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", "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 * On first ENXIO error, initiate an asynchronous forcible unmount. 285 * Used to unmount filesystems whose underlying media has gone away. 286 * 287 * Return true if a cleanup is in progress. 288 */ 289 int 290 ffs_fsfail_cleanup(struct ufsmount *ump, int error) 291 { 292 int retval; 293 294 UFS_LOCK(ump); 295 retval = ffs_fsfail_cleanup_locked(ump, error); 296 UFS_UNLOCK(ump); 297 return (retval); 298 } 299 300 int 301 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error) 302 { 303 mtx_assert(UFS_MTX(ump), MA_OWNED); 304 if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) { 305 ump->um_flags |= UM_FSFAIL_CLEANUP; 306 /* 307 * Queue an async forced unmount. 308 */ 309 vfs_ref(ump->um_mountp); 310 dounmount(ump->um_mountp, 311 MNT_FORCE | MNT_RECURSE | MNT_DEFERRED, curthread); 312 printf("UFS: forcibly unmounting %s from %s\n", 313 ump->um_mountp->mnt_stat.f_mntfromname, 314 ump->um_mountp->mnt_stat.f_mntonname); 315 } 316 return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0); 317 } 318 319 /* 320 * Wrapper used during ENXIO cleanup to allocate empty buffers when 321 * the kernel is unable to read the real one. They are needed so that 322 * the soft updates code can use them to unwind its dependencies. 323 */ 324 int 325 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno, 326 daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt, 327 struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *), 328 struct buf **bpp) 329 { 330 int error; 331 332 flags |= GB_CVTENXIO; 333 error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt, 334 cred, flags, ckhashfunc, bpp); 335 if (error != 0 && ffs_fsfail_cleanup(ump, error)) { 336 error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp); 337 KASSERT(error == 0, ("getblkx failed")); 338 vfs_bio_bzero_buf(*bpp, 0, size); 339 } 340 return (error); 341 } 342 343 static int 344 ffs_mount(struct mount *mp) 345 { 346 struct vnode *devvp, *odevvp; 347 struct thread *td; 348 struct ufsmount *ump = NULL; 349 struct fs *fs; 350 int error, flags; 351 int error1 __diagused; 352 uint64_t mntorflags, saved_mnt_flag; 353 accmode_t accmode; 354 struct nameidata ndp; 355 char *fspec; 356 bool mounted_softdep; 357 358 td = curthread; 359 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 360 return (EINVAL); 361 if (uma_inode == NULL) { 362 uma_inode = uma_zcreate("FFS inode", 363 sizeof(struct inode), NULL, NULL, NULL, NULL, 364 UMA_ALIGN_PTR, 0); 365 uma_ufs1 = uma_zcreate("FFS1 dinode", 366 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 367 UMA_ALIGN_PTR, 0); 368 uma_ufs2 = uma_zcreate("FFS2 dinode", 369 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 370 UMA_ALIGN_PTR, 0); 371 VFS_SMR_ZONE_SET(uma_inode); 372 } 373 374 vfs_deleteopt(mp->mnt_optnew, "groupquota"); 375 vfs_deleteopt(mp->mnt_optnew, "userquota"); 376 377 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 378 if (error) 379 return (error); 380 381 mntorflags = 0; 382 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0) 383 mntorflags |= MNT_UNTRUSTED; 384 385 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 386 mntorflags |= MNT_ACLS; 387 388 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 389 mntorflags |= MNT_SNAPSHOT; 390 /* 391 * Once we have set the MNT_SNAPSHOT flag, do not 392 * persist "snapshot" in the options list. 393 */ 394 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 395 vfs_deleteopt(mp->mnt_opt, "snapshot"); 396 } 397 398 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 399 if (mntorflags & MNT_ACLS) { 400 vfs_mount_error(mp, 401 "\"acls\" and \"nfsv4acls\" options " 402 "are mutually exclusive"); 403 return (EINVAL); 404 } 405 mntorflags |= MNT_NFS4ACLS; 406 } 407 408 MNT_ILOCK(mp); 409 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP; 410 mp->mnt_flag |= mntorflags; 411 MNT_IUNLOCK(mp); 412 413 /* 414 * If this is a snapshot request, take the snapshot. 415 */ 416 if (mp->mnt_flag & MNT_SNAPSHOT) 417 return (ffs_snapshot(mp, fspec)); 418 419 /* 420 * Must not call namei() while owning busy ref. 421 */ 422 if (mp->mnt_flag & MNT_UPDATE) 423 vfs_unbusy(mp); 424 425 /* 426 * Not an update, or updating the name: look up the name 427 * and verify that it refers to a sensible disk device. 428 */ 429 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec); 430 error = namei(&ndp); 431 if ((mp->mnt_flag & MNT_UPDATE) != 0) { 432 /* 433 * Unmount does not start if MNT_UPDATE is set. Mount 434 * update busies mp before setting MNT_UPDATE. We 435 * must be able to retain our busy ref successfully, 436 * without sleep. 437 */ 438 error1 = vfs_busy(mp, MBF_NOWAIT); 439 MPASS(error1 == 0); 440 } 441 if (error != 0) 442 return (error); 443 NDFREE_PNBUF(&ndp); 444 if (!vn_isdisk_error(ndp.ni_vp, &error)) { 445 vput(ndp.ni_vp); 446 return (error); 447 } 448 449 /* 450 * If mount by non-root, then verify that user has necessary 451 * permissions on the device. 452 */ 453 accmode = VREAD; 454 if ((mp->mnt_flag & MNT_RDONLY) == 0) 455 accmode |= VWRITE; 456 error = VOP_ACCESS(ndp.ni_vp, accmode, td->td_ucred, td); 457 if (error) 458 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 459 if (error) { 460 vput(ndp.ni_vp); 461 return (error); 462 } 463 464 /* 465 * New mount 466 * 467 * We need the name for the mount point (also used for 468 * "last mounted on") copied in. If an error occurs, 469 * the mount point is discarded by the upper level code. 470 * Note that vfs_mount_alloc() populates f_mntonname for us. 471 */ 472 if ((mp->mnt_flag & MNT_UPDATE) == 0) { 473 if ((error = ffs_mountfs(ndp.ni_vp, mp, td)) != 0) { 474 vrele(ndp.ni_vp); 475 return (error); 476 } 477 } else { 478 /* 479 * When updating, check whether changing from read-only to 480 * read/write; if there is no device name, that's all we do. 481 */ 482 ump = VFSTOUFS(mp); 483 fs = ump->um_fs; 484 odevvp = ump->um_odevvp; 485 devvp = ump->um_devvp; 486 487 /* 488 * If it's not the same vnode, or at least the same device 489 * then it's not correct. 490 */ 491 if (ndp.ni_vp->v_rdev != ump->um_odevvp->v_rdev) 492 error = EINVAL; /* needs translation */ 493 vput(ndp.ni_vp); 494 if (error) 495 return (error); 496 if (fs->fs_ronly == 0 && 497 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 498 /* 499 * Flush any dirty data and suspend filesystem. 500 */ 501 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 502 return (error); 503 error = vfs_write_suspend_umnt(mp); 504 if (error != 0) 505 return (error); 506 507 fs->fs_ronly = 1; 508 if (MOUNTEDSOFTDEP(mp)) { 509 MNT_ILOCK(mp); 510 mp->mnt_flag &= ~MNT_SOFTDEP; 511 MNT_IUNLOCK(mp); 512 mounted_softdep = true; 513 } else 514 mounted_softdep = false; 515 516 /* 517 * Check for and optionally get rid of files open 518 * for writing. 519 */ 520 flags = WRITECLOSE; 521 if (mp->mnt_flag & MNT_FORCE) 522 flags |= FORCECLOSE; 523 if (mounted_softdep) { 524 error = softdep_flushfiles(mp, flags, td); 525 } else { 526 error = ffs_flushfiles(mp, flags, td); 527 } 528 if (error) { 529 fs->fs_ronly = 0; 530 if (mounted_softdep) { 531 MNT_ILOCK(mp); 532 mp->mnt_flag |= MNT_SOFTDEP; 533 MNT_IUNLOCK(mp); 534 } 535 vfs_write_resume(mp, 0); 536 return (error); 537 } 538 539 if (fs->fs_pendingblocks != 0 || 540 fs->fs_pendinginodes != 0) { 541 printf("WARNING: %s Update error: blocks %jd " 542 "files %d\n", fs->fs_fsmnt, 543 (intmax_t)fs->fs_pendingblocks, 544 fs->fs_pendinginodes); 545 fs->fs_pendingblocks = 0; 546 fs->fs_pendinginodes = 0; 547 } 548 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 549 fs->fs_clean = 1; 550 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 551 fs->fs_ronly = 0; 552 fs->fs_clean = 0; 553 if (mounted_softdep) { 554 MNT_ILOCK(mp); 555 mp->mnt_flag |= MNT_SOFTDEP; 556 MNT_IUNLOCK(mp); 557 } 558 vfs_write_resume(mp, 0); 559 return (error); 560 } 561 if (mounted_softdep) 562 softdep_unmount(mp); 563 g_topology_lock(); 564 /* 565 * Drop our write and exclusive access. 566 */ 567 g_access(ump->um_cp, 0, -1, -1); 568 g_topology_unlock(); 569 MNT_ILOCK(mp); 570 mp->mnt_flag |= MNT_RDONLY; 571 MNT_IUNLOCK(mp); 572 /* 573 * Allow the writers to note that filesystem 574 * is ro now. 575 */ 576 vfs_write_resume(mp, 0); 577 } 578 if ((mp->mnt_flag & MNT_RELOAD) && 579 (error = ffs_reload(mp, 0)) != 0) 580 return (error); 581 if (fs->fs_ronly && 582 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 583 /* 584 * If upgrade to read-write by non-root, then verify 585 * that user has necessary permissions on the device. 586 */ 587 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY); 588 error = VOP_ACCESS(odevvp, VREAD | VWRITE, 589 td->td_ucred, td); 590 if (error) 591 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 592 VOP_UNLOCK(odevvp); 593 if (error) { 594 return (error); 595 } 596 fs->fs_flags &= ~FS_UNCLEAN; 597 if (fs->fs_clean == 0) { 598 fs->fs_flags |= FS_UNCLEAN; 599 if ((mp->mnt_flag & MNT_FORCE) || 600 ((fs->fs_flags & 601 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 602 (fs->fs_flags & FS_DOSOFTDEP))) { 603 printf("WARNING: %s was not properly " 604 "dismounted\n", fs->fs_fsmnt); 605 } else { 606 vfs_mount_error(mp, 607 "R/W mount of %s denied. %s.%s", 608 fs->fs_fsmnt, 609 "Filesystem is not clean - run fsck", 610 (fs->fs_flags & FS_SUJ) == 0 ? "" : 611 " Forced mount will invalidate" 612 " journal contents"); 613 return (EPERM); 614 } 615 } 616 g_topology_lock(); 617 /* 618 * Request exclusive write access. 619 */ 620 error = g_access(ump->um_cp, 0, 1, 1); 621 g_topology_unlock(); 622 if (error) 623 return (error); 624 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 625 return (error); 626 error = vfs_write_suspend_umnt(mp); 627 if (error != 0) 628 return (error); 629 fs->fs_ronly = 0; 630 MNT_ILOCK(mp); 631 saved_mnt_flag = MNT_RDONLY; 632 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag & 633 MNT_ASYNC) != 0) 634 saved_mnt_flag |= MNT_ASYNC; 635 mp->mnt_flag &= ~saved_mnt_flag; 636 MNT_IUNLOCK(mp); 637 fs->fs_mtime = time_second; 638 /* check to see if we need to start softdep */ 639 if ((fs->fs_flags & FS_DOSOFTDEP) && 640 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 641 fs->fs_ronly = 1; 642 MNT_ILOCK(mp); 643 mp->mnt_flag |= saved_mnt_flag; 644 MNT_IUNLOCK(mp); 645 vfs_write_resume(mp, 0); 646 return (error); 647 } 648 fs->fs_clean = 0; 649 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 650 fs->fs_ronly = 1; 651 if ((fs->fs_flags & FS_DOSOFTDEP) != 0) 652 softdep_unmount(mp); 653 MNT_ILOCK(mp); 654 mp->mnt_flag |= saved_mnt_flag; 655 MNT_IUNLOCK(mp); 656 vfs_write_resume(mp, 0); 657 return (error); 658 } 659 if (fs->fs_snapinum[0] != 0) 660 ffs_snapshot_mount(mp); 661 vfs_write_resume(mp, 0); 662 } 663 /* 664 * Soft updates is incompatible with "async", 665 * so if we are doing softupdates stop the user 666 * from setting the async flag in an update. 667 * Softdep_mount() clears it in an initial mount 668 * or ro->rw remount. 669 */ 670 if (MOUNTEDSOFTDEP(mp)) { 671 /* XXX: Reset too late ? */ 672 MNT_ILOCK(mp); 673 mp->mnt_flag &= ~MNT_ASYNC; 674 MNT_IUNLOCK(mp); 675 } 676 /* 677 * Keep MNT_ACLS flag if it is stored in superblock. 678 */ 679 if ((fs->fs_flags & FS_ACLS) != 0) { 680 /* XXX: Set too late ? */ 681 MNT_ILOCK(mp); 682 mp->mnt_flag |= MNT_ACLS; 683 MNT_IUNLOCK(mp); 684 } 685 686 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 687 /* XXX: Set too late ? */ 688 MNT_ILOCK(mp); 689 mp->mnt_flag |= MNT_NFS4ACLS; 690 MNT_IUNLOCK(mp); 691 } 692 693 } 694 695 MNT_ILOCK(mp); 696 /* 697 * This is racy versus lookup, see ufs_fplookup_vexec for details. 698 */ 699 if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0) 700 panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp); 701 if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0) 702 mp->mnt_kern_flag |= MNTK_FPLOOKUP; 703 MNT_IUNLOCK(mp); 704 705 vfs_mountedfrom(mp, fspec); 706 return (0); 707 } 708 709 /* 710 * Compatibility with old mount system call. 711 */ 712 713 static int 714 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) 715 { 716 struct ufs_args args; 717 int error; 718 719 if (data == NULL) 720 return (EINVAL); 721 error = copyin(data, &args, sizeof args); 722 if (error) 723 return (error); 724 725 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 726 ma = mount_arg(ma, "export", &args.export, sizeof(args.export)); 727 error = kernel_mount(ma, flags); 728 729 return (error); 730 } 731 732 /* 733 * Reload all incore data for a filesystem (used after running fsck on 734 * the root filesystem and finding things to fix). If the 'force' flag 735 * is 0, the filesystem must be mounted read-only. 736 * 737 * Things to do to update the mount: 738 * 1) invalidate all cached meta-data. 739 * 2) re-read superblock from disk. 740 * 3) re-read summary information from disk. 741 * 4) invalidate all inactive vnodes. 742 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary 743 * writers, if requested. 744 * 6) invalidate all cached file data. 745 * 7) re-read inode data for all active vnodes. 746 */ 747 int 748 ffs_reload(struct mount *mp, int flags) 749 { 750 struct vnode *vp, *mvp, *devvp; 751 struct inode *ip; 752 void *space; 753 struct buf *bp; 754 struct fs *fs, *newfs; 755 struct ufsmount *ump; 756 ufs2_daddr_t sblockloc; 757 int i, blks, error; 758 u_long size; 759 int32_t *lp; 760 761 ump = VFSTOUFS(mp); 762 763 MNT_ILOCK(mp); 764 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { 765 MNT_IUNLOCK(mp); 766 return (EINVAL); 767 } 768 MNT_IUNLOCK(mp); 769 770 /* 771 * Step 1: invalidate all cached meta-data. 772 */ 773 devvp = VFSTOUFS(mp)->um_devvp; 774 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 775 if (vinvalbuf(devvp, 0, 0, 0) != 0) 776 panic("ffs_reload: dirty1"); 777 VOP_UNLOCK(devvp); 778 779 /* 780 * Step 2: re-read superblock from disk. 781 */ 782 fs = VFSTOUFS(mp)->um_fs; 783 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 784 NOCRED, &bp)) != 0) 785 return (error); 786 newfs = (struct fs *)bp->b_data; 787 if ((newfs->fs_magic != FS_UFS1_MAGIC && 788 newfs->fs_magic != FS_UFS2_MAGIC) || 789 newfs->fs_bsize > MAXBSIZE || 790 newfs->fs_bsize < sizeof(struct fs)) { 791 brelse(bp); 792 return (EIO); /* XXX needs translation */ 793 } 794 /* 795 * Preserve the summary information, read-only status, and 796 * superblock location by copying these fields into our new 797 * superblock before using it to update the existing superblock. 798 */ 799 newfs->fs_si = fs->fs_si; 800 newfs->fs_ronly = fs->fs_ronly; 801 sblockloc = fs->fs_sblockloc; 802 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 803 brelse(bp); 804 ump->um_maxsymlinklen = fs->fs_maxsymlinklen; 805 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 806 UFS_LOCK(ump); 807 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 808 printf("WARNING: %s: reload pending error: blocks %jd " 809 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 810 fs->fs_pendinginodes); 811 fs->fs_pendingblocks = 0; 812 fs->fs_pendinginodes = 0; 813 } 814 UFS_UNLOCK(ump); 815 816 /* 817 * Step 3: re-read summary information from disk. 818 */ 819 size = fs->fs_cssize; 820 blks = howmany(size, fs->fs_fsize); 821 if (fs->fs_contigsumsize > 0) 822 size += fs->fs_ncg * sizeof(int32_t); 823 size += fs->fs_ncg * sizeof(u_int8_t); 824 free(fs->fs_csp, M_UFSMNT); 825 space = malloc(size, M_UFSMNT, M_WAITOK); 826 fs->fs_csp = space; 827 for (i = 0; i < blks; i += fs->fs_frag) { 828 size = fs->fs_bsize; 829 if (i + fs->fs_frag > blks) 830 size = (blks - i) * fs->fs_fsize; 831 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 832 NOCRED, &bp); 833 if (error) 834 return (error); 835 bcopy(bp->b_data, space, (u_int)size); 836 space = (char *)space + size; 837 brelse(bp); 838 } 839 /* 840 * We no longer know anything about clusters per cylinder group. 841 */ 842 if (fs->fs_contigsumsize > 0) { 843 fs->fs_maxcluster = lp = space; 844 for (i = 0; i < fs->fs_ncg; i++) 845 *lp++ = fs->fs_contigsumsize; 846 space = lp; 847 } 848 size = fs->fs_ncg * sizeof(u_int8_t); 849 fs->fs_contigdirs = (u_int8_t *)space; 850 bzero(fs->fs_contigdirs, size); 851 if ((flags & FFSR_UNSUSPEND) != 0) { 852 MNT_ILOCK(mp); 853 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 854 wakeup(&mp->mnt_flag); 855 MNT_IUNLOCK(mp); 856 } 857 858 loop: 859 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 860 /* 861 * Skip syncer vnode. 862 */ 863 if (vp->v_type == VNON) { 864 VI_UNLOCK(vp); 865 continue; 866 } 867 /* 868 * Step 4: invalidate all cached file data. 869 */ 870 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 871 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 872 goto loop; 873 } 874 if (vinvalbuf(vp, 0, 0, 0)) 875 panic("ffs_reload: dirty2"); 876 /* 877 * Step 5: re-read inode data for all active vnodes. 878 */ 879 ip = VTOI(vp); 880 error = 881 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 882 (int)fs->fs_bsize, NOCRED, &bp); 883 if (error) { 884 vput(vp); 885 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 886 return (error); 887 } 888 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) { 889 brelse(bp); 890 vput(vp); 891 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 892 return (error); 893 } 894 ip->i_effnlink = ip->i_nlink; 895 brelse(bp); 896 vput(vp); 897 } 898 return (0); 899 } 900 901 /* 902 * Common code for mount and mountroot 903 */ 904 static int 905 ffs_mountfs(odevvp, mp, td) 906 struct vnode *odevvp; 907 struct mount *mp; 908 struct thread *td; 909 { 910 struct ufsmount *ump; 911 struct fs *fs; 912 struct cdev *dev; 913 int error, i, len, ronly; 914 struct ucred *cred; 915 struct g_consumer *cp; 916 struct mount *nmp; 917 struct vnode *devvp; 918 int candelete, canspeedup; 919 off_t loc; 920 921 fs = NULL; 922 ump = NULL; 923 cred = td ? td->td_ucred : NOCRED; 924 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 925 926 devvp = mntfs_allocvp(mp, odevvp); 927 VOP_UNLOCK(odevvp); 928 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 929 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); 930 dev = devvp->v_rdev; 931 KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data")); 932 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, 933 (uintptr_t)mp) == 0) { 934 mntfs_freevp(devvp); 935 return (EBUSY); 936 } 937 g_topology_lock(); 938 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 939 g_topology_unlock(); 940 if (error != 0) { 941 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 942 mntfs_freevp(devvp); 943 return (error); 944 } 945 dev_ref(dev); 946 devvp->v_bufobj.bo_ops = &ffs_ops; 947 BO_LOCK(&odevvp->v_bufobj); 948 odevvp->v_bufobj.bo_flag |= BO_NOBUFS; 949 BO_UNLOCK(&odevvp->v_bufobj); 950 VOP_UNLOCK(devvp); 951 if (dev->si_iosize_max != 0) 952 mp->mnt_iosize_max = dev->si_iosize_max; 953 if (mp->mnt_iosize_max > maxphys) 954 mp->mnt_iosize_max = maxphys; 955 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 956 error = EINVAL; 957 vfs_mount_error(mp, 958 "Invalid sectorsize %d for superblock size %d", 959 cp->provider->sectorsize, SBLOCKSIZE); 960 goto out; 961 } 962 /* fetch the superblock and summary information */ 963 loc = STDSB; 964 if ((mp->mnt_flag & (MNT_ROOTFS | MNT_FORCE)) != 0) 965 loc = STDSB_NOHASHFAIL; 966 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0) 967 goto out; 968 fs->fs_flags &= ~FS_UNCLEAN; 969 if (fs->fs_clean == 0) { 970 fs->fs_flags |= FS_UNCLEAN; 971 if (ronly || (mp->mnt_flag & MNT_FORCE) || 972 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 973 (fs->fs_flags & FS_DOSOFTDEP))) { 974 printf("WARNING: %s was not properly dismounted\n", 975 fs->fs_fsmnt); 976 } else { 977 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 978 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 979 (fs->fs_flags & FS_SUJ) == 0 ? "" : 980 " Forced mount will invalidate journal contents"); 981 error = EPERM; 982 goto out; 983 } 984 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 985 (mp->mnt_flag & MNT_FORCE)) { 986 printf("WARNING: %s: lost blocks %jd files %d\n", 987 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 988 fs->fs_pendinginodes); 989 fs->fs_pendingblocks = 0; 990 fs->fs_pendinginodes = 0; 991 } 992 } 993 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 994 printf("WARNING: %s: mount pending error: blocks %jd " 995 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 996 fs->fs_pendinginodes); 997 fs->fs_pendingblocks = 0; 998 fs->fs_pendinginodes = 0; 999 } 1000 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 1001 #ifdef UFS_GJOURNAL 1002 /* 1003 * Get journal provider name. 1004 */ 1005 len = 1024; 1006 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); 1007 if (g_io_getattr("GJOURNAL::provider", cp, &len, 1008 mp->mnt_gjprovider) == 0) { 1009 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, 1010 M_UFSMNT, M_WAITOK); 1011 MNT_ILOCK(mp); 1012 mp->mnt_flag |= MNT_GJOURNAL; 1013 MNT_IUNLOCK(mp); 1014 } else { 1015 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1016 printf("WARNING: %s: GJOURNAL flag on fs " 1017 "but no gjournal provider below\n", 1018 mp->mnt_stat.f_mntonname); 1019 free(mp->mnt_gjprovider, M_UFSMNT); 1020 mp->mnt_gjprovider = NULL; 1021 } 1022 #else 1023 printf("WARNING: %s: GJOURNAL flag on fs but no " 1024 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 1025 #endif 1026 } else { 1027 mp->mnt_gjprovider = NULL; 1028 } 1029 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1030 ump->um_cp = cp; 1031 ump->um_bo = &devvp->v_bufobj; 1032 ump->um_fs = fs; 1033 if (fs->fs_magic == FS_UFS1_MAGIC) { 1034 ump->um_fstype = UFS1; 1035 ump->um_balloc = ffs_balloc_ufs1; 1036 } else { 1037 ump->um_fstype = UFS2; 1038 ump->um_balloc = ffs_balloc_ufs2; 1039 } 1040 ump->um_blkatoff = ffs_blkatoff; 1041 ump->um_truncate = ffs_truncate; 1042 ump->um_update = ffs_update; 1043 ump->um_valloc = ffs_valloc; 1044 ump->um_vfree = ffs_vfree; 1045 ump->um_ifree = ffs_ifree; 1046 ump->um_rdonly = ffs_rdonly; 1047 ump->um_snapgone = ffs_snapgone; 1048 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) 1049 ump->um_check_blkno = ffs_check_blkno; 1050 else 1051 ump->um_check_blkno = NULL; 1052 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 1053 sx_init(&ump->um_checkpath_lock, "uchpth"); 1054 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); 1055 fs->fs_ronly = ronly; 1056 fs->fs_active = NULL; 1057 mp->mnt_data = ump; 1058 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 1059 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 1060 nmp = NULL; 1061 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 1062 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 1063 if (nmp) 1064 vfs_rel(nmp); 1065 vfs_getnewfsid(mp); 1066 } 1067 ump->um_maxsymlinklen = fs->fs_maxsymlinklen; 1068 MNT_ILOCK(mp); 1069 mp->mnt_flag |= MNT_LOCAL; 1070 MNT_IUNLOCK(mp); 1071 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 1072 #ifdef MAC 1073 MNT_ILOCK(mp); 1074 mp->mnt_flag |= MNT_MULTILABEL; 1075 MNT_IUNLOCK(mp); 1076 #else 1077 printf("WARNING: %s: multilabel flag on fs but " 1078 "no MAC support\n", mp->mnt_stat.f_mntonname); 1079 #endif 1080 } 1081 if ((fs->fs_flags & FS_ACLS) != 0) { 1082 #ifdef UFS_ACL 1083 MNT_ILOCK(mp); 1084 1085 if (mp->mnt_flag & MNT_NFS4ACLS) 1086 printf("WARNING: %s: ACLs flag on fs conflicts with " 1087 "\"nfsv4acls\" mount option; option ignored\n", 1088 mp->mnt_stat.f_mntonname); 1089 mp->mnt_flag &= ~MNT_NFS4ACLS; 1090 mp->mnt_flag |= MNT_ACLS; 1091 1092 MNT_IUNLOCK(mp); 1093 #else 1094 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 1095 mp->mnt_stat.f_mntonname); 1096 #endif 1097 } 1098 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 1099 #ifdef UFS_ACL 1100 MNT_ILOCK(mp); 1101 1102 if (mp->mnt_flag & MNT_ACLS) 1103 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 1104 "with \"acls\" mount option; option ignored\n", 1105 mp->mnt_stat.f_mntonname); 1106 mp->mnt_flag &= ~MNT_ACLS; 1107 mp->mnt_flag |= MNT_NFS4ACLS; 1108 1109 MNT_IUNLOCK(mp); 1110 #else 1111 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 1112 "ACLs support\n", mp->mnt_stat.f_mntonname); 1113 #endif 1114 } 1115 if ((fs->fs_flags & FS_TRIM) != 0) { 1116 len = sizeof(int); 1117 if (g_io_getattr("GEOM::candelete", cp, &len, 1118 &candelete) == 0) { 1119 if (candelete) 1120 ump->um_flags |= UM_CANDELETE; 1121 else 1122 printf("WARNING: %s: TRIM flag on fs but disk " 1123 "does not support TRIM\n", 1124 mp->mnt_stat.f_mntonname); 1125 } else { 1126 printf("WARNING: %s: TRIM flag on fs but disk does " 1127 "not confirm that it supports TRIM\n", 1128 mp->mnt_stat.f_mntonname); 1129 } 1130 if (((ump->um_flags) & UM_CANDELETE) != 0) { 1131 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1132 taskqueue_thread_enqueue, &ump->um_trim_tq); 1133 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1134 "%s trim", mp->mnt_stat.f_mntonname); 1135 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, 1136 &ump->um_trimlisthashsize); 1137 } 1138 } 1139 1140 len = sizeof(int); 1141 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) { 1142 if (canspeedup) 1143 ump->um_flags |= UM_CANSPEEDUP; 1144 } 1145 1146 ump->um_mountp = mp; 1147 ump->um_dev = dev; 1148 ump->um_devvp = devvp; 1149 ump->um_odevvp = odevvp; 1150 ump->um_nindir = fs->fs_nindir; 1151 ump->um_bptrtodb = fs->fs_fsbtodb; 1152 ump->um_seqinc = fs->fs_frag; 1153 for (i = 0; i < MAXQUOTAS; i++) 1154 ump->um_quotas[i] = NULLVP; 1155 #ifdef UFS_EXTATTR 1156 ufs_extattr_uepm_init(&ump->um_extattr); 1157 #endif 1158 /* 1159 * Set FS local "last mounted on" information (NULL pad) 1160 */ 1161 bzero(fs->fs_fsmnt, MAXMNTLEN); 1162 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1163 mp->mnt_stat.f_iosize = fs->fs_bsize; 1164 1165 if (mp->mnt_flag & MNT_ROOTFS) { 1166 /* 1167 * Root mount; update timestamp in mount structure. 1168 * this will be used by the common root mount code 1169 * to update the system clock. 1170 */ 1171 mp->mnt_time = fs->fs_time; 1172 } 1173 1174 if (ronly == 0) { 1175 fs->fs_mtime = time_second; 1176 if ((fs->fs_flags & FS_DOSOFTDEP) && 1177 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1178 ffs_flushfiles(mp, FORCECLOSE, td); 1179 goto out; 1180 } 1181 if (fs->fs_snapinum[0] != 0) 1182 ffs_snapshot_mount(mp); 1183 fs->fs_fmod = 1; 1184 fs->fs_clean = 0; 1185 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1186 } 1187 /* 1188 * Initialize filesystem state information in mount struct. 1189 */ 1190 MNT_ILOCK(mp); 1191 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1192 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1193 MNT_IUNLOCK(mp); 1194 #ifdef UFS_EXTATTR 1195 #ifdef UFS_EXTATTR_AUTOSTART 1196 /* 1197 * 1198 * Auto-starting does the following: 1199 * - check for /.attribute in the fs, and extattr_start if so 1200 * - for each file in .attribute, enable that file with 1201 * an attribute of the same name. 1202 * Not clear how to report errors -- probably eat them. 1203 * This would all happen while the filesystem was busy/not 1204 * available, so would effectively be "atomic". 1205 */ 1206 (void) ufs_extattr_autostart(mp, td); 1207 #endif /* !UFS_EXTATTR_AUTOSTART */ 1208 #endif /* !UFS_EXTATTR */ 1209 return (0); 1210 out: 1211 if (fs != NULL) { 1212 free(fs->fs_csp, M_UFSMNT); 1213 free(fs->fs_si, M_UFSMNT); 1214 free(fs, M_UFSMNT); 1215 } 1216 if (cp != NULL) { 1217 g_topology_lock(); 1218 g_vfs_close(cp); 1219 g_topology_unlock(); 1220 } 1221 if (ump != NULL) { 1222 mtx_destroy(UFS_MTX(ump)); 1223 sx_destroy(&ump->um_checkpath_lock); 1224 if (mp->mnt_gjprovider != NULL) { 1225 free(mp->mnt_gjprovider, M_UFSMNT); 1226 mp->mnt_gjprovider = NULL; 1227 } 1228 MPASS(ump->um_softdep == NULL); 1229 free(ump, M_UFSMNT); 1230 mp->mnt_data = NULL; 1231 } 1232 BO_LOCK(&odevvp->v_bufobj); 1233 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1234 BO_UNLOCK(&odevvp->v_bufobj); 1235 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1236 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1237 mntfs_freevp(devvp); 1238 dev_rel(dev); 1239 return (error); 1240 } 1241 1242 /* 1243 * A read function for use by filesystem-layer routines. 1244 */ 1245 static int 1246 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) 1247 { 1248 struct buf *bp; 1249 int error; 1250 1251 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); 1252 *bufp = malloc(size, M_UFSMNT, M_WAITOK); 1253 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, 1254 &bp)) != 0) 1255 return (error); 1256 bcopy(bp->b_data, *bufp, size); 1257 bp->b_flags |= B_INVAL | B_NOCACHE; 1258 brelse(bp); 1259 return (0); 1260 } 1261 1262 static int bigcgs = 0; 1263 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1264 1265 /* 1266 * Sanity checks for loading old filesystem superblocks. 1267 * See ffs_oldfscompat_write below for unwound actions. 1268 * 1269 * XXX - Parts get retired eventually. 1270 * Unfortunately new bits get added. 1271 */ 1272 static void 1273 ffs_oldfscompat_read(fs, ump, sblockloc) 1274 struct fs *fs; 1275 struct ufsmount *ump; 1276 ufs2_daddr_t sblockloc; 1277 { 1278 off_t maxfilesize; 1279 1280 /* 1281 * If not yet done, update fs_flags location and value of fs_sblockloc. 1282 */ 1283 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1284 fs->fs_flags = fs->fs_old_flags; 1285 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1286 fs->fs_sblockloc = sblockloc; 1287 } 1288 /* 1289 * If not yet done, update UFS1 superblock with new wider fields. 1290 */ 1291 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1292 fs->fs_maxbsize = fs->fs_bsize; 1293 fs->fs_time = fs->fs_old_time; 1294 fs->fs_size = fs->fs_old_size; 1295 fs->fs_dsize = fs->fs_old_dsize; 1296 fs->fs_csaddr = fs->fs_old_csaddr; 1297 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1298 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1299 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1300 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1301 } 1302 if (fs->fs_magic == FS_UFS1_MAGIC && 1303 fs->fs_old_inodefmt < FS_44INODEFMT) { 1304 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1305 fs->fs_qbmask = ~fs->fs_bmask; 1306 fs->fs_qfmask = ~fs->fs_fmask; 1307 } 1308 if (fs->fs_magic == FS_UFS1_MAGIC) { 1309 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1310 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1311 if (fs->fs_maxfilesize > maxfilesize) 1312 fs->fs_maxfilesize = maxfilesize; 1313 } 1314 /* Compatibility for old filesystems */ 1315 if (fs->fs_avgfilesize <= 0) 1316 fs->fs_avgfilesize = AVFILESIZ; 1317 if (fs->fs_avgfpdir <= 0) 1318 fs->fs_avgfpdir = AFPDIR; 1319 if (bigcgs) { 1320 fs->fs_save_cgsize = fs->fs_cgsize; 1321 fs->fs_cgsize = fs->fs_bsize; 1322 } 1323 } 1324 1325 /* 1326 * Unwinding superblock updates for old filesystems. 1327 * See ffs_oldfscompat_read above for details. 1328 * 1329 * XXX - Parts get retired eventually. 1330 * Unfortunately new bits get added. 1331 */ 1332 void 1333 ffs_oldfscompat_write(fs, ump) 1334 struct fs *fs; 1335 struct ufsmount *ump; 1336 { 1337 1338 /* 1339 * Copy back UFS2 updated fields that UFS1 inspects. 1340 */ 1341 if (fs->fs_magic == FS_UFS1_MAGIC) { 1342 fs->fs_old_time = fs->fs_time; 1343 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1344 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1345 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1346 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1347 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1348 } 1349 if (bigcgs) { 1350 fs->fs_cgsize = fs->fs_save_cgsize; 1351 fs->fs_save_cgsize = 0; 1352 } 1353 } 1354 1355 /* 1356 * unmount system call 1357 */ 1358 static int 1359 ffs_unmount(mp, mntflags) 1360 struct mount *mp; 1361 int mntflags; 1362 { 1363 struct thread *td; 1364 struct ufsmount *ump = VFSTOUFS(mp); 1365 struct fs *fs; 1366 int error, flags, susp; 1367 #ifdef UFS_EXTATTR 1368 int e_restart; 1369 #endif 1370 1371 flags = 0; 1372 td = curthread; 1373 fs = ump->um_fs; 1374 if (mntflags & MNT_FORCE) 1375 flags |= FORCECLOSE; 1376 susp = fs->fs_ronly == 0; 1377 #ifdef UFS_EXTATTR 1378 if ((error = ufs_extattr_stop(mp, td))) { 1379 if (error != EOPNOTSUPP) 1380 printf("WARNING: unmount %s: ufs_extattr_stop " 1381 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1382 error); 1383 e_restart = 0; 1384 } else { 1385 ufs_extattr_uepm_destroy(&ump->um_extattr); 1386 e_restart = 1; 1387 } 1388 #endif 1389 if (susp) { 1390 error = vfs_write_suspend_umnt(mp); 1391 if (error != 0) 1392 goto fail1; 1393 } 1394 if (MOUNTEDSOFTDEP(mp)) 1395 error = softdep_flushfiles(mp, flags, td); 1396 else 1397 error = ffs_flushfiles(mp, flags, td); 1398 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) 1399 goto fail; 1400 1401 UFS_LOCK(ump); 1402 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1403 printf("WARNING: unmount %s: pending error: blocks %jd " 1404 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1405 fs->fs_pendinginodes); 1406 fs->fs_pendingblocks = 0; 1407 fs->fs_pendinginodes = 0; 1408 } 1409 UFS_UNLOCK(ump); 1410 if (MOUNTEDSOFTDEP(mp)) 1411 softdep_unmount(mp); 1412 MPASS(ump->um_softdep == NULL); 1413 if (fs->fs_ronly == 0) { 1414 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1415 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1416 if (ffs_fsfail_cleanup(ump, error)) 1417 error = 0; 1418 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) { 1419 fs->fs_clean = 0; 1420 goto fail; 1421 } 1422 } 1423 if (susp) 1424 vfs_write_resume(mp, VR_START_WRITE); 1425 if (ump->um_trim_tq != NULL) { 1426 while (ump->um_trim_inflight != 0) 1427 pause("ufsutr", hz); 1428 taskqueue_drain_all(ump->um_trim_tq); 1429 taskqueue_free(ump->um_trim_tq); 1430 free (ump->um_trimhash, M_TRIM); 1431 } 1432 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1433 g_topology_lock(); 1434 g_vfs_close(ump->um_cp); 1435 g_topology_unlock(); 1436 BO_LOCK(&ump->um_odevvp->v_bufobj); 1437 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1438 BO_UNLOCK(&ump->um_odevvp->v_bufobj); 1439 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1440 mntfs_freevp(ump->um_devvp); 1441 vrele(ump->um_odevvp); 1442 dev_rel(ump->um_dev); 1443 mtx_destroy(UFS_MTX(ump)); 1444 sx_destroy(&ump->um_checkpath_lock); 1445 if (mp->mnt_gjprovider != NULL) { 1446 free(mp->mnt_gjprovider, M_UFSMNT); 1447 mp->mnt_gjprovider = NULL; 1448 } 1449 free(fs->fs_csp, M_UFSMNT); 1450 free(fs->fs_si, M_UFSMNT); 1451 free(fs, M_UFSMNT); 1452 free(ump, M_UFSMNT); 1453 mp->mnt_data = NULL; 1454 MNT_ILOCK(mp); 1455 mp->mnt_flag &= ~MNT_LOCAL; 1456 MNT_IUNLOCK(mp); 1457 if (td->td_su == mp) { 1458 td->td_su = NULL; 1459 vfs_rel(mp); 1460 } 1461 return (error); 1462 1463 fail: 1464 if (susp) 1465 vfs_write_resume(mp, VR_START_WRITE); 1466 fail1: 1467 #ifdef UFS_EXTATTR 1468 if (e_restart) { 1469 ufs_extattr_uepm_init(&ump->um_extattr); 1470 #ifdef UFS_EXTATTR_AUTOSTART 1471 (void) ufs_extattr_autostart(mp, td); 1472 #endif 1473 } 1474 #endif 1475 1476 return (error); 1477 } 1478 1479 /* 1480 * Flush out all the files in a filesystem. 1481 */ 1482 int 1483 ffs_flushfiles(mp, flags, td) 1484 struct mount *mp; 1485 int flags; 1486 struct thread *td; 1487 { 1488 struct ufsmount *ump; 1489 int qerror, error; 1490 1491 ump = VFSTOUFS(mp); 1492 qerror = 0; 1493 #ifdef QUOTA 1494 if (mp->mnt_flag & MNT_QUOTA) { 1495 int i; 1496 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1497 if (error) 1498 return (error); 1499 for (i = 0; i < MAXQUOTAS; i++) { 1500 error = quotaoff(td, mp, i); 1501 if (error != 0) { 1502 if ((flags & EARLYFLUSH) == 0) 1503 return (error); 1504 else 1505 qerror = error; 1506 } 1507 } 1508 1509 /* 1510 * Here we fall through to vflush again to ensure that 1511 * we have gotten rid of all the system vnodes, unless 1512 * quotas must not be closed. 1513 */ 1514 } 1515 #endif 1516 /* devvp is not locked there */ 1517 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1518 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1519 return (error); 1520 ffs_snapshot_unmount(mp); 1521 flags |= FORCECLOSE; 1522 /* 1523 * Here we fall through to vflush again to ensure 1524 * that we have gotten rid of all the system vnodes. 1525 */ 1526 } 1527 1528 /* 1529 * Do not close system files if quotas were not closed, to be 1530 * able to sync the remaining dquots. The freeblks softupdate 1531 * workitems might hold a reference on a dquot, preventing 1532 * quotaoff() from completing. Next round of 1533 * softdep_flushworklist() iteration should process the 1534 * blockers, allowing the next run of quotaoff() to finally 1535 * flush held dquots. 1536 * 1537 * Otherwise, flush all the files. 1538 */ 1539 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1540 return (error); 1541 1542 /* 1543 * Flush filesystem metadata. 1544 */ 1545 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1546 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1547 VOP_UNLOCK(ump->um_devvp); 1548 return (error); 1549 } 1550 1551 /* 1552 * Get filesystem statistics. 1553 */ 1554 static int 1555 ffs_statfs(mp, sbp) 1556 struct mount *mp; 1557 struct statfs *sbp; 1558 { 1559 struct ufsmount *ump; 1560 struct fs *fs; 1561 1562 ump = VFSTOUFS(mp); 1563 fs = ump->um_fs; 1564 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1565 panic("ffs_statfs"); 1566 sbp->f_version = STATFS_VERSION; 1567 sbp->f_bsize = fs->fs_fsize; 1568 sbp->f_iosize = fs->fs_bsize; 1569 sbp->f_blocks = fs->fs_dsize; 1570 UFS_LOCK(ump); 1571 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1572 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1573 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1574 dbtofsb(fs, fs->fs_pendingblocks); 1575 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; 1576 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1577 UFS_UNLOCK(ump); 1578 sbp->f_namemax = UFS_MAXNAMLEN; 1579 return (0); 1580 } 1581 1582 static bool 1583 sync_doupdate(struct inode *ip) 1584 { 1585 1586 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1587 IN_UPDATE)) != 0); 1588 } 1589 1590 static int 1591 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) 1592 { 1593 struct inode *ip; 1594 1595 /* 1596 * Flags are safe to access because ->v_data invalidation 1597 * is held off by listmtx. 1598 */ 1599 if (vp->v_type == VNON) 1600 return (false); 1601 ip = VTOI(vp); 1602 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) 1603 return (false); 1604 return (true); 1605 } 1606 1607 /* 1608 * For a lazy sync, we only care about access times, quotas and the 1609 * superblock. Other filesystem changes are already converted to 1610 * cylinder group blocks or inode blocks updates and are written to 1611 * disk by syncer. 1612 */ 1613 static int 1614 ffs_sync_lazy(mp) 1615 struct mount *mp; 1616 { 1617 struct vnode *mvp, *vp; 1618 struct inode *ip; 1619 int allerror, error; 1620 1621 allerror = 0; 1622 if ((mp->mnt_flag & MNT_NOATIME) != 0) { 1623 #ifdef QUOTA 1624 qsync(mp); 1625 #endif 1626 goto sbupdate; 1627 } 1628 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { 1629 if (vp->v_type == VNON) { 1630 VI_UNLOCK(vp); 1631 continue; 1632 } 1633 ip = VTOI(vp); 1634 1635 /* 1636 * The IN_ACCESS flag is converted to IN_MODIFIED by 1637 * ufs_close() and ufs_getattr() by the calls to 1638 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1639 * Test also all the other timestamp flags too, to pick up 1640 * any other cases that could be missed. 1641 */ 1642 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1643 VI_UNLOCK(vp); 1644 continue; 1645 } 1646 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0) 1647 continue; 1648 #ifdef QUOTA 1649 qsyncvp(vp); 1650 #endif 1651 if (sync_doupdate(ip)) 1652 error = ffs_update(vp, 0); 1653 if (error != 0) 1654 allerror = error; 1655 vput(vp); 1656 } 1657 sbupdate: 1658 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1659 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1660 allerror = error; 1661 return (allerror); 1662 } 1663 1664 /* 1665 * Go through the disk queues to initiate sandbagged IO; 1666 * go through the inodes to write those that have been modified; 1667 * initiate the writing of the super block if it has been modified. 1668 * 1669 * Note: we are always called with the filesystem marked busy using 1670 * vfs_busy(). 1671 */ 1672 static int 1673 ffs_sync(mp, waitfor) 1674 struct mount *mp; 1675 int waitfor; 1676 { 1677 struct vnode *mvp, *vp, *devvp; 1678 struct thread *td; 1679 struct inode *ip; 1680 struct ufsmount *ump = VFSTOUFS(mp); 1681 struct fs *fs; 1682 int error, count, lockreq, allerror = 0; 1683 int suspend; 1684 int suspended; 1685 int secondary_writes; 1686 int secondary_accwrites; 1687 int softdep_deps; 1688 int softdep_accdeps; 1689 struct bufobj *bo; 1690 1691 suspend = 0; 1692 suspended = 0; 1693 td = curthread; 1694 fs = ump->um_fs; 1695 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) 1696 panic("%s: ffs_sync: modification on read-only filesystem", 1697 fs->fs_fsmnt); 1698 if (waitfor == MNT_LAZY) { 1699 if (!rebooting) 1700 return (ffs_sync_lazy(mp)); 1701 waitfor = MNT_NOWAIT; 1702 } 1703 1704 /* 1705 * Write back each (modified) inode. 1706 */ 1707 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1708 if (waitfor == MNT_SUSPEND) { 1709 suspend = 1; 1710 waitfor = MNT_WAIT; 1711 } 1712 if (waitfor == MNT_WAIT) 1713 lockreq = LK_EXCLUSIVE; 1714 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1715 loop: 1716 /* Grab snapshot of secondary write counts */ 1717 MNT_ILOCK(mp); 1718 secondary_writes = mp->mnt_secondary_writes; 1719 secondary_accwrites = mp->mnt_secondary_accwrites; 1720 MNT_IUNLOCK(mp); 1721 1722 /* Grab snapshot of softdep dependency counts */ 1723 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1724 1725 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1726 /* 1727 * Depend on the vnode interlock to keep things stable enough 1728 * for a quick test. Since there might be hundreds of 1729 * thousands of vnodes, we cannot afford even a subroutine 1730 * call unless there's a good chance that we have work to do. 1731 */ 1732 if (vp->v_type == VNON) { 1733 VI_UNLOCK(vp); 1734 continue; 1735 } 1736 ip = VTOI(vp); 1737 if ((ip->i_flag & 1738 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1739 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1740 VI_UNLOCK(vp); 1741 continue; 1742 } 1743 if ((error = vget(vp, lockreq)) != 0) { 1744 if (error == ENOENT || error == ENOLCK) { 1745 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1746 goto loop; 1747 } 1748 continue; 1749 } 1750 #ifdef QUOTA 1751 qsyncvp(vp); 1752 #endif 1753 for (;;) { 1754 error = ffs_syncvnode(vp, waitfor, 0); 1755 if (error == ERELOOKUP) 1756 continue; 1757 if (error != 0) 1758 allerror = error; 1759 break; 1760 } 1761 vput(vp); 1762 } 1763 /* 1764 * Force stale filesystem control information to be flushed. 1765 */ 1766 if (waitfor == MNT_WAIT || rebooting) { 1767 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1768 allerror = error; 1769 if (ffs_fsfail_cleanup(ump, allerror)) 1770 allerror = 0; 1771 /* Flushed work items may create new vnodes to clean */ 1772 if (allerror == 0 && count) 1773 goto loop; 1774 } 1775 1776 devvp = ump->um_devvp; 1777 bo = &devvp->v_bufobj; 1778 BO_LOCK(bo); 1779 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1780 BO_UNLOCK(bo); 1781 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1782 error = VOP_FSYNC(devvp, waitfor, td); 1783 VOP_UNLOCK(devvp); 1784 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1785 error = ffs_sbupdate(ump, waitfor, 0); 1786 if (error != 0) 1787 allerror = error; 1788 if (ffs_fsfail_cleanup(ump, allerror)) 1789 allerror = 0; 1790 if (allerror == 0 && waitfor == MNT_WAIT) 1791 goto loop; 1792 } else if (suspend != 0) { 1793 if (softdep_check_suspend(mp, 1794 devvp, 1795 softdep_deps, 1796 softdep_accdeps, 1797 secondary_writes, 1798 secondary_accwrites) != 0) { 1799 MNT_IUNLOCK(mp); 1800 goto loop; /* More work needed */ 1801 } 1802 mtx_assert(MNT_MTX(mp), MA_OWNED); 1803 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1804 MNT_IUNLOCK(mp); 1805 suspended = 1; 1806 } else 1807 BO_UNLOCK(bo); 1808 /* 1809 * Write back modified superblock. 1810 */ 1811 if (fs->fs_fmod != 0 && 1812 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1813 allerror = error; 1814 if (ffs_fsfail_cleanup(ump, allerror)) 1815 allerror = 0; 1816 return (allerror); 1817 } 1818 1819 int 1820 ffs_vget(mp, ino, flags, vpp) 1821 struct mount *mp; 1822 ino_t ino; 1823 int flags; 1824 struct vnode **vpp; 1825 { 1826 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1827 } 1828 1829 int 1830 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1831 struct mount *mp; 1832 ino_t ino; 1833 int flags; 1834 struct vnode **vpp; 1835 int ffs_flags; 1836 { 1837 struct fs *fs; 1838 struct inode *ip; 1839 struct ufsmount *ump; 1840 struct buf *bp; 1841 struct vnode *vp; 1842 daddr_t dbn; 1843 int error; 1844 1845 MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 || 1846 (flags & LK_EXCLUSIVE) != 0); 1847 1848 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1849 if (error != 0) 1850 return (error); 1851 if (*vpp != NULL) { 1852 if ((ffs_flags & FFSV_REPLACE) == 0 || 1853 ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 || 1854 !VN_IS_DOOMED(*vpp))) 1855 return (0); 1856 vgone(*vpp); 1857 vput(*vpp); 1858 } 1859 1860 /* 1861 * We must promote to an exclusive lock for vnode creation. This 1862 * can happen if lookup is passed LOCKSHARED. 1863 */ 1864 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1865 flags &= ~LK_TYPE_MASK; 1866 flags |= LK_EXCLUSIVE; 1867 } 1868 1869 /* 1870 * We do not lock vnode creation as it is believed to be too 1871 * expensive for such rare case as simultaneous creation of vnode 1872 * for same ino by different processes. We just allow them to race 1873 * and check later to decide who wins. Let the race begin! 1874 */ 1875 1876 ump = VFSTOUFS(mp); 1877 fs = ump->um_fs; 1878 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO); 1879 1880 /* Allocate a new vnode/inode. */ 1881 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1882 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1883 if (error) { 1884 *vpp = NULL; 1885 uma_zfree_smr(uma_inode, ip); 1886 return (error); 1887 } 1888 /* 1889 * FFS supports recursive locking. 1890 */ 1891 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); 1892 VN_LOCK_AREC(vp); 1893 vp->v_data = ip; 1894 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1895 ip->i_vnode = vp; 1896 ip->i_ump = ump; 1897 ip->i_number = ino; 1898 ip->i_ea_refs = 0; 1899 ip->i_nextclustercg = -1; 1900 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; 1901 ip->i_mode = 0; /* ensure error cases below throw away vnode */ 1902 cluster_init_vn(&ip->i_clusterw); 1903 #ifdef DIAGNOSTIC 1904 ufs_init_trackers(ip); 1905 #endif 1906 #ifdef QUOTA 1907 { 1908 int i; 1909 for (i = 0; i < MAXQUOTAS; i++) 1910 ip->i_dquot[i] = NODQUOT; 1911 } 1912 #endif 1913 1914 if (ffs_flags & FFSV_FORCEINSMQ) 1915 vp->v_vflag |= VV_FORCEINSMQ; 1916 error = insmntque(vp, mp); 1917 if (error != 0) { 1918 uma_zfree_smr(uma_inode, ip); 1919 *vpp = NULL; 1920 return (error); 1921 } 1922 vp->v_vflag &= ~VV_FORCEINSMQ; 1923 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1924 if (error != 0) 1925 return (error); 1926 if (*vpp != NULL) { 1927 /* 1928 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) 1929 * operate on empty inode, which must not be found by 1930 * other threads until fully filled. Vnode for empty 1931 * inode must be not re-inserted on the hash by other 1932 * thread, after removal by us at the beginning. 1933 */ 1934 MPASS((ffs_flags & FFSV_REPLACE) == 0); 1935 return (0); 1936 } 1937 1938 /* Read in the disk contents for the inode, copy into the inode. */ 1939 dbn = fsbtodb(fs, ino_to_fsba(fs, ino)); 1940 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, 1941 NULL, NULL, 0, NOCRED, 0, NULL, &bp); 1942 if (error != 0) { 1943 /* 1944 * The inode does not contain anything useful, so it would 1945 * be misleading to leave it on its hash chain. With mode 1946 * still zero, it will be unlinked and returned to the free 1947 * list by vput(). 1948 */ 1949 vgone(vp); 1950 vput(vp); 1951 *vpp = NULL; 1952 return (error); 1953 } 1954 if (I_IS_UFS1(ip)) 1955 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1956 else 1957 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1958 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { 1959 bqrelse(bp); 1960 vgone(vp); 1961 vput(vp); 1962 *vpp = NULL; 1963 return (error); 1964 } 1965 if (DOINGSOFTDEP(vp) && (!fs->fs_ronly || 1966 (ffs_flags & FFSV_FORCEINODEDEP) != 0)) 1967 softdep_load_inodeblock(ip); 1968 else 1969 ip->i_effnlink = ip->i_nlink; 1970 bqrelse(bp); 1971 1972 /* 1973 * Initialize the vnode from the inode, check for aliases. 1974 * Note that the underlying vnode may have changed. 1975 */ 1976 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, 1977 &vp); 1978 if (error) { 1979 vgone(vp); 1980 vput(vp); 1981 *vpp = NULL; 1982 return (error); 1983 } 1984 1985 /* 1986 * Finish inode initialization. 1987 */ 1988 if (vp->v_type != VFIFO) { 1989 /* FFS supports shared locking for all files except fifos. */ 1990 VN_LOCK_ASHARE(vp); 1991 } 1992 1993 /* 1994 * Set up a generation number for this inode if it does not 1995 * already have one. This should only happen on old filesystems. 1996 */ 1997 if (ip->i_gen == 0) { 1998 while (ip->i_gen == 0) 1999 ip->i_gen = arc4random(); 2000 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 2001 UFS_INODE_SET_FLAG(ip, IN_MODIFIED); 2002 DIP_SET(ip, i_gen, ip->i_gen); 2003 } 2004 } 2005 #ifdef MAC 2006 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 2007 /* 2008 * If this vnode is already allocated, and we're running 2009 * multi-label, attempt to perform a label association 2010 * from the extended attributes on the inode. 2011 */ 2012 error = mac_vnode_associate_extattr(mp, vp); 2013 if (error) { 2014 /* ufs_inactive will release ip->i_devvp ref. */ 2015 vgone(vp); 2016 vput(vp); 2017 *vpp = NULL; 2018 return (error); 2019 } 2020 } 2021 #endif 2022 2023 *vpp = vp; 2024 return (0); 2025 } 2026 2027 /* 2028 * File handle to vnode 2029 * 2030 * Have to be really careful about stale file handles: 2031 * - check that the inode number is valid 2032 * - for UFS2 check that the inode number is initialized 2033 * - call ffs_vget() to get the locked inode 2034 * - check for an unallocated inode (i_mode == 0) 2035 * - check that the given client host has export rights and return 2036 * those rights via. exflagsp and credanonp 2037 */ 2038 static int 2039 ffs_fhtovp(mp, fhp, flags, vpp) 2040 struct mount *mp; 2041 struct fid *fhp; 2042 int flags; 2043 struct vnode **vpp; 2044 { 2045 struct ufid *ufhp; 2046 2047 ufhp = (struct ufid *)fhp; 2048 return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags, 2049 vpp, 0)); 2050 } 2051 2052 int 2053 ffs_inotovp(mp, ino, gen, lflags, vpp, ffs_flags) 2054 struct mount *mp; 2055 ino_t ino; 2056 u_int64_t gen; 2057 int lflags; 2058 struct vnode **vpp; 2059 int ffs_flags; 2060 { 2061 struct ufsmount *ump; 2062 struct vnode *nvp; 2063 struct inode *ip; 2064 struct fs *fs; 2065 struct cg *cgp; 2066 struct buf *bp; 2067 u_int cg; 2068 int error; 2069 2070 ump = VFSTOUFS(mp); 2071 fs = ump->um_fs; 2072 *vpp = NULL; 2073 2074 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) 2075 return (ESTALE); 2076 2077 /* 2078 * Need to check if inode is initialized because UFS2 does lazy 2079 * initialization and nfs_fhtovp can offer arbitrary inode numbers. 2080 */ 2081 if (fs->fs_magic == FS_UFS2_MAGIC) { 2082 cg = ino_to_cg(fs, ino); 2083 error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp); 2084 if (error != 0) 2085 return (error); 2086 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { 2087 brelse(bp); 2088 return (ESTALE); 2089 } 2090 brelse(bp); 2091 } 2092 2093 error = ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags); 2094 if (error != 0) 2095 return (error); 2096 2097 ip = VTOI(nvp); 2098 if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) { 2099 if (ip->i_mode == 0) 2100 vgone(nvp); 2101 vput(nvp); 2102 return (ESTALE); 2103 } 2104 2105 vnode_create_vobject(nvp, DIP(ip, i_size), curthread); 2106 *vpp = nvp; 2107 return (0); 2108 } 2109 2110 /* 2111 * Initialize the filesystem. 2112 */ 2113 static int 2114 ffs_init(vfsp) 2115 struct vfsconf *vfsp; 2116 { 2117 2118 ffs_susp_initialize(); 2119 softdep_initialize(); 2120 return (ufs_init(vfsp)); 2121 } 2122 2123 /* 2124 * Undo the work of ffs_init(). 2125 */ 2126 static int 2127 ffs_uninit(vfsp) 2128 struct vfsconf *vfsp; 2129 { 2130 int ret; 2131 2132 ret = ufs_uninit(vfsp); 2133 softdep_uninitialize(); 2134 ffs_susp_uninitialize(); 2135 taskqueue_drain_all(taskqueue_thread); 2136 return (ret); 2137 } 2138 2139 /* 2140 * Structure used to pass information from ffs_sbupdate to its 2141 * helper routine ffs_use_bwrite. 2142 */ 2143 struct devfd { 2144 struct ufsmount *ump; 2145 struct buf *sbbp; 2146 int waitfor; 2147 int suspended; 2148 int error; 2149 }; 2150 2151 /* 2152 * Write a superblock and associated information back to disk. 2153 */ 2154 int 2155 ffs_sbupdate(ump, waitfor, suspended) 2156 struct ufsmount *ump; 2157 int waitfor; 2158 int suspended; 2159 { 2160 struct fs *fs; 2161 struct buf *sbbp; 2162 struct devfd devfd; 2163 2164 fs = ump->um_fs; 2165 if (fs->fs_ronly == 1 && 2166 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 2167 (MNT_RDONLY | MNT_UPDATE)) 2168 panic("ffs_sbupdate: write read-only filesystem"); 2169 /* 2170 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 2171 */ 2172 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 2173 (int)fs->fs_sbsize, 0, 0, 0); 2174 /* 2175 * Initialize info needed for write function. 2176 */ 2177 devfd.ump = ump; 2178 devfd.sbbp = sbbp; 2179 devfd.waitfor = waitfor; 2180 devfd.suspended = suspended; 2181 devfd.error = 0; 2182 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); 2183 } 2184 2185 /* 2186 * Write function for use by filesystem-layer routines. 2187 */ 2188 static int 2189 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) 2190 { 2191 struct devfd *devfdp; 2192 struct ufsmount *ump; 2193 struct buf *bp; 2194 struct fs *fs; 2195 int error; 2196 2197 devfdp = devfd; 2198 ump = devfdp->ump; 2199 fs = ump->um_fs; 2200 /* 2201 * Writing the superblock summary information. 2202 */ 2203 if (loc != fs->fs_sblockloc) { 2204 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); 2205 bcopy(buf, bp->b_data, (u_int)size); 2206 if (devfdp->suspended) 2207 bp->b_flags |= B_VALIDSUSPWRT; 2208 if (devfdp->waitfor != MNT_WAIT) 2209 bawrite(bp); 2210 else if ((error = bwrite(bp)) != 0) 2211 devfdp->error = error; 2212 return (0); 2213 } 2214 /* 2215 * Writing the superblock itself. We need to do special checks for it. 2216 */ 2217 bp = devfdp->sbbp; 2218 if (ffs_fsfail_cleanup(ump, devfdp->error)) 2219 devfdp->error = 0; 2220 if (devfdp->error != 0) { 2221 brelse(bp); 2222 return (devfdp->error); 2223 } 2224 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 2225 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2226 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2227 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 2228 fs->fs_sblockloc = SBLOCK_UFS1; 2229 } 2230 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 2231 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2232 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2233 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 2234 fs->fs_sblockloc = SBLOCK_UFS2; 2235 } 2236 if (MOUNTEDSOFTDEP(ump->um_mountp)) 2237 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 2238 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 2239 fs = (struct fs *)bp->b_data; 2240 ffs_oldfscompat_write(fs, ump); 2241 fs->fs_si = NULL; 2242 /* Recalculate the superblock hash */ 2243 fs->fs_ckhash = ffs_calc_sbhash(fs); 2244 if (devfdp->suspended) 2245 bp->b_flags |= B_VALIDSUSPWRT; 2246 if (devfdp->waitfor != MNT_WAIT) 2247 bawrite(bp); 2248 else if ((error = bwrite(bp)) != 0) 2249 devfdp->error = error; 2250 return (devfdp->error); 2251 } 2252 2253 static int 2254 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 2255 int attrnamespace, const char *attrname) 2256 { 2257 2258 #ifdef UFS_EXTATTR 2259 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 2260 attrname)); 2261 #else 2262 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 2263 attrname)); 2264 #endif 2265 } 2266 2267 static void 2268 ffs_ifree(struct ufsmount *ump, struct inode *ip) 2269 { 2270 2271 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 2272 uma_zfree(uma_ufs1, ip->i_din1); 2273 else if (ip->i_din2 != NULL) 2274 uma_zfree(uma_ufs2, ip->i_din2); 2275 uma_zfree_smr(uma_inode, ip); 2276 } 2277 2278 static int dobkgrdwrite = 1; 2279 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 2280 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 2281 2282 /* 2283 * Complete a background write started from bwrite. 2284 */ 2285 static void 2286 ffs_backgroundwritedone(struct buf *bp) 2287 { 2288 struct bufobj *bufobj; 2289 struct buf *origbp; 2290 2291 #ifdef SOFTUPDATES 2292 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0) 2293 softdep_handle_error(bp); 2294 #endif 2295 2296 /* 2297 * Find the original buffer that we are writing. 2298 */ 2299 bufobj = bp->b_bufobj; 2300 BO_LOCK(bufobj); 2301 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2302 panic("backgroundwritedone: lost buffer"); 2303 2304 /* 2305 * We should mark the cylinder group buffer origbp as 2306 * dirty, to not lose the failed write. 2307 */ 2308 if ((bp->b_ioflags & BIO_ERROR) != 0) 2309 origbp->b_vflags |= BV_BKGRDERR; 2310 BO_UNLOCK(bufobj); 2311 /* 2312 * Process dependencies then return any unfinished ones. 2313 */ 2314 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2315 buf_complete(bp); 2316 #ifdef SOFTUPDATES 2317 if (!LIST_EMPTY(&bp->b_dep)) 2318 softdep_move_dependencies(bp, origbp); 2319 #endif 2320 /* 2321 * This buffer is marked B_NOCACHE so when it is released 2322 * by biodone it will be tossed. Clear B_IOSTARTED in case of error. 2323 */ 2324 bp->b_flags |= B_NOCACHE; 2325 bp->b_flags &= ~(B_CACHE | B_IOSTARTED); 2326 pbrelvp(bp); 2327 2328 /* 2329 * Prevent brelse() from trying to keep and re-dirtying bp on 2330 * errors. It causes b_bufobj dereference in 2331 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2332 * pbrelvp() above. 2333 */ 2334 if ((bp->b_ioflags & BIO_ERROR) != 0) 2335 bp->b_flags |= B_INVAL; 2336 bufdone(bp); 2337 BO_LOCK(bufobj); 2338 /* 2339 * Clear the BV_BKGRDINPROG flag in the original buffer 2340 * and awaken it if it is waiting for the write to complete. 2341 * If BV_BKGRDINPROG is not set in the original buffer it must 2342 * have been released and re-instantiated - which is not legal. 2343 */ 2344 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2345 ("backgroundwritedone: lost buffer2")); 2346 origbp->b_vflags &= ~BV_BKGRDINPROG; 2347 if (origbp->b_vflags & BV_BKGRDWAIT) { 2348 origbp->b_vflags &= ~BV_BKGRDWAIT; 2349 wakeup(&origbp->b_xflags); 2350 } 2351 BO_UNLOCK(bufobj); 2352 } 2353 2354 /* 2355 * Write, release buffer on completion. (Done by iodone 2356 * if async). Do not bother writing anything if the buffer 2357 * is invalid. 2358 * 2359 * Note that we set B_CACHE here, indicating that buffer is 2360 * fully valid and thus cacheable. This is true even of NFS 2361 * now so we set it generally. This could be set either here 2362 * or in biodone() since the I/O is synchronous. We put it 2363 * here. 2364 */ 2365 static int 2366 ffs_bufwrite(struct buf *bp) 2367 { 2368 struct buf *newbp; 2369 struct cg *cgp; 2370 2371 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2372 if (bp->b_flags & B_INVAL) { 2373 brelse(bp); 2374 return (0); 2375 } 2376 2377 if (!BUF_ISLOCKED(bp)) 2378 panic("bufwrite: buffer is not busy???"); 2379 /* 2380 * If a background write is already in progress, delay 2381 * writing this block if it is asynchronous. Otherwise 2382 * wait for the background write to complete. 2383 */ 2384 BO_LOCK(bp->b_bufobj); 2385 if (bp->b_vflags & BV_BKGRDINPROG) { 2386 if (bp->b_flags & B_ASYNC) { 2387 BO_UNLOCK(bp->b_bufobj); 2388 bdwrite(bp); 2389 return (0); 2390 } 2391 bp->b_vflags |= BV_BKGRDWAIT; 2392 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2393 "bwrbg", 0); 2394 if (bp->b_vflags & BV_BKGRDINPROG) 2395 panic("bufwrite: still writing"); 2396 } 2397 bp->b_vflags &= ~BV_BKGRDERR; 2398 BO_UNLOCK(bp->b_bufobj); 2399 2400 /* 2401 * If this buffer is marked for background writing and we 2402 * do not have to wait for it, make a copy and write the 2403 * copy so as to leave this buffer ready for further use. 2404 * 2405 * This optimization eats a lot of memory. If we have a page 2406 * or buffer shortfall we can't do it. 2407 */ 2408 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2409 (bp->b_flags & B_ASYNC) && 2410 !vm_page_count_severe() && 2411 !buf_dirty_count_severe()) { 2412 KASSERT(bp->b_iodone == NULL, 2413 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2414 2415 /* get a new block */ 2416 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2417 if (newbp == NULL) 2418 goto normal_write; 2419 2420 KASSERT(buf_mapped(bp), ("Unmapped cg")); 2421 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2422 BO_LOCK(bp->b_bufobj); 2423 bp->b_vflags |= BV_BKGRDINPROG; 2424 BO_UNLOCK(bp->b_bufobj); 2425 newbp->b_xflags |= 2426 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; 2427 newbp->b_lblkno = bp->b_lblkno; 2428 newbp->b_blkno = bp->b_blkno; 2429 newbp->b_offset = bp->b_offset; 2430 newbp->b_iodone = ffs_backgroundwritedone; 2431 newbp->b_flags |= B_ASYNC; 2432 newbp->b_flags &= ~B_INVAL; 2433 pbgetvp(bp->b_vp, newbp); 2434 2435 #ifdef SOFTUPDATES 2436 /* 2437 * Move over the dependencies. If there are rollbacks, 2438 * leave the parent buffer dirtied as it will need to 2439 * be written again. 2440 */ 2441 if (LIST_EMPTY(&bp->b_dep) || 2442 softdep_move_dependencies(bp, newbp) == 0) 2443 bundirty(bp); 2444 #else 2445 bundirty(bp); 2446 #endif 2447 2448 /* 2449 * Initiate write on the copy, release the original. The 2450 * BKGRDINPROG flag prevents it from going away until 2451 * the background write completes. We have to recalculate 2452 * its check hash in case the buffer gets freed and then 2453 * reconstituted from the buffer cache during a later read. 2454 */ 2455 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2456 cgp = (struct cg *)bp->b_data; 2457 cgp->cg_ckhash = 0; 2458 cgp->cg_ckhash = 2459 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2460 } 2461 bqrelse(bp); 2462 bp = newbp; 2463 } else 2464 /* Mark the buffer clean */ 2465 bundirty(bp); 2466 2467 /* Let the normal bufwrite do the rest for us */ 2468 normal_write: 2469 /* 2470 * If we are writing a cylinder group, update its time. 2471 */ 2472 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2473 cgp = (struct cg *)bp->b_data; 2474 cgp->cg_old_time = cgp->cg_time = time_second; 2475 } 2476 return (bufwrite(bp)); 2477 } 2478 2479 static void 2480 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2481 { 2482 struct vnode *vp; 2483 struct buf *tbp; 2484 int error, nocopy; 2485 2486 /* 2487 * This is the bufobj strategy for the private VCHR vnodes 2488 * used by FFS to access the underlying storage device. 2489 * We override the default bufobj strategy and thus bypass 2490 * VOP_STRATEGY() for these vnodes. 2491 */ 2492 vp = bo2vnode(bo); 2493 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR || 2494 bp->b_vp->v_rdev == NULL || 2495 bp->b_vp->v_rdev->si_mountpt == NULL || 2496 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL || 2497 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp, 2498 ("ffs_geom_strategy() with wrong vp")); 2499 if (bp->b_iocmd == BIO_WRITE) { 2500 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2501 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2502 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2503 panic("ffs_geom_strategy: bad I/O"); 2504 nocopy = bp->b_flags & B_NOCOPY; 2505 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2506 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2507 vp->v_rdev->si_snapdata != NULL) { 2508 if ((bp->b_flags & B_CLUSTER) != 0) { 2509 runningbufwakeup(bp); 2510 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2511 b_cluster.cluster_entry) { 2512 error = ffs_copyonwrite(vp, tbp); 2513 if (error != 0 && 2514 error != EOPNOTSUPP) { 2515 bp->b_error = error; 2516 bp->b_ioflags |= BIO_ERROR; 2517 bp->b_flags &= ~B_BARRIER; 2518 bufdone(bp); 2519 return; 2520 } 2521 } 2522 bp->b_runningbufspace = bp->b_bufsize; 2523 atomic_add_long(&runningbufspace, 2524 bp->b_runningbufspace); 2525 } else { 2526 error = ffs_copyonwrite(vp, bp); 2527 if (error != 0 && error != EOPNOTSUPP) { 2528 bp->b_error = error; 2529 bp->b_ioflags |= BIO_ERROR; 2530 bp->b_flags &= ~B_BARRIER; 2531 bufdone(bp); 2532 return; 2533 } 2534 } 2535 } 2536 #ifdef SOFTUPDATES 2537 if ((bp->b_flags & B_CLUSTER) != 0) { 2538 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2539 b_cluster.cluster_entry) { 2540 if (!LIST_EMPTY(&tbp->b_dep)) 2541 buf_start(tbp); 2542 } 2543 } else { 2544 if (!LIST_EMPTY(&bp->b_dep)) 2545 buf_start(bp); 2546 } 2547 2548 #endif 2549 /* 2550 * Check for metadata that needs check-hashes and update them. 2551 */ 2552 switch (bp->b_xflags & BX_FSPRIV) { 2553 case BX_CYLGRP: 2554 ((struct cg *)bp->b_data)->cg_ckhash = 0; 2555 ((struct cg *)bp->b_data)->cg_ckhash = 2556 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2557 break; 2558 2559 case BX_SUPERBLOCK: 2560 case BX_INODE: 2561 case BX_INDIR: 2562 case BX_DIR: 2563 printf("Check-hash write is unimplemented!!!\n"); 2564 break; 2565 2566 case 0: 2567 break; 2568 2569 default: 2570 printf("multiple buffer types 0x%b\n", 2571 (u_int)(bp->b_xflags & BX_FSPRIV), 2572 PRINT_UFS_BUF_XFLAGS); 2573 break; 2574 } 2575 } 2576 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable) 2577 bp->b_xflags |= BX_CVTENXIO; 2578 g_vfs_strategy(bo, bp); 2579 } 2580 2581 int 2582 ffs_own_mount(const struct mount *mp) 2583 { 2584 2585 if (mp->mnt_op == &ufs_vfsops) 2586 return (1); 2587 return (0); 2588 } 2589 2590 #ifdef DDB 2591 #ifdef SOFTUPDATES 2592 2593 /* defined in ffs_softdep.c */ 2594 extern void db_print_ffs(struct ufsmount *ump); 2595 2596 DB_SHOW_COMMAND(ffs, db_show_ffs) 2597 { 2598 struct mount *mp; 2599 struct ufsmount *ump; 2600 2601 if (have_addr) { 2602 ump = VFSTOUFS((struct mount *)addr); 2603 db_print_ffs(ump); 2604 return; 2605 } 2606 2607 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2608 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2609 db_print_ffs(VFSTOUFS(mp)); 2610 } 2611 } 2612 2613 #endif /* SOFTUPDATES */ 2614 #endif /* DDB */ 2615