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/sysctl.h> 62 #include <sys/vmmeter.h> 63 64 #include <security/mac/mac_framework.h> 65 66 #include <ufs/ufs/dir.h> 67 #include <ufs/ufs/extattr.h> 68 #include <ufs/ufs/gjournal.h> 69 #include <ufs/ufs/quota.h> 70 #include <ufs/ufs/ufsmount.h> 71 #include <ufs/ufs/inode.h> 72 #include <ufs/ufs/ufs_extern.h> 73 74 #include <ufs/ffs/fs.h> 75 #include <ufs/ffs/ffs_extern.h> 76 77 #include <vm/vm.h> 78 #include <vm/uma.h> 79 #include <vm/vm_page.h> 80 81 #include <geom/geom.h> 82 #include <geom/geom_vfs.h> 83 84 #include <ddb/ddb.h> 85 86 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 87 VFS_SMR_DECLARE; 88 89 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 90 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 91 ufs2_daddr_t); 92 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 93 static int ffs_sync_lazy(struct mount *mp); 94 static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size); 95 static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size); 96 97 static vfs_init_t ffs_init; 98 static vfs_uninit_t ffs_uninit; 99 static vfs_extattrctl_t ffs_extattrctl; 100 static vfs_cmount_t ffs_cmount; 101 static vfs_unmount_t ffs_unmount; 102 static vfs_mount_t ffs_mount; 103 static vfs_statfs_t ffs_statfs; 104 static vfs_fhtovp_t ffs_fhtovp; 105 static vfs_sync_t ffs_sync; 106 107 static struct vfsops ufs_vfsops = { 108 .vfs_extattrctl = ffs_extattrctl, 109 .vfs_fhtovp = ffs_fhtovp, 110 .vfs_init = ffs_init, 111 .vfs_mount = ffs_mount, 112 .vfs_cmount = ffs_cmount, 113 .vfs_quotactl = ufs_quotactl, 114 .vfs_root = vfs_cache_root, 115 .vfs_cachedroot = ufs_root, 116 .vfs_statfs = ffs_statfs, 117 .vfs_sync = ffs_sync, 118 .vfs_uninit = ffs_uninit, 119 .vfs_unmount = ffs_unmount, 120 .vfs_vget = ffs_vget, 121 .vfs_susp_clean = process_deferred_inactive, 122 }; 123 124 VFS_SET(ufs_vfsops, ufs, 0); 125 MODULE_VERSION(ufs, 1); 126 127 static b_strategy_t ffs_geom_strategy; 128 static b_write_t ffs_bufwrite; 129 130 static struct buf_ops ffs_ops = { 131 .bop_name = "FFS", 132 .bop_write = ffs_bufwrite, 133 .bop_strategy = ffs_geom_strategy, 134 .bop_sync = bufsync, 135 #ifdef NO_FFS_SNAPSHOT 136 .bop_bdflush = bufbdflush, 137 #else 138 .bop_bdflush = ffs_bdflush, 139 #endif 140 }; 141 142 /* 143 * Note that userquota and groupquota options are not currently used 144 * by UFS/FFS code and generally mount(8) does not pass those options 145 * from userland, but they can be passed by loader(8) via 146 * vfs.root.mountfrom.options. 147 */ 148 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 149 "noclusterw", "noexec", "export", "force", "from", "groupquota", 150 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir", 151 "nosymfollow", "sync", "union", "userquota", "untrusted", NULL }; 152 153 static int ffs_enxio_enable = 1; 154 SYSCTL_DECL(_vfs_ffs); 155 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN, 156 &ffs_enxio_enable, 0, 157 "enable mapping of other disk I/O errors to ENXIO"); 158 159 /* 160 * Return buffer with the contents of block "offset" from the beginning of 161 * directory "ip". If "res" is non-zero, fill it in with a pointer to the 162 * remaining space in the directory. 163 */ 164 static int 165 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp) 166 { 167 struct inode *ip; 168 struct fs *fs; 169 struct buf *bp; 170 ufs_lbn_t lbn; 171 int bsize, error; 172 173 ip = VTOI(vp); 174 fs = ITOFS(ip); 175 lbn = lblkno(fs, offset); 176 bsize = blksize(fs, ip, lbn); 177 178 *bpp = NULL; 179 error = bread(vp, lbn, bsize, NOCRED, &bp); 180 if (error) { 181 return (error); 182 } 183 if (res) 184 *res = (char *)bp->b_data + blkoff(fs, offset); 185 *bpp = bp; 186 return (0); 187 } 188 189 /* 190 * Load up the contents of an inode and copy the appropriate pieces 191 * to the incore copy. 192 */ 193 static int 194 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino) 195 { 196 struct ufs1_dinode *dip1; 197 struct ufs2_dinode *dip2; 198 int error; 199 200 if (I_IS_UFS1(ip)) { 201 dip1 = ip->i_din1; 202 *dip1 = 203 *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino)); 204 ip->i_mode = dip1->di_mode; 205 ip->i_nlink = dip1->di_nlink; 206 ip->i_effnlink = dip1->di_nlink; 207 ip->i_size = dip1->di_size; 208 ip->i_flags = dip1->di_flags; 209 ip->i_gen = dip1->di_gen; 210 ip->i_uid = dip1->di_uid; 211 ip->i_gid = dip1->di_gid; 212 return (0); 213 } 214 dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino)); 215 if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 && 216 !ffs_fsfail_cleanup(ITOUMP(ip), error)) { 217 printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt, 218 (intmax_t)ino); 219 return (error); 220 } 221 *ip->i_din2 = *dip2; 222 dip2 = ip->i_din2; 223 ip->i_mode = dip2->di_mode; 224 ip->i_nlink = dip2->di_nlink; 225 ip->i_effnlink = dip2->di_nlink; 226 ip->i_size = dip2->di_size; 227 ip->i_flags = dip2->di_flags; 228 ip->i_gen = dip2->di_gen; 229 ip->i_uid = dip2->di_uid; 230 ip->i_gid = dip2->di_gid; 231 return (0); 232 } 233 234 /* 235 * Verify that a filesystem block number is a valid data block. 236 * This routine is only called on untrusted filesystems. 237 */ 238 static int 239 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize) 240 { 241 struct fs *fs; 242 struct ufsmount *ump; 243 ufs2_daddr_t end_daddr; 244 int cg, havemtx; 245 246 KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0, 247 ("ffs_check_blkno called on a trusted file system")); 248 ump = VFSTOUFS(mp); 249 fs = ump->um_fs; 250 cg = dtog(fs, daddr); 251 end_daddr = daddr + numfrags(fs, blksize); 252 /* 253 * Verify that the block number is a valid data block. Also check 254 * that it does not point to an inode block or a superblock. Accept 255 * blocks that are unalloacted (0) or part of snapshot metadata 256 * (BLK_NOCOPY or BLK_SNAP). 257 * 258 * Thus, the block must be in a valid range for the filesystem and 259 * either in the space before a backup superblock (except the first 260 * cylinder group where that space is used by the bootstrap code) or 261 * after the inode blocks and before the end of the cylinder group. 262 */ 263 if ((uint64_t)daddr <= BLK_SNAP || 264 ((uint64_t)end_daddr <= fs->fs_size && 265 ((cg > 0 && end_daddr <= cgsblock(fs, cg)) || 266 (daddr >= cgdmin(fs, cg) && 267 end_daddr <= cgbase(fs, cg) + fs->fs_fpg)))) 268 return (0); 269 if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0) 270 UFS_LOCK(ump); 271 if (ppsratecheck(&ump->um_last_integritymsg, 272 &ump->um_secs_integritymsg, 1)) { 273 UFS_UNLOCK(ump); 274 uprintf("\n%s: inode %jd, out-of-range indirect block " 275 "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr); 276 if (havemtx) 277 UFS_LOCK(ump); 278 } else if (!havemtx) 279 UFS_UNLOCK(ump); 280 return (EINTEGRITY); 281 } 282 283 /* 284 * Initiate a forcible unmount. 285 * Used to unmount filesystems whose underlying media has gone away. 286 */ 287 static void 288 ffs_fsfail_unmount(void *v, int pending) 289 { 290 struct fsfail_task *etp; 291 struct mount *mp; 292 293 etp = v; 294 295 /* 296 * Find our mount and get a ref on it, then try to unmount. 297 */ 298 mp = vfs_getvfs(&etp->fsid); 299 if (mp != NULL) 300 dounmount(mp, MNT_FORCE, curthread); 301 free(etp, M_UFSMNT); 302 } 303 304 /* 305 * On first ENXIO error, start a task that forcibly unmounts the filesystem. 306 * 307 * Return true if a cleanup is in progress. 308 */ 309 int 310 ffs_fsfail_cleanup(struct ufsmount *ump, int error) 311 { 312 int retval; 313 314 UFS_LOCK(ump); 315 retval = ffs_fsfail_cleanup_locked(ump, error); 316 UFS_UNLOCK(ump); 317 return (retval); 318 } 319 320 int 321 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error) 322 { 323 struct fsfail_task *etp; 324 struct task *tp; 325 326 mtx_assert(UFS_MTX(ump), MA_OWNED); 327 if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) { 328 ump->um_flags |= UM_FSFAIL_CLEANUP; 329 /* 330 * Queue an async forced unmount. 331 */ 332 etp = ump->um_fsfail_task; 333 ump->um_fsfail_task = NULL; 334 if (etp != NULL) { 335 tp = &etp->task; 336 TASK_INIT(tp, 0, ffs_fsfail_unmount, etp); 337 taskqueue_enqueue(taskqueue_thread, tp); 338 printf("UFS: forcibly unmounting %s from %s\n", 339 ump->um_mountp->mnt_stat.f_mntfromname, 340 ump->um_mountp->mnt_stat.f_mntonname); 341 } 342 } 343 return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0); 344 } 345 346 /* 347 * Wrapper used during ENXIO cleanup to allocate empty buffers when 348 * the kernel is unable to read the real one. They are needed so that 349 * the soft updates code can use them to unwind its dependencies. 350 */ 351 int 352 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno, 353 daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt, 354 struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *), 355 struct buf **bpp) 356 { 357 int error; 358 359 flags |= GB_CVTENXIO; 360 error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt, 361 cred, flags, ckhashfunc, bpp); 362 if (error != 0 && ffs_fsfail_cleanup(ump, error)) { 363 error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp); 364 KASSERT(error == 0, ("getblkx failed")); 365 vfs_bio_bzero_buf(*bpp, 0, size); 366 } 367 return (error); 368 } 369 370 static int 371 ffs_mount(struct mount *mp) 372 { 373 struct vnode *devvp, *odevvp; 374 struct thread *td; 375 struct ufsmount *ump = NULL; 376 struct fs *fs; 377 pid_t fsckpid = 0; 378 int error, error1, flags; 379 uint64_t mntorflags, saved_mnt_flag; 380 accmode_t accmode; 381 struct nameidata ndp; 382 char *fspec; 383 384 td = curthread; 385 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 386 return (EINVAL); 387 if (uma_inode == NULL) { 388 uma_inode = uma_zcreate("FFS inode", 389 sizeof(struct inode), NULL, NULL, NULL, NULL, 390 UMA_ALIGN_PTR, 0); 391 uma_ufs1 = uma_zcreate("FFS1 dinode", 392 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 393 UMA_ALIGN_PTR, 0); 394 uma_ufs2 = uma_zcreate("FFS2 dinode", 395 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 396 UMA_ALIGN_PTR, 0); 397 VFS_SMR_ZONE_SET(uma_inode); 398 } 399 400 vfs_deleteopt(mp->mnt_optnew, "groupquota"); 401 vfs_deleteopt(mp->mnt_optnew, "userquota"); 402 403 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 404 if (error) 405 return (error); 406 407 mntorflags = 0; 408 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0) 409 mntorflags |= MNT_UNTRUSTED; 410 411 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 412 mntorflags |= MNT_ACLS; 413 414 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 415 mntorflags |= MNT_SNAPSHOT; 416 /* 417 * Once we have set the MNT_SNAPSHOT flag, do not 418 * persist "snapshot" in the options list. 419 */ 420 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 421 vfs_deleteopt(mp->mnt_opt, "snapshot"); 422 } 423 424 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 && 425 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { 426 /* 427 * Once we have set the restricted PID, do not 428 * persist "fsckpid" in the options list. 429 */ 430 vfs_deleteopt(mp->mnt_optnew, "fsckpid"); 431 vfs_deleteopt(mp->mnt_opt, "fsckpid"); 432 if (mp->mnt_flag & MNT_UPDATE) { 433 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && 434 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 435 vfs_mount_error(mp, 436 "Checker enable: Must be read-only"); 437 return (EINVAL); 438 } 439 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 440 vfs_mount_error(mp, 441 "Checker enable: Must be read-only"); 442 return (EINVAL); 443 } 444 /* Set to -1 if we are done */ 445 if (fsckpid == 0) 446 fsckpid = -1; 447 } 448 449 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 450 if (mntorflags & MNT_ACLS) { 451 vfs_mount_error(mp, 452 "\"acls\" and \"nfsv4acls\" options " 453 "are mutually exclusive"); 454 return (EINVAL); 455 } 456 mntorflags |= MNT_NFS4ACLS; 457 } 458 459 MNT_ILOCK(mp); 460 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP; 461 mp->mnt_flag |= mntorflags; 462 MNT_IUNLOCK(mp); 463 /* 464 * If updating, check whether changing from read-only to 465 * read/write; if there is no device name, that's all we do. 466 */ 467 if (mp->mnt_flag & MNT_UPDATE) { 468 ump = VFSTOUFS(mp); 469 fs = ump->um_fs; 470 odevvp = ump->um_odevvp; 471 devvp = ump->um_devvp; 472 if (fsckpid == -1 && ump->um_fsckpid > 0) { 473 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || 474 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) 475 return (error); 476 g_topology_lock(); 477 /* 478 * Return to normal read-only mode. 479 */ 480 error = g_access(ump->um_cp, 0, -1, 0); 481 g_topology_unlock(); 482 ump->um_fsckpid = 0; 483 } 484 if (fs->fs_ronly == 0 && 485 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 486 /* 487 * Flush any dirty data and suspend filesystem. 488 */ 489 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 490 return (error); 491 error = vfs_write_suspend_umnt(mp); 492 if (error != 0) 493 return (error); 494 /* 495 * Check for and optionally get rid of files open 496 * for writing. 497 */ 498 flags = WRITECLOSE; 499 if (mp->mnt_flag & MNT_FORCE) 500 flags |= FORCECLOSE; 501 if (MOUNTEDSOFTDEP(mp)) { 502 error = softdep_flushfiles(mp, flags, td); 503 } else { 504 error = ffs_flushfiles(mp, flags, td); 505 } 506 if (error) { 507 vfs_write_resume(mp, 0); 508 return (error); 509 } 510 if (fs->fs_pendingblocks != 0 || 511 fs->fs_pendinginodes != 0) { 512 printf("WARNING: %s Update error: blocks %jd " 513 "files %d\n", fs->fs_fsmnt, 514 (intmax_t)fs->fs_pendingblocks, 515 fs->fs_pendinginodes); 516 fs->fs_pendingblocks = 0; 517 fs->fs_pendinginodes = 0; 518 } 519 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 520 fs->fs_clean = 1; 521 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 522 fs->fs_ronly = 0; 523 fs->fs_clean = 0; 524 vfs_write_resume(mp, 0); 525 return (error); 526 } 527 if (MOUNTEDSOFTDEP(mp)) 528 softdep_unmount(mp); 529 g_topology_lock(); 530 /* 531 * Drop our write and exclusive access. 532 */ 533 g_access(ump->um_cp, 0, -1, -1); 534 g_topology_unlock(); 535 fs->fs_ronly = 1; 536 MNT_ILOCK(mp); 537 mp->mnt_flag |= MNT_RDONLY; 538 MNT_IUNLOCK(mp); 539 /* 540 * Allow the writers to note that filesystem 541 * is ro now. 542 */ 543 vfs_write_resume(mp, 0); 544 } 545 if ((mp->mnt_flag & MNT_RELOAD) && 546 (error = ffs_reload(mp, td, 0)) != 0) 547 return (error); 548 if (fs->fs_ronly && 549 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 550 /* 551 * If we are running a checker, do not allow upgrade. 552 */ 553 if (ump->um_fsckpid > 0) { 554 vfs_mount_error(mp, 555 "Active checker, cannot upgrade to write"); 556 return (EINVAL); 557 } 558 /* 559 * If upgrade to read-write by non-root, then verify 560 * that user has necessary permissions on the device. 561 */ 562 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY); 563 error = VOP_ACCESS(odevvp, VREAD | VWRITE, 564 td->td_ucred, td); 565 if (error) 566 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 567 VOP_UNLOCK(odevvp); 568 if (error) { 569 return (error); 570 } 571 fs->fs_flags &= ~FS_UNCLEAN; 572 if (fs->fs_clean == 0) { 573 fs->fs_flags |= FS_UNCLEAN; 574 if ((mp->mnt_flag & MNT_FORCE) || 575 ((fs->fs_flags & 576 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 577 (fs->fs_flags & FS_DOSOFTDEP))) { 578 printf("WARNING: %s was not properly " 579 "dismounted\n", fs->fs_fsmnt); 580 } else { 581 vfs_mount_error(mp, 582 "R/W mount of %s denied. %s.%s", 583 fs->fs_fsmnt, 584 "Filesystem is not clean - run fsck", 585 (fs->fs_flags & FS_SUJ) == 0 ? "" : 586 " Forced mount will invalidate" 587 " journal contents"); 588 return (EPERM); 589 } 590 } 591 g_topology_lock(); 592 /* 593 * Request exclusive write access. 594 */ 595 error = g_access(ump->um_cp, 0, 1, 1); 596 g_topology_unlock(); 597 if (error) 598 return (error); 599 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 600 return (error); 601 error = vfs_write_suspend_umnt(mp); 602 if (error != 0) 603 return (error); 604 fs->fs_ronly = 0; 605 MNT_ILOCK(mp); 606 saved_mnt_flag = MNT_RDONLY; 607 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag & 608 MNT_ASYNC) != 0) 609 saved_mnt_flag |= MNT_ASYNC; 610 mp->mnt_flag &= ~saved_mnt_flag; 611 MNT_IUNLOCK(mp); 612 fs->fs_mtime = time_second; 613 /* check to see if we need to start softdep */ 614 if ((fs->fs_flags & FS_DOSOFTDEP) && 615 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 616 fs->fs_ronly = 1; 617 MNT_ILOCK(mp); 618 mp->mnt_flag |= saved_mnt_flag; 619 MNT_IUNLOCK(mp); 620 vfs_write_resume(mp, 0); 621 return (error); 622 } 623 fs->fs_clean = 0; 624 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 625 fs->fs_ronly = 1; 626 MNT_ILOCK(mp); 627 mp->mnt_flag |= saved_mnt_flag; 628 MNT_IUNLOCK(mp); 629 vfs_write_resume(mp, 0); 630 return (error); 631 } 632 if (fs->fs_snapinum[0] != 0) 633 ffs_snapshot_mount(mp); 634 vfs_write_resume(mp, 0); 635 } 636 /* 637 * Soft updates is incompatible with "async", 638 * so if we are doing softupdates stop the user 639 * from setting the async flag in an update. 640 * Softdep_mount() clears it in an initial mount 641 * or ro->rw remount. 642 */ 643 if (MOUNTEDSOFTDEP(mp)) { 644 /* XXX: Reset too late ? */ 645 MNT_ILOCK(mp); 646 mp->mnt_flag &= ~MNT_ASYNC; 647 MNT_IUNLOCK(mp); 648 } 649 /* 650 * Keep MNT_ACLS flag if it is stored in superblock. 651 */ 652 if ((fs->fs_flags & FS_ACLS) != 0) { 653 /* XXX: Set too late ? */ 654 MNT_ILOCK(mp); 655 mp->mnt_flag |= MNT_ACLS; 656 MNT_IUNLOCK(mp); 657 } 658 659 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 660 /* XXX: Set too late ? */ 661 MNT_ILOCK(mp); 662 mp->mnt_flag |= MNT_NFS4ACLS; 663 MNT_IUNLOCK(mp); 664 } 665 /* 666 * If this is a request from fsck to clean up the filesystem, 667 * then allow the specified pid to proceed. 668 */ 669 if (fsckpid > 0) { 670 if (ump->um_fsckpid != 0) { 671 vfs_mount_error(mp, 672 "Active checker already running on %s", 673 fs->fs_fsmnt); 674 return (EINVAL); 675 } 676 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 677 ("soft updates enabled on read-only file system")); 678 g_topology_lock(); 679 /* 680 * Request write access. 681 */ 682 error = g_access(ump->um_cp, 0, 1, 0); 683 g_topology_unlock(); 684 if (error) { 685 vfs_mount_error(mp, 686 "Checker activation failed on %s", 687 fs->fs_fsmnt); 688 return (error); 689 } 690 ump->um_fsckpid = fsckpid; 691 if (fs->fs_snapinum[0] != 0) 692 ffs_snapshot_mount(mp); 693 fs->fs_mtime = time_second; 694 fs->fs_fmod = 1; 695 fs->fs_clean = 0; 696 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 697 } 698 699 /* 700 * If this is a snapshot request, take the snapshot. 701 */ 702 if (mp->mnt_flag & MNT_SNAPSHOT) 703 return (ffs_snapshot(mp, fspec)); 704 705 /* 706 * Must not call namei() while owning busy ref. 707 */ 708 vfs_unbusy(mp); 709 } 710 711 /* 712 * Not an update, or updating the name: look up the name 713 * and verify that it refers to a sensible disk device. 714 */ 715 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 716 error = namei(&ndp); 717 if ((mp->mnt_flag & MNT_UPDATE) != 0) { 718 /* 719 * Unmount does not start if MNT_UPDATE is set. Mount 720 * update busies mp before setting MNT_UPDATE. We 721 * must be able to retain our busy ref succesfully, 722 * without sleep. 723 */ 724 error1 = vfs_busy(mp, MBF_NOWAIT); 725 MPASS(error1 == 0); 726 } 727 if (error != 0) 728 return (error); 729 NDFREE(&ndp, NDF_ONLY_PNBUF); 730 devvp = ndp.ni_vp; 731 if (!vn_isdisk_error(devvp, &error)) { 732 vput(devvp); 733 return (error); 734 } 735 736 /* 737 * If mount by non-root, then verify that user has necessary 738 * permissions on the device. 739 */ 740 accmode = VREAD; 741 if ((mp->mnt_flag & MNT_RDONLY) == 0) 742 accmode |= VWRITE; 743 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 744 if (error) 745 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 746 if (error) { 747 vput(devvp); 748 return (error); 749 } 750 751 if (mp->mnt_flag & MNT_UPDATE) { 752 /* 753 * Update only 754 * 755 * If it's not the same vnode, or at least the same device 756 * then it's not correct. 757 */ 758 759 if (devvp->v_rdev != ump->um_devvp->v_rdev) 760 error = EINVAL; /* needs translation */ 761 vput(devvp); 762 if (error) 763 return (error); 764 } else { 765 /* 766 * New mount 767 * 768 * We need the name for the mount point (also used for 769 * "last mounted on") copied in. If an error occurs, 770 * the mount point is discarded by the upper level code. 771 * Note that vfs_mount_alloc() populates f_mntonname for us. 772 */ 773 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 774 vrele(devvp); 775 return (error); 776 } 777 if (fsckpid > 0) { 778 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 779 ("soft updates enabled on read-only file system")); 780 ump = VFSTOUFS(mp); 781 fs = ump->um_fs; 782 g_topology_lock(); 783 /* 784 * Request write access. 785 */ 786 error = g_access(ump->um_cp, 0, 1, 0); 787 g_topology_unlock(); 788 if (error) { 789 printf("WARNING: %s: Checker activation " 790 "failed\n", fs->fs_fsmnt); 791 } else { 792 ump->um_fsckpid = fsckpid; 793 if (fs->fs_snapinum[0] != 0) 794 ffs_snapshot_mount(mp); 795 fs->fs_mtime = time_second; 796 fs->fs_clean = 0; 797 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 798 } 799 } 800 } 801 802 MNT_ILOCK(mp); 803 /* 804 * This is racy versus lookup, see ufs_fplookup_vexec for details. 805 */ 806 if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0) 807 panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp); 808 if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0) 809 mp->mnt_kern_flag |= MNTK_FPLOOKUP; 810 MNT_IUNLOCK(mp); 811 812 vfs_mountedfrom(mp, fspec); 813 return (0); 814 } 815 816 /* 817 * Compatibility with old mount system call. 818 */ 819 820 static int 821 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) 822 { 823 struct ufs_args args; 824 int error; 825 826 if (data == NULL) 827 return (EINVAL); 828 error = copyin(data, &args, sizeof args); 829 if (error) 830 return (error); 831 832 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 833 ma = mount_arg(ma, "export", &args.export, sizeof(args.export)); 834 error = kernel_mount(ma, flags); 835 836 return (error); 837 } 838 839 /* 840 * Reload all incore data for a filesystem (used after running fsck on 841 * the root filesystem and finding things to fix). If the 'force' flag 842 * is 0, the filesystem must be mounted read-only. 843 * 844 * Things to do to update the mount: 845 * 1) invalidate all cached meta-data. 846 * 2) re-read superblock from disk. 847 * 3) re-read summary information from disk. 848 * 4) invalidate all inactive vnodes. 849 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary 850 * writers, if requested. 851 * 6) invalidate all cached file data. 852 * 7) re-read inode data for all active vnodes. 853 */ 854 int 855 ffs_reload(struct mount *mp, struct thread *td, int flags) 856 { 857 struct vnode *vp, *mvp, *devvp; 858 struct inode *ip; 859 void *space; 860 struct buf *bp; 861 struct fs *fs, *newfs; 862 struct ufsmount *ump; 863 ufs2_daddr_t sblockloc; 864 int i, blks, error; 865 u_long size; 866 int32_t *lp; 867 868 ump = VFSTOUFS(mp); 869 870 MNT_ILOCK(mp); 871 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { 872 MNT_IUNLOCK(mp); 873 return (EINVAL); 874 } 875 MNT_IUNLOCK(mp); 876 877 /* 878 * Step 1: invalidate all cached meta-data. 879 */ 880 devvp = VFSTOUFS(mp)->um_devvp; 881 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 882 if (vinvalbuf(devvp, 0, 0, 0) != 0) 883 panic("ffs_reload: dirty1"); 884 VOP_UNLOCK(devvp); 885 886 /* 887 * Step 2: re-read superblock from disk. 888 */ 889 fs = VFSTOUFS(mp)->um_fs; 890 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 891 NOCRED, &bp)) != 0) 892 return (error); 893 newfs = (struct fs *)bp->b_data; 894 if ((newfs->fs_magic != FS_UFS1_MAGIC && 895 newfs->fs_magic != FS_UFS2_MAGIC) || 896 newfs->fs_bsize > MAXBSIZE || 897 newfs->fs_bsize < sizeof(struct fs)) { 898 brelse(bp); 899 return (EIO); /* XXX needs translation */ 900 } 901 /* 902 * Preserve the summary information, read-only status, and 903 * superblock location by copying these fields into our new 904 * superblock before using it to update the existing superblock. 905 */ 906 newfs->fs_si = fs->fs_si; 907 newfs->fs_ronly = fs->fs_ronly; 908 sblockloc = fs->fs_sblockloc; 909 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 910 brelse(bp); 911 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 912 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 913 UFS_LOCK(ump); 914 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 915 printf("WARNING: %s: reload pending error: blocks %jd " 916 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 917 fs->fs_pendinginodes); 918 fs->fs_pendingblocks = 0; 919 fs->fs_pendinginodes = 0; 920 } 921 UFS_UNLOCK(ump); 922 923 /* 924 * Step 3: re-read summary information from disk. 925 */ 926 size = fs->fs_cssize; 927 blks = howmany(size, fs->fs_fsize); 928 if (fs->fs_contigsumsize > 0) 929 size += fs->fs_ncg * sizeof(int32_t); 930 size += fs->fs_ncg * sizeof(u_int8_t); 931 free(fs->fs_csp, M_UFSMNT); 932 space = malloc(size, M_UFSMNT, M_WAITOK); 933 fs->fs_csp = space; 934 for (i = 0; i < blks; i += fs->fs_frag) { 935 size = fs->fs_bsize; 936 if (i + fs->fs_frag > blks) 937 size = (blks - i) * fs->fs_fsize; 938 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 939 NOCRED, &bp); 940 if (error) 941 return (error); 942 bcopy(bp->b_data, space, (u_int)size); 943 space = (char *)space + size; 944 brelse(bp); 945 } 946 /* 947 * We no longer know anything about clusters per cylinder group. 948 */ 949 if (fs->fs_contigsumsize > 0) { 950 fs->fs_maxcluster = lp = space; 951 for (i = 0; i < fs->fs_ncg; i++) 952 *lp++ = fs->fs_contigsumsize; 953 space = lp; 954 } 955 size = fs->fs_ncg * sizeof(u_int8_t); 956 fs->fs_contigdirs = (u_int8_t *)space; 957 bzero(fs->fs_contigdirs, size); 958 if ((flags & FFSR_UNSUSPEND) != 0) { 959 MNT_ILOCK(mp); 960 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 961 wakeup(&mp->mnt_flag); 962 MNT_IUNLOCK(mp); 963 } 964 965 loop: 966 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 967 /* 968 * Skip syncer vnode. 969 */ 970 if (vp->v_type == VNON) { 971 VI_UNLOCK(vp); 972 continue; 973 } 974 /* 975 * Step 4: invalidate all cached file data. 976 */ 977 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 978 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 979 goto loop; 980 } 981 if (vinvalbuf(vp, 0, 0, 0)) 982 panic("ffs_reload: dirty2"); 983 /* 984 * Step 5: re-read inode data for all active vnodes. 985 */ 986 ip = VTOI(vp); 987 error = 988 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 989 (int)fs->fs_bsize, NOCRED, &bp); 990 if (error) { 991 vput(vp); 992 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 993 return (error); 994 } 995 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) { 996 brelse(bp); 997 vput(vp); 998 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 999 return (error); 1000 } 1001 ip->i_effnlink = ip->i_nlink; 1002 brelse(bp); 1003 vput(vp); 1004 } 1005 return (0); 1006 } 1007 1008 /* 1009 * Common code for mount and mountroot 1010 */ 1011 static int 1012 ffs_mountfs(odevvp, mp, td) 1013 struct vnode *odevvp; 1014 struct mount *mp; 1015 struct thread *td; 1016 { 1017 struct ufsmount *ump; 1018 struct fs *fs; 1019 struct cdev *dev; 1020 int error, i, len, ronly; 1021 struct ucred *cred; 1022 struct g_consumer *cp; 1023 struct mount *nmp; 1024 struct vnode *devvp; 1025 struct fsfail_task *etp; 1026 int candelete, canspeedup; 1027 off_t loc; 1028 1029 fs = NULL; 1030 ump = NULL; 1031 cred = td ? td->td_ucred : NOCRED; 1032 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 1033 1034 devvp = mntfs_allocvp(mp, odevvp); 1035 VOP_UNLOCK(odevvp); 1036 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); 1037 dev = devvp->v_rdev; 1038 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, 1039 (uintptr_t)mp) == 0) { 1040 mntfs_freevp(devvp); 1041 return (EBUSY); 1042 } 1043 g_topology_lock(); 1044 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 1045 g_topology_unlock(); 1046 if (error != 0) { 1047 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1048 mntfs_freevp(devvp); 1049 return (error); 1050 } 1051 dev_ref(dev); 1052 devvp->v_bufobj.bo_ops = &ffs_ops; 1053 BO_LOCK(&odevvp->v_bufobj); 1054 odevvp->v_bufobj.bo_flag |= BO_NOBUFS; 1055 BO_UNLOCK(&odevvp->v_bufobj); 1056 if (dev->si_iosize_max != 0) 1057 mp->mnt_iosize_max = dev->si_iosize_max; 1058 if (mp->mnt_iosize_max > maxphys) 1059 mp->mnt_iosize_max = maxphys; 1060 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 1061 error = EINVAL; 1062 vfs_mount_error(mp, 1063 "Invalid sectorsize %d for superblock size %d", 1064 cp->provider->sectorsize, SBLOCKSIZE); 1065 goto out; 1066 } 1067 /* fetch the superblock and summary information */ 1068 loc = STDSB; 1069 if ((mp->mnt_flag & MNT_ROOTFS) != 0) 1070 loc = STDSB_NOHASHFAIL; 1071 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0) 1072 goto out; 1073 fs->fs_flags &= ~FS_UNCLEAN; 1074 if (fs->fs_clean == 0) { 1075 fs->fs_flags |= FS_UNCLEAN; 1076 if (ronly || (mp->mnt_flag & MNT_FORCE) || 1077 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 1078 (fs->fs_flags & FS_DOSOFTDEP))) { 1079 printf("WARNING: %s was not properly dismounted\n", 1080 fs->fs_fsmnt); 1081 } else { 1082 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 1083 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 1084 (fs->fs_flags & FS_SUJ) == 0 ? "" : 1085 " Forced mount will invalidate journal contents"); 1086 error = EPERM; 1087 goto out; 1088 } 1089 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 1090 (mp->mnt_flag & MNT_FORCE)) { 1091 printf("WARNING: %s: lost blocks %jd files %d\n", 1092 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1093 fs->fs_pendinginodes); 1094 fs->fs_pendingblocks = 0; 1095 fs->fs_pendinginodes = 0; 1096 } 1097 } 1098 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1099 printf("WARNING: %s: mount pending error: blocks %jd " 1100 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1101 fs->fs_pendinginodes); 1102 fs->fs_pendingblocks = 0; 1103 fs->fs_pendinginodes = 0; 1104 } 1105 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 1106 #ifdef UFS_GJOURNAL 1107 /* 1108 * Get journal provider name. 1109 */ 1110 len = 1024; 1111 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); 1112 if (g_io_getattr("GJOURNAL::provider", cp, &len, 1113 mp->mnt_gjprovider) == 0) { 1114 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, 1115 M_UFSMNT, M_WAITOK); 1116 MNT_ILOCK(mp); 1117 mp->mnt_flag |= MNT_GJOURNAL; 1118 MNT_IUNLOCK(mp); 1119 } else { 1120 printf("WARNING: %s: GJOURNAL flag on fs " 1121 "but no gjournal provider below\n", 1122 mp->mnt_stat.f_mntonname); 1123 free(mp->mnt_gjprovider, M_UFSMNT); 1124 mp->mnt_gjprovider = NULL; 1125 } 1126 #else 1127 printf("WARNING: %s: GJOURNAL flag on fs but no " 1128 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 1129 #endif 1130 } else { 1131 mp->mnt_gjprovider = NULL; 1132 } 1133 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1134 ump->um_cp = cp; 1135 ump->um_bo = &devvp->v_bufobj; 1136 ump->um_fs = fs; 1137 if (fs->fs_magic == FS_UFS1_MAGIC) { 1138 ump->um_fstype = UFS1; 1139 ump->um_balloc = ffs_balloc_ufs1; 1140 } else { 1141 ump->um_fstype = UFS2; 1142 ump->um_balloc = ffs_balloc_ufs2; 1143 } 1144 ump->um_blkatoff = ffs_blkatoff; 1145 ump->um_truncate = ffs_truncate; 1146 ump->um_update = ffs_update; 1147 ump->um_valloc = ffs_valloc; 1148 ump->um_vfree = ffs_vfree; 1149 ump->um_ifree = ffs_ifree; 1150 ump->um_rdonly = ffs_rdonly; 1151 ump->um_snapgone = ffs_snapgone; 1152 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) 1153 ump->um_check_blkno = ffs_check_blkno; 1154 else 1155 ump->um_check_blkno = NULL; 1156 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 1157 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); 1158 fs->fs_ronly = ronly; 1159 fs->fs_active = NULL; 1160 mp->mnt_data = ump; 1161 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 1162 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 1163 nmp = NULL; 1164 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 1165 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 1166 if (nmp) 1167 vfs_rel(nmp); 1168 vfs_getnewfsid(mp); 1169 } 1170 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 1171 MNT_ILOCK(mp); 1172 mp->mnt_flag |= MNT_LOCAL; 1173 MNT_IUNLOCK(mp); 1174 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 1175 #ifdef MAC 1176 MNT_ILOCK(mp); 1177 mp->mnt_flag |= MNT_MULTILABEL; 1178 MNT_IUNLOCK(mp); 1179 #else 1180 printf("WARNING: %s: multilabel flag on fs but " 1181 "no MAC support\n", mp->mnt_stat.f_mntonname); 1182 #endif 1183 } 1184 if ((fs->fs_flags & FS_ACLS) != 0) { 1185 #ifdef UFS_ACL 1186 MNT_ILOCK(mp); 1187 1188 if (mp->mnt_flag & MNT_NFS4ACLS) 1189 printf("WARNING: %s: ACLs flag on fs conflicts with " 1190 "\"nfsv4acls\" mount option; option ignored\n", 1191 mp->mnt_stat.f_mntonname); 1192 mp->mnt_flag &= ~MNT_NFS4ACLS; 1193 mp->mnt_flag |= MNT_ACLS; 1194 1195 MNT_IUNLOCK(mp); 1196 #else 1197 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 1198 mp->mnt_stat.f_mntonname); 1199 #endif 1200 } 1201 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 1202 #ifdef UFS_ACL 1203 MNT_ILOCK(mp); 1204 1205 if (mp->mnt_flag & MNT_ACLS) 1206 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 1207 "with \"acls\" mount option; option ignored\n", 1208 mp->mnt_stat.f_mntonname); 1209 mp->mnt_flag &= ~MNT_ACLS; 1210 mp->mnt_flag |= MNT_NFS4ACLS; 1211 1212 MNT_IUNLOCK(mp); 1213 #else 1214 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 1215 "ACLs support\n", mp->mnt_stat.f_mntonname); 1216 #endif 1217 } 1218 if ((fs->fs_flags & FS_TRIM) != 0) { 1219 len = sizeof(int); 1220 if (g_io_getattr("GEOM::candelete", cp, &len, 1221 &candelete) == 0) { 1222 if (candelete) 1223 ump->um_flags |= UM_CANDELETE; 1224 else 1225 printf("WARNING: %s: TRIM flag on fs but disk " 1226 "does not support TRIM\n", 1227 mp->mnt_stat.f_mntonname); 1228 } else { 1229 printf("WARNING: %s: TRIM flag on fs but disk does " 1230 "not confirm that it supports TRIM\n", 1231 mp->mnt_stat.f_mntonname); 1232 } 1233 if (((ump->um_flags) & UM_CANDELETE) != 0) { 1234 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1235 taskqueue_thread_enqueue, &ump->um_trim_tq); 1236 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1237 "%s trim", mp->mnt_stat.f_mntonname); 1238 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, 1239 &ump->um_trimlisthashsize); 1240 } 1241 } 1242 1243 len = sizeof(int); 1244 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) { 1245 if (canspeedup) 1246 ump->um_flags |= UM_CANSPEEDUP; 1247 } 1248 1249 ump->um_mountp = mp; 1250 ump->um_dev = dev; 1251 ump->um_devvp = devvp; 1252 ump->um_odevvp = odevvp; 1253 ump->um_nindir = fs->fs_nindir; 1254 ump->um_bptrtodb = fs->fs_fsbtodb; 1255 ump->um_seqinc = fs->fs_frag; 1256 for (i = 0; i < MAXQUOTAS; i++) 1257 ump->um_quotas[i] = NULLVP; 1258 #ifdef UFS_EXTATTR 1259 ufs_extattr_uepm_init(&ump->um_extattr); 1260 #endif 1261 /* 1262 * Set FS local "last mounted on" information (NULL pad) 1263 */ 1264 bzero(fs->fs_fsmnt, MAXMNTLEN); 1265 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1266 mp->mnt_stat.f_iosize = fs->fs_bsize; 1267 1268 if (mp->mnt_flag & MNT_ROOTFS) { 1269 /* 1270 * Root mount; update timestamp in mount structure. 1271 * this will be used by the common root mount code 1272 * to update the system clock. 1273 */ 1274 mp->mnt_time = fs->fs_time; 1275 } 1276 1277 if (ronly == 0) { 1278 fs->fs_mtime = time_second; 1279 if ((fs->fs_flags & FS_DOSOFTDEP) && 1280 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1281 ffs_flushfiles(mp, FORCECLOSE, td); 1282 goto out; 1283 } 1284 if (fs->fs_snapinum[0] != 0) 1285 ffs_snapshot_mount(mp); 1286 fs->fs_fmod = 1; 1287 fs->fs_clean = 0; 1288 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1289 } 1290 /* 1291 * Initialize filesystem state information in mount struct. 1292 */ 1293 MNT_ILOCK(mp); 1294 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1295 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1296 MNT_IUNLOCK(mp); 1297 #ifdef UFS_EXTATTR 1298 #ifdef UFS_EXTATTR_AUTOSTART 1299 /* 1300 * 1301 * Auto-starting does the following: 1302 * - check for /.attribute in the fs, and extattr_start if so 1303 * - for each file in .attribute, enable that file with 1304 * an attribute of the same name. 1305 * Not clear how to report errors -- probably eat them. 1306 * This would all happen while the filesystem was busy/not 1307 * available, so would effectively be "atomic". 1308 */ 1309 (void) ufs_extattr_autostart(mp, td); 1310 #endif /* !UFS_EXTATTR_AUTOSTART */ 1311 #endif /* !UFS_EXTATTR */ 1312 etp = malloc(sizeof *ump->um_fsfail_task, M_UFSMNT, M_WAITOK | M_ZERO); 1313 etp->fsid = mp->mnt_stat.f_fsid; 1314 ump->um_fsfail_task = etp; 1315 return (0); 1316 out: 1317 if (fs != NULL) { 1318 free(fs->fs_csp, M_UFSMNT); 1319 free(fs->fs_si, M_UFSMNT); 1320 free(fs, M_UFSMNT); 1321 } 1322 if (cp != NULL) { 1323 g_topology_lock(); 1324 g_vfs_close(cp); 1325 g_topology_unlock(); 1326 } 1327 if (ump) { 1328 mtx_destroy(UFS_MTX(ump)); 1329 if (mp->mnt_gjprovider != NULL) { 1330 free(mp->mnt_gjprovider, M_UFSMNT); 1331 mp->mnt_gjprovider = NULL; 1332 } 1333 free(ump, M_UFSMNT); 1334 mp->mnt_data = NULL; 1335 } 1336 BO_LOCK(&odevvp->v_bufobj); 1337 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1338 BO_UNLOCK(&odevvp->v_bufobj); 1339 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1340 mntfs_freevp(devvp); 1341 dev_rel(dev); 1342 return (error); 1343 } 1344 1345 /* 1346 * A read function for use by filesystem-layer routines. 1347 */ 1348 static int 1349 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) 1350 { 1351 struct buf *bp; 1352 int error; 1353 1354 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); 1355 *bufp = malloc(size, M_UFSMNT, M_WAITOK); 1356 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, 1357 &bp)) != 0) 1358 return (error); 1359 bcopy(bp->b_data, *bufp, size); 1360 bp->b_flags |= B_INVAL | B_NOCACHE; 1361 brelse(bp); 1362 return (0); 1363 } 1364 1365 static int bigcgs = 0; 1366 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1367 1368 /* 1369 * Sanity checks for loading old filesystem superblocks. 1370 * See ffs_oldfscompat_write below for unwound actions. 1371 * 1372 * XXX - Parts get retired eventually. 1373 * Unfortunately new bits get added. 1374 */ 1375 static void 1376 ffs_oldfscompat_read(fs, ump, sblockloc) 1377 struct fs *fs; 1378 struct ufsmount *ump; 1379 ufs2_daddr_t sblockloc; 1380 { 1381 off_t maxfilesize; 1382 1383 /* 1384 * If not yet done, update fs_flags location and value of fs_sblockloc. 1385 */ 1386 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1387 fs->fs_flags = fs->fs_old_flags; 1388 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1389 fs->fs_sblockloc = sblockloc; 1390 } 1391 /* 1392 * If not yet done, update UFS1 superblock with new wider fields. 1393 */ 1394 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1395 fs->fs_maxbsize = fs->fs_bsize; 1396 fs->fs_time = fs->fs_old_time; 1397 fs->fs_size = fs->fs_old_size; 1398 fs->fs_dsize = fs->fs_old_dsize; 1399 fs->fs_csaddr = fs->fs_old_csaddr; 1400 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1401 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1402 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1403 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1404 } 1405 if (fs->fs_magic == FS_UFS1_MAGIC && 1406 fs->fs_old_inodefmt < FS_44INODEFMT) { 1407 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1408 fs->fs_qbmask = ~fs->fs_bmask; 1409 fs->fs_qfmask = ~fs->fs_fmask; 1410 } 1411 if (fs->fs_magic == FS_UFS1_MAGIC) { 1412 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1413 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1414 if (fs->fs_maxfilesize > maxfilesize) 1415 fs->fs_maxfilesize = maxfilesize; 1416 } 1417 /* Compatibility for old filesystems */ 1418 if (fs->fs_avgfilesize <= 0) 1419 fs->fs_avgfilesize = AVFILESIZ; 1420 if (fs->fs_avgfpdir <= 0) 1421 fs->fs_avgfpdir = AFPDIR; 1422 if (bigcgs) { 1423 fs->fs_save_cgsize = fs->fs_cgsize; 1424 fs->fs_cgsize = fs->fs_bsize; 1425 } 1426 } 1427 1428 /* 1429 * Unwinding superblock updates for old filesystems. 1430 * See ffs_oldfscompat_read above for details. 1431 * 1432 * XXX - Parts get retired eventually. 1433 * Unfortunately new bits get added. 1434 */ 1435 void 1436 ffs_oldfscompat_write(fs, ump) 1437 struct fs *fs; 1438 struct ufsmount *ump; 1439 { 1440 1441 /* 1442 * Copy back UFS2 updated fields that UFS1 inspects. 1443 */ 1444 if (fs->fs_magic == FS_UFS1_MAGIC) { 1445 fs->fs_old_time = fs->fs_time; 1446 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1447 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1448 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1449 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1450 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1451 } 1452 if (bigcgs) { 1453 fs->fs_cgsize = fs->fs_save_cgsize; 1454 fs->fs_save_cgsize = 0; 1455 } 1456 } 1457 1458 /* 1459 * unmount system call 1460 */ 1461 static int 1462 ffs_unmount(mp, mntflags) 1463 struct mount *mp; 1464 int mntflags; 1465 { 1466 struct thread *td; 1467 struct ufsmount *ump = VFSTOUFS(mp); 1468 struct fs *fs; 1469 int error, flags, susp; 1470 #ifdef UFS_EXTATTR 1471 int e_restart; 1472 #endif 1473 1474 flags = 0; 1475 td = curthread; 1476 fs = ump->um_fs; 1477 if (mntflags & MNT_FORCE) 1478 flags |= FORCECLOSE; 1479 susp = fs->fs_ronly == 0; 1480 #ifdef UFS_EXTATTR 1481 if ((error = ufs_extattr_stop(mp, td))) { 1482 if (error != EOPNOTSUPP) 1483 printf("WARNING: unmount %s: ufs_extattr_stop " 1484 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1485 error); 1486 e_restart = 0; 1487 } else { 1488 ufs_extattr_uepm_destroy(&ump->um_extattr); 1489 e_restart = 1; 1490 } 1491 #endif 1492 if (susp) { 1493 error = vfs_write_suspend_umnt(mp); 1494 if (error != 0) 1495 goto fail1; 1496 } 1497 if (MOUNTEDSOFTDEP(mp)) 1498 error = softdep_flushfiles(mp, flags, td); 1499 else 1500 error = ffs_flushfiles(mp, flags, td); 1501 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) 1502 goto fail; 1503 1504 UFS_LOCK(ump); 1505 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1506 printf("WARNING: unmount %s: pending error: blocks %jd " 1507 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1508 fs->fs_pendinginodes); 1509 fs->fs_pendingblocks = 0; 1510 fs->fs_pendinginodes = 0; 1511 } 1512 UFS_UNLOCK(ump); 1513 if (MOUNTEDSOFTDEP(mp)) 1514 softdep_unmount(mp); 1515 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { 1516 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1517 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1518 if (ffs_fsfail_cleanup(ump, error)) 1519 error = 0; 1520 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) { 1521 fs->fs_clean = 0; 1522 goto fail; 1523 } 1524 } 1525 if (susp) 1526 vfs_write_resume(mp, VR_START_WRITE); 1527 if (ump->um_trim_tq != NULL) { 1528 while (ump->um_trim_inflight != 0) 1529 pause("ufsutr", hz); 1530 taskqueue_drain_all(ump->um_trim_tq); 1531 taskqueue_free(ump->um_trim_tq); 1532 free (ump->um_trimhash, M_TRIM); 1533 } 1534 g_topology_lock(); 1535 if (ump->um_fsckpid > 0) { 1536 /* 1537 * Return to normal read-only mode. 1538 */ 1539 error = g_access(ump->um_cp, 0, -1, 0); 1540 ump->um_fsckpid = 0; 1541 } 1542 g_vfs_close(ump->um_cp); 1543 g_topology_unlock(); 1544 BO_LOCK(&ump->um_odevvp->v_bufobj); 1545 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1546 BO_UNLOCK(&ump->um_odevvp->v_bufobj); 1547 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1548 mntfs_freevp(ump->um_devvp); 1549 vrele(ump->um_odevvp); 1550 dev_rel(ump->um_dev); 1551 mtx_destroy(UFS_MTX(ump)); 1552 if (mp->mnt_gjprovider != NULL) { 1553 free(mp->mnt_gjprovider, M_UFSMNT); 1554 mp->mnt_gjprovider = NULL; 1555 } 1556 free(fs->fs_csp, M_UFSMNT); 1557 free(fs->fs_si, M_UFSMNT); 1558 free(fs, M_UFSMNT); 1559 if (ump->um_fsfail_task != NULL) 1560 free(ump->um_fsfail_task, M_UFSMNT); 1561 free(ump, M_UFSMNT); 1562 mp->mnt_data = NULL; 1563 MNT_ILOCK(mp); 1564 mp->mnt_flag &= ~MNT_LOCAL; 1565 MNT_IUNLOCK(mp); 1566 if (td->td_su == mp) { 1567 td->td_su = NULL; 1568 vfs_rel(mp); 1569 } 1570 return (error); 1571 1572 fail: 1573 if (susp) 1574 vfs_write_resume(mp, VR_START_WRITE); 1575 fail1: 1576 #ifdef UFS_EXTATTR 1577 if (e_restart) { 1578 ufs_extattr_uepm_init(&ump->um_extattr); 1579 #ifdef UFS_EXTATTR_AUTOSTART 1580 (void) ufs_extattr_autostart(mp, td); 1581 #endif 1582 } 1583 #endif 1584 1585 return (error); 1586 } 1587 1588 /* 1589 * Flush out all the files in a filesystem. 1590 */ 1591 int 1592 ffs_flushfiles(mp, flags, td) 1593 struct mount *mp; 1594 int flags; 1595 struct thread *td; 1596 { 1597 struct ufsmount *ump; 1598 int qerror, error; 1599 1600 ump = VFSTOUFS(mp); 1601 qerror = 0; 1602 #ifdef QUOTA 1603 if (mp->mnt_flag & MNT_QUOTA) { 1604 int i; 1605 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1606 if (error) 1607 return (error); 1608 for (i = 0; i < MAXQUOTAS; i++) { 1609 error = quotaoff(td, mp, i); 1610 if (error != 0) { 1611 if ((flags & EARLYFLUSH) == 0) 1612 return (error); 1613 else 1614 qerror = error; 1615 } 1616 } 1617 1618 /* 1619 * Here we fall through to vflush again to ensure that 1620 * we have gotten rid of all the system vnodes, unless 1621 * quotas must not be closed. 1622 */ 1623 } 1624 #endif 1625 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1626 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1627 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1628 return (error); 1629 ffs_snapshot_unmount(mp); 1630 flags |= FORCECLOSE; 1631 /* 1632 * Here we fall through to vflush again to ensure 1633 * that we have gotten rid of all the system vnodes. 1634 */ 1635 } 1636 1637 /* 1638 * Do not close system files if quotas were not closed, to be 1639 * able to sync the remaining dquots. The freeblks softupdate 1640 * workitems might hold a reference on a dquot, preventing 1641 * quotaoff() from completing. Next round of 1642 * softdep_flushworklist() iteration should process the 1643 * blockers, allowing the next run of quotaoff() to finally 1644 * flush held dquots. 1645 * 1646 * Otherwise, flush all the files. 1647 */ 1648 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1649 return (error); 1650 1651 /* 1652 * Flush filesystem metadata. 1653 */ 1654 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1655 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1656 VOP_UNLOCK(ump->um_devvp); 1657 return (error); 1658 } 1659 1660 /* 1661 * Get filesystem statistics. 1662 */ 1663 static int 1664 ffs_statfs(mp, sbp) 1665 struct mount *mp; 1666 struct statfs *sbp; 1667 { 1668 struct ufsmount *ump; 1669 struct fs *fs; 1670 1671 ump = VFSTOUFS(mp); 1672 fs = ump->um_fs; 1673 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1674 panic("ffs_statfs"); 1675 sbp->f_version = STATFS_VERSION; 1676 sbp->f_bsize = fs->fs_fsize; 1677 sbp->f_iosize = fs->fs_bsize; 1678 sbp->f_blocks = fs->fs_dsize; 1679 UFS_LOCK(ump); 1680 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1681 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1682 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1683 dbtofsb(fs, fs->fs_pendingblocks); 1684 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; 1685 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1686 UFS_UNLOCK(ump); 1687 sbp->f_namemax = UFS_MAXNAMLEN; 1688 return (0); 1689 } 1690 1691 static bool 1692 sync_doupdate(struct inode *ip) 1693 { 1694 1695 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1696 IN_UPDATE)) != 0); 1697 } 1698 1699 static int 1700 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) 1701 { 1702 struct inode *ip; 1703 1704 /* 1705 * Flags are safe to access because ->v_data invalidation 1706 * is held off by listmtx. 1707 */ 1708 if (vp->v_type == VNON) 1709 return (false); 1710 ip = VTOI(vp); 1711 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) 1712 return (false); 1713 return (true); 1714 } 1715 1716 /* 1717 * For a lazy sync, we only care about access times, quotas and the 1718 * superblock. Other filesystem changes are already converted to 1719 * cylinder group blocks or inode blocks updates and are written to 1720 * disk by syncer. 1721 */ 1722 static int 1723 ffs_sync_lazy(mp) 1724 struct mount *mp; 1725 { 1726 struct vnode *mvp, *vp; 1727 struct inode *ip; 1728 struct thread *td; 1729 int allerror, error; 1730 1731 allerror = 0; 1732 td = curthread; 1733 if ((mp->mnt_flag & MNT_NOATIME) != 0) { 1734 #ifdef QUOTA 1735 qsync(mp); 1736 #endif 1737 goto sbupdate; 1738 } 1739 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { 1740 if (vp->v_type == VNON) { 1741 VI_UNLOCK(vp); 1742 continue; 1743 } 1744 ip = VTOI(vp); 1745 1746 /* 1747 * The IN_ACCESS flag is converted to IN_MODIFIED by 1748 * ufs_close() and ufs_getattr() by the calls to 1749 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1750 * Test also all the other timestamp flags too, to pick up 1751 * any other cases that could be missed. 1752 */ 1753 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1754 VI_UNLOCK(vp); 1755 continue; 1756 } 1757 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0) 1758 continue; 1759 #ifdef QUOTA 1760 qsyncvp(vp); 1761 #endif 1762 if (sync_doupdate(ip)) 1763 error = ffs_update(vp, 0); 1764 if (error != 0) 1765 allerror = error; 1766 vput(vp); 1767 } 1768 sbupdate: 1769 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1770 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1771 allerror = error; 1772 return (allerror); 1773 } 1774 1775 /* 1776 * Go through the disk queues to initiate sandbagged IO; 1777 * go through the inodes to write those that have been modified; 1778 * initiate the writing of the super block if it has been modified. 1779 * 1780 * Note: we are always called with the filesystem marked busy using 1781 * vfs_busy(). 1782 */ 1783 static int 1784 ffs_sync(mp, waitfor) 1785 struct mount *mp; 1786 int waitfor; 1787 { 1788 struct vnode *mvp, *vp, *devvp; 1789 struct thread *td; 1790 struct inode *ip; 1791 struct ufsmount *ump = VFSTOUFS(mp); 1792 struct fs *fs; 1793 int error, count, lockreq, allerror = 0; 1794 int suspend; 1795 int suspended; 1796 int secondary_writes; 1797 int secondary_accwrites; 1798 int softdep_deps; 1799 int softdep_accdeps; 1800 struct bufobj *bo; 1801 1802 suspend = 0; 1803 suspended = 0; 1804 td = curthread; 1805 fs = ump->um_fs; 1806 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) 1807 panic("%s: ffs_sync: modification on read-only filesystem", 1808 fs->fs_fsmnt); 1809 if (waitfor == MNT_LAZY) { 1810 if (!rebooting) 1811 return (ffs_sync_lazy(mp)); 1812 waitfor = MNT_NOWAIT; 1813 } 1814 1815 /* 1816 * Write back each (modified) inode. 1817 */ 1818 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1819 if (waitfor == MNT_SUSPEND) { 1820 suspend = 1; 1821 waitfor = MNT_WAIT; 1822 } 1823 if (waitfor == MNT_WAIT) 1824 lockreq = LK_EXCLUSIVE; 1825 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1826 loop: 1827 /* Grab snapshot of secondary write counts */ 1828 MNT_ILOCK(mp); 1829 secondary_writes = mp->mnt_secondary_writes; 1830 secondary_accwrites = mp->mnt_secondary_accwrites; 1831 MNT_IUNLOCK(mp); 1832 1833 /* Grab snapshot of softdep dependency counts */ 1834 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1835 1836 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1837 /* 1838 * Depend on the vnode interlock to keep things stable enough 1839 * for a quick test. Since there might be hundreds of 1840 * thousands of vnodes, we cannot afford even a subroutine 1841 * call unless there's a good chance that we have work to do. 1842 */ 1843 if (vp->v_type == VNON) { 1844 VI_UNLOCK(vp); 1845 continue; 1846 } 1847 ip = VTOI(vp); 1848 if ((ip->i_flag & 1849 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1850 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1851 VI_UNLOCK(vp); 1852 continue; 1853 } 1854 if ((error = vget(vp, lockreq)) != 0) { 1855 if (error == ENOENT || error == ENOLCK) { 1856 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1857 goto loop; 1858 } 1859 continue; 1860 } 1861 #ifdef QUOTA 1862 qsyncvp(vp); 1863 #endif 1864 for (;;) { 1865 error = ffs_syncvnode(vp, waitfor, 0); 1866 if (error == ERELOOKUP) 1867 continue; 1868 if (error != 0) 1869 allerror = error; 1870 break; 1871 } 1872 vput(vp); 1873 } 1874 /* 1875 * Force stale filesystem control information to be flushed. 1876 */ 1877 if (waitfor == MNT_WAIT || rebooting) { 1878 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1879 allerror = error; 1880 if (ffs_fsfail_cleanup(ump, allerror)) 1881 allerror = 0; 1882 /* Flushed work items may create new vnodes to clean */ 1883 if (allerror == 0 && count) 1884 goto loop; 1885 } 1886 1887 devvp = ump->um_devvp; 1888 bo = &devvp->v_bufobj; 1889 BO_LOCK(bo); 1890 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1891 BO_UNLOCK(bo); 1892 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1893 error = VOP_FSYNC(devvp, waitfor, td); 1894 VOP_UNLOCK(devvp); 1895 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1896 error = ffs_sbupdate(ump, waitfor, 0); 1897 if (error != 0) 1898 allerror = error; 1899 if (ffs_fsfail_cleanup(ump, allerror)) 1900 allerror = 0; 1901 if (allerror == 0 && waitfor == MNT_WAIT) 1902 goto loop; 1903 } else if (suspend != 0) { 1904 if (softdep_check_suspend(mp, 1905 devvp, 1906 softdep_deps, 1907 softdep_accdeps, 1908 secondary_writes, 1909 secondary_accwrites) != 0) { 1910 MNT_IUNLOCK(mp); 1911 goto loop; /* More work needed */ 1912 } 1913 mtx_assert(MNT_MTX(mp), MA_OWNED); 1914 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1915 MNT_IUNLOCK(mp); 1916 suspended = 1; 1917 } else 1918 BO_UNLOCK(bo); 1919 /* 1920 * Write back modified superblock. 1921 */ 1922 if (fs->fs_fmod != 0 && 1923 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1924 allerror = error; 1925 if (ffs_fsfail_cleanup(ump, allerror)) 1926 allerror = 0; 1927 return (allerror); 1928 } 1929 1930 int 1931 ffs_vget(mp, ino, flags, vpp) 1932 struct mount *mp; 1933 ino_t ino; 1934 int flags; 1935 struct vnode **vpp; 1936 { 1937 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1938 } 1939 1940 int 1941 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1942 struct mount *mp; 1943 ino_t ino; 1944 int flags; 1945 struct vnode **vpp; 1946 int ffs_flags; 1947 { 1948 struct fs *fs; 1949 struct inode *ip; 1950 struct ufsmount *ump; 1951 struct buf *bp; 1952 struct vnode *vp; 1953 daddr_t dbn; 1954 int error; 1955 1956 MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0); 1957 1958 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1959 if (error != 0) 1960 return (error); 1961 if (*vpp != NULL) { 1962 if ((ffs_flags & FFSV_REPLACE) == 0) 1963 return (0); 1964 vgone(*vpp); 1965 vput(*vpp); 1966 } 1967 1968 /* 1969 * We must promote to an exclusive lock for vnode creation. This 1970 * can happen if lookup is passed LOCKSHARED. 1971 */ 1972 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1973 flags &= ~LK_TYPE_MASK; 1974 flags |= LK_EXCLUSIVE; 1975 } 1976 1977 /* 1978 * We do not lock vnode creation as it is believed to be too 1979 * expensive for such rare case as simultaneous creation of vnode 1980 * for same ino by different processes. We just allow them to race 1981 * and check later to decide who wins. Let the race begin! 1982 */ 1983 1984 ump = VFSTOUFS(mp); 1985 fs = ump->um_fs; 1986 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO); 1987 1988 /* Allocate a new vnode/inode. */ 1989 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1990 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1991 if (error) { 1992 *vpp = NULL; 1993 uma_zfree_smr(uma_inode, ip); 1994 return (error); 1995 } 1996 /* 1997 * FFS supports recursive locking. 1998 */ 1999 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 2000 VN_LOCK_AREC(vp); 2001 vp->v_data = ip; 2002 vp->v_bufobj.bo_bsize = fs->fs_bsize; 2003 ip->i_vnode = vp; 2004 ip->i_ump = ump; 2005 ip->i_number = ino; 2006 ip->i_ea_refs = 0; 2007 ip->i_nextclustercg = -1; 2008 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; 2009 ip->i_mode = 0; /* ensure error cases below throw away vnode */ 2010 #ifdef DIAGNOSTIC 2011 ufs_init_trackers(ip); 2012 #endif 2013 #ifdef QUOTA 2014 { 2015 int i; 2016 for (i = 0; i < MAXQUOTAS; i++) 2017 ip->i_dquot[i] = NODQUOT; 2018 } 2019 #endif 2020 2021 if (ffs_flags & FFSV_FORCEINSMQ) 2022 vp->v_vflag |= VV_FORCEINSMQ; 2023 error = insmntque(vp, mp); 2024 if (error != 0) { 2025 uma_zfree_smr(uma_inode, ip); 2026 *vpp = NULL; 2027 return (error); 2028 } 2029 vp->v_vflag &= ~VV_FORCEINSMQ; 2030 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 2031 if (error != 0) 2032 return (error); 2033 if (*vpp != NULL) { 2034 /* 2035 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) 2036 * operate on empty inode, which must not be found by 2037 * other threads until fully filled. Vnode for empty 2038 * inode must be not re-inserted on the hash by other 2039 * thread, after removal by us at the beginning. 2040 */ 2041 MPASS((ffs_flags & FFSV_REPLACE) == 0); 2042 return (0); 2043 } 2044 2045 /* Read in the disk contents for the inode, copy into the inode. */ 2046 dbn = fsbtodb(fs, ino_to_fsba(fs, ino)); 2047 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, 2048 NULL, NULL, 0, NOCRED, 0, NULL, &bp); 2049 if (error != 0) { 2050 /* 2051 * The inode does not contain anything useful, so it would 2052 * be misleading to leave it on its hash chain. With mode 2053 * still zero, it will be unlinked and returned to the free 2054 * list by vput(). 2055 */ 2056 vgone(vp); 2057 vput(vp); 2058 *vpp = NULL; 2059 return (error); 2060 } 2061 if (I_IS_UFS1(ip)) 2062 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 2063 else 2064 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 2065 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { 2066 bqrelse(bp); 2067 vgone(vp); 2068 vput(vp); 2069 *vpp = NULL; 2070 return (error); 2071 } 2072 if (DOINGSOFTDEP(vp)) 2073 softdep_load_inodeblock(ip); 2074 else 2075 ip->i_effnlink = ip->i_nlink; 2076 bqrelse(bp); 2077 2078 /* 2079 * Initialize the vnode from the inode, check for aliases. 2080 * Note that the underlying vnode may have changed. 2081 */ 2082 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, 2083 &vp); 2084 if (error) { 2085 vgone(vp); 2086 vput(vp); 2087 *vpp = NULL; 2088 return (error); 2089 } 2090 2091 /* 2092 * Finish inode initialization. 2093 */ 2094 if (vp->v_type != VFIFO) { 2095 /* FFS supports shared locking for all files except fifos. */ 2096 VN_LOCK_ASHARE(vp); 2097 } 2098 2099 /* 2100 * Set up a generation number for this inode if it does not 2101 * already have one. This should only happen on old filesystems. 2102 */ 2103 if (ip->i_gen == 0) { 2104 while (ip->i_gen == 0) 2105 ip->i_gen = arc4random(); 2106 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 2107 UFS_INODE_SET_FLAG(ip, IN_MODIFIED); 2108 DIP_SET(ip, i_gen, ip->i_gen); 2109 } 2110 } 2111 #ifdef MAC 2112 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 2113 /* 2114 * If this vnode is already allocated, and we're running 2115 * multi-label, attempt to perform a label association 2116 * from the extended attributes on the inode. 2117 */ 2118 error = mac_vnode_associate_extattr(mp, vp); 2119 if (error) { 2120 /* ufs_inactive will release ip->i_devvp ref. */ 2121 vgone(vp); 2122 vput(vp); 2123 *vpp = NULL; 2124 return (error); 2125 } 2126 } 2127 #endif 2128 2129 *vpp = vp; 2130 return (0); 2131 } 2132 2133 /* 2134 * File handle to vnode 2135 * 2136 * Have to be really careful about stale file handles: 2137 * - check that the inode number is valid 2138 * - for UFS2 check that the inode number is initialized 2139 * - call ffs_vget() to get the locked inode 2140 * - check for an unallocated inode (i_mode == 0) 2141 * - check that the given client host has export rights and return 2142 * those rights via. exflagsp and credanonp 2143 */ 2144 static int 2145 ffs_fhtovp(mp, fhp, flags, vpp) 2146 struct mount *mp; 2147 struct fid *fhp; 2148 int flags; 2149 struct vnode **vpp; 2150 { 2151 struct ufid *ufhp; 2152 struct ufsmount *ump; 2153 struct fs *fs; 2154 struct cg *cgp; 2155 struct buf *bp; 2156 ino_t ino; 2157 u_int cg; 2158 int error; 2159 2160 ufhp = (struct ufid *)fhp; 2161 ino = ufhp->ufid_ino; 2162 ump = VFSTOUFS(mp); 2163 fs = ump->um_fs; 2164 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) 2165 return (ESTALE); 2166 /* 2167 * Need to check if inode is initialized because UFS2 does lazy 2168 * initialization and nfs_fhtovp can offer arbitrary inode numbers. 2169 */ 2170 if (fs->fs_magic != FS_UFS2_MAGIC) 2171 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 2172 cg = ino_to_cg(fs, ino); 2173 if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) 2174 return (error); 2175 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { 2176 brelse(bp); 2177 return (ESTALE); 2178 } 2179 brelse(bp); 2180 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 2181 } 2182 2183 /* 2184 * Initialize the filesystem. 2185 */ 2186 static int 2187 ffs_init(vfsp) 2188 struct vfsconf *vfsp; 2189 { 2190 2191 ffs_susp_initialize(); 2192 softdep_initialize(); 2193 return (ufs_init(vfsp)); 2194 } 2195 2196 /* 2197 * Undo the work of ffs_init(). 2198 */ 2199 static int 2200 ffs_uninit(vfsp) 2201 struct vfsconf *vfsp; 2202 { 2203 int ret; 2204 2205 ret = ufs_uninit(vfsp); 2206 softdep_uninitialize(); 2207 ffs_susp_uninitialize(); 2208 taskqueue_drain_all(taskqueue_thread); 2209 return (ret); 2210 } 2211 2212 /* 2213 * Structure used to pass information from ffs_sbupdate to its 2214 * helper routine ffs_use_bwrite. 2215 */ 2216 struct devfd { 2217 struct ufsmount *ump; 2218 struct buf *sbbp; 2219 int waitfor; 2220 int suspended; 2221 int error; 2222 }; 2223 2224 /* 2225 * Write a superblock and associated information back to disk. 2226 */ 2227 int 2228 ffs_sbupdate(ump, waitfor, suspended) 2229 struct ufsmount *ump; 2230 int waitfor; 2231 int suspended; 2232 { 2233 struct fs *fs; 2234 struct buf *sbbp; 2235 struct devfd devfd; 2236 2237 fs = ump->um_fs; 2238 if (fs->fs_ronly == 1 && 2239 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 2240 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) 2241 panic("ffs_sbupdate: write read-only filesystem"); 2242 /* 2243 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 2244 */ 2245 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 2246 (int)fs->fs_sbsize, 0, 0, 0); 2247 /* 2248 * Initialize info needed for write function. 2249 */ 2250 devfd.ump = ump; 2251 devfd.sbbp = sbbp; 2252 devfd.waitfor = waitfor; 2253 devfd.suspended = suspended; 2254 devfd.error = 0; 2255 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); 2256 } 2257 2258 /* 2259 * Write function for use by filesystem-layer routines. 2260 */ 2261 static int 2262 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) 2263 { 2264 struct devfd *devfdp; 2265 struct ufsmount *ump; 2266 struct buf *bp; 2267 struct fs *fs; 2268 int error; 2269 2270 devfdp = devfd; 2271 ump = devfdp->ump; 2272 fs = ump->um_fs; 2273 /* 2274 * Writing the superblock summary information. 2275 */ 2276 if (loc != fs->fs_sblockloc) { 2277 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); 2278 bcopy(buf, bp->b_data, (u_int)size); 2279 if (devfdp->suspended) 2280 bp->b_flags |= B_VALIDSUSPWRT; 2281 if (devfdp->waitfor != MNT_WAIT) 2282 bawrite(bp); 2283 else if ((error = bwrite(bp)) != 0) 2284 devfdp->error = error; 2285 return (0); 2286 } 2287 /* 2288 * Writing the superblock itself. We need to do special checks for it. 2289 */ 2290 bp = devfdp->sbbp; 2291 if (ffs_fsfail_cleanup(ump, devfdp->error)) 2292 devfdp->error = 0; 2293 if (devfdp->error != 0) { 2294 brelse(bp); 2295 return (devfdp->error); 2296 } 2297 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 2298 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2299 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2300 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 2301 fs->fs_sblockloc = SBLOCK_UFS1; 2302 } 2303 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 2304 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2305 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2306 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 2307 fs->fs_sblockloc = SBLOCK_UFS2; 2308 } 2309 if (MOUNTEDSOFTDEP(ump->um_mountp)) 2310 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 2311 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 2312 fs = (struct fs *)bp->b_data; 2313 ffs_oldfscompat_write(fs, ump); 2314 fs->fs_si = NULL; 2315 /* Recalculate the superblock hash */ 2316 fs->fs_ckhash = ffs_calc_sbhash(fs); 2317 if (devfdp->suspended) 2318 bp->b_flags |= B_VALIDSUSPWRT; 2319 if (devfdp->waitfor != MNT_WAIT) 2320 bawrite(bp); 2321 else if ((error = bwrite(bp)) != 0) 2322 devfdp->error = error; 2323 return (devfdp->error); 2324 } 2325 2326 static int 2327 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 2328 int attrnamespace, const char *attrname) 2329 { 2330 2331 #ifdef UFS_EXTATTR 2332 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 2333 attrname)); 2334 #else 2335 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 2336 attrname)); 2337 #endif 2338 } 2339 2340 static void 2341 ffs_ifree(struct ufsmount *ump, struct inode *ip) 2342 { 2343 2344 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 2345 uma_zfree(uma_ufs1, ip->i_din1); 2346 else if (ip->i_din2 != NULL) 2347 uma_zfree(uma_ufs2, ip->i_din2); 2348 uma_zfree_smr(uma_inode, ip); 2349 } 2350 2351 static int dobkgrdwrite = 1; 2352 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 2353 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 2354 2355 /* 2356 * Complete a background write started from bwrite. 2357 */ 2358 static void 2359 ffs_backgroundwritedone(struct buf *bp) 2360 { 2361 struct bufobj *bufobj; 2362 struct buf *origbp; 2363 2364 #ifdef SOFTUPDATES 2365 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0) 2366 softdep_handle_error(bp); 2367 #endif 2368 2369 /* 2370 * Find the original buffer that we are writing. 2371 */ 2372 bufobj = bp->b_bufobj; 2373 BO_LOCK(bufobj); 2374 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2375 panic("backgroundwritedone: lost buffer"); 2376 2377 /* 2378 * We should mark the cylinder group buffer origbp as 2379 * dirty, to not lose the failed write. 2380 */ 2381 if ((bp->b_ioflags & BIO_ERROR) != 0) 2382 origbp->b_vflags |= BV_BKGRDERR; 2383 BO_UNLOCK(bufobj); 2384 /* 2385 * Process dependencies then return any unfinished ones. 2386 */ 2387 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2388 buf_complete(bp); 2389 #ifdef SOFTUPDATES 2390 if (!LIST_EMPTY(&bp->b_dep)) 2391 softdep_move_dependencies(bp, origbp); 2392 #endif 2393 /* 2394 * This buffer is marked B_NOCACHE so when it is released 2395 * by biodone it will be tossed. 2396 */ 2397 bp->b_flags |= B_NOCACHE; 2398 bp->b_flags &= ~B_CACHE; 2399 pbrelvp(bp); 2400 2401 /* 2402 * Prevent brelse() from trying to keep and re-dirtying bp on 2403 * errors. It causes b_bufobj dereference in 2404 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2405 * pbrelvp() above. 2406 */ 2407 if ((bp->b_ioflags & BIO_ERROR) != 0) 2408 bp->b_flags |= B_INVAL; 2409 bufdone(bp); 2410 BO_LOCK(bufobj); 2411 /* 2412 * Clear the BV_BKGRDINPROG flag in the original buffer 2413 * and awaken it if it is waiting for the write to complete. 2414 * If BV_BKGRDINPROG is not set in the original buffer it must 2415 * have been released and re-instantiated - which is not legal. 2416 */ 2417 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2418 ("backgroundwritedone: lost buffer2")); 2419 origbp->b_vflags &= ~BV_BKGRDINPROG; 2420 if (origbp->b_vflags & BV_BKGRDWAIT) { 2421 origbp->b_vflags &= ~BV_BKGRDWAIT; 2422 wakeup(&origbp->b_xflags); 2423 } 2424 BO_UNLOCK(bufobj); 2425 } 2426 2427 /* 2428 * Write, release buffer on completion. (Done by iodone 2429 * if async). Do not bother writing anything if the buffer 2430 * is invalid. 2431 * 2432 * Note that we set B_CACHE here, indicating that buffer is 2433 * fully valid and thus cacheable. This is true even of NFS 2434 * now so we set it generally. This could be set either here 2435 * or in biodone() since the I/O is synchronous. We put it 2436 * here. 2437 */ 2438 static int 2439 ffs_bufwrite(struct buf *bp) 2440 { 2441 struct buf *newbp; 2442 struct cg *cgp; 2443 2444 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2445 if (bp->b_flags & B_INVAL) { 2446 brelse(bp); 2447 return (0); 2448 } 2449 2450 if (!BUF_ISLOCKED(bp)) 2451 panic("bufwrite: buffer is not busy???"); 2452 /* 2453 * If a background write is already in progress, delay 2454 * writing this block if it is asynchronous. Otherwise 2455 * wait for the background write to complete. 2456 */ 2457 BO_LOCK(bp->b_bufobj); 2458 if (bp->b_vflags & BV_BKGRDINPROG) { 2459 if (bp->b_flags & B_ASYNC) { 2460 BO_UNLOCK(bp->b_bufobj); 2461 bdwrite(bp); 2462 return (0); 2463 } 2464 bp->b_vflags |= BV_BKGRDWAIT; 2465 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2466 "bwrbg", 0); 2467 if (bp->b_vflags & BV_BKGRDINPROG) 2468 panic("bufwrite: still writing"); 2469 } 2470 bp->b_vflags &= ~BV_BKGRDERR; 2471 BO_UNLOCK(bp->b_bufobj); 2472 2473 /* 2474 * If this buffer is marked for background writing and we 2475 * do not have to wait for it, make a copy and write the 2476 * copy so as to leave this buffer ready for further use. 2477 * 2478 * This optimization eats a lot of memory. If we have a page 2479 * or buffer shortfall we can't do it. 2480 */ 2481 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2482 (bp->b_flags & B_ASYNC) && 2483 !vm_page_count_severe() && 2484 !buf_dirty_count_severe()) { 2485 KASSERT(bp->b_iodone == NULL, 2486 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2487 2488 /* get a new block */ 2489 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2490 if (newbp == NULL) 2491 goto normal_write; 2492 2493 KASSERT(buf_mapped(bp), ("Unmapped cg")); 2494 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2495 BO_LOCK(bp->b_bufobj); 2496 bp->b_vflags |= BV_BKGRDINPROG; 2497 BO_UNLOCK(bp->b_bufobj); 2498 newbp->b_xflags |= 2499 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; 2500 newbp->b_lblkno = bp->b_lblkno; 2501 newbp->b_blkno = bp->b_blkno; 2502 newbp->b_offset = bp->b_offset; 2503 newbp->b_iodone = ffs_backgroundwritedone; 2504 newbp->b_flags |= B_ASYNC; 2505 newbp->b_flags &= ~B_INVAL; 2506 pbgetvp(bp->b_vp, newbp); 2507 2508 #ifdef SOFTUPDATES 2509 /* 2510 * Move over the dependencies. If there are rollbacks, 2511 * leave the parent buffer dirtied as it will need to 2512 * be written again. 2513 */ 2514 if (LIST_EMPTY(&bp->b_dep) || 2515 softdep_move_dependencies(bp, newbp) == 0) 2516 bundirty(bp); 2517 #else 2518 bundirty(bp); 2519 #endif 2520 2521 /* 2522 * Initiate write on the copy, release the original. The 2523 * BKGRDINPROG flag prevents it from going away until 2524 * the background write completes. We have to recalculate 2525 * its check hash in case the buffer gets freed and then 2526 * reconstituted from the buffer cache during a later read. 2527 */ 2528 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2529 cgp = (struct cg *)bp->b_data; 2530 cgp->cg_ckhash = 0; 2531 cgp->cg_ckhash = 2532 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2533 } 2534 bqrelse(bp); 2535 bp = newbp; 2536 } else 2537 /* Mark the buffer clean */ 2538 bundirty(bp); 2539 2540 /* Let the normal bufwrite do the rest for us */ 2541 normal_write: 2542 /* 2543 * If we are writing a cylinder group, update its time. 2544 */ 2545 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2546 cgp = (struct cg *)bp->b_data; 2547 cgp->cg_old_time = cgp->cg_time = time_second; 2548 } 2549 return (bufwrite(bp)); 2550 } 2551 2552 static void 2553 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2554 { 2555 struct vnode *vp; 2556 struct buf *tbp; 2557 int error, nocopy; 2558 2559 /* 2560 * This is the bufobj strategy for the private VCHR vnodes 2561 * used by FFS to access the underlying storage device. 2562 * We override the default bufobj strategy and thus bypass 2563 * VOP_STRATEGY() for these vnodes. 2564 */ 2565 vp = bo2vnode(bo); 2566 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR || 2567 bp->b_vp->v_rdev == NULL || 2568 bp->b_vp->v_rdev->si_mountpt == NULL || 2569 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL || 2570 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp, 2571 ("ffs_geom_strategy() with wrong vp")); 2572 if (bp->b_iocmd == BIO_WRITE) { 2573 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2574 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2575 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2576 panic("ffs_geom_strategy: bad I/O"); 2577 nocopy = bp->b_flags & B_NOCOPY; 2578 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2579 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2580 vp->v_rdev->si_snapdata != NULL) { 2581 if ((bp->b_flags & B_CLUSTER) != 0) { 2582 runningbufwakeup(bp); 2583 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2584 b_cluster.cluster_entry) { 2585 error = ffs_copyonwrite(vp, tbp); 2586 if (error != 0 && 2587 error != EOPNOTSUPP) { 2588 bp->b_error = error; 2589 bp->b_ioflags |= BIO_ERROR; 2590 bp->b_flags &= ~B_BARRIER; 2591 bufdone(bp); 2592 return; 2593 } 2594 } 2595 bp->b_runningbufspace = bp->b_bufsize; 2596 atomic_add_long(&runningbufspace, 2597 bp->b_runningbufspace); 2598 } else { 2599 error = ffs_copyonwrite(vp, bp); 2600 if (error != 0 && error != EOPNOTSUPP) { 2601 bp->b_error = error; 2602 bp->b_ioflags |= BIO_ERROR; 2603 bp->b_flags &= ~B_BARRIER; 2604 bufdone(bp); 2605 return; 2606 } 2607 } 2608 } 2609 #ifdef SOFTUPDATES 2610 if ((bp->b_flags & B_CLUSTER) != 0) { 2611 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2612 b_cluster.cluster_entry) { 2613 if (!LIST_EMPTY(&tbp->b_dep)) 2614 buf_start(tbp); 2615 } 2616 } else { 2617 if (!LIST_EMPTY(&bp->b_dep)) 2618 buf_start(bp); 2619 } 2620 2621 #endif 2622 /* 2623 * Check for metadata that needs check-hashes and update them. 2624 */ 2625 switch (bp->b_xflags & BX_FSPRIV) { 2626 case BX_CYLGRP: 2627 ((struct cg *)bp->b_data)->cg_ckhash = 0; 2628 ((struct cg *)bp->b_data)->cg_ckhash = 2629 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2630 break; 2631 2632 case BX_SUPERBLOCK: 2633 case BX_INODE: 2634 case BX_INDIR: 2635 case BX_DIR: 2636 printf("Check-hash write is unimplemented!!!\n"); 2637 break; 2638 2639 case 0: 2640 break; 2641 2642 default: 2643 printf("multiple buffer types 0x%b\n", 2644 (u_int)(bp->b_xflags & BX_FSPRIV), 2645 PRINT_UFS_BUF_XFLAGS); 2646 break; 2647 } 2648 } 2649 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable) 2650 bp->b_xflags |= BX_CVTENXIO; 2651 g_vfs_strategy(bo, bp); 2652 } 2653 2654 int 2655 ffs_own_mount(const struct mount *mp) 2656 { 2657 2658 if (mp->mnt_op == &ufs_vfsops) 2659 return (1); 2660 return (0); 2661 } 2662 2663 #ifdef DDB 2664 #ifdef SOFTUPDATES 2665 2666 /* defined in ffs_softdep.c */ 2667 extern void db_print_ffs(struct ufsmount *ump); 2668 2669 DB_SHOW_COMMAND(ffs, db_show_ffs) 2670 { 2671 struct mount *mp; 2672 struct ufsmount *ump; 2673 2674 if (have_addr) { 2675 ump = VFSTOUFS((struct mount *)addr); 2676 db_print_ffs(ump); 2677 return; 2678 } 2679 2680 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2681 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2682 db_print_ffs(VFSTOUFS(mp)); 2683 } 2684 } 2685 2686 #endif /* SOFTUPDATES */ 2687 #endif /* DDB */ 2688