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_quota.h" 36 #include "opt_ufs.h" 37 #include "opt_ffs.h" 38 #include "opt_ddb.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 #include <ddb/ddb.h> 75 76 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 77 78 static int ffs_reload(struct mount *, struct thread *); 79 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 80 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 81 ufs2_daddr_t); 82 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 83 static vfs_init_t ffs_init; 84 static vfs_uninit_t ffs_uninit; 85 static vfs_extattrctl_t ffs_extattrctl; 86 static vfs_cmount_t ffs_cmount; 87 static vfs_unmount_t ffs_unmount; 88 static vfs_mount_t ffs_mount; 89 static vfs_statfs_t ffs_statfs; 90 static vfs_fhtovp_t ffs_fhtovp; 91 static vfs_sync_t ffs_sync; 92 93 static struct vfsops ufs_vfsops = { 94 .vfs_extattrctl = ffs_extattrctl, 95 .vfs_fhtovp = ffs_fhtovp, 96 .vfs_init = ffs_init, 97 .vfs_mount = ffs_mount, 98 .vfs_cmount = ffs_cmount, 99 .vfs_quotactl = ufs_quotactl, 100 .vfs_root = ufs_root, 101 .vfs_statfs = ffs_statfs, 102 .vfs_sync = ffs_sync, 103 .vfs_uninit = ffs_uninit, 104 .vfs_unmount = ffs_unmount, 105 .vfs_vget = ffs_vget, 106 .vfs_susp_clean = process_deferred_inactive, 107 }; 108 109 VFS_SET(ufs_vfsops, ufs, 0); 110 MODULE_VERSION(ufs, 1); 111 112 static b_strategy_t ffs_geom_strategy; 113 static b_write_t ffs_bufwrite; 114 115 static struct buf_ops ffs_ops = { 116 .bop_name = "FFS", 117 .bop_write = ffs_bufwrite, 118 .bop_strategy = ffs_geom_strategy, 119 .bop_sync = bufsync, 120 #ifdef NO_FFS_SNAPSHOT 121 .bop_bdflush = bufbdflush, 122 #else 123 .bop_bdflush = ffs_bdflush, 124 #endif 125 }; 126 127 /* 128 * Note that userquota and groupquota options are not currently used 129 * by UFS/FFS code and generally mount(8) does not pass those options 130 * from userland, but they can be passed by loader(8) via 131 * vfs.root.mountfrom.options. 132 */ 133 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 134 "noclusterw", "noexec", "export", "force", "from", "groupquota", 135 "multilabel", "nfsv4acls", "snapshot", "nosuid", "suiddir", "nosymfollow", 136 "sync", "union", "userquota", NULL }; 137 138 static int 139 ffs_mount(struct mount *mp) 140 { 141 struct vnode *devvp; 142 struct thread *td; 143 struct ufsmount *ump = 0; 144 struct fs *fs; 145 int error, flags; 146 u_int mntorflags; 147 accmode_t accmode; 148 struct nameidata ndp; 149 char *fspec; 150 151 td = curthread; 152 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 153 return (EINVAL); 154 if (uma_inode == NULL) { 155 uma_inode = uma_zcreate("FFS inode", 156 sizeof(struct inode), NULL, NULL, NULL, NULL, 157 UMA_ALIGN_PTR, 0); 158 uma_ufs1 = uma_zcreate("FFS1 dinode", 159 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 160 UMA_ALIGN_PTR, 0); 161 uma_ufs2 = uma_zcreate("FFS2 dinode", 162 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 163 UMA_ALIGN_PTR, 0); 164 } 165 166 vfs_deleteopt(mp->mnt_optnew, "groupquota"); 167 vfs_deleteopt(mp->mnt_optnew, "userquota"); 168 169 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 170 if (error) 171 return (error); 172 173 mntorflags = 0; 174 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 175 mntorflags |= MNT_ACLS; 176 177 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 178 mntorflags |= MNT_SNAPSHOT; 179 /* 180 * Once we have set the MNT_SNAPSHOT flag, do not 181 * persist "snapshot" in the options list. 182 */ 183 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 184 vfs_deleteopt(mp->mnt_opt, "snapshot"); 185 } 186 187 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 188 if (mntorflags & MNT_ACLS) { 189 printf("WARNING: \"acls\" and \"nfsv4acls\" " 190 "options are mutually exclusive\n"); 191 return (EINVAL); 192 } 193 mntorflags |= MNT_NFS4ACLS; 194 } 195 196 MNT_ILOCK(mp); 197 mp->mnt_flag |= mntorflags; 198 MNT_IUNLOCK(mp); 199 /* 200 * If updating, check whether changing from read-only to 201 * read/write; if there is no device name, that's all we do. 202 */ 203 if (mp->mnt_flag & MNT_UPDATE) { 204 ump = VFSTOUFS(mp); 205 fs = ump->um_fs; 206 devvp = ump->um_devvp; 207 if (fs->fs_ronly == 0 && 208 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 209 /* 210 * Flush any dirty data and suspend filesystem. 211 */ 212 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 213 return (error); 214 for (;;) { 215 vn_finished_write(mp); 216 if ((error = vfs_write_suspend(mp)) != 0) 217 return (error); 218 MNT_ILOCK(mp); 219 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 220 /* 221 * Allow the secondary writes 222 * to proceed. 223 */ 224 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 225 MNTK_SUSPEND2); 226 wakeup(&mp->mnt_flag); 227 MNT_IUNLOCK(mp); 228 /* 229 * Allow the curthread to 230 * ignore the suspension to 231 * synchronize on-disk state. 232 */ 233 td->td_pflags |= TDP_IGNSUSP; 234 break; 235 } 236 MNT_IUNLOCK(mp); 237 vn_start_write(NULL, &mp, V_WAIT); 238 } 239 /* 240 * Check for and optionally get rid of files open 241 * for writing. 242 */ 243 flags = WRITECLOSE; 244 if (mp->mnt_flag & MNT_FORCE) 245 flags |= FORCECLOSE; 246 if (mp->mnt_flag & MNT_SOFTDEP) { 247 error = softdep_flushfiles(mp, flags, td); 248 } else { 249 error = ffs_flushfiles(mp, flags, td); 250 } 251 if (error) { 252 vfs_write_resume(mp); 253 return (error); 254 } 255 if (fs->fs_pendingblocks != 0 || 256 fs->fs_pendinginodes != 0) { 257 printf("%s: %s: blocks %jd files %d\n", 258 fs->fs_fsmnt, "update error", 259 (intmax_t)fs->fs_pendingblocks, 260 fs->fs_pendinginodes); 261 fs->fs_pendingblocks = 0; 262 fs->fs_pendinginodes = 0; 263 } 264 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 265 fs->fs_clean = 1; 266 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 267 fs->fs_ronly = 0; 268 fs->fs_clean = 0; 269 vfs_write_resume(mp); 270 return (error); 271 } 272 DROP_GIANT(); 273 g_topology_lock(); 274 g_access(ump->um_cp, 0, -1, 0); 275 g_topology_unlock(); 276 PICKUP_GIANT(); 277 fs->fs_ronly = 1; 278 MNT_ILOCK(mp); 279 mp->mnt_flag |= MNT_RDONLY; 280 MNT_IUNLOCK(mp); 281 /* 282 * Allow the writers to note that filesystem 283 * is ro now. 284 */ 285 vfs_write_resume(mp); 286 } 287 if ((mp->mnt_flag & MNT_RELOAD) && 288 (error = ffs_reload(mp, td)) != 0) 289 return (error); 290 if (fs->fs_ronly && 291 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 292 /* 293 * If upgrade to read-write by non-root, then verify 294 * that user has necessary permissions on the device. 295 */ 296 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 297 error = VOP_ACCESS(devvp, VREAD | VWRITE, 298 td->td_ucred, td); 299 if (error) 300 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 301 if (error) { 302 VOP_UNLOCK(devvp, 0); 303 return (error); 304 } 305 VOP_UNLOCK(devvp, 0); 306 fs->fs_flags &= ~FS_UNCLEAN; 307 if (fs->fs_clean == 0) { 308 fs->fs_flags |= FS_UNCLEAN; 309 if ((mp->mnt_flag & MNT_FORCE) || 310 ((fs->fs_flags & 311 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 312 (fs->fs_flags & FS_DOSOFTDEP))) { 313 printf("WARNING: %s was not %s\n", 314 fs->fs_fsmnt, "properly dismounted"); 315 } else { 316 printf( 317 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 318 fs->fs_fsmnt); 319 if (fs->fs_flags & FS_SUJ) 320 printf( 321 "WARNING: Forced mount will invalidate journal contents\n"); 322 return (EPERM); 323 } 324 } 325 DROP_GIANT(); 326 g_topology_lock(); 327 /* 328 * If we're the root device, we may not have an E count 329 * yet, get it now. 330 */ 331 if (ump->um_cp->ace == 0) 332 error = g_access(ump->um_cp, 0, 1, 1); 333 else 334 error = g_access(ump->um_cp, 0, 1, 0); 335 g_topology_unlock(); 336 PICKUP_GIANT(); 337 if (error) 338 return (error); 339 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 340 return (error); 341 fs->fs_ronly = 0; 342 MNT_ILOCK(mp); 343 mp->mnt_flag &= ~MNT_RDONLY; 344 MNT_IUNLOCK(mp); 345 fs->fs_mtime = time_second; 346 /* check to see if we need to start softdep */ 347 if ((fs->fs_flags & FS_DOSOFTDEP) && 348 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 349 vn_finished_write(mp); 350 return (error); 351 } 352 fs->fs_clean = 0; 353 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 354 vn_finished_write(mp); 355 return (error); 356 } 357 if (fs->fs_snapinum[0] != 0) 358 ffs_snapshot_mount(mp); 359 vn_finished_write(mp); 360 } 361 /* 362 * Soft updates is incompatible with "async", 363 * so if we are doing softupdates stop the user 364 * from setting the async flag in an update. 365 * Softdep_mount() clears it in an initial mount 366 * or ro->rw remount. 367 */ 368 if (mp->mnt_flag & MNT_SOFTDEP) { 369 /* XXX: Reset too late ? */ 370 MNT_ILOCK(mp); 371 mp->mnt_flag &= ~MNT_ASYNC; 372 MNT_IUNLOCK(mp); 373 } 374 /* 375 * Keep MNT_ACLS flag if it is stored in superblock. 376 */ 377 if ((fs->fs_flags & FS_ACLS) != 0) { 378 /* XXX: Set too late ? */ 379 MNT_ILOCK(mp); 380 mp->mnt_flag |= MNT_ACLS; 381 MNT_IUNLOCK(mp); 382 } 383 384 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 385 /* XXX: Set too late ? */ 386 MNT_ILOCK(mp); 387 mp->mnt_flag |= MNT_NFS4ACLS; 388 MNT_IUNLOCK(mp); 389 } 390 391 /* 392 * If this is a snapshot request, take the snapshot. 393 */ 394 if (mp->mnt_flag & MNT_SNAPSHOT) 395 return (ffs_snapshot(mp, fspec)); 396 } 397 398 /* 399 * Not an update, or updating the name: look up the name 400 * and verify that it refers to a sensible disk device. 401 */ 402 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 403 if ((error = namei(&ndp)) != 0) 404 return (error); 405 NDFREE(&ndp, NDF_ONLY_PNBUF); 406 devvp = ndp.ni_vp; 407 if (!vn_isdisk(devvp, &error)) { 408 vput(devvp); 409 return (error); 410 } 411 412 /* 413 * If mount by non-root, then verify that user has necessary 414 * permissions on the device. 415 */ 416 accmode = VREAD; 417 if ((mp->mnt_flag & MNT_RDONLY) == 0) 418 accmode |= VWRITE; 419 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 420 if (error) 421 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 422 if (error) { 423 vput(devvp); 424 return (error); 425 } 426 427 if (mp->mnt_flag & MNT_UPDATE) { 428 /* 429 * Update only 430 * 431 * If it's not the same vnode, or at least the same device 432 * then it's not correct. 433 */ 434 435 if (devvp->v_rdev != ump->um_devvp->v_rdev) 436 error = EINVAL; /* needs translation */ 437 vput(devvp); 438 if (error) 439 return (error); 440 } else { 441 /* 442 * New mount 443 * 444 * We need the name for the mount point (also used for 445 * "last mounted on") copied in. If an error occurs, 446 * the mount point is discarded by the upper level code. 447 * Note that vfs_mount() populates f_mntonname for us. 448 */ 449 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 450 vrele(devvp); 451 return (error); 452 } 453 } 454 vfs_mountedfrom(mp, fspec); 455 return (0); 456 } 457 458 /* 459 * Compatibility with old mount system call. 460 */ 461 462 static int 463 ffs_cmount(struct mntarg *ma, void *data, int flags) 464 { 465 struct ufs_args args; 466 struct export_args exp; 467 int error; 468 469 if (data == NULL) 470 return (EINVAL); 471 error = copyin(data, &args, sizeof args); 472 if (error) 473 return (error); 474 vfs_oexport_conv(&args.export, &exp); 475 476 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 477 ma = mount_arg(ma, "export", &exp, sizeof(exp)); 478 error = kernel_mount(ma, flags); 479 480 return (error); 481 } 482 483 /* 484 * Reload all incore data for a filesystem (used after running fsck on 485 * the root filesystem and finding things to fix). The filesystem must 486 * be mounted read-only. 487 * 488 * Things to do to update the mount: 489 * 1) invalidate all cached meta-data. 490 * 2) re-read superblock from disk. 491 * 3) re-read summary information from disk. 492 * 4) invalidate all inactive vnodes. 493 * 5) invalidate all cached file data. 494 * 6) re-read inode data for all active vnodes. 495 */ 496 static int 497 ffs_reload(struct mount *mp, struct thread *td) 498 { 499 struct vnode *vp, *mvp, *devvp; 500 struct inode *ip; 501 void *space; 502 struct buf *bp; 503 struct fs *fs, *newfs; 504 struct ufsmount *ump; 505 ufs2_daddr_t sblockloc; 506 int i, blks, size, error; 507 int32_t *lp; 508 509 if ((mp->mnt_flag & MNT_RDONLY) == 0) 510 return (EINVAL); 511 ump = VFSTOUFS(mp); 512 /* 513 * Step 1: invalidate all cached meta-data. 514 */ 515 devvp = VFSTOUFS(mp)->um_devvp; 516 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 517 if (vinvalbuf(devvp, 0, 0, 0) != 0) 518 panic("ffs_reload: dirty1"); 519 VOP_UNLOCK(devvp, 0); 520 521 /* 522 * Step 2: re-read superblock from disk. 523 */ 524 fs = VFSTOUFS(mp)->um_fs; 525 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 526 NOCRED, &bp)) != 0) 527 return (error); 528 newfs = (struct fs *)bp->b_data; 529 if ((newfs->fs_magic != FS_UFS1_MAGIC && 530 newfs->fs_magic != FS_UFS2_MAGIC) || 531 newfs->fs_bsize > MAXBSIZE || 532 newfs->fs_bsize < sizeof(struct fs)) { 533 brelse(bp); 534 return (EIO); /* XXX needs translation */ 535 } 536 /* 537 * Copy pointer fields back into superblock before copying in XXX 538 * new superblock. These should really be in the ufsmount. XXX 539 * Note that important parameters (eg fs_ncg) are unchanged. 540 */ 541 newfs->fs_csp = fs->fs_csp; 542 newfs->fs_maxcluster = fs->fs_maxcluster; 543 newfs->fs_contigdirs = fs->fs_contigdirs; 544 newfs->fs_active = fs->fs_active; 545 /* The file system is still read-only. */ 546 newfs->fs_ronly = 1; 547 sblockloc = fs->fs_sblockloc; 548 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 549 brelse(bp); 550 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 551 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 552 UFS_LOCK(ump); 553 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 554 printf("%s: reload pending error: blocks %jd files %d\n", 555 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 556 fs->fs_pendinginodes); 557 fs->fs_pendingblocks = 0; 558 fs->fs_pendinginodes = 0; 559 } 560 UFS_UNLOCK(ump); 561 562 /* 563 * Step 3: re-read summary information from disk. 564 */ 565 blks = howmany(fs->fs_cssize, fs->fs_fsize); 566 space = fs->fs_csp; 567 for (i = 0; i < blks; i += fs->fs_frag) { 568 size = fs->fs_bsize; 569 if (i + fs->fs_frag > blks) 570 size = (blks - i) * fs->fs_fsize; 571 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 572 NOCRED, &bp); 573 if (error) 574 return (error); 575 bcopy(bp->b_data, space, (u_int)size); 576 space = (char *)space + size; 577 brelse(bp); 578 } 579 /* 580 * We no longer know anything about clusters per cylinder group. 581 */ 582 if (fs->fs_contigsumsize > 0) { 583 lp = fs->fs_maxcluster; 584 for (i = 0; i < fs->fs_ncg; i++) 585 *lp++ = fs->fs_contigsumsize; 586 } 587 588 loop: 589 MNT_ILOCK(mp); 590 MNT_VNODE_FOREACH(vp, mp, mvp) { 591 VI_LOCK(vp); 592 if (vp->v_iflag & VI_DOOMED) { 593 VI_UNLOCK(vp); 594 continue; 595 } 596 MNT_IUNLOCK(mp); 597 /* 598 * Step 4: invalidate all cached file data. 599 */ 600 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 601 MNT_VNODE_FOREACH_ABORT(mp, mvp); 602 goto loop; 603 } 604 if (vinvalbuf(vp, 0, 0, 0)) 605 panic("ffs_reload: dirty2"); 606 /* 607 * Step 5: re-read inode data for all active vnodes. 608 */ 609 ip = VTOI(vp); 610 error = 611 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 612 (int)fs->fs_bsize, NOCRED, &bp); 613 if (error) { 614 VOP_UNLOCK(vp, 0); 615 vrele(vp); 616 MNT_VNODE_FOREACH_ABORT(mp, mvp); 617 return (error); 618 } 619 ffs_load_inode(bp, ip, fs, ip->i_number); 620 ip->i_effnlink = ip->i_nlink; 621 brelse(bp); 622 VOP_UNLOCK(vp, 0); 623 vrele(vp); 624 MNT_ILOCK(mp); 625 } 626 MNT_IUNLOCK(mp); 627 return (0); 628 } 629 630 /* 631 * Possible superblock locations ordered from most to least likely. 632 */ 633 static int sblock_try[] = SBLOCKSEARCH; 634 635 /* 636 * Common code for mount and mountroot 637 */ 638 static int 639 ffs_mountfs(devvp, mp, td) 640 struct vnode *devvp; 641 struct mount *mp; 642 struct thread *td; 643 { 644 struct ufsmount *ump; 645 struct buf *bp; 646 struct fs *fs; 647 struct cdev *dev; 648 void *space; 649 ufs2_daddr_t sblockloc; 650 int error, i, blks, size, ronly; 651 int32_t *lp; 652 struct ucred *cred; 653 struct g_consumer *cp; 654 struct mount *nmp; 655 656 bp = NULL; 657 ump = NULL; 658 cred = td ? td->td_ucred : NOCRED; 659 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 660 661 dev = devvp->v_rdev; 662 dev_ref(dev); 663 DROP_GIANT(); 664 g_topology_lock(); 665 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 666 667 /* 668 * If we are a root mount, drop the E flag so fsck can do its magic. 669 * We will pick it up again when we remount R/W. 670 */ 671 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 672 error = g_access(cp, 0, 0, -1); 673 g_topology_unlock(); 674 PICKUP_GIANT(); 675 VOP_UNLOCK(devvp, 0); 676 if (error) 677 goto out; 678 if (devvp->v_rdev->si_iosize_max != 0) 679 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 680 if (mp->mnt_iosize_max > MAXPHYS) 681 mp->mnt_iosize_max = MAXPHYS; 682 683 devvp->v_bufobj.bo_ops = &ffs_ops; 684 685 fs = NULL; 686 sblockloc = 0; 687 /* 688 * Try reading the superblock in each of its possible locations. 689 */ 690 for (i = 0; sblock_try[i] != -1; i++) { 691 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 692 error = EINVAL; 693 vfs_mount_error(mp, 694 "Invalid sectorsize %d for superblock size %d", 695 cp->provider->sectorsize, SBLOCKSIZE); 696 goto out; 697 } 698 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE, 699 cred, &bp)) != 0) 700 goto out; 701 fs = (struct fs *)bp->b_data; 702 sblockloc = sblock_try[i]; 703 if ((fs->fs_magic == FS_UFS1_MAGIC || 704 (fs->fs_magic == FS_UFS2_MAGIC && 705 (fs->fs_sblockloc == sblockloc || 706 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 707 fs->fs_bsize <= MAXBSIZE && 708 fs->fs_bsize >= sizeof(struct fs)) 709 break; 710 brelse(bp); 711 bp = NULL; 712 } 713 if (sblock_try[i] == -1) { 714 error = EINVAL; /* XXX needs translation */ 715 goto out; 716 } 717 fs->fs_fmod = 0; 718 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 719 fs->fs_flags &= ~FS_UNCLEAN; 720 if (fs->fs_clean == 0) { 721 fs->fs_flags |= FS_UNCLEAN; 722 if (ronly || (mp->mnt_flag & MNT_FORCE) || 723 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 724 (fs->fs_flags & FS_DOSOFTDEP))) { 725 printf("WARNING: %s was not properly dismounted\n", 726 fs->fs_fsmnt); 727 } else { 728 printf( 729 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 730 fs->fs_fsmnt); 731 if (fs->fs_flags & FS_SUJ) 732 printf( 733 "WARNING: Forced mount will invalidate journal contents\n"); 734 error = EPERM; 735 goto out; 736 } 737 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 738 (mp->mnt_flag & MNT_FORCE)) { 739 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 740 (intmax_t)fs->fs_pendingblocks, 741 fs->fs_pendinginodes); 742 fs->fs_pendingblocks = 0; 743 fs->fs_pendinginodes = 0; 744 } 745 } 746 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 747 printf("%s: mount pending error: blocks %jd files %d\n", 748 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 749 fs->fs_pendinginodes); 750 fs->fs_pendingblocks = 0; 751 fs->fs_pendinginodes = 0; 752 } 753 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 754 #ifdef UFS_GJOURNAL 755 /* 756 * Get journal provider name. 757 */ 758 size = 1024; 759 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK); 760 if (g_io_getattr("GJOURNAL::provider", cp, &size, 761 mp->mnt_gjprovider) == 0) { 762 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size, 763 M_UFSMNT, M_WAITOK); 764 MNT_ILOCK(mp); 765 mp->mnt_flag |= MNT_GJOURNAL; 766 MNT_IUNLOCK(mp); 767 } else { 768 printf( 769 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n", 770 mp->mnt_stat.f_mntonname); 771 free(mp->mnt_gjprovider, M_UFSMNT); 772 mp->mnt_gjprovider = NULL; 773 } 774 #else 775 printf( 776 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n", 777 mp->mnt_stat.f_mntonname); 778 #endif 779 } else { 780 mp->mnt_gjprovider = NULL; 781 } 782 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 783 ump->um_cp = cp; 784 ump->um_bo = &devvp->v_bufobj; 785 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 786 if (fs->fs_magic == FS_UFS1_MAGIC) { 787 ump->um_fstype = UFS1; 788 ump->um_balloc = ffs_balloc_ufs1; 789 } else { 790 ump->um_fstype = UFS2; 791 ump->um_balloc = ffs_balloc_ufs2; 792 } 793 ump->um_blkatoff = ffs_blkatoff; 794 ump->um_truncate = ffs_truncate; 795 ump->um_update = ffs_update; 796 ump->um_valloc = ffs_valloc; 797 ump->um_vfree = ffs_vfree; 798 ump->um_ifree = ffs_ifree; 799 ump->um_rdonly = ffs_rdonly; 800 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 801 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 802 if (fs->fs_sbsize < SBLOCKSIZE) 803 bp->b_flags |= B_INVAL | B_NOCACHE; 804 brelse(bp); 805 bp = NULL; 806 fs = ump->um_fs; 807 ffs_oldfscompat_read(fs, ump, sblockloc); 808 fs->fs_ronly = ronly; 809 size = fs->fs_cssize; 810 blks = howmany(size, fs->fs_fsize); 811 if (fs->fs_contigsumsize > 0) 812 size += fs->fs_ncg * sizeof(int32_t); 813 size += fs->fs_ncg * sizeof(u_int8_t); 814 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 815 fs->fs_csp = space; 816 for (i = 0; i < blks; i += fs->fs_frag) { 817 size = fs->fs_bsize; 818 if (i + fs->fs_frag > blks) 819 size = (blks - i) * fs->fs_fsize; 820 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 821 cred, &bp)) != 0) { 822 free(fs->fs_csp, M_UFSMNT); 823 goto out; 824 } 825 bcopy(bp->b_data, space, (u_int)size); 826 space = (char *)space + size; 827 brelse(bp); 828 bp = NULL; 829 } 830 if (fs->fs_contigsumsize > 0) { 831 fs->fs_maxcluster = lp = space; 832 for (i = 0; i < fs->fs_ncg; i++) 833 *lp++ = fs->fs_contigsumsize; 834 space = lp; 835 } 836 size = fs->fs_ncg * sizeof(u_int8_t); 837 fs->fs_contigdirs = (u_int8_t *)space; 838 bzero(fs->fs_contigdirs, size); 839 fs->fs_active = NULL; 840 mp->mnt_data = ump; 841 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 842 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 843 nmp = NULL; 844 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 845 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 846 if (nmp) 847 vfs_rel(nmp); 848 vfs_getnewfsid(mp); 849 } 850 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 851 MNT_ILOCK(mp); 852 mp->mnt_flag |= MNT_LOCAL; 853 MNT_IUNLOCK(mp); 854 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 855 #ifdef MAC 856 MNT_ILOCK(mp); 857 mp->mnt_flag |= MNT_MULTILABEL; 858 MNT_IUNLOCK(mp); 859 #else 860 printf( 861 "WARNING: %s: multilabel flag on fs but no MAC support\n", 862 mp->mnt_stat.f_mntonname); 863 #endif 864 } 865 if ((fs->fs_flags & FS_ACLS) != 0) { 866 #ifdef UFS_ACL 867 MNT_ILOCK(mp); 868 869 if (mp->mnt_flag & MNT_NFS4ACLS) 870 printf("WARNING: ACLs flag on fs conflicts with " 871 "\"nfsv4acls\" mount option; option ignored\n"); 872 mp->mnt_flag &= ~MNT_NFS4ACLS; 873 mp->mnt_flag |= MNT_ACLS; 874 875 MNT_IUNLOCK(mp); 876 #else 877 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 878 mp->mnt_stat.f_mntonname); 879 #endif 880 } 881 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 882 #ifdef UFS_ACL 883 MNT_ILOCK(mp); 884 885 if (mp->mnt_flag & MNT_ACLS) 886 printf("WARNING: NFSv4 ACLs flag on fs conflicts with " 887 "\"acls\" mount option; option ignored\n"); 888 mp->mnt_flag &= ~MNT_ACLS; 889 mp->mnt_flag |= MNT_NFS4ACLS; 890 891 MNT_IUNLOCK(mp); 892 #else 893 printf( 894 "WARNING: %s: NFSv4 ACLs flag on fs but no ACLs support\n", 895 mp->mnt_stat.f_mntonname); 896 #endif 897 } 898 899 ump->um_mountp = mp; 900 ump->um_dev = dev; 901 ump->um_devvp = devvp; 902 ump->um_nindir = fs->fs_nindir; 903 ump->um_bptrtodb = fs->fs_fsbtodb; 904 ump->um_seqinc = fs->fs_frag; 905 for (i = 0; i < MAXQUOTAS; i++) 906 ump->um_quotas[i] = NULLVP; 907 #ifdef UFS_EXTATTR 908 ufs_extattr_uepm_init(&ump->um_extattr); 909 #endif 910 /* 911 * Set FS local "last mounted on" information (NULL pad) 912 */ 913 bzero(fs->fs_fsmnt, MAXMNTLEN); 914 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 915 mp->mnt_stat.f_iosize = fs->fs_bsize; 916 917 if( mp->mnt_flag & MNT_ROOTFS) { 918 /* 919 * Root mount; update timestamp in mount structure. 920 * this will be used by the common root mount code 921 * to update the system clock. 922 */ 923 mp->mnt_time = fs->fs_time; 924 } 925 926 if (ronly == 0) { 927 fs->fs_mtime = time_second; 928 if ((fs->fs_flags & FS_DOSOFTDEP) && 929 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 930 free(fs->fs_csp, M_UFSMNT); 931 goto out; 932 } 933 if (fs->fs_snapinum[0] != 0) 934 ffs_snapshot_mount(mp); 935 fs->fs_fmod = 1; 936 fs->fs_clean = 0; 937 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 938 } 939 /* 940 * Initialize filesystem stat information in mount struct. 941 */ 942 MNT_ILOCK(mp); 943 mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED | 944 MNTK_EXTENDED_SHARED; 945 MNT_IUNLOCK(mp); 946 #ifdef UFS_EXTATTR 947 #ifdef UFS_EXTATTR_AUTOSTART 948 /* 949 * 950 * Auto-starting does the following: 951 * - check for /.attribute in the fs, and extattr_start if so 952 * - for each file in .attribute, enable that file with 953 * an attribute of the same name. 954 * Not clear how to report errors -- probably eat them. 955 * This would all happen while the filesystem was busy/not 956 * available, so would effectively be "atomic". 957 */ 958 (void) ufs_extattr_autostart(mp, td); 959 #endif /* !UFS_EXTATTR_AUTOSTART */ 960 #endif /* !UFS_EXTATTR */ 961 return (0); 962 out: 963 if (bp) 964 brelse(bp); 965 if (cp != NULL) { 966 DROP_GIANT(); 967 g_topology_lock(); 968 g_vfs_close(cp); 969 g_topology_unlock(); 970 PICKUP_GIANT(); 971 } 972 if (ump) { 973 mtx_destroy(UFS_MTX(ump)); 974 if (mp->mnt_gjprovider != NULL) { 975 free(mp->mnt_gjprovider, M_UFSMNT); 976 mp->mnt_gjprovider = NULL; 977 } 978 free(ump->um_fs, M_UFSMNT); 979 free(ump, M_UFSMNT); 980 mp->mnt_data = NULL; 981 } 982 dev_rel(dev); 983 return (error); 984 } 985 986 #include <sys/sysctl.h> 987 static int bigcgs = 0; 988 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 989 990 /* 991 * Sanity checks for loading old filesystem superblocks. 992 * See ffs_oldfscompat_write below for unwound actions. 993 * 994 * XXX - Parts get retired eventually. 995 * Unfortunately new bits get added. 996 */ 997 static void 998 ffs_oldfscompat_read(fs, ump, sblockloc) 999 struct fs *fs; 1000 struct ufsmount *ump; 1001 ufs2_daddr_t sblockloc; 1002 { 1003 off_t maxfilesize; 1004 1005 /* 1006 * If not yet done, update fs_flags location and value of fs_sblockloc. 1007 */ 1008 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1009 fs->fs_flags = fs->fs_old_flags; 1010 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1011 fs->fs_sblockloc = sblockloc; 1012 } 1013 /* 1014 * If not yet done, update UFS1 superblock with new wider fields. 1015 */ 1016 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1017 fs->fs_maxbsize = fs->fs_bsize; 1018 fs->fs_time = fs->fs_old_time; 1019 fs->fs_size = fs->fs_old_size; 1020 fs->fs_dsize = fs->fs_old_dsize; 1021 fs->fs_csaddr = fs->fs_old_csaddr; 1022 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1023 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1024 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1025 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1026 } 1027 if (fs->fs_magic == FS_UFS1_MAGIC && 1028 fs->fs_old_inodefmt < FS_44INODEFMT) { 1029 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1030 fs->fs_qbmask = ~fs->fs_bmask; 1031 fs->fs_qfmask = ~fs->fs_fmask; 1032 } 1033 if (fs->fs_magic == FS_UFS1_MAGIC) { 1034 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1035 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1036 if (fs->fs_maxfilesize > maxfilesize) 1037 fs->fs_maxfilesize = maxfilesize; 1038 } 1039 /* Compatibility for old filesystems */ 1040 if (fs->fs_avgfilesize <= 0) 1041 fs->fs_avgfilesize = AVFILESIZ; 1042 if (fs->fs_avgfpdir <= 0) 1043 fs->fs_avgfpdir = AFPDIR; 1044 if (bigcgs) { 1045 fs->fs_save_cgsize = fs->fs_cgsize; 1046 fs->fs_cgsize = fs->fs_bsize; 1047 } 1048 } 1049 1050 /* 1051 * Unwinding superblock updates for old filesystems. 1052 * See ffs_oldfscompat_read above for details. 1053 * 1054 * XXX - Parts get retired eventually. 1055 * Unfortunately new bits get added. 1056 */ 1057 void 1058 ffs_oldfscompat_write(fs, ump) 1059 struct fs *fs; 1060 struct ufsmount *ump; 1061 { 1062 1063 /* 1064 * Copy back UFS2 updated fields that UFS1 inspects. 1065 */ 1066 if (fs->fs_magic == FS_UFS1_MAGIC) { 1067 fs->fs_old_time = fs->fs_time; 1068 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1069 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1070 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1071 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1072 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1073 } 1074 if (bigcgs) { 1075 fs->fs_cgsize = fs->fs_save_cgsize; 1076 fs->fs_save_cgsize = 0; 1077 } 1078 } 1079 1080 /* 1081 * unmount system call 1082 */ 1083 static int 1084 ffs_unmount(mp, mntflags) 1085 struct mount *mp; 1086 int mntflags; 1087 { 1088 struct thread *td; 1089 struct ufsmount *ump = VFSTOUFS(mp); 1090 struct fs *fs; 1091 int error, flags, susp; 1092 #ifdef UFS_EXTATTR 1093 int e_restart; 1094 #endif 1095 1096 flags = 0; 1097 td = curthread; 1098 fs = ump->um_fs; 1099 if (mntflags & MNT_FORCE) { 1100 flags |= FORCECLOSE; 1101 susp = fs->fs_ronly != 0; 1102 } else 1103 susp = 0; 1104 #ifdef UFS_EXTATTR 1105 if ((error = ufs_extattr_stop(mp, td))) { 1106 if (error != EOPNOTSUPP) 1107 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 1108 error); 1109 e_restart = 0; 1110 } else { 1111 ufs_extattr_uepm_destroy(&ump->um_extattr); 1112 e_restart = 1; 1113 } 1114 #endif 1115 if (susp) { 1116 /* 1117 * dounmount already called vn_start_write(). 1118 */ 1119 for (;;) { 1120 vn_finished_write(mp); 1121 if ((error = vfs_write_suspend(mp)) != 0) 1122 return (error); 1123 MNT_ILOCK(mp); 1124 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 1125 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 1126 MNTK_SUSPEND2); 1127 wakeup(&mp->mnt_flag); 1128 MNT_IUNLOCK(mp); 1129 td->td_pflags |= TDP_IGNSUSP; 1130 break; 1131 } 1132 MNT_IUNLOCK(mp); 1133 vn_start_write(NULL, &mp, V_WAIT); 1134 } 1135 } 1136 if (mp->mnt_flag & MNT_SOFTDEP) 1137 error = softdep_flushfiles(mp, flags, td); 1138 else 1139 error = ffs_flushfiles(mp, flags, td); 1140 if (error != 0 && error != ENXIO) 1141 goto fail; 1142 1143 UFS_LOCK(ump); 1144 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1145 printf("%s: unmount pending error: blocks %jd files %d\n", 1146 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1147 fs->fs_pendinginodes); 1148 fs->fs_pendingblocks = 0; 1149 fs->fs_pendinginodes = 0; 1150 } 1151 UFS_UNLOCK(ump); 1152 softdep_unmount(mp); 1153 if (fs->fs_ronly == 0) { 1154 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1155 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1156 if (error && error != ENXIO) { 1157 fs->fs_clean = 0; 1158 goto fail; 1159 } 1160 } 1161 if (susp) { 1162 vfs_write_resume(mp); 1163 vn_start_write(NULL, &mp, V_WAIT); 1164 } 1165 DROP_GIANT(); 1166 g_topology_lock(); 1167 g_vfs_close(ump->um_cp); 1168 g_topology_unlock(); 1169 PICKUP_GIANT(); 1170 vrele(ump->um_devvp); 1171 dev_rel(ump->um_dev); 1172 mtx_destroy(UFS_MTX(ump)); 1173 if (mp->mnt_gjprovider != NULL) { 1174 free(mp->mnt_gjprovider, M_UFSMNT); 1175 mp->mnt_gjprovider = NULL; 1176 } 1177 free(fs->fs_csp, M_UFSMNT); 1178 free(fs, M_UFSMNT); 1179 free(ump, M_UFSMNT); 1180 mp->mnt_data = NULL; 1181 MNT_ILOCK(mp); 1182 mp->mnt_flag &= ~MNT_LOCAL; 1183 MNT_IUNLOCK(mp); 1184 return (error); 1185 1186 fail: 1187 if (susp) { 1188 vfs_write_resume(mp); 1189 vn_start_write(NULL, &mp, V_WAIT); 1190 } 1191 #ifdef UFS_EXTATTR 1192 if (e_restart) { 1193 ufs_extattr_uepm_init(&ump->um_extattr); 1194 #ifdef UFS_EXTATTR_AUTOSTART 1195 (void) ufs_extattr_autostart(mp, td); 1196 #endif 1197 } 1198 #endif 1199 1200 return (error); 1201 } 1202 1203 /* 1204 * Flush out all the files in a filesystem. 1205 */ 1206 int 1207 ffs_flushfiles(mp, flags, td) 1208 struct mount *mp; 1209 int flags; 1210 struct thread *td; 1211 { 1212 struct ufsmount *ump; 1213 int error; 1214 1215 ump = VFSTOUFS(mp); 1216 #ifdef QUOTA 1217 if (mp->mnt_flag & MNT_QUOTA) { 1218 int i; 1219 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1220 if (error) 1221 return (error); 1222 for (i = 0; i < MAXQUOTAS; i++) { 1223 quotaoff(td, mp, i); 1224 } 1225 /* 1226 * Here we fall through to vflush again to ensure 1227 * that we have gotten rid of all the system vnodes. 1228 */ 1229 } 1230 #endif 1231 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1232 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1233 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1234 return (error); 1235 ffs_snapshot_unmount(mp); 1236 flags |= FORCECLOSE; 1237 /* 1238 * Here we fall through to vflush again to ensure 1239 * that we have gotten rid of all the system vnodes. 1240 */ 1241 } 1242 /* 1243 * Flush all the files. 1244 */ 1245 if ((error = vflush(mp, 0, flags, td)) != 0) 1246 return (error); 1247 /* 1248 * Flush filesystem metadata. 1249 */ 1250 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1251 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1252 VOP_UNLOCK(ump->um_devvp, 0); 1253 return (error); 1254 } 1255 1256 /* 1257 * Get filesystem statistics. 1258 */ 1259 static int 1260 ffs_statfs(mp, sbp) 1261 struct mount *mp; 1262 struct statfs *sbp; 1263 { 1264 struct ufsmount *ump; 1265 struct fs *fs; 1266 1267 ump = VFSTOUFS(mp); 1268 fs = ump->um_fs; 1269 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1270 panic("ffs_statfs"); 1271 sbp->f_version = STATFS_VERSION; 1272 sbp->f_bsize = fs->fs_fsize; 1273 sbp->f_iosize = fs->fs_bsize; 1274 sbp->f_blocks = fs->fs_dsize; 1275 UFS_LOCK(ump); 1276 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1277 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1278 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1279 dbtofsb(fs, fs->fs_pendingblocks); 1280 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1281 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1282 UFS_UNLOCK(ump); 1283 sbp->f_namemax = NAME_MAX; 1284 return (0); 1285 } 1286 1287 /* 1288 * Go through the disk queues to initiate sandbagged IO; 1289 * go through the inodes to write those that have been modified; 1290 * initiate the writing of the super block if it has been modified. 1291 * 1292 * Note: we are always called with the filesystem marked `MPBUSY'. 1293 */ 1294 static int 1295 ffs_sync(mp, waitfor) 1296 struct mount *mp; 1297 int waitfor; 1298 { 1299 struct vnode *mvp, *vp, *devvp; 1300 struct thread *td; 1301 struct inode *ip; 1302 struct ufsmount *ump = VFSTOUFS(mp); 1303 struct fs *fs; 1304 int error, count, wait, lockreq, allerror = 0; 1305 int suspend; 1306 int suspended; 1307 int secondary_writes; 1308 int secondary_accwrites; 1309 int softdep_deps; 1310 int softdep_accdeps; 1311 struct bufobj *bo; 1312 1313 td = curthread; 1314 fs = ump->um_fs; 1315 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1316 printf("fs = %s\n", fs->fs_fsmnt); 1317 panic("ffs_sync: rofs mod"); 1318 } 1319 /* 1320 * Write back each (modified) inode. 1321 */ 1322 wait = 0; 1323 suspend = 0; 1324 suspended = 0; 1325 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1326 if (waitfor == MNT_SUSPEND) { 1327 suspend = 1; 1328 waitfor = MNT_WAIT; 1329 } 1330 if (waitfor == MNT_WAIT) { 1331 wait = 1; 1332 lockreq = LK_EXCLUSIVE; 1333 } 1334 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1335 MNT_ILOCK(mp); 1336 loop: 1337 /* Grab snapshot of secondary write counts */ 1338 secondary_writes = mp->mnt_secondary_writes; 1339 secondary_accwrites = mp->mnt_secondary_accwrites; 1340 1341 /* Grab snapshot of softdep dependency counts */ 1342 MNT_IUNLOCK(mp); 1343 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1344 MNT_ILOCK(mp); 1345 1346 MNT_VNODE_FOREACH(vp, mp, mvp) { 1347 /* 1348 * Depend on the mntvnode_slock to keep things stable enough 1349 * for a quick test. Since there might be hundreds of 1350 * thousands of vnodes, we cannot afford even a subroutine 1351 * call unless there's a good chance that we have work to do. 1352 */ 1353 VI_LOCK(vp); 1354 if (vp->v_iflag & VI_DOOMED) { 1355 VI_UNLOCK(vp); 1356 continue; 1357 } 1358 ip = VTOI(vp); 1359 if (vp->v_type == VNON || ((ip->i_flag & 1360 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1361 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1362 VI_UNLOCK(vp); 1363 continue; 1364 } 1365 MNT_IUNLOCK(mp); 1366 if ((error = vget(vp, lockreq, td)) != 0) { 1367 MNT_ILOCK(mp); 1368 if (error == ENOENT || error == ENOLCK) { 1369 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1370 goto loop; 1371 } 1372 continue; 1373 } 1374 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1375 allerror = error; 1376 vput(vp); 1377 MNT_ILOCK(mp); 1378 } 1379 MNT_IUNLOCK(mp); 1380 /* 1381 * Force stale filesystem control information to be flushed. 1382 */ 1383 if (waitfor == MNT_WAIT) { 1384 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1385 allerror = error; 1386 /* Flushed work items may create new vnodes to clean */ 1387 if (allerror == 0 && count) { 1388 MNT_ILOCK(mp); 1389 goto loop; 1390 } 1391 } 1392 #ifdef QUOTA 1393 qsync(mp); 1394 #endif 1395 devvp = ump->um_devvp; 1396 bo = &devvp->v_bufobj; 1397 BO_LOCK(bo); 1398 if (waitfor != MNT_LAZY && 1399 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1400 BO_UNLOCK(bo); 1401 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1402 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1403 allerror = error; 1404 VOP_UNLOCK(devvp, 0); 1405 if (allerror == 0 && waitfor == MNT_WAIT) { 1406 MNT_ILOCK(mp); 1407 goto loop; 1408 } 1409 } else if (suspend != 0) { 1410 if (softdep_check_suspend(mp, 1411 devvp, 1412 softdep_deps, 1413 softdep_accdeps, 1414 secondary_writes, 1415 secondary_accwrites) != 0) 1416 goto loop; /* More work needed */ 1417 mtx_assert(MNT_MTX(mp), MA_OWNED); 1418 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1419 MNT_IUNLOCK(mp); 1420 suspended = 1; 1421 } else 1422 BO_UNLOCK(bo); 1423 /* 1424 * Write back modified superblock. 1425 */ 1426 if (fs->fs_fmod != 0 && 1427 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1428 allerror = error; 1429 return (allerror); 1430 } 1431 1432 int 1433 ffs_vget(mp, ino, flags, vpp) 1434 struct mount *mp; 1435 ino_t ino; 1436 int flags; 1437 struct vnode **vpp; 1438 { 1439 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1440 } 1441 1442 int 1443 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1444 struct mount *mp; 1445 ino_t ino; 1446 int flags; 1447 struct vnode **vpp; 1448 int ffs_flags; 1449 { 1450 struct fs *fs; 1451 struct inode *ip; 1452 struct ufsmount *ump; 1453 struct buf *bp; 1454 struct vnode *vp; 1455 struct cdev *dev; 1456 int error; 1457 1458 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1459 if (error || *vpp != NULL) 1460 return (error); 1461 1462 /* 1463 * We must promote to an exclusive lock for vnode creation. This 1464 * can happen if lookup is passed LOCKSHARED. 1465 */ 1466 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1467 flags &= ~LK_TYPE_MASK; 1468 flags |= LK_EXCLUSIVE; 1469 } 1470 1471 /* 1472 * We do not lock vnode creation as it is believed to be too 1473 * expensive for such rare case as simultaneous creation of vnode 1474 * for same ino by different processes. We just allow them to race 1475 * and check later to decide who wins. Let the race begin! 1476 */ 1477 1478 ump = VFSTOUFS(mp); 1479 dev = ump->um_dev; 1480 fs = ump->um_fs; 1481 1482 /* 1483 * If this malloc() is performed after the getnewvnode() 1484 * it might block, leaving a vnode with a NULL v_data to be 1485 * found by ffs_sync() if a sync happens to fire right then, 1486 * which will cause a panic because ffs_sync() blindly 1487 * dereferences vp->v_data (as well it should). 1488 */ 1489 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1490 1491 /* Allocate a new vnode/inode. */ 1492 if (fs->fs_magic == FS_UFS1_MAGIC) 1493 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1494 else 1495 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1496 if (error) { 1497 *vpp = NULL; 1498 uma_zfree(uma_inode, ip); 1499 return (error); 1500 } 1501 /* 1502 * FFS supports recursive locking. 1503 */ 1504 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1505 VN_LOCK_AREC(vp); 1506 vp->v_data = ip; 1507 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1508 ip->i_vnode = vp; 1509 ip->i_ump = ump; 1510 ip->i_fs = fs; 1511 ip->i_dev = dev; 1512 ip->i_number = ino; 1513 ip->i_ea_refs = 0; 1514 #ifdef QUOTA 1515 { 1516 int i; 1517 for (i = 0; i < MAXQUOTAS; i++) 1518 ip->i_dquot[i] = NODQUOT; 1519 } 1520 #endif 1521 1522 if (ffs_flags & FFSV_FORCEINSMQ) 1523 vp->v_vflag |= VV_FORCEINSMQ; 1524 error = insmntque(vp, mp); 1525 if (error != 0) { 1526 uma_zfree(uma_inode, ip); 1527 *vpp = NULL; 1528 return (error); 1529 } 1530 vp->v_vflag &= ~VV_FORCEINSMQ; 1531 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1532 if (error || *vpp != NULL) 1533 return (error); 1534 1535 /* Read in the disk contents for the inode, copy into the inode. */ 1536 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1537 (int)fs->fs_bsize, NOCRED, &bp); 1538 if (error) { 1539 /* 1540 * The inode does not contain anything useful, so it would 1541 * be misleading to leave it on its hash chain. With mode 1542 * still zero, it will be unlinked and returned to the free 1543 * list by vput(). 1544 */ 1545 brelse(bp); 1546 vput(vp); 1547 *vpp = NULL; 1548 return (error); 1549 } 1550 if (ip->i_ump->um_fstype == UFS1) 1551 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1552 else 1553 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1554 ffs_load_inode(bp, ip, fs, ino); 1555 if (DOINGSOFTDEP(vp)) 1556 softdep_load_inodeblock(ip); 1557 else 1558 ip->i_effnlink = ip->i_nlink; 1559 bqrelse(bp); 1560 1561 /* 1562 * Initialize the vnode from the inode, check for aliases. 1563 * Note that the underlying vnode may have changed. 1564 */ 1565 if (ip->i_ump->um_fstype == UFS1) 1566 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1567 else 1568 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1569 if (error) { 1570 vput(vp); 1571 *vpp = NULL; 1572 return (error); 1573 } 1574 1575 /* 1576 * Finish inode initialization. 1577 */ 1578 if (vp->v_type != VFIFO) { 1579 /* FFS supports shared locking for all files except fifos. */ 1580 VN_LOCK_ASHARE(vp); 1581 } 1582 1583 /* 1584 * Set up a generation number for this inode if it does not 1585 * already have one. This should only happen on old filesystems. 1586 */ 1587 if (ip->i_gen == 0) { 1588 ip->i_gen = arc4random() / 2 + 1; 1589 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1590 ip->i_flag |= IN_MODIFIED; 1591 DIP_SET(ip, i_gen, ip->i_gen); 1592 } 1593 } 1594 #ifdef MAC 1595 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1596 /* 1597 * If this vnode is already allocated, and we're running 1598 * multi-label, attempt to perform a label association 1599 * from the extended attributes on the inode. 1600 */ 1601 error = mac_vnode_associate_extattr(mp, vp); 1602 if (error) { 1603 /* ufs_inactive will release ip->i_devvp ref. */ 1604 vput(vp); 1605 *vpp = NULL; 1606 return (error); 1607 } 1608 } 1609 #endif 1610 1611 *vpp = vp; 1612 return (0); 1613 } 1614 1615 /* 1616 * File handle to vnode 1617 * 1618 * Have to be really careful about stale file handles: 1619 * - check that the inode number is valid 1620 * - call ffs_vget() to get the locked inode 1621 * - check for an unallocated inode (i_mode == 0) 1622 * - check that the given client host has export rights and return 1623 * those rights via. exflagsp and credanonp 1624 */ 1625 static int 1626 ffs_fhtovp(mp, fhp, vpp) 1627 struct mount *mp; 1628 struct fid *fhp; 1629 struct vnode **vpp; 1630 { 1631 struct ufid *ufhp; 1632 struct fs *fs; 1633 1634 ufhp = (struct ufid *)fhp; 1635 fs = VFSTOUFS(mp)->um_fs; 1636 if (ufhp->ufid_ino < ROOTINO || 1637 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1638 return (ESTALE); 1639 return (ufs_fhtovp(mp, ufhp, vpp)); 1640 } 1641 1642 /* 1643 * Initialize the filesystem. 1644 */ 1645 static int 1646 ffs_init(vfsp) 1647 struct vfsconf *vfsp; 1648 { 1649 1650 softdep_initialize(); 1651 return (ufs_init(vfsp)); 1652 } 1653 1654 /* 1655 * Undo the work of ffs_init(). 1656 */ 1657 static int 1658 ffs_uninit(vfsp) 1659 struct vfsconf *vfsp; 1660 { 1661 int ret; 1662 1663 ret = ufs_uninit(vfsp); 1664 softdep_uninitialize(); 1665 return (ret); 1666 } 1667 1668 /* 1669 * Write a superblock and associated information back to disk. 1670 */ 1671 int 1672 ffs_sbupdate(mp, waitfor, suspended) 1673 struct ufsmount *mp; 1674 int waitfor; 1675 int suspended; 1676 { 1677 struct fs *fs = mp->um_fs; 1678 struct buf *sbbp; 1679 struct buf *bp; 1680 int blks; 1681 void *space; 1682 int i, size, error, allerror = 0; 1683 1684 if (fs->fs_ronly == 1 && 1685 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1686 (MNT_RDONLY | MNT_UPDATE)) 1687 panic("ffs_sbupdate: write read-only filesystem"); 1688 /* 1689 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1690 */ 1691 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1692 0, 0, 0); 1693 /* 1694 * First write back the summary information. 1695 */ 1696 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1697 space = fs->fs_csp; 1698 for (i = 0; i < blks; i += fs->fs_frag) { 1699 size = fs->fs_bsize; 1700 if (i + fs->fs_frag > blks) 1701 size = (blks - i) * fs->fs_fsize; 1702 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1703 size, 0, 0, 0); 1704 bcopy(space, bp->b_data, (u_int)size); 1705 space = (char *)space + size; 1706 if (suspended) 1707 bp->b_flags |= B_VALIDSUSPWRT; 1708 if (waitfor != MNT_WAIT) 1709 bawrite(bp); 1710 else if ((error = bwrite(bp)) != 0) 1711 allerror = error; 1712 } 1713 /* 1714 * Now write back the superblock itself. If any errors occurred 1715 * up to this point, then fail so that the superblock avoids 1716 * being written out as clean. 1717 */ 1718 if (allerror) { 1719 brelse(sbbp); 1720 return (allerror); 1721 } 1722 bp = sbbp; 1723 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1724 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1725 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1726 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1727 fs->fs_sblockloc = SBLOCK_UFS1; 1728 } 1729 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1730 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1731 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1732 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1733 fs->fs_sblockloc = SBLOCK_UFS2; 1734 } 1735 fs->fs_fmod = 0; 1736 fs->fs_time = time_second; 1737 if (fs->fs_flags & FS_DOSOFTDEP) 1738 softdep_setup_sbupdate(mp, (struct fs *)bp->b_data, bp); 1739 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1740 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1741 if (suspended) 1742 bp->b_flags |= B_VALIDSUSPWRT; 1743 if (waitfor != MNT_WAIT) 1744 bawrite(bp); 1745 else if ((error = bwrite(bp)) != 0) 1746 allerror = error; 1747 return (allerror); 1748 } 1749 1750 static int 1751 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1752 int attrnamespace, const char *attrname) 1753 { 1754 1755 #ifdef UFS_EXTATTR 1756 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1757 attrname)); 1758 #else 1759 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1760 attrname)); 1761 #endif 1762 } 1763 1764 static void 1765 ffs_ifree(struct ufsmount *ump, struct inode *ip) 1766 { 1767 1768 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1769 uma_zfree(uma_ufs1, ip->i_din1); 1770 else if (ip->i_din2 != NULL) 1771 uma_zfree(uma_ufs2, ip->i_din2); 1772 uma_zfree(uma_inode, ip); 1773 } 1774 1775 static int dobkgrdwrite = 1; 1776 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1777 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1778 1779 /* 1780 * Complete a background write started from bwrite. 1781 */ 1782 static void 1783 ffs_backgroundwritedone(struct buf *bp) 1784 { 1785 struct bufobj *bufobj; 1786 struct buf *origbp; 1787 1788 /* 1789 * Find the original buffer that we are writing. 1790 */ 1791 bufobj = bp->b_bufobj; 1792 BO_LOCK(bufobj); 1793 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1794 panic("backgroundwritedone: lost buffer"); 1795 /* Grab an extra reference to be dropped by the bufdone() below. */ 1796 bufobj_wrefl(bufobj); 1797 BO_UNLOCK(bufobj); 1798 /* 1799 * Process dependencies then return any unfinished ones. 1800 */ 1801 if (!LIST_EMPTY(&bp->b_dep)) 1802 buf_complete(bp); 1803 #ifdef SOFTUPDATES 1804 if (!LIST_EMPTY(&bp->b_dep)) 1805 softdep_move_dependencies(bp, origbp); 1806 #endif 1807 /* 1808 * This buffer is marked B_NOCACHE so when it is released 1809 * by biodone it will be tossed. 1810 */ 1811 bp->b_flags |= B_NOCACHE; 1812 bp->b_flags &= ~B_CACHE; 1813 bufdone(bp); 1814 BO_LOCK(bufobj); 1815 /* 1816 * Clear the BV_BKGRDINPROG flag in the original buffer 1817 * and awaken it if it is waiting for the write to complete. 1818 * If BV_BKGRDINPROG is not set in the original buffer it must 1819 * have been released and re-instantiated - which is not legal. 1820 */ 1821 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1822 ("backgroundwritedone: lost buffer2")); 1823 origbp->b_vflags &= ~BV_BKGRDINPROG; 1824 if (origbp->b_vflags & BV_BKGRDWAIT) { 1825 origbp->b_vflags &= ~BV_BKGRDWAIT; 1826 wakeup(&origbp->b_xflags); 1827 } 1828 BO_UNLOCK(bufobj); 1829 } 1830 1831 1832 /* 1833 * Write, release buffer on completion. (Done by iodone 1834 * if async). Do not bother writing anything if the buffer 1835 * is invalid. 1836 * 1837 * Note that we set B_CACHE here, indicating that buffer is 1838 * fully valid and thus cacheable. This is true even of NFS 1839 * now so we set it generally. This could be set either here 1840 * or in biodone() since the I/O is synchronous. We put it 1841 * here. 1842 */ 1843 static int 1844 ffs_bufwrite(struct buf *bp) 1845 { 1846 int oldflags, s; 1847 struct buf *newbp; 1848 1849 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1850 if (bp->b_flags & B_INVAL) { 1851 brelse(bp); 1852 return (0); 1853 } 1854 1855 oldflags = bp->b_flags; 1856 1857 if (!BUF_ISLOCKED(bp)) 1858 panic("bufwrite: buffer is not busy???"); 1859 s = splbio(); 1860 /* 1861 * If a background write is already in progress, delay 1862 * writing this block if it is asynchronous. Otherwise 1863 * wait for the background write to complete. 1864 */ 1865 BO_LOCK(bp->b_bufobj); 1866 if (bp->b_vflags & BV_BKGRDINPROG) { 1867 if (bp->b_flags & B_ASYNC) { 1868 BO_UNLOCK(bp->b_bufobj); 1869 splx(s); 1870 bdwrite(bp); 1871 return (0); 1872 } 1873 bp->b_vflags |= BV_BKGRDWAIT; 1874 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1875 if (bp->b_vflags & BV_BKGRDINPROG) 1876 panic("bufwrite: still writing"); 1877 } 1878 BO_UNLOCK(bp->b_bufobj); 1879 1880 /* 1881 * If this buffer is marked for background writing and we 1882 * do not have to wait for it, make a copy and write the 1883 * copy so as to leave this buffer ready for further use. 1884 * 1885 * This optimization eats a lot of memory. If we have a page 1886 * or buffer shortfall we can't do it. 1887 */ 1888 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1889 (bp->b_flags & B_ASYNC) && 1890 !vm_page_count_severe() && 1891 !buf_dirty_count_severe()) { 1892 KASSERT(bp->b_iodone == NULL, 1893 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1894 1895 /* get a new block */ 1896 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 1897 if (newbp == NULL) 1898 goto normal_write; 1899 1900 /* 1901 * set it to be identical to the old block. We have to 1902 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1903 * to avoid confusing the splay tree and gbincore(). 1904 */ 1905 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1906 newbp->b_lblkno = bp->b_lblkno; 1907 newbp->b_xflags |= BX_BKGRDMARKER; 1908 BO_LOCK(bp->b_bufobj); 1909 bp->b_vflags |= BV_BKGRDINPROG; 1910 bgetvp(bp->b_vp, newbp); 1911 BO_UNLOCK(bp->b_bufobj); 1912 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1913 newbp->b_blkno = bp->b_blkno; 1914 newbp->b_offset = bp->b_offset; 1915 newbp->b_iodone = ffs_backgroundwritedone; 1916 newbp->b_flags |= B_ASYNC; 1917 newbp->b_flags &= ~B_INVAL; 1918 1919 #ifdef SOFTUPDATES 1920 /* 1921 * Move over the dependencies. If there are rollbacks, 1922 * leave the parent buffer dirtied as it will need to 1923 * be written again. 1924 */ 1925 if (LIST_EMPTY(&bp->b_dep) || 1926 softdep_move_dependencies(bp, newbp) == 0) 1927 bundirty(bp); 1928 #else 1929 bundirty(bp); 1930 #endif 1931 1932 /* 1933 * Initiate write on the copy, release the original to 1934 * the B_LOCKED queue so that it cannot go away until 1935 * the background write completes. If not locked it could go 1936 * away and then be reconstituted while it was being written. 1937 * If the reconstituted buffer were written, we could end up 1938 * with two background copies being written at the same time. 1939 */ 1940 bqrelse(bp); 1941 bp = newbp; 1942 } else 1943 /* Mark the buffer clean */ 1944 bundirty(bp); 1945 1946 1947 /* Let the normal bufwrite do the rest for us */ 1948 normal_write: 1949 return (bufwrite(bp)); 1950 } 1951 1952 1953 static void 1954 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1955 { 1956 struct vnode *vp; 1957 int error; 1958 struct buf *tbp; 1959 int nocopy; 1960 1961 vp = bo->__bo_vnode; 1962 if (bp->b_iocmd == BIO_WRITE) { 1963 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1964 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1965 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1966 panic("ffs_geom_strategy: bad I/O"); 1967 nocopy = bp->b_flags & B_NOCOPY; 1968 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 1969 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 1970 vp->v_rdev->si_snapdata != NULL) { 1971 if ((bp->b_flags & B_CLUSTER) != 0) { 1972 runningbufwakeup(bp); 1973 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1974 b_cluster.cluster_entry) { 1975 error = ffs_copyonwrite(vp, tbp); 1976 if (error != 0 && 1977 error != EOPNOTSUPP) { 1978 bp->b_error = error; 1979 bp->b_ioflags |= BIO_ERROR; 1980 bufdone(bp); 1981 return; 1982 } 1983 } 1984 bp->b_runningbufspace = bp->b_bufsize; 1985 atomic_add_long(&runningbufspace, 1986 bp->b_runningbufspace); 1987 } else { 1988 error = ffs_copyonwrite(vp, bp); 1989 if (error != 0 && error != EOPNOTSUPP) { 1990 bp->b_error = error; 1991 bp->b_ioflags |= BIO_ERROR; 1992 bufdone(bp); 1993 return; 1994 } 1995 } 1996 } 1997 #ifdef SOFTUPDATES 1998 if ((bp->b_flags & B_CLUSTER) != 0) { 1999 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2000 b_cluster.cluster_entry) { 2001 if (!LIST_EMPTY(&tbp->b_dep)) 2002 buf_start(tbp); 2003 } 2004 } else { 2005 if (!LIST_EMPTY(&bp->b_dep)) 2006 buf_start(bp); 2007 } 2008 2009 #endif 2010 } 2011 g_vfs_strategy(bo, bp); 2012 } 2013 2014 #ifdef DDB 2015 2016 static void 2017 db_print_ffs(struct ufsmount *ump) 2018 { 2019 db_printf("mp %p %s devvp %p fs %p su_wl %d su_wl_in %d su_deps %d " 2020 "su_req %d\n", 2021 ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname, 2022 ump->um_devvp, ump->um_fs, ump->softdep_on_worklist, 2023 ump->softdep_on_worklist_inprogress, ump->softdep_deps, 2024 ump->softdep_req); 2025 } 2026 2027 DB_SHOW_COMMAND(ffs, db_show_ffs) 2028 { 2029 struct mount *mp; 2030 struct ufsmount *ump; 2031 2032 if (have_addr) { 2033 ump = VFSTOUFS((struct mount *)addr); 2034 db_print_ffs(ump); 2035 return; 2036 } 2037 2038 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2039 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2040 db_print_ffs(VFSTOUFS(mp)); 2041 } 2042 } 2043 2044 #endif /* DDB */ 2045