1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1991, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_quota.h" 38 #include "opt_ufs.h" 39 #include "opt_ffs.h" 40 #include "opt_ddb.h" 41 42 #include <sys/param.h> 43 #include <sys/gsb_crc32.h> 44 #include <sys/systm.h> 45 #include <sys/namei.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/taskqueue.h> 49 #include <sys/kernel.h> 50 #include <sys/ktr.h> 51 #include <sys/vnode.h> 52 #include <sys/mount.h> 53 #include <sys/bio.h> 54 #include <sys/buf.h> 55 #include <sys/conf.h> 56 #include <sys/fcntl.h> 57 #include <sys/ioccom.h> 58 #include <sys/malloc.h> 59 #include <sys/mutex.h> 60 #include <sys/rwlock.h> 61 #include <sys/vmmeter.h> 62 63 #include <security/mac/mac_framework.h> 64 65 #include <ufs/ufs/dir.h> 66 #include <ufs/ufs/extattr.h> 67 #include <ufs/ufs/gjournal.h> 68 #include <ufs/ufs/quota.h> 69 #include <ufs/ufs/ufsmount.h> 70 #include <ufs/ufs/inode.h> 71 #include <ufs/ufs/ufs_extern.h> 72 73 #include <ufs/ffs/fs.h> 74 #include <ufs/ffs/ffs_extern.h> 75 76 #include <vm/vm.h> 77 #include <vm/uma.h> 78 #include <vm/vm_page.h> 79 80 #include <geom/geom.h> 81 #include <geom/geom_vfs.h> 82 83 #include <ddb/ddb.h> 84 85 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 86 87 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 88 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 89 ufs2_daddr_t); 90 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 91 static int ffs_sync_lazy(struct mount *mp); 92 static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size); 93 static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size); 94 95 static vfs_init_t ffs_init; 96 static vfs_uninit_t ffs_uninit; 97 static vfs_extattrctl_t ffs_extattrctl; 98 static vfs_cmount_t ffs_cmount; 99 static vfs_unmount_t ffs_unmount; 100 static vfs_mount_t ffs_mount; 101 static vfs_statfs_t ffs_statfs; 102 static vfs_fhtovp_t ffs_fhtovp; 103 static vfs_sync_t ffs_sync; 104 105 static struct vfsops ufs_vfsops = { 106 .vfs_extattrctl = ffs_extattrctl, 107 .vfs_fhtovp = ffs_fhtovp, 108 .vfs_init = ffs_init, 109 .vfs_mount = ffs_mount, 110 .vfs_cmount = ffs_cmount, 111 .vfs_quotactl = ufs_quotactl, 112 .vfs_root = vfs_cache_root, 113 .vfs_cachedroot = ufs_root, 114 .vfs_statfs = ffs_statfs, 115 .vfs_sync = ffs_sync, 116 .vfs_uninit = ffs_uninit, 117 .vfs_unmount = ffs_unmount, 118 .vfs_vget = ffs_vget, 119 .vfs_susp_clean = process_deferred_inactive, 120 }; 121 122 VFS_SET(ufs_vfsops, ufs, 0); 123 MODULE_VERSION(ufs, 1); 124 125 static b_strategy_t ffs_geom_strategy; 126 static b_write_t ffs_bufwrite; 127 128 static struct buf_ops ffs_ops = { 129 .bop_name = "FFS", 130 .bop_write = ffs_bufwrite, 131 .bop_strategy = ffs_geom_strategy, 132 .bop_sync = bufsync, 133 #ifdef NO_FFS_SNAPSHOT 134 .bop_bdflush = bufbdflush, 135 #else 136 .bop_bdflush = ffs_bdflush, 137 #endif 138 }; 139 140 /* 141 * Note that userquota and groupquota options are not currently used 142 * by UFS/FFS code and generally mount(8) does not pass those options 143 * from userland, but they can be passed by loader(8) via 144 * vfs.root.mountfrom.options. 145 */ 146 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 147 "noclusterw", "noexec", "export", "force", "from", "groupquota", 148 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir", 149 "nosymfollow", "sync", "union", "userquota", "untrusted", NULL }; 150 151 static int 152 ffs_mount(struct mount *mp) 153 { 154 struct vnode *devvp; 155 struct thread *td; 156 struct ufsmount *ump = NULL; 157 struct fs *fs; 158 pid_t fsckpid = 0; 159 int error, error1, flags; 160 uint64_t mntorflags, saved_mnt_flag; 161 accmode_t accmode; 162 struct nameidata ndp; 163 char *fspec; 164 165 td = curthread; 166 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 167 return (EINVAL); 168 if (uma_inode == NULL) { 169 uma_inode = uma_zcreate("FFS inode", 170 sizeof(struct inode), NULL, NULL, NULL, NULL, 171 UMA_ALIGN_PTR, 0); 172 uma_ufs1 = uma_zcreate("FFS1 dinode", 173 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 174 UMA_ALIGN_PTR, 0); 175 uma_ufs2 = uma_zcreate("FFS2 dinode", 176 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 177 UMA_ALIGN_PTR, 0); 178 } 179 180 vfs_deleteopt(mp->mnt_optnew, "groupquota"); 181 vfs_deleteopt(mp->mnt_optnew, "userquota"); 182 183 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 184 if (error) 185 return (error); 186 187 mntorflags = 0; 188 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0) 189 mntorflags |= MNT_UNTRUSTED; 190 191 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 192 mntorflags |= MNT_ACLS; 193 194 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 195 mntorflags |= MNT_SNAPSHOT; 196 /* 197 * Once we have set the MNT_SNAPSHOT flag, do not 198 * persist "snapshot" in the options list. 199 */ 200 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 201 vfs_deleteopt(mp->mnt_opt, "snapshot"); 202 } 203 204 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 && 205 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { 206 /* 207 * Once we have set the restricted PID, do not 208 * persist "fsckpid" in the options list. 209 */ 210 vfs_deleteopt(mp->mnt_optnew, "fsckpid"); 211 vfs_deleteopt(mp->mnt_opt, "fsckpid"); 212 if (mp->mnt_flag & MNT_UPDATE) { 213 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && 214 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 215 vfs_mount_error(mp, 216 "Checker enable: Must be read-only"); 217 return (EINVAL); 218 } 219 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 220 vfs_mount_error(mp, 221 "Checker enable: Must be read-only"); 222 return (EINVAL); 223 } 224 /* Set to -1 if we are done */ 225 if (fsckpid == 0) 226 fsckpid = -1; 227 } 228 229 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 230 if (mntorflags & MNT_ACLS) { 231 vfs_mount_error(mp, 232 "\"acls\" and \"nfsv4acls\" options " 233 "are mutually exclusive"); 234 return (EINVAL); 235 } 236 mntorflags |= MNT_NFS4ACLS; 237 } 238 239 MNT_ILOCK(mp); 240 mp->mnt_flag |= mntorflags; 241 MNT_IUNLOCK(mp); 242 /* 243 * If updating, check whether changing from read-only to 244 * read/write; if there is no device name, that's all we do. 245 */ 246 if (mp->mnt_flag & MNT_UPDATE) { 247 ump = VFSTOUFS(mp); 248 fs = ump->um_fs; 249 devvp = ump->um_devvp; 250 if (fsckpid == -1 && ump->um_fsckpid > 0) { 251 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || 252 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) 253 return (error); 254 g_topology_lock(); 255 /* 256 * Return to normal read-only mode. 257 */ 258 error = g_access(ump->um_cp, 0, -1, 0); 259 g_topology_unlock(); 260 ump->um_fsckpid = 0; 261 } 262 if (fs->fs_ronly == 0 && 263 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 264 /* 265 * Flush any dirty data and suspend filesystem. 266 */ 267 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 268 return (error); 269 error = vfs_write_suspend_umnt(mp); 270 if (error != 0) 271 return (error); 272 /* 273 * Check for and optionally get rid of files open 274 * for writing. 275 */ 276 flags = WRITECLOSE; 277 if (mp->mnt_flag & MNT_FORCE) 278 flags |= FORCECLOSE; 279 if (MOUNTEDSOFTDEP(mp)) { 280 error = softdep_flushfiles(mp, flags, td); 281 } else { 282 error = ffs_flushfiles(mp, flags, td); 283 } 284 if (error) { 285 vfs_write_resume(mp, 0); 286 return (error); 287 } 288 if (fs->fs_pendingblocks != 0 || 289 fs->fs_pendinginodes != 0) { 290 printf("WARNING: %s Update error: blocks %jd " 291 "files %d\n", fs->fs_fsmnt, 292 (intmax_t)fs->fs_pendingblocks, 293 fs->fs_pendinginodes); 294 fs->fs_pendingblocks = 0; 295 fs->fs_pendinginodes = 0; 296 } 297 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 298 fs->fs_clean = 1; 299 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 300 fs->fs_ronly = 0; 301 fs->fs_clean = 0; 302 vfs_write_resume(mp, 0); 303 return (error); 304 } 305 if (MOUNTEDSOFTDEP(mp)) 306 softdep_unmount(mp); 307 g_topology_lock(); 308 /* 309 * Drop our write and exclusive access. 310 */ 311 g_access(ump->um_cp, 0, -1, -1); 312 g_topology_unlock(); 313 fs->fs_ronly = 1; 314 MNT_ILOCK(mp); 315 mp->mnt_flag |= MNT_RDONLY; 316 MNT_IUNLOCK(mp); 317 /* 318 * Allow the writers to note that filesystem 319 * is ro now. 320 */ 321 vfs_write_resume(mp, 0); 322 } 323 if ((mp->mnt_flag & MNT_RELOAD) && 324 (error = ffs_reload(mp, td, 0)) != 0) 325 return (error); 326 if (fs->fs_ronly && 327 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 328 /* 329 * If we are running a checker, do not allow upgrade. 330 */ 331 if (ump->um_fsckpid > 0) { 332 vfs_mount_error(mp, 333 "Active checker, cannot upgrade to write"); 334 return (EINVAL); 335 } 336 /* 337 * If upgrade to read-write by non-root, then verify 338 * that user has necessary permissions on the device. 339 */ 340 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 341 error = VOP_ACCESS(devvp, VREAD | VWRITE, 342 td->td_ucred, td); 343 if (error) 344 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 345 if (error) { 346 VOP_UNLOCK(devvp); 347 return (error); 348 } 349 VOP_UNLOCK(devvp); 350 fs->fs_flags &= ~FS_UNCLEAN; 351 if (fs->fs_clean == 0) { 352 fs->fs_flags |= FS_UNCLEAN; 353 if ((mp->mnt_flag & MNT_FORCE) || 354 ((fs->fs_flags & 355 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 356 (fs->fs_flags & FS_DOSOFTDEP))) { 357 printf("WARNING: %s was not properly " 358 "dismounted\n", fs->fs_fsmnt); 359 } else { 360 vfs_mount_error(mp, 361 "R/W mount of %s denied. %s.%s", 362 fs->fs_fsmnt, 363 "Filesystem is not clean - run fsck", 364 (fs->fs_flags & FS_SUJ) == 0 ? "" : 365 " Forced mount will invalidate" 366 " journal contents"); 367 return (EPERM); 368 } 369 } 370 g_topology_lock(); 371 /* 372 * Request exclusive write access. 373 */ 374 error = g_access(ump->um_cp, 0, 1, 1); 375 g_topology_unlock(); 376 if (error) 377 return (error); 378 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 379 return (error); 380 error = vfs_write_suspend_umnt(mp); 381 if (error != 0) 382 return (error); 383 fs->fs_ronly = 0; 384 MNT_ILOCK(mp); 385 saved_mnt_flag = MNT_RDONLY; 386 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag & 387 MNT_ASYNC) != 0) 388 saved_mnt_flag |= MNT_ASYNC; 389 mp->mnt_flag &= ~saved_mnt_flag; 390 MNT_IUNLOCK(mp); 391 fs->fs_mtime = time_second; 392 /* check to see if we need to start softdep */ 393 if ((fs->fs_flags & FS_DOSOFTDEP) && 394 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 395 fs->fs_ronly = 1; 396 MNT_ILOCK(mp); 397 mp->mnt_flag |= saved_mnt_flag; 398 MNT_IUNLOCK(mp); 399 vfs_write_resume(mp, 0); 400 return (error); 401 } 402 fs->fs_clean = 0; 403 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 404 fs->fs_ronly = 1; 405 MNT_ILOCK(mp); 406 mp->mnt_flag |= saved_mnt_flag; 407 MNT_IUNLOCK(mp); 408 vfs_write_resume(mp, 0); 409 return (error); 410 } 411 if (fs->fs_snapinum[0] != 0) 412 ffs_snapshot_mount(mp); 413 vfs_write_resume(mp, 0); 414 } 415 /* 416 * Soft updates is incompatible with "async", 417 * so if we are doing softupdates stop the user 418 * from setting the async flag in an update. 419 * Softdep_mount() clears it in an initial mount 420 * or ro->rw remount. 421 */ 422 if (MOUNTEDSOFTDEP(mp)) { 423 /* XXX: Reset too late ? */ 424 MNT_ILOCK(mp); 425 mp->mnt_flag &= ~MNT_ASYNC; 426 MNT_IUNLOCK(mp); 427 } 428 /* 429 * Keep MNT_ACLS flag if it is stored in superblock. 430 */ 431 if ((fs->fs_flags & FS_ACLS) != 0) { 432 /* XXX: Set too late ? */ 433 MNT_ILOCK(mp); 434 mp->mnt_flag |= MNT_ACLS; 435 MNT_IUNLOCK(mp); 436 } 437 438 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 439 /* XXX: Set too late ? */ 440 MNT_ILOCK(mp); 441 mp->mnt_flag |= MNT_NFS4ACLS; 442 MNT_IUNLOCK(mp); 443 } 444 /* 445 * If this is a request from fsck to clean up the filesystem, 446 * then allow the specified pid to proceed. 447 */ 448 if (fsckpid > 0) { 449 if (ump->um_fsckpid != 0) { 450 vfs_mount_error(mp, 451 "Active checker already running on %s", 452 fs->fs_fsmnt); 453 return (EINVAL); 454 } 455 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 456 ("soft updates enabled on read-only file system")); 457 g_topology_lock(); 458 /* 459 * Request write access. 460 */ 461 error = g_access(ump->um_cp, 0, 1, 0); 462 g_topology_unlock(); 463 if (error) { 464 vfs_mount_error(mp, 465 "Checker activation failed on %s", 466 fs->fs_fsmnt); 467 return (error); 468 } 469 ump->um_fsckpid = fsckpid; 470 if (fs->fs_snapinum[0] != 0) 471 ffs_snapshot_mount(mp); 472 fs->fs_mtime = time_second; 473 fs->fs_fmod = 1; 474 fs->fs_clean = 0; 475 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 476 } 477 478 /* 479 * If this is a snapshot request, take the snapshot. 480 */ 481 if (mp->mnt_flag & MNT_SNAPSHOT) 482 return (ffs_snapshot(mp, fspec)); 483 484 /* 485 * Must not call namei() while owning busy ref. 486 */ 487 vfs_unbusy(mp); 488 } 489 490 /* 491 * Not an update, or updating the name: look up the name 492 * and verify that it refers to a sensible disk device. 493 */ 494 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 495 error = namei(&ndp); 496 if ((mp->mnt_flag & MNT_UPDATE) != 0) { 497 /* 498 * Unmount does not start if MNT_UPDATE is set. Mount 499 * update busies mp before setting MNT_UPDATE. We 500 * must be able to retain our busy ref succesfully, 501 * without sleep. 502 */ 503 error1 = vfs_busy(mp, MBF_NOWAIT); 504 MPASS(error1 == 0); 505 } 506 if (error != 0) 507 return (error); 508 NDFREE(&ndp, NDF_ONLY_PNBUF); 509 devvp = ndp.ni_vp; 510 if (!vn_isdisk(devvp, &error)) { 511 vput(devvp); 512 return (error); 513 } 514 515 /* 516 * If mount by non-root, then verify that user has necessary 517 * permissions on the device. 518 */ 519 accmode = VREAD; 520 if ((mp->mnt_flag & MNT_RDONLY) == 0) 521 accmode |= VWRITE; 522 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 523 if (error) 524 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 525 if (error) { 526 vput(devvp); 527 return (error); 528 } 529 530 if (mp->mnt_flag & MNT_UPDATE) { 531 /* 532 * Update only 533 * 534 * If it's not the same vnode, or at least the same device 535 * then it's not correct. 536 */ 537 538 if (devvp->v_rdev != ump->um_devvp->v_rdev) 539 error = EINVAL; /* needs translation */ 540 vput(devvp); 541 if (error) 542 return (error); 543 } else { 544 /* 545 * New mount 546 * 547 * We need the name for the mount point (also used for 548 * "last mounted on") copied in. If an error occurs, 549 * the mount point is discarded by the upper level code. 550 * Note that vfs_mount_alloc() populates f_mntonname for us. 551 */ 552 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 553 vrele(devvp); 554 return (error); 555 } 556 if (fsckpid > 0) { 557 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 558 ("soft updates enabled on read-only file system")); 559 ump = VFSTOUFS(mp); 560 fs = ump->um_fs; 561 g_topology_lock(); 562 /* 563 * Request write access. 564 */ 565 error = g_access(ump->um_cp, 0, 1, 0); 566 g_topology_unlock(); 567 if (error) { 568 printf("WARNING: %s: Checker activation " 569 "failed\n", fs->fs_fsmnt); 570 } else { 571 ump->um_fsckpid = fsckpid; 572 if (fs->fs_snapinum[0] != 0) 573 ffs_snapshot_mount(mp); 574 fs->fs_mtime = time_second; 575 fs->fs_clean = 0; 576 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 577 } 578 } 579 } 580 vfs_mountedfrom(mp, fspec); 581 return (0); 582 } 583 584 /* 585 * Compatibility with old mount system call. 586 */ 587 588 static int 589 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) 590 { 591 struct ufs_args args; 592 struct export_args exp; 593 int error; 594 595 if (data == NULL) 596 return (EINVAL); 597 error = copyin(data, &args, sizeof args); 598 if (error) 599 return (error); 600 vfs_oexport_conv(&args.export, &exp); 601 602 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 603 ma = mount_arg(ma, "export", &exp, sizeof(exp)); 604 error = kernel_mount(ma, flags); 605 606 return (error); 607 } 608 609 /* 610 * Reload all incore data for a filesystem (used after running fsck on 611 * the root filesystem and finding things to fix). If the 'force' flag 612 * is 0, the filesystem must be mounted read-only. 613 * 614 * Things to do to update the mount: 615 * 1) invalidate all cached meta-data. 616 * 2) re-read superblock from disk. 617 * 3) re-read summary information from disk. 618 * 4) invalidate all inactive vnodes. 619 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary 620 * writers, if requested. 621 * 6) invalidate all cached file data. 622 * 7) re-read inode data for all active vnodes. 623 */ 624 int 625 ffs_reload(struct mount *mp, struct thread *td, int flags) 626 { 627 struct vnode *vp, *mvp, *devvp; 628 struct inode *ip; 629 void *space; 630 struct buf *bp; 631 struct fs *fs, *newfs; 632 struct ufsmount *ump; 633 ufs2_daddr_t sblockloc; 634 int i, blks, error; 635 u_long size; 636 int32_t *lp; 637 638 ump = VFSTOUFS(mp); 639 640 MNT_ILOCK(mp); 641 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { 642 MNT_IUNLOCK(mp); 643 return (EINVAL); 644 } 645 MNT_IUNLOCK(mp); 646 647 /* 648 * Step 1: invalidate all cached meta-data. 649 */ 650 devvp = VFSTOUFS(mp)->um_devvp; 651 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 652 if (vinvalbuf(devvp, 0, 0, 0) != 0) 653 panic("ffs_reload: dirty1"); 654 VOP_UNLOCK(devvp); 655 656 /* 657 * Step 2: re-read superblock from disk. 658 */ 659 fs = VFSTOUFS(mp)->um_fs; 660 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 661 NOCRED, &bp)) != 0) 662 return (error); 663 newfs = (struct fs *)bp->b_data; 664 if ((newfs->fs_magic != FS_UFS1_MAGIC && 665 newfs->fs_magic != FS_UFS2_MAGIC) || 666 newfs->fs_bsize > MAXBSIZE || 667 newfs->fs_bsize < sizeof(struct fs)) { 668 brelse(bp); 669 return (EIO); /* XXX needs translation */ 670 } 671 /* 672 * Copy pointer fields back into superblock before copying in XXX 673 * new superblock. These should really be in the ufsmount. XXX 674 * Note that important parameters (eg fs_ncg) are unchanged. 675 */ 676 newfs->fs_csp = fs->fs_csp; 677 newfs->fs_maxcluster = fs->fs_maxcluster; 678 newfs->fs_contigdirs = fs->fs_contigdirs; 679 newfs->fs_active = fs->fs_active; 680 newfs->fs_ronly = fs->fs_ronly; 681 sblockloc = fs->fs_sblockloc; 682 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 683 brelse(bp); 684 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 685 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 686 UFS_LOCK(ump); 687 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 688 printf("WARNING: %s: reload pending error: blocks %jd " 689 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 690 fs->fs_pendinginodes); 691 fs->fs_pendingblocks = 0; 692 fs->fs_pendinginodes = 0; 693 } 694 UFS_UNLOCK(ump); 695 696 /* 697 * Step 3: re-read summary information from disk. 698 */ 699 size = fs->fs_cssize; 700 blks = howmany(size, fs->fs_fsize); 701 if (fs->fs_contigsumsize > 0) 702 size += fs->fs_ncg * sizeof(int32_t); 703 size += fs->fs_ncg * sizeof(u_int8_t); 704 free(fs->fs_csp, M_UFSMNT); 705 space = malloc(size, M_UFSMNT, M_WAITOK); 706 fs->fs_csp = space; 707 for (i = 0; i < blks; i += fs->fs_frag) { 708 size = fs->fs_bsize; 709 if (i + fs->fs_frag > blks) 710 size = (blks - i) * fs->fs_fsize; 711 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 712 NOCRED, &bp); 713 if (error) 714 return (error); 715 bcopy(bp->b_data, space, (u_int)size); 716 space = (char *)space + size; 717 brelse(bp); 718 } 719 /* 720 * We no longer know anything about clusters per cylinder group. 721 */ 722 if (fs->fs_contigsumsize > 0) { 723 fs->fs_maxcluster = lp = space; 724 for (i = 0; i < fs->fs_ncg; i++) 725 *lp++ = fs->fs_contigsumsize; 726 space = lp; 727 } 728 size = fs->fs_ncg * sizeof(u_int8_t); 729 fs->fs_contigdirs = (u_int8_t *)space; 730 bzero(fs->fs_contigdirs, size); 731 if ((flags & FFSR_UNSUSPEND) != 0) { 732 MNT_ILOCK(mp); 733 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 734 wakeup(&mp->mnt_flag); 735 MNT_IUNLOCK(mp); 736 } 737 738 loop: 739 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 740 /* 741 * Skip syncer vnode. 742 */ 743 if (vp->v_type == VNON) { 744 VI_UNLOCK(vp); 745 continue; 746 } 747 /* 748 * Step 4: invalidate all cached file data. 749 */ 750 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 751 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 752 goto loop; 753 } 754 if (vinvalbuf(vp, 0, 0, 0)) 755 panic("ffs_reload: dirty2"); 756 /* 757 * Step 5: re-read inode data for all active vnodes. 758 */ 759 ip = VTOI(vp); 760 error = 761 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 762 (int)fs->fs_bsize, NOCRED, &bp); 763 if (error) { 764 vput(vp); 765 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 766 return (error); 767 } 768 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) { 769 brelse(bp); 770 vput(vp); 771 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 772 return (error); 773 } 774 ip->i_effnlink = ip->i_nlink; 775 brelse(bp); 776 vput(vp); 777 } 778 return (0); 779 } 780 781 /* 782 * Common code for mount and mountroot 783 */ 784 static int 785 ffs_mountfs(devvp, mp, td) 786 struct vnode *devvp; 787 struct mount *mp; 788 struct thread *td; 789 { 790 struct ufsmount *ump; 791 struct fs *fs; 792 struct cdev *dev; 793 int error, i, len, ronly; 794 struct ucred *cred; 795 struct g_consumer *cp; 796 struct mount *nmp; 797 int candelete; 798 off_t loc; 799 800 fs = NULL; 801 ump = NULL; 802 cred = td ? td->td_ucred : NOCRED; 803 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 804 805 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); 806 dev = devvp->v_rdev; 807 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, 808 (uintptr_t)mp) == 0) { 809 VOP_UNLOCK(devvp); 810 return (EBUSY); 811 } 812 g_topology_lock(); 813 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 814 g_topology_unlock(); 815 if (error != 0) { 816 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 817 VOP_UNLOCK(devvp); 818 return (error); 819 } 820 dev_ref(dev); 821 devvp->v_bufobj.bo_ops = &ffs_ops; 822 VOP_UNLOCK(devvp); 823 if (dev->si_iosize_max != 0) 824 mp->mnt_iosize_max = dev->si_iosize_max; 825 if (mp->mnt_iosize_max > MAXPHYS) 826 mp->mnt_iosize_max = MAXPHYS; 827 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 828 error = EINVAL; 829 vfs_mount_error(mp, 830 "Invalid sectorsize %d for superblock size %d", 831 cp->provider->sectorsize, SBLOCKSIZE); 832 goto out; 833 } 834 /* fetch the superblock and summary information */ 835 loc = STDSB; 836 if ((mp->mnt_flag & MNT_ROOTFS) != 0) 837 loc = STDSB_NOHASHFAIL; 838 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0) 839 goto out; 840 /* none of these types of check-hashes are maintained by this kernel */ 841 fs->fs_metackhash &= ~(CK_INDIR | CK_DIR); 842 /* no support for any undefined flags */ 843 fs->fs_flags &= FS_SUPPORTED; 844 fs->fs_flags &= ~FS_UNCLEAN; 845 if (fs->fs_clean == 0) { 846 fs->fs_flags |= FS_UNCLEAN; 847 if (ronly || (mp->mnt_flag & MNT_FORCE) || 848 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 849 (fs->fs_flags & FS_DOSOFTDEP))) { 850 printf("WARNING: %s was not properly dismounted\n", 851 fs->fs_fsmnt); 852 } else { 853 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 854 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 855 (fs->fs_flags & FS_SUJ) == 0 ? "" : 856 " Forced mount will invalidate journal contents"); 857 error = EPERM; 858 goto out; 859 } 860 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 861 (mp->mnt_flag & MNT_FORCE)) { 862 printf("WARNING: %s: lost blocks %jd files %d\n", 863 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 864 fs->fs_pendinginodes); 865 fs->fs_pendingblocks = 0; 866 fs->fs_pendinginodes = 0; 867 } 868 } 869 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 870 printf("WARNING: %s: mount pending error: blocks %jd " 871 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 872 fs->fs_pendinginodes); 873 fs->fs_pendingblocks = 0; 874 fs->fs_pendinginodes = 0; 875 } 876 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 877 #ifdef UFS_GJOURNAL 878 /* 879 * Get journal provider name. 880 */ 881 len = 1024; 882 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); 883 if (g_io_getattr("GJOURNAL::provider", cp, &len, 884 mp->mnt_gjprovider) == 0) { 885 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, 886 M_UFSMNT, M_WAITOK); 887 MNT_ILOCK(mp); 888 mp->mnt_flag |= MNT_GJOURNAL; 889 MNT_IUNLOCK(mp); 890 } else { 891 printf("WARNING: %s: GJOURNAL flag on fs " 892 "but no gjournal provider below\n", 893 mp->mnt_stat.f_mntonname); 894 free(mp->mnt_gjprovider, M_UFSMNT); 895 mp->mnt_gjprovider = NULL; 896 } 897 #else 898 printf("WARNING: %s: GJOURNAL flag on fs but no " 899 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 900 #endif 901 } else { 902 mp->mnt_gjprovider = NULL; 903 } 904 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 905 ump->um_cp = cp; 906 ump->um_bo = &devvp->v_bufobj; 907 ump->um_fs = fs; 908 if (fs->fs_magic == FS_UFS1_MAGIC) { 909 ump->um_fstype = UFS1; 910 ump->um_balloc = ffs_balloc_ufs1; 911 } else { 912 ump->um_fstype = UFS2; 913 ump->um_balloc = ffs_balloc_ufs2; 914 } 915 ump->um_blkatoff = ffs_blkatoff; 916 ump->um_truncate = ffs_truncate; 917 ump->um_update = ffs_update; 918 ump->um_valloc = ffs_valloc; 919 ump->um_vfree = ffs_vfree; 920 ump->um_ifree = ffs_ifree; 921 ump->um_rdonly = ffs_rdonly; 922 ump->um_snapgone = ffs_snapgone; 923 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) 924 ump->um_check_blkno = ffs_check_blkno; 925 else 926 ump->um_check_blkno = NULL; 927 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 928 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); 929 fs->fs_ronly = ronly; 930 fs->fs_active = NULL; 931 mp->mnt_data = ump; 932 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 933 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 934 nmp = NULL; 935 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 936 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 937 if (nmp) 938 vfs_rel(nmp); 939 vfs_getnewfsid(mp); 940 } 941 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 942 MNT_ILOCK(mp); 943 mp->mnt_flag |= MNT_LOCAL; 944 MNT_IUNLOCK(mp); 945 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 946 #ifdef MAC 947 MNT_ILOCK(mp); 948 mp->mnt_flag |= MNT_MULTILABEL; 949 MNT_IUNLOCK(mp); 950 #else 951 printf("WARNING: %s: multilabel flag on fs but " 952 "no MAC support\n", mp->mnt_stat.f_mntonname); 953 #endif 954 } 955 if ((fs->fs_flags & FS_ACLS) != 0) { 956 #ifdef UFS_ACL 957 MNT_ILOCK(mp); 958 959 if (mp->mnt_flag & MNT_NFS4ACLS) 960 printf("WARNING: %s: ACLs flag on fs conflicts with " 961 "\"nfsv4acls\" mount option; option ignored\n", 962 mp->mnt_stat.f_mntonname); 963 mp->mnt_flag &= ~MNT_NFS4ACLS; 964 mp->mnt_flag |= MNT_ACLS; 965 966 MNT_IUNLOCK(mp); 967 #else 968 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 969 mp->mnt_stat.f_mntonname); 970 #endif 971 } 972 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 973 #ifdef UFS_ACL 974 MNT_ILOCK(mp); 975 976 if (mp->mnt_flag & MNT_ACLS) 977 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 978 "with \"acls\" mount option; option ignored\n", 979 mp->mnt_stat.f_mntonname); 980 mp->mnt_flag &= ~MNT_ACLS; 981 mp->mnt_flag |= MNT_NFS4ACLS; 982 983 MNT_IUNLOCK(mp); 984 #else 985 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 986 "ACLs support\n", mp->mnt_stat.f_mntonname); 987 #endif 988 } 989 if ((fs->fs_flags & FS_TRIM) != 0) { 990 len = sizeof(int); 991 if (g_io_getattr("GEOM::candelete", cp, &len, 992 &candelete) == 0) { 993 if (candelete) 994 ump->um_flags |= UM_CANDELETE; 995 else 996 printf("WARNING: %s: TRIM flag on fs but disk " 997 "does not support TRIM\n", 998 mp->mnt_stat.f_mntonname); 999 } else { 1000 printf("WARNING: %s: TRIM flag on fs but disk does " 1001 "not confirm that it supports TRIM\n", 1002 mp->mnt_stat.f_mntonname); 1003 } 1004 if (((ump->um_flags) & UM_CANDELETE) != 0) { 1005 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1006 taskqueue_thread_enqueue, &ump->um_trim_tq); 1007 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1008 "%s trim", mp->mnt_stat.f_mntonname); 1009 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, 1010 &ump->um_trimlisthashsize); 1011 } 1012 } 1013 1014 ump->um_mountp = mp; 1015 ump->um_dev = dev; 1016 ump->um_devvp = devvp; 1017 ump->um_nindir = fs->fs_nindir; 1018 ump->um_bptrtodb = fs->fs_fsbtodb; 1019 ump->um_seqinc = fs->fs_frag; 1020 for (i = 0; i < MAXQUOTAS; i++) 1021 ump->um_quotas[i] = NULLVP; 1022 #ifdef UFS_EXTATTR 1023 ufs_extattr_uepm_init(&ump->um_extattr); 1024 #endif 1025 /* 1026 * Set FS local "last mounted on" information (NULL pad) 1027 */ 1028 bzero(fs->fs_fsmnt, MAXMNTLEN); 1029 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1030 mp->mnt_stat.f_iosize = fs->fs_bsize; 1031 1032 if (mp->mnt_flag & MNT_ROOTFS) { 1033 /* 1034 * Root mount; update timestamp in mount structure. 1035 * this will be used by the common root mount code 1036 * to update the system clock. 1037 */ 1038 mp->mnt_time = fs->fs_time; 1039 } 1040 1041 if (ronly == 0) { 1042 fs->fs_mtime = time_second; 1043 if ((fs->fs_flags & FS_DOSOFTDEP) && 1044 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1045 ffs_flushfiles(mp, FORCECLOSE, td); 1046 goto out; 1047 } 1048 if (fs->fs_snapinum[0] != 0) 1049 ffs_snapshot_mount(mp); 1050 fs->fs_fmod = 1; 1051 fs->fs_clean = 0; 1052 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1053 } 1054 /* 1055 * Initialize filesystem state information in mount struct. 1056 */ 1057 MNT_ILOCK(mp); 1058 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1059 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1060 MNT_IUNLOCK(mp); 1061 #ifdef UFS_EXTATTR 1062 #ifdef UFS_EXTATTR_AUTOSTART 1063 /* 1064 * 1065 * Auto-starting does the following: 1066 * - check for /.attribute in the fs, and extattr_start if so 1067 * - for each file in .attribute, enable that file with 1068 * an attribute of the same name. 1069 * Not clear how to report errors -- probably eat them. 1070 * This would all happen while the filesystem was busy/not 1071 * available, so would effectively be "atomic". 1072 */ 1073 (void) ufs_extattr_autostart(mp, td); 1074 #endif /* !UFS_EXTATTR_AUTOSTART */ 1075 #endif /* !UFS_EXTATTR */ 1076 return (0); 1077 out: 1078 if (fs != NULL) { 1079 free(fs->fs_csp, M_UFSMNT); 1080 free(fs, M_UFSMNT); 1081 } 1082 if (cp != NULL) { 1083 g_topology_lock(); 1084 g_vfs_close(cp); 1085 g_topology_unlock(); 1086 } 1087 if (ump) { 1088 mtx_destroy(UFS_MTX(ump)); 1089 if (mp->mnt_gjprovider != NULL) { 1090 free(mp->mnt_gjprovider, M_UFSMNT); 1091 mp->mnt_gjprovider = NULL; 1092 } 1093 free(ump, M_UFSMNT); 1094 mp->mnt_data = NULL; 1095 } 1096 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1097 dev_rel(dev); 1098 return (error); 1099 } 1100 1101 /* 1102 * A read function for use by filesystem-layer routines. 1103 */ 1104 static int 1105 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) 1106 { 1107 struct buf *bp; 1108 int error; 1109 1110 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); 1111 *bufp = malloc(size, M_UFSMNT, M_WAITOK); 1112 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, 1113 &bp)) != 0) 1114 return (error); 1115 bcopy(bp->b_data, *bufp, size); 1116 bp->b_flags |= B_INVAL | B_NOCACHE; 1117 brelse(bp); 1118 return (0); 1119 } 1120 1121 #include <sys/sysctl.h> 1122 static int bigcgs = 0; 1123 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1124 1125 /* 1126 * Sanity checks for loading old filesystem superblocks. 1127 * See ffs_oldfscompat_write below for unwound actions. 1128 * 1129 * XXX - Parts get retired eventually. 1130 * Unfortunately new bits get added. 1131 */ 1132 static void 1133 ffs_oldfscompat_read(fs, ump, sblockloc) 1134 struct fs *fs; 1135 struct ufsmount *ump; 1136 ufs2_daddr_t sblockloc; 1137 { 1138 off_t maxfilesize; 1139 1140 /* 1141 * If not yet done, update fs_flags location and value of fs_sblockloc. 1142 */ 1143 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1144 fs->fs_flags = fs->fs_old_flags; 1145 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1146 fs->fs_sblockloc = sblockloc; 1147 } 1148 /* 1149 * If not yet done, update UFS1 superblock with new wider fields. 1150 */ 1151 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1152 fs->fs_maxbsize = fs->fs_bsize; 1153 fs->fs_time = fs->fs_old_time; 1154 fs->fs_size = fs->fs_old_size; 1155 fs->fs_dsize = fs->fs_old_dsize; 1156 fs->fs_csaddr = fs->fs_old_csaddr; 1157 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1158 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1159 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1160 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1161 } 1162 if (fs->fs_magic == FS_UFS1_MAGIC && 1163 fs->fs_old_inodefmt < FS_44INODEFMT) { 1164 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1165 fs->fs_qbmask = ~fs->fs_bmask; 1166 fs->fs_qfmask = ~fs->fs_fmask; 1167 } 1168 if (fs->fs_magic == FS_UFS1_MAGIC) { 1169 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1170 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1171 if (fs->fs_maxfilesize > maxfilesize) 1172 fs->fs_maxfilesize = maxfilesize; 1173 } 1174 /* Compatibility for old filesystems */ 1175 if (fs->fs_avgfilesize <= 0) 1176 fs->fs_avgfilesize = AVFILESIZ; 1177 if (fs->fs_avgfpdir <= 0) 1178 fs->fs_avgfpdir = AFPDIR; 1179 if (bigcgs) { 1180 fs->fs_save_cgsize = fs->fs_cgsize; 1181 fs->fs_cgsize = fs->fs_bsize; 1182 } 1183 } 1184 1185 /* 1186 * Unwinding superblock updates for old filesystems. 1187 * See ffs_oldfscompat_read above for details. 1188 * 1189 * XXX - Parts get retired eventually. 1190 * Unfortunately new bits get added. 1191 */ 1192 void 1193 ffs_oldfscompat_write(fs, ump) 1194 struct fs *fs; 1195 struct ufsmount *ump; 1196 { 1197 1198 /* 1199 * Copy back UFS2 updated fields that UFS1 inspects. 1200 */ 1201 if (fs->fs_magic == FS_UFS1_MAGIC) { 1202 fs->fs_old_time = fs->fs_time; 1203 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1204 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1205 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1206 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1207 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1208 } 1209 if (bigcgs) { 1210 fs->fs_cgsize = fs->fs_save_cgsize; 1211 fs->fs_save_cgsize = 0; 1212 } 1213 } 1214 1215 /* 1216 * unmount system call 1217 */ 1218 static int 1219 ffs_unmount(mp, mntflags) 1220 struct mount *mp; 1221 int mntflags; 1222 { 1223 struct thread *td; 1224 struct ufsmount *ump = VFSTOUFS(mp); 1225 struct fs *fs; 1226 int error, flags, susp; 1227 #ifdef UFS_EXTATTR 1228 int e_restart; 1229 #endif 1230 1231 flags = 0; 1232 td = curthread; 1233 fs = ump->um_fs; 1234 susp = 0; 1235 if (mntflags & MNT_FORCE) { 1236 flags |= FORCECLOSE; 1237 susp = fs->fs_ronly == 0; 1238 } 1239 #ifdef UFS_EXTATTR 1240 if ((error = ufs_extattr_stop(mp, td))) { 1241 if (error != EOPNOTSUPP) 1242 printf("WARNING: unmount %s: ufs_extattr_stop " 1243 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1244 error); 1245 e_restart = 0; 1246 } else { 1247 ufs_extattr_uepm_destroy(&ump->um_extattr); 1248 e_restart = 1; 1249 } 1250 #endif 1251 if (susp) { 1252 error = vfs_write_suspend_umnt(mp); 1253 if (error != 0) 1254 goto fail1; 1255 } 1256 if (MOUNTEDSOFTDEP(mp)) 1257 error = softdep_flushfiles(mp, flags, td); 1258 else 1259 error = ffs_flushfiles(mp, flags, td); 1260 if (error != 0 && error != ENXIO) 1261 goto fail; 1262 1263 UFS_LOCK(ump); 1264 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1265 printf("WARNING: unmount %s: pending error: blocks %jd " 1266 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1267 fs->fs_pendinginodes); 1268 fs->fs_pendingblocks = 0; 1269 fs->fs_pendinginodes = 0; 1270 } 1271 UFS_UNLOCK(ump); 1272 if (MOUNTEDSOFTDEP(mp)) 1273 softdep_unmount(mp); 1274 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { 1275 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1276 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1277 if (error && error != ENXIO) { 1278 fs->fs_clean = 0; 1279 goto fail; 1280 } 1281 } 1282 if (susp) 1283 vfs_write_resume(mp, VR_START_WRITE); 1284 if (ump->um_trim_tq != NULL) { 1285 while (ump->um_trim_inflight != 0) 1286 pause("ufsutr", hz); 1287 taskqueue_drain_all(ump->um_trim_tq); 1288 taskqueue_free(ump->um_trim_tq); 1289 free (ump->um_trimhash, M_TRIM); 1290 } 1291 g_topology_lock(); 1292 if (ump->um_fsckpid > 0) { 1293 /* 1294 * Return to normal read-only mode. 1295 */ 1296 error = g_access(ump->um_cp, 0, -1, 0); 1297 ump->um_fsckpid = 0; 1298 } 1299 g_vfs_close(ump->um_cp); 1300 g_topology_unlock(); 1301 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1302 vrele(ump->um_devvp); 1303 dev_rel(ump->um_dev); 1304 mtx_destroy(UFS_MTX(ump)); 1305 if (mp->mnt_gjprovider != NULL) { 1306 free(mp->mnt_gjprovider, M_UFSMNT); 1307 mp->mnt_gjprovider = NULL; 1308 } 1309 free(fs->fs_csp, M_UFSMNT); 1310 free(fs, M_UFSMNT); 1311 free(ump, M_UFSMNT); 1312 mp->mnt_data = NULL; 1313 MNT_ILOCK(mp); 1314 mp->mnt_flag &= ~MNT_LOCAL; 1315 MNT_IUNLOCK(mp); 1316 if (td->td_su == mp) { 1317 td->td_su = NULL; 1318 vfs_rel(mp); 1319 } 1320 return (error); 1321 1322 fail: 1323 if (susp) 1324 vfs_write_resume(mp, VR_START_WRITE); 1325 fail1: 1326 #ifdef UFS_EXTATTR 1327 if (e_restart) { 1328 ufs_extattr_uepm_init(&ump->um_extattr); 1329 #ifdef UFS_EXTATTR_AUTOSTART 1330 (void) ufs_extattr_autostart(mp, td); 1331 #endif 1332 } 1333 #endif 1334 1335 return (error); 1336 } 1337 1338 /* 1339 * Flush out all the files in a filesystem. 1340 */ 1341 int 1342 ffs_flushfiles(mp, flags, td) 1343 struct mount *mp; 1344 int flags; 1345 struct thread *td; 1346 { 1347 struct ufsmount *ump; 1348 int qerror, error; 1349 1350 ump = VFSTOUFS(mp); 1351 qerror = 0; 1352 #ifdef QUOTA 1353 if (mp->mnt_flag & MNT_QUOTA) { 1354 int i; 1355 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1356 if (error) 1357 return (error); 1358 for (i = 0; i < MAXQUOTAS; i++) { 1359 error = quotaoff(td, mp, i); 1360 if (error != 0) { 1361 if ((flags & EARLYFLUSH) == 0) 1362 return (error); 1363 else 1364 qerror = error; 1365 } 1366 } 1367 1368 /* 1369 * Here we fall through to vflush again to ensure that 1370 * we have gotten rid of all the system vnodes, unless 1371 * quotas must not be closed. 1372 */ 1373 } 1374 #endif 1375 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1376 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1377 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1378 return (error); 1379 ffs_snapshot_unmount(mp); 1380 flags |= FORCECLOSE; 1381 /* 1382 * Here we fall through to vflush again to ensure 1383 * that we have gotten rid of all the system vnodes. 1384 */ 1385 } 1386 1387 /* 1388 * Do not close system files if quotas were not closed, to be 1389 * able to sync the remaining dquots. The freeblks softupdate 1390 * workitems might hold a reference on a dquot, preventing 1391 * quotaoff() from completing. Next round of 1392 * softdep_flushworklist() iteration should process the 1393 * blockers, allowing the next run of quotaoff() to finally 1394 * flush held dquots. 1395 * 1396 * Otherwise, flush all the files. 1397 */ 1398 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1399 return (error); 1400 1401 /* 1402 * Flush filesystem metadata. 1403 */ 1404 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1405 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1406 VOP_UNLOCK(ump->um_devvp); 1407 return (error); 1408 } 1409 1410 /* 1411 * Get filesystem statistics. 1412 */ 1413 static int 1414 ffs_statfs(mp, sbp) 1415 struct mount *mp; 1416 struct statfs *sbp; 1417 { 1418 struct ufsmount *ump; 1419 struct fs *fs; 1420 1421 ump = VFSTOUFS(mp); 1422 fs = ump->um_fs; 1423 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1424 panic("ffs_statfs"); 1425 sbp->f_version = STATFS_VERSION; 1426 sbp->f_bsize = fs->fs_fsize; 1427 sbp->f_iosize = fs->fs_bsize; 1428 sbp->f_blocks = fs->fs_dsize; 1429 UFS_LOCK(ump); 1430 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1431 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1432 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1433 dbtofsb(fs, fs->fs_pendingblocks); 1434 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; 1435 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1436 UFS_UNLOCK(ump); 1437 sbp->f_namemax = UFS_MAXNAMLEN; 1438 return (0); 1439 } 1440 1441 static bool 1442 sync_doupdate(struct inode *ip) 1443 { 1444 1445 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1446 IN_UPDATE)) != 0); 1447 } 1448 1449 static int 1450 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) 1451 { 1452 struct inode *ip; 1453 1454 /* 1455 * Flags are safe to access because ->v_data invalidation 1456 * is held off by listmtx. 1457 */ 1458 if (vp->v_type == VNON) 1459 return (false); 1460 ip = VTOI(vp); 1461 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) 1462 return (false); 1463 return (true); 1464 } 1465 1466 /* 1467 * For a lazy sync, we only care about access times, quotas and the 1468 * superblock. Other filesystem changes are already converted to 1469 * cylinder group blocks or inode blocks updates and are written to 1470 * disk by syncer. 1471 */ 1472 static int 1473 ffs_sync_lazy(mp) 1474 struct mount *mp; 1475 { 1476 struct vnode *mvp, *vp; 1477 struct inode *ip; 1478 struct thread *td; 1479 int allerror, error; 1480 1481 allerror = 0; 1482 td = curthread; 1483 if ((mp->mnt_flag & MNT_NOATIME) != 0) { 1484 #ifdef QUOTA 1485 qsync(mp); 1486 #endif 1487 goto sbupdate; 1488 } 1489 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { 1490 if (vp->v_type == VNON) { 1491 VI_UNLOCK(vp); 1492 continue; 1493 } 1494 ip = VTOI(vp); 1495 1496 /* 1497 * The IN_ACCESS flag is converted to IN_MODIFIED by 1498 * ufs_close() and ufs_getattr() by the calls to 1499 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1500 * Test also all the other timestamp flags too, to pick up 1501 * any other cases that could be missed. 1502 */ 1503 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1504 VI_UNLOCK(vp); 1505 continue; 1506 } 1507 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, 1508 td)) != 0) 1509 continue; 1510 #ifdef QUOTA 1511 qsyncvp(vp); 1512 #endif 1513 if (sync_doupdate(ip)) 1514 error = ffs_update(vp, 0); 1515 if (error != 0) 1516 allerror = error; 1517 vput(vp); 1518 } 1519 sbupdate: 1520 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1521 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1522 allerror = error; 1523 return (allerror); 1524 } 1525 1526 /* 1527 * Go through the disk queues to initiate sandbagged IO; 1528 * go through the inodes to write those that have been modified; 1529 * initiate the writing of the super block if it has been modified. 1530 * 1531 * Note: we are always called with the filesystem marked busy using 1532 * vfs_busy(). 1533 */ 1534 static int 1535 ffs_sync(mp, waitfor) 1536 struct mount *mp; 1537 int waitfor; 1538 { 1539 struct vnode *mvp, *vp, *devvp; 1540 struct thread *td; 1541 struct inode *ip; 1542 struct ufsmount *ump = VFSTOUFS(mp); 1543 struct fs *fs; 1544 int error, count, lockreq, allerror = 0; 1545 int suspend; 1546 int suspended; 1547 int secondary_writes; 1548 int secondary_accwrites; 1549 int softdep_deps; 1550 int softdep_accdeps; 1551 struct bufobj *bo; 1552 1553 suspend = 0; 1554 suspended = 0; 1555 td = curthread; 1556 fs = ump->um_fs; 1557 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) 1558 panic("%s: ffs_sync: modification on read-only filesystem", 1559 fs->fs_fsmnt); 1560 if (waitfor == MNT_LAZY) { 1561 if (!rebooting) 1562 return (ffs_sync_lazy(mp)); 1563 waitfor = MNT_NOWAIT; 1564 } 1565 1566 /* 1567 * Write back each (modified) inode. 1568 */ 1569 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1570 if (waitfor == MNT_SUSPEND) { 1571 suspend = 1; 1572 waitfor = MNT_WAIT; 1573 } 1574 if (waitfor == MNT_WAIT) 1575 lockreq = LK_EXCLUSIVE; 1576 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1577 loop: 1578 /* Grab snapshot of secondary write counts */ 1579 MNT_ILOCK(mp); 1580 secondary_writes = mp->mnt_secondary_writes; 1581 secondary_accwrites = mp->mnt_secondary_accwrites; 1582 MNT_IUNLOCK(mp); 1583 1584 /* Grab snapshot of softdep dependency counts */ 1585 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1586 1587 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1588 /* 1589 * Depend on the vnode interlock to keep things stable enough 1590 * for a quick test. Since there might be hundreds of 1591 * thousands of vnodes, we cannot afford even a subroutine 1592 * call unless there's a good chance that we have work to do. 1593 */ 1594 if (vp->v_type == VNON) { 1595 VI_UNLOCK(vp); 1596 continue; 1597 } 1598 ip = VTOI(vp); 1599 if ((ip->i_flag & 1600 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1601 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1602 VI_UNLOCK(vp); 1603 continue; 1604 } 1605 if ((error = vget(vp, lockreq, td)) != 0) { 1606 if (error == ENOENT || error == ENOLCK) { 1607 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1608 goto loop; 1609 } 1610 continue; 1611 } 1612 #ifdef QUOTA 1613 qsyncvp(vp); 1614 #endif 1615 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0) 1616 allerror = error; 1617 vput(vp); 1618 } 1619 /* 1620 * Force stale filesystem control information to be flushed. 1621 */ 1622 if (waitfor == MNT_WAIT || rebooting) { 1623 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1624 allerror = error; 1625 /* Flushed work items may create new vnodes to clean */ 1626 if (allerror == 0 && count) 1627 goto loop; 1628 } 1629 1630 devvp = ump->um_devvp; 1631 bo = &devvp->v_bufobj; 1632 BO_LOCK(bo); 1633 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1634 BO_UNLOCK(bo); 1635 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1636 error = VOP_FSYNC(devvp, waitfor, td); 1637 VOP_UNLOCK(devvp); 1638 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1639 error = ffs_sbupdate(ump, waitfor, 0); 1640 if (error != 0) 1641 allerror = error; 1642 if (allerror == 0 && waitfor == MNT_WAIT) 1643 goto loop; 1644 } else if (suspend != 0) { 1645 if (softdep_check_suspend(mp, 1646 devvp, 1647 softdep_deps, 1648 softdep_accdeps, 1649 secondary_writes, 1650 secondary_accwrites) != 0) { 1651 MNT_IUNLOCK(mp); 1652 goto loop; /* More work needed */ 1653 } 1654 mtx_assert(MNT_MTX(mp), MA_OWNED); 1655 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1656 MNT_IUNLOCK(mp); 1657 suspended = 1; 1658 } else 1659 BO_UNLOCK(bo); 1660 /* 1661 * Write back modified superblock. 1662 */ 1663 if (fs->fs_fmod != 0 && 1664 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1665 allerror = error; 1666 return (allerror); 1667 } 1668 1669 int 1670 ffs_vget(mp, ino, flags, vpp) 1671 struct mount *mp; 1672 ino_t ino; 1673 int flags; 1674 struct vnode **vpp; 1675 { 1676 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1677 } 1678 1679 int 1680 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1681 struct mount *mp; 1682 ino_t ino; 1683 int flags; 1684 struct vnode **vpp; 1685 int ffs_flags; 1686 { 1687 struct fs *fs; 1688 struct inode *ip; 1689 struct ufsmount *ump; 1690 struct buf *bp; 1691 struct vnode *vp; 1692 int error; 1693 1694 MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0); 1695 1696 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1697 if (error != 0) 1698 return (error); 1699 if (*vpp != NULL) { 1700 if ((ffs_flags & FFSV_REPLACE) == 0) 1701 return (0); 1702 vgone(*vpp); 1703 vput(*vpp); 1704 } 1705 1706 /* 1707 * We must promote to an exclusive lock for vnode creation. This 1708 * can happen if lookup is passed LOCKSHARED. 1709 */ 1710 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1711 flags &= ~LK_TYPE_MASK; 1712 flags |= LK_EXCLUSIVE; 1713 } 1714 1715 /* 1716 * We do not lock vnode creation as it is believed to be too 1717 * expensive for such rare case as simultaneous creation of vnode 1718 * for same ino by different processes. We just allow them to race 1719 * and check later to decide who wins. Let the race begin! 1720 */ 1721 1722 ump = VFSTOUFS(mp); 1723 fs = ump->um_fs; 1724 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1725 1726 /* Allocate a new vnode/inode. */ 1727 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1728 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1729 if (error) { 1730 *vpp = NULL; 1731 uma_zfree(uma_inode, ip); 1732 return (error); 1733 } 1734 /* 1735 * FFS supports recursive locking. 1736 */ 1737 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1738 VN_LOCK_AREC(vp); 1739 vp->v_data = ip; 1740 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1741 ip->i_vnode = vp; 1742 ip->i_ump = ump; 1743 ip->i_number = ino; 1744 ip->i_ea_refs = 0; 1745 ip->i_nextclustercg = -1; 1746 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; 1747 ip->i_mode = 0; /* ensure error cases below throw away vnode */ 1748 #ifdef QUOTA 1749 { 1750 int i; 1751 for (i = 0; i < MAXQUOTAS; i++) 1752 ip->i_dquot[i] = NODQUOT; 1753 } 1754 #endif 1755 1756 if (ffs_flags & FFSV_FORCEINSMQ) 1757 vp->v_vflag |= VV_FORCEINSMQ; 1758 error = insmntque(vp, mp); 1759 if (error != 0) { 1760 uma_zfree(uma_inode, ip); 1761 *vpp = NULL; 1762 return (error); 1763 } 1764 vp->v_vflag &= ~VV_FORCEINSMQ; 1765 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1766 if (error != 0) 1767 return (error); 1768 if (*vpp != NULL) { 1769 /* 1770 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) 1771 * operate on empty inode, which must not be found by 1772 * other threads until fully filled. Vnode for empty 1773 * inode must be not re-inserted on the hash by other 1774 * thread, after removal by us at the beginning. 1775 */ 1776 MPASS((ffs_flags & FFSV_REPLACE) == 0); 1777 return (0); 1778 } 1779 1780 /* Read in the disk contents for the inode, copy into the inode. */ 1781 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1782 (int)fs->fs_bsize, NOCRED, &bp); 1783 if (error) { 1784 /* 1785 * The inode does not contain anything useful, so it would 1786 * be misleading to leave it on its hash chain. With mode 1787 * still zero, it will be unlinked and returned to the free 1788 * list by vput(). 1789 */ 1790 vput(vp); 1791 *vpp = NULL; 1792 return (error); 1793 } 1794 if (I_IS_UFS1(ip)) 1795 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1796 else 1797 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1798 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { 1799 bqrelse(bp); 1800 vput(vp); 1801 *vpp = NULL; 1802 return (error); 1803 } 1804 if (DOINGSOFTDEP(vp)) 1805 softdep_load_inodeblock(ip); 1806 else 1807 ip->i_effnlink = ip->i_nlink; 1808 bqrelse(bp); 1809 1810 /* 1811 * Initialize the vnode from the inode, check for aliases. 1812 * Note that the underlying vnode may have changed. 1813 */ 1814 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, 1815 &vp); 1816 if (error) { 1817 vput(vp); 1818 *vpp = NULL; 1819 return (error); 1820 } 1821 1822 /* 1823 * Finish inode initialization. 1824 */ 1825 if (vp->v_type != VFIFO) { 1826 /* FFS supports shared locking for all files except fifos. */ 1827 VN_LOCK_ASHARE(vp); 1828 } 1829 1830 /* 1831 * Set up a generation number for this inode if it does not 1832 * already have one. This should only happen on old filesystems. 1833 */ 1834 if (ip->i_gen == 0) { 1835 while (ip->i_gen == 0) 1836 ip->i_gen = arc4random(); 1837 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1838 UFS_INODE_SET_FLAG(ip, IN_MODIFIED); 1839 DIP_SET(ip, i_gen, ip->i_gen); 1840 } 1841 } 1842 #ifdef MAC 1843 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1844 /* 1845 * If this vnode is already allocated, and we're running 1846 * multi-label, attempt to perform a label association 1847 * from the extended attributes on the inode. 1848 */ 1849 error = mac_vnode_associate_extattr(mp, vp); 1850 if (error) { 1851 /* ufs_inactive will release ip->i_devvp ref. */ 1852 vput(vp); 1853 *vpp = NULL; 1854 return (error); 1855 } 1856 } 1857 #endif 1858 1859 *vpp = vp; 1860 return (0); 1861 } 1862 1863 /* 1864 * File handle to vnode 1865 * 1866 * Have to be really careful about stale file handles: 1867 * - check that the inode number is valid 1868 * - for UFS2 check that the inode number is initialized 1869 * - call ffs_vget() to get the locked inode 1870 * - check for an unallocated inode (i_mode == 0) 1871 * - check that the given client host has export rights and return 1872 * those rights via. exflagsp and credanonp 1873 */ 1874 static int 1875 ffs_fhtovp(mp, fhp, flags, vpp) 1876 struct mount *mp; 1877 struct fid *fhp; 1878 int flags; 1879 struct vnode **vpp; 1880 { 1881 struct ufid *ufhp; 1882 struct ufsmount *ump; 1883 struct fs *fs; 1884 struct cg *cgp; 1885 struct buf *bp; 1886 ino_t ino; 1887 u_int cg; 1888 int error; 1889 1890 ufhp = (struct ufid *)fhp; 1891 ino = ufhp->ufid_ino; 1892 ump = VFSTOUFS(mp); 1893 fs = ump->um_fs; 1894 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) 1895 return (ESTALE); 1896 /* 1897 * Need to check if inode is initialized because UFS2 does lazy 1898 * initialization and nfs_fhtovp can offer arbitrary inode numbers. 1899 */ 1900 if (fs->fs_magic != FS_UFS2_MAGIC) 1901 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 1902 cg = ino_to_cg(fs, ino); 1903 if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) 1904 return (error); 1905 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { 1906 brelse(bp); 1907 return (ESTALE); 1908 } 1909 brelse(bp); 1910 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 1911 } 1912 1913 /* 1914 * Initialize the filesystem. 1915 */ 1916 static int 1917 ffs_init(vfsp) 1918 struct vfsconf *vfsp; 1919 { 1920 1921 ffs_susp_initialize(); 1922 softdep_initialize(); 1923 return (ufs_init(vfsp)); 1924 } 1925 1926 /* 1927 * Undo the work of ffs_init(). 1928 */ 1929 static int 1930 ffs_uninit(vfsp) 1931 struct vfsconf *vfsp; 1932 { 1933 int ret; 1934 1935 ret = ufs_uninit(vfsp); 1936 softdep_uninitialize(); 1937 ffs_susp_uninitialize(); 1938 return (ret); 1939 } 1940 1941 /* 1942 * Structure used to pass information from ffs_sbupdate to its 1943 * helper routine ffs_use_bwrite. 1944 */ 1945 struct devfd { 1946 struct ufsmount *ump; 1947 struct buf *sbbp; 1948 int waitfor; 1949 int suspended; 1950 int error; 1951 }; 1952 1953 /* 1954 * Write a superblock and associated information back to disk. 1955 */ 1956 int 1957 ffs_sbupdate(ump, waitfor, suspended) 1958 struct ufsmount *ump; 1959 int waitfor; 1960 int suspended; 1961 { 1962 struct fs *fs; 1963 struct buf *sbbp; 1964 struct devfd devfd; 1965 1966 fs = ump->um_fs; 1967 if (fs->fs_ronly == 1 && 1968 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1969 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) 1970 panic("ffs_sbupdate: write read-only filesystem"); 1971 /* 1972 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1973 */ 1974 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 1975 (int)fs->fs_sbsize, 0, 0, 0); 1976 /* 1977 * Initialize info needed for write function. 1978 */ 1979 devfd.ump = ump; 1980 devfd.sbbp = sbbp; 1981 devfd.waitfor = waitfor; 1982 devfd.suspended = suspended; 1983 devfd.error = 0; 1984 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); 1985 } 1986 1987 /* 1988 * Write function for use by filesystem-layer routines. 1989 */ 1990 static int 1991 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) 1992 { 1993 struct devfd *devfdp; 1994 struct ufsmount *ump; 1995 struct buf *bp; 1996 struct fs *fs; 1997 int error; 1998 1999 devfdp = devfd; 2000 ump = devfdp->ump; 2001 fs = ump->um_fs; 2002 /* 2003 * Writing the superblock summary information. 2004 */ 2005 if (loc != fs->fs_sblockloc) { 2006 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); 2007 bcopy(buf, bp->b_data, (u_int)size); 2008 if (devfdp->suspended) 2009 bp->b_flags |= B_VALIDSUSPWRT; 2010 if (devfdp->waitfor != MNT_WAIT) 2011 bawrite(bp); 2012 else if ((error = bwrite(bp)) != 0) 2013 devfdp->error = error; 2014 return (0); 2015 } 2016 /* 2017 * Writing the superblock itself. We need to do special checks for it. 2018 */ 2019 bp = devfdp->sbbp; 2020 if (devfdp->error != 0) { 2021 brelse(bp); 2022 return (devfdp->error); 2023 } 2024 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 2025 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2026 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2027 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 2028 fs->fs_sblockloc = SBLOCK_UFS1; 2029 } 2030 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 2031 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2032 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2033 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 2034 fs->fs_sblockloc = SBLOCK_UFS2; 2035 } 2036 if (MOUNTEDSOFTDEP(ump->um_mountp)) 2037 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 2038 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 2039 fs = (struct fs *)bp->b_data; 2040 ffs_oldfscompat_write(fs, ump); 2041 /* 2042 * Because we may have made changes to the superblock, we need to 2043 * recompute its check-hash. 2044 */ 2045 fs->fs_ckhash = ffs_calc_sbhash(fs); 2046 if (devfdp->suspended) 2047 bp->b_flags |= B_VALIDSUSPWRT; 2048 if (devfdp->waitfor != MNT_WAIT) 2049 bawrite(bp); 2050 else if ((error = bwrite(bp)) != 0) 2051 devfdp->error = error; 2052 return (devfdp->error); 2053 } 2054 2055 static int 2056 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 2057 int attrnamespace, const char *attrname) 2058 { 2059 2060 #ifdef UFS_EXTATTR 2061 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 2062 attrname)); 2063 #else 2064 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 2065 attrname)); 2066 #endif 2067 } 2068 2069 static void 2070 ffs_ifree(struct ufsmount *ump, struct inode *ip) 2071 { 2072 2073 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 2074 uma_zfree(uma_ufs1, ip->i_din1); 2075 else if (ip->i_din2 != NULL) 2076 uma_zfree(uma_ufs2, ip->i_din2); 2077 uma_zfree(uma_inode, ip); 2078 } 2079 2080 static int dobkgrdwrite = 1; 2081 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 2082 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 2083 2084 /* 2085 * Complete a background write started from bwrite. 2086 */ 2087 static void 2088 ffs_backgroundwritedone(struct buf *bp) 2089 { 2090 struct bufobj *bufobj; 2091 struct buf *origbp; 2092 2093 /* 2094 * Find the original buffer that we are writing. 2095 */ 2096 bufobj = bp->b_bufobj; 2097 BO_LOCK(bufobj); 2098 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2099 panic("backgroundwritedone: lost buffer"); 2100 2101 /* 2102 * We should mark the cylinder group buffer origbp as 2103 * dirty, to not loose the failed write. 2104 */ 2105 if ((bp->b_ioflags & BIO_ERROR) != 0) 2106 origbp->b_vflags |= BV_BKGRDERR; 2107 BO_UNLOCK(bufobj); 2108 /* 2109 * Process dependencies then return any unfinished ones. 2110 */ 2111 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2112 buf_complete(bp); 2113 #ifdef SOFTUPDATES 2114 if (!LIST_EMPTY(&bp->b_dep)) 2115 softdep_move_dependencies(bp, origbp); 2116 #endif 2117 /* 2118 * This buffer is marked B_NOCACHE so when it is released 2119 * by biodone it will be tossed. 2120 */ 2121 bp->b_flags |= B_NOCACHE; 2122 bp->b_flags &= ~B_CACHE; 2123 pbrelvp(bp); 2124 2125 /* 2126 * Prevent brelse() from trying to keep and re-dirtying bp on 2127 * errors. It causes b_bufobj dereference in 2128 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2129 * pbrelvp() above. 2130 */ 2131 if ((bp->b_ioflags & BIO_ERROR) != 0) 2132 bp->b_flags |= B_INVAL; 2133 bufdone(bp); 2134 BO_LOCK(bufobj); 2135 /* 2136 * Clear the BV_BKGRDINPROG flag in the original buffer 2137 * and awaken it if it is waiting for the write to complete. 2138 * If BV_BKGRDINPROG is not set in the original buffer it must 2139 * have been released and re-instantiated - which is not legal. 2140 */ 2141 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2142 ("backgroundwritedone: lost buffer2")); 2143 origbp->b_vflags &= ~BV_BKGRDINPROG; 2144 if (origbp->b_vflags & BV_BKGRDWAIT) { 2145 origbp->b_vflags &= ~BV_BKGRDWAIT; 2146 wakeup(&origbp->b_xflags); 2147 } 2148 BO_UNLOCK(bufobj); 2149 } 2150 2151 2152 /* 2153 * Write, release buffer on completion. (Done by iodone 2154 * if async). Do not bother writing anything if the buffer 2155 * is invalid. 2156 * 2157 * Note that we set B_CACHE here, indicating that buffer is 2158 * fully valid and thus cacheable. This is true even of NFS 2159 * now so we set it generally. This could be set either here 2160 * or in biodone() since the I/O is synchronous. We put it 2161 * here. 2162 */ 2163 static int 2164 ffs_bufwrite(struct buf *bp) 2165 { 2166 struct buf *newbp; 2167 struct cg *cgp; 2168 2169 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2170 if (bp->b_flags & B_INVAL) { 2171 brelse(bp); 2172 return (0); 2173 } 2174 2175 if (!BUF_ISLOCKED(bp)) 2176 panic("bufwrite: buffer is not busy???"); 2177 /* 2178 * If a background write is already in progress, delay 2179 * writing this block if it is asynchronous. Otherwise 2180 * wait for the background write to complete. 2181 */ 2182 BO_LOCK(bp->b_bufobj); 2183 if (bp->b_vflags & BV_BKGRDINPROG) { 2184 if (bp->b_flags & B_ASYNC) { 2185 BO_UNLOCK(bp->b_bufobj); 2186 bdwrite(bp); 2187 return (0); 2188 } 2189 bp->b_vflags |= BV_BKGRDWAIT; 2190 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2191 "bwrbg", 0); 2192 if (bp->b_vflags & BV_BKGRDINPROG) 2193 panic("bufwrite: still writing"); 2194 } 2195 bp->b_vflags &= ~BV_BKGRDERR; 2196 BO_UNLOCK(bp->b_bufobj); 2197 2198 /* 2199 * If this buffer is marked for background writing and we 2200 * do not have to wait for it, make a copy and write the 2201 * copy so as to leave this buffer ready for further use. 2202 * 2203 * This optimization eats a lot of memory. If we have a page 2204 * or buffer shortfall we can't do it. 2205 */ 2206 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2207 (bp->b_flags & B_ASYNC) && 2208 !vm_page_count_severe() && 2209 !buf_dirty_count_severe()) { 2210 KASSERT(bp->b_iodone == NULL, 2211 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2212 2213 /* get a new block */ 2214 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2215 if (newbp == NULL) 2216 goto normal_write; 2217 2218 KASSERT(buf_mapped(bp), ("Unmapped cg")); 2219 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2220 BO_LOCK(bp->b_bufobj); 2221 bp->b_vflags |= BV_BKGRDINPROG; 2222 BO_UNLOCK(bp->b_bufobj); 2223 newbp->b_xflags |= 2224 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; 2225 newbp->b_lblkno = bp->b_lblkno; 2226 newbp->b_blkno = bp->b_blkno; 2227 newbp->b_offset = bp->b_offset; 2228 newbp->b_iodone = ffs_backgroundwritedone; 2229 newbp->b_flags |= B_ASYNC; 2230 newbp->b_flags &= ~B_INVAL; 2231 pbgetvp(bp->b_vp, newbp); 2232 2233 #ifdef SOFTUPDATES 2234 /* 2235 * Move over the dependencies. If there are rollbacks, 2236 * leave the parent buffer dirtied as it will need to 2237 * be written again. 2238 */ 2239 if (LIST_EMPTY(&bp->b_dep) || 2240 softdep_move_dependencies(bp, newbp) == 0) 2241 bundirty(bp); 2242 #else 2243 bundirty(bp); 2244 #endif 2245 2246 /* 2247 * Initiate write on the copy, release the original. The 2248 * BKGRDINPROG flag prevents it from going away until 2249 * the background write completes. We have to recalculate 2250 * its check hash in case the buffer gets freed and then 2251 * reconstituted from the buffer cache during a later read. 2252 */ 2253 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2254 cgp = (struct cg *)bp->b_data; 2255 cgp->cg_ckhash = 0; 2256 cgp->cg_ckhash = 2257 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2258 } 2259 bqrelse(bp); 2260 bp = newbp; 2261 } else 2262 /* Mark the buffer clean */ 2263 bundirty(bp); 2264 2265 2266 /* Let the normal bufwrite do the rest for us */ 2267 normal_write: 2268 /* 2269 * If we are writing a cylinder group, update its time. 2270 */ 2271 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2272 cgp = (struct cg *)bp->b_data; 2273 cgp->cg_old_time = cgp->cg_time = time_second; 2274 } 2275 return (bufwrite(bp)); 2276 } 2277 2278 2279 static void 2280 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2281 { 2282 struct vnode *vp; 2283 struct buf *tbp; 2284 int error, nocopy; 2285 2286 vp = bo2vnode(bo); 2287 if (bp->b_iocmd == BIO_WRITE) { 2288 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2289 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2290 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2291 panic("ffs_geom_strategy: bad I/O"); 2292 nocopy = bp->b_flags & B_NOCOPY; 2293 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2294 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2295 vp->v_rdev->si_snapdata != NULL) { 2296 if ((bp->b_flags & B_CLUSTER) != 0) { 2297 runningbufwakeup(bp); 2298 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2299 b_cluster.cluster_entry) { 2300 error = ffs_copyonwrite(vp, tbp); 2301 if (error != 0 && 2302 error != EOPNOTSUPP) { 2303 bp->b_error = error; 2304 bp->b_ioflags |= BIO_ERROR; 2305 bufdone(bp); 2306 return; 2307 } 2308 } 2309 bp->b_runningbufspace = bp->b_bufsize; 2310 atomic_add_long(&runningbufspace, 2311 bp->b_runningbufspace); 2312 } else { 2313 error = ffs_copyonwrite(vp, bp); 2314 if (error != 0 && error != EOPNOTSUPP) { 2315 bp->b_error = error; 2316 bp->b_ioflags |= BIO_ERROR; 2317 bufdone(bp); 2318 return; 2319 } 2320 } 2321 } 2322 #ifdef SOFTUPDATES 2323 if ((bp->b_flags & B_CLUSTER) != 0) { 2324 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2325 b_cluster.cluster_entry) { 2326 if (!LIST_EMPTY(&tbp->b_dep)) 2327 buf_start(tbp); 2328 } 2329 } else { 2330 if (!LIST_EMPTY(&bp->b_dep)) 2331 buf_start(bp); 2332 } 2333 2334 #endif 2335 /* 2336 * Check for metadata that needs check-hashes and update them. 2337 */ 2338 switch (bp->b_xflags & BX_FSPRIV) { 2339 case BX_CYLGRP: 2340 ((struct cg *)bp->b_data)->cg_ckhash = 0; 2341 ((struct cg *)bp->b_data)->cg_ckhash = 2342 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2343 break; 2344 2345 case BX_SUPERBLOCK: 2346 case BX_INODE: 2347 case BX_INDIR: 2348 case BX_DIR: 2349 printf("Check-hash write is unimplemented!!!\n"); 2350 break; 2351 2352 case 0: 2353 break; 2354 2355 default: 2356 printf("multiple buffer types 0x%b\n", 2357 (u_int)(bp->b_xflags & BX_FSPRIV), 2358 PRINT_UFS_BUF_XFLAGS); 2359 break; 2360 } 2361 } 2362 g_vfs_strategy(bo, bp); 2363 } 2364 2365 int 2366 ffs_own_mount(const struct mount *mp) 2367 { 2368 2369 if (mp->mnt_op == &ufs_vfsops) 2370 return (1); 2371 return (0); 2372 } 2373 2374 #ifdef DDB 2375 #ifdef SOFTUPDATES 2376 2377 /* defined in ffs_softdep.c */ 2378 extern void db_print_ffs(struct ufsmount *ump); 2379 2380 DB_SHOW_COMMAND(ffs, db_show_ffs) 2381 { 2382 struct mount *mp; 2383 struct ufsmount *ump; 2384 2385 if (have_addr) { 2386 ump = VFSTOUFS((struct mount *)addr); 2387 db_print_ffs(ump); 2388 return; 2389 } 2390 2391 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2392 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2393 db_print_ffs(VFSTOUFS(mp)); 2394 } 2395 } 2396 2397 #endif /* SOFTUPDATES */ 2398 #endif /* DDB */ 2399