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 /* none of these types of check-hashes are maintained by this kernel */ 1074 fs->fs_metackhash &= ~(CK_INDIR | CK_DIR); 1075 /* no support for any undefined flags */ 1076 fs->fs_flags &= FS_SUPPORTED; 1077 fs->fs_flags &= ~FS_UNCLEAN; 1078 if (fs->fs_clean == 0) { 1079 fs->fs_flags |= FS_UNCLEAN; 1080 if (ronly || (mp->mnt_flag & MNT_FORCE) || 1081 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 1082 (fs->fs_flags & FS_DOSOFTDEP))) { 1083 printf("WARNING: %s was not properly dismounted\n", 1084 fs->fs_fsmnt); 1085 } else { 1086 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 1087 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 1088 (fs->fs_flags & FS_SUJ) == 0 ? "" : 1089 " Forced mount will invalidate journal contents"); 1090 error = EPERM; 1091 goto out; 1092 } 1093 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 1094 (mp->mnt_flag & MNT_FORCE)) { 1095 printf("WARNING: %s: lost blocks %jd files %d\n", 1096 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1097 fs->fs_pendinginodes); 1098 fs->fs_pendingblocks = 0; 1099 fs->fs_pendinginodes = 0; 1100 } 1101 } 1102 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1103 printf("WARNING: %s: mount pending error: blocks %jd " 1104 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1105 fs->fs_pendinginodes); 1106 fs->fs_pendingblocks = 0; 1107 fs->fs_pendinginodes = 0; 1108 } 1109 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 1110 #ifdef UFS_GJOURNAL 1111 /* 1112 * Get journal provider name. 1113 */ 1114 len = 1024; 1115 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); 1116 if (g_io_getattr("GJOURNAL::provider", cp, &len, 1117 mp->mnt_gjprovider) == 0) { 1118 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, 1119 M_UFSMNT, M_WAITOK); 1120 MNT_ILOCK(mp); 1121 mp->mnt_flag |= MNT_GJOURNAL; 1122 MNT_IUNLOCK(mp); 1123 } else { 1124 printf("WARNING: %s: GJOURNAL flag on fs " 1125 "but no gjournal provider below\n", 1126 mp->mnt_stat.f_mntonname); 1127 free(mp->mnt_gjprovider, M_UFSMNT); 1128 mp->mnt_gjprovider = NULL; 1129 } 1130 #else 1131 printf("WARNING: %s: GJOURNAL flag on fs but no " 1132 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 1133 #endif 1134 } else { 1135 mp->mnt_gjprovider = NULL; 1136 } 1137 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1138 ump->um_cp = cp; 1139 ump->um_bo = &devvp->v_bufobj; 1140 ump->um_fs = fs; 1141 if (fs->fs_magic == FS_UFS1_MAGIC) { 1142 ump->um_fstype = UFS1; 1143 ump->um_balloc = ffs_balloc_ufs1; 1144 } else { 1145 ump->um_fstype = UFS2; 1146 ump->um_balloc = ffs_balloc_ufs2; 1147 } 1148 ump->um_blkatoff = ffs_blkatoff; 1149 ump->um_truncate = ffs_truncate; 1150 ump->um_update = ffs_update; 1151 ump->um_valloc = ffs_valloc; 1152 ump->um_vfree = ffs_vfree; 1153 ump->um_ifree = ffs_ifree; 1154 ump->um_rdonly = ffs_rdonly; 1155 ump->um_snapgone = ffs_snapgone; 1156 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) 1157 ump->um_check_blkno = ffs_check_blkno; 1158 else 1159 ump->um_check_blkno = NULL; 1160 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 1161 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); 1162 fs->fs_ronly = ronly; 1163 fs->fs_active = NULL; 1164 mp->mnt_data = ump; 1165 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 1166 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 1167 nmp = NULL; 1168 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 1169 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 1170 if (nmp) 1171 vfs_rel(nmp); 1172 vfs_getnewfsid(mp); 1173 } 1174 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 1175 MNT_ILOCK(mp); 1176 mp->mnt_flag |= MNT_LOCAL; 1177 MNT_IUNLOCK(mp); 1178 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 1179 #ifdef MAC 1180 MNT_ILOCK(mp); 1181 mp->mnt_flag |= MNT_MULTILABEL; 1182 MNT_IUNLOCK(mp); 1183 #else 1184 printf("WARNING: %s: multilabel flag on fs but " 1185 "no MAC support\n", mp->mnt_stat.f_mntonname); 1186 #endif 1187 } 1188 if ((fs->fs_flags & FS_ACLS) != 0) { 1189 #ifdef UFS_ACL 1190 MNT_ILOCK(mp); 1191 1192 if (mp->mnt_flag & MNT_NFS4ACLS) 1193 printf("WARNING: %s: ACLs flag on fs conflicts with " 1194 "\"nfsv4acls\" mount option; option ignored\n", 1195 mp->mnt_stat.f_mntonname); 1196 mp->mnt_flag &= ~MNT_NFS4ACLS; 1197 mp->mnt_flag |= MNT_ACLS; 1198 1199 MNT_IUNLOCK(mp); 1200 #else 1201 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 1202 mp->mnt_stat.f_mntonname); 1203 #endif 1204 } 1205 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 1206 #ifdef UFS_ACL 1207 MNT_ILOCK(mp); 1208 1209 if (mp->mnt_flag & MNT_ACLS) 1210 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 1211 "with \"acls\" mount option; option ignored\n", 1212 mp->mnt_stat.f_mntonname); 1213 mp->mnt_flag &= ~MNT_ACLS; 1214 mp->mnt_flag |= MNT_NFS4ACLS; 1215 1216 MNT_IUNLOCK(mp); 1217 #else 1218 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 1219 "ACLs support\n", mp->mnt_stat.f_mntonname); 1220 #endif 1221 } 1222 if ((fs->fs_flags & FS_TRIM) != 0) { 1223 len = sizeof(int); 1224 if (g_io_getattr("GEOM::candelete", cp, &len, 1225 &candelete) == 0) { 1226 if (candelete) 1227 ump->um_flags |= UM_CANDELETE; 1228 else 1229 printf("WARNING: %s: TRIM flag on fs but disk " 1230 "does not support TRIM\n", 1231 mp->mnt_stat.f_mntonname); 1232 } else { 1233 printf("WARNING: %s: TRIM flag on fs but disk does " 1234 "not confirm that it supports TRIM\n", 1235 mp->mnt_stat.f_mntonname); 1236 } 1237 if (((ump->um_flags) & UM_CANDELETE) != 0) { 1238 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1239 taskqueue_thread_enqueue, &ump->um_trim_tq); 1240 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1241 "%s trim", mp->mnt_stat.f_mntonname); 1242 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, 1243 &ump->um_trimlisthashsize); 1244 } 1245 } 1246 1247 len = sizeof(int); 1248 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) { 1249 if (canspeedup) 1250 ump->um_flags |= UM_CANSPEEDUP; 1251 } 1252 1253 ump->um_mountp = mp; 1254 ump->um_dev = dev; 1255 ump->um_devvp = devvp; 1256 ump->um_odevvp = odevvp; 1257 ump->um_nindir = fs->fs_nindir; 1258 ump->um_bptrtodb = fs->fs_fsbtodb; 1259 ump->um_seqinc = fs->fs_frag; 1260 for (i = 0; i < MAXQUOTAS; i++) 1261 ump->um_quotas[i] = NULLVP; 1262 #ifdef UFS_EXTATTR 1263 ufs_extattr_uepm_init(&ump->um_extattr); 1264 #endif 1265 /* 1266 * Set FS local "last mounted on" information (NULL pad) 1267 */ 1268 bzero(fs->fs_fsmnt, MAXMNTLEN); 1269 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1270 mp->mnt_stat.f_iosize = fs->fs_bsize; 1271 1272 if (mp->mnt_flag & MNT_ROOTFS) { 1273 /* 1274 * Root mount; update timestamp in mount structure. 1275 * this will be used by the common root mount code 1276 * to update the system clock. 1277 */ 1278 mp->mnt_time = fs->fs_time; 1279 } 1280 1281 if (ronly == 0) { 1282 fs->fs_mtime = time_second; 1283 if ((fs->fs_flags & FS_DOSOFTDEP) && 1284 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1285 ffs_flushfiles(mp, FORCECLOSE, td); 1286 goto out; 1287 } 1288 if (fs->fs_snapinum[0] != 0) 1289 ffs_snapshot_mount(mp); 1290 fs->fs_fmod = 1; 1291 fs->fs_clean = 0; 1292 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1293 } 1294 /* 1295 * Initialize filesystem state information in mount struct. 1296 */ 1297 MNT_ILOCK(mp); 1298 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1299 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1300 MNT_IUNLOCK(mp); 1301 #ifdef UFS_EXTATTR 1302 #ifdef UFS_EXTATTR_AUTOSTART 1303 /* 1304 * 1305 * Auto-starting does the following: 1306 * - check for /.attribute in the fs, and extattr_start if so 1307 * - for each file in .attribute, enable that file with 1308 * an attribute of the same name. 1309 * Not clear how to report errors -- probably eat them. 1310 * This would all happen while the filesystem was busy/not 1311 * available, so would effectively be "atomic". 1312 */ 1313 (void) ufs_extattr_autostart(mp, td); 1314 #endif /* !UFS_EXTATTR_AUTOSTART */ 1315 #endif /* !UFS_EXTATTR */ 1316 etp = malloc(sizeof *ump->um_fsfail_task, M_UFSMNT, M_WAITOK | M_ZERO); 1317 etp->fsid = mp->mnt_stat.f_fsid; 1318 ump->um_fsfail_task = etp; 1319 return (0); 1320 out: 1321 if (fs != NULL) { 1322 free(fs->fs_csp, M_UFSMNT); 1323 free(fs->fs_si, M_UFSMNT); 1324 free(fs, M_UFSMNT); 1325 } 1326 if (cp != NULL) { 1327 g_topology_lock(); 1328 g_vfs_close(cp); 1329 g_topology_unlock(); 1330 } 1331 if (ump) { 1332 mtx_destroy(UFS_MTX(ump)); 1333 if (mp->mnt_gjprovider != NULL) { 1334 free(mp->mnt_gjprovider, M_UFSMNT); 1335 mp->mnt_gjprovider = NULL; 1336 } 1337 free(ump, M_UFSMNT); 1338 mp->mnt_data = NULL; 1339 } 1340 BO_LOCK(&odevvp->v_bufobj); 1341 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1342 BO_UNLOCK(&odevvp->v_bufobj); 1343 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1344 mntfs_freevp(devvp); 1345 dev_rel(dev); 1346 return (error); 1347 } 1348 1349 /* 1350 * A read function for use by filesystem-layer routines. 1351 */ 1352 static int 1353 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) 1354 { 1355 struct buf *bp; 1356 int error; 1357 1358 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); 1359 *bufp = malloc(size, M_UFSMNT, M_WAITOK); 1360 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, 1361 &bp)) != 0) 1362 return (error); 1363 bcopy(bp->b_data, *bufp, size); 1364 bp->b_flags |= B_INVAL | B_NOCACHE; 1365 brelse(bp); 1366 return (0); 1367 } 1368 1369 static int bigcgs = 0; 1370 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1371 1372 /* 1373 * Sanity checks for loading old filesystem superblocks. 1374 * See ffs_oldfscompat_write below for unwound actions. 1375 * 1376 * XXX - Parts get retired eventually. 1377 * Unfortunately new bits get added. 1378 */ 1379 static void 1380 ffs_oldfscompat_read(fs, ump, sblockloc) 1381 struct fs *fs; 1382 struct ufsmount *ump; 1383 ufs2_daddr_t sblockloc; 1384 { 1385 off_t maxfilesize; 1386 1387 /* 1388 * If not yet done, update fs_flags location and value of fs_sblockloc. 1389 */ 1390 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1391 fs->fs_flags = fs->fs_old_flags; 1392 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1393 fs->fs_sblockloc = sblockloc; 1394 } 1395 /* 1396 * If not yet done, update UFS1 superblock with new wider fields. 1397 */ 1398 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1399 fs->fs_maxbsize = fs->fs_bsize; 1400 fs->fs_time = fs->fs_old_time; 1401 fs->fs_size = fs->fs_old_size; 1402 fs->fs_dsize = fs->fs_old_dsize; 1403 fs->fs_csaddr = fs->fs_old_csaddr; 1404 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1405 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1406 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1407 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1408 } 1409 if (fs->fs_magic == FS_UFS1_MAGIC && 1410 fs->fs_old_inodefmt < FS_44INODEFMT) { 1411 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1412 fs->fs_qbmask = ~fs->fs_bmask; 1413 fs->fs_qfmask = ~fs->fs_fmask; 1414 } 1415 if (fs->fs_magic == FS_UFS1_MAGIC) { 1416 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1417 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1418 if (fs->fs_maxfilesize > maxfilesize) 1419 fs->fs_maxfilesize = maxfilesize; 1420 } 1421 /* Compatibility for old filesystems */ 1422 if (fs->fs_avgfilesize <= 0) 1423 fs->fs_avgfilesize = AVFILESIZ; 1424 if (fs->fs_avgfpdir <= 0) 1425 fs->fs_avgfpdir = AFPDIR; 1426 if (bigcgs) { 1427 fs->fs_save_cgsize = fs->fs_cgsize; 1428 fs->fs_cgsize = fs->fs_bsize; 1429 } 1430 } 1431 1432 /* 1433 * Unwinding superblock updates for old filesystems. 1434 * See ffs_oldfscompat_read above for details. 1435 * 1436 * XXX - Parts get retired eventually. 1437 * Unfortunately new bits get added. 1438 */ 1439 void 1440 ffs_oldfscompat_write(fs, ump) 1441 struct fs *fs; 1442 struct ufsmount *ump; 1443 { 1444 1445 /* 1446 * Copy back UFS2 updated fields that UFS1 inspects. 1447 */ 1448 if (fs->fs_magic == FS_UFS1_MAGIC) { 1449 fs->fs_old_time = fs->fs_time; 1450 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1451 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1452 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1453 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1454 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1455 } 1456 if (bigcgs) { 1457 fs->fs_cgsize = fs->fs_save_cgsize; 1458 fs->fs_save_cgsize = 0; 1459 } 1460 } 1461 1462 /* 1463 * unmount system call 1464 */ 1465 static int 1466 ffs_unmount(mp, mntflags) 1467 struct mount *mp; 1468 int mntflags; 1469 { 1470 struct thread *td; 1471 struct ufsmount *ump = VFSTOUFS(mp); 1472 struct fs *fs; 1473 int error, flags, susp; 1474 #ifdef UFS_EXTATTR 1475 int e_restart; 1476 #endif 1477 1478 flags = 0; 1479 td = curthread; 1480 fs = ump->um_fs; 1481 if (mntflags & MNT_FORCE) 1482 flags |= FORCECLOSE; 1483 susp = fs->fs_ronly == 0; 1484 #ifdef UFS_EXTATTR 1485 if ((error = ufs_extattr_stop(mp, td))) { 1486 if (error != EOPNOTSUPP) 1487 printf("WARNING: unmount %s: ufs_extattr_stop " 1488 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1489 error); 1490 e_restart = 0; 1491 } else { 1492 ufs_extattr_uepm_destroy(&ump->um_extattr); 1493 e_restart = 1; 1494 } 1495 #endif 1496 if (susp) { 1497 error = vfs_write_suspend_umnt(mp); 1498 if (error != 0) 1499 goto fail1; 1500 } 1501 if (MOUNTEDSOFTDEP(mp)) 1502 error = softdep_flushfiles(mp, flags, td); 1503 else 1504 error = ffs_flushfiles(mp, flags, td); 1505 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) 1506 goto fail; 1507 1508 UFS_LOCK(ump); 1509 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1510 printf("WARNING: unmount %s: pending error: blocks %jd " 1511 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1512 fs->fs_pendinginodes); 1513 fs->fs_pendingblocks = 0; 1514 fs->fs_pendinginodes = 0; 1515 } 1516 UFS_UNLOCK(ump); 1517 if (MOUNTEDSOFTDEP(mp)) 1518 softdep_unmount(mp); 1519 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { 1520 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1521 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1522 if (ffs_fsfail_cleanup(ump, error)) 1523 error = 0; 1524 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) { 1525 fs->fs_clean = 0; 1526 goto fail; 1527 } 1528 } 1529 if (susp) 1530 vfs_write_resume(mp, VR_START_WRITE); 1531 if (ump->um_trim_tq != NULL) { 1532 while (ump->um_trim_inflight != 0) 1533 pause("ufsutr", hz); 1534 taskqueue_drain_all(ump->um_trim_tq); 1535 taskqueue_free(ump->um_trim_tq); 1536 free (ump->um_trimhash, M_TRIM); 1537 } 1538 g_topology_lock(); 1539 if (ump->um_fsckpid > 0) { 1540 /* 1541 * Return to normal read-only mode. 1542 */ 1543 error = g_access(ump->um_cp, 0, -1, 0); 1544 ump->um_fsckpid = 0; 1545 } 1546 g_vfs_close(ump->um_cp); 1547 g_topology_unlock(); 1548 BO_LOCK(&ump->um_odevvp->v_bufobj); 1549 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1550 BO_UNLOCK(&ump->um_odevvp->v_bufobj); 1551 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1552 mntfs_freevp(ump->um_devvp); 1553 vrele(ump->um_odevvp); 1554 dev_rel(ump->um_dev); 1555 mtx_destroy(UFS_MTX(ump)); 1556 if (mp->mnt_gjprovider != NULL) { 1557 free(mp->mnt_gjprovider, M_UFSMNT); 1558 mp->mnt_gjprovider = NULL; 1559 } 1560 free(fs->fs_csp, M_UFSMNT); 1561 free(fs->fs_si, M_UFSMNT); 1562 free(fs, M_UFSMNT); 1563 if (ump->um_fsfail_task != NULL) 1564 free(ump->um_fsfail_task, M_UFSMNT); 1565 free(ump, M_UFSMNT); 1566 mp->mnt_data = NULL; 1567 MNT_ILOCK(mp); 1568 mp->mnt_flag &= ~MNT_LOCAL; 1569 MNT_IUNLOCK(mp); 1570 if (td->td_su == mp) { 1571 td->td_su = NULL; 1572 vfs_rel(mp); 1573 } 1574 return (error); 1575 1576 fail: 1577 if (susp) 1578 vfs_write_resume(mp, VR_START_WRITE); 1579 fail1: 1580 #ifdef UFS_EXTATTR 1581 if (e_restart) { 1582 ufs_extattr_uepm_init(&ump->um_extattr); 1583 #ifdef UFS_EXTATTR_AUTOSTART 1584 (void) ufs_extattr_autostart(mp, td); 1585 #endif 1586 } 1587 #endif 1588 1589 return (error); 1590 } 1591 1592 /* 1593 * Flush out all the files in a filesystem. 1594 */ 1595 int 1596 ffs_flushfiles(mp, flags, td) 1597 struct mount *mp; 1598 int flags; 1599 struct thread *td; 1600 { 1601 struct ufsmount *ump; 1602 int qerror, error; 1603 1604 ump = VFSTOUFS(mp); 1605 qerror = 0; 1606 #ifdef QUOTA 1607 if (mp->mnt_flag & MNT_QUOTA) { 1608 int i; 1609 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1610 if (error) 1611 return (error); 1612 for (i = 0; i < MAXQUOTAS; i++) { 1613 error = quotaoff(td, mp, i); 1614 if (error != 0) { 1615 if ((flags & EARLYFLUSH) == 0) 1616 return (error); 1617 else 1618 qerror = error; 1619 } 1620 } 1621 1622 /* 1623 * Here we fall through to vflush again to ensure that 1624 * we have gotten rid of all the system vnodes, unless 1625 * quotas must not be closed. 1626 */ 1627 } 1628 #endif 1629 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1630 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1631 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1632 return (error); 1633 ffs_snapshot_unmount(mp); 1634 flags |= FORCECLOSE; 1635 /* 1636 * Here we fall through to vflush again to ensure 1637 * that we have gotten rid of all the system vnodes. 1638 */ 1639 } 1640 1641 /* 1642 * Do not close system files if quotas were not closed, to be 1643 * able to sync the remaining dquots. The freeblks softupdate 1644 * workitems might hold a reference on a dquot, preventing 1645 * quotaoff() from completing. Next round of 1646 * softdep_flushworklist() iteration should process the 1647 * blockers, allowing the next run of quotaoff() to finally 1648 * flush held dquots. 1649 * 1650 * Otherwise, flush all the files. 1651 */ 1652 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1653 return (error); 1654 1655 /* 1656 * Flush filesystem metadata. 1657 */ 1658 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1659 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1660 VOP_UNLOCK(ump->um_devvp); 1661 return (error); 1662 } 1663 1664 /* 1665 * Get filesystem statistics. 1666 */ 1667 static int 1668 ffs_statfs(mp, sbp) 1669 struct mount *mp; 1670 struct statfs *sbp; 1671 { 1672 struct ufsmount *ump; 1673 struct fs *fs; 1674 1675 ump = VFSTOUFS(mp); 1676 fs = ump->um_fs; 1677 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1678 panic("ffs_statfs"); 1679 sbp->f_version = STATFS_VERSION; 1680 sbp->f_bsize = fs->fs_fsize; 1681 sbp->f_iosize = fs->fs_bsize; 1682 sbp->f_blocks = fs->fs_dsize; 1683 UFS_LOCK(ump); 1684 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1685 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1686 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1687 dbtofsb(fs, fs->fs_pendingblocks); 1688 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; 1689 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1690 UFS_UNLOCK(ump); 1691 sbp->f_namemax = UFS_MAXNAMLEN; 1692 return (0); 1693 } 1694 1695 static bool 1696 sync_doupdate(struct inode *ip) 1697 { 1698 1699 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1700 IN_UPDATE)) != 0); 1701 } 1702 1703 static int 1704 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) 1705 { 1706 struct inode *ip; 1707 1708 /* 1709 * Flags are safe to access because ->v_data invalidation 1710 * is held off by listmtx. 1711 */ 1712 if (vp->v_type == VNON) 1713 return (false); 1714 ip = VTOI(vp); 1715 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) 1716 return (false); 1717 return (true); 1718 } 1719 1720 /* 1721 * For a lazy sync, we only care about access times, quotas and the 1722 * superblock. Other filesystem changes are already converted to 1723 * cylinder group blocks or inode blocks updates and are written to 1724 * disk by syncer. 1725 */ 1726 static int 1727 ffs_sync_lazy(mp) 1728 struct mount *mp; 1729 { 1730 struct vnode *mvp, *vp; 1731 struct inode *ip; 1732 struct thread *td; 1733 int allerror, error; 1734 1735 allerror = 0; 1736 td = curthread; 1737 if ((mp->mnt_flag & MNT_NOATIME) != 0) { 1738 #ifdef QUOTA 1739 qsync(mp); 1740 #endif 1741 goto sbupdate; 1742 } 1743 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { 1744 if (vp->v_type == VNON) { 1745 VI_UNLOCK(vp); 1746 continue; 1747 } 1748 ip = VTOI(vp); 1749 1750 /* 1751 * The IN_ACCESS flag is converted to IN_MODIFIED by 1752 * ufs_close() and ufs_getattr() by the calls to 1753 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1754 * Test also all the other timestamp flags too, to pick up 1755 * any other cases that could be missed. 1756 */ 1757 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1758 VI_UNLOCK(vp); 1759 continue; 1760 } 1761 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0) 1762 continue; 1763 #ifdef QUOTA 1764 qsyncvp(vp); 1765 #endif 1766 if (sync_doupdate(ip)) 1767 error = ffs_update(vp, 0); 1768 if (error != 0) 1769 allerror = error; 1770 vput(vp); 1771 } 1772 sbupdate: 1773 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1774 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1775 allerror = error; 1776 return (allerror); 1777 } 1778 1779 /* 1780 * Go through the disk queues to initiate sandbagged IO; 1781 * go through the inodes to write those that have been modified; 1782 * initiate the writing of the super block if it has been modified. 1783 * 1784 * Note: we are always called with the filesystem marked busy using 1785 * vfs_busy(). 1786 */ 1787 static int 1788 ffs_sync(mp, waitfor) 1789 struct mount *mp; 1790 int waitfor; 1791 { 1792 struct vnode *mvp, *vp, *devvp; 1793 struct thread *td; 1794 struct inode *ip; 1795 struct ufsmount *ump = VFSTOUFS(mp); 1796 struct fs *fs; 1797 int error, count, lockreq, allerror = 0; 1798 int suspend; 1799 int suspended; 1800 int secondary_writes; 1801 int secondary_accwrites; 1802 int softdep_deps; 1803 int softdep_accdeps; 1804 struct bufobj *bo; 1805 1806 suspend = 0; 1807 suspended = 0; 1808 td = curthread; 1809 fs = ump->um_fs; 1810 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) 1811 panic("%s: ffs_sync: modification on read-only filesystem", 1812 fs->fs_fsmnt); 1813 if (waitfor == MNT_LAZY) { 1814 if (!rebooting) 1815 return (ffs_sync_lazy(mp)); 1816 waitfor = MNT_NOWAIT; 1817 } 1818 1819 /* 1820 * Write back each (modified) inode. 1821 */ 1822 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1823 if (waitfor == MNT_SUSPEND) { 1824 suspend = 1; 1825 waitfor = MNT_WAIT; 1826 } 1827 if (waitfor == MNT_WAIT) 1828 lockreq = LK_EXCLUSIVE; 1829 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1830 loop: 1831 /* Grab snapshot of secondary write counts */ 1832 MNT_ILOCK(mp); 1833 secondary_writes = mp->mnt_secondary_writes; 1834 secondary_accwrites = mp->mnt_secondary_accwrites; 1835 MNT_IUNLOCK(mp); 1836 1837 /* Grab snapshot of softdep dependency counts */ 1838 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1839 1840 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1841 /* 1842 * Depend on the vnode interlock to keep things stable enough 1843 * for a quick test. Since there might be hundreds of 1844 * thousands of vnodes, we cannot afford even a subroutine 1845 * call unless there's a good chance that we have work to do. 1846 */ 1847 if (vp->v_type == VNON) { 1848 VI_UNLOCK(vp); 1849 continue; 1850 } 1851 ip = VTOI(vp); 1852 if ((ip->i_flag & 1853 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1854 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1855 VI_UNLOCK(vp); 1856 continue; 1857 } 1858 if ((error = vget(vp, lockreq)) != 0) { 1859 if (error == ENOENT || error == ENOLCK) { 1860 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1861 goto loop; 1862 } 1863 continue; 1864 } 1865 #ifdef QUOTA 1866 qsyncvp(vp); 1867 #endif 1868 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0) 1869 allerror = error; 1870 vput(vp); 1871 } 1872 /* 1873 * Force stale filesystem control information to be flushed. 1874 */ 1875 if (waitfor == MNT_WAIT || rebooting) { 1876 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1877 allerror = error; 1878 if (ffs_fsfail_cleanup(ump, allerror)) 1879 allerror = 0; 1880 /* Flushed work items may create new vnodes to clean */ 1881 if (allerror == 0 && count) 1882 goto loop; 1883 } 1884 1885 devvp = ump->um_devvp; 1886 bo = &devvp->v_bufobj; 1887 BO_LOCK(bo); 1888 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1889 BO_UNLOCK(bo); 1890 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1891 error = VOP_FSYNC(devvp, waitfor, td); 1892 VOP_UNLOCK(devvp); 1893 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1894 error = ffs_sbupdate(ump, waitfor, 0); 1895 if (error != 0) 1896 allerror = error; 1897 if (ffs_fsfail_cleanup(ump, allerror)) 1898 allerror = 0; 1899 if (allerror == 0 && waitfor == MNT_WAIT) 1900 goto loop; 1901 } else if (suspend != 0) { 1902 if (softdep_check_suspend(mp, 1903 devvp, 1904 softdep_deps, 1905 softdep_accdeps, 1906 secondary_writes, 1907 secondary_accwrites) != 0) { 1908 MNT_IUNLOCK(mp); 1909 goto loop; /* More work needed */ 1910 } 1911 mtx_assert(MNT_MTX(mp), MA_OWNED); 1912 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1913 MNT_IUNLOCK(mp); 1914 suspended = 1; 1915 } else 1916 BO_UNLOCK(bo); 1917 /* 1918 * Write back modified superblock. 1919 */ 1920 if (fs->fs_fmod != 0 && 1921 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1922 allerror = error; 1923 if (ffs_fsfail_cleanup(ump, allerror)) 1924 allerror = 0; 1925 return (allerror); 1926 } 1927 1928 int 1929 ffs_vget(mp, ino, flags, vpp) 1930 struct mount *mp; 1931 ino_t ino; 1932 int flags; 1933 struct vnode **vpp; 1934 { 1935 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1936 } 1937 1938 int 1939 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1940 struct mount *mp; 1941 ino_t ino; 1942 int flags; 1943 struct vnode **vpp; 1944 int ffs_flags; 1945 { 1946 struct fs *fs; 1947 struct inode *ip; 1948 struct ufsmount *ump; 1949 struct buf *bp; 1950 struct vnode *vp; 1951 daddr_t dbn; 1952 int error; 1953 1954 MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0); 1955 1956 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1957 if (error != 0) 1958 return (error); 1959 if (*vpp != NULL) { 1960 if ((ffs_flags & FFSV_REPLACE) == 0) 1961 return (0); 1962 vgone(*vpp); 1963 vput(*vpp); 1964 } 1965 1966 /* 1967 * We must promote to an exclusive lock for vnode creation. This 1968 * can happen if lookup is passed LOCKSHARED. 1969 */ 1970 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1971 flags &= ~LK_TYPE_MASK; 1972 flags |= LK_EXCLUSIVE; 1973 } 1974 1975 /* 1976 * We do not lock vnode creation as it is believed to be too 1977 * expensive for such rare case as simultaneous creation of vnode 1978 * for same ino by different processes. We just allow them to race 1979 * and check later to decide who wins. Let the race begin! 1980 */ 1981 1982 ump = VFSTOUFS(mp); 1983 fs = ump->um_fs; 1984 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO); 1985 1986 /* Allocate a new vnode/inode. */ 1987 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1988 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1989 if (error) { 1990 *vpp = NULL; 1991 uma_zfree_smr(uma_inode, ip); 1992 return (error); 1993 } 1994 /* 1995 * FFS supports recursive locking. 1996 */ 1997 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1998 VN_LOCK_AREC(vp); 1999 vp->v_data = ip; 2000 vp->v_bufobj.bo_bsize = fs->fs_bsize; 2001 ip->i_vnode = vp; 2002 ip->i_ump = ump; 2003 ip->i_number = ino; 2004 ip->i_ea_refs = 0; 2005 ip->i_nextclustercg = -1; 2006 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; 2007 ip->i_mode = 0; /* ensure error cases below throw away vnode */ 2008 #ifdef QUOTA 2009 { 2010 int i; 2011 for (i = 0; i < MAXQUOTAS; i++) 2012 ip->i_dquot[i] = NODQUOT; 2013 } 2014 #endif 2015 2016 if (ffs_flags & FFSV_FORCEINSMQ) 2017 vp->v_vflag |= VV_FORCEINSMQ; 2018 error = insmntque(vp, mp); 2019 if (error != 0) { 2020 uma_zfree_smr(uma_inode, ip); 2021 *vpp = NULL; 2022 return (error); 2023 } 2024 vp->v_vflag &= ~VV_FORCEINSMQ; 2025 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 2026 if (error != 0) 2027 return (error); 2028 if (*vpp != NULL) { 2029 /* 2030 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) 2031 * operate on empty inode, which must not be found by 2032 * other threads until fully filled. Vnode for empty 2033 * inode must be not re-inserted on the hash by other 2034 * thread, after removal by us at the beginning. 2035 */ 2036 MPASS((ffs_flags & FFSV_REPLACE) == 0); 2037 return (0); 2038 } 2039 2040 /* Read in the disk contents for the inode, copy into the inode. */ 2041 dbn = fsbtodb(fs, ino_to_fsba(fs, ino)); 2042 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, 2043 NULL, NULL, 0, NOCRED, 0, NULL, &bp); 2044 if (error != 0) { 2045 /* 2046 * The inode does not contain anything useful, so it would 2047 * be misleading to leave it on its hash chain. With mode 2048 * still zero, it will be unlinked and returned to the free 2049 * list by vput(). 2050 */ 2051 vgone(vp); 2052 vput(vp); 2053 *vpp = NULL; 2054 return (error); 2055 } 2056 if (I_IS_UFS1(ip)) 2057 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 2058 else 2059 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 2060 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { 2061 bqrelse(bp); 2062 vgone(vp); 2063 vput(vp); 2064 *vpp = NULL; 2065 return (error); 2066 } 2067 if (DOINGSOFTDEP(vp)) 2068 softdep_load_inodeblock(ip); 2069 else 2070 ip->i_effnlink = ip->i_nlink; 2071 bqrelse(bp); 2072 2073 /* 2074 * Initialize the vnode from the inode, check for aliases. 2075 * Note that the underlying vnode may have changed. 2076 */ 2077 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, 2078 &vp); 2079 if (error) { 2080 vgone(vp); 2081 vput(vp); 2082 *vpp = NULL; 2083 return (error); 2084 } 2085 2086 /* 2087 * Finish inode initialization. 2088 */ 2089 if (vp->v_type != VFIFO) { 2090 /* FFS supports shared locking for all files except fifos. */ 2091 VN_LOCK_ASHARE(vp); 2092 } 2093 2094 /* 2095 * Set up a generation number for this inode if it does not 2096 * already have one. This should only happen on old filesystems. 2097 */ 2098 if (ip->i_gen == 0) { 2099 while (ip->i_gen == 0) 2100 ip->i_gen = arc4random(); 2101 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 2102 UFS_INODE_SET_FLAG(ip, IN_MODIFIED); 2103 DIP_SET(ip, i_gen, ip->i_gen); 2104 } 2105 } 2106 #ifdef MAC 2107 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 2108 /* 2109 * If this vnode is already allocated, and we're running 2110 * multi-label, attempt to perform a label association 2111 * from the extended attributes on the inode. 2112 */ 2113 error = mac_vnode_associate_extattr(mp, vp); 2114 if (error) { 2115 /* ufs_inactive will release ip->i_devvp ref. */ 2116 vgone(vp); 2117 vput(vp); 2118 *vpp = NULL; 2119 return (error); 2120 } 2121 } 2122 #endif 2123 2124 *vpp = vp; 2125 return (0); 2126 } 2127 2128 /* 2129 * File handle to vnode 2130 * 2131 * Have to be really careful about stale file handles: 2132 * - check that the inode number is valid 2133 * - for UFS2 check that the inode number is initialized 2134 * - call ffs_vget() to get the locked inode 2135 * - check for an unallocated inode (i_mode == 0) 2136 * - check that the given client host has export rights and return 2137 * those rights via. exflagsp and credanonp 2138 */ 2139 static int 2140 ffs_fhtovp(mp, fhp, flags, vpp) 2141 struct mount *mp; 2142 struct fid *fhp; 2143 int flags; 2144 struct vnode **vpp; 2145 { 2146 struct ufid *ufhp; 2147 struct ufsmount *ump; 2148 struct fs *fs; 2149 struct cg *cgp; 2150 struct buf *bp; 2151 ino_t ino; 2152 u_int cg; 2153 int error; 2154 2155 ufhp = (struct ufid *)fhp; 2156 ino = ufhp->ufid_ino; 2157 ump = VFSTOUFS(mp); 2158 fs = ump->um_fs; 2159 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) 2160 return (ESTALE); 2161 /* 2162 * Need to check if inode is initialized because UFS2 does lazy 2163 * initialization and nfs_fhtovp can offer arbitrary inode numbers. 2164 */ 2165 if (fs->fs_magic != FS_UFS2_MAGIC) 2166 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 2167 cg = ino_to_cg(fs, ino); 2168 if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) 2169 return (error); 2170 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { 2171 brelse(bp); 2172 return (ESTALE); 2173 } 2174 brelse(bp); 2175 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 2176 } 2177 2178 /* 2179 * Initialize the filesystem. 2180 */ 2181 static int 2182 ffs_init(vfsp) 2183 struct vfsconf *vfsp; 2184 { 2185 2186 ffs_susp_initialize(); 2187 softdep_initialize(); 2188 return (ufs_init(vfsp)); 2189 } 2190 2191 /* 2192 * Undo the work of ffs_init(). 2193 */ 2194 static int 2195 ffs_uninit(vfsp) 2196 struct vfsconf *vfsp; 2197 { 2198 int ret; 2199 2200 ret = ufs_uninit(vfsp); 2201 softdep_uninitialize(); 2202 ffs_susp_uninitialize(); 2203 taskqueue_drain_all(taskqueue_thread); 2204 return (ret); 2205 } 2206 2207 /* 2208 * Structure used to pass information from ffs_sbupdate to its 2209 * helper routine ffs_use_bwrite. 2210 */ 2211 struct devfd { 2212 struct ufsmount *ump; 2213 struct buf *sbbp; 2214 int waitfor; 2215 int suspended; 2216 int error; 2217 }; 2218 2219 /* 2220 * Write a superblock and associated information back to disk. 2221 */ 2222 int 2223 ffs_sbupdate(ump, waitfor, suspended) 2224 struct ufsmount *ump; 2225 int waitfor; 2226 int suspended; 2227 { 2228 struct fs *fs; 2229 struct buf *sbbp; 2230 struct devfd devfd; 2231 2232 fs = ump->um_fs; 2233 if (fs->fs_ronly == 1 && 2234 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 2235 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) 2236 panic("ffs_sbupdate: write read-only filesystem"); 2237 /* 2238 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 2239 */ 2240 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 2241 (int)fs->fs_sbsize, 0, 0, 0); 2242 /* 2243 * Initialize info needed for write function. 2244 */ 2245 devfd.ump = ump; 2246 devfd.sbbp = sbbp; 2247 devfd.waitfor = waitfor; 2248 devfd.suspended = suspended; 2249 devfd.error = 0; 2250 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); 2251 } 2252 2253 /* 2254 * Write function for use by filesystem-layer routines. 2255 */ 2256 static int 2257 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) 2258 { 2259 struct devfd *devfdp; 2260 struct ufsmount *ump; 2261 struct buf *bp; 2262 struct fs *fs; 2263 int error; 2264 2265 devfdp = devfd; 2266 ump = devfdp->ump; 2267 fs = ump->um_fs; 2268 /* 2269 * Writing the superblock summary information. 2270 */ 2271 if (loc != fs->fs_sblockloc) { 2272 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); 2273 bcopy(buf, bp->b_data, (u_int)size); 2274 if (devfdp->suspended) 2275 bp->b_flags |= B_VALIDSUSPWRT; 2276 if (devfdp->waitfor != MNT_WAIT) 2277 bawrite(bp); 2278 else if ((error = bwrite(bp)) != 0) 2279 devfdp->error = error; 2280 return (0); 2281 } 2282 /* 2283 * Writing the superblock itself. We need to do special checks for it. 2284 */ 2285 bp = devfdp->sbbp; 2286 if (ffs_fsfail_cleanup(ump, devfdp->error)) 2287 devfdp->error = 0; 2288 if (devfdp->error != 0) { 2289 brelse(bp); 2290 return (devfdp->error); 2291 } 2292 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 2293 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2294 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2295 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 2296 fs->fs_sblockloc = SBLOCK_UFS1; 2297 } 2298 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 2299 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2300 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2301 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 2302 fs->fs_sblockloc = SBLOCK_UFS2; 2303 } 2304 if (MOUNTEDSOFTDEP(ump->um_mountp)) 2305 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 2306 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 2307 fs = (struct fs *)bp->b_data; 2308 ffs_oldfscompat_write(fs, ump); 2309 fs->fs_si = NULL; 2310 /* Recalculate the superblock hash */ 2311 fs->fs_ckhash = ffs_calc_sbhash(fs); 2312 if (devfdp->suspended) 2313 bp->b_flags |= B_VALIDSUSPWRT; 2314 if (devfdp->waitfor != MNT_WAIT) 2315 bawrite(bp); 2316 else if ((error = bwrite(bp)) != 0) 2317 devfdp->error = error; 2318 return (devfdp->error); 2319 } 2320 2321 static int 2322 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 2323 int attrnamespace, const char *attrname) 2324 { 2325 2326 #ifdef UFS_EXTATTR 2327 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 2328 attrname)); 2329 #else 2330 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 2331 attrname)); 2332 #endif 2333 } 2334 2335 static void 2336 ffs_ifree(struct ufsmount *ump, struct inode *ip) 2337 { 2338 2339 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 2340 uma_zfree(uma_ufs1, ip->i_din1); 2341 else if (ip->i_din2 != NULL) 2342 uma_zfree(uma_ufs2, ip->i_din2); 2343 uma_zfree_smr(uma_inode, ip); 2344 } 2345 2346 static int dobkgrdwrite = 1; 2347 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 2348 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 2349 2350 /* 2351 * Complete a background write started from bwrite. 2352 */ 2353 static void 2354 ffs_backgroundwritedone(struct buf *bp) 2355 { 2356 struct bufobj *bufobj; 2357 struct buf *origbp; 2358 2359 #ifdef SOFTUPDATES 2360 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0) 2361 softdep_handle_error(bp); 2362 #endif 2363 2364 /* 2365 * Find the original buffer that we are writing. 2366 */ 2367 bufobj = bp->b_bufobj; 2368 BO_LOCK(bufobj); 2369 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2370 panic("backgroundwritedone: lost buffer"); 2371 2372 /* 2373 * We should mark the cylinder group buffer origbp as 2374 * dirty, to not lose the failed write. 2375 */ 2376 if ((bp->b_ioflags & BIO_ERROR) != 0) 2377 origbp->b_vflags |= BV_BKGRDERR; 2378 BO_UNLOCK(bufobj); 2379 /* 2380 * Process dependencies then return any unfinished ones. 2381 */ 2382 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2383 buf_complete(bp); 2384 #ifdef SOFTUPDATES 2385 if (!LIST_EMPTY(&bp->b_dep)) 2386 softdep_move_dependencies(bp, origbp); 2387 #endif 2388 /* 2389 * This buffer is marked B_NOCACHE so when it is released 2390 * by biodone it will be tossed. 2391 */ 2392 bp->b_flags |= B_NOCACHE; 2393 bp->b_flags &= ~B_CACHE; 2394 pbrelvp(bp); 2395 2396 /* 2397 * Prevent brelse() from trying to keep and re-dirtying bp on 2398 * errors. It causes b_bufobj dereference in 2399 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2400 * pbrelvp() above. 2401 */ 2402 if ((bp->b_ioflags & BIO_ERROR) != 0) 2403 bp->b_flags |= B_INVAL; 2404 bufdone(bp); 2405 BO_LOCK(bufobj); 2406 /* 2407 * Clear the BV_BKGRDINPROG flag in the original buffer 2408 * and awaken it if it is waiting for the write to complete. 2409 * If BV_BKGRDINPROG is not set in the original buffer it must 2410 * have been released and re-instantiated - which is not legal. 2411 */ 2412 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2413 ("backgroundwritedone: lost buffer2")); 2414 origbp->b_vflags &= ~BV_BKGRDINPROG; 2415 if (origbp->b_vflags & BV_BKGRDWAIT) { 2416 origbp->b_vflags &= ~BV_BKGRDWAIT; 2417 wakeup(&origbp->b_xflags); 2418 } 2419 BO_UNLOCK(bufobj); 2420 } 2421 2422 /* 2423 * Write, release buffer on completion. (Done by iodone 2424 * if async). Do not bother writing anything if the buffer 2425 * is invalid. 2426 * 2427 * Note that we set B_CACHE here, indicating that buffer is 2428 * fully valid and thus cacheable. This is true even of NFS 2429 * now so we set it generally. This could be set either here 2430 * or in biodone() since the I/O is synchronous. We put it 2431 * here. 2432 */ 2433 static int 2434 ffs_bufwrite(struct buf *bp) 2435 { 2436 struct buf *newbp; 2437 struct cg *cgp; 2438 2439 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2440 if (bp->b_flags & B_INVAL) { 2441 brelse(bp); 2442 return (0); 2443 } 2444 2445 if (!BUF_ISLOCKED(bp)) 2446 panic("bufwrite: buffer is not busy???"); 2447 /* 2448 * If a background write is already in progress, delay 2449 * writing this block if it is asynchronous. Otherwise 2450 * wait for the background write to complete. 2451 */ 2452 BO_LOCK(bp->b_bufobj); 2453 if (bp->b_vflags & BV_BKGRDINPROG) { 2454 if (bp->b_flags & B_ASYNC) { 2455 BO_UNLOCK(bp->b_bufobj); 2456 bdwrite(bp); 2457 return (0); 2458 } 2459 bp->b_vflags |= BV_BKGRDWAIT; 2460 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2461 "bwrbg", 0); 2462 if (bp->b_vflags & BV_BKGRDINPROG) 2463 panic("bufwrite: still writing"); 2464 } 2465 bp->b_vflags &= ~BV_BKGRDERR; 2466 BO_UNLOCK(bp->b_bufobj); 2467 2468 /* 2469 * If this buffer is marked for background writing and we 2470 * do not have to wait for it, make a copy and write the 2471 * copy so as to leave this buffer ready for further use. 2472 * 2473 * This optimization eats a lot of memory. If we have a page 2474 * or buffer shortfall we can't do it. 2475 */ 2476 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2477 (bp->b_flags & B_ASYNC) && 2478 !vm_page_count_severe() && 2479 !buf_dirty_count_severe()) { 2480 KASSERT(bp->b_iodone == NULL, 2481 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2482 2483 /* get a new block */ 2484 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2485 if (newbp == NULL) 2486 goto normal_write; 2487 2488 KASSERT(buf_mapped(bp), ("Unmapped cg")); 2489 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2490 BO_LOCK(bp->b_bufobj); 2491 bp->b_vflags |= BV_BKGRDINPROG; 2492 BO_UNLOCK(bp->b_bufobj); 2493 newbp->b_xflags |= 2494 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; 2495 newbp->b_lblkno = bp->b_lblkno; 2496 newbp->b_blkno = bp->b_blkno; 2497 newbp->b_offset = bp->b_offset; 2498 newbp->b_iodone = ffs_backgroundwritedone; 2499 newbp->b_flags |= B_ASYNC; 2500 newbp->b_flags &= ~B_INVAL; 2501 pbgetvp(bp->b_vp, newbp); 2502 2503 #ifdef SOFTUPDATES 2504 /* 2505 * Move over the dependencies. If there are rollbacks, 2506 * leave the parent buffer dirtied as it will need to 2507 * be written again. 2508 */ 2509 if (LIST_EMPTY(&bp->b_dep) || 2510 softdep_move_dependencies(bp, newbp) == 0) 2511 bundirty(bp); 2512 #else 2513 bundirty(bp); 2514 #endif 2515 2516 /* 2517 * Initiate write on the copy, release the original. The 2518 * BKGRDINPROG flag prevents it from going away until 2519 * the background write completes. We have to recalculate 2520 * its check hash in case the buffer gets freed and then 2521 * reconstituted from the buffer cache during a later read. 2522 */ 2523 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2524 cgp = (struct cg *)bp->b_data; 2525 cgp->cg_ckhash = 0; 2526 cgp->cg_ckhash = 2527 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2528 } 2529 bqrelse(bp); 2530 bp = newbp; 2531 } else 2532 /* Mark the buffer clean */ 2533 bundirty(bp); 2534 2535 /* Let the normal bufwrite do the rest for us */ 2536 normal_write: 2537 /* 2538 * If we are writing a cylinder group, update its time. 2539 */ 2540 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2541 cgp = (struct cg *)bp->b_data; 2542 cgp->cg_old_time = cgp->cg_time = time_second; 2543 } 2544 return (bufwrite(bp)); 2545 } 2546 2547 static void 2548 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2549 { 2550 struct vnode *vp; 2551 struct buf *tbp; 2552 int error, nocopy; 2553 2554 /* 2555 * This is the bufobj strategy for the private VCHR vnodes 2556 * used by FFS to access the underlying storage device. 2557 * We override the default bufobj strategy and thus bypass 2558 * VOP_STRATEGY() for these vnodes. 2559 */ 2560 vp = bo2vnode(bo); 2561 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR || 2562 bp->b_vp->v_rdev == NULL || 2563 bp->b_vp->v_rdev->si_mountpt == NULL || 2564 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL || 2565 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp, 2566 ("ffs_geom_strategy() with wrong vp")); 2567 if (bp->b_iocmd == BIO_WRITE) { 2568 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2569 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2570 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2571 panic("ffs_geom_strategy: bad I/O"); 2572 nocopy = bp->b_flags & B_NOCOPY; 2573 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2574 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2575 vp->v_rdev->si_snapdata != NULL) { 2576 if ((bp->b_flags & B_CLUSTER) != 0) { 2577 runningbufwakeup(bp); 2578 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2579 b_cluster.cluster_entry) { 2580 error = ffs_copyonwrite(vp, tbp); 2581 if (error != 0 && 2582 error != EOPNOTSUPP) { 2583 bp->b_error = error; 2584 bp->b_ioflags |= BIO_ERROR; 2585 bufdone(bp); 2586 return; 2587 } 2588 } 2589 bp->b_runningbufspace = bp->b_bufsize; 2590 atomic_add_long(&runningbufspace, 2591 bp->b_runningbufspace); 2592 } else { 2593 error = ffs_copyonwrite(vp, bp); 2594 if (error != 0 && error != EOPNOTSUPP) { 2595 bp->b_error = error; 2596 bp->b_ioflags |= BIO_ERROR; 2597 bufdone(bp); 2598 return; 2599 } 2600 } 2601 } 2602 #ifdef SOFTUPDATES 2603 if ((bp->b_flags & B_CLUSTER) != 0) { 2604 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2605 b_cluster.cluster_entry) { 2606 if (!LIST_EMPTY(&tbp->b_dep)) 2607 buf_start(tbp); 2608 } 2609 } else { 2610 if (!LIST_EMPTY(&bp->b_dep)) 2611 buf_start(bp); 2612 } 2613 2614 #endif 2615 /* 2616 * Check for metadata that needs check-hashes and update them. 2617 */ 2618 switch (bp->b_xflags & BX_FSPRIV) { 2619 case BX_CYLGRP: 2620 ((struct cg *)bp->b_data)->cg_ckhash = 0; 2621 ((struct cg *)bp->b_data)->cg_ckhash = 2622 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2623 break; 2624 2625 case BX_SUPERBLOCK: 2626 case BX_INODE: 2627 case BX_INDIR: 2628 case BX_DIR: 2629 printf("Check-hash write is unimplemented!!!\n"); 2630 break; 2631 2632 case 0: 2633 break; 2634 2635 default: 2636 printf("multiple buffer types 0x%b\n", 2637 (u_int)(bp->b_xflags & BX_FSPRIV), 2638 PRINT_UFS_BUF_XFLAGS); 2639 break; 2640 } 2641 } 2642 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable) 2643 bp->b_xflags |= BX_CVTENXIO; 2644 g_vfs_strategy(bo, bp); 2645 } 2646 2647 int 2648 ffs_own_mount(const struct mount *mp) 2649 { 2650 2651 if (mp->mnt_op == &ufs_vfsops) 2652 return (1); 2653 return (0); 2654 } 2655 2656 #ifdef DDB 2657 #ifdef SOFTUPDATES 2658 2659 /* defined in ffs_softdep.c */ 2660 extern void db_print_ffs(struct ufsmount *ump); 2661 2662 DB_SHOW_COMMAND(ffs, db_show_ffs) 2663 { 2664 struct mount *mp; 2665 struct ufsmount *ump; 2666 2667 if (have_addr) { 2668 ump = VFSTOUFS((struct mount *)addr); 2669 db_print_ffs(ump); 2670 return; 2671 } 2672 2673 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2674 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2675 db_print_ffs(VFSTOUFS(mp)); 2676 } 2677 } 2678 2679 #endif /* SOFTUPDATES */ 2680 #endif /* DDB */ 2681