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