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 #include "opt_ddb.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/priv.h> 45 #include <sys/proc.h> 46 #include <sys/kernel.h> 47 #include <sys/vnode.h> 48 #include <sys/mount.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/fcntl.h> 53 #include <sys/malloc.h> 54 #include <sys/mutex.h> 55 56 #include <security/mac/mac_framework.h> 57 58 #include <ufs/ufs/extattr.h> 59 #include <ufs/ufs/gjournal.h> 60 #include <ufs/ufs/quota.h> 61 #include <ufs/ufs/ufsmount.h> 62 #include <ufs/ufs/inode.h> 63 #include <ufs/ufs/ufs_extern.h> 64 65 #include <ufs/ffs/fs.h> 66 #include <ufs/ffs/ffs_extern.h> 67 68 #include <vm/vm.h> 69 #include <vm/uma.h> 70 #include <vm/vm_page.h> 71 72 #include <geom/geom.h> 73 #include <geom/geom_vfs.h> 74 75 #include <ddb/ddb.h> 76 77 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 78 79 static int ffs_reload(struct mount *, struct thread *); 80 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 81 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 82 ufs2_daddr_t); 83 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *); 84 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 85 static vfs_init_t ffs_init; 86 static vfs_uninit_t ffs_uninit; 87 static vfs_extattrctl_t ffs_extattrctl; 88 static vfs_cmount_t ffs_cmount; 89 static vfs_unmount_t ffs_unmount; 90 static vfs_mount_t ffs_mount; 91 static vfs_statfs_t ffs_statfs; 92 static vfs_fhtovp_t ffs_fhtovp; 93 static vfs_sync_t ffs_sync; 94 95 static struct vfsops ufs_vfsops = { 96 .vfs_extattrctl = ffs_extattrctl, 97 .vfs_fhtovp = ffs_fhtovp, 98 .vfs_init = ffs_init, 99 .vfs_mount = ffs_mount, 100 .vfs_cmount = ffs_cmount, 101 .vfs_quotactl = ufs_quotactl, 102 .vfs_root = ufs_root, 103 .vfs_statfs = ffs_statfs, 104 .vfs_sync = ffs_sync, 105 .vfs_uninit = ffs_uninit, 106 .vfs_unmount = ffs_unmount, 107 .vfs_vget = ffs_vget, 108 .vfs_susp_clean = process_deferred_inactive, 109 }; 110 111 VFS_SET(ufs_vfsops, ufs, 0); 112 MODULE_VERSION(ufs, 1); 113 114 static b_strategy_t ffs_geom_strategy; 115 static b_write_t ffs_bufwrite; 116 117 static struct buf_ops ffs_ops = { 118 .bop_name = "FFS", 119 .bop_write = ffs_bufwrite, 120 .bop_strategy = ffs_geom_strategy, 121 .bop_sync = bufsync, 122 #ifdef NO_FFS_SNAPSHOT 123 .bop_bdflush = bufbdflush, 124 #else 125 .bop_bdflush = ffs_bdflush, 126 #endif 127 }; 128 129 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 130 "noclusterw", "noexec", "export", "force", "from", "multilabel", 131 "snapshot", "nosuid", "suiddir", "nosymfollow", "sync", 132 "union", NULL }; 133 134 static int 135 ffs_mount(struct mount *mp, struct thread *td) 136 { 137 struct vnode *devvp; 138 struct ufsmount *ump = 0; 139 struct fs *fs; 140 int error, flags; 141 u_int mntorflags, mntandnotflags; 142 accmode_t accmode; 143 struct nameidata ndp; 144 char *fspec; 145 146 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 147 return (EINVAL); 148 if (uma_inode == NULL) { 149 uma_inode = uma_zcreate("FFS inode", 150 sizeof(struct inode), NULL, NULL, NULL, NULL, 151 UMA_ALIGN_PTR, 0); 152 uma_ufs1 = uma_zcreate("FFS1 dinode", 153 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 154 UMA_ALIGN_PTR, 0); 155 uma_ufs2 = uma_zcreate("FFS2 dinode", 156 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 157 UMA_ALIGN_PTR, 0); 158 } 159 160 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 161 if (error) 162 return (error); 163 164 mntorflags = 0; 165 mntandnotflags = 0; 166 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 167 mntorflags |= MNT_ACLS; 168 169 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 170 mntorflags |= MNT_SNAPSHOT; 171 /* 172 * Once we have set the MNT_SNAPSHOT flag, do not 173 * persist "snapshot" in the options list. 174 */ 175 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 176 vfs_deleteopt(mp->mnt_opt, "snapshot"); 177 } 178 179 MNT_ILOCK(mp); 180 mp->mnt_flag = (mp->mnt_flag | mntorflags) & ~mntandnotflags; 181 MNT_IUNLOCK(mp); 182 /* 183 * If updating, check whether changing from read-only to 184 * read/write; if there is no device name, that's all we do. 185 */ 186 if (mp->mnt_flag & MNT_UPDATE) { 187 ump = VFSTOUFS(mp); 188 fs = ump->um_fs; 189 devvp = ump->um_devvp; 190 if (fs->fs_ronly == 0 && 191 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 192 /* 193 * Flush any dirty data and suspend filesystem. 194 */ 195 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 196 return (error); 197 for (;;) { 198 vn_finished_write(mp); 199 if ((error = vfs_write_suspend(mp)) != 0) 200 return (error); 201 MNT_ILOCK(mp); 202 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 203 /* 204 * Allow the secondary writes 205 * to proceed. 206 */ 207 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 208 MNTK_SUSPEND2); 209 wakeup(&mp->mnt_flag); 210 MNT_IUNLOCK(mp); 211 /* 212 * Allow the curthread to 213 * ignore the suspension to 214 * synchronize on-disk state. 215 */ 216 curthread->td_pflags |= TDP_IGNSUSP; 217 break; 218 } 219 MNT_IUNLOCK(mp); 220 vn_start_write(NULL, &mp, V_WAIT); 221 } 222 /* 223 * Check for and optionally get rid of files open 224 * for writing. 225 */ 226 flags = WRITECLOSE; 227 if (mp->mnt_flag & MNT_FORCE) 228 flags |= FORCECLOSE; 229 if (mp->mnt_flag & MNT_SOFTDEP) { 230 error = softdep_flushfiles(mp, flags, td); 231 } else { 232 error = ffs_flushfiles(mp, flags, td); 233 } 234 if (error) { 235 vfs_write_resume(mp); 236 return (error); 237 } 238 if (fs->fs_pendingblocks != 0 || 239 fs->fs_pendinginodes != 0) { 240 printf("%s: %s: blocks %jd files %d\n", 241 fs->fs_fsmnt, "update error", 242 (intmax_t)fs->fs_pendingblocks, 243 fs->fs_pendinginodes); 244 fs->fs_pendingblocks = 0; 245 fs->fs_pendinginodes = 0; 246 } 247 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 248 fs->fs_clean = 1; 249 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 250 fs->fs_ronly = 0; 251 fs->fs_clean = 0; 252 vfs_write_resume(mp); 253 return (error); 254 } 255 DROP_GIANT(); 256 g_topology_lock(); 257 g_access(ump->um_cp, 0, -1, 0); 258 g_topology_unlock(); 259 PICKUP_GIANT(); 260 fs->fs_ronly = 1; 261 MNT_ILOCK(mp); 262 mp->mnt_flag |= MNT_RDONLY; 263 MNT_IUNLOCK(mp); 264 /* 265 * Allow the writers to note that filesystem 266 * is ro now. 267 */ 268 vfs_write_resume(mp); 269 } 270 if ((mp->mnt_flag & MNT_RELOAD) && 271 (error = ffs_reload(mp, td)) != 0) 272 return (error); 273 if (fs->fs_ronly && 274 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 275 /* 276 * If upgrade to read-write by non-root, then verify 277 * that user has necessary permissions on the device. 278 */ 279 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 280 error = VOP_ACCESS(devvp, VREAD | VWRITE, 281 td->td_ucred, td); 282 if (error) 283 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 284 if (error) { 285 VOP_UNLOCK(devvp, 0); 286 return (error); 287 } 288 VOP_UNLOCK(devvp, 0); 289 fs->fs_flags &= ~FS_UNCLEAN; 290 if (fs->fs_clean == 0) { 291 fs->fs_flags |= FS_UNCLEAN; 292 if ((mp->mnt_flag & MNT_FORCE) || 293 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 294 (fs->fs_flags & FS_DOSOFTDEP))) { 295 printf("WARNING: %s was not %s\n", 296 fs->fs_fsmnt, "properly dismounted"); 297 } else { 298 printf( 299 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 300 fs->fs_fsmnt); 301 return (EPERM); 302 } 303 } 304 DROP_GIANT(); 305 g_topology_lock(); 306 /* 307 * If we're the root device, we may not have an E count 308 * yet, get it now. 309 */ 310 if (ump->um_cp->ace == 0) 311 error = g_access(ump->um_cp, 0, 1, 1); 312 else 313 error = g_access(ump->um_cp, 0, 1, 0); 314 g_topology_unlock(); 315 PICKUP_GIANT(); 316 if (error) 317 return (error); 318 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 319 return (error); 320 fs->fs_ronly = 0; 321 MNT_ILOCK(mp); 322 mp->mnt_flag &= ~MNT_RDONLY; 323 MNT_IUNLOCK(mp); 324 fs->fs_clean = 0; 325 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 326 vn_finished_write(mp); 327 return (error); 328 } 329 /* check to see if we need to start softdep */ 330 if ((fs->fs_flags & FS_DOSOFTDEP) && 331 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 332 vn_finished_write(mp); 333 return (error); 334 } 335 if (fs->fs_snapinum[0] != 0) 336 ffs_snapshot_mount(mp); 337 vn_finished_write(mp); 338 } 339 /* 340 * Soft updates is incompatible with "async", 341 * so if we are doing softupdates stop the user 342 * from setting the async flag in an update. 343 * Softdep_mount() clears it in an initial mount 344 * or ro->rw remount. 345 */ 346 if (mp->mnt_flag & MNT_SOFTDEP) { 347 /* XXX: Reset too late ? */ 348 MNT_ILOCK(mp); 349 mp->mnt_flag &= ~MNT_ASYNC; 350 MNT_IUNLOCK(mp); 351 } 352 /* 353 * Keep MNT_ACLS flag if it is stored in superblock. 354 */ 355 if ((fs->fs_flags & FS_ACLS) != 0) { 356 /* XXX: Set too late ? */ 357 MNT_ILOCK(mp); 358 mp->mnt_flag |= MNT_ACLS; 359 MNT_IUNLOCK(mp); 360 } 361 362 /* 363 * If this is a snapshot request, take the snapshot. 364 */ 365 if (mp->mnt_flag & MNT_SNAPSHOT) 366 return (ffs_snapshot(mp, fspec)); 367 } 368 369 /* 370 * Not an update, or updating the name: look up the name 371 * and verify that it refers to a sensible disk device. 372 */ 373 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 374 if ((error = namei(&ndp)) != 0) 375 return (error); 376 NDFREE(&ndp, NDF_ONLY_PNBUF); 377 devvp = ndp.ni_vp; 378 if (!vn_isdisk(devvp, &error)) { 379 vput(devvp); 380 return (error); 381 } 382 383 /* 384 * If mount by non-root, then verify that user has necessary 385 * permissions on the device. 386 */ 387 accmode = VREAD; 388 if ((mp->mnt_flag & MNT_RDONLY) == 0) 389 accmode |= VWRITE; 390 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 391 if (error) 392 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 393 if (error) { 394 vput(devvp); 395 return (error); 396 } 397 398 if (mp->mnt_flag & MNT_UPDATE) { 399 /* 400 * Update only 401 * 402 * If it's not the same vnode, or at least the same device 403 * then it's not correct. 404 */ 405 406 if (devvp->v_rdev != ump->um_devvp->v_rdev) 407 error = EINVAL; /* needs translation */ 408 vput(devvp); 409 if (error) 410 return (error); 411 } else { 412 /* 413 * New mount 414 * 415 * We need the name for the mount point (also used for 416 * "last mounted on") copied in. If an error occurs, 417 * the mount point is discarded by the upper level code. 418 * Note that vfs_mount() populates f_mntonname for us. 419 */ 420 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 421 vrele(devvp); 422 return (error); 423 } 424 } 425 vfs_mountedfrom(mp, fspec); 426 return (0); 427 } 428 429 /* 430 * Compatibility with old mount system call. 431 */ 432 433 static int 434 ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td) 435 { 436 struct ufs_args args; 437 int error; 438 439 if (data == NULL) 440 return (EINVAL); 441 error = copyin(data, &args, sizeof args); 442 if (error) 443 return (error); 444 445 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 446 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 447 error = kernel_mount(ma, flags); 448 449 return (error); 450 } 451 452 /* 453 * Reload all incore data for a filesystem (used after running fsck on 454 * the root filesystem and finding things to fix). The filesystem must 455 * be mounted read-only. 456 * 457 * Things to do to update the mount: 458 * 1) invalidate all cached meta-data. 459 * 2) re-read superblock from disk. 460 * 3) re-read summary information from disk. 461 * 4) invalidate all inactive vnodes. 462 * 5) invalidate all cached file data. 463 * 6) re-read inode data for all active vnodes. 464 */ 465 static int 466 ffs_reload(struct mount *mp, struct thread *td) 467 { 468 struct vnode *vp, *mvp, *devvp; 469 struct inode *ip; 470 void *space; 471 struct buf *bp; 472 struct fs *fs, *newfs; 473 struct ufsmount *ump; 474 ufs2_daddr_t sblockloc; 475 int i, blks, size, error; 476 int32_t *lp; 477 478 if ((mp->mnt_flag & MNT_RDONLY) == 0) 479 return (EINVAL); 480 ump = VFSTOUFS(mp); 481 /* 482 * Step 1: invalidate all cached meta-data. 483 */ 484 devvp = VFSTOUFS(mp)->um_devvp; 485 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 486 if (vinvalbuf(devvp, 0, 0, 0) != 0) 487 panic("ffs_reload: dirty1"); 488 VOP_UNLOCK(devvp, 0); 489 490 /* 491 * Step 2: re-read superblock from disk. 492 */ 493 fs = VFSTOUFS(mp)->um_fs; 494 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 495 NOCRED, &bp)) != 0) 496 return (error); 497 newfs = (struct fs *)bp->b_data; 498 if ((newfs->fs_magic != FS_UFS1_MAGIC && 499 newfs->fs_magic != FS_UFS2_MAGIC) || 500 newfs->fs_bsize > MAXBSIZE || 501 newfs->fs_bsize < sizeof(struct fs)) { 502 brelse(bp); 503 return (EIO); /* XXX needs translation */ 504 } 505 /* 506 * Copy pointer fields back into superblock before copying in XXX 507 * new superblock. These should really be in the ufsmount. XXX 508 * Note that important parameters (eg fs_ncg) are unchanged. 509 */ 510 newfs->fs_csp = fs->fs_csp; 511 newfs->fs_maxcluster = fs->fs_maxcluster; 512 newfs->fs_contigdirs = fs->fs_contigdirs; 513 newfs->fs_active = fs->fs_active; 514 /* The file system is still read-only. */ 515 newfs->fs_ronly = 1; 516 sblockloc = fs->fs_sblockloc; 517 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 518 brelse(bp); 519 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 520 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 521 UFS_LOCK(ump); 522 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 523 printf("%s: reload pending error: blocks %jd files %d\n", 524 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 525 fs->fs_pendinginodes); 526 fs->fs_pendingblocks = 0; 527 fs->fs_pendinginodes = 0; 528 } 529 UFS_UNLOCK(ump); 530 531 /* 532 * Step 3: re-read summary information from disk. 533 */ 534 blks = howmany(fs->fs_cssize, fs->fs_fsize); 535 space = fs->fs_csp; 536 for (i = 0; i < blks; i += fs->fs_frag) { 537 size = fs->fs_bsize; 538 if (i + fs->fs_frag > blks) 539 size = (blks - i) * fs->fs_fsize; 540 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 541 NOCRED, &bp); 542 if (error) 543 return (error); 544 bcopy(bp->b_data, space, (u_int)size); 545 space = (char *)space + size; 546 brelse(bp); 547 } 548 /* 549 * We no longer know anything about clusters per cylinder group. 550 */ 551 if (fs->fs_contigsumsize > 0) { 552 lp = fs->fs_maxcluster; 553 for (i = 0; i < fs->fs_ncg; i++) 554 *lp++ = fs->fs_contigsumsize; 555 } 556 557 loop: 558 MNT_ILOCK(mp); 559 MNT_VNODE_FOREACH(vp, mp, mvp) { 560 VI_LOCK(vp); 561 if (vp->v_iflag & VI_DOOMED) { 562 VI_UNLOCK(vp); 563 continue; 564 } 565 MNT_IUNLOCK(mp); 566 /* 567 * Step 4: invalidate all cached file data. 568 */ 569 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 570 MNT_VNODE_FOREACH_ABORT(mp, mvp); 571 goto loop; 572 } 573 if (vinvalbuf(vp, 0, 0, 0)) 574 panic("ffs_reload: dirty2"); 575 /* 576 * Step 5: re-read inode data for all active vnodes. 577 */ 578 ip = VTOI(vp); 579 error = 580 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 581 (int)fs->fs_bsize, NOCRED, &bp); 582 if (error) { 583 VOP_UNLOCK(vp, 0); 584 vrele(vp); 585 MNT_VNODE_FOREACH_ABORT(mp, mvp); 586 return (error); 587 } 588 ffs_load_inode(bp, ip, fs, ip->i_number); 589 ip->i_effnlink = ip->i_nlink; 590 brelse(bp); 591 VOP_UNLOCK(vp, 0); 592 vrele(vp); 593 MNT_ILOCK(mp); 594 } 595 MNT_IUNLOCK(mp); 596 return (0); 597 } 598 599 /* 600 * Possible superblock locations ordered from most to least likely. 601 */ 602 static int sblock_try[] = SBLOCKSEARCH; 603 604 /* 605 * Common code for mount and mountroot 606 */ 607 static int 608 ffs_mountfs(devvp, mp, td) 609 struct vnode *devvp; 610 struct mount *mp; 611 struct thread *td; 612 { 613 struct ufsmount *ump; 614 struct buf *bp; 615 struct fs *fs; 616 struct cdev *dev; 617 void *space; 618 ufs2_daddr_t sblockloc; 619 int error, i, blks, size, ronly; 620 int32_t *lp; 621 struct ucred *cred; 622 struct g_consumer *cp; 623 struct mount *nmp; 624 625 bp = NULL; 626 ump = NULL; 627 cred = td ? td->td_ucred : NOCRED; 628 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 629 630 dev = devvp->v_rdev; 631 dev_ref(dev); 632 DROP_GIANT(); 633 g_topology_lock(); 634 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 635 636 /* 637 * If we are a root mount, drop the E flag so fsck can do its magic. 638 * We will pick it up again when we remount R/W. 639 */ 640 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 641 error = g_access(cp, 0, 0, -1); 642 g_topology_unlock(); 643 PICKUP_GIANT(); 644 VOP_UNLOCK(devvp, 0); 645 if (error) 646 goto out; 647 if (devvp->v_rdev->si_iosize_max != 0) 648 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 649 if (mp->mnt_iosize_max > MAXPHYS) 650 mp->mnt_iosize_max = MAXPHYS; 651 652 devvp->v_bufobj.bo_private = cp; 653 devvp->v_bufobj.bo_ops = &ffs_ops; 654 655 fs = NULL; 656 sblockloc = 0; 657 /* 658 * Try reading the superblock in each of its possible locations. 659 */ 660 for (i = 0; sblock_try[i] != -1; i++) { 661 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 662 error = EINVAL; 663 vfs_mount_error(mp, 664 "Invalid sectorsize %d for superblock size %d", 665 cp->provider->sectorsize, SBLOCKSIZE); 666 goto out; 667 } 668 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE, 669 cred, &bp)) != 0) 670 goto out; 671 fs = (struct fs *)bp->b_data; 672 sblockloc = sblock_try[i]; 673 if ((fs->fs_magic == FS_UFS1_MAGIC || 674 (fs->fs_magic == FS_UFS2_MAGIC && 675 (fs->fs_sblockloc == sblockloc || 676 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 677 fs->fs_bsize <= MAXBSIZE && 678 fs->fs_bsize >= sizeof(struct fs)) 679 break; 680 brelse(bp); 681 bp = NULL; 682 } 683 if (sblock_try[i] == -1) { 684 error = EINVAL; /* XXX needs translation */ 685 goto out; 686 } 687 fs->fs_fmod = 0; 688 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 689 fs->fs_flags &= ~FS_UNCLEAN; 690 if (fs->fs_clean == 0) { 691 fs->fs_flags |= FS_UNCLEAN; 692 if (ronly || (mp->mnt_flag & MNT_FORCE) || 693 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 694 (fs->fs_flags & FS_DOSOFTDEP))) { 695 printf( 696 "WARNING: %s was not properly dismounted\n", 697 fs->fs_fsmnt); 698 } else { 699 printf( 700 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 701 fs->fs_fsmnt); 702 error = EPERM; 703 goto out; 704 } 705 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 706 (mp->mnt_flag & MNT_FORCE)) { 707 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 708 (intmax_t)fs->fs_pendingblocks, 709 fs->fs_pendinginodes); 710 fs->fs_pendingblocks = 0; 711 fs->fs_pendinginodes = 0; 712 } 713 } 714 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 715 printf("%s: mount pending error: blocks %jd files %d\n", 716 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 717 fs->fs_pendinginodes); 718 fs->fs_pendingblocks = 0; 719 fs->fs_pendinginodes = 0; 720 } 721 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 722 #ifdef UFS_GJOURNAL 723 /* 724 * Get journal provider name. 725 */ 726 size = 1024; 727 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK); 728 if (g_io_getattr("GJOURNAL::provider", cp, &size, 729 mp->mnt_gjprovider) == 0) { 730 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size, 731 M_UFSMNT, M_WAITOK); 732 MNT_ILOCK(mp); 733 mp->mnt_flag |= MNT_GJOURNAL; 734 MNT_IUNLOCK(mp); 735 } else { 736 printf( 737 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n", 738 mp->mnt_stat.f_mntonname); 739 free(mp->mnt_gjprovider, M_UFSMNT); 740 mp->mnt_gjprovider = NULL; 741 } 742 #else 743 printf( 744 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n", 745 mp->mnt_stat.f_mntonname); 746 #endif 747 } else { 748 mp->mnt_gjprovider = NULL; 749 } 750 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 751 ump->um_cp = cp; 752 ump->um_bo = &devvp->v_bufobj; 753 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 754 if (fs->fs_magic == FS_UFS1_MAGIC) { 755 ump->um_fstype = UFS1; 756 ump->um_balloc = ffs_balloc_ufs1; 757 } else { 758 ump->um_fstype = UFS2; 759 ump->um_balloc = ffs_balloc_ufs2; 760 } 761 ump->um_blkatoff = ffs_blkatoff; 762 ump->um_truncate = ffs_truncate; 763 ump->um_update = ffs_update; 764 ump->um_valloc = ffs_valloc; 765 ump->um_vfree = ffs_vfree; 766 ump->um_ifree = ffs_ifree; 767 ump->um_rdonly = ffs_rdonly; 768 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 769 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 770 if (fs->fs_sbsize < SBLOCKSIZE) 771 bp->b_flags |= B_INVAL | B_NOCACHE; 772 brelse(bp); 773 bp = NULL; 774 fs = ump->um_fs; 775 ffs_oldfscompat_read(fs, ump, sblockloc); 776 fs->fs_ronly = ronly; 777 size = fs->fs_cssize; 778 blks = howmany(size, fs->fs_fsize); 779 if (fs->fs_contigsumsize > 0) 780 size += fs->fs_ncg * sizeof(int32_t); 781 size += fs->fs_ncg * sizeof(u_int8_t); 782 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 783 fs->fs_csp = space; 784 for (i = 0; i < blks; i += fs->fs_frag) { 785 size = fs->fs_bsize; 786 if (i + fs->fs_frag > blks) 787 size = (blks - i) * fs->fs_fsize; 788 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 789 cred, &bp)) != 0) { 790 free(fs->fs_csp, M_UFSMNT); 791 goto out; 792 } 793 bcopy(bp->b_data, space, (u_int)size); 794 space = (char *)space + size; 795 brelse(bp); 796 bp = NULL; 797 } 798 if (fs->fs_contigsumsize > 0) { 799 fs->fs_maxcluster = lp = space; 800 for (i = 0; i < fs->fs_ncg; i++) 801 *lp++ = fs->fs_contigsumsize; 802 space = lp; 803 } 804 size = fs->fs_ncg * sizeof(u_int8_t); 805 fs->fs_contigdirs = (u_int8_t *)space; 806 bzero(fs->fs_contigdirs, size); 807 fs->fs_active = NULL; 808 mp->mnt_data = ump; 809 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 810 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 811 nmp = NULL; 812 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 813 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 814 if (nmp) 815 vfs_rel(nmp); 816 vfs_getnewfsid(mp); 817 } 818 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 819 MNT_ILOCK(mp); 820 mp->mnt_flag |= MNT_LOCAL; 821 MNT_IUNLOCK(mp); 822 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 823 #ifdef MAC 824 MNT_ILOCK(mp); 825 mp->mnt_flag |= MNT_MULTILABEL; 826 MNT_IUNLOCK(mp); 827 #else 828 printf( 829 "WARNING: %s: multilabel flag on fs but no MAC support\n", 830 mp->mnt_stat.f_mntonname); 831 #endif 832 } 833 if ((fs->fs_flags & FS_ACLS) != 0) { 834 #ifdef UFS_ACL 835 MNT_ILOCK(mp); 836 mp->mnt_flag |= MNT_ACLS; 837 MNT_IUNLOCK(mp); 838 #else 839 printf( 840 "WARNING: %s: ACLs flag on fs but no ACLs support\n", 841 mp->mnt_stat.f_mntonname); 842 #endif 843 } 844 ump->um_mountp = mp; 845 ump->um_dev = dev; 846 ump->um_devvp = devvp; 847 ump->um_nindir = fs->fs_nindir; 848 ump->um_bptrtodb = fs->fs_fsbtodb; 849 ump->um_seqinc = fs->fs_frag; 850 for (i = 0; i < MAXQUOTAS; i++) 851 ump->um_quotas[i] = NULLVP; 852 #ifdef UFS_EXTATTR 853 ufs_extattr_uepm_init(&ump->um_extattr); 854 #endif 855 /* 856 * Set FS local "last mounted on" information (NULL pad) 857 */ 858 bzero(fs->fs_fsmnt, MAXMNTLEN); 859 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 860 861 if( mp->mnt_flag & MNT_ROOTFS) { 862 /* 863 * Root mount; update timestamp in mount structure. 864 * this will be used by the common root mount code 865 * to update the system clock. 866 */ 867 mp->mnt_time = fs->fs_time; 868 } 869 870 if (ronly == 0) { 871 if ((fs->fs_flags & FS_DOSOFTDEP) && 872 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 873 free(fs->fs_csp, M_UFSMNT); 874 goto out; 875 } 876 if (fs->fs_snapinum[0] != 0) 877 ffs_snapshot_mount(mp); 878 fs->fs_fmod = 1; 879 fs->fs_clean = 0; 880 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 881 } 882 /* 883 * Initialize filesystem stat information in mount struct. 884 */ 885 MNT_ILOCK(mp); 886 mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED; 887 MNT_IUNLOCK(mp); 888 #ifdef UFS_EXTATTR 889 #ifdef UFS_EXTATTR_AUTOSTART 890 /* 891 * 892 * Auto-starting does the following: 893 * - check for /.attribute in the fs, and extattr_start if so 894 * - for each file in .attribute, enable that file with 895 * an attribute of the same name. 896 * Not clear how to report errors -- probably eat them. 897 * This would all happen while the filesystem was busy/not 898 * available, so would effectively be "atomic". 899 */ 900 mp->mnt_stat.f_iosize = fs->fs_bsize; 901 (void) ufs_extattr_autostart(mp, td); 902 #endif /* !UFS_EXTATTR_AUTOSTART */ 903 #endif /* !UFS_EXTATTR */ 904 return (0); 905 out: 906 if (bp) 907 brelse(bp); 908 if (cp != NULL) { 909 DROP_GIANT(); 910 g_topology_lock(); 911 g_vfs_close(cp); 912 g_topology_unlock(); 913 PICKUP_GIANT(); 914 } 915 if (ump) { 916 mtx_destroy(UFS_MTX(ump)); 917 if (mp->mnt_gjprovider != NULL) { 918 free(mp->mnt_gjprovider, M_UFSMNT); 919 mp->mnt_gjprovider = NULL; 920 } 921 free(ump->um_fs, M_UFSMNT); 922 free(ump, M_UFSMNT); 923 mp->mnt_data = NULL; 924 } 925 dev_rel(dev); 926 return (error); 927 } 928 929 #include <sys/sysctl.h> 930 static int bigcgs = 0; 931 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 932 933 /* 934 * Sanity checks for loading old filesystem superblocks. 935 * See ffs_oldfscompat_write below for unwound actions. 936 * 937 * XXX - Parts get retired eventually. 938 * Unfortunately new bits get added. 939 */ 940 static void 941 ffs_oldfscompat_read(fs, ump, sblockloc) 942 struct fs *fs; 943 struct ufsmount *ump; 944 ufs2_daddr_t sblockloc; 945 { 946 off_t maxfilesize; 947 948 /* 949 * If not yet done, update fs_flags location and value of fs_sblockloc. 950 */ 951 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 952 fs->fs_flags = fs->fs_old_flags; 953 fs->fs_old_flags |= FS_FLAGS_UPDATED; 954 fs->fs_sblockloc = sblockloc; 955 } 956 /* 957 * If not yet done, update UFS1 superblock with new wider fields. 958 */ 959 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 960 fs->fs_maxbsize = fs->fs_bsize; 961 fs->fs_time = fs->fs_old_time; 962 fs->fs_size = fs->fs_old_size; 963 fs->fs_dsize = fs->fs_old_dsize; 964 fs->fs_csaddr = fs->fs_old_csaddr; 965 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 966 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 967 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 968 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 969 } 970 if (fs->fs_magic == FS_UFS1_MAGIC && 971 fs->fs_old_inodefmt < FS_44INODEFMT) { 972 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 973 fs->fs_qbmask = ~fs->fs_bmask; 974 fs->fs_qfmask = ~fs->fs_fmask; 975 } 976 if (fs->fs_magic == FS_UFS1_MAGIC) { 977 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 978 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 979 if (fs->fs_maxfilesize > maxfilesize) 980 fs->fs_maxfilesize = maxfilesize; 981 } 982 /* Compatibility for old filesystems */ 983 if (fs->fs_avgfilesize <= 0) 984 fs->fs_avgfilesize = AVFILESIZ; 985 if (fs->fs_avgfpdir <= 0) 986 fs->fs_avgfpdir = AFPDIR; 987 if (bigcgs) { 988 fs->fs_save_cgsize = fs->fs_cgsize; 989 fs->fs_cgsize = fs->fs_bsize; 990 } 991 } 992 993 /* 994 * Unwinding superblock updates for old filesystems. 995 * See ffs_oldfscompat_read above for details. 996 * 997 * XXX - Parts get retired eventually. 998 * Unfortunately new bits get added. 999 */ 1000 static void 1001 ffs_oldfscompat_write(fs, ump) 1002 struct fs *fs; 1003 struct ufsmount *ump; 1004 { 1005 1006 /* 1007 * Copy back UFS2 updated fields that UFS1 inspects. 1008 */ 1009 if (fs->fs_magic == FS_UFS1_MAGIC) { 1010 fs->fs_old_time = fs->fs_time; 1011 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1012 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1013 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1014 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1015 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1016 } 1017 if (bigcgs) { 1018 fs->fs_cgsize = fs->fs_save_cgsize; 1019 fs->fs_save_cgsize = 0; 1020 } 1021 } 1022 1023 /* 1024 * unmount system call 1025 */ 1026 static int 1027 ffs_unmount(mp, mntflags, td) 1028 struct mount *mp; 1029 int mntflags; 1030 struct thread *td; 1031 { 1032 struct ufsmount *ump = VFSTOUFS(mp); 1033 struct fs *fs; 1034 int error, flags, susp; 1035 #ifdef UFS_EXTATTR 1036 int e_restart; 1037 #endif 1038 1039 flags = 0; 1040 fs = ump->um_fs; 1041 if (mntflags & MNT_FORCE) { 1042 flags |= FORCECLOSE; 1043 susp = fs->fs_ronly != 0; 1044 } else 1045 susp = 0; 1046 #ifdef UFS_EXTATTR 1047 if ((error = ufs_extattr_stop(mp, td))) { 1048 if (error != EOPNOTSUPP) 1049 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 1050 error); 1051 e_restart = 0; 1052 } else { 1053 ufs_extattr_uepm_destroy(&ump->um_extattr); 1054 e_restart = 1; 1055 } 1056 #endif 1057 if (susp) { 1058 /* 1059 * dounmount already called vn_start_write(). 1060 */ 1061 for (;;) { 1062 vn_finished_write(mp); 1063 if ((error = vfs_write_suspend(mp)) != 0) 1064 return (error); 1065 MNT_ILOCK(mp); 1066 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 1067 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 1068 MNTK_SUSPEND2); 1069 wakeup(&mp->mnt_flag); 1070 MNT_IUNLOCK(mp); 1071 curthread->td_pflags |= TDP_IGNSUSP; 1072 break; 1073 } 1074 MNT_IUNLOCK(mp); 1075 vn_start_write(NULL, &mp, V_WAIT); 1076 } 1077 } 1078 if (mp->mnt_flag & MNT_SOFTDEP) { 1079 if ((error = softdep_flushfiles(mp, flags, td)) != 0) 1080 goto fail; 1081 } else { 1082 if ((error = ffs_flushfiles(mp, flags, td)) != 0) 1083 goto fail; 1084 } 1085 UFS_LOCK(ump); 1086 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1087 printf("%s: unmount pending error: blocks %jd files %d\n", 1088 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1089 fs->fs_pendinginodes); 1090 fs->fs_pendingblocks = 0; 1091 fs->fs_pendinginodes = 0; 1092 } 1093 UFS_UNLOCK(ump); 1094 if (fs->fs_ronly == 0) { 1095 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1096 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1097 if (error) { 1098 fs->fs_clean = 0; 1099 goto fail; 1100 } 1101 } 1102 if (susp) { 1103 vfs_write_resume(mp); 1104 vn_start_write(NULL, &mp, V_WAIT); 1105 } 1106 DROP_GIANT(); 1107 g_topology_lock(); 1108 g_vfs_close(ump->um_cp); 1109 g_topology_unlock(); 1110 PICKUP_GIANT(); 1111 vrele(ump->um_devvp); 1112 dev_rel(ump->um_dev); 1113 mtx_destroy(UFS_MTX(ump)); 1114 if (mp->mnt_gjprovider != NULL) { 1115 free(mp->mnt_gjprovider, M_UFSMNT); 1116 mp->mnt_gjprovider = NULL; 1117 } 1118 free(fs->fs_csp, M_UFSMNT); 1119 free(fs, M_UFSMNT); 1120 free(ump, M_UFSMNT); 1121 mp->mnt_data = NULL; 1122 MNT_ILOCK(mp); 1123 mp->mnt_flag &= ~MNT_LOCAL; 1124 MNT_IUNLOCK(mp); 1125 return (error); 1126 1127 fail: 1128 if (susp) { 1129 vfs_write_resume(mp); 1130 vn_start_write(NULL, &mp, V_WAIT); 1131 } 1132 #ifdef UFS_EXTATTR 1133 if (e_restart) { 1134 ufs_extattr_uepm_init(&ump->um_extattr); 1135 #ifdef UFS_EXTATTR_AUTOSTART 1136 (void) ufs_extattr_autostart(mp, td); 1137 #endif 1138 } 1139 #endif 1140 1141 return (error); 1142 } 1143 1144 /* 1145 * Flush out all the files in a filesystem. 1146 */ 1147 int 1148 ffs_flushfiles(mp, flags, td) 1149 struct mount *mp; 1150 int flags; 1151 struct thread *td; 1152 { 1153 struct ufsmount *ump; 1154 int error; 1155 1156 ump = VFSTOUFS(mp); 1157 #ifdef QUOTA 1158 if (mp->mnt_flag & MNT_QUOTA) { 1159 int i; 1160 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1161 if (error) 1162 return (error); 1163 for (i = 0; i < MAXQUOTAS; i++) { 1164 quotaoff(td, mp, i); 1165 } 1166 /* 1167 * Here we fall through to vflush again to ensure 1168 * that we have gotten rid of all the system vnodes. 1169 */ 1170 } 1171 #endif 1172 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1173 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1174 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1175 return (error); 1176 ffs_snapshot_unmount(mp); 1177 flags |= FORCECLOSE; 1178 /* 1179 * Here we fall through to vflush again to ensure 1180 * that we have gotten rid of all the system vnodes. 1181 */ 1182 } 1183 /* 1184 * Flush all the files. 1185 */ 1186 if ((error = vflush(mp, 0, flags, td)) != 0) 1187 return (error); 1188 /* 1189 * Flush filesystem metadata. 1190 */ 1191 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1192 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1193 VOP_UNLOCK(ump->um_devvp, 0); 1194 return (error); 1195 } 1196 1197 /* 1198 * Get filesystem statistics. 1199 */ 1200 static int 1201 ffs_statfs(mp, sbp, td) 1202 struct mount *mp; 1203 struct statfs *sbp; 1204 struct thread *td; 1205 { 1206 struct ufsmount *ump; 1207 struct fs *fs; 1208 1209 ump = VFSTOUFS(mp); 1210 fs = ump->um_fs; 1211 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1212 panic("ffs_statfs"); 1213 sbp->f_version = STATFS_VERSION; 1214 sbp->f_bsize = fs->fs_fsize; 1215 sbp->f_iosize = fs->fs_bsize; 1216 sbp->f_blocks = fs->fs_dsize; 1217 UFS_LOCK(ump); 1218 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1219 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1220 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1221 dbtofsb(fs, fs->fs_pendingblocks); 1222 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1223 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1224 UFS_UNLOCK(ump); 1225 sbp->f_namemax = NAME_MAX; 1226 return (0); 1227 } 1228 1229 /* 1230 * Go through the disk queues to initiate sandbagged IO; 1231 * go through the inodes to write those that have been modified; 1232 * initiate the writing of the super block if it has been modified. 1233 * 1234 * Note: we are always called with the filesystem marked `MPBUSY'. 1235 */ 1236 static int 1237 ffs_sync(mp, waitfor, td) 1238 struct mount *mp; 1239 int waitfor; 1240 struct thread *td; 1241 { 1242 struct vnode *mvp, *vp, *devvp; 1243 struct inode *ip; 1244 struct ufsmount *ump = VFSTOUFS(mp); 1245 struct fs *fs; 1246 int error, count, wait, lockreq, allerror = 0; 1247 int suspend; 1248 int suspended; 1249 int secondary_writes; 1250 int secondary_accwrites; 1251 int softdep_deps; 1252 int softdep_accdeps; 1253 struct bufobj *bo; 1254 1255 fs = ump->um_fs; 1256 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1257 printf("fs = %s\n", fs->fs_fsmnt); 1258 panic("ffs_sync: rofs mod"); 1259 } 1260 /* 1261 * Write back each (modified) inode. 1262 */ 1263 wait = 0; 1264 suspend = 0; 1265 suspended = 0; 1266 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1267 if (waitfor == MNT_SUSPEND) { 1268 suspend = 1; 1269 waitfor = MNT_WAIT; 1270 } 1271 if (waitfor == MNT_WAIT) { 1272 wait = 1; 1273 lockreq = LK_EXCLUSIVE; 1274 } 1275 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1276 MNT_ILOCK(mp); 1277 loop: 1278 /* Grab snapshot of secondary write counts */ 1279 secondary_writes = mp->mnt_secondary_writes; 1280 secondary_accwrites = mp->mnt_secondary_accwrites; 1281 1282 /* Grab snapshot of softdep dependency counts */ 1283 MNT_IUNLOCK(mp); 1284 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1285 MNT_ILOCK(mp); 1286 1287 MNT_VNODE_FOREACH(vp, mp, mvp) { 1288 /* 1289 * Depend on the mntvnode_slock to keep things stable enough 1290 * for a quick test. Since there might be hundreds of 1291 * thousands of vnodes, we cannot afford even a subroutine 1292 * call unless there's a good chance that we have work to do. 1293 */ 1294 VI_LOCK(vp); 1295 if (vp->v_iflag & VI_DOOMED) { 1296 VI_UNLOCK(vp); 1297 continue; 1298 } 1299 ip = VTOI(vp); 1300 if (vp->v_type == VNON || ((ip->i_flag & 1301 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1302 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1303 VI_UNLOCK(vp); 1304 continue; 1305 } 1306 MNT_IUNLOCK(mp); 1307 if ((error = vget(vp, lockreq, td)) != 0) { 1308 MNT_ILOCK(mp); 1309 if (error == ENOENT || error == ENOLCK) { 1310 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1311 goto loop; 1312 } 1313 continue; 1314 } 1315 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1316 allerror = error; 1317 vput(vp); 1318 MNT_ILOCK(mp); 1319 } 1320 MNT_IUNLOCK(mp); 1321 /* 1322 * Force stale filesystem control information to be flushed. 1323 */ 1324 if (waitfor == MNT_WAIT) { 1325 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1326 allerror = error; 1327 /* Flushed work items may create new vnodes to clean */ 1328 if (allerror == 0 && count) { 1329 MNT_ILOCK(mp); 1330 goto loop; 1331 } 1332 } 1333 #ifdef QUOTA 1334 qsync(mp); 1335 #endif 1336 devvp = ump->um_devvp; 1337 bo = &devvp->v_bufobj; 1338 BO_LOCK(bo); 1339 if (waitfor != MNT_LAZY && 1340 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1341 BO_UNLOCK(bo); 1342 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1343 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1344 allerror = error; 1345 VOP_UNLOCK(devvp, 0); 1346 if (allerror == 0 && waitfor == MNT_WAIT) { 1347 MNT_ILOCK(mp); 1348 goto loop; 1349 } 1350 } else if (suspend != 0) { 1351 if (softdep_check_suspend(mp, 1352 devvp, 1353 softdep_deps, 1354 softdep_accdeps, 1355 secondary_writes, 1356 secondary_accwrites) != 0) 1357 goto loop; /* More work needed */ 1358 mtx_assert(MNT_MTX(mp), MA_OWNED); 1359 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1360 MNT_IUNLOCK(mp); 1361 suspended = 1; 1362 } else 1363 BO_UNLOCK(bo); 1364 /* 1365 * Write back modified superblock. 1366 */ 1367 if (fs->fs_fmod != 0 && 1368 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1369 allerror = error; 1370 return (allerror); 1371 } 1372 1373 int 1374 ffs_vget(mp, ino, flags, vpp) 1375 struct mount *mp; 1376 ino_t ino; 1377 int flags; 1378 struct vnode **vpp; 1379 { 1380 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1381 } 1382 1383 int 1384 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1385 struct mount *mp; 1386 ino_t ino; 1387 int flags; 1388 struct vnode **vpp; 1389 int ffs_flags; 1390 { 1391 struct fs *fs; 1392 struct inode *ip; 1393 struct ufsmount *ump; 1394 struct buf *bp; 1395 struct vnode *vp; 1396 struct cdev *dev; 1397 int error; 1398 1399 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1400 if (error || *vpp != NULL) 1401 return (error); 1402 1403 /* 1404 * We must promote to an exclusive lock for vnode creation. This 1405 * can happen if lookup is passed LOCKSHARED. 1406 */ 1407 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1408 flags &= ~LK_TYPE_MASK; 1409 flags |= LK_EXCLUSIVE; 1410 } 1411 1412 /* 1413 * We do not lock vnode creation as it is believed to be too 1414 * expensive for such rare case as simultaneous creation of vnode 1415 * for same ino by different processes. We just allow them to race 1416 * and check later to decide who wins. Let the race begin! 1417 */ 1418 1419 ump = VFSTOUFS(mp); 1420 dev = ump->um_dev; 1421 fs = ump->um_fs; 1422 1423 /* 1424 * If this malloc() is performed after the getnewvnode() 1425 * it might block, leaving a vnode with a NULL v_data to be 1426 * found by ffs_sync() if a sync happens to fire right then, 1427 * which will cause a panic because ffs_sync() blindly 1428 * dereferences vp->v_data (as well it should). 1429 */ 1430 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1431 1432 /* Allocate a new vnode/inode. */ 1433 if (fs->fs_magic == FS_UFS1_MAGIC) 1434 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1435 else 1436 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1437 if (error) { 1438 *vpp = NULL; 1439 uma_zfree(uma_inode, ip); 1440 return (error); 1441 } 1442 /* 1443 * FFS supports recursive and shared locking. 1444 */ 1445 VN_LOCK_AREC(vp); 1446 VN_LOCK_ASHARE(vp); 1447 vp->v_data = ip; 1448 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1449 ip->i_vnode = vp; 1450 ip->i_ump = ump; 1451 ip->i_fs = fs; 1452 ip->i_dev = dev; 1453 ip->i_number = ino; 1454 #ifdef QUOTA 1455 { 1456 int i; 1457 for (i = 0; i < MAXQUOTAS; i++) 1458 ip->i_dquot[i] = NODQUOT; 1459 } 1460 #endif 1461 1462 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1463 if (ffs_flags & FFSV_FORCEINSMQ) 1464 vp->v_vflag |= VV_FORCEINSMQ; 1465 error = insmntque(vp, mp); 1466 if (error != 0) { 1467 uma_zfree(uma_inode, ip); 1468 *vpp = NULL; 1469 return (error); 1470 } 1471 vp->v_vflag &= ~VV_FORCEINSMQ; 1472 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1473 if (error || *vpp != NULL) 1474 return (error); 1475 1476 /* Read in the disk contents for the inode, copy into the inode. */ 1477 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1478 (int)fs->fs_bsize, NOCRED, &bp); 1479 if (error) { 1480 /* 1481 * The inode does not contain anything useful, so it would 1482 * be misleading to leave it on its hash chain. With mode 1483 * still zero, it will be unlinked and returned to the free 1484 * list by vput(). 1485 */ 1486 brelse(bp); 1487 vput(vp); 1488 *vpp = NULL; 1489 return (error); 1490 } 1491 if (ip->i_ump->um_fstype == UFS1) 1492 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1493 else 1494 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1495 ffs_load_inode(bp, ip, fs, ino); 1496 if (DOINGSOFTDEP(vp)) 1497 softdep_load_inodeblock(ip); 1498 else 1499 ip->i_effnlink = ip->i_nlink; 1500 bqrelse(bp); 1501 1502 /* 1503 * Initialize the vnode from the inode, check for aliases. 1504 * Note that the underlying vnode may have changed. 1505 */ 1506 if (ip->i_ump->um_fstype == UFS1) 1507 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1508 else 1509 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1510 if (error) { 1511 vput(vp); 1512 *vpp = NULL; 1513 return (error); 1514 } 1515 1516 /* 1517 * Finish inode initialization. 1518 */ 1519 1520 /* 1521 * Set up a generation number for this inode if it does not 1522 * already have one. This should only happen on old filesystems. 1523 */ 1524 if (ip->i_gen == 0) { 1525 ip->i_gen = arc4random() / 2 + 1; 1526 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1527 ip->i_flag |= IN_MODIFIED; 1528 DIP_SET(ip, i_gen, ip->i_gen); 1529 } 1530 } 1531 /* 1532 * Ensure that uid and gid are correct. This is a temporary 1533 * fix until fsck has been changed to do the update. 1534 */ 1535 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1536 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1537 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1538 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1539 } /* XXX */ 1540 1541 #ifdef MAC 1542 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1543 /* 1544 * If this vnode is already allocated, and we're running 1545 * multi-label, attempt to perform a label association 1546 * from the extended attributes on the inode. 1547 */ 1548 error = mac_vnode_associate_extattr(mp, vp); 1549 if (error) { 1550 /* ufs_inactive will release ip->i_devvp ref. */ 1551 vput(vp); 1552 *vpp = NULL; 1553 return (error); 1554 } 1555 } 1556 #endif 1557 1558 *vpp = vp; 1559 return (0); 1560 } 1561 1562 /* 1563 * File handle to vnode 1564 * 1565 * Have to be really careful about stale file handles: 1566 * - check that the inode number is valid 1567 * - call ffs_vget() to get the locked inode 1568 * - check for an unallocated inode (i_mode == 0) 1569 * - check that the given client host has export rights and return 1570 * those rights via. exflagsp and credanonp 1571 */ 1572 static int 1573 ffs_fhtovp(mp, fhp, vpp) 1574 struct mount *mp; 1575 struct fid *fhp; 1576 struct vnode **vpp; 1577 { 1578 struct ufid *ufhp; 1579 struct fs *fs; 1580 1581 ufhp = (struct ufid *)fhp; 1582 fs = VFSTOUFS(mp)->um_fs; 1583 if (ufhp->ufid_ino < ROOTINO || 1584 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1585 return (ESTALE); 1586 return (ufs_fhtovp(mp, ufhp, vpp)); 1587 } 1588 1589 /* 1590 * Initialize the filesystem. 1591 */ 1592 static int 1593 ffs_init(vfsp) 1594 struct vfsconf *vfsp; 1595 { 1596 1597 softdep_initialize(); 1598 return (ufs_init(vfsp)); 1599 } 1600 1601 /* 1602 * Undo the work of ffs_init(). 1603 */ 1604 static int 1605 ffs_uninit(vfsp) 1606 struct vfsconf *vfsp; 1607 { 1608 int ret; 1609 1610 ret = ufs_uninit(vfsp); 1611 softdep_uninitialize(); 1612 return (ret); 1613 } 1614 1615 /* 1616 * Write a superblock and associated information back to disk. 1617 */ 1618 int 1619 ffs_sbupdate(mp, waitfor, suspended) 1620 struct ufsmount *mp; 1621 int waitfor; 1622 int suspended; 1623 { 1624 struct fs *fs = mp->um_fs; 1625 struct buf *sbbp; 1626 struct buf *bp; 1627 int blks; 1628 void *space; 1629 int i, size, error, allerror = 0; 1630 1631 if (fs->fs_ronly == 1 && 1632 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1633 (MNT_RDONLY | MNT_UPDATE)) 1634 panic("ffs_sbupdate: write read-only filesystem"); 1635 /* 1636 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1637 */ 1638 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1639 0, 0, 0); 1640 /* 1641 * First write back the summary information. 1642 */ 1643 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1644 space = fs->fs_csp; 1645 for (i = 0; i < blks; i += fs->fs_frag) { 1646 size = fs->fs_bsize; 1647 if (i + fs->fs_frag > blks) 1648 size = (blks - i) * fs->fs_fsize; 1649 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1650 size, 0, 0, 0); 1651 bcopy(space, bp->b_data, (u_int)size); 1652 space = (char *)space + size; 1653 if (suspended) 1654 bp->b_flags |= B_VALIDSUSPWRT; 1655 if (waitfor != MNT_WAIT) 1656 bawrite(bp); 1657 else if ((error = bwrite(bp)) != 0) 1658 allerror = error; 1659 } 1660 /* 1661 * Now write back the superblock itself. If any errors occurred 1662 * up to this point, then fail so that the superblock avoids 1663 * being written out as clean. 1664 */ 1665 if (allerror) { 1666 brelse(sbbp); 1667 return (allerror); 1668 } 1669 bp = sbbp; 1670 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1671 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1672 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1673 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1674 fs->fs_sblockloc = SBLOCK_UFS1; 1675 } 1676 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1677 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1678 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1679 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1680 fs->fs_sblockloc = SBLOCK_UFS2; 1681 } 1682 fs->fs_fmod = 0; 1683 fs->fs_time = time_second; 1684 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1685 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1686 if (suspended) 1687 bp->b_flags |= B_VALIDSUSPWRT; 1688 if (waitfor != MNT_WAIT) 1689 bawrite(bp); 1690 else if ((error = bwrite(bp)) != 0) 1691 allerror = error; 1692 return (allerror); 1693 } 1694 1695 static int 1696 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1697 int attrnamespace, const char *attrname, struct thread *td) 1698 { 1699 1700 #ifdef UFS_EXTATTR 1701 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1702 attrname, td)); 1703 #else 1704 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1705 attrname, td)); 1706 #endif 1707 } 1708 1709 static void 1710 ffs_ifree(struct ufsmount *ump, struct inode *ip) 1711 { 1712 1713 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1714 uma_zfree(uma_ufs1, ip->i_din1); 1715 else if (ip->i_din2 != NULL) 1716 uma_zfree(uma_ufs2, ip->i_din2); 1717 uma_zfree(uma_inode, ip); 1718 } 1719 1720 static int dobkgrdwrite = 1; 1721 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1722 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1723 1724 /* 1725 * Complete a background write started from bwrite. 1726 */ 1727 static void 1728 ffs_backgroundwritedone(struct buf *bp) 1729 { 1730 struct bufobj *bufobj; 1731 struct buf *origbp; 1732 1733 /* 1734 * Find the original buffer that we are writing. 1735 */ 1736 bufobj = bp->b_bufobj; 1737 BO_LOCK(bufobj); 1738 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1739 panic("backgroundwritedone: lost buffer"); 1740 /* Grab an extra reference to be dropped by the bufdone() below. */ 1741 bufobj_wrefl(bufobj); 1742 BO_UNLOCK(bufobj); 1743 /* 1744 * Process dependencies then return any unfinished ones. 1745 */ 1746 if (!LIST_EMPTY(&bp->b_dep)) 1747 buf_complete(bp); 1748 #ifdef SOFTUPDATES 1749 if (!LIST_EMPTY(&bp->b_dep)) 1750 softdep_move_dependencies(bp, origbp); 1751 #endif 1752 /* 1753 * This buffer is marked B_NOCACHE so when it is released 1754 * by biodone it will be tossed. 1755 */ 1756 bp->b_flags |= B_NOCACHE; 1757 bp->b_flags &= ~B_CACHE; 1758 bufdone(bp); 1759 BO_LOCK(bufobj); 1760 /* 1761 * Clear the BV_BKGRDINPROG flag in the original buffer 1762 * and awaken it if it is waiting for the write to complete. 1763 * If BV_BKGRDINPROG is not set in the original buffer it must 1764 * have been released and re-instantiated - which is not legal. 1765 */ 1766 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1767 ("backgroundwritedone: lost buffer2")); 1768 origbp->b_vflags &= ~BV_BKGRDINPROG; 1769 if (origbp->b_vflags & BV_BKGRDWAIT) { 1770 origbp->b_vflags &= ~BV_BKGRDWAIT; 1771 wakeup(&origbp->b_xflags); 1772 } 1773 BO_UNLOCK(bufobj); 1774 } 1775 1776 1777 /* 1778 * Write, release buffer on completion. (Done by iodone 1779 * if async). Do not bother writing anything if the buffer 1780 * is invalid. 1781 * 1782 * Note that we set B_CACHE here, indicating that buffer is 1783 * fully valid and thus cacheable. This is true even of NFS 1784 * now so we set it generally. This could be set either here 1785 * or in biodone() since the I/O is synchronous. We put it 1786 * here. 1787 */ 1788 static int 1789 ffs_bufwrite(struct buf *bp) 1790 { 1791 int oldflags, s; 1792 struct buf *newbp; 1793 1794 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1795 if (bp->b_flags & B_INVAL) { 1796 brelse(bp); 1797 return (0); 1798 } 1799 1800 oldflags = bp->b_flags; 1801 1802 if (!BUF_ISLOCKED(bp)) 1803 panic("bufwrite: buffer is not busy???"); 1804 s = splbio(); 1805 /* 1806 * If a background write is already in progress, delay 1807 * writing this block if it is asynchronous. Otherwise 1808 * wait for the background write to complete. 1809 */ 1810 BO_LOCK(bp->b_bufobj); 1811 if (bp->b_vflags & BV_BKGRDINPROG) { 1812 if (bp->b_flags & B_ASYNC) { 1813 BO_UNLOCK(bp->b_bufobj); 1814 splx(s); 1815 bdwrite(bp); 1816 return (0); 1817 } 1818 bp->b_vflags |= BV_BKGRDWAIT; 1819 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1820 if (bp->b_vflags & BV_BKGRDINPROG) 1821 panic("bufwrite: still writing"); 1822 } 1823 BO_UNLOCK(bp->b_bufobj); 1824 1825 /* Mark the buffer clean */ 1826 bundirty(bp); 1827 1828 /* 1829 * If this buffer is marked for background writing and we 1830 * do not have to wait for it, make a copy and write the 1831 * copy so as to leave this buffer ready for further use. 1832 * 1833 * This optimization eats a lot of memory. If we have a page 1834 * or buffer shortfall we can't do it. 1835 */ 1836 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1837 (bp->b_flags & B_ASYNC) && 1838 !vm_page_count_severe() && 1839 !buf_dirty_count_severe()) { 1840 KASSERT(bp->b_iodone == NULL, 1841 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1842 1843 /* get a new block */ 1844 newbp = geteblk(bp->b_bufsize); 1845 1846 /* 1847 * set it to be identical to the old block. We have to 1848 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1849 * to avoid confusing the splay tree and gbincore(). 1850 */ 1851 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1852 newbp->b_lblkno = bp->b_lblkno; 1853 newbp->b_xflags |= BX_BKGRDMARKER; 1854 BO_LOCK(bp->b_bufobj); 1855 bp->b_vflags |= BV_BKGRDINPROG; 1856 bgetvp(bp->b_vp, newbp); 1857 BO_UNLOCK(bp->b_bufobj); 1858 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1859 newbp->b_blkno = bp->b_blkno; 1860 newbp->b_offset = bp->b_offset; 1861 newbp->b_iodone = ffs_backgroundwritedone; 1862 newbp->b_flags |= B_ASYNC; 1863 newbp->b_flags &= ~B_INVAL; 1864 1865 #ifdef SOFTUPDATES 1866 /* move over the dependencies */ 1867 if (!LIST_EMPTY(&bp->b_dep)) 1868 softdep_move_dependencies(bp, newbp); 1869 #endif 1870 1871 /* 1872 * Initiate write on the copy, release the original to 1873 * the B_LOCKED queue so that it cannot go away until 1874 * the background write completes. If not locked it could go 1875 * away and then be reconstituted while it was being written. 1876 * If the reconstituted buffer were written, we could end up 1877 * with two background copies being written at the same time. 1878 */ 1879 bqrelse(bp); 1880 bp = newbp; 1881 } 1882 1883 /* Let the normal bufwrite do the rest for us */ 1884 return (bufwrite(bp)); 1885 } 1886 1887 1888 static void 1889 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1890 { 1891 struct vnode *vp; 1892 int error; 1893 struct buf *tbp; 1894 1895 vp = bo->__bo_vnode; 1896 if (bp->b_iocmd == BIO_WRITE) { 1897 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1898 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1899 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1900 panic("ffs_geom_strategy: bad I/O"); 1901 bp->b_flags &= ~B_VALIDSUSPWRT; 1902 if ((vp->v_vflag & VV_COPYONWRITE) && 1903 vp->v_rdev->si_snapdata != NULL) { 1904 if ((bp->b_flags & B_CLUSTER) != 0) { 1905 runningbufwakeup(bp); 1906 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1907 b_cluster.cluster_entry) { 1908 error = ffs_copyonwrite(vp, tbp); 1909 if (error != 0 && 1910 error != EOPNOTSUPP) { 1911 bp->b_error = error; 1912 bp->b_ioflags |= BIO_ERROR; 1913 bufdone(bp); 1914 return; 1915 } 1916 } 1917 bp->b_runningbufspace = bp->b_bufsize; 1918 atomic_add_int(&runningbufspace, 1919 bp->b_runningbufspace); 1920 } else { 1921 error = ffs_copyonwrite(vp, bp); 1922 if (error != 0 && error != EOPNOTSUPP) { 1923 bp->b_error = error; 1924 bp->b_ioflags |= BIO_ERROR; 1925 bufdone(bp); 1926 return; 1927 } 1928 } 1929 } 1930 #ifdef SOFTUPDATES 1931 if ((bp->b_flags & B_CLUSTER) != 0) { 1932 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1933 b_cluster.cluster_entry) { 1934 if (!LIST_EMPTY(&tbp->b_dep)) 1935 buf_start(tbp); 1936 } 1937 } else { 1938 if (!LIST_EMPTY(&bp->b_dep)) 1939 buf_start(bp); 1940 } 1941 1942 #endif 1943 } 1944 g_vfs_strategy(bo, bp); 1945 } 1946 1947 #ifdef DDB 1948 1949 static void 1950 db_print_ffs(struct ufsmount *ump) 1951 { 1952 db_printf("mp %p %s devvp %p fs %p su_wl %d su_wl_in %d su_deps %d " 1953 "su_req %d\n", 1954 ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname, 1955 ump->um_devvp, ump->um_fs, ump->softdep_on_worklist, 1956 ump->softdep_on_worklist_inprogress, ump->softdep_deps, 1957 ump->softdep_req); 1958 } 1959 1960 DB_SHOW_COMMAND(ffs, db_show_ffs) 1961 { 1962 struct mount *mp; 1963 struct ufsmount *ump; 1964 1965 if (have_addr) { 1966 ump = VFSTOUFS((struct mount *)addr); 1967 db_print_ffs(ump); 1968 return; 1969 } 1970 1971 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1972 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 1973 db_print_ffs(VFSTOUFS(mp)); 1974 } 1975 } 1976 1977 #endif /* DDB */ 1978