 631 l, &vp); 632 if (error) { 633 DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc" 634 " returned %d\n", error)); 635 continue; 636 } 637 ip = VTOI(vp); 638 if (dip->di_size != ip->i_size) 639 lfs_truncate(vp, dip->di_size, 0, NOCRED, l); 640 /* Get mode, link count, size, and times */ 641 memcpy(ip->i_din.ffs1_din, dip, 642 offsetof(struct ufs1_dinode, di_db[0])); 643 644 /* Then the rest, except di_blocks */ 645 ip->i_flags = ip->i_ffs1_flags = dip->di_flags; 646 ip->i_gen = ip->i_ffs1_gen = dip->di_gen; 647 ip->i_uid = ip->i_ffs1_uid = dip->di_uid; 648 ip->i_gid = ip->i_ffs1_gid = dip->di_gid; 649 650 ip->i_mode = ip->i_ffs1_mode; 651 ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink; 652 ip->i_size = ip->i_ffs1_size; 653 654 LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE); 655 656 /* Re-initialize to get type right */ 657 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, 658 &vp); 659 vput(vp); 660 661 /* Record change in location */ 662 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp); 663 daddr = ifp->if_daddr; 664 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno); 665 error = LFS_BWRITE_LOG(ibp); /* Ifile */ 666 /* And do segment accounting */ 667 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) { 668 if (daddr > 0) { 669 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), 670 ibp); 671 sup->su_nbytes -= sizeof (struct ufs1_dinode); 672 LFS_WRITESEGENTRY(sup, fs, 673 dtosn(fs, daddr), 674 ibp); 675 } 676 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 677 ibp); 678 sup->su_nbytes += sizeof (struct ufs1_dinode); 679 LFS_WRITESEGENTRY(sup, fs, 680 dtosn(fs, dbtofsb(fs, dbp->b_blkno)), 681 ibp); 682 } 683 } 684 } 685 dbp->b_flags |= B_AGE; 686 brelse(dbp); 687 688 return 0; 689 } 690 691 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */ 692 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */ 693 694 static daddr_t 695 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial, 696 struct ucred *cred, int flags, int *pseg_flags, struct lwp *l) 697 { 698 struct vnode *devvp; 699 struct buf *bp, *dbp; 700 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */ 701 SEGSUM *ssp; 702 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */ 703 daddr_t oldoffset; 704 int32_t *iaddr; /* XXX ondisk32 */ 705 FINFO *fip; 706 SEGUSE *sup; 707 size_t size; 708 709 devvp = VTOI(fs->lfs_ivnode)->i_devvp; 710 /* 711 * If the segment has a superblock and we're at the top 712 * of the segment, skip the superblock. 713 */ 714 if (sntod(fs, dtosn(fs, offset)) == offset) { 715 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 716 if (sup->su_flags & SEGUSE_SUPERBLOCK) 717 offset += btofsb(fs, LFS_SBPAD); 718 brelse(bp); 719 } 720 721 /* Read in the segment summary */ 722 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp); 723 if (error) 724 return -1; 725 726 /* Check summary checksum */ 727 ssp = (SEGSUM *)bp->b_data; 728 if (flags & CHECK_CKSUM) { 729 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum, 730 fs->lfs_sumsize - 731 sizeof(ssp->ss_sumsum))) { 732 DLOG((DLOG_RF, "Sumsum error at 0x%" PRIx64 "\n", offset)); 733 offset = -1; 734 goto err1; 735 } 736 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) { 737 DLOG((DLOG_RF, "Empty pseg at 0x%" PRIx64 "\n", offset)); 738 offset = -1; 739 goto err1; 740 } 741 if (ssp->ss_create < fs->lfs_tstamp) { 742 DLOG((DLOG_RF, "Old data at 0x%" PRIx64 "\n", offset)); 743 offset = -1; 744 goto err1; 745 } 746 } 747 if (fs->lfs_version > 1) { 748 if (ssp->ss_serial != nextserial) { 749 DLOG((DLOG_RF, "Unexpected serial number at 0x%" PRIx64 750 "\n", offset)); 751 offset = -1; 752 goto err1; 753 } 754 if (ssp->ss_ident != fs->lfs_ident) { 755 DLOG((DLOG_RF, "Incorrect fsid (0x%x vs 0x%x) at 0x%" 756 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset)); 757 offset = -1; 758 goto err1; 759 } 760 } 761 if (pseg_flags) 762 *pseg_flags = ssp->ss_flags; 763 oldoffset = offset; 764 offset += btofsb(fs, fs->lfs_sumsize); 765 766 ninos = howmany(ssp->ss_ninos, INOPB(fs)); 767 /* XXX ondisk32 */ 768 iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t)); 769 if (flags & CHECK_CKSUM) { 770 /* Count blocks */ 771 nblocks = 0; 772 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 773 for (i = 0; i < ssp->ss_nfinfo; ++i) { 774 nblocks += fip->fi_nblocks; 775 if (fip->fi_nblocks <= 0) 776 break; 777 /* XXX ondisk32 */ 778 fip = (FINFO *)(((char *)fip) + FINFOSIZE + 779 (fip->fi_nblocks * sizeof(int32_t))); 780 } 781 nblocks += ninos; 782 /* Create the sum array */ 783 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long), 784 M_SEGMENT, M_WAITOK); 785 } 786 787 /* Handle individual blocks */ 788 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs)); 789 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) { 790 /* Inode block? */ 791 if (ninos && *iaddr == offset) { 792 if (flags & CHECK_CKSUM) { 793 /* Read in the head and add to the buffer */ 794 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize, 795 cred, &dbp); 796 if (error) { 797 offset = -1; 798 goto err2; 799 } 800 (*dp++) = ((u_long *)(dbp->b_data))[0]; 801 dbp->b_flags |= B_AGE; 802 brelse(dbp); 803 } 804 if (flags & CHECK_UPDATE) { 805 if ((error = update_inoblk(fs, offset, cred, l)) 806 != 0) { 807 offset = -1; 808 goto err2; 809 } 810 } 811 offset += btofsb(fs, fs->lfs_ibsize); 812 --iaddr; 813 --ninos; 814 --i; /* compensate */ 815 continue; 816 } 817 size = fs->lfs_bsize; 818 for (j = 0; j < fip->fi_nblocks; ++j) { 819 if (j == fip->fi_nblocks - 1) 820 size = fip->fi_lastlength; 821 if (flags & CHECK_CKSUM) { 822 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp); 823 if (error) { 824 offset = -1; 825 goto err2; 826 } 827 (*dp++) = ((u_long *)(dbp->b_data))[0]; 828 dbp->b_flags |= B_AGE; 829 brelse(dbp); 830 } 831 /* Account for and update any direct blocks */ 832 if ((flags & CHECK_UPDATE) && 833 fip->fi_ino > LFS_IFILE_INUM && 834 fip->fi_blocks[j] >= 0) { 835 update_meta(fs, fip->fi_ino, fip->fi_version, 836 fip->fi_blocks[j], offset, size, l); 837 } 838 offset += btofsb(fs, size); 839 } 840 /* XXX ondisk32 */ 841 fip = (FINFO *)(((char *)fip) + FINFOSIZE 842 + fip->fi_nblocks * sizeof(int32_t)); 843 } 844 /* Checksum the array, compare */ 845 if ((flags & CHECK_CKSUM) && 846 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long))) 847 { 848 DLOG((DLOG_RF, "Datasum error at 0x%" PRIx64 849 " (wanted %x got %x)\n", 850 offset, ssp->ss_datasum, cksum(datap, nblocks * 851 sizeof(u_long)))); 852 offset = -1; 853 goto err2; 854 } 855 856 /* If we're at the end of the segment, move to the next */ 857 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) != 858 dtosn(fs, offset)) { 859 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) { 860 offset = -1; 861 goto err2; 862 } 863 offset = ssp->ss_next; 864 DLOG((DLOG_RF, "LFS roll forward: moving to offset 0x%" PRIx64 865 " -> segment %d\n", offset, dtosn(fs,offset))); 866 } 867 868 if (flags & CHECK_UPDATE) { 869 fs->lfs_avail -= (offset - oldoffset); 870 /* Don't clog the buffer queue */ 871 simple_lock(&lfs_subsys_lock); 872 if (locked_queue_count > LFS_MAX_BUFS || 873 locked_queue_bytes > LFS_MAX_BYTES) { 874 lfs_flush(fs, SEGM_CKP, 0); 875 } 876 simple_unlock(&lfs_subsys_lock); 877 } 878 879 err2: 880 if (flags & CHECK_CKSUM) 881 free(datap, M_SEGMENT); 882 err1: 883 bp->b_flags |= B_AGE; 884 brelse(bp); 885 886 /* XXX should we update the serial number even for bad psegs? */ 887 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1) 888 fs->lfs_serial = nextserial; 889 return offset; 890 } 891 892 /* 893 * Common code for mount and mountroot 894 * LFS specific 895 */ 896 int 897 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l) 898 { 899 struct dlfs *tdfs, *dfs, *adfs; 900 struct lfs *fs; 901 struct ufsmount *ump; 902 struct vnode *vp; 903 struct buf *bp, *abp; 904 struct partinfo dpart; 905 struct proc *p; 906 dev_t dev; 907 int error, i, ronly, secsize, fsbsize; 908 struct ucred *cred; 909 CLEANERINFO *cip; 910 SEGUSE *sup; 911 int flags, dirty, do_rollforward; 912 daddr_t offset, oldoffset, lastgoodpseg, sb_addr; 913 int sn, curseg; 914 915 p = l ? l->l_proc : NULL; 916 cred = p ? p->p_ucred : NOCRED; 917 918 /* 919 * Flush out any old buffers remaining from a previous use. 920 */ 921 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 922 error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0); 923 VOP_UNLOCK(devvp, 0); 924 if (error) 925 return (error); 926 927 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 928 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, l) != 0) 929 secsize = DEV_BSIZE; 930 else 931 secsize = dpart.disklab->d_secsize; 932 933 /* Don't free random space on error. */ 934 bp = NULL; 935 abp = NULL; 936 ump = NULL; 937 938 sb_addr = LFS_LABELPAD / secsize; 939 while (1) { 940 /* Read in the superblock. */ 941 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp); 942 if (error) 943 goto out; 944 dfs = (struct dlfs *)bp->b_data; 945 946 /* Check the basics. */ 947 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize > MAXBSIZE || 948 dfs->dlfs_version > LFS_VERSION || 949 dfs->dlfs_bsize < sizeof(struct dlfs)) { 950 DLOG((DLOG_MOUNT, "lfs_mountfs: primary superblock sanity failed\n")); 951 error = EINVAL; /* XXX needs translation */ 952 goto out; 953 } 954 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT) { 955 DLOG((DLOG_MOUNT, "lfs_mountfs: unknown inode format %d\n", 956 dfs->dlfs_inodefmt)); 957 error = EINVAL; 958 goto out; 959 } 960 961 if (dfs->dlfs_version == 1) 962 fsbsize = secsize; 963 else { 964 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb + 965 dfs->dlfs_fsbtodb); 966 /* 967 * Could be, if the frag size is large enough, that we 968 * don't have the "real" primary superblock. If that's 969 * the case, get the real one, and try again. 970 */ 971 if (sb_addr != dfs->dlfs_sboffs[0] << 972 dfs->dlfs_fsbtodb) { 973 DLOG((DLOG_MOUNT, "lfs_mountfs: sb daddr" 974 " 0x%llx is not right, trying 0x%llx\n", 975 (long long)sb_addr, 976 (long long)(dfs->dlfs_sboffs[0] << 977 dfs->dlfs_fsbtodb))); 978 sb_addr = dfs->dlfs_sboffs[0] << 979 dfs->dlfs_fsbtodb; 980 brelse(bp); 981 continue; 982 } 983 } 984 break; 985 } 986 987 /* 988 * Check the second superblock to see which is newer; then mount 989 * using the older of the two. This is necessary to ensure that 990 * the filesystem is valid if it was not unmounted cleanly. 991 */ 992 993 if (dfs->dlfs_sboffs[1] && 994 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize) 995 { 996 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize), 997 LFS_SBPAD, cred, &abp); 998 if (error) 999 goto out; 1000 adfs = (struct dlfs *)abp->b_data; 1001 1002 if (dfs->dlfs_version == 1) { 1003 /* 1s resolution comparison */ 1004 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp) 1005 tdfs = adfs; 1006 else 1007 tdfs = dfs; 1008 } else { 1009 /* monotonic infinite-resolution comparison */ 1010 if (adfs->dlfs_serial < dfs->dlfs_serial) 1011 tdfs = adfs; 1012 else 1013 tdfs = dfs; 1014 } 1015 1016 /* Check the basics. */ 1017 if (tdfs->dlfs_magic != LFS_MAGIC || 1018 tdfs->dlfs_bsize > MAXBSIZE || 1019 tdfs->dlfs_version > LFS_VERSION || 1020 tdfs->dlfs_bsize < sizeof(struct dlfs)) { 1021 DLOG((DLOG_MOUNT, "lfs_mountfs: alt superblock" 1022 " sanity failed\n")); 1023 error = EINVAL; /* XXX needs translation */ 1024 goto out; 1025 } 1026 } else { 1027 DLOG((DLOG_MOUNT, "lfs_mountfs: invalid alt superblock" 1028 " daddr=0x%x\n", dfs->dlfs_sboffs[1])); 1029 error = EINVAL; 1030 goto out; 1031 } 1032 1033 /* Allocate the mount structure, copy the superblock into it. */ 1034 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO); 1035 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs)); 1036 1037 /* Compatibility */ 1038 if (fs->lfs_version < 2) { 1039 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 1040 fs->lfs_ibsize = fs->lfs_bsize; 1041 fs->lfs_start = fs->lfs_sboffs[0]; 1042 fs->lfs_tstamp = fs->lfs_otstamp; 1043 fs->lfs_fsbtodb = 0; 1044 } 1045 1046 /* 1047 * If we aren't going to be able to write meaningfully to this 1048 * filesystem, and were not mounted readonly, bomb out now. 1049 */ 1050 if (fsbtob(fs, LFS_NRESERVE(fs)) > LFS_MAX_BYTES && !ronly) { 1051 DLOG((DLOG_MOUNT, "lfs_mount: to mount this filesystem read/write," 1052 " we need BUFPAGES >= %lld\n", 1053 (long long)((bufmem_hiwater / bufmem_lowater) * 1054 LFS_INVERSE_MAX_BYTES( 1055 fsbtob(fs, LFS_NRESERVE(fs))) >> PAGE_SHIFT))); 1056 free(fs, M_UFSMNT); 1057 error = EFBIG; /* XXX needs translation */ 1058 goto out; 1059 } 1060 1061 /* Before rolling forward, lock so vget will sleep for other procs */ 1062 fs->lfs_flags = LFS_NOTYET; 1063 fs->lfs_rfpid = p->p_pid; 1064 1065 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1066 ump->um_lfs = fs; 1067 ump->um_ops = &lfs_ufsops; 1068 ump->um_fstype = UFS1; 1069 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */ 1070 bp->b_flags |= B_INVAL; 1071 abp->b_flags |= B_INVAL; 1072 } 1073 brelse(bp); 1074 bp = NULL; 1075 brelse(abp); 1076 abp = NULL; 1077 1078 /* Set up the I/O information */ 1079 fs->lfs_devbsize = secsize; 1080 fs->lfs_iocount = 0; 1081 fs->lfs_diropwait = 0; 1082 fs->lfs_activesb = 0; 1083 fs->lfs_uinodes = 0; 1084 fs->lfs_ravail = 0; 1085 fs->lfs_favail = 0; 1086 fs->lfs_sbactive = 0; 1087 1088 /* Set up the ifile and lock aflags */ 1089 fs->lfs_doifile = 0; 1090 fs->lfs_writer = 0; 1091 fs->lfs_dirops = 0; 1092 fs->lfs_nadirop = 0; 1093 fs->lfs_seglock = 0; 1094 fs->lfs_pdflush = 0; 1095 fs->lfs_sleepers = 0; 1096 fs->lfs_pages = 0; 1097 simple_lock_init(&fs->lfs_interlock); 1098 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0); 1099 lockinit(&fs->lfs_iflock, PINOD, "lfs_iflock", 0, 0); 1100 1101 /* Set the file system readonly/modify bits. */ 1102 fs->lfs_ronly = ronly; 1103 if (ronly == 0) 1104 fs->lfs_fmod = 1; 1105 1106 /* Initialize the mount structure. */ 1107 dev = devvp->v_rdev; 1108 mp->mnt_data = ump; 1109 mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev; 1110 mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_LFS); 1111 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 1112 mp->mnt_stat.f_namemax = LFS_MAXNAMLEN; 1113 mp->mnt_stat.f_iosize = fs->lfs_bsize; 1114 mp->mnt_flag |= MNT_LOCAL; 1115 mp->mnt_fs_bshift = fs->lfs_bshift; 1116 ump->um_flags = 0; 1117 ump->um_mountp = mp; 1118 ump->um_dev = dev; 1119 ump->um_devvp = devvp; 1120 ump->um_bptrtodb = fs->lfs_fsbtodb; 1121 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag); 1122 ump->um_nindir = fs->lfs_nindir; 1123 ump->um_lognindir = ffs(fs->lfs_nindir) - 1; 1124 for (i = 0; i < MAXQUOTAS; i++) 1125 ump->um_quotas[i] = NULLVP; 1126 ump->um_maxsymlinklen = fs->lfs_maxsymlinklen; 1127 ump->um_dirblksiz = DIRBLKSIZ; 1128 ump->um_maxfilesize = fs->lfs_maxfilesize; 1129 if (ump->um_maxsymlinklen > 0) 1130 mp->mnt_iflag |= IMNT_DTYPE; 1131 devvp->v_specmountpoint = mp; 1132 1133 /* Set up reserved memory for pageout */ 1134 lfs_setup_resblks(fs); 1135 /* Set up vdirop tailq */ 1136 TAILQ_INIT(&fs->lfs_dchainhd); 1137 /* and paging tailq */ 1138 TAILQ_INIT(&fs->lfs_pchainhd); 1139 1140 /* 1141 * We use the ifile vnode for almost every operation. Instead of 1142 * retrieving it from the hash table each time we retrieve it here, 1143 * artificially increment the reference count and keep a pointer 1144 * to it in the incore copy of the superblock. 1145 */ 1146 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) { 1147 DLOG((DLOG_MOUNT, "lfs_mountfs: ifile vget failed, error=%d\n", error)); 1148 goto out; 1149 } 1150 fs->lfs_ivnode = vp; 1151 VREF(vp); 1152 1153 /* Set up segment usage flags for the autocleaner. */ 1154 fs->lfs_nactive = 0; 1155 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *), 1156 M_SEGMENT, M_WAITOK); 1157 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 1158 M_SEGMENT, M_WAITOK); 1159 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t), 1160 M_SEGMENT, M_WAITOK); 1161 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t)); 1162 for (i = 0; i < fs->lfs_nseg; i++) { 1163 int changed; 1164 1165 LFS_SEGENTRY(sup, fs, i, bp); 1166 changed = 0; 1167 if (!ronly) { 1168 if (sup->su_nbytes == 0 && 1169 !(sup->su_flags & SEGUSE_EMPTY)) { 1170 sup->su_flags |= SEGUSE_EMPTY; 1171 ++changed; 1172 } else if (!(sup->su_nbytes == 0) && 1173 (sup->su_flags & SEGUSE_EMPTY)) { 1174 sup->su_flags &= ~SEGUSE_EMPTY; 1175 ++changed; 1176 } 1177 if (sup->su_flags & (SEGUSE_ACTIVE|SEGUSE_INVAL)) { 1178 sup->su_flags &= ~(SEGUSE_ACTIVE|SEGUSE_INVAL); 1179 ++changed; 1180 } 1181 } 1182 fs->lfs_suflags[0][i] = sup->su_flags; 1183 if (changed) 1184 LFS_WRITESEGENTRY(sup, fs, i, bp); 1185 else 1186 brelse(bp); 1187 } 1188 1189 /* 1190 * Roll forward. 1191 * 1192 * We don't roll forward for v1 filesystems, because 1193 * of the danger that the clock was turned back between the last 1194 * checkpoint and crash. This would roll forward garbage. 1195 * 1196 * v2 filesystems don't have this problem because they use a 1197 * monotonically increasing serial number instead of a timestamp. 1198 */ 1199 do_rollforward = (!(fs->lfs_pflags & LFS_PF_CLEAN) && 1200 lfs_do_rfw && fs->lfs_version > 1); 1201 if (do_rollforward) { 1202 u_int64_t nextserial; 1203 /* 1204 * Phase I: Find the address of the last good partial 1205 * segment that was written after the checkpoint. Mark 1206 * the segments in question dirty, so they won't be 1207 * reallocated. 1208 */ 1209 lastgoodpseg = oldoffset = offset = fs->lfs_offset; 1210 flags = 0x0; 1211 DLOG((DLOG_RF, "LFS roll forward phase 1: start at offset 0x%" 1212 PRIx64 "\n", offset)); 1213 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp); 1214 if (!(sup->su_flags & SEGUSE_DIRTY)) 1215 --fs->lfs_nclean; 1216 sup->su_flags |= SEGUSE_DIRTY; 1217 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp); 1218 nextserial = fs->lfs_serial + 1; 1219 while ((offset = check_segsum(fs, offset, nextserial, 1220 cred, CHECK_CKSUM, &flags, l)) > 0) { 1221 nextserial++; 1222 if (sntod(fs, oldoffset) != sntod(fs, offset)) { 1223 LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset), 1224 bp); 1225 if (!(sup->su_flags & SEGUSE_DIRTY)) 1226 --fs->lfs_nclean; 1227 sup->su_flags |= SEGUSE_DIRTY; 1228 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset), 1229 bp); 1230 } 1231 1232 DLOG((DLOG_RF, "LFS roll forward phase 1: offset=0x%" 1233 PRIx64 "\n", offset)); 1234 if (flags & SS_DIROP) { 1235 DLOG((DLOG_RF, "lfs_mountfs: dirops at 0x%" 1236 PRIx64 "\n", oldoffset)); 1237 if (!(flags & SS_CONT)) 1238 DLOG((DLOG_RF, "lfs_mountfs: dirops end " 1239 "at 0x%" PRIx64 "\n", oldoffset)); 1240 } 1241 if (!(flags & SS_CONT)) 1242 lastgoodpseg = offset; 1243 oldoffset = offset; 1244 } 1245 if (flags & SS_CONT) { 1246 DLOG((DLOG_RF, "LFS roll forward: warning: incomplete " 1247 "dirops discarded\n")); 1248 } 1249 DLOG((DLOG_RF, "LFS roll forward phase 1: completed: " 1250 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg)); 1251 oldoffset = fs->lfs_offset; 1252 if (fs->lfs_offset != lastgoodpseg) { 1253 /* Don't overwrite what we're trying to preserve */ 1254 offset = fs->lfs_offset; 1255 fs->lfs_offset = lastgoodpseg; 1256 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset)); 1257 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) { 1258 sn = (sn + 1) % fs->lfs_nseg; 1259 if (sn == curseg) 1260 panic("lfs_mountfs: no clean segments"); 1261 LFS_SEGENTRY(sup, fs, sn, bp); 1262 dirty = (sup->su_flags & SEGUSE_DIRTY); 1263 brelse(bp); 1264 if (!dirty) 1265 break; 1266 } 1267 fs->lfs_nextseg = sntod(fs, sn); 1268 1269 /* 1270 * Phase II: Roll forward from the first superblock. 1271 */ 1272 while (offset != lastgoodpseg) { 1273 DLOG((DLOG_RF, "LFS roll forward phase 2: 0x%" 1274 PRIx64 "\n", offset)); 1275 offset = check_segsum(fs, offset, 1276 fs->lfs_serial + 1, cred, CHECK_UPDATE, 1277 NULL, l); 1278 } 1279 1280 /* 1281 * Finish: flush our changes to disk. 1282 */ 1283 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC); 1284 DLOG((DLOG_RF, "lfs_mountfs: roll forward ", 1285 "recovered %lld blocks\n", 1286 (long long)(lastgoodpseg - oldoffset))); 1287 } 1288 DLOG((DLOG_RF, "LFS roll forward complete\n")); 1289 } 1290 /* If writing, sb is not clean; record in case of immediate crash */ 1291 if (!fs->lfs_ronly) { 1292 fs->lfs_pflags &= ~LFS_PF_CLEAN; 1293 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1294 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1295 } 1296 1297 /* Allow vget now that roll-forward is complete */ 1298 fs->lfs_flags &= ~(LFS_NOTYET); 1299 wakeup(&fs->lfs_flags); 1300 1301 /* 1302 * Initialize the ifile cleaner info with information from 1303 * the superblock. 1304 */ 1305 LFS_CLEANERINFO(cip, fs, bp); 1306 cip->clean = fs->lfs_nclean; 1307 cip->dirty = fs->lfs_nseg - fs->lfs_nclean; 1308 cip->avail = fs->lfs_avail; 1309 cip->bfree = fs->lfs_bfree; 1310 (void) LFS_BWRITE_LOG(bp); /* Ifile */ 1311 1312 /* 1313 * Mark the current segment as ACTIVE, since we're going to 1314 * be writing to it. 1315 */ 1316 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); 1317 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1318 fs->lfs_nactive++; 1319 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */ 1320 1321 /* Now that roll-forward is done, unlock the Ifile */ 1322 vput(vp); 1323 1324 /* Comment on ifile size if it is too large */ 1325 warn_ifile_size(fs); 1326 1327 /* Start the pagedaemon-anticipating daemon */ 1328 if (lfs_writer_daemon == 0 && 1329 kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0) 1330 panic("fork lfs_writer"); 1331 1332 return (0); 1333 1334 out: 1335 if (bp) 1336 brelse(bp); 1337 if (abp) 1338 brelse(abp); 1339 if (ump) { 1340 free(ump->um_lfs, M_UFSMNT); 1341 free(ump, M_UFSMNT); 1342 mp->mnt_data = NULL; 1343 } 1344 1345 return (error); 1346 } 1347 1348 /* 1349 * unmount system call 1350 */ 1351 int 1352 lfs_unmount(struct mount *mp, int mntflags, struct lwp *l) 1353 { 1354 struct ufsmount *ump; 1355 struct lfs *fs; 1356 int error, flags, ronly; 1357 int s; 1358 1359 flags = 0; 1360 if (mntflags & MNT_FORCE) 1361 flags |= FORCECLOSE; 1362 1363 ump = VFSTOUFS(mp); 1364 fs = ump->um_lfs; 1365 1366 /* Write everything we've got */ 1367 lfs_segwrite(mp, SEGM_CKP); 1368 1369 /* wake up the cleaner so it can die */ 1370 wakeup(&fs->lfs_nextseg); 1371 wakeup(&lfs_allclean_wakeup); 1372 simple_lock(&fs->lfs_interlock); 1373 while (fs->lfs_sleepers) 1374 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0, 1375 &fs->lfs_interlock); 1376 simple_unlock(&fs->lfs_interlock); 1377 1378 #ifdef QUOTA 1379 if (mp->mnt_flag & MNT_QUOTA) { 1380 int i; 1381 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags); 1382 if (error) 1383 return (error); 1384 for (i = 0; i < MAXQUOTAS; i++) { 1385 if (ump->um_quotas[i] == NULLVP) 1386 continue; 1387 quotaoff(l, mp, i); 1388 } 1389 /* 1390 * Here we fall through to vflush again to ensure 1391 * that we have gotten rid of all the system vnodes. 1392 */ 1393 } 1394 #endif 1395 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0) 1396 return (error); 1397 if ((error = VFS_SYNC(mp, 1, l->l_proc->p_ucred, l)) != 0) 1398 return (error); 1399 s = splbio(); 1400 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd)) 1401 panic("lfs_unmount: still dirty blocks on ifile vnode"); 1402 splx(s); 1403 1404 /* Comment on ifile size if it has become too large */ 1405 if (!(fs->lfs_flags & LFS_WARNED)) 1406 warn_ifile_size(fs); 1407 1408 /* Explicitly write the superblock, to update serial and pflags */ 1409 fs->lfs_pflags |= LFS_PF_CLEAN; 1410 lfs_writesuper(fs, fs->lfs_sboffs[0]); 1411 lfs_writesuper(fs, fs->lfs_sboffs[1]); 1412 simple_lock(&fs->lfs_interlock); 1413 while (fs->lfs_iocount) 1414 ltsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0, 1415 &fs->lfs_interlock); 1416 simple_unlock(&fs->lfs_interlock); 1417 1418 /* Finish with the Ifile, now that we're done with it */ 1419 vrele(fs->lfs_ivnode); 1420 vgone(fs->lfs_ivnode); 1421 1422 ronly = !fs->lfs_ronly; 1423 if (ump->um_devvp->v_type != VBAD) 1424 ump->um_devvp->v_specmountpoint = NULL; 1425 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1426 error = VOP_CLOSE(ump->um_devvp, 1427 ronly ? FREAD : FREAD|FWRITE, NOCRED, l); 1428 vput(ump->um_devvp); 1429 1430 /* Complain about page leakage */ 1431 if (fs->lfs_pages > 0) 1432 printf("lfs_unmount: still claim %d pages (%d in subsystem)\n", 1433 fs->lfs_pages, lfs_subsys_pages); 1434 1435 /* Free per-mount data structures */ 1436 free(fs->lfs_suflags[0], M_SEGMENT); 1437 free(fs->lfs_suflags[1], M_SEGMENT); 1438 free(fs->lfs_suflags, M_SEGMENT); 1439 lfs_free_resblks(fs); 1440 free(fs, M_UFSMNT); 1441 free(ump, M_UFSMNT); 1442 1443 mp->mnt_data = NULL; 1444 mp->mnt_flag &= ~MNT_LOCAL; 1445 return (error); 1446 } 1447 1448 /* 1449 * Get file system statistics. 1450 * 1451 * NB: We don't lock to access the superblock here, because it's not 1452 * really that important if we get it wrong. 1453 */ 1454 int 1455 lfs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l) 1456 { 1457 struct lfs *fs; 1458 struct ufsmount *ump; 1459 1460 ump = VFSTOUFS(mp); 1461 fs = ump->um_lfs; 1462 if (fs->lfs_magic != LFS_MAGIC) 1463 panic("lfs_statvfs: magic"); 1464 1465 sbp->f_bsize = fs->lfs_bsize; 1466 sbp->f_frsize = fs->lfs_fsize; 1467 sbp->f_iosize = fs->lfs_bsize; 1468 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs) - VTOI(fs->lfs_ivnode)->i_lfs_effnblks); 1469 1470 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs)); 1471 KASSERT(sbp->f_bfree <= fs->lfs_dsize); 1472 if (sbp->f_bfree < 0) 1473 sbp->f_bfree = 0; 1474 1475 sbp->f_bresvd = fsbtofrags(fs, LFS_EST_RSVD(fs)); 1476 if (sbp->f_bfree > sbp->f_bresvd) 1477 sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd; 1478 else 1479 sbp->f_bavail = 0; 1480 1481 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs); 1482 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles; 1483 sbp->f_favail = sbp->f_ffree; 1484 sbp->f_fresvd = 0; 1485 copy_statvfs_info(sbp, mp); 1486 return (0); 1487 } 1488 1489 /* 1490 * Go through the disk queues to initiate sandbagged IO; 1491 * go through the inodes to write those that have been modified; 1492 * initiate the writing of the super block if it has been modified. 1493 * 1494 * Note: we are always called with the filesystem marked `MPBUSY'. 1495 */ 1496 int 1497 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct lwp *l) 1498 { 1499 int error; 1500 struct lfs *fs; 1501 1502 fs = VFSTOUFS(mp)->um_lfs; 1503 if (fs->lfs_ronly) 1504 return 0; 1505 lfs_writer_enter(fs, "lfs_dirops"); 1506 1507 /* All syncs must be checkpoints until roll-forward is implemented. */ 1508 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0)); 1509 lfs_writer_leave(fs); 1510 #ifdef QUOTA 1511 qsync(mp); 1512 #endif 1513 return (error); 1514 } 1515 1516 extern struct lock ufs_hashlock; 1517 1518 /* 1519 * Look up an LFS dinode number to find its incore vnode. If not already 1520 * in core, read it in from the specified device. Return the inode locked. 1521 * Detection and handling of mount points must be done by the calling routine. 1522 */ 1523 int 1524 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1525 { 1526 struct lfs *fs; 1527 struct ufs1_dinode *dip; 1528 struct inode *ip; 1529 struct buf *bp; 1530 struct ifile *ifp; 1531 struct vnode *vp; 1532 struct ufsmount *ump; 1533 daddr_t daddr; 1534 dev_t dev; 1535 int error, retries; 1536 struct timespec ts; 1537 1538 ump = VFSTOUFS(mp); 1539 dev = ump->um_dev; 1540 fs = ump->um_lfs; 1541 1542 /* 1543 * If the filesystem is not completely mounted yet, suspend 1544 * any access requests (wait for roll-forward to complete). 1545 */ 1546 simple_lock(&fs->lfs_interlock); 1547 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid) 1548 ltsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0, 1549 &fs->lfs_interlock); 1550 simple_unlock(&fs->lfs_interlock); 1551 1552 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) 1553 return (0); 1554 1555 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) { 1556 *vpp = NULL; 1557 return (error); 1558 } 1559 1560 do { 1561 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) { 1562 ungetnewvnode(vp); 1563 return (0); 1564 } 1565 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0)); 1566 1567 /* Translate the inode number to a disk address. */ 1568 if (ino == LFS_IFILE_INUM) 1569 daddr = fs->lfs_idaddr; 1570 else { 1571 /* XXX bounds-check this too */ 1572 LFS_IENTRY(ifp, fs, ino, bp); 1573 daddr = ifp->if_daddr; 1574 if (fs->lfs_version > 1) { 1575 ts.tv_sec = ifp->if_atime_sec; 1576 ts.tv_nsec = ifp->if_atime_nsec; 1577 } 1578 1579 brelse(bp); 1580 if (daddr == LFS_UNUSED_DADDR) { 1581 *vpp = NULLVP; 1582 ungetnewvnode(vp); 1583 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1584 return (ENOENT); 1585 } 1586 } 1587 1588 /* Allocate/init new vnode/inode. */ 1589 lfs_vcreate(mp, ino, vp); 1590 1591 /* 1592 * Put it onto its hash chain and lock it so that other requests for 1593 * this inode will block if they arrive while we are sleeping waiting 1594 * for old data structures to be purged or for the contents of the 1595 * disk portion of this inode to be read. 1596 */ 1597 ip = VTOI(vp); 1598 ufs_ihashins(ip); 1599 lockmgr(&ufs_hashlock, LK_RELEASE, 0); 1600 1601 /* 1602 * XXX 1603 * This may not need to be here, logically it should go down with 1604 * the i_devvp initialization. 1605 * Ask Kirk. 1606 */ 1607 ip->i_lfs = ump->um_lfs; 1608 1609 /* Read in the disk contents for the inode, copy into the inode. */ 1610 retries = 0; 1611 again: 1612 error = bread(ump->um_devvp, fsbtodb(fs, daddr), 1613 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize), 1614 NOCRED, &bp); 1615 if (error) { 1616 /* 1617 * The inode does not contain anything useful, so it would 1618 * be misleading to leave it on its hash chain. With mode 1619 * still zero, it will be unlinked and returned to the free 1620 * list by vput(). 1621 */ 1622 vput(vp); 1623 brelse(bp); 1624 *vpp = NULL; 1625 return (error); 1626 } 1627 1628 dip = lfs_ifind(fs, ino, bp); 1629 if (dip == NULL) { 1630 /* Assume write has not completed yet; try again */ 1631 bp->b_flags |= B_INVAL; 1632 brelse(bp); 1633 ++retries; 1634 if (retries > LFS_IFIND_RETRIES) { 1635 #ifdef DEBUG 1636 /* If the seglock is held look at the bpp to see 1637 what is there anyway */ 1638 simple_lock(&fs->lfs_interlock); 1639 if (fs->lfs_seglock > 0) { 1640 struct buf **bpp; 1641 struct ufs1_dinode *dp; 1642 int i; 1643 1644 for (bpp = fs->lfs_sp->bpp; 1645 bpp != fs->lfs_sp->cbpp; ++bpp) { 1646 if ((*bpp)->b_vp == fs->lfs_ivnode && 1647 bpp != fs->lfs_sp->bpp) { 1648 /* Inode block */ 1649 printf("lfs_vget: block 0x%" PRIx64 ": ", 1650 (*bpp)->b_blkno); 1651 dp = (struct ufs1_dinode *)(*bpp)->b_data; 1652 for (i = 0; i < INOPB(fs); i++) 1653 if (dp[i].di_u.inumber) 1654 printf("%d ", dp[i].di_u.inumber); 1655 printf("\n"); 1656 } 1657 } 1658 } 1659 simple_unlock(&fs->lfs_interlock); 1660 #endif /* DEBUG */ 1661 panic("lfs_vget: dinode not found"); 1662 } 1663 simple_lock(&fs->lfs_interlock); 1664 if (fs->lfs_iocount) { 1665 DLOG((DLOG_VNODE, "lfs_vget: dinode %d not found, retrying...\n", ino)); 1666 (void)ltsleep(&fs->lfs_iocount, PRIBIO + 1, 1667 "lfs ifind", 1, &fs->lfs_interlock); 1668 } else 1669 retries = LFS_IFIND_RETRIES; 1670 simple_unlock(&fs->lfs_interlock); 1671 goto again; 1672 } 1673 *ip->i_din.ffs1_din = *dip; 1674 brelse(bp); 1675 1676 if (fs->lfs_version > 1) { 1677 ip->i_ffs1_atime = ts.tv_sec; 1678 ip->i_ffs1_atimensec = ts.tv_nsec; 1679 } 1680 1681 lfs_vinit(mp, &vp); 1682 1683 *vpp = vp; 1684 1685 KASSERT(VOP_ISLOCKED(vp)); 1686 1687 return (0); 1688 } 1689 1690 /* 1691 * File handle to vnode 1692 */ 1693 int 1694 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) 1695 { 1696 struct lfid *lfhp; 1697 struct buf *bp; 1698 IFILE *ifp; 1699 int32_t daddr; 1700 struct lfs *fs; 1701 1702 lfhp = (struct lfid *)fhp; 1703 if (lfhp->lfid_ino < LFS_IFILE_INUM) 1704 return ESTALE; 1705 1706 fs = VFSTOUFS(mp)->um_lfs; 1707 if (lfhp->lfid_ident != fs->lfs_ident) 1708 return ESTALE; 1709 1710 if (lfhp->lfid_ino > 1711 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) - 1712 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb) 1713 return ESTALE; 1714 1715 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) { 1716 LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp); 1717 daddr = ifp->if_daddr; 1718 brelse(bp); 1719 if (daddr == LFS_UNUSED_DADDR) 1720 return ESTALE; 1721 } 1722 1723 return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp)); 1724 } 1725 1726 /* 1727 * Vnode pointer to File handle 1728 */ 1729 /* ARGSUSED */ 1730 int 1731 lfs_vptofh(struct vnode *vp, struct fid *fhp) 1732 { 1733 struct inode *ip; 1734 struct lfid *lfhp; 1735 1736 ip = VTOI(vp); 1737 lfhp = (struct lfid *)fhp; 1738 lfhp->lfid_len = sizeof(struct lfid); 1739 lfhp->lfid_ino = ip->i_number; 1740 lfhp->lfid_gen = ip->i_gen; 1741 lfhp->lfid_ident = ip->i_lfs->lfs_ident; 1742 return (0); 1743 } 1744 1745 static int 1746 sysctl_lfs_dostats(SYSCTLFN_ARGS) 1747 { 1748 extern struct lfs_stats lfs_stats; 1749 extern int lfs_dostats; 1750 int error; 1751 1752 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1753 if (error || newp == NULL) 1754 return (error); 1755 1756 if (lfs_dostats == 0) 1757 memset(&lfs_stats, 0, sizeof(lfs_stats)); 1758 1759 return (0); 1760 } 1761 1762 struct shortlong { 1763 const char *sname; 1764 const char *lname; 1765 }; 1766 1767 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup") 1768 { 1769 int i; 1770 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead, 1771 lfs_fs_pagetrip; 1772 #ifdef DEBUG 1773 extern int lfs_debug_log_subsys[DLOG_MAX]; 1774 struct shortlong dlog_names[DLOG_MAX] = { /* Must match lfs.h ! */ 1775 { "rollforward", "Debug roll-forward code" }, 1776 { "alloc", "Debug inode allocation and free list" }, 1777 { "avail", "Debug space-available-now accounting" }, 1778 { "flush", "Debug flush triggers" }, 1779 { "lockedlist", "Debug locked list accounting" }, 1780 { "vnode_verbose", "Verbose per-vnode-written debugging" }, 1781 { "vnode", "Debug vnode use during segment write" }, 1782 { "segment", "Debug segment writing" }, 1783 { "seguse", "Debug segment used-bytes accounting" }, 1784 { "cleaner", "Debug cleaning routines" }, 1785 { "mount", "Debug mount/unmount routines" }, 1786 { "pagecache", "Debug UBC interactions" }, 1787 { "dirop", "Debug directory-operation accounting" }, 1788 { "malloc", "Debug private malloc accounting" }, 1789 }; 1790 #endif /* DEBUG */ 1791 struct shortlong stat_names[] = { /* Must match lfs.h! */ 1792 { "segsused", "Number of new segments allocated" }, 1793 { "psegwrites", "Number of partial-segment writes" }, 1794 { "psyncwrites", "Number of synchronous partial-segment" 1795 " writes" }, 1796 { "pcleanwrites", "Number of partial-segment writes by the" 1797 " cleaner" }, 1798 { "blocktot", "Number of blocks written" }, 1799 { "cleanblocks", "Number of blocks written by the cleaner" }, 1800 { "ncheckpoints", "Number of checkpoints made" }, 1801 { "nwrites", "Number of whole writes" }, 1802 { "nsync_writes", "Number of synchronous writes" }, 1803 { "wait_exceeded", "Number of times writer waited for" 1804 " cleaner" }, 1805 { "write_exceeded", "Number of times writer invoked flush" }, 1806 { "flush_invoked", "Number of times flush was invoked" }, 1807 { "vflush_invoked", "Number of time vflush was called" }, 1808 { "clean_inlocked", "Number of vnodes skipped for VXLOCK" }, 1809 { "clean_vnlocked", "Number of vnodes skipped for vget failure" }, 1810 { "segs_reclaimed", "Number of segments reclaimed" }, 1811 }; 1812 1813 sysctl_createv(clog, 0, NULL, NULL, 1814 CTLFLAG_PERMANENT, 1815 CTLTYPE_NODE, "vfs", NULL, 1816 NULL, 0, NULL, 0, 1817 CTL_VFS, CTL_EOL); 1818 sysctl_createv(clog, 0, NULL, NULL, 1819 CTLFLAG_PERMANENT, 1820 CTLTYPE_NODE, "lfs", 1821 SYSCTL_DESCR("Log-structured file system"), 1822 NULL, 0, NULL, 0, 1823 CTL_VFS, 5, CTL_EOL); 1824 /* 1825 * XXX the "5" above could be dynamic, thereby eliminating one 1826 * more instance of the "number to vfs" mapping problem, but 1827 * "5" is the order as taken from sys/mount.h 1828 */ 1829 1830 sysctl_createv(clog, 0, NULL, NULL, 1831 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1832 CTLTYPE_INT, "flushindir", NULL, 1833 NULL, 0, &lfs_writeindir, 0, 1834 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL); 1835 sysctl_createv(clog, 0, NULL, NULL, 1836 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1837 CTLTYPE_INT, "clean_vnhead", NULL, 1838 NULL, 0, &lfs_clean_vnhead, 0, 1839 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL); 1840 sysctl_createv(clog, 0, NULL, NULL, 1841 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1842 CTLTYPE_INT, "dostats", 1843 SYSCTL_DESCR("Maintain statistics on LFS operations"), 1844 sysctl_lfs_dostats, 0, &lfs_dostats, 0, 1845 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL); 1846 sysctl_createv(clog, 0, NULL, NULL, 1847 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1848 CTLTYPE_INT, "pagetrip", 1849 SYSCTL_DESCR("How many dirty pages in fs triggers" 1850 " a flush"), 1851 NULL, 0, &lfs_fs_pagetrip, 0, 1852 CTL_VFS, 5, LFS_FS_PAGETRIP, CTL_EOL); 1853 sysctl_createv(clog, 0, NULL, NULL, 1854 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1855 CTLTYPE_INT, "rfw", 1856 SYSCTL_DESCR("Use in-kernel roll-forward on mount"), 1857 NULL, 0, &lfs_do_rfw, 0, 1858 CTL_VFS, 5, LFS_DO_RFW, CTL_EOL); 1859 1860 sysctl_createv(clog, 0, NULL, NULL, 1861 CTLFLAG_PERMANENT, 1862 CTLTYPE_NODE, "stats", 1863 SYSCTL_DESCR("Debugging options"), 1864 NULL, 0, NULL, 0, 1865 CTL_VFS, 5, LFS_STATS, CTL_EOL); 1866 for (i = 0; i < sizeof(struct lfs_stats) / sizeof(u_int); i++) { 1867 sysctl_createv(clog, 0, NULL, NULL, 1868 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1869 CTLTYPE_INT, stat_names[i].sname, 1870 SYSCTL_DESCR(stat_names[i].lname), 1871 NULL, 0, &(((u_int *)&lfs_stats.segsused)[i]), 1872 0, CTL_VFS, 5, LFS_STATS, i, CTL_EOL); 1873 } 1874 1875 #ifdef DEBUG 1876 sysctl_createv(clog, 0, NULL, NULL, 1877 CTLFLAG_PERMANENT, 1878 CTLTYPE_NODE, "debug", 1879 SYSCTL_DESCR("Debugging options"), 1880 NULL, 0, NULL, 0, 1881 CTL_VFS, 5, LFS_DEBUGLOG, CTL_EOL); 1882 for (i = 0; i < DLOG_MAX; i++) { 1883 sysctl_createv(clog, 0, NULL, NULL, 1884 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1885 CTLTYPE_INT, dlog_names[i].sname, 1886 SYSCTL_DESCR(dlog_names[i].lname), 1887 NULL, 0, &(lfs_debug_log_subsys[i]), 0, 1888 CTL_VFS, 5, LFS_DEBUGLOG, i, CTL_EOL); 1889 } 1890 #endif 1891 } 1892 1893 /* 1894 * ufs_bmaparray callback function for writing. 1895 * 1896 * Since blocks will be written to the new segment anyway, 1897 * we don't care about current daddr of them. 1898 */ 1899 static boolean_t 1900 lfs_issequential_hole(const struct ufsmount *ump, 1901 daddr_t daddr0, daddr_t daddr1) 1902 { 1903 daddr0 = (daddr_t)((int32_t)daddr0); /* XXX ondisk32 */ 1904 daddr1 = (daddr_t)((int32_t)daddr1); /* XXX ondisk32 */ 1905 1906 KASSERT(daddr0 == UNWRITTEN || 1907 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR)); 1908 KASSERT(daddr1 == UNWRITTEN || 1909 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR)); 1910 1911 /* NOTE: all we want to know here is 'hole or not'. */ 1912 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */ 1913 1914 /* 1915 * treat UNWRITTENs and all resident blocks as 'contiguous' 1916 */ 1917 if (daddr0 != 0 && daddr1 != 0) 1918 return TRUE; 1919 1920 /* 1921 * both are in hole? 1922 */ 1923 if (daddr0 == 0 && daddr1 == 0) 1924 return TRUE; /* all holes are 'contiguous' for us. */ 1925 1926 return FALSE; 1927 } 1928 1929 /* 1930 * lfs_gop_write functions exactly like genfs_gop_write, except that 1931 * (1) it requires the seglock to be held by its caller, and sp->fip 1932 * to be properly initialized (it will return without re-initializing 1933 * sp->fip, and without calling lfs_writeseg). 1934 * (2) it uses the remaining space in the segment, rather than VOP_BMAP, 1935 * to determine how large a block it can write at once (though it does 1936 * still use VOP_BMAP to find holes in the file); 1937 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks 1938 * (leaving lfs_writeseg to deal with the cluster blocks, so we might 1939 * now have clusters of clusters, ick.) 1940 */ 1941 static int 1942 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags) 1943 { 1944 int i, s, error, run; 1945 int fs_bshift; 1946 vaddr_t kva; 1947 off_t eof, offset, startoffset = 0; 1948 size_t bytes, iobytes, skipbytes; 1949 daddr_t lbn, blkno; 1950 struct vm_page *pg; 1951 struct buf *mbp, *bp; 1952 struct vnode *devvp = VTOI(vp)->i_devvp; 1953 struct inode *ip = VTOI(vp); 1954 struct lfs *fs = ip->i_lfs; 1955 struct segment *sp = fs->lfs_sp; 1956 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist); 1957 1958 ASSERT_SEGLOCK(fs); 1959 1960 /* The Ifile lives in the buffer cache */ 1961 KASSERT(vp != fs->lfs_ivnode); 1962 1963 /* 1964 * We don't want to fill the disk before the cleaner has a chance 1965 * to make room for us. If we're in danger of doing that, fail 1966 * with EAGAIN. The caller will have to notice this, unlock 1967 * so the cleaner can run, relock and try again. 1968 */ 1969 if (LFS_STARVED_FOR_SEGS(fs)) 1970 goto tryagain; 1971 1972 /* 1973 * Sometimes things slip past the filters in lfs_putpages, 1974 * and the pagedaemon tries to write pages---problem is 1975 * that the pagedaemon never acquires the segment lock. 1976 * 1977 * Alternatively, pages that were clean when we called 1978 * genfs_putpages may have become dirty in the meantime. In this 1979 * case the segment header is not properly set up for blocks 1980 * to be added to it. 1981 * 1982 * Unbusy and unclean the pages, and put them on the ACTIVE 1983 * queue under the hypothesis that they couldn't have got here 1984 * unless they were modified *quite* recently. 1985 * 1986 * XXXUBC that last statement is an oversimplification of course. 1987 */ 1988 if (!LFS_SEGLOCK_HELD(fs) || 1989 (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) || 1990 (pgs[0]->offset & fs->lfs_bmask) != 0) { 1991 goto tryagain; 1992 } 1993 1994 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x", 1995 vp, pgs, npages, flags); 1996 1997 GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE); 1998 1999 if (vp->v_type == VREG) 2000 fs_bshift = vp->v_mount->mnt_fs_bshift; 2001 else 2002 fs_bshift = DEV_BSHIFT; 2003 error = 0; 2004 pg = pgs[0]; 2005 startoffset = pg->offset; 2006 KASSERT(eof >= 0); 2007 if (startoffset >= eof) { 2008 goto tryagain; 2009 } else 2010 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset); 2011 skipbytes = 0; 2012 2013 KASSERT(bytes != 0); 2014 2015 /* Swap PG_DELWRI for PG_PAGEOUT */ 2016 for (i = 0; i < npages; i++) 2017 if (pgs[i]->flags & PG_DELWRI) { 2018 KASSERT(!(pgs[i]->flags & PG_PAGEOUT)); 2019 pgs[i]->flags &= ~PG_DELWRI; 2020 pgs[i]->flags |= PG_PAGEOUT; 2021 uvmexp.paging++; 2022 uvm_lock_pageq(); 2023 uvm_pageunwire(pgs[i]); 2024 uvm_unlock_pageq(); 2025 } 2026 2027 /* 2028 * Check to make sure we're starting on a block boundary. 2029 * We'll check later to make sure we always write entire 2030 * blocks (or fragments). 2031 */ 2032 if (startoffset & fs->lfs_bmask) 2033 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n", 2034 startoffset, fs->lfs_bmask, 2035 startoffset & fs->lfs_bmask); 2036 KASSERT((startoffset & fs->lfs_bmask) == 0); 2037 if (bytes & fs->lfs_ffmask) { 2038 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes); 2039 panic("lfs_gop_write: non-integer blocks"); 2040 } 2041 2042 /* 2043 * We could deadlock here on pager_map with UVMPAGER_MAPIN_WAITOK. 2044 * If we would, write what we have and try again. If we don't 2045 * have anything to write, we'll have to sleep. 2046 */ 2047 if ((kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2048 (((SEGSUM *)(sp->segsum))->ss_nfinfo < 1 ? 2049 UVMPAGER_MAPIN_WAITOK : 0))) == 0x0) { 2050 int vers; 2051 2052 DLOG((DLOG_PAGE, "lfs_gop_write: forcing write\n")); 2053 #if 0 2054 " with nfinfo=%d at offset 0x%x\n", 2055 (int)((SEGSUM *)(sp->segsum))->ss_nfinfo, 2056 (unsigned)fs->lfs_offset)); 2057 #endif 2058 if (sp->fip->fi_nblocks == 0) { 2059 /* Don't write zero-length finfos */ 2060 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2061 sp->sum_bytes_left += FINFOSIZE; 2062 } else 2063 lfs_updatemeta(sp); 2064 2065 vers = sp->fip->fi_version; 2066 (void) lfs_writeseg(fs, sp); 2067 2068 sp->fip->fi_version = vers; 2069 sp->fip->fi_ino = ip->i_number; 2070 /* Add the current file to the segment summary. */ 2071 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2072 sp->sum_bytes_left -= FINFOSIZE; 2073 2074 /* 2075 * Having given up all of the pager_map we were holding, 2076 * we can now wait for aiodoned to reclaim it for us 2077 * without fear of deadlock. 2078 */ 2079 kva = uvm_pagermapin(pgs, npages, UVMPAGER_MAPIN_WRITE | 2080 UVMPAGER_MAPIN_WAITOK); 2081 } 2082 2083 s = splbio(); 2084 simple_lock(&global_v_numoutput_slock); 2085 vp->v_numoutput += 2; /* one for biodone, one for aiodone */ 2086 simple_unlock(&global_v_numoutput_slock); 2087 splx(s); 2088 2089 mbp = getiobuf(); 2090 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x", 2091 vp, mbp, vp->v_numoutput, bytes); 2092 mbp->b_bufsize = npages << PAGE_SHIFT; 2093 mbp->b_data = (void *)kva; 2094 mbp->b_resid = mbp->b_bcount = bytes; 2095 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL; 2096 mbp->b_iodone = uvm_aio_biodone; 2097 mbp->b_vp = vp; 2098 2099 bp = NULL; 2100 for (offset = startoffset; 2101 bytes > 0; 2102 offset += iobytes, bytes -= iobytes) { 2103 lbn = offset >> fs_bshift; 2104 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run, 2105 lfs_issequential_hole); 2106 if (error) { 2107 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d", 2108 error,0,0,0); 2109 skipbytes += bytes; 2110 bytes = 0; 2111 break; 2112 } 2113 2114 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset, 2115 bytes); 2116 if (blkno == (daddr_t)-1) { 2117 skipbytes += iobytes; 2118 continue; 2119 } 2120 2121 /* 2122 * Discover how much we can really pack into this buffer. 2123 */ 2124 /* If no room in the current segment, finish it up */ 2125 if (sp->sum_bytes_left < sizeof(int32_t) || 2126 sp->seg_bytes_left < (1 << fs->lfs_bshift)) { 2127 int vers; 2128 2129 lfs_updatemeta(sp); 2130 2131 vers = sp->fip->fi_version; 2132 (void) lfs_writeseg(fs, sp); 2133 2134 sp->fip->fi_version = vers; 2135 sp->fip->fi_ino = ip->i_number; 2136 /* Add the current file to the segment summary. */ 2137 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2138 sp->sum_bytes_left -= FINFOSIZE; 2139 } 2140 /* Check both for space in segment and space in segsum */ 2141 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift) 2142 << fs_bshift); 2143 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t)) 2144 << fs_bshift); 2145 KASSERT(iobytes > 0); 2146 2147 /* if it's really one i/o, don't make a second buf */ 2148 if (offset == startoffset && iobytes == bytes) { 2149 bp = mbp; 2150 /* correct overcount if there is no second buffer */ 2151 s = splbio(); 2152 simple_lock(&global_v_numoutput_slock); 2153 --vp->v_numoutput; 2154 simple_unlock(&global_v_numoutput_slock); 2155 splx(s); 2156 } else { 2157 bp = getiobuf(); 2158 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d", 2159 vp, bp, vp->v_numoutput, 0); 2160 bp->b_data = (char *)kva + 2161 (vaddr_t)(offset - pg->offset); 2162 bp->b_resid = bp->b_bcount = iobytes; 2163 bp->b_flags = B_BUSY|B_WRITE|B_CALL; 2164 bp->b_iodone = uvm_aio_biodone1; 2165 } 2166 2167 /* XXX This is silly ... is this necessary? */ 2168 bp->b_vp = NULL; 2169 s = splbio(); 2170 bgetvp(vp, bp); 2171 splx(s); 2172 2173 bp->b_lblkno = lblkno(fs, offset); 2174 bp->b_private = mbp; 2175 if (devvp->v_type == VBLK) { 2176 bp->b_dev = devvp->v_rdev; 2177 } 2178 VOP_BWRITE(bp); 2179 while (lfs_gatherblock(sp, bp, NULL)) 2180 continue; 2181 } 2182 2183 if (skipbytes) { 2184 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0); 2185 s = splbio(); 2186 if (error) { 2187 mbp->b_flags |= B_ERROR; 2188 mbp->b_error = error; 2189 } 2190 mbp->b_resid -= skipbytes; 2191 if (mbp->b_resid == 0) { 2192 biodone(mbp); 2193 } 2194 splx(s); 2195 } 2196 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0); 2197 return (0); 2198 2199 tryagain: 2200 /* 2201 * We can't write the pages, for whatever reason. 2202 * Clean up after ourselves, and make the caller try again. 2203 */ 2204 simple_lock(&vp->v_interlock); 2205 2206 /* Tell why we're here, if we know */ 2207 if (ip->i_lfs_iflags & LFSI_NO_GOP_WRITE) 2208 DLOG((DLOG_PAGE, "lfs_gop_write: clean pages dirtied\n")); 2209 else if ((pgs[0]->offset & fs->lfs_bmask) != 0) 2210 DLOG((DLOG_PAGE, "lfs_gop_write: not on block boundary\n")); 2211 else if (startoffset >= eof) 2212 DLOG((DLOG_PAGE, "lfs_gop_write: ino %d start 0x%" PRIx64 2213 " eof 0x%" PRIx64 " npages=%d\n", VTOI(vp)->i_number, 2214 pgs[0]->offset, eof, npages)); 2215 else if (LFS_STARVED_FOR_SEGS(fs)) 2216 DLOG((DLOG_PAGE, "lfs_gop_write: avail too low\n")); 2217 else 2218 DLOG((DLOG_PAGE, "lfs_gop_write: seglock not held\n")); 2219 2220 uvm_lock_pageq(); 2221 for (i = 0; i < npages; i++) { 2222 pg = pgs[i]; 2223 2224 if (pg->flags & PG_PAGEOUT) 2225 uvmexp.paging--; 2226 if (pg->flags & PG_DELWRI) { 2227 uvm_pageunwire(pg); 2228 } 2229 uvm_pageactivate(pg); 2230 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED); 2231 DLOG((DLOG_PAGE, "pg[%d] = %p (vp %p off %" PRIx64 ")\n", i, pg, 2232 vp, pg->offset)); 2233 DLOG((DLOG_PAGE, "pg[%d]->flags = %x\n", i, pg->flags)); 2234 DLOG((DLOG_PAGE, "pg[%d]->pqflags = %x\n", i, pg->pqflags)); 2235 DLOG((DLOG_PAGE, "pg[%d]->uanon = %p\n", i, pg->uanon)); 2236 DLOG((DLOG_PAGE, "pg[%d]->uobject = %p\n", i, pg->uobject)); 2237 DLOG((DLOG_PAGE, "pg[%d]->wire_count = %d\n", i, 2238 pg->wire_count)); 2239 DLOG((DLOG_PAGE, "pg[%d]->loan_count = %d\n", i, 2240 pg->loan_count)); 2241 } 2242 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */ 2243 uvm_page_unbusy(pgs, npages); 2244 uvm_unlock_pageq(); 2245 simple_unlock(&vp->v_interlock); 2246 return EAGAIN; 2247 } 2248 2249 /* 2250 * finish vnode/inode initialization. 2251 * used by lfs_vget and lfs_fastvget. 2252 */ 2253 void 2254 lfs_vinit(struct mount *mp, struct vnode **vpp) 2255 { 2256 struct vnode *vp = *vpp; 2257 struct inode *ip = VTOI(vp); 2258 struct ufsmount *ump = VFSTOUFS(mp); 2259 int i; 2260 2261 ip->i_mode = ip->i_ffs1_mode; 2262 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink; 2263 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size; 2264 ip->i_flags = ip->i_ffs1_flags; 2265 ip->i_gen = ip->i_ffs1_gen; 2266 ip->i_uid = ip->i_ffs1_uid; 2267 ip->i_gid = ip->i_ffs1_gid; 2268 2269 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 2270 2271 /* 2272 * Initialize the vnode from the inode, check for aliases. In all 2273 * cases re-init ip, the underlying vnode/inode may have changed. 2274 */ 2275 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp); 2276 ip = VTOI(vp); 2277 2278 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 2279 if (vp->v_type != VLNK || ip->i_size >= ip->i_ump->um_maxsymlinklen) { 2280 struct lfs *fs = ump->um_lfs; 2281 #ifdef DEBUG 2282 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift; 2283 i < NDADDR; i++) { 2284 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 2285 i == 0) 2286 continue; 2287 if (ip->i_ffs1_db[i] != 0) { 2288 inconsistent: 2289 lfs_dump_dinode(ip->i_din.ffs1_din); 2290 panic("inconsistent inode"); 2291 } 2292 } 2293 for ( ; i < NDADDR + NIADDR; i++) { 2294 if (ip->i_ffs1_ib[i - NDADDR] != 0) { 2295 goto inconsistent; 2296 } 2297 } 2298 #endif /* DEBUG */ 2299 for (i = 0; i < NDADDR; i++) 2300 if (ip->i_ffs1_db[i] != 0) 2301 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 2302 } 2303 2304 #ifdef DIAGNOSTIC 2305 if (vp->v_type == VNON) { 2306 # ifdef DEBUG 2307 lfs_dump_dinode(ip->i_din.ffs1_din); 2308 # endif 2309 panic("lfs_vinit: ino %llu is type VNON! (ifmt=%o)\n", 2310 (unsigned long long)ip->i_number, 2311 (ip->i_mode & IFMT) >> 12); 2312 } 2313 #endif /* DIAGNOSTIC */ 2314 2315 /* 2316 * Finish inode initialization now that aliasing has been resolved. 2317 */ 2318 2319 ip->i_devvp = ump->um_devvp; 2320 VREF(ip->i_devvp); 2321 genfs_node_init(vp, &lfs_genfsops); 2322 uvm_vnp_setsize(vp, ip->i_size); 2323 2324 /* Initialize hiblk from file size */ 2325 ip->i_lfs_hiblk = lblkno(ip->i_lfs, ip->i_size + ip->i_lfs->lfs_bsize - 1) - 1; 2326 2327 *vpp = vp; 2328 } 2329 2330 /* 2331 * Warn if the inode portion of the Ifile is too large to be contained 2332 * in the buffer cache, according to LFS_MAX_BUFS / LFS_MAX_BYTES. 2333 * XXX the estimates don't take multiple LFSs into account. 2334 */ 2335 static void 2336 warn_ifile_size(struct lfs *fs) 2337 { 2338 KASSERT(LFS_MAX_BUFS > 0); 2339 KASSERT(LFS_MAX_BYTES > 0); 2340 if (((fs->lfs_ivnode->v_size >> fs->lfs_bshift) - fs->lfs_segtabsz) > 2341 LFS_MAX_BUFS) { 2342 simple_lock(&fs->lfs_interlock); 2343 fs->lfs_flags |= LFS_WARNED; 2344 simple_unlock(&fs->lfs_interlock); 2345 log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %" 2346 PRId64 " cannot fit in %d buffers\n", 2347 fs->lfs_ivnode->v_size - 2348 (fs->lfs_segtabsz << fs->lfs_bshift), 2349 LFS_MAX_BUFS); 2350 log(LOG_WARNING, "lfs_mountfs: please consider increasing NBUF" 2351 " to at least %" PRId64 "\n", 2352 LFS_INVERSE_MAX_BUFS((fs->lfs_ivnode->v_size >> 2353 fs->lfs_bshift) - 2354 fs->lfs_segtabsz)); 2355 } else if ((fs->lfs_ivnode->v_size >> fs->lfs_bshift) > LFS_MAX_BUFS) { 2356 /* Same thing but LOG_NOTICE */ 2357 simple_lock(&fs->lfs_interlock); 2358 fs->lfs_flags |= LFS_WARNED; 2359 simple_unlock(&fs->lfs_interlock); 2360 log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %" 2361 PRId64 " cannot fit in %d buffers\n", 2362 fs->lfs_ivnode->v_size, LFS_MAX_BUFS); 2363 log(LOG_NOTICE, "lfs_mountfs: please consider increasing NBUF" 2364 " to at least %" PRId64 "\n", 2365 LFS_INVERSE_MAX_BUFS(fs->lfs_ivnode->v_size >> 2366 fs->lfs_bshift)); 2367 } 2368 2369 if (fs->lfs_ivnode->v_size - (fs->lfs_segtabsz << fs->lfs_bshift) > 2370 LFS_MAX_BYTES) { 2371 simple_lock(&fs->lfs_interlock); 2372 fs->lfs_flags |= LFS_WARNED; 2373 simple_unlock(&fs->lfs_interlock); 2374 log(LOG_WARNING, "lfs_mountfs: inode part of ifile of length %" 2375 PRId64 " cannot fit in %lu bytes\n", 2376 fs->lfs_ivnode->v_size - (fs->lfs_segtabsz << 2377 fs->lfs_bshift), 2378 LFS_MAX_BYTES); 2379 log(LOG_WARNING, "lfs_mountfs: please consider increasing" 2380 " BUFPAGES to at least %" PRId64 "\n", 2381 LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size - 2382 (fs->lfs_segtabsz << 2383 fs->lfs_bshift)) >> 2384 PAGE_SHIFT); 2385 } else if(fs->lfs_ivnode->v_size > LFS_MAX_BYTES) { 2386 simple_lock(&fs->lfs_interlock); 2387 fs->lfs_flags |= LFS_WARNED; 2388 simple_unlock(&fs->lfs_interlock); 2389 log(LOG_NOTICE, "lfs_mountfs: entire ifile of length %" PRId64 2390 " cannot fit in %lu buffer bytes\n", 2391 fs->lfs_ivnode->v_size, LFS_MAX_BYTES); 2392 log(LOG_NOTICE, "lfs_mountfs: please consider increasing" 2393 " BUFPAGES to at least %" PRId64 "\n", 2394 LFS_INVERSE_MAX_BYTES(fs->lfs_ivnode->v_size - 2395 (fs->lfs_segtabsz << 2396 fs->lfs_bshift)) >> 2397 PAGE_SHIFT); 2398 } 2399 } 2400 2401 /* 2402 * Resize the filesystem to contain the specified number of segments. 2403 */ 2404 int 2405 lfs_resize_fs(struct lfs *fs, int newnsegs) 2406 { 2407 SEGUSE *sup; 2408 struct buf *bp, *obp; 2409 daddr_t olast, nlast, ilast, noff, start, end; 2410 struct vnode *ivp; 2411 struct inode *ip; 2412 int error, badnews, inc, oldnsegs; 2413 int sbbytes, csbbytes, gain, cgain; 2414 int i; 2415 2416 /* Only support v2 and up */ 2417 if (fs->lfs_version < 2) 2418 return EOPNOTSUPP; 2419 2420 /* If we're doing nothing, do it fast */ 2421 oldnsegs = fs->lfs_nseg; 2422 if (newnsegs == oldnsegs) 2423 return 0; 2424 2425 /* We always have to have two superblocks */ 2426 if (newnsegs <= dtosn(fs, fs->lfs_sboffs[1])) 2427 return EFBIG; 2428 2429 ivp = fs->lfs_ivnode; 2430 ip = VTOI(ivp); 2431 error = 0; 2432 2433 /* Take the segment lock so no one else calls lfs_newseg() */ 2434 lfs_seglock(fs, SEGM_PROT); 2435 2436 /* 2437 * Make sure the segments we're going to be losing, if any, 2438 * are in fact empty. We hold the seglock, so their status 2439 * cannot change underneath us. Count the superblocks we lose, 2440 * while we're at it. 2441 */ 2442 sbbytes = csbbytes = 0; 2443 cgain = 0; 2444 for (i = newnsegs; i < oldnsegs; i++) { 2445 LFS_SEGENTRY(sup, fs, i, bp); 2446 badnews = sup->su_nbytes || !(sup->su_flags & SEGUSE_INVAL); 2447 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2448 sbbytes += LFS_SBPAD; 2449 if (!(sup->su_flags & SEGUSE_DIRTY)) { 2450 ++cgain; 2451 if (sup->su_flags & SEGUSE_SUPERBLOCK) 2452 csbbytes += LFS_SBPAD; 2453 } 2454 brelse(bp); 2455 if (badnews) { 2456 error = EBUSY; 2457 goto out; 2458 } 2459 } 2460 2461 /* Note old and new segment table endpoints, and old ifile size */ 2462 olast = fs->lfs_cleansz + fs->lfs_segtabsz; 2463 nlast = howmany(newnsegs, fs->lfs_sepb) + fs->lfs_cleansz; 2464 ilast = ivp->v_size >> fs->lfs_bshift; 2465 noff = nlast - olast; 2466 2467 /* 2468 * Make sure no one can use the Ifile while we change it around. 2469 * Even after taking the iflock we need to make sure no one still 2470 * is holding Ifile buffers, so we get each one, to drain them. 2471 * (XXX this could be done better.) 2472 */ 2473 simple_lock(&fs->lfs_interlock); 2474 lockmgr(&fs->lfs_iflock, LK_EXCLUSIVE, &fs->lfs_interlock); 2475 simple_unlock(&fs->lfs_interlock); 2476 vn_lock(ivp, LK_EXCLUSIVE | LK_RETRY); 2477 for (i = 0; i < ilast; i++) { 2478 bread(ivp, i, fs->lfs_bsize, NOCRED, &bp); 2479 brelse(bp); 2480 } 2481 2482 /* Allocate new Ifile blocks */ 2483 for (i = ilast; i < ilast + noff; i++) { 2484 if (lfs_balloc(ivp, i * fs->lfs_bsize, fs->lfs_bsize, NOCRED, 0, 2485 &bp) != 0) 2486 panic("balloc extending ifile"); 2487 memset(bp->b_data, 0, fs->lfs_bsize); 2488 VOP_BWRITE(bp); 2489 } 2490 2491 /* Register new ifile size */ 2492 ip->i_size += noff * fs->lfs_bsize; 2493 ip->i_ffs1_size = ip->i_size; 2494 uvm_vnp_setsize(ivp, ip->i_size); 2495 2496 /* Copy the inode table to its new position */ 2497 if (noff != 0) { 2498 if (noff < 0) { 2499 start = nlast; 2500 end = ilast + noff; 2501 inc = 1; 2502 } else { 2503 start = ilast + noff - 1; 2504 end = nlast - 1; 2505 inc = -1; 2506 } 2507 for (i = start; i != end; i += inc) { 2508 if (bread(ivp, i, fs->lfs_bsize, NOCRED, &bp) != 0) 2509 panic("resize: bread dst blk failed"); 2510 if (bread(ivp, i - noff, fs->lfs_bsize, NOCRED, &obp)) 2511 panic("resize: bread src blk failed"); 2512 memcpy(bp->b_data, obp->b_data, fs->lfs_bsize); 2513 VOP_BWRITE(bp); 2514 brelse(obp); 2515 } 2516 } 2517 2518 /* If we are expanding, write the new empty SEGUSE entries */ 2519 if (newnsegs > oldnsegs) { 2520 for (i = oldnsegs; i < newnsegs; i++) { 2521 if ((error = bread(ivp, i / fs->lfs_sepb + 2522 fs->lfs_cleansz, 2523 fs->lfs_bsize, NOCRED, &bp)) != 0) 2524 panic("lfs: ifile read: %d", error); 2525 while ((i + 1) % fs->lfs_sepb && i < newnsegs) { 2526 sup = &((SEGUSE *)bp->b_data)[i % fs->lfs_sepb]; 2527 memset(sup, 0, sizeof(*sup)); 2528 i++; 2529 } 2530 VOP_BWRITE(bp); 2531 } 2532 } 2533 2534 /* Zero out unused superblock offsets */ 2535 for (i = 2; i < LFS_MAXNUMSB; i++) 2536 if (dtosn(fs, fs->lfs_sboffs[i]) >= newnsegs) 2537 fs->lfs_sboffs[i] = 0x0; 2538 2539 /* 2540 * Correct superblock entries that depend on fs size. 2541 * The computations of these are as follows: 2542 * 2543 * size = segtod(fs, nseg) 2544 * dsize = segtod(fs, nseg - minfreeseg) - btofsb(#super * LFS_SBPAD) 2545 * bfree = dsize - btofsb(fs, bsize * nseg / 2) - blocks_actually_used 2546 * avail = segtod(fs, nclean) - btofsb(#clean_super * LFS_SBPAD) 2547 * + (segtod(fs, 1) - (offset - curseg)) 2548 * - segtod(fs, minfreeseg - (minfreeseg / 2)) 2549 * 2550 * XXX - we should probably adjust minfreeseg as well. 2551 */ 2552 gain = (newnsegs - oldnsegs); 2553 fs->lfs_nseg = newnsegs; 2554 fs->lfs_segtabsz = nlast - fs->lfs_cleansz; 2555 fs->lfs_size += gain * btofsb(fs, fs->lfs_ssize); 2556 fs->lfs_dsize += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes); 2557 fs->lfs_bfree += gain * btofsb(fs, fs->lfs_ssize) - btofsb(fs, sbbytes) 2558 - gain * btofsb(fs, fs->lfs_bsize / 2); 2559 if (gain > 0) { 2560 fs->lfs_nclean += gain; 2561 fs->lfs_avail += gain * btofsb(fs, fs->lfs_ssize); 2562 } else { 2563 fs->lfs_nclean -= cgain; 2564 fs->lfs_avail -= cgain * btofsb(fs, fs->lfs_ssize) - 2565 btofsb(fs, csbbytes); 2566 } 2567 2568 /* Resize segment flag cache */ 2569 fs->lfs_suflags[0] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2570 fs->lfs_nseg * sizeof(u_int32_t), 2571 M_SEGMENT, M_WAITOK); 2572 fs->lfs_suflags[1] = (u_int32_t *)realloc(fs->lfs_suflags[0], 2573 fs->lfs_nseg * sizeof(u_int32_t), 2574 M_SEGMENT, M_WAITOK); 2575 for (i = oldnsegs; i < newnsegs; i++) 2576 fs->lfs_suflags[0][i] = fs->lfs_suflags[1][i] = 0x0; 2577 2578 /* Truncate Ifile if necessary */ 2579 if (noff < 0) 2580 lfs_truncate(ivp, ivp->v_size + (noff << fs->lfs_bshift), 0, 2581 NOCRED, curlwp); 2582 2583 /* Update cleaner info so the cleaner can die */ 2584 bread(ivp, 0, fs->lfs_bsize, NOCRED, &bp); 2585 ((CLEANERINFO *)bp->b_data)->clean = fs->lfs_nclean; 2586 ((CLEANERINFO *)bp->b_data)->dirty = fs->lfs_nseg - fs->lfs_nclean; 2587 VOP_BWRITE(bp); 2588 2589 /* Let Ifile accesses proceed */ 2590 VOP_UNLOCK(ivp, 0); 2591 simple_lock(&fs->lfs_interlock); 2592 lockmgr(&fs->lfs_iflock, LK_RELEASE, &fs->lfs_interlock); 2593 simple_unlock(&fs->lfs_interlock); 2594 2595 out: 2596 lfs_segunlock(fs); 2597 return error; 2598 } 2599
Indexes created Thu Sep 25 16:09:42 GMT 2025