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 * 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 pid_t fsckpid = 0; 351 int error, error1, flags; 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, "fsckpid", NULL, NULL) == 0 && 399 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { 400 /* 401 * Once we have set the restricted PID, do not 402 * persist "fsckpid" in the options list. 403 */ 404 vfs_deleteopt(mp->mnt_optnew, "fsckpid"); 405 vfs_deleteopt(mp->mnt_opt, "fsckpid"); 406 if (mp->mnt_flag & MNT_UPDATE) { 407 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && 408 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 409 vfs_mount_error(mp, 410 "Checker enable: Must be read-only"); 411 return (EINVAL); 412 } 413 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 414 vfs_mount_error(mp, 415 "Checker enable: Must be read-only"); 416 return (EINVAL); 417 } 418 /* Set to -1 if we are done */ 419 if (fsckpid == 0) 420 fsckpid = -1; 421 } 422 423 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 424 if (mntorflags & MNT_ACLS) { 425 vfs_mount_error(mp, 426 "\"acls\" and \"nfsv4acls\" options " 427 "are mutually exclusive"); 428 return (EINVAL); 429 } 430 mntorflags |= MNT_NFS4ACLS; 431 } 432 433 MNT_ILOCK(mp); 434 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP; 435 mp->mnt_flag |= mntorflags; 436 MNT_IUNLOCK(mp); 437 /* 438 * If updating, check whether changing from read-only to 439 * read/write; if there is no device name, that's all we do. 440 */ 441 if (mp->mnt_flag & MNT_UPDATE) { 442 ump = VFSTOUFS(mp); 443 fs = ump->um_fs; 444 odevvp = ump->um_odevvp; 445 devvp = ump->um_devvp; 446 if (fsckpid == -1 && ump->um_fsckpid > 0) { 447 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || 448 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) 449 return (error); 450 g_topology_lock(); 451 /* 452 * Return to normal read-only mode. 453 */ 454 error = g_access(ump->um_cp, 0, -1, 0); 455 g_topology_unlock(); 456 ump->um_fsckpid = 0; 457 } 458 if (fs->fs_ronly == 0 && 459 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 460 /* 461 * Flush any dirty data and suspend filesystem. 462 */ 463 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 464 return (error); 465 error = vfs_write_suspend_umnt(mp); 466 if (error != 0) 467 return (error); 468 469 fs->fs_ronly = 1; 470 if (MOUNTEDSOFTDEP(mp)) { 471 MNT_ILOCK(mp); 472 mp->mnt_flag &= ~MNT_SOFTDEP; 473 MNT_IUNLOCK(mp); 474 mounted_softdep = true; 475 } else 476 mounted_softdep = false; 477 478 /* 479 * Check for and optionally get rid of files open 480 * for writing. 481 */ 482 flags = WRITECLOSE; 483 if (mp->mnt_flag & MNT_FORCE) 484 flags |= FORCECLOSE; 485 if (mounted_softdep) { 486 error = softdep_flushfiles(mp, flags, td); 487 } else { 488 error = ffs_flushfiles(mp, flags, td); 489 } 490 if (error) { 491 fs->fs_ronly = 0; 492 if (mounted_softdep) { 493 MNT_ILOCK(mp); 494 mp->mnt_flag |= MNT_SOFTDEP; 495 MNT_IUNLOCK(mp); 496 } 497 vfs_write_resume(mp, 0); 498 return (error); 499 } 500 501 if (fs->fs_pendingblocks != 0 || 502 fs->fs_pendinginodes != 0) { 503 printf("WARNING: %s Update error: blocks %jd " 504 "files %d\n", fs->fs_fsmnt, 505 (intmax_t)fs->fs_pendingblocks, 506 fs->fs_pendinginodes); 507 fs->fs_pendingblocks = 0; 508 fs->fs_pendinginodes = 0; 509 } 510 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 511 fs->fs_clean = 1; 512 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 513 fs->fs_ronly = 0; 514 fs->fs_clean = 0; 515 if (mounted_softdep) { 516 MNT_ILOCK(mp); 517 mp->mnt_flag |= MNT_SOFTDEP; 518 MNT_IUNLOCK(mp); 519 } 520 vfs_write_resume(mp, 0); 521 return (error); 522 } 523 if (mounted_softdep) 524 softdep_unmount(mp); 525 g_topology_lock(); 526 /* 527 * Drop our write and exclusive access. 528 */ 529 g_access(ump->um_cp, 0, -1, -1); 530 g_topology_unlock(); 531 MNT_ILOCK(mp); 532 mp->mnt_flag |= MNT_RDONLY; 533 MNT_IUNLOCK(mp); 534 /* 535 * Allow the writers to note that filesystem 536 * is ro now. 537 */ 538 vfs_write_resume(mp, 0); 539 } 540 if ((mp->mnt_flag & MNT_RELOAD) && 541 (error = ffs_reload(mp, td, 0)) != 0) 542 return (error); 543 if (fs->fs_ronly && 544 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 545 /* 546 * If we are running a checker, do not allow upgrade. 547 */ 548 if (ump->um_fsckpid > 0) { 549 vfs_mount_error(mp, 550 "Active checker, cannot upgrade to write"); 551 return (EINVAL); 552 } 553 /* 554 * If upgrade to read-write by non-root, then verify 555 * that user has necessary permissions on the device. 556 */ 557 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY); 558 error = VOP_ACCESS(odevvp, VREAD | VWRITE, 559 td->td_ucred, td); 560 if (error) 561 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 562 VOP_UNLOCK(odevvp); 563 if (error) { 564 return (error); 565 } 566 fs->fs_flags &= ~FS_UNCLEAN; 567 if (fs->fs_clean == 0) { 568 fs->fs_flags |= FS_UNCLEAN; 569 if ((mp->mnt_flag & MNT_FORCE) || 570 ((fs->fs_flags & 571 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 572 (fs->fs_flags & FS_DOSOFTDEP))) { 573 printf("WARNING: %s was not properly " 574 "dismounted\n", fs->fs_fsmnt); 575 } else { 576 vfs_mount_error(mp, 577 "R/W mount of %s denied. %s.%s", 578 fs->fs_fsmnt, 579 "Filesystem is not clean - run fsck", 580 (fs->fs_flags & FS_SUJ) == 0 ? "" : 581 " Forced mount will invalidate" 582 " journal contents"); 583 return (EPERM); 584 } 585 } 586 g_topology_lock(); 587 /* 588 * Request exclusive write access. 589 */ 590 error = g_access(ump->um_cp, 0, 1, 1); 591 g_topology_unlock(); 592 if (error) 593 return (error); 594 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 595 return (error); 596 error = vfs_write_suspend_umnt(mp); 597 if (error != 0) 598 return (error); 599 fs->fs_ronly = 0; 600 MNT_ILOCK(mp); 601 saved_mnt_flag = MNT_RDONLY; 602 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag & 603 MNT_ASYNC) != 0) 604 saved_mnt_flag |= MNT_ASYNC; 605 mp->mnt_flag &= ~saved_mnt_flag; 606 MNT_IUNLOCK(mp); 607 fs->fs_mtime = time_second; 608 /* check to see if we need to start softdep */ 609 if ((fs->fs_flags & FS_DOSOFTDEP) && 610 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 611 fs->fs_ronly = 1; 612 MNT_ILOCK(mp); 613 mp->mnt_flag |= saved_mnt_flag; 614 MNT_IUNLOCK(mp); 615 vfs_write_resume(mp, 0); 616 return (error); 617 } 618 fs->fs_clean = 0; 619 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 620 fs->fs_ronly = 1; 621 if ((fs->fs_flags & FS_DOSOFTDEP) != 0) 622 softdep_unmount(mp); 623 MNT_ILOCK(mp); 624 mp->mnt_flag |= saved_mnt_flag; 625 MNT_IUNLOCK(mp); 626 vfs_write_resume(mp, 0); 627 return (error); 628 } 629 if (fs->fs_snapinum[0] != 0) 630 ffs_snapshot_mount(mp); 631 vfs_write_resume(mp, 0); 632 } 633 /* 634 * Soft updates is incompatible with "async", 635 * so if we are doing softupdates stop the user 636 * from setting the async flag in an update. 637 * Softdep_mount() clears it in an initial mount 638 * or ro->rw remount. 639 */ 640 if (MOUNTEDSOFTDEP(mp)) { 641 /* XXX: Reset too late ? */ 642 MNT_ILOCK(mp); 643 mp->mnt_flag &= ~MNT_ASYNC; 644 MNT_IUNLOCK(mp); 645 } 646 /* 647 * Keep MNT_ACLS flag if it is stored in superblock. 648 */ 649 if ((fs->fs_flags & FS_ACLS) != 0) { 650 /* XXX: Set too late ? */ 651 MNT_ILOCK(mp); 652 mp->mnt_flag |= MNT_ACLS; 653 MNT_IUNLOCK(mp); 654 } 655 656 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 657 /* XXX: Set too late ? */ 658 MNT_ILOCK(mp); 659 mp->mnt_flag |= MNT_NFS4ACLS; 660 MNT_IUNLOCK(mp); 661 } 662 /* 663 * If this is a request from fsck to clean up the filesystem, 664 * then allow the specified pid to proceed. 665 */ 666 if (fsckpid > 0) { 667 if (ump->um_fsckpid != 0) { 668 vfs_mount_error(mp, 669 "Active checker already running on %s", 670 fs->fs_fsmnt); 671 return (EINVAL); 672 } 673 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 674 ("soft updates enabled on read-only file system")); 675 g_topology_lock(); 676 /* 677 * Request write access. 678 */ 679 error = g_access(ump->um_cp, 0, 1, 0); 680 g_topology_unlock(); 681 if (error) { 682 vfs_mount_error(mp, 683 "Checker activation failed on %s", 684 fs->fs_fsmnt); 685 return (error); 686 } 687 ump->um_fsckpid = fsckpid; 688 if (fs->fs_snapinum[0] != 0) 689 ffs_snapshot_mount(mp); 690 fs->fs_mtime = time_second; 691 fs->fs_fmod = 1; 692 fs->fs_clean = 0; 693 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 694 } 695 696 /* 697 * If this is a snapshot request, take the snapshot. 698 */ 699 if (mp->mnt_flag & MNT_SNAPSHOT) 700 return (ffs_snapshot(mp, fspec)); 701 702 /* 703 * Must not call namei() while owning busy ref. 704 */ 705 vfs_unbusy(mp); 706 } 707 708 /* 709 * Not an update, or updating the name: look up the name 710 * and verify that it refers to a sensible disk device. 711 */ 712 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 713 error = namei(&ndp); 714 if ((mp->mnt_flag & MNT_UPDATE) != 0) { 715 /* 716 * Unmount does not start if MNT_UPDATE is set. Mount 717 * update busies mp before setting MNT_UPDATE. We 718 * must be able to retain our busy ref succesfully, 719 * without sleep. 720 */ 721 error1 = vfs_busy(mp, MBF_NOWAIT); 722 MPASS(error1 == 0); 723 } 724 if (error != 0) 725 return (error); 726 NDFREE(&ndp, NDF_ONLY_PNBUF); 727 devvp = ndp.ni_vp; 728 if (!vn_isdisk_error(devvp, &error)) { 729 vput(devvp); 730 return (error); 731 } 732 733 /* 734 * If mount by non-root, then verify that user has necessary 735 * permissions on the device. 736 */ 737 accmode = VREAD; 738 if ((mp->mnt_flag & MNT_RDONLY) == 0) 739 accmode |= VWRITE; 740 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 741 if (error) 742 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 743 if (error) { 744 vput(devvp); 745 return (error); 746 } 747 748 if (mp->mnt_flag & MNT_UPDATE) { 749 /* 750 * Update only 751 * 752 * If it's not the same vnode, or at least the same device 753 * then it's not correct. 754 */ 755 756 if (devvp->v_rdev != ump->um_devvp->v_rdev) 757 error = EINVAL; /* needs translation */ 758 vput(devvp); 759 if (error) 760 return (error); 761 } else { 762 /* 763 * New mount 764 * 765 * We need the name for the mount point (also used for 766 * "last mounted on") copied in. If an error occurs, 767 * the mount point is discarded by the upper level code. 768 * Note that vfs_mount_alloc() populates f_mntonname for us. 769 */ 770 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 771 vrele(devvp); 772 return (error); 773 } 774 if (fsckpid > 0) { 775 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 776 ("soft updates enabled on read-only file system")); 777 ump = VFSTOUFS(mp); 778 fs = ump->um_fs; 779 g_topology_lock(); 780 /* 781 * Request write access. 782 */ 783 error = g_access(ump->um_cp, 0, 1, 0); 784 g_topology_unlock(); 785 if (error) { 786 printf("WARNING: %s: Checker activation " 787 "failed\n", fs->fs_fsmnt); 788 } else { 789 ump->um_fsckpid = fsckpid; 790 if (fs->fs_snapinum[0] != 0) 791 ffs_snapshot_mount(mp); 792 fs->fs_mtime = time_second; 793 fs->fs_clean = 0; 794 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 795 } 796 } 797 } 798 799 MNT_ILOCK(mp); 800 /* 801 * This is racy versus lookup, see ufs_fplookup_vexec for details. 802 */ 803 if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0) 804 panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp); 805 if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0) 806 mp->mnt_kern_flag |= MNTK_FPLOOKUP; 807 MNT_IUNLOCK(mp); 808 809 vfs_mountedfrom(mp, fspec); 810 return (0); 811 } 812 813 /* 814 * Compatibility with old mount system call. 815 */ 816 817 static int 818 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) 819 { 820 struct ufs_args args; 821 int error; 822 823 if (data == NULL) 824 return (EINVAL); 825 error = copyin(data, &args, sizeof args); 826 if (error) 827 return (error); 828 829 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 830 ma = mount_arg(ma, "export", &args.export, sizeof(args.export)); 831 error = kernel_mount(ma, flags); 832 833 return (error); 834 } 835 836 /* 837 * Reload all incore data for a filesystem (used after running fsck on 838 * the root filesystem and finding things to fix). If the 'force' flag 839 * is 0, the filesystem must be mounted read-only. 840 * 841 * Things to do to update the mount: 842 * 1) invalidate all cached meta-data. 843 * 2) re-read superblock from disk. 844 * 3) re-read summary information from disk. 845 * 4) invalidate all inactive vnodes. 846 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary 847 * writers, if requested. 848 * 6) invalidate all cached file data. 849 * 7) re-read inode data for all active vnodes. 850 */ 851 int 852 ffs_reload(struct mount *mp, struct thread *td, int flags) 853 { 854 struct vnode *vp, *mvp, *devvp; 855 struct inode *ip; 856 void *space; 857 struct buf *bp; 858 struct fs *fs, *newfs; 859 struct ufsmount *ump; 860 ufs2_daddr_t sblockloc; 861 int i, blks, error; 862 u_long size; 863 int32_t *lp; 864 865 ump = VFSTOUFS(mp); 866 867 MNT_ILOCK(mp); 868 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { 869 MNT_IUNLOCK(mp); 870 return (EINVAL); 871 } 872 MNT_IUNLOCK(mp); 873 874 /* 875 * Step 1: invalidate all cached meta-data. 876 */ 877 devvp = VFSTOUFS(mp)->um_devvp; 878 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 879 if (vinvalbuf(devvp, 0, 0, 0) != 0) 880 panic("ffs_reload: dirty1"); 881 VOP_UNLOCK(devvp); 882 883 /* 884 * Step 2: re-read superblock from disk. 885 */ 886 fs = VFSTOUFS(mp)->um_fs; 887 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 888 NOCRED, &bp)) != 0) 889 return (error); 890 newfs = (struct fs *)bp->b_data; 891 if ((newfs->fs_magic != FS_UFS1_MAGIC && 892 newfs->fs_magic != FS_UFS2_MAGIC) || 893 newfs->fs_bsize > MAXBSIZE || 894 newfs->fs_bsize < sizeof(struct fs)) { 895 brelse(bp); 896 return (EIO); /* XXX needs translation */ 897 } 898 /* 899 * Preserve the summary information, read-only status, and 900 * superblock location by copying these fields into our new 901 * superblock before using it to update the existing superblock. 902 */ 903 newfs->fs_si = fs->fs_si; 904 newfs->fs_ronly = fs->fs_ronly; 905 sblockloc = fs->fs_sblockloc; 906 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 907 brelse(bp); 908 ump->um_maxsymlinklen = fs->fs_maxsymlinklen; 909 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 910 UFS_LOCK(ump); 911 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 912 printf("WARNING: %s: reload pending error: blocks %jd " 913 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 914 fs->fs_pendinginodes); 915 fs->fs_pendingblocks = 0; 916 fs->fs_pendinginodes = 0; 917 } 918 UFS_UNLOCK(ump); 919 920 /* 921 * Step 3: re-read summary information from disk. 922 */ 923 size = fs->fs_cssize; 924 blks = howmany(size, fs->fs_fsize); 925 if (fs->fs_contigsumsize > 0) 926 size += fs->fs_ncg * sizeof(int32_t); 927 size += fs->fs_ncg * sizeof(u_int8_t); 928 free(fs->fs_csp, M_UFSMNT); 929 space = malloc(size, M_UFSMNT, M_WAITOK); 930 fs->fs_csp = space; 931 for (i = 0; i < blks; i += fs->fs_frag) { 932 size = fs->fs_bsize; 933 if (i + fs->fs_frag > blks) 934 size = (blks - i) * fs->fs_fsize; 935 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 936 NOCRED, &bp); 937 if (error) 938 return (error); 939 bcopy(bp->b_data, space, (u_int)size); 940 space = (char *)space + size; 941 brelse(bp); 942 } 943 /* 944 * We no longer know anything about clusters per cylinder group. 945 */ 946 if (fs->fs_contigsumsize > 0) { 947 fs->fs_maxcluster = lp = space; 948 for (i = 0; i < fs->fs_ncg; i++) 949 *lp++ = fs->fs_contigsumsize; 950 space = lp; 951 } 952 size = fs->fs_ncg * sizeof(u_int8_t); 953 fs->fs_contigdirs = (u_int8_t *)space; 954 bzero(fs->fs_contigdirs, size); 955 if ((flags & FFSR_UNSUSPEND) != 0) { 956 MNT_ILOCK(mp); 957 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 958 wakeup(&mp->mnt_flag); 959 MNT_IUNLOCK(mp); 960 } 961 962 loop: 963 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 964 /* 965 * Skip syncer vnode. 966 */ 967 if (vp->v_type == VNON) { 968 VI_UNLOCK(vp); 969 continue; 970 } 971 /* 972 * Step 4: invalidate all cached file data. 973 */ 974 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 975 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 976 goto loop; 977 } 978 if (vinvalbuf(vp, 0, 0, 0)) 979 panic("ffs_reload: dirty2"); 980 /* 981 * Step 5: re-read inode data for all active vnodes. 982 */ 983 ip = VTOI(vp); 984 error = 985 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 986 (int)fs->fs_bsize, NOCRED, &bp); 987 if (error) { 988 vput(vp); 989 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 990 return (error); 991 } 992 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) { 993 brelse(bp); 994 vput(vp); 995 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 996 return (error); 997 } 998 ip->i_effnlink = ip->i_nlink; 999 brelse(bp); 1000 vput(vp); 1001 } 1002 return (0); 1003 } 1004 1005 /* 1006 * Common code for mount and mountroot 1007 */ 1008 static int 1009 ffs_mountfs(odevvp, mp, td) 1010 struct vnode *odevvp; 1011 struct mount *mp; 1012 struct thread *td; 1013 { 1014 struct ufsmount *ump; 1015 struct fs *fs; 1016 struct cdev *dev; 1017 int error, i, len, ronly; 1018 struct ucred *cred; 1019 struct g_consumer *cp; 1020 struct mount *nmp; 1021 struct vnode *devvp; 1022 int candelete, canspeedup; 1023 off_t loc; 1024 1025 fs = NULL; 1026 ump = NULL; 1027 cred = td ? td->td_ucred : NOCRED; 1028 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 1029 1030 devvp = mntfs_allocvp(mp, odevvp); 1031 VOP_UNLOCK(odevvp); 1032 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); 1033 dev = devvp->v_rdev; 1034 KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data")); 1035 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, 1036 (uintptr_t)mp) == 0) { 1037 mntfs_freevp(devvp); 1038 return (EBUSY); 1039 } 1040 g_topology_lock(); 1041 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 1042 g_topology_unlock(); 1043 if (error != 0) { 1044 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1045 mntfs_freevp(devvp); 1046 return (error); 1047 } 1048 dev_ref(dev); 1049 devvp->v_bufobj.bo_ops = &ffs_ops; 1050 BO_LOCK(&odevvp->v_bufobj); 1051 odevvp->v_bufobj.bo_flag |= BO_NOBUFS; 1052 BO_UNLOCK(&odevvp->v_bufobj); 1053 if (dev->si_iosize_max != 0) 1054 mp->mnt_iosize_max = dev->si_iosize_max; 1055 if (mp->mnt_iosize_max > maxphys) 1056 mp->mnt_iosize_max = maxphys; 1057 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 1058 error = EINVAL; 1059 vfs_mount_error(mp, 1060 "Invalid sectorsize %d for superblock size %d", 1061 cp->provider->sectorsize, SBLOCKSIZE); 1062 goto out; 1063 } 1064 /* fetch the superblock and summary information */ 1065 loc = STDSB; 1066 if ((mp->mnt_flag & MNT_ROOTFS) != 0) 1067 loc = STDSB_NOHASHFAIL; 1068 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0) 1069 goto out; 1070 fs->fs_flags &= ~FS_UNCLEAN; 1071 if (fs->fs_clean == 0) { 1072 fs->fs_flags |= FS_UNCLEAN; 1073 if (ronly || (mp->mnt_flag & MNT_FORCE) || 1074 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 1075 (fs->fs_flags & FS_DOSOFTDEP))) { 1076 printf("WARNING: %s was not properly dismounted\n", 1077 fs->fs_fsmnt); 1078 } else { 1079 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 1080 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 1081 (fs->fs_flags & FS_SUJ) == 0 ? "" : 1082 " Forced mount will invalidate journal contents"); 1083 error = EPERM; 1084 goto out; 1085 } 1086 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 1087 (mp->mnt_flag & MNT_FORCE)) { 1088 printf("WARNING: %s: lost blocks %jd files %d\n", 1089 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1090 fs->fs_pendinginodes); 1091 fs->fs_pendingblocks = 0; 1092 fs->fs_pendinginodes = 0; 1093 } 1094 } 1095 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1096 printf("WARNING: %s: mount pending error: blocks %jd " 1097 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1098 fs->fs_pendinginodes); 1099 fs->fs_pendingblocks = 0; 1100 fs->fs_pendinginodes = 0; 1101 } 1102 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 1103 #ifdef UFS_GJOURNAL 1104 /* 1105 * Get journal provider name. 1106 */ 1107 len = 1024; 1108 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); 1109 if (g_io_getattr("GJOURNAL::provider", cp, &len, 1110 mp->mnt_gjprovider) == 0) { 1111 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, 1112 M_UFSMNT, M_WAITOK); 1113 MNT_ILOCK(mp); 1114 mp->mnt_flag |= MNT_GJOURNAL; 1115 MNT_IUNLOCK(mp); 1116 } else { 1117 printf("WARNING: %s: GJOURNAL flag on fs " 1118 "but no gjournal provider below\n", 1119 mp->mnt_stat.f_mntonname); 1120 free(mp->mnt_gjprovider, M_UFSMNT); 1121 mp->mnt_gjprovider = NULL; 1122 } 1123 #else 1124 printf("WARNING: %s: GJOURNAL flag on fs but no " 1125 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 1126 #endif 1127 } else { 1128 mp->mnt_gjprovider = NULL; 1129 } 1130 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 1131 ump->um_cp = cp; 1132 ump->um_bo = &devvp->v_bufobj; 1133 ump->um_fs = fs; 1134 if (fs->fs_magic == FS_UFS1_MAGIC) { 1135 ump->um_fstype = UFS1; 1136 ump->um_balloc = ffs_balloc_ufs1; 1137 } else { 1138 ump->um_fstype = UFS2; 1139 ump->um_balloc = ffs_balloc_ufs2; 1140 } 1141 ump->um_blkatoff = ffs_blkatoff; 1142 ump->um_truncate = ffs_truncate; 1143 ump->um_update = ffs_update; 1144 ump->um_valloc = ffs_valloc; 1145 ump->um_vfree = ffs_vfree; 1146 ump->um_ifree = ffs_ifree; 1147 ump->um_rdonly = ffs_rdonly; 1148 ump->um_snapgone = ffs_snapgone; 1149 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) 1150 ump->um_check_blkno = ffs_check_blkno; 1151 else 1152 ump->um_check_blkno = NULL; 1153 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 1154 sx_init(&ump->um_checkpath_lock, "uchpth"); 1155 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); 1156 fs->fs_ronly = ronly; 1157 fs->fs_active = NULL; 1158 mp->mnt_data = ump; 1159 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 1160 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 1161 nmp = NULL; 1162 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 1163 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 1164 if (nmp) 1165 vfs_rel(nmp); 1166 vfs_getnewfsid(mp); 1167 } 1168 ump->um_maxsymlinklen = fs->fs_maxsymlinklen; 1169 MNT_ILOCK(mp); 1170 mp->mnt_flag |= MNT_LOCAL; 1171 MNT_IUNLOCK(mp); 1172 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 1173 #ifdef MAC 1174 MNT_ILOCK(mp); 1175 mp->mnt_flag |= MNT_MULTILABEL; 1176 MNT_IUNLOCK(mp); 1177 #else 1178 printf("WARNING: %s: multilabel flag on fs but " 1179 "no MAC support\n", mp->mnt_stat.f_mntonname); 1180 #endif 1181 } 1182 if ((fs->fs_flags & FS_ACLS) != 0) { 1183 #ifdef UFS_ACL 1184 MNT_ILOCK(mp); 1185 1186 if (mp->mnt_flag & MNT_NFS4ACLS) 1187 printf("WARNING: %s: ACLs flag on fs conflicts with " 1188 "\"nfsv4acls\" mount option; option ignored\n", 1189 mp->mnt_stat.f_mntonname); 1190 mp->mnt_flag &= ~MNT_NFS4ACLS; 1191 mp->mnt_flag |= MNT_ACLS; 1192 1193 MNT_IUNLOCK(mp); 1194 #else 1195 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 1196 mp->mnt_stat.f_mntonname); 1197 #endif 1198 } 1199 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 1200 #ifdef UFS_ACL 1201 MNT_ILOCK(mp); 1202 1203 if (mp->mnt_flag & MNT_ACLS) 1204 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 1205 "with \"acls\" mount option; option ignored\n", 1206 mp->mnt_stat.f_mntonname); 1207 mp->mnt_flag &= ~MNT_ACLS; 1208 mp->mnt_flag |= MNT_NFS4ACLS; 1209 1210 MNT_IUNLOCK(mp); 1211 #else 1212 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 1213 "ACLs support\n", mp->mnt_stat.f_mntonname); 1214 #endif 1215 } 1216 if ((fs->fs_flags & FS_TRIM) != 0) { 1217 len = sizeof(int); 1218 if (g_io_getattr("GEOM::candelete", cp, &len, 1219 &candelete) == 0) { 1220 if (candelete) 1221 ump->um_flags |= UM_CANDELETE; 1222 else 1223 printf("WARNING: %s: TRIM flag on fs but disk " 1224 "does not support TRIM\n", 1225 mp->mnt_stat.f_mntonname); 1226 } else { 1227 printf("WARNING: %s: TRIM flag on fs but disk does " 1228 "not confirm that it supports TRIM\n", 1229 mp->mnt_stat.f_mntonname); 1230 } 1231 if (((ump->um_flags) & UM_CANDELETE) != 0) { 1232 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1233 taskqueue_thread_enqueue, &ump->um_trim_tq); 1234 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1235 "%s trim", mp->mnt_stat.f_mntonname); 1236 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, 1237 &ump->um_trimlisthashsize); 1238 } 1239 } 1240 1241 len = sizeof(int); 1242 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) { 1243 if (canspeedup) 1244 ump->um_flags |= UM_CANSPEEDUP; 1245 } 1246 1247 ump->um_mountp = mp; 1248 ump->um_dev = dev; 1249 ump->um_devvp = devvp; 1250 ump->um_odevvp = odevvp; 1251 ump->um_nindir = fs->fs_nindir; 1252 ump->um_bptrtodb = fs->fs_fsbtodb; 1253 ump->um_seqinc = fs->fs_frag; 1254 for (i = 0; i < MAXQUOTAS; i++) 1255 ump->um_quotas[i] = NULLVP; 1256 #ifdef UFS_EXTATTR 1257 ufs_extattr_uepm_init(&ump->um_extattr); 1258 #endif 1259 /* 1260 * Set FS local "last mounted on" information (NULL pad) 1261 */ 1262 bzero(fs->fs_fsmnt, MAXMNTLEN); 1263 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1264 mp->mnt_stat.f_iosize = fs->fs_bsize; 1265 1266 if (mp->mnt_flag & MNT_ROOTFS) { 1267 /* 1268 * Root mount; update timestamp in mount structure. 1269 * this will be used by the common root mount code 1270 * to update the system clock. 1271 */ 1272 mp->mnt_time = fs->fs_time; 1273 } 1274 1275 if (ronly == 0) { 1276 fs->fs_mtime = time_second; 1277 if ((fs->fs_flags & FS_DOSOFTDEP) && 1278 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1279 ffs_flushfiles(mp, FORCECLOSE, td); 1280 goto out; 1281 } 1282 if (fs->fs_snapinum[0] != 0) 1283 ffs_snapshot_mount(mp); 1284 fs->fs_fmod = 1; 1285 fs->fs_clean = 0; 1286 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1287 } 1288 /* 1289 * Initialize filesystem state information in mount struct. 1290 */ 1291 MNT_ILOCK(mp); 1292 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1293 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1294 MNT_IUNLOCK(mp); 1295 #ifdef UFS_EXTATTR 1296 #ifdef UFS_EXTATTR_AUTOSTART 1297 /* 1298 * 1299 * Auto-starting does the following: 1300 * - check for /.attribute in the fs, and extattr_start if so 1301 * - for each file in .attribute, enable that file with 1302 * an attribute of the same name. 1303 * Not clear how to report errors -- probably eat them. 1304 * This would all happen while the filesystem was busy/not 1305 * available, so would effectively be "atomic". 1306 */ 1307 (void) ufs_extattr_autostart(mp, td); 1308 #endif /* !UFS_EXTATTR_AUTOSTART */ 1309 #endif /* !UFS_EXTATTR */ 1310 return (0); 1311 out: 1312 if (fs != NULL) { 1313 free(fs->fs_csp, M_UFSMNT); 1314 free(fs->fs_si, M_UFSMNT); 1315 free(fs, M_UFSMNT); 1316 } 1317 if (cp != NULL) { 1318 g_topology_lock(); 1319 g_vfs_close(cp); 1320 g_topology_unlock(); 1321 } 1322 if (ump != NULL) { 1323 mtx_destroy(UFS_MTX(ump)); 1324 sx_destroy(&ump->um_checkpath_lock); 1325 if (mp->mnt_gjprovider != NULL) { 1326 free(mp->mnt_gjprovider, M_UFSMNT); 1327 mp->mnt_gjprovider = NULL; 1328 } 1329 MPASS(ump->um_softdep == NULL); 1330 free(ump, M_UFSMNT); 1331 mp->mnt_data = NULL; 1332 } 1333 BO_LOCK(&odevvp->v_bufobj); 1334 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1335 BO_UNLOCK(&odevvp->v_bufobj); 1336 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1337 mntfs_freevp(devvp); 1338 dev_rel(dev); 1339 return (error); 1340 } 1341 1342 /* 1343 * A read function for use by filesystem-layer routines. 1344 */ 1345 static int 1346 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) 1347 { 1348 struct buf *bp; 1349 int error; 1350 1351 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); 1352 *bufp = malloc(size, M_UFSMNT, M_WAITOK); 1353 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, 1354 &bp)) != 0) 1355 return (error); 1356 bcopy(bp->b_data, *bufp, size); 1357 bp->b_flags |= B_INVAL | B_NOCACHE; 1358 brelse(bp); 1359 return (0); 1360 } 1361 1362 static int bigcgs = 0; 1363 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1364 1365 /* 1366 * Sanity checks for loading old filesystem superblocks. 1367 * See ffs_oldfscompat_write below for unwound actions. 1368 * 1369 * XXX - Parts get retired eventually. 1370 * Unfortunately new bits get added. 1371 */ 1372 static void 1373 ffs_oldfscompat_read(fs, ump, sblockloc) 1374 struct fs *fs; 1375 struct ufsmount *ump; 1376 ufs2_daddr_t sblockloc; 1377 { 1378 off_t maxfilesize; 1379 1380 /* 1381 * If not yet done, update fs_flags location and value of fs_sblockloc. 1382 */ 1383 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1384 fs->fs_flags = fs->fs_old_flags; 1385 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1386 fs->fs_sblockloc = sblockloc; 1387 } 1388 /* 1389 * If not yet done, update UFS1 superblock with new wider fields. 1390 */ 1391 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1392 fs->fs_maxbsize = fs->fs_bsize; 1393 fs->fs_time = fs->fs_old_time; 1394 fs->fs_size = fs->fs_old_size; 1395 fs->fs_dsize = fs->fs_old_dsize; 1396 fs->fs_csaddr = fs->fs_old_csaddr; 1397 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1398 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1399 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1400 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1401 } 1402 if (fs->fs_magic == FS_UFS1_MAGIC && 1403 fs->fs_old_inodefmt < FS_44INODEFMT) { 1404 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1405 fs->fs_qbmask = ~fs->fs_bmask; 1406 fs->fs_qfmask = ~fs->fs_fmask; 1407 } 1408 if (fs->fs_magic == FS_UFS1_MAGIC) { 1409 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1410 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1411 if (fs->fs_maxfilesize > maxfilesize) 1412 fs->fs_maxfilesize = maxfilesize; 1413 } 1414 /* Compatibility for old filesystems */ 1415 if (fs->fs_avgfilesize <= 0) 1416 fs->fs_avgfilesize = AVFILESIZ; 1417 if (fs->fs_avgfpdir <= 0) 1418 fs->fs_avgfpdir = AFPDIR; 1419 if (bigcgs) { 1420 fs->fs_save_cgsize = fs->fs_cgsize; 1421 fs->fs_cgsize = fs->fs_bsize; 1422 } 1423 } 1424 1425 /* 1426 * Unwinding superblock updates for old filesystems. 1427 * See ffs_oldfscompat_read above for details. 1428 * 1429 * XXX - Parts get retired eventually. 1430 * Unfortunately new bits get added. 1431 */ 1432 void 1433 ffs_oldfscompat_write(fs, ump) 1434 struct fs *fs; 1435 struct ufsmount *ump; 1436 { 1437 1438 /* 1439 * Copy back UFS2 updated fields that UFS1 inspects. 1440 */ 1441 if (fs->fs_magic == FS_UFS1_MAGIC) { 1442 fs->fs_old_time = fs->fs_time; 1443 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1444 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1445 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1446 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1447 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1448 } 1449 if (bigcgs) { 1450 fs->fs_cgsize = fs->fs_save_cgsize; 1451 fs->fs_save_cgsize = 0; 1452 } 1453 } 1454 1455 /* 1456 * unmount system call 1457 */ 1458 static int 1459 ffs_unmount(mp, mntflags) 1460 struct mount *mp; 1461 int mntflags; 1462 { 1463 struct thread *td; 1464 struct ufsmount *ump = VFSTOUFS(mp); 1465 struct fs *fs; 1466 int error, flags, susp; 1467 #ifdef UFS_EXTATTR 1468 int e_restart; 1469 #endif 1470 1471 flags = 0; 1472 td = curthread; 1473 fs = ump->um_fs; 1474 if (mntflags & MNT_FORCE) 1475 flags |= FORCECLOSE; 1476 susp = fs->fs_ronly == 0; 1477 #ifdef UFS_EXTATTR 1478 if ((error = ufs_extattr_stop(mp, td))) { 1479 if (error != EOPNOTSUPP) 1480 printf("WARNING: unmount %s: ufs_extattr_stop " 1481 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1482 error); 1483 e_restart = 0; 1484 } else { 1485 ufs_extattr_uepm_destroy(&ump->um_extattr); 1486 e_restart = 1; 1487 } 1488 #endif 1489 if (susp) { 1490 error = vfs_write_suspend_umnt(mp); 1491 if (error != 0) 1492 goto fail1; 1493 } 1494 if (MOUNTEDSOFTDEP(mp)) 1495 error = softdep_flushfiles(mp, flags, td); 1496 else 1497 error = ffs_flushfiles(mp, flags, td); 1498 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) 1499 goto fail; 1500 1501 UFS_LOCK(ump); 1502 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1503 printf("WARNING: unmount %s: pending error: blocks %jd " 1504 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1505 fs->fs_pendinginodes); 1506 fs->fs_pendingblocks = 0; 1507 fs->fs_pendinginodes = 0; 1508 } 1509 UFS_UNLOCK(ump); 1510 if (MOUNTEDSOFTDEP(mp)) 1511 softdep_unmount(mp); 1512 MPASS(ump->um_softdep == NULL); 1513 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { 1514 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1515 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1516 if (ffs_fsfail_cleanup(ump, error)) 1517 error = 0; 1518 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) { 1519 fs->fs_clean = 0; 1520 goto fail; 1521 } 1522 } 1523 if (susp) 1524 vfs_write_resume(mp, VR_START_WRITE); 1525 if (ump->um_trim_tq != NULL) { 1526 while (ump->um_trim_inflight != 0) 1527 pause("ufsutr", hz); 1528 taskqueue_drain_all(ump->um_trim_tq); 1529 taskqueue_free(ump->um_trim_tq); 1530 free (ump->um_trimhash, M_TRIM); 1531 } 1532 g_topology_lock(); 1533 if (ump->um_fsckpid > 0) { 1534 /* 1535 * Return to normal read-only mode. 1536 */ 1537 error = g_access(ump->um_cp, 0, -1, 0); 1538 ump->um_fsckpid = 0; 1539 } 1540 g_vfs_close(ump->um_cp); 1541 g_topology_unlock(); 1542 BO_LOCK(&ump->um_odevvp->v_bufobj); 1543 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; 1544 BO_UNLOCK(&ump->um_odevvp->v_bufobj); 1545 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1546 mntfs_freevp(ump->um_devvp); 1547 vrele(ump->um_odevvp); 1548 dev_rel(ump->um_dev); 1549 mtx_destroy(UFS_MTX(ump)); 1550 sx_destroy(&ump->um_checkpath_lock); 1551 if (mp->mnt_gjprovider != NULL) { 1552 free(mp->mnt_gjprovider, M_UFSMNT); 1553 mp->mnt_gjprovider = NULL; 1554 } 1555 free(fs->fs_csp, M_UFSMNT); 1556 free(fs->fs_si, M_UFSMNT); 1557 free(fs, M_UFSMNT); 1558 free(ump, M_UFSMNT); 1559 mp->mnt_data = NULL; 1560 MNT_ILOCK(mp); 1561 mp->mnt_flag &= ~MNT_LOCAL; 1562 MNT_IUNLOCK(mp); 1563 if (td->td_su == mp) { 1564 td->td_su = NULL; 1565 vfs_rel(mp); 1566 } 1567 return (error); 1568 1569 fail: 1570 if (susp) 1571 vfs_write_resume(mp, VR_START_WRITE); 1572 fail1: 1573 #ifdef UFS_EXTATTR 1574 if (e_restart) { 1575 ufs_extattr_uepm_init(&ump->um_extattr); 1576 #ifdef UFS_EXTATTR_AUTOSTART 1577 (void) ufs_extattr_autostart(mp, td); 1578 #endif 1579 } 1580 #endif 1581 1582 return (error); 1583 } 1584 1585 /* 1586 * Flush out all the files in a filesystem. 1587 */ 1588 int 1589 ffs_flushfiles(mp, flags, td) 1590 struct mount *mp; 1591 int flags; 1592 struct thread *td; 1593 { 1594 struct ufsmount *ump; 1595 int qerror, error; 1596 1597 ump = VFSTOUFS(mp); 1598 qerror = 0; 1599 #ifdef QUOTA 1600 if (mp->mnt_flag & MNT_QUOTA) { 1601 int i; 1602 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1603 if (error) 1604 return (error); 1605 for (i = 0; i < MAXQUOTAS; i++) { 1606 error = quotaoff(td, mp, i); 1607 if (error != 0) { 1608 if ((flags & EARLYFLUSH) == 0) 1609 return (error); 1610 else 1611 qerror = error; 1612 } 1613 } 1614 1615 /* 1616 * Here we fall through to vflush again to ensure that 1617 * we have gotten rid of all the system vnodes, unless 1618 * quotas must not be closed. 1619 */ 1620 } 1621 #endif 1622 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1623 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1624 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1625 return (error); 1626 ffs_snapshot_unmount(mp); 1627 flags |= FORCECLOSE; 1628 /* 1629 * Here we fall through to vflush again to ensure 1630 * that we have gotten rid of all the system vnodes. 1631 */ 1632 } 1633 1634 /* 1635 * Do not close system files if quotas were not closed, to be 1636 * able to sync the remaining dquots. The freeblks softupdate 1637 * workitems might hold a reference on a dquot, preventing 1638 * quotaoff() from completing. Next round of 1639 * softdep_flushworklist() iteration should process the 1640 * blockers, allowing the next run of quotaoff() to finally 1641 * flush held dquots. 1642 * 1643 * Otherwise, flush all the files. 1644 */ 1645 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1646 return (error); 1647 1648 /* 1649 * Flush filesystem metadata. 1650 */ 1651 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1652 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1653 VOP_UNLOCK(ump->um_devvp); 1654 return (error); 1655 } 1656 1657 /* 1658 * Get filesystem statistics. 1659 */ 1660 static int 1661 ffs_statfs(mp, sbp) 1662 struct mount *mp; 1663 struct statfs *sbp; 1664 { 1665 struct ufsmount *ump; 1666 struct fs *fs; 1667 1668 ump = VFSTOUFS(mp); 1669 fs = ump->um_fs; 1670 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1671 panic("ffs_statfs"); 1672 sbp->f_version = STATFS_VERSION; 1673 sbp->f_bsize = fs->fs_fsize; 1674 sbp->f_iosize = fs->fs_bsize; 1675 sbp->f_blocks = fs->fs_dsize; 1676 UFS_LOCK(ump); 1677 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1678 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1679 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1680 dbtofsb(fs, fs->fs_pendingblocks); 1681 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; 1682 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1683 UFS_UNLOCK(ump); 1684 sbp->f_namemax = UFS_MAXNAMLEN; 1685 return (0); 1686 } 1687 1688 static bool 1689 sync_doupdate(struct inode *ip) 1690 { 1691 1692 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1693 IN_UPDATE)) != 0); 1694 } 1695 1696 static int 1697 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) 1698 { 1699 struct inode *ip; 1700 1701 /* 1702 * Flags are safe to access because ->v_data invalidation 1703 * is held off by listmtx. 1704 */ 1705 if (vp->v_type == VNON) 1706 return (false); 1707 ip = VTOI(vp); 1708 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) 1709 return (false); 1710 return (true); 1711 } 1712 1713 /* 1714 * For a lazy sync, we only care about access times, quotas and the 1715 * superblock. Other filesystem changes are already converted to 1716 * cylinder group blocks or inode blocks updates and are written to 1717 * disk by syncer. 1718 */ 1719 static int 1720 ffs_sync_lazy(mp) 1721 struct mount *mp; 1722 { 1723 struct vnode *mvp, *vp; 1724 struct inode *ip; 1725 struct thread *td; 1726 int allerror, error; 1727 1728 allerror = 0; 1729 td = curthread; 1730 if ((mp->mnt_flag & MNT_NOATIME) != 0) { 1731 #ifdef QUOTA 1732 qsync(mp); 1733 #endif 1734 goto sbupdate; 1735 } 1736 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { 1737 if (vp->v_type == VNON) { 1738 VI_UNLOCK(vp); 1739 continue; 1740 } 1741 ip = VTOI(vp); 1742 1743 /* 1744 * The IN_ACCESS flag is converted to IN_MODIFIED by 1745 * ufs_close() and ufs_getattr() by the calls to 1746 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1747 * Test also all the other timestamp flags too, to pick up 1748 * any other cases that could be missed. 1749 */ 1750 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1751 VI_UNLOCK(vp); 1752 continue; 1753 } 1754 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0) 1755 continue; 1756 #ifdef QUOTA 1757 qsyncvp(vp); 1758 #endif 1759 if (sync_doupdate(ip)) 1760 error = ffs_update(vp, 0); 1761 if (error != 0) 1762 allerror = error; 1763 vput(vp); 1764 } 1765 sbupdate: 1766 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1767 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1768 allerror = error; 1769 return (allerror); 1770 } 1771 1772 /* 1773 * Go through the disk queues to initiate sandbagged IO; 1774 * go through the inodes to write those that have been modified; 1775 * initiate the writing of the super block if it has been modified. 1776 * 1777 * Note: we are always called with the filesystem marked busy using 1778 * vfs_busy(). 1779 */ 1780 static int 1781 ffs_sync(mp, waitfor) 1782 struct mount *mp; 1783 int waitfor; 1784 { 1785 struct vnode *mvp, *vp, *devvp; 1786 struct thread *td; 1787 struct inode *ip; 1788 struct ufsmount *ump = VFSTOUFS(mp); 1789 struct fs *fs; 1790 int error, count, lockreq, allerror = 0; 1791 int suspend; 1792 int suspended; 1793 int secondary_writes; 1794 int secondary_accwrites; 1795 int softdep_deps; 1796 int softdep_accdeps; 1797 struct bufobj *bo; 1798 1799 suspend = 0; 1800 suspended = 0; 1801 td = curthread; 1802 fs = ump->um_fs; 1803 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) 1804 panic("%s: ffs_sync: modification on read-only filesystem", 1805 fs->fs_fsmnt); 1806 if (waitfor == MNT_LAZY) { 1807 if (!rebooting) 1808 return (ffs_sync_lazy(mp)); 1809 waitfor = MNT_NOWAIT; 1810 } 1811 1812 /* 1813 * Write back each (modified) inode. 1814 */ 1815 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1816 if (waitfor == MNT_SUSPEND) { 1817 suspend = 1; 1818 waitfor = MNT_WAIT; 1819 } 1820 if (waitfor == MNT_WAIT) 1821 lockreq = LK_EXCLUSIVE; 1822 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1823 loop: 1824 /* Grab snapshot of secondary write counts */ 1825 MNT_ILOCK(mp); 1826 secondary_writes = mp->mnt_secondary_writes; 1827 secondary_accwrites = mp->mnt_secondary_accwrites; 1828 MNT_IUNLOCK(mp); 1829 1830 /* Grab snapshot of softdep dependency counts */ 1831 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1832 1833 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1834 /* 1835 * Depend on the vnode interlock to keep things stable enough 1836 * for a quick test. Since there might be hundreds of 1837 * thousands of vnodes, we cannot afford even a subroutine 1838 * call unless there's a good chance that we have work to do. 1839 */ 1840 if (vp->v_type == VNON) { 1841 VI_UNLOCK(vp); 1842 continue; 1843 } 1844 ip = VTOI(vp); 1845 if ((ip->i_flag & 1846 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1847 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1848 VI_UNLOCK(vp); 1849 continue; 1850 } 1851 if ((error = vget(vp, lockreq)) != 0) { 1852 if (error == ENOENT || error == ENOLCK) { 1853 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1854 goto loop; 1855 } 1856 continue; 1857 } 1858 #ifdef QUOTA 1859 qsyncvp(vp); 1860 #endif 1861 for (;;) { 1862 error = ffs_syncvnode(vp, waitfor, 0); 1863 if (error == ERELOOKUP) 1864 continue; 1865 if (error != 0) 1866 allerror = error; 1867 break; 1868 } 1869 vput(vp); 1870 } 1871 /* 1872 * Force stale filesystem control information to be flushed. 1873 */ 1874 if (waitfor == MNT_WAIT || rebooting) { 1875 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1876 allerror = error; 1877 if (ffs_fsfail_cleanup(ump, allerror)) 1878 allerror = 0; 1879 /* Flushed work items may create new vnodes to clean */ 1880 if (allerror == 0 && count) 1881 goto loop; 1882 } 1883 1884 devvp = ump->um_devvp; 1885 bo = &devvp->v_bufobj; 1886 BO_LOCK(bo); 1887 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1888 BO_UNLOCK(bo); 1889 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1890 error = VOP_FSYNC(devvp, waitfor, td); 1891 VOP_UNLOCK(devvp); 1892 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1893 error = ffs_sbupdate(ump, waitfor, 0); 1894 if (error != 0) 1895 allerror = error; 1896 if (ffs_fsfail_cleanup(ump, allerror)) 1897 allerror = 0; 1898 if (allerror == 0 && waitfor == MNT_WAIT) 1899 goto loop; 1900 } else if (suspend != 0) { 1901 if (softdep_check_suspend(mp, 1902 devvp, 1903 softdep_deps, 1904 softdep_accdeps, 1905 secondary_writes, 1906 secondary_accwrites) != 0) { 1907 MNT_IUNLOCK(mp); 1908 goto loop; /* More work needed */ 1909 } 1910 mtx_assert(MNT_MTX(mp), MA_OWNED); 1911 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1912 MNT_IUNLOCK(mp); 1913 suspended = 1; 1914 } else 1915 BO_UNLOCK(bo); 1916 /* 1917 * Write back modified superblock. 1918 */ 1919 if (fs->fs_fmod != 0 && 1920 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1921 allerror = error; 1922 if (ffs_fsfail_cleanup(ump, allerror)) 1923 allerror = 0; 1924 return (allerror); 1925 } 1926 1927 int 1928 ffs_vget(mp, ino, flags, vpp) 1929 struct mount *mp; 1930 ino_t ino; 1931 int flags; 1932 struct vnode **vpp; 1933 { 1934 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1935 } 1936 1937 int 1938 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1939 struct mount *mp; 1940 ino_t ino; 1941 int flags; 1942 struct vnode **vpp; 1943 int ffs_flags; 1944 { 1945 struct fs *fs; 1946 struct inode *ip; 1947 struct ufsmount *ump; 1948 struct buf *bp; 1949 struct vnode *vp; 1950 daddr_t dbn; 1951 int error; 1952 1953 MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 || 1954 (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 ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 || 1962 !VN_IS_DOOMED(*vpp))) 1963 return (0); 1964 vgone(*vpp); 1965 vput(*vpp); 1966 } 1967 1968 /* 1969 * We must promote to an exclusive lock for vnode creation. This 1970 * can happen if lookup is passed LOCKSHARED. 1971 */ 1972 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1973 flags &= ~LK_TYPE_MASK; 1974 flags |= LK_EXCLUSIVE; 1975 } 1976 1977 /* 1978 * We do not lock vnode creation as it is believed to be too 1979 * expensive for such rare case as simultaneous creation of vnode 1980 * for same ino by different processes. We just allow them to race 1981 * and check later to decide who wins. Let the race begin! 1982 */ 1983 1984 ump = VFSTOUFS(mp); 1985 fs = ump->um_fs; 1986 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO); 1987 1988 /* Allocate a new vnode/inode. */ 1989 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1990 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1991 if (error) { 1992 *vpp = NULL; 1993 uma_zfree_smr(uma_inode, ip); 1994 return (error); 1995 } 1996 /* 1997 * FFS supports recursive locking. 1998 */ 1999 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 2000 VN_LOCK_AREC(vp); 2001 vp->v_data = ip; 2002 vp->v_bufobj.bo_bsize = fs->fs_bsize; 2003 ip->i_vnode = vp; 2004 ip->i_ump = ump; 2005 ip->i_number = ino; 2006 ip->i_ea_refs = 0; 2007 ip->i_nextclustercg = -1; 2008 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; 2009 ip->i_mode = 0; /* ensure error cases below throw away vnode */ 2010 cluster_init_vn(&ip->i_clusterw); 2011 #ifdef DIAGNOSTIC 2012 ufs_init_trackers(ip); 2013 #endif 2014 #ifdef QUOTA 2015 { 2016 int i; 2017 for (i = 0; i < MAXQUOTAS; i++) 2018 ip->i_dquot[i] = NODQUOT; 2019 } 2020 #endif 2021 2022 if (ffs_flags & FFSV_FORCEINSMQ) 2023 vp->v_vflag |= VV_FORCEINSMQ; 2024 error = insmntque(vp, mp); 2025 if (error != 0) { 2026 uma_zfree_smr(uma_inode, ip); 2027 *vpp = NULL; 2028 return (error); 2029 } 2030 vp->v_vflag &= ~VV_FORCEINSMQ; 2031 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 2032 if (error != 0) 2033 return (error); 2034 if (*vpp != NULL) { 2035 /* 2036 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) 2037 * operate on empty inode, which must not be found by 2038 * other threads until fully filled. Vnode for empty 2039 * inode must be not re-inserted on the hash by other 2040 * thread, after removal by us at the beginning. 2041 */ 2042 MPASS((ffs_flags & FFSV_REPLACE) == 0); 2043 return (0); 2044 } 2045 2046 /* Read in the disk contents for the inode, copy into the inode. */ 2047 dbn = fsbtodb(fs, ino_to_fsba(fs, ino)); 2048 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize, 2049 NULL, NULL, 0, NOCRED, 0, NULL, &bp); 2050 if (error != 0) { 2051 /* 2052 * The inode does not contain anything useful, so it would 2053 * be misleading to leave it on its hash chain. With mode 2054 * still zero, it will be unlinked and returned to the free 2055 * list by vput(). 2056 */ 2057 vgone(vp); 2058 vput(vp); 2059 *vpp = NULL; 2060 return (error); 2061 } 2062 if (I_IS_UFS1(ip)) 2063 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 2064 else 2065 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 2066 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { 2067 bqrelse(bp); 2068 vgone(vp); 2069 vput(vp); 2070 *vpp = NULL; 2071 return (error); 2072 } 2073 if (DOINGSOFTDEP(vp) && (!fs->fs_ronly || 2074 (ffs_flags & FFSV_FORCEINODEDEP) != 0)) 2075 softdep_load_inodeblock(ip); 2076 else 2077 ip->i_effnlink = ip->i_nlink; 2078 bqrelse(bp); 2079 2080 /* 2081 * Initialize the vnode from the inode, check for aliases. 2082 * Note that the underlying vnode may have changed. 2083 */ 2084 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, 2085 &vp); 2086 if (error) { 2087 vgone(vp); 2088 vput(vp); 2089 *vpp = NULL; 2090 return (error); 2091 } 2092 2093 /* 2094 * Finish inode initialization. 2095 */ 2096 if (vp->v_type != VFIFO) { 2097 /* FFS supports shared locking for all files except fifos. */ 2098 VN_LOCK_ASHARE(vp); 2099 } 2100 2101 /* 2102 * Set up a generation number for this inode if it does not 2103 * already have one. This should only happen on old filesystems. 2104 */ 2105 if (ip->i_gen == 0) { 2106 while (ip->i_gen == 0) 2107 ip->i_gen = arc4random(); 2108 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 2109 UFS_INODE_SET_FLAG(ip, IN_MODIFIED); 2110 DIP_SET(ip, i_gen, ip->i_gen); 2111 } 2112 } 2113 #ifdef MAC 2114 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 2115 /* 2116 * If this vnode is already allocated, and we're running 2117 * multi-label, attempt to perform a label association 2118 * from the extended attributes on the inode. 2119 */ 2120 error = mac_vnode_associate_extattr(mp, vp); 2121 if (error) { 2122 /* ufs_inactive will release ip->i_devvp ref. */ 2123 vgone(vp); 2124 vput(vp); 2125 *vpp = NULL; 2126 return (error); 2127 } 2128 } 2129 #endif 2130 2131 *vpp = vp; 2132 return (0); 2133 } 2134 2135 /* 2136 * File handle to vnode 2137 * 2138 * Have to be really careful about stale file handles: 2139 * - check that the inode number is valid 2140 * - for UFS2 check that the inode number is initialized 2141 * - call ffs_vget() to get the locked inode 2142 * - check for an unallocated inode (i_mode == 0) 2143 * - check that the given client host has export rights and return 2144 * those rights via. exflagsp and credanonp 2145 */ 2146 static int 2147 ffs_fhtovp(mp, fhp, flags, vpp) 2148 struct mount *mp; 2149 struct fid *fhp; 2150 int flags; 2151 struct vnode **vpp; 2152 { 2153 struct ufid *ufhp; 2154 2155 ufhp = (struct ufid *)fhp; 2156 return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags, 2157 vpp, 0)); 2158 } 2159 2160 int 2161 ffs_inotovp(mp, ino, gen, lflags, vpp, ffs_flags) 2162 struct mount *mp; 2163 ino_t ino; 2164 u_int64_t gen; 2165 int lflags; 2166 struct vnode **vpp; 2167 int ffs_flags; 2168 { 2169 struct ufsmount *ump; 2170 struct vnode *nvp; 2171 struct inode *ip; 2172 struct fs *fs; 2173 struct cg *cgp; 2174 struct buf *bp; 2175 u_int cg; 2176 int error; 2177 2178 ump = VFSTOUFS(mp); 2179 fs = ump->um_fs; 2180 *vpp = NULL; 2181 2182 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) 2183 return (ESTALE); 2184 2185 /* 2186 * Need to check if inode is initialized because UFS2 does lazy 2187 * initialization and nfs_fhtovp can offer arbitrary inode numbers. 2188 */ 2189 if (fs->fs_magic == FS_UFS2_MAGIC) { 2190 cg = ino_to_cg(fs, ino); 2191 error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp); 2192 if (error != 0) 2193 return (error); 2194 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { 2195 brelse(bp); 2196 return (ESTALE); 2197 } 2198 brelse(bp); 2199 } 2200 2201 error = ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags); 2202 if (error != 0) 2203 return (error); 2204 2205 ip = VTOI(nvp); 2206 if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) { 2207 if (ip->i_mode == 0) 2208 vgone(nvp); 2209 vput(nvp); 2210 return (ESTALE); 2211 } 2212 2213 vnode_create_vobject(nvp, DIP(ip, i_size), curthread); 2214 *vpp = nvp; 2215 return (0); 2216 } 2217 2218 /* 2219 * Initialize the filesystem. 2220 */ 2221 static int 2222 ffs_init(vfsp) 2223 struct vfsconf *vfsp; 2224 { 2225 2226 ffs_susp_initialize(); 2227 softdep_initialize(); 2228 return (ufs_init(vfsp)); 2229 } 2230 2231 /* 2232 * Undo the work of ffs_init(). 2233 */ 2234 static int 2235 ffs_uninit(vfsp) 2236 struct vfsconf *vfsp; 2237 { 2238 int ret; 2239 2240 ret = ufs_uninit(vfsp); 2241 softdep_uninitialize(); 2242 ffs_susp_uninitialize(); 2243 taskqueue_drain_all(taskqueue_thread); 2244 return (ret); 2245 } 2246 2247 /* 2248 * Structure used to pass information from ffs_sbupdate to its 2249 * helper routine ffs_use_bwrite. 2250 */ 2251 struct devfd { 2252 struct ufsmount *ump; 2253 struct buf *sbbp; 2254 int waitfor; 2255 int suspended; 2256 int error; 2257 }; 2258 2259 /* 2260 * Write a superblock and associated information back to disk. 2261 */ 2262 int 2263 ffs_sbupdate(ump, waitfor, suspended) 2264 struct ufsmount *ump; 2265 int waitfor; 2266 int suspended; 2267 { 2268 struct fs *fs; 2269 struct buf *sbbp; 2270 struct devfd devfd; 2271 2272 fs = ump->um_fs; 2273 if (fs->fs_ronly == 1 && 2274 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 2275 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) 2276 panic("ffs_sbupdate: write read-only filesystem"); 2277 /* 2278 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 2279 */ 2280 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 2281 (int)fs->fs_sbsize, 0, 0, 0); 2282 /* 2283 * Initialize info needed for write function. 2284 */ 2285 devfd.ump = ump; 2286 devfd.sbbp = sbbp; 2287 devfd.waitfor = waitfor; 2288 devfd.suspended = suspended; 2289 devfd.error = 0; 2290 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); 2291 } 2292 2293 /* 2294 * Write function for use by filesystem-layer routines. 2295 */ 2296 static int 2297 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) 2298 { 2299 struct devfd *devfdp; 2300 struct ufsmount *ump; 2301 struct buf *bp; 2302 struct fs *fs; 2303 int error; 2304 2305 devfdp = devfd; 2306 ump = devfdp->ump; 2307 fs = ump->um_fs; 2308 /* 2309 * Writing the superblock summary information. 2310 */ 2311 if (loc != fs->fs_sblockloc) { 2312 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); 2313 bcopy(buf, bp->b_data, (u_int)size); 2314 if (devfdp->suspended) 2315 bp->b_flags |= B_VALIDSUSPWRT; 2316 if (devfdp->waitfor != MNT_WAIT) 2317 bawrite(bp); 2318 else if ((error = bwrite(bp)) != 0) 2319 devfdp->error = error; 2320 return (0); 2321 } 2322 /* 2323 * Writing the superblock itself. We need to do special checks for it. 2324 */ 2325 bp = devfdp->sbbp; 2326 if (ffs_fsfail_cleanup(ump, devfdp->error)) 2327 devfdp->error = 0; 2328 if (devfdp->error != 0) { 2329 brelse(bp); 2330 return (devfdp->error); 2331 } 2332 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 2333 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2334 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2335 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 2336 fs->fs_sblockloc = SBLOCK_UFS1; 2337 } 2338 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 2339 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 2340 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 2341 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 2342 fs->fs_sblockloc = SBLOCK_UFS2; 2343 } 2344 if (MOUNTEDSOFTDEP(ump->um_mountp)) 2345 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 2346 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 2347 fs = (struct fs *)bp->b_data; 2348 ffs_oldfscompat_write(fs, ump); 2349 fs->fs_si = NULL; 2350 /* Recalculate the superblock hash */ 2351 fs->fs_ckhash = ffs_calc_sbhash(fs); 2352 if (devfdp->suspended) 2353 bp->b_flags |= B_VALIDSUSPWRT; 2354 if (devfdp->waitfor != MNT_WAIT) 2355 bawrite(bp); 2356 else if ((error = bwrite(bp)) != 0) 2357 devfdp->error = error; 2358 return (devfdp->error); 2359 } 2360 2361 static int 2362 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 2363 int attrnamespace, const char *attrname) 2364 { 2365 2366 #ifdef UFS_EXTATTR 2367 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 2368 attrname)); 2369 #else 2370 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 2371 attrname)); 2372 #endif 2373 } 2374 2375 static void 2376 ffs_ifree(struct ufsmount *ump, struct inode *ip) 2377 { 2378 2379 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 2380 uma_zfree(uma_ufs1, ip->i_din1); 2381 else if (ip->i_din2 != NULL) 2382 uma_zfree(uma_ufs2, ip->i_din2); 2383 uma_zfree_smr(uma_inode, ip); 2384 } 2385 2386 static int dobkgrdwrite = 1; 2387 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 2388 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 2389 2390 /* 2391 * Complete a background write started from bwrite. 2392 */ 2393 static void 2394 ffs_backgroundwritedone(struct buf *bp) 2395 { 2396 struct bufobj *bufobj; 2397 struct buf *origbp; 2398 2399 #ifdef SOFTUPDATES 2400 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0) 2401 softdep_handle_error(bp); 2402 #endif 2403 2404 /* 2405 * Find the original buffer that we are writing. 2406 */ 2407 bufobj = bp->b_bufobj; 2408 BO_LOCK(bufobj); 2409 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2410 panic("backgroundwritedone: lost buffer"); 2411 2412 /* 2413 * We should mark the cylinder group buffer origbp as 2414 * dirty, to not lose the failed write. 2415 */ 2416 if ((bp->b_ioflags & BIO_ERROR) != 0) 2417 origbp->b_vflags |= BV_BKGRDERR; 2418 BO_UNLOCK(bufobj); 2419 /* 2420 * Process dependencies then return any unfinished ones. 2421 */ 2422 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2423 buf_complete(bp); 2424 #ifdef SOFTUPDATES 2425 if (!LIST_EMPTY(&bp->b_dep)) 2426 softdep_move_dependencies(bp, origbp); 2427 #endif 2428 /* 2429 * This buffer is marked B_NOCACHE so when it is released 2430 * by biodone it will be tossed. Clear B_IOSTARTED in case of error. 2431 */ 2432 bp->b_flags |= B_NOCACHE; 2433 bp->b_flags &= ~(B_CACHE | B_IOSTARTED); 2434 pbrelvp(bp); 2435 2436 /* 2437 * Prevent brelse() from trying to keep and re-dirtying bp on 2438 * errors. It causes b_bufobj dereference in 2439 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2440 * pbrelvp() above. 2441 */ 2442 if ((bp->b_ioflags & BIO_ERROR) != 0) 2443 bp->b_flags |= B_INVAL; 2444 bufdone(bp); 2445 BO_LOCK(bufobj); 2446 /* 2447 * Clear the BV_BKGRDINPROG flag in the original buffer 2448 * and awaken it if it is waiting for the write to complete. 2449 * If BV_BKGRDINPROG is not set in the original buffer it must 2450 * have been released and re-instantiated - which is not legal. 2451 */ 2452 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2453 ("backgroundwritedone: lost buffer2")); 2454 origbp->b_vflags &= ~BV_BKGRDINPROG; 2455 if (origbp->b_vflags & BV_BKGRDWAIT) { 2456 origbp->b_vflags &= ~BV_BKGRDWAIT; 2457 wakeup(&origbp->b_xflags); 2458 } 2459 BO_UNLOCK(bufobj); 2460 } 2461 2462 /* 2463 * Write, release buffer on completion. (Done by iodone 2464 * if async). Do not bother writing anything if the buffer 2465 * is invalid. 2466 * 2467 * Note that we set B_CACHE here, indicating that buffer is 2468 * fully valid and thus cacheable. This is true even of NFS 2469 * now so we set it generally. This could be set either here 2470 * or in biodone() since the I/O is synchronous. We put it 2471 * here. 2472 */ 2473 static int 2474 ffs_bufwrite(struct buf *bp) 2475 { 2476 struct buf *newbp; 2477 struct cg *cgp; 2478 2479 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2480 if (bp->b_flags & B_INVAL) { 2481 brelse(bp); 2482 return (0); 2483 } 2484 2485 if (!BUF_ISLOCKED(bp)) 2486 panic("bufwrite: buffer is not busy???"); 2487 /* 2488 * If a background write is already in progress, delay 2489 * writing this block if it is asynchronous. Otherwise 2490 * wait for the background write to complete. 2491 */ 2492 BO_LOCK(bp->b_bufobj); 2493 if (bp->b_vflags & BV_BKGRDINPROG) { 2494 if (bp->b_flags & B_ASYNC) { 2495 BO_UNLOCK(bp->b_bufobj); 2496 bdwrite(bp); 2497 return (0); 2498 } 2499 bp->b_vflags |= BV_BKGRDWAIT; 2500 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2501 "bwrbg", 0); 2502 if (bp->b_vflags & BV_BKGRDINPROG) 2503 panic("bufwrite: still writing"); 2504 } 2505 bp->b_vflags &= ~BV_BKGRDERR; 2506 BO_UNLOCK(bp->b_bufobj); 2507 2508 /* 2509 * If this buffer is marked for background writing and we 2510 * do not have to wait for it, make a copy and write the 2511 * copy so as to leave this buffer ready for further use. 2512 * 2513 * This optimization eats a lot of memory. If we have a page 2514 * or buffer shortfall we can't do it. 2515 */ 2516 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2517 (bp->b_flags & B_ASYNC) && 2518 !vm_page_count_severe() && 2519 !buf_dirty_count_severe()) { 2520 KASSERT(bp->b_iodone == NULL, 2521 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2522 2523 /* get a new block */ 2524 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2525 if (newbp == NULL) 2526 goto normal_write; 2527 2528 KASSERT(buf_mapped(bp), ("Unmapped cg")); 2529 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2530 BO_LOCK(bp->b_bufobj); 2531 bp->b_vflags |= BV_BKGRDINPROG; 2532 BO_UNLOCK(bp->b_bufobj); 2533 newbp->b_xflags |= 2534 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; 2535 newbp->b_lblkno = bp->b_lblkno; 2536 newbp->b_blkno = bp->b_blkno; 2537 newbp->b_offset = bp->b_offset; 2538 newbp->b_iodone = ffs_backgroundwritedone; 2539 newbp->b_flags |= B_ASYNC; 2540 newbp->b_flags &= ~B_INVAL; 2541 pbgetvp(bp->b_vp, newbp); 2542 2543 #ifdef SOFTUPDATES 2544 /* 2545 * Move over the dependencies. If there are rollbacks, 2546 * leave the parent buffer dirtied as it will need to 2547 * be written again. 2548 */ 2549 if (LIST_EMPTY(&bp->b_dep) || 2550 softdep_move_dependencies(bp, newbp) == 0) 2551 bundirty(bp); 2552 #else 2553 bundirty(bp); 2554 #endif 2555 2556 /* 2557 * Initiate write on the copy, release the original. The 2558 * BKGRDINPROG flag prevents it from going away until 2559 * the background write completes. We have to recalculate 2560 * its check hash in case the buffer gets freed and then 2561 * reconstituted from the buffer cache during a later read. 2562 */ 2563 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2564 cgp = (struct cg *)bp->b_data; 2565 cgp->cg_ckhash = 0; 2566 cgp->cg_ckhash = 2567 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2568 } 2569 bqrelse(bp); 2570 bp = newbp; 2571 } else 2572 /* Mark the buffer clean */ 2573 bundirty(bp); 2574 2575 /* Let the normal bufwrite do the rest for us */ 2576 normal_write: 2577 /* 2578 * If we are writing a cylinder group, update its time. 2579 */ 2580 if ((bp->b_xflags & BX_CYLGRP) != 0) { 2581 cgp = (struct cg *)bp->b_data; 2582 cgp->cg_old_time = cgp->cg_time = time_second; 2583 } 2584 return (bufwrite(bp)); 2585 } 2586 2587 static void 2588 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2589 { 2590 struct vnode *vp; 2591 struct buf *tbp; 2592 int error, nocopy; 2593 2594 /* 2595 * This is the bufobj strategy for the private VCHR vnodes 2596 * used by FFS to access the underlying storage device. 2597 * We override the default bufobj strategy and thus bypass 2598 * VOP_STRATEGY() for these vnodes. 2599 */ 2600 vp = bo2vnode(bo); 2601 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR || 2602 bp->b_vp->v_rdev == NULL || 2603 bp->b_vp->v_rdev->si_mountpt == NULL || 2604 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL || 2605 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp, 2606 ("ffs_geom_strategy() with wrong vp")); 2607 if (bp->b_iocmd == BIO_WRITE) { 2608 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2609 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2610 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2611 panic("ffs_geom_strategy: bad I/O"); 2612 nocopy = bp->b_flags & B_NOCOPY; 2613 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2614 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2615 vp->v_rdev->si_snapdata != NULL) { 2616 if ((bp->b_flags & B_CLUSTER) != 0) { 2617 runningbufwakeup(bp); 2618 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2619 b_cluster.cluster_entry) { 2620 error = ffs_copyonwrite(vp, tbp); 2621 if (error != 0 && 2622 error != EOPNOTSUPP) { 2623 bp->b_error = error; 2624 bp->b_ioflags |= BIO_ERROR; 2625 bp->b_flags &= ~B_BARRIER; 2626 bufdone(bp); 2627 return; 2628 } 2629 } 2630 bp->b_runningbufspace = bp->b_bufsize; 2631 atomic_add_long(&runningbufspace, 2632 bp->b_runningbufspace); 2633 } else { 2634 error = ffs_copyonwrite(vp, bp); 2635 if (error != 0 && error != EOPNOTSUPP) { 2636 bp->b_error = error; 2637 bp->b_ioflags |= BIO_ERROR; 2638 bp->b_flags &= ~B_BARRIER; 2639 bufdone(bp); 2640 return; 2641 } 2642 } 2643 } 2644 #ifdef SOFTUPDATES 2645 if ((bp->b_flags & B_CLUSTER) != 0) { 2646 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2647 b_cluster.cluster_entry) { 2648 if (!LIST_EMPTY(&tbp->b_dep)) 2649 buf_start(tbp); 2650 } 2651 } else { 2652 if (!LIST_EMPTY(&bp->b_dep)) 2653 buf_start(bp); 2654 } 2655 2656 #endif 2657 /* 2658 * Check for metadata that needs check-hashes and update them. 2659 */ 2660 switch (bp->b_xflags & BX_FSPRIV) { 2661 case BX_CYLGRP: 2662 ((struct cg *)bp->b_data)->cg_ckhash = 0; 2663 ((struct cg *)bp->b_data)->cg_ckhash = 2664 calculate_crc32c(~0L, bp->b_data, bp->b_bcount); 2665 break; 2666 2667 case BX_SUPERBLOCK: 2668 case BX_INODE: 2669 case BX_INDIR: 2670 case BX_DIR: 2671 printf("Check-hash write is unimplemented!!!\n"); 2672 break; 2673 2674 case 0: 2675 break; 2676 2677 default: 2678 printf("multiple buffer types 0x%b\n", 2679 (u_int)(bp->b_xflags & BX_FSPRIV), 2680 PRINT_UFS_BUF_XFLAGS); 2681 break; 2682 } 2683 } 2684 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable) 2685 bp->b_xflags |= BX_CVTENXIO; 2686 g_vfs_strategy(bo, bp); 2687 } 2688 2689 int 2690 ffs_own_mount(const struct mount *mp) 2691 { 2692 2693 if (mp->mnt_op == &ufs_vfsops) 2694 return (1); 2695 return (0); 2696 } 2697 2698 #ifdef DDB 2699 #ifdef SOFTUPDATES 2700 2701 /* defined in ffs_softdep.c */ 2702 extern void db_print_ffs(struct ufsmount *ump); 2703 2704 DB_SHOW_COMMAND(ffs, db_show_ffs) 2705 { 2706 struct mount *mp; 2707 struct ufsmount *ump; 2708 2709 if (have_addr) { 2710 ump = VFSTOUFS((struct mount *)addr); 2711 db_print_ffs(ump); 2712 return; 2713 } 2714 2715 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2716 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2717 db_print_ffs(VFSTOUFS(mp)); 2718 } 2719 } 2720 2721 #endif /* SOFTUPDATES */ 2722 #endif /* DDB */ 2723