1 /*- 2 * Copyright (c) 1989, 1991, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_mac.h" 36 #include "opt_quota.h" 37 #include "opt_ufs.h" 38 #include "opt_ffs.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/namei.h> 43 #include <sys/priv.h> 44 #include <sys/proc.h> 45 #include <sys/kernel.h> 46 #include <sys/vnode.h> 47 #include <sys/mount.h> 48 #include <sys/bio.h> 49 #include <sys/buf.h> 50 #include <sys/conf.h> 51 #include <sys/fcntl.h> 52 #include <sys/malloc.h> 53 #include <sys/mutex.h> 54 55 #include <security/mac/mac_framework.h> 56 57 #include <ufs/ufs/extattr.h> 58 #include <ufs/ufs/gjournal.h> 59 #include <ufs/ufs/quota.h> 60 #include <ufs/ufs/ufsmount.h> 61 #include <ufs/ufs/inode.h> 62 #include <ufs/ufs/ufs_extern.h> 63 64 #include <ufs/ffs/fs.h> 65 #include <ufs/ffs/ffs_extern.h> 66 67 #include <vm/vm.h> 68 #include <vm/uma.h> 69 #include <vm/vm_page.h> 70 71 #include <geom/geom.h> 72 #include <geom/geom_vfs.h> 73 74 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 75 76 static int ffs_reload(struct mount *, struct thread *); 77 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 78 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 79 ufs2_daddr_t); 80 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *); 81 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 82 static vfs_init_t ffs_init; 83 static vfs_uninit_t ffs_uninit; 84 static vfs_extattrctl_t ffs_extattrctl; 85 static vfs_cmount_t ffs_cmount; 86 static vfs_unmount_t ffs_unmount; 87 static vfs_mount_t ffs_mount; 88 static vfs_statfs_t ffs_statfs; 89 static vfs_fhtovp_t ffs_fhtovp; 90 static vfs_vptofh_t ffs_vptofh; 91 static vfs_sync_t ffs_sync; 92 93 static struct vfsops ufs_vfsops = { 94 .vfs_extattrctl = ffs_extattrctl, 95 .vfs_fhtovp = ffs_fhtovp, 96 .vfs_init = ffs_init, 97 .vfs_mount = ffs_mount, 98 .vfs_cmount = ffs_cmount, 99 .vfs_quotactl = ufs_quotactl, 100 .vfs_root = ufs_root, 101 .vfs_statfs = ffs_statfs, 102 .vfs_sync = ffs_sync, 103 .vfs_uninit = ffs_uninit, 104 .vfs_unmount = ffs_unmount, 105 .vfs_vget = ffs_vget, 106 .vfs_vptofh = ffs_vptofh, 107 }; 108 109 VFS_SET(ufs_vfsops, ufs, 0); 110 MODULE_VERSION(ufs, 1); 111 112 static b_strategy_t ffs_geom_strategy; 113 static b_write_t ffs_bufwrite; 114 115 static struct buf_ops ffs_ops = { 116 .bop_name = "FFS", 117 .bop_write = ffs_bufwrite, 118 .bop_strategy = ffs_geom_strategy, 119 .bop_sync = bufsync, 120 }; 121 122 static const char *ffs_opts[] = { "acls", "async", "atime", "clusterr", 123 "clusterw", "exec", "export", "force", "from", "multilabel", 124 "snapshot", "suid", "suiddir", "symfollow", "sync", 125 "union", NULL }; 126 127 static int 128 ffs_mount(struct mount *mp, struct thread *td) 129 { 130 struct vnode *devvp; 131 struct ufsmount *ump = 0; 132 struct fs *fs; 133 int error, flags; 134 u_int mntorflags, mntandnotflags; 135 mode_t accessmode; 136 struct nameidata ndp; 137 char *fspec; 138 139 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 140 return (EINVAL); 141 if (uma_inode == NULL) { 142 uma_inode = uma_zcreate("FFS inode", 143 sizeof(struct inode), NULL, NULL, NULL, NULL, 144 UMA_ALIGN_PTR, 0); 145 uma_ufs1 = uma_zcreate("FFS1 dinode", 146 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 147 UMA_ALIGN_PTR, 0); 148 uma_ufs2 = uma_zcreate("FFS2 dinode", 149 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 150 UMA_ALIGN_PTR, 0); 151 } 152 153 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 154 if (error) 155 return (error); 156 157 mntorflags = 0; 158 mntandnotflags = 0; 159 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 160 mntorflags |= MNT_ACLS; 161 162 if (vfs_getopt(mp->mnt_optnew, "async", NULL, NULL) == 0) 163 mntorflags |= MNT_ASYNC; 164 165 if (vfs_getopt(mp->mnt_optnew, "force", NULL, NULL) == 0) 166 mntorflags |= MNT_FORCE; 167 168 if (vfs_getopt(mp->mnt_optnew, "multilabel", NULL, NULL) == 0) 169 mntorflags |= MNT_MULTILABEL; 170 171 if (vfs_getopt(mp->mnt_optnew, "noasync", NULL, NULL) == 0) 172 mntandnotflags |= MNT_ASYNC; 173 174 if (vfs_getopt(mp->mnt_optnew, "noatime", NULL, NULL) == 0) 175 mntorflags |= MNT_NOATIME; 176 177 if (vfs_getopt(mp->mnt_optnew, "noclusterr", NULL, NULL) == 0) 178 mntorflags |= MNT_NOCLUSTERR; 179 180 if (vfs_getopt(mp->mnt_optnew, "noclusterw", NULL, NULL) == 0) 181 mntorflags |= MNT_NOCLUSTERW; 182 183 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) 184 mntorflags |= MNT_SNAPSHOT; 185 186 MNT_ILOCK(mp); 187 mp->mnt_flag = (mp->mnt_flag | mntorflags) & ~mntandnotflags; 188 MNT_IUNLOCK(mp); 189 /* 190 * If updating, check whether changing from read-only to 191 * read/write; if there is no device name, that's all we do. 192 */ 193 if (mp->mnt_flag & MNT_UPDATE) { 194 ump = VFSTOUFS(mp); 195 fs = ump->um_fs; 196 devvp = ump->um_devvp; 197 if (fs->fs_ronly == 0 && 198 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 199 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 200 return (error); 201 /* 202 * Flush any dirty data. 203 */ 204 if ((error = ffs_sync(mp, MNT_WAIT, td)) != 0) { 205 vn_finished_write(mp); 206 return (error); 207 } 208 /* 209 * Check for and optionally get rid of files open 210 * for writing. 211 */ 212 flags = WRITECLOSE; 213 if (mp->mnt_flag & MNT_FORCE) 214 flags |= FORCECLOSE; 215 if (mp->mnt_flag & MNT_SOFTDEP) { 216 error = softdep_flushfiles(mp, flags, td); 217 } else { 218 error = ffs_flushfiles(mp, flags, td); 219 } 220 if (error) { 221 vn_finished_write(mp); 222 return (error); 223 } 224 if (fs->fs_pendingblocks != 0 || 225 fs->fs_pendinginodes != 0) { 226 printf("%s: %s: blocks %jd files %d\n", 227 fs->fs_fsmnt, "update error", 228 (intmax_t)fs->fs_pendingblocks, 229 fs->fs_pendinginodes); 230 fs->fs_pendingblocks = 0; 231 fs->fs_pendinginodes = 0; 232 } 233 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 234 fs->fs_clean = 1; 235 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 236 fs->fs_ronly = 0; 237 fs->fs_clean = 0; 238 vn_finished_write(mp); 239 return (error); 240 } 241 vn_finished_write(mp); 242 DROP_GIANT(); 243 g_topology_lock(); 244 g_access(ump->um_cp, 0, -1, 0); 245 g_topology_unlock(); 246 PICKUP_GIANT(); 247 fs->fs_ronly = 1; 248 MNT_ILOCK(mp); 249 mp->mnt_flag |= MNT_RDONLY; 250 MNT_IUNLOCK(mp); 251 } 252 if ((mp->mnt_flag & MNT_RELOAD) && 253 (error = ffs_reload(mp, td)) != 0) 254 return (error); 255 if (fs->fs_ronly && 256 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 257 /* 258 * If upgrade to read-write by non-root, then verify 259 * that user has necessary permissions on the device. 260 */ 261 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 262 error = VOP_ACCESS(devvp, VREAD | VWRITE, 263 td->td_ucred, td); 264 if (error) 265 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 266 if (error) { 267 VOP_UNLOCK(devvp, 0, td); 268 return (error); 269 } 270 VOP_UNLOCK(devvp, 0, td); 271 fs->fs_flags &= ~FS_UNCLEAN; 272 if (fs->fs_clean == 0) { 273 fs->fs_flags |= FS_UNCLEAN; 274 if ((mp->mnt_flag & MNT_FORCE) || 275 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 276 (fs->fs_flags & FS_DOSOFTDEP))) { 277 printf("WARNING: %s was not %s\n", 278 fs->fs_fsmnt, "properly dismounted"); 279 } else { 280 printf( 281 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 282 fs->fs_fsmnt); 283 return (EPERM); 284 } 285 } 286 DROP_GIANT(); 287 g_topology_lock(); 288 /* 289 * If we're the root device, we may not have an E count 290 * yet, get it now. 291 */ 292 if (ump->um_cp->ace == 0) 293 error = g_access(ump->um_cp, 0, 1, 1); 294 else 295 error = g_access(ump->um_cp, 0, 1, 0); 296 g_topology_unlock(); 297 PICKUP_GIANT(); 298 if (error) 299 return (error); 300 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 301 return (error); 302 fs->fs_ronly = 0; 303 MNT_ILOCK(mp); 304 mp->mnt_flag &= ~MNT_RDONLY; 305 MNT_IUNLOCK(mp); 306 fs->fs_clean = 0; 307 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 308 vn_finished_write(mp); 309 return (error); 310 } 311 /* check to see if we need to start softdep */ 312 if ((fs->fs_flags & FS_DOSOFTDEP) && 313 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 314 vn_finished_write(mp); 315 return (error); 316 } 317 if (fs->fs_snapinum[0] != 0) 318 ffs_snapshot_mount(mp); 319 vn_finished_write(mp); 320 } 321 /* 322 * Soft updates is incompatible with "async", 323 * so if we are doing softupdates stop the user 324 * from setting the async flag in an update. 325 * Softdep_mount() clears it in an initial mount 326 * or ro->rw remount. 327 */ 328 if (mp->mnt_flag & MNT_SOFTDEP) { 329 /* XXX: Reset too late ? */ 330 MNT_ILOCK(mp); 331 mp->mnt_flag &= ~MNT_ASYNC; 332 MNT_IUNLOCK(mp); 333 } 334 /* 335 * Keep MNT_ACLS flag if it is stored in superblock. 336 */ 337 if ((fs->fs_flags & FS_ACLS) != 0) { 338 /* XXX: Set too late ? */ 339 MNT_ILOCK(mp); 340 mp->mnt_flag |= MNT_ACLS; 341 MNT_IUNLOCK(mp); 342 } 343 344 /* 345 * If this is a snapshot request, take the snapshot. 346 */ 347 if (mp->mnt_flag & MNT_SNAPSHOT) 348 return (ffs_snapshot(mp, fspec)); 349 } 350 351 /* 352 * Not an update, or updating the name: look up the name 353 * and verify that it refers to a sensible disk device. 354 */ 355 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 356 if ((error = namei(&ndp)) != 0) 357 return (error); 358 NDFREE(&ndp, NDF_ONLY_PNBUF); 359 devvp = ndp.ni_vp; 360 if (!vn_isdisk(devvp, &error)) { 361 vput(devvp); 362 return (error); 363 } 364 365 /* 366 * If mount by non-root, then verify that user has necessary 367 * permissions on the device. 368 */ 369 accessmode = VREAD; 370 if ((mp->mnt_flag & MNT_RDONLY) == 0) 371 accessmode |= VWRITE; 372 error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td); 373 if (error) 374 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 375 if (error) { 376 vput(devvp); 377 return (error); 378 } 379 380 if (mp->mnt_flag & MNT_UPDATE) { 381 /* 382 * Update only 383 * 384 * If it's not the same vnode, or at least the same device 385 * then it's not correct. 386 */ 387 388 if (devvp->v_rdev != ump->um_devvp->v_rdev) 389 error = EINVAL; /* needs translation */ 390 vput(devvp); 391 if (error) 392 return (error); 393 } else { 394 /* 395 * New mount 396 * 397 * We need the name for the mount point (also used for 398 * "last mounted on") copied in. If an error occurs, 399 * the mount point is discarded by the upper level code. 400 * Note that vfs_mount() populates f_mntonname for us. 401 */ 402 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 403 vrele(devvp); 404 return (error); 405 } 406 } 407 vfs_mountedfrom(mp, fspec); 408 return (0); 409 } 410 411 /* 412 * Compatibility with old mount system call. 413 */ 414 415 static int 416 ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td) 417 { 418 struct ufs_args args; 419 int error; 420 421 if (data == NULL) 422 return (EINVAL); 423 error = copyin(data, &args, sizeof args); 424 if (error) 425 return (error); 426 427 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 428 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 429 error = kernel_mount(ma, flags); 430 431 return (error); 432 } 433 434 /* 435 * Reload all incore data for a filesystem (used after running fsck on 436 * the root filesystem and finding things to fix). The filesystem must 437 * be mounted read-only. 438 * 439 * Things to do to update the mount: 440 * 1) invalidate all cached meta-data. 441 * 2) re-read superblock from disk. 442 * 3) re-read summary information from disk. 443 * 4) invalidate all inactive vnodes. 444 * 5) invalidate all cached file data. 445 * 6) re-read inode data for all active vnodes. 446 */ 447 static int 448 ffs_reload(struct mount *mp, struct thread *td) 449 { 450 struct vnode *vp, *mvp, *devvp; 451 struct inode *ip; 452 void *space; 453 struct buf *bp; 454 struct fs *fs, *newfs; 455 struct ufsmount *ump; 456 ufs2_daddr_t sblockloc; 457 int i, blks, size, error; 458 int32_t *lp; 459 460 if ((mp->mnt_flag & MNT_RDONLY) == 0) 461 return (EINVAL); 462 ump = VFSTOUFS(mp); 463 /* 464 * Step 1: invalidate all cached meta-data. 465 */ 466 devvp = VFSTOUFS(mp)->um_devvp; 467 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 468 if (vinvalbuf(devvp, 0, td, 0, 0) != 0) 469 panic("ffs_reload: dirty1"); 470 VOP_UNLOCK(devvp, 0, td); 471 472 /* 473 * Step 2: re-read superblock from disk. 474 */ 475 fs = VFSTOUFS(mp)->um_fs; 476 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 477 NOCRED, &bp)) != 0) 478 return (error); 479 newfs = (struct fs *)bp->b_data; 480 if ((newfs->fs_magic != FS_UFS1_MAGIC && 481 newfs->fs_magic != FS_UFS2_MAGIC) || 482 newfs->fs_bsize > MAXBSIZE || 483 newfs->fs_bsize < sizeof(struct fs)) { 484 brelse(bp); 485 return (EIO); /* XXX needs translation */ 486 } 487 /* 488 * Copy pointer fields back into superblock before copying in XXX 489 * new superblock. These should really be in the ufsmount. XXX 490 * Note that important parameters (eg fs_ncg) are unchanged. 491 */ 492 newfs->fs_csp = fs->fs_csp; 493 newfs->fs_maxcluster = fs->fs_maxcluster; 494 newfs->fs_contigdirs = fs->fs_contigdirs; 495 newfs->fs_active = fs->fs_active; 496 /* The file system is still read-only. */ 497 newfs->fs_ronly = 1; 498 sblockloc = fs->fs_sblockloc; 499 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 500 brelse(bp); 501 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 502 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 503 UFS_LOCK(ump); 504 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 505 printf("%s: reload pending error: blocks %jd files %d\n", 506 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 507 fs->fs_pendinginodes); 508 fs->fs_pendingblocks = 0; 509 fs->fs_pendinginodes = 0; 510 } 511 UFS_UNLOCK(ump); 512 513 /* 514 * Step 3: re-read summary information from disk. 515 */ 516 blks = howmany(fs->fs_cssize, fs->fs_fsize); 517 space = fs->fs_csp; 518 for (i = 0; i < blks; i += fs->fs_frag) { 519 size = fs->fs_bsize; 520 if (i + fs->fs_frag > blks) 521 size = (blks - i) * fs->fs_fsize; 522 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 523 NOCRED, &bp); 524 if (error) 525 return (error); 526 bcopy(bp->b_data, space, (u_int)size); 527 space = (char *)space + size; 528 brelse(bp); 529 } 530 /* 531 * We no longer know anything about clusters per cylinder group. 532 */ 533 if (fs->fs_contigsumsize > 0) { 534 lp = fs->fs_maxcluster; 535 for (i = 0; i < fs->fs_ncg; i++) 536 *lp++ = fs->fs_contigsumsize; 537 } 538 539 loop: 540 MNT_ILOCK(mp); 541 MNT_VNODE_FOREACH(vp, mp, mvp) { 542 VI_LOCK(vp); 543 if (vp->v_iflag & VI_DOOMED) { 544 VI_UNLOCK(vp); 545 continue; 546 } 547 MNT_IUNLOCK(mp); 548 /* 549 * Step 4: invalidate all cached file data. 550 */ 551 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 552 MNT_VNODE_FOREACH_ABORT(mp, mvp); 553 goto loop; 554 } 555 if (vinvalbuf(vp, 0, td, 0, 0)) 556 panic("ffs_reload: dirty2"); 557 /* 558 * Step 5: re-read inode data for all active vnodes. 559 */ 560 ip = VTOI(vp); 561 error = 562 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 563 (int)fs->fs_bsize, NOCRED, &bp); 564 if (error) { 565 VOP_UNLOCK(vp, 0, td); 566 vrele(vp); 567 MNT_VNODE_FOREACH_ABORT(mp, mvp); 568 return (error); 569 } 570 ffs_load_inode(bp, ip, fs, ip->i_number); 571 ip->i_effnlink = ip->i_nlink; 572 brelse(bp); 573 VOP_UNLOCK(vp, 0, td); 574 vrele(vp); 575 MNT_ILOCK(mp); 576 } 577 MNT_IUNLOCK(mp); 578 return (0); 579 } 580 581 /* 582 * Possible superblock locations ordered from most to least likely. 583 */ 584 static int sblock_try[] = SBLOCKSEARCH; 585 586 /* 587 * Common code for mount and mountroot 588 */ 589 static int 590 ffs_mountfs(devvp, mp, td) 591 struct vnode *devvp; 592 struct mount *mp; 593 struct thread *td; 594 { 595 struct ufsmount *ump; 596 struct buf *bp; 597 struct fs *fs; 598 struct cdev *dev; 599 void *space; 600 ufs2_daddr_t sblockloc; 601 int error, i, blks, size, ronly; 602 int32_t *lp; 603 struct ucred *cred; 604 struct g_consumer *cp; 605 struct mount *nmp; 606 607 dev = devvp->v_rdev; 608 cred = td ? td->td_ucred : NOCRED; 609 610 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 611 DROP_GIANT(); 612 g_topology_lock(); 613 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 614 615 /* 616 * If we are a root mount, drop the E flag so fsck can do its magic. 617 * We will pick it up again when we remount R/W. 618 */ 619 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 620 error = g_access(cp, 0, 0, -1); 621 g_topology_unlock(); 622 PICKUP_GIANT(); 623 VOP_UNLOCK(devvp, 0, td); 624 if (error) 625 return (error); 626 if (devvp->v_rdev->si_iosize_max != 0) 627 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 628 if (mp->mnt_iosize_max > MAXPHYS) 629 mp->mnt_iosize_max = MAXPHYS; 630 631 devvp->v_bufobj.bo_private = cp; 632 devvp->v_bufobj.bo_ops = &ffs_ops; 633 634 bp = NULL; 635 ump = NULL; 636 fs = NULL; 637 sblockloc = 0; 638 /* 639 * Try reading the superblock in each of its possible locations. 640 */ 641 for (i = 0; sblock_try[i] != -1; i++) { 642 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 643 error = EINVAL; 644 vfs_mount_error(mp, 645 "Invalid sectorsize %d for superblock size %d", 646 cp->provider->sectorsize, SBLOCKSIZE); 647 goto out; 648 } 649 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE, 650 cred, &bp)) != 0) 651 goto out; 652 fs = (struct fs *)bp->b_data; 653 sblockloc = sblock_try[i]; 654 if ((fs->fs_magic == FS_UFS1_MAGIC || 655 (fs->fs_magic == FS_UFS2_MAGIC && 656 (fs->fs_sblockloc == sblockloc || 657 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 658 fs->fs_bsize <= MAXBSIZE && 659 fs->fs_bsize >= sizeof(struct fs)) 660 break; 661 brelse(bp); 662 bp = NULL; 663 } 664 if (sblock_try[i] == -1) { 665 error = EINVAL; /* XXX needs translation */ 666 goto out; 667 } 668 fs->fs_fmod = 0; 669 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 670 fs->fs_flags &= ~FS_UNCLEAN; 671 if (fs->fs_clean == 0) { 672 fs->fs_flags |= FS_UNCLEAN; 673 if (ronly || (mp->mnt_flag & MNT_FORCE) || 674 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 675 (fs->fs_flags & FS_DOSOFTDEP))) { 676 printf( 677 "WARNING: %s was not properly dismounted\n", 678 fs->fs_fsmnt); 679 } else { 680 printf( 681 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 682 fs->fs_fsmnt); 683 error = EPERM; 684 goto out; 685 } 686 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 687 (mp->mnt_flag & MNT_FORCE)) { 688 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 689 (intmax_t)fs->fs_pendingblocks, 690 fs->fs_pendinginodes); 691 fs->fs_pendingblocks = 0; 692 fs->fs_pendinginodes = 0; 693 } 694 } 695 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 696 printf("%s: mount pending error: blocks %jd files %d\n", 697 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 698 fs->fs_pendinginodes); 699 fs->fs_pendingblocks = 0; 700 fs->fs_pendinginodes = 0; 701 } 702 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 703 #ifdef UFS_GJOURNAL 704 /* 705 * Get journal provider name. 706 */ 707 size = 1024; 708 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK); 709 if (g_io_getattr("GJOURNAL::provider", cp, &size, 710 mp->mnt_gjprovider) == 0) { 711 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size, 712 M_UFSMNT, M_WAITOK); 713 MNT_ILOCK(mp); 714 mp->mnt_flag |= MNT_GJOURNAL; 715 MNT_IUNLOCK(mp); 716 } else { 717 printf( 718 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n", 719 mp->mnt_stat.f_mntonname); 720 free(mp->mnt_gjprovider, M_UFSMNT); 721 mp->mnt_gjprovider = NULL; 722 } 723 #else 724 printf( 725 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n", 726 mp->mnt_stat.f_mntonname); 727 #endif 728 } else { 729 mp->mnt_gjprovider = NULL; 730 } 731 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 732 ump->um_cp = cp; 733 ump->um_bo = &devvp->v_bufobj; 734 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 735 if (fs->fs_magic == FS_UFS1_MAGIC) { 736 ump->um_fstype = UFS1; 737 ump->um_balloc = ffs_balloc_ufs1; 738 } else { 739 ump->um_fstype = UFS2; 740 ump->um_balloc = ffs_balloc_ufs2; 741 } 742 ump->um_blkatoff = ffs_blkatoff; 743 ump->um_truncate = ffs_truncate; 744 ump->um_update = ffs_update; 745 ump->um_valloc = ffs_valloc; 746 ump->um_vfree = ffs_vfree; 747 ump->um_ifree = ffs_ifree; 748 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 749 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 750 if (fs->fs_sbsize < SBLOCKSIZE) 751 bp->b_flags |= B_INVAL | B_NOCACHE; 752 brelse(bp); 753 bp = NULL; 754 fs = ump->um_fs; 755 ffs_oldfscompat_read(fs, ump, sblockloc); 756 fs->fs_ronly = ronly; 757 size = fs->fs_cssize; 758 blks = howmany(size, fs->fs_fsize); 759 if (fs->fs_contigsumsize > 0) 760 size += fs->fs_ncg * sizeof(int32_t); 761 size += fs->fs_ncg * sizeof(u_int8_t); 762 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 763 fs->fs_csp = space; 764 for (i = 0; i < blks; i += fs->fs_frag) { 765 size = fs->fs_bsize; 766 if (i + fs->fs_frag > blks) 767 size = (blks - i) * fs->fs_fsize; 768 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 769 cred, &bp)) != 0) { 770 free(fs->fs_csp, M_UFSMNT); 771 goto out; 772 } 773 bcopy(bp->b_data, space, (u_int)size); 774 space = (char *)space + size; 775 brelse(bp); 776 bp = NULL; 777 } 778 if (fs->fs_contigsumsize > 0) { 779 fs->fs_maxcluster = lp = space; 780 for (i = 0; i < fs->fs_ncg; i++) 781 *lp++ = fs->fs_contigsumsize; 782 space = lp; 783 } 784 size = fs->fs_ncg * sizeof(u_int8_t); 785 fs->fs_contigdirs = (u_int8_t *)space; 786 bzero(fs->fs_contigdirs, size); 787 fs->fs_active = NULL; 788 mp->mnt_data = (qaddr_t)ump; 789 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 790 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 791 nmp = NULL; 792 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 793 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 794 if (nmp) 795 vfs_rel(nmp); 796 vfs_getnewfsid(mp); 797 } 798 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 799 MNT_ILOCK(mp); 800 mp->mnt_flag |= MNT_LOCAL; 801 MNT_IUNLOCK(mp); 802 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 803 #ifdef MAC 804 MNT_ILOCK(mp); 805 mp->mnt_flag |= MNT_MULTILABEL; 806 MNT_IUNLOCK(mp); 807 #else 808 printf( 809 "WARNING: %s: multilabel flag on fs but no MAC support\n", 810 mp->mnt_stat.f_mntonname); 811 #endif 812 } 813 if ((fs->fs_flags & FS_ACLS) != 0) { 814 #ifdef UFS_ACL 815 MNT_ILOCK(mp); 816 mp->mnt_flag |= MNT_ACLS; 817 MNT_IUNLOCK(mp); 818 #else 819 printf( 820 "WARNING: %s: ACLs flag on fs but no ACLs support\n", 821 mp->mnt_stat.f_mntonname); 822 #endif 823 } 824 ump->um_mountp = mp; 825 ump->um_dev = dev; 826 ump->um_devvp = devvp; 827 ump->um_nindir = fs->fs_nindir; 828 ump->um_bptrtodb = fs->fs_fsbtodb; 829 ump->um_seqinc = fs->fs_frag; 830 for (i = 0; i < MAXQUOTAS; i++) 831 ump->um_quotas[i] = NULLVP; 832 #ifdef UFS_EXTATTR 833 ufs_extattr_uepm_init(&ump->um_extattr); 834 #endif 835 /* 836 * Set FS local "last mounted on" information (NULL pad) 837 */ 838 bzero(fs->fs_fsmnt, MAXMNTLEN); 839 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 840 841 if( mp->mnt_flag & MNT_ROOTFS) { 842 /* 843 * Root mount; update timestamp in mount structure. 844 * this will be used by the common root mount code 845 * to update the system clock. 846 */ 847 mp->mnt_time = fs->fs_time; 848 } 849 850 if (ronly == 0) { 851 if ((fs->fs_flags & FS_DOSOFTDEP) && 852 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 853 free(fs->fs_csp, M_UFSMNT); 854 goto out; 855 } 856 if (fs->fs_snapinum[0] != 0) 857 ffs_snapshot_mount(mp); 858 fs->fs_fmod = 1; 859 fs->fs_clean = 0; 860 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 861 } 862 /* 863 * Initialize filesystem stat information in mount struct. 864 */ 865 #ifdef UFS_EXTATTR 866 #ifdef UFS_EXTATTR_AUTOSTART 867 /* 868 * 869 * Auto-starting does the following: 870 * - check for /.attribute in the fs, and extattr_start if so 871 * - for each file in .attribute, enable that file with 872 * an attribute of the same name. 873 * Not clear how to report errors -- probably eat them. 874 * This would all happen while the filesystem was busy/not 875 * available, so would effectively be "atomic". 876 */ 877 (void) ufs_extattr_autostart(mp, td); 878 #endif /* !UFS_EXTATTR_AUTOSTART */ 879 #endif /* !UFS_EXTATTR */ 880 #ifdef QUOTA 881 /* 882 * Our bufobj must require giant for snapshots when quotas are 883 * enabled. 884 */ 885 BO_LOCK(&devvp->v_bufobj); 886 devvp->v_bufobj.bo_flag |= BO_NEEDSGIANT; 887 BO_UNLOCK(&devvp->v_bufobj); 888 #else 889 MNT_ILOCK(mp); 890 mp->mnt_kern_flag |= MNTK_MPSAFE; 891 MNT_IUNLOCK(mp); 892 #endif 893 return (0); 894 out: 895 if (bp) 896 brelse(bp); 897 if (cp != NULL) { 898 DROP_GIANT(); 899 g_topology_lock(); 900 g_vfs_close(cp, td); 901 g_topology_unlock(); 902 PICKUP_GIANT(); 903 } 904 if (ump) { 905 mtx_destroy(UFS_MTX(ump)); 906 if (mp->mnt_gjprovider != NULL) { 907 free(mp->mnt_gjprovider, M_UFSMNT); 908 mp->mnt_gjprovider = NULL; 909 } 910 free(ump->um_fs, M_UFSMNT); 911 free(ump, M_UFSMNT); 912 mp->mnt_data = (qaddr_t)0; 913 } 914 return (error); 915 } 916 917 #include <sys/sysctl.h> 918 static int bigcgs = 0; 919 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 920 921 /* 922 * Sanity checks for loading old filesystem superblocks. 923 * See ffs_oldfscompat_write below for unwound actions. 924 * 925 * XXX - Parts get retired eventually. 926 * Unfortunately new bits get added. 927 */ 928 static void 929 ffs_oldfscompat_read(fs, ump, sblockloc) 930 struct fs *fs; 931 struct ufsmount *ump; 932 ufs2_daddr_t sblockloc; 933 { 934 off_t maxfilesize; 935 936 /* 937 * If not yet done, update fs_flags location and value of fs_sblockloc. 938 */ 939 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 940 fs->fs_flags = fs->fs_old_flags; 941 fs->fs_old_flags |= FS_FLAGS_UPDATED; 942 fs->fs_sblockloc = sblockloc; 943 } 944 /* 945 * If not yet done, update UFS1 superblock with new wider fields. 946 */ 947 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 948 fs->fs_maxbsize = fs->fs_bsize; 949 fs->fs_time = fs->fs_old_time; 950 fs->fs_size = fs->fs_old_size; 951 fs->fs_dsize = fs->fs_old_dsize; 952 fs->fs_csaddr = fs->fs_old_csaddr; 953 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 954 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 955 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 956 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 957 } 958 if (fs->fs_magic == FS_UFS1_MAGIC && 959 fs->fs_old_inodefmt < FS_44INODEFMT) { 960 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 961 fs->fs_qbmask = ~fs->fs_bmask; 962 fs->fs_qfmask = ~fs->fs_fmask; 963 } 964 if (fs->fs_magic == FS_UFS1_MAGIC) { 965 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 966 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 967 if (fs->fs_maxfilesize > maxfilesize) 968 fs->fs_maxfilesize = maxfilesize; 969 } 970 /* Compatibility for old filesystems */ 971 if (fs->fs_avgfilesize <= 0) 972 fs->fs_avgfilesize = AVFILESIZ; 973 if (fs->fs_avgfpdir <= 0) 974 fs->fs_avgfpdir = AFPDIR; 975 if (bigcgs) { 976 fs->fs_save_cgsize = fs->fs_cgsize; 977 fs->fs_cgsize = fs->fs_bsize; 978 } 979 } 980 981 /* 982 * Unwinding superblock updates for old filesystems. 983 * See ffs_oldfscompat_read above for details. 984 * 985 * XXX - Parts get retired eventually. 986 * Unfortunately new bits get added. 987 */ 988 static void 989 ffs_oldfscompat_write(fs, ump) 990 struct fs *fs; 991 struct ufsmount *ump; 992 { 993 994 /* 995 * Copy back UFS2 updated fields that UFS1 inspects. 996 */ 997 if (fs->fs_magic == FS_UFS1_MAGIC) { 998 fs->fs_old_time = fs->fs_time; 999 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1000 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1001 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1002 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1003 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1004 } 1005 if (bigcgs) { 1006 fs->fs_cgsize = fs->fs_save_cgsize; 1007 fs->fs_save_cgsize = 0; 1008 } 1009 } 1010 1011 /* 1012 * unmount system call 1013 */ 1014 static int 1015 ffs_unmount(mp, mntflags, td) 1016 struct mount *mp; 1017 int mntflags; 1018 struct thread *td; 1019 { 1020 struct ufsmount *ump = VFSTOUFS(mp); 1021 struct fs *fs; 1022 int error, flags; 1023 1024 flags = 0; 1025 if (mntflags & MNT_FORCE) { 1026 flags |= FORCECLOSE; 1027 } 1028 #ifdef UFS_EXTATTR 1029 if ((error = ufs_extattr_stop(mp, td))) { 1030 if (error != EOPNOTSUPP) 1031 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 1032 error); 1033 } else { 1034 ufs_extattr_uepm_destroy(&ump->um_extattr); 1035 } 1036 #endif 1037 if (mp->mnt_flag & MNT_SOFTDEP) { 1038 if ((error = softdep_flushfiles(mp, flags, td)) != 0) 1039 return (error); 1040 } else { 1041 if ((error = ffs_flushfiles(mp, flags, td)) != 0) 1042 return (error); 1043 } 1044 fs = ump->um_fs; 1045 UFS_LOCK(ump); 1046 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1047 printf("%s: unmount pending error: blocks %jd files %d\n", 1048 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1049 fs->fs_pendinginodes); 1050 fs->fs_pendingblocks = 0; 1051 fs->fs_pendinginodes = 0; 1052 } 1053 UFS_UNLOCK(ump); 1054 if (fs->fs_ronly == 0) { 1055 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1056 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1057 if (error) { 1058 fs->fs_clean = 0; 1059 return (error); 1060 } 1061 } 1062 DROP_GIANT(); 1063 g_topology_lock(); 1064 g_vfs_close(ump->um_cp, td); 1065 g_topology_unlock(); 1066 PICKUP_GIANT(); 1067 vrele(ump->um_devvp); 1068 mtx_destroy(UFS_MTX(ump)); 1069 if (mp->mnt_gjprovider != NULL) { 1070 free(mp->mnt_gjprovider, M_UFSMNT); 1071 mp->mnt_gjprovider = NULL; 1072 } 1073 free(fs->fs_csp, M_UFSMNT); 1074 free(fs, M_UFSMNT); 1075 free(ump, M_UFSMNT); 1076 mp->mnt_data = (qaddr_t)0; 1077 MNT_ILOCK(mp); 1078 mp->mnt_flag &= ~MNT_LOCAL; 1079 MNT_IUNLOCK(mp); 1080 return (error); 1081 } 1082 1083 /* 1084 * Flush out all the files in a filesystem. 1085 */ 1086 int 1087 ffs_flushfiles(mp, flags, td) 1088 struct mount *mp; 1089 int flags; 1090 struct thread *td; 1091 { 1092 struct ufsmount *ump; 1093 int error; 1094 1095 ump = VFSTOUFS(mp); 1096 #ifdef QUOTA 1097 if (mp->mnt_flag & MNT_QUOTA) { 1098 int i; 1099 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1100 if (error) 1101 return (error); 1102 for (i = 0; i < MAXQUOTAS; i++) { 1103 if (ump->um_quotas[i] == NULLVP) 1104 continue; 1105 quotaoff(td, mp, i); 1106 } 1107 /* 1108 * Here we fall through to vflush again to ensure 1109 * that we have gotten rid of all the system vnodes. 1110 */ 1111 } 1112 #endif 1113 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1114 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1115 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1116 return (error); 1117 ffs_snapshot_unmount(mp); 1118 flags |= FORCECLOSE; 1119 /* 1120 * Here we fall through to vflush again to ensure 1121 * that we have gotten rid of all the system vnodes. 1122 */ 1123 } 1124 /* 1125 * Flush all the files. 1126 */ 1127 if ((error = vflush(mp, 0, flags, td)) != 0) 1128 return (error); 1129 /* 1130 * Flush filesystem metadata. 1131 */ 1132 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td); 1133 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1134 VOP_UNLOCK(ump->um_devvp, 0, td); 1135 return (error); 1136 } 1137 1138 /* 1139 * Get filesystem statistics. 1140 */ 1141 static int 1142 ffs_statfs(mp, sbp, td) 1143 struct mount *mp; 1144 struct statfs *sbp; 1145 struct thread *td; 1146 { 1147 struct ufsmount *ump; 1148 struct fs *fs; 1149 1150 ump = VFSTOUFS(mp); 1151 fs = ump->um_fs; 1152 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1153 panic("ffs_statfs"); 1154 sbp->f_version = STATFS_VERSION; 1155 sbp->f_bsize = fs->fs_fsize; 1156 sbp->f_iosize = fs->fs_bsize; 1157 sbp->f_blocks = fs->fs_dsize; 1158 UFS_LOCK(ump); 1159 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1160 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1161 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1162 dbtofsb(fs, fs->fs_pendingblocks); 1163 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1164 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1165 UFS_UNLOCK(ump); 1166 sbp->f_namemax = NAME_MAX; 1167 return (0); 1168 } 1169 1170 /* 1171 * Go through the disk queues to initiate sandbagged IO; 1172 * go through the inodes to write those that have been modified; 1173 * initiate the writing of the super block if it has been modified. 1174 * 1175 * Note: we are always called with the filesystem marked `MPBUSY'. 1176 */ 1177 static int 1178 ffs_sync(mp, waitfor, td) 1179 struct mount *mp; 1180 int waitfor; 1181 struct thread *td; 1182 { 1183 struct vnode *mvp, *vp, *devvp; 1184 struct inode *ip; 1185 struct ufsmount *ump = VFSTOUFS(mp); 1186 struct fs *fs; 1187 int error, count, wait, lockreq, allerror = 0; 1188 int suspend; 1189 int suspended; 1190 int secondary_writes; 1191 int secondary_accwrites; 1192 int softdep_deps; 1193 int softdep_accdeps; 1194 struct bufobj *bo; 1195 1196 fs = ump->um_fs; 1197 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1198 printf("fs = %s\n", fs->fs_fsmnt); 1199 panic("ffs_sync: rofs mod"); 1200 } 1201 /* 1202 * Write back each (modified) inode. 1203 */ 1204 wait = 0; 1205 suspend = 0; 1206 suspended = 0; 1207 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1208 if (waitfor == MNT_SUSPEND) { 1209 suspend = 1; 1210 waitfor = MNT_WAIT; 1211 } 1212 if (waitfor == MNT_WAIT) { 1213 wait = 1; 1214 lockreq = LK_EXCLUSIVE; 1215 } 1216 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1217 MNT_ILOCK(mp); 1218 loop: 1219 /* Grab snapshot of secondary write counts */ 1220 secondary_writes = mp->mnt_secondary_writes; 1221 secondary_accwrites = mp->mnt_secondary_accwrites; 1222 1223 /* Grab snapshot of softdep dependency counts */ 1224 MNT_IUNLOCK(mp); 1225 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1226 MNT_ILOCK(mp); 1227 1228 MNT_VNODE_FOREACH(vp, mp, mvp) { 1229 /* 1230 * Depend on the mntvnode_slock to keep things stable enough 1231 * for a quick test. Since there might be hundreds of 1232 * thousands of vnodes, we cannot afford even a subroutine 1233 * call unless there's a good chance that we have work to do. 1234 */ 1235 VI_LOCK(vp); 1236 if (vp->v_iflag & VI_DOOMED) { 1237 VI_UNLOCK(vp); 1238 continue; 1239 } 1240 ip = VTOI(vp); 1241 if (vp->v_type == VNON || ((ip->i_flag & 1242 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1243 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1244 VI_UNLOCK(vp); 1245 continue; 1246 } 1247 MNT_IUNLOCK(mp); 1248 if ((error = vget(vp, lockreq, td)) != 0) { 1249 MNT_ILOCK(mp); 1250 if (error == ENOENT || error == ENOLCK) { 1251 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1252 goto loop; 1253 } 1254 continue; 1255 } 1256 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1257 allerror = error; 1258 vput(vp); 1259 MNT_ILOCK(mp); 1260 } 1261 MNT_IUNLOCK(mp); 1262 /* 1263 * Force stale filesystem control information to be flushed. 1264 */ 1265 if (waitfor == MNT_WAIT) { 1266 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1267 allerror = error; 1268 /* Flushed work items may create new vnodes to clean */ 1269 if (allerror == 0 && count) { 1270 MNT_ILOCK(mp); 1271 goto loop; 1272 } 1273 } 1274 #ifdef QUOTA 1275 qsync(mp); 1276 #endif 1277 devvp = ump->um_devvp; 1278 VI_LOCK(devvp); 1279 bo = &devvp->v_bufobj; 1280 if (waitfor != MNT_LAZY && 1281 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1282 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td); 1283 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1284 allerror = error; 1285 VOP_UNLOCK(devvp, 0, td); 1286 if (allerror == 0 && waitfor == MNT_WAIT) { 1287 MNT_ILOCK(mp); 1288 goto loop; 1289 } 1290 } else if (suspend != 0) { 1291 if (softdep_check_suspend(mp, 1292 devvp, 1293 softdep_deps, 1294 softdep_accdeps, 1295 secondary_writes, 1296 secondary_accwrites) != 0) 1297 goto loop; /* More work needed */ 1298 mtx_assert(MNT_MTX(mp), MA_OWNED); 1299 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1300 MNT_IUNLOCK(mp); 1301 suspended = 1; 1302 } else 1303 VI_UNLOCK(devvp); 1304 /* 1305 * Write back modified superblock. 1306 */ 1307 if (fs->fs_fmod != 0 && 1308 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1309 allerror = error; 1310 return (allerror); 1311 } 1312 1313 int 1314 ffs_vget(mp, ino, flags, vpp) 1315 struct mount *mp; 1316 ino_t ino; 1317 int flags; 1318 struct vnode **vpp; 1319 { 1320 struct fs *fs; 1321 struct inode *ip; 1322 struct ufsmount *ump; 1323 struct buf *bp; 1324 struct vnode *vp; 1325 struct cdev *dev; 1326 int error; 1327 1328 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1329 if (error || *vpp != NULL) 1330 return (error); 1331 1332 /* 1333 * We must promote to an exclusive lock for vnode creation. This 1334 * can happen if lookup is passed LOCKSHARED. 1335 */ 1336 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1337 flags &= ~LK_TYPE_MASK; 1338 flags |= LK_EXCLUSIVE; 1339 } 1340 1341 /* 1342 * We do not lock vnode creation as it is believed to be too 1343 * expensive for such rare case as simultaneous creation of vnode 1344 * for same ino by different processes. We just allow them to race 1345 * and check later to decide who wins. Let the race begin! 1346 */ 1347 1348 ump = VFSTOUFS(mp); 1349 dev = ump->um_dev; 1350 fs = ump->um_fs; 1351 1352 /* 1353 * If this MALLOC() is performed after the getnewvnode() 1354 * it might block, leaving a vnode with a NULL v_data to be 1355 * found by ffs_sync() if a sync happens to fire right then, 1356 * which will cause a panic because ffs_sync() blindly 1357 * dereferences vp->v_data (as well it should). 1358 */ 1359 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1360 1361 /* Allocate a new vnode/inode. */ 1362 if (fs->fs_magic == FS_UFS1_MAGIC) 1363 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1364 else 1365 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1366 if (error) { 1367 *vpp = NULL; 1368 uma_zfree(uma_inode, ip); 1369 return (error); 1370 } 1371 /* 1372 * FFS supports recursive and shared locking. 1373 */ 1374 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 1375 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 1376 vp->v_data = ip; 1377 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1378 ip->i_vnode = vp; 1379 ip->i_ump = ump; 1380 ip->i_fs = fs; 1381 ip->i_dev = dev; 1382 ip->i_number = ino; 1383 #ifdef QUOTA 1384 { 1385 int i; 1386 for (i = 0; i < MAXQUOTAS; i++) 1387 ip->i_dquot[i] = NODQUOT; 1388 } 1389 #endif 1390 1391 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1392 if (error || *vpp != NULL) 1393 return (error); 1394 1395 /* Read in the disk contents for the inode, copy into the inode. */ 1396 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1397 (int)fs->fs_bsize, NOCRED, &bp); 1398 if (error) { 1399 /* 1400 * The inode does not contain anything useful, so it would 1401 * be misleading to leave it on its hash chain. With mode 1402 * still zero, it will be unlinked and returned to the free 1403 * list by vput(). 1404 */ 1405 brelse(bp); 1406 vput(vp); 1407 *vpp = NULL; 1408 return (error); 1409 } 1410 if (ip->i_ump->um_fstype == UFS1) 1411 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1412 else 1413 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1414 ffs_load_inode(bp, ip, fs, ino); 1415 if (DOINGSOFTDEP(vp)) 1416 softdep_load_inodeblock(ip); 1417 else 1418 ip->i_effnlink = ip->i_nlink; 1419 bqrelse(bp); 1420 1421 /* 1422 * Initialize the vnode from the inode, check for aliases. 1423 * Note that the underlying vnode may have changed. 1424 */ 1425 if (ip->i_ump->um_fstype == UFS1) 1426 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1427 else 1428 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1429 if (error) { 1430 vput(vp); 1431 *vpp = NULL; 1432 return (error); 1433 } 1434 1435 /* 1436 * Finish inode initialization. 1437 */ 1438 1439 /* 1440 * Set up a generation number for this inode if it does not 1441 * already have one. This should only happen on old filesystems. 1442 */ 1443 if (ip->i_gen == 0) { 1444 ip->i_gen = arc4random() / 2 + 1; 1445 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1446 ip->i_flag |= IN_MODIFIED; 1447 DIP_SET(ip, i_gen, ip->i_gen); 1448 } 1449 } 1450 /* 1451 * Ensure that uid and gid are correct. This is a temporary 1452 * fix until fsck has been changed to do the update. 1453 */ 1454 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1455 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1456 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1457 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1458 } /* XXX */ 1459 1460 #ifdef MAC 1461 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1462 /* 1463 * If this vnode is already allocated, and we're running 1464 * multi-label, attempt to perform a label association 1465 * from the extended attributes on the inode. 1466 */ 1467 error = mac_associate_vnode_extattr(mp, vp); 1468 if (error) { 1469 /* ufs_inactive will release ip->i_devvp ref. */ 1470 vput(vp); 1471 *vpp = NULL; 1472 return (error); 1473 } 1474 } 1475 #endif 1476 1477 *vpp = vp; 1478 return (0); 1479 } 1480 1481 /* 1482 * File handle to vnode 1483 * 1484 * Have to be really careful about stale file handles: 1485 * - check that the inode number is valid 1486 * - call ffs_vget() to get the locked inode 1487 * - check for an unallocated inode (i_mode == 0) 1488 * - check that the given client host has export rights and return 1489 * those rights via. exflagsp and credanonp 1490 */ 1491 static int 1492 ffs_fhtovp(mp, fhp, vpp) 1493 struct mount *mp; 1494 struct fid *fhp; 1495 struct vnode **vpp; 1496 { 1497 struct ufid *ufhp; 1498 struct fs *fs; 1499 1500 ufhp = (struct ufid *)fhp; 1501 fs = VFSTOUFS(mp)->um_fs; 1502 if (ufhp->ufid_ino < ROOTINO || 1503 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1504 return (ESTALE); 1505 return (ufs_fhtovp(mp, ufhp, vpp)); 1506 } 1507 1508 /* 1509 * Vnode pointer to File handle 1510 */ 1511 /* ARGSUSED */ 1512 static int 1513 ffs_vptofh(vp, fhp) 1514 struct vnode *vp; 1515 struct fid *fhp; 1516 { 1517 struct inode *ip; 1518 struct ufid *ufhp; 1519 1520 ip = VTOI(vp); 1521 ufhp = (struct ufid *)fhp; 1522 ufhp->ufid_len = sizeof(struct ufid); 1523 ufhp->ufid_ino = ip->i_number; 1524 ufhp->ufid_gen = ip->i_gen; 1525 return (0); 1526 } 1527 1528 /* 1529 * Initialize the filesystem. 1530 */ 1531 static int 1532 ffs_init(vfsp) 1533 struct vfsconf *vfsp; 1534 { 1535 1536 softdep_initialize(); 1537 return (ufs_init(vfsp)); 1538 } 1539 1540 /* 1541 * Undo the work of ffs_init(). 1542 */ 1543 static int 1544 ffs_uninit(vfsp) 1545 struct vfsconf *vfsp; 1546 { 1547 int ret; 1548 1549 ret = ufs_uninit(vfsp); 1550 softdep_uninitialize(); 1551 return (ret); 1552 } 1553 1554 /* 1555 * Write a superblock and associated information back to disk. 1556 */ 1557 int 1558 ffs_sbupdate(mp, waitfor, suspended) 1559 struct ufsmount *mp; 1560 int waitfor; 1561 int suspended; 1562 { 1563 struct fs *fs = mp->um_fs; 1564 struct buf *sbbp; 1565 struct buf *bp; 1566 int blks; 1567 void *space; 1568 int i, size, error, allerror = 0; 1569 1570 if (fs->fs_ronly == 1 && 1571 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1572 (MNT_RDONLY | MNT_UPDATE)) 1573 panic("ffs_sbupdate: write read-only filesystem"); 1574 /* 1575 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1576 */ 1577 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1578 0, 0, 0); 1579 /* 1580 * First write back the summary information. 1581 */ 1582 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1583 space = fs->fs_csp; 1584 for (i = 0; i < blks; i += fs->fs_frag) { 1585 size = fs->fs_bsize; 1586 if (i + fs->fs_frag > blks) 1587 size = (blks - i) * fs->fs_fsize; 1588 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1589 size, 0, 0, 0); 1590 bcopy(space, bp->b_data, (u_int)size); 1591 space = (char *)space + size; 1592 if (suspended) 1593 bp->b_flags |= B_VALIDSUSPWRT; 1594 if (waitfor != MNT_WAIT) 1595 bawrite(bp); 1596 else if ((error = bwrite(bp)) != 0) 1597 allerror = error; 1598 } 1599 /* 1600 * Now write back the superblock itself. If any errors occurred 1601 * up to this point, then fail so that the superblock avoids 1602 * being written out as clean. 1603 */ 1604 if (allerror) { 1605 brelse(sbbp); 1606 return (allerror); 1607 } 1608 bp = sbbp; 1609 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1610 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1611 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1612 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1613 fs->fs_sblockloc = SBLOCK_UFS1; 1614 } 1615 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1616 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1617 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1618 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1619 fs->fs_sblockloc = SBLOCK_UFS2; 1620 } 1621 fs->fs_fmod = 0; 1622 fs->fs_time = time_second; 1623 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1624 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1625 if (suspended) 1626 bp->b_flags |= B_VALIDSUSPWRT; 1627 if (waitfor != MNT_WAIT) 1628 bawrite(bp); 1629 else if ((error = bwrite(bp)) != 0) 1630 allerror = error; 1631 return (allerror); 1632 } 1633 1634 static int 1635 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1636 int attrnamespace, const char *attrname, struct thread *td) 1637 { 1638 1639 #ifdef UFS_EXTATTR 1640 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1641 attrname, td)); 1642 #else 1643 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1644 attrname, td)); 1645 #endif 1646 } 1647 1648 static void 1649 ffs_ifree(struct ufsmount *ump, struct inode *ip) 1650 { 1651 1652 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1653 uma_zfree(uma_ufs1, ip->i_din1); 1654 else if (ip->i_din2 != NULL) 1655 uma_zfree(uma_ufs2, ip->i_din2); 1656 uma_zfree(uma_inode, ip); 1657 } 1658 1659 static int dobkgrdwrite = 1; 1660 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1661 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1662 1663 /* 1664 * Complete a background write started from bwrite. 1665 */ 1666 static void 1667 ffs_backgroundwritedone(struct buf *bp) 1668 { 1669 struct bufobj *bufobj; 1670 struct buf *origbp; 1671 1672 /* 1673 * Find the original buffer that we are writing. 1674 */ 1675 bufobj = bp->b_bufobj; 1676 BO_LOCK(bufobj); 1677 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1678 panic("backgroundwritedone: lost buffer"); 1679 /* Grab an extra reference to be dropped by the bufdone() below. */ 1680 bufobj_wrefl(bufobj); 1681 BO_UNLOCK(bufobj); 1682 /* 1683 * Process dependencies then return any unfinished ones. 1684 */ 1685 if (LIST_FIRST(&bp->b_dep) != NULL) 1686 buf_complete(bp); 1687 #ifdef SOFTUPDATES 1688 if (LIST_FIRST(&bp->b_dep) != NULL) 1689 softdep_move_dependencies(bp, origbp); 1690 #endif 1691 /* 1692 * This buffer is marked B_NOCACHE so when it is released 1693 * by biodone it will be tossed. 1694 */ 1695 bp->b_flags |= B_NOCACHE; 1696 bp->b_flags &= ~B_CACHE; 1697 bufdone(bp); 1698 BO_LOCK(bufobj); 1699 /* 1700 * Clear the BV_BKGRDINPROG flag in the original buffer 1701 * and awaken it if it is waiting for the write to complete. 1702 * If BV_BKGRDINPROG is not set in the original buffer it must 1703 * have been released and re-instantiated - which is not legal. 1704 */ 1705 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1706 ("backgroundwritedone: lost buffer2")); 1707 origbp->b_vflags &= ~BV_BKGRDINPROG; 1708 if (origbp->b_vflags & BV_BKGRDWAIT) { 1709 origbp->b_vflags &= ~BV_BKGRDWAIT; 1710 wakeup(&origbp->b_xflags); 1711 } 1712 BO_UNLOCK(bufobj); 1713 } 1714 1715 1716 /* 1717 * Write, release buffer on completion. (Done by iodone 1718 * if async). Do not bother writing anything if the buffer 1719 * is invalid. 1720 * 1721 * Note that we set B_CACHE here, indicating that buffer is 1722 * fully valid and thus cacheable. This is true even of NFS 1723 * now so we set it generally. This could be set either here 1724 * or in biodone() since the I/O is synchronous. We put it 1725 * here. 1726 */ 1727 static int 1728 ffs_bufwrite(struct buf *bp) 1729 { 1730 int oldflags, s; 1731 struct buf *newbp; 1732 1733 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1734 if (bp->b_flags & B_INVAL) { 1735 brelse(bp); 1736 return (0); 1737 } 1738 1739 oldflags = bp->b_flags; 1740 1741 if (BUF_REFCNT(bp) == 0) 1742 panic("bufwrite: buffer is not busy???"); 1743 s = splbio(); 1744 /* 1745 * If a background write is already in progress, delay 1746 * writing this block if it is asynchronous. Otherwise 1747 * wait for the background write to complete. 1748 */ 1749 BO_LOCK(bp->b_bufobj); 1750 if (bp->b_vflags & BV_BKGRDINPROG) { 1751 if (bp->b_flags & B_ASYNC) { 1752 BO_UNLOCK(bp->b_bufobj); 1753 splx(s); 1754 bdwrite(bp); 1755 return (0); 1756 } 1757 bp->b_vflags |= BV_BKGRDWAIT; 1758 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1759 if (bp->b_vflags & BV_BKGRDINPROG) 1760 panic("bufwrite: still writing"); 1761 } 1762 BO_UNLOCK(bp->b_bufobj); 1763 1764 /* Mark the buffer clean */ 1765 bundirty(bp); 1766 1767 /* 1768 * If this buffer is marked for background writing and we 1769 * do not have to wait for it, make a copy and write the 1770 * copy so as to leave this buffer ready for further use. 1771 * 1772 * This optimization eats a lot of memory. If we have a page 1773 * or buffer shortfall we can't do it. 1774 */ 1775 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1776 (bp->b_flags & B_ASYNC) && 1777 !vm_page_count_severe() && 1778 !buf_dirty_count_severe()) { 1779 KASSERT(bp->b_iodone == NULL, 1780 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1781 1782 /* get a new block */ 1783 newbp = geteblk(bp->b_bufsize); 1784 1785 /* 1786 * set it to be identical to the old block. We have to 1787 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1788 * to avoid confusing the splay tree and gbincore(). 1789 */ 1790 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1791 newbp->b_lblkno = bp->b_lblkno; 1792 newbp->b_xflags |= BX_BKGRDMARKER; 1793 BO_LOCK(bp->b_bufobj); 1794 bp->b_vflags |= BV_BKGRDINPROG; 1795 bgetvp(bp->b_vp, newbp); 1796 BO_UNLOCK(bp->b_bufobj); 1797 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1798 newbp->b_blkno = bp->b_blkno; 1799 newbp->b_offset = bp->b_offset; 1800 newbp->b_iodone = ffs_backgroundwritedone; 1801 newbp->b_flags |= B_ASYNC; 1802 newbp->b_flags &= ~B_INVAL; 1803 1804 #ifdef SOFTUPDATES 1805 /* move over the dependencies */ 1806 if (LIST_FIRST(&bp->b_dep) != NULL) 1807 softdep_move_dependencies(bp, newbp); 1808 #endif 1809 1810 /* 1811 * Initiate write on the copy, release the original to 1812 * the B_LOCKED queue so that it cannot go away until 1813 * the background write completes. If not locked it could go 1814 * away and then be reconstituted while it was being written. 1815 * If the reconstituted buffer were written, we could end up 1816 * with two background copies being written at the same time. 1817 */ 1818 bqrelse(bp); 1819 bp = newbp; 1820 } 1821 1822 /* Let the normal bufwrite do the rest for us */ 1823 return (bufwrite(bp)); 1824 } 1825 1826 1827 static void 1828 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1829 { 1830 struct vnode *vp; 1831 int error; 1832 struct buf *tbp; 1833 1834 vp = bo->__bo_vnode; 1835 if (bp->b_iocmd == BIO_WRITE) { 1836 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1837 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1838 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1839 panic("ffs_geom_strategy: bad I/O"); 1840 bp->b_flags &= ~B_VALIDSUSPWRT; 1841 if ((vp->v_vflag & VV_COPYONWRITE) && 1842 vp->v_rdev->si_snapdata != NULL) { 1843 if ((bp->b_flags & B_CLUSTER) != 0) { 1844 runningbufwakeup(bp); 1845 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1846 b_cluster.cluster_entry) { 1847 error = ffs_copyonwrite(vp, tbp); 1848 if (error != 0 && 1849 error != EOPNOTSUPP) { 1850 bp->b_error = error; 1851 bp->b_ioflags |= BIO_ERROR; 1852 bufdone(bp); 1853 return; 1854 } 1855 } 1856 bp->b_runningbufspace = bp->b_bufsize; 1857 atomic_add_int(&runningbufspace, 1858 bp->b_runningbufspace); 1859 } else { 1860 error = ffs_copyonwrite(vp, bp); 1861 if (error != 0 && error != EOPNOTSUPP) { 1862 bp->b_error = error; 1863 bp->b_ioflags |= BIO_ERROR; 1864 bufdone(bp); 1865 return; 1866 } 1867 } 1868 } 1869 #ifdef SOFTUPDATES 1870 if ((bp->b_flags & B_CLUSTER) != 0) { 1871 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1872 b_cluster.cluster_entry) { 1873 if (LIST_FIRST(&tbp->b_dep) != NULL) 1874 buf_start(tbp); 1875 } 1876 } else { 1877 if (LIST_FIRST(&bp->b_dep) != NULL) 1878 buf_start(bp); 1879 } 1880 1881 #endif 1882 } 1883 g_vfs_strategy(bo, bp); 1884 } 1885