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_oldfscompat_write(struct fs *, struct ufsmount *); 83 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 84 static vfs_init_t ffs_init; 85 static vfs_uninit_t ffs_uninit; 86 static vfs_extattrctl_t ffs_extattrctl; 87 static vfs_cmount_t ffs_cmount; 88 static vfs_unmount_t ffs_unmount; 89 static vfs_mount_t ffs_mount; 90 static vfs_statfs_t ffs_statfs; 91 static vfs_fhtovp_t ffs_fhtovp; 92 static vfs_sync_t ffs_sync; 93 94 static struct vfsops ufs_vfsops = { 95 .vfs_extattrctl = ffs_extattrctl, 96 .vfs_fhtovp = ffs_fhtovp, 97 .vfs_init = ffs_init, 98 .vfs_mount = ffs_mount, 99 .vfs_cmount = ffs_cmount, 100 .vfs_quotactl = ufs_quotactl, 101 .vfs_root = ufs_root, 102 .vfs_statfs = ffs_statfs, 103 .vfs_sync = ffs_sync, 104 .vfs_uninit = ffs_uninit, 105 .vfs_unmount = ffs_unmount, 106 .vfs_vget = ffs_vget, 107 .vfs_susp_clean = process_deferred_inactive, 108 }; 109 110 VFS_SET(ufs_vfsops, ufs, 0); 111 MODULE_VERSION(ufs, 1); 112 113 static b_strategy_t ffs_geom_strategy; 114 static b_write_t ffs_bufwrite; 115 116 static struct buf_ops ffs_ops = { 117 .bop_name = "FFS", 118 .bop_write = ffs_bufwrite, 119 .bop_strategy = ffs_geom_strategy, 120 .bop_sync = bufsync, 121 #ifdef NO_FFS_SNAPSHOT 122 .bop_bdflush = bufbdflush, 123 #else 124 .bop_bdflush = ffs_bdflush, 125 #endif 126 }; 127 128 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 129 "noclusterw", "noexec", "export", "force", "from", "multilabel", 130 "snapshot", "nosuid", "suiddir", "nosymfollow", "sync", 131 "union", "nfsv4acls", NULL }; 132 133 static int 134 ffs_mount(struct mount *mp) 135 { 136 struct vnode *devvp; 137 struct thread *td; 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 td = curthread; 147 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 148 return (EINVAL); 149 if (uma_inode == NULL) { 150 uma_inode = uma_zcreate("FFS inode", 151 sizeof(struct inode), NULL, NULL, NULL, NULL, 152 UMA_ALIGN_PTR, 0); 153 uma_ufs1 = uma_zcreate("FFS1 dinode", 154 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 155 UMA_ALIGN_PTR, 0); 156 uma_ufs2 = uma_zcreate("FFS2 dinode", 157 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 158 UMA_ALIGN_PTR, 0); 159 } 160 161 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 162 if (error) 163 return (error); 164 165 mntorflags = 0; 166 mntandnotflags = 0; 167 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 168 mntorflags |= MNT_ACLS; 169 170 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 171 mntorflags |= MNT_SNAPSHOT; 172 /* 173 * Once we have set the MNT_SNAPSHOT flag, do not 174 * persist "snapshot" in the options list. 175 */ 176 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 177 vfs_deleteopt(mp->mnt_opt, "snapshot"); 178 } 179 180 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 181 if (mntorflags & MNT_ACLS) { 182 printf("WARNING: \"acls\" and \"nfsv4acls\" " 183 "options are mutually exclusive\n"); 184 return (EINVAL); 185 } 186 mntorflags |= MNT_NFS4ACLS; 187 } 188 189 MNT_ILOCK(mp); 190 mp->mnt_flag = (mp->mnt_flag | mntorflags) & ~mntandnotflags; 191 MNT_IUNLOCK(mp); 192 /* 193 * If updating, check whether changing from read-only to 194 * read/write; if there is no device name, that's all we do. 195 */ 196 if (mp->mnt_flag & MNT_UPDATE) { 197 ump = VFSTOUFS(mp); 198 fs = ump->um_fs; 199 devvp = ump->um_devvp; 200 if (fs->fs_ronly == 0 && 201 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 202 /* 203 * Flush any dirty data and suspend filesystem. 204 */ 205 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 206 return (error); 207 for (;;) { 208 vn_finished_write(mp); 209 if ((error = vfs_write_suspend(mp)) != 0) 210 return (error); 211 MNT_ILOCK(mp); 212 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 213 /* 214 * Allow the secondary writes 215 * to proceed. 216 */ 217 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 218 MNTK_SUSPEND2); 219 wakeup(&mp->mnt_flag); 220 MNT_IUNLOCK(mp); 221 /* 222 * Allow the curthread to 223 * ignore the suspension to 224 * synchronize on-disk state. 225 */ 226 td->td_pflags |= TDP_IGNSUSP; 227 break; 228 } 229 MNT_IUNLOCK(mp); 230 vn_start_write(NULL, &mp, V_WAIT); 231 } 232 /* 233 * Check for and optionally get rid of files open 234 * for writing. 235 */ 236 flags = WRITECLOSE; 237 if (mp->mnt_flag & MNT_FORCE) 238 flags |= FORCECLOSE; 239 if (mp->mnt_flag & MNT_SOFTDEP) { 240 error = softdep_flushfiles(mp, flags, td); 241 } else { 242 error = ffs_flushfiles(mp, flags, td); 243 } 244 if (error) { 245 vfs_write_resume(mp); 246 return (error); 247 } 248 if (fs->fs_pendingblocks != 0 || 249 fs->fs_pendinginodes != 0) { 250 printf("%s: %s: blocks %jd files %d\n", 251 fs->fs_fsmnt, "update error", 252 (intmax_t)fs->fs_pendingblocks, 253 fs->fs_pendinginodes); 254 fs->fs_pendingblocks = 0; 255 fs->fs_pendinginodes = 0; 256 } 257 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 258 fs->fs_clean = 1; 259 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 260 fs->fs_ronly = 0; 261 fs->fs_clean = 0; 262 vfs_write_resume(mp); 263 return (error); 264 } 265 DROP_GIANT(); 266 g_topology_lock(); 267 g_access(ump->um_cp, 0, -1, 0); 268 g_topology_unlock(); 269 PICKUP_GIANT(); 270 fs->fs_ronly = 1; 271 MNT_ILOCK(mp); 272 mp->mnt_flag |= MNT_RDONLY; 273 MNT_IUNLOCK(mp); 274 /* 275 * Allow the writers to note that filesystem 276 * is ro now. 277 */ 278 vfs_write_resume(mp); 279 } 280 if ((mp->mnt_flag & MNT_RELOAD) && 281 (error = ffs_reload(mp, td)) != 0) 282 return (error); 283 if (fs->fs_ronly && 284 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 285 /* 286 * If upgrade to read-write by non-root, then verify 287 * that user has necessary permissions on the device. 288 */ 289 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 290 error = VOP_ACCESS(devvp, VREAD | VWRITE, 291 td->td_ucred, td); 292 if (error) 293 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 294 if (error) { 295 VOP_UNLOCK(devvp, 0); 296 return (error); 297 } 298 VOP_UNLOCK(devvp, 0); 299 fs->fs_flags &= ~FS_UNCLEAN; 300 if (fs->fs_clean == 0) { 301 fs->fs_flags |= FS_UNCLEAN; 302 if ((mp->mnt_flag & MNT_FORCE) || 303 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 304 (fs->fs_flags & FS_DOSOFTDEP))) { 305 printf("WARNING: %s was not %s\n", 306 fs->fs_fsmnt, "properly dismounted"); 307 } else { 308 printf( 309 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 310 fs->fs_fsmnt); 311 return (EPERM); 312 } 313 } 314 DROP_GIANT(); 315 g_topology_lock(); 316 /* 317 * If we're the root device, we may not have an E count 318 * yet, get it now. 319 */ 320 if (ump->um_cp->ace == 0) 321 error = g_access(ump->um_cp, 0, 1, 1); 322 else 323 error = g_access(ump->um_cp, 0, 1, 0); 324 g_topology_unlock(); 325 PICKUP_GIANT(); 326 if (error) 327 return (error); 328 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 329 return (error); 330 fs->fs_ronly = 0; 331 MNT_ILOCK(mp); 332 mp->mnt_flag &= ~MNT_RDONLY; 333 MNT_IUNLOCK(mp); 334 fs->fs_clean = 0; 335 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 336 vn_finished_write(mp); 337 return (error); 338 } 339 /* check to see if we need to start softdep */ 340 if ((fs->fs_flags & FS_DOSOFTDEP) && 341 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 342 vn_finished_write(mp); 343 return (error); 344 } 345 if (fs->fs_snapinum[0] != 0) 346 ffs_snapshot_mount(mp); 347 vn_finished_write(mp); 348 } 349 /* 350 * Soft updates is incompatible with "async", 351 * so if we are doing softupdates stop the user 352 * from setting the async flag in an update. 353 * Softdep_mount() clears it in an initial mount 354 * or ro->rw remount. 355 */ 356 if (mp->mnt_flag & MNT_SOFTDEP) { 357 /* XXX: Reset too late ? */ 358 MNT_ILOCK(mp); 359 mp->mnt_flag &= ~MNT_ASYNC; 360 MNT_IUNLOCK(mp); 361 } 362 /* 363 * Keep MNT_ACLS flag if it is stored in superblock. 364 */ 365 if ((fs->fs_flags & FS_ACLS) != 0) { 366 /* XXX: Set too late ? */ 367 MNT_ILOCK(mp); 368 mp->mnt_flag |= MNT_ACLS; 369 MNT_IUNLOCK(mp); 370 } 371 372 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 373 /* XXX: Set too late ? */ 374 MNT_ILOCK(mp); 375 mp->mnt_flag |= MNT_NFS4ACLS; 376 MNT_IUNLOCK(mp); 377 } 378 379 /* 380 * If this is a snapshot request, take the snapshot. 381 */ 382 if (mp->mnt_flag & MNT_SNAPSHOT) 383 return (ffs_snapshot(mp, fspec)); 384 } 385 386 /* 387 * Not an update, or updating the name: look up the name 388 * and verify that it refers to a sensible disk device. 389 */ 390 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 391 if ((error = namei(&ndp)) != 0) 392 return (error); 393 NDFREE(&ndp, NDF_ONLY_PNBUF); 394 devvp = ndp.ni_vp; 395 if (!vn_isdisk(devvp, &error)) { 396 vput(devvp); 397 return (error); 398 } 399 400 /* 401 * If mount by non-root, then verify that user has necessary 402 * permissions on the device. 403 */ 404 accmode = VREAD; 405 if ((mp->mnt_flag & MNT_RDONLY) == 0) 406 accmode |= VWRITE; 407 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 408 if (error) 409 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 410 if (error) { 411 vput(devvp); 412 return (error); 413 } 414 415 if (mp->mnt_flag & MNT_UPDATE) { 416 /* 417 * Update only 418 * 419 * If it's not the same vnode, or at least the same device 420 * then it's not correct. 421 */ 422 423 if (devvp->v_rdev != ump->um_devvp->v_rdev) 424 error = EINVAL; /* needs translation */ 425 vput(devvp); 426 if (error) 427 return (error); 428 } else { 429 /* 430 * New mount 431 * 432 * We need the name for the mount point (also used for 433 * "last mounted on") copied in. If an error occurs, 434 * the mount point is discarded by the upper level code. 435 * Note that vfs_mount() populates f_mntonname for us. 436 */ 437 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 438 vrele(devvp); 439 return (error); 440 } 441 } 442 vfs_mountedfrom(mp, fspec); 443 return (0); 444 } 445 446 /* 447 * Compatibility with old mount system call. 448 */ 449 450 static int 451 ffs_cmount(struct mntarg *ma, void *data, int flags) 452 { 453 struct ufs_args args; 454 int error; 455 456 if (data == NULL) 457 return (EINVAL); 458 error = copyin(data, &args, sizeof args); 459 if (error) 460 return (error); 461 462 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 463 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 464 error = kernel_mount(ma, flags); 465 466 return (error); 467 } 468 469 /* 470 * Reload all incore data for a filesystem (used after running fsck on 471 * the root filesystem and finding things to fix). The filesystem must 472 * be mounted read-only. 473 * 474 * Things to do to update the mount: 475 * 1) invalidate all cached meta-data. 476 * 2) re-read superblock from disk. 477 * 3) re-read summary information from disk. 478 * 4) invalidate all inactive vnodes. 479 * 5) invalidate all cached file data. 480 * 6) re-read inode data for all active vnodes. 481 */ 482 static int 483 ffs_reload(struct mount *mp, struct thread *td) 484 { 485 struct vnode *vp, *mvp, *devvp; 486 struct inode *ip; 487 void *space; 488 struct buf *bp; 489 struct fs *fs, *newfs; 490 struct ufsmount *ump; 491 ufs2_daddr_t sblockloc; 492 int i, blks, size, error; 493 int32_t *lp; 494 495 if ((mp->mnt_flag & MNT_RDONLY) == 0) 496 return (EINVAL); 497 ump = VFSTOUFS(mp); 498 /* 499 * Step 1: invalidate all cached meta-data. 500 */ 501 devvp = VFSTOUFS(mp)->um_devvp; 502 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 503 if (vinvalbuf(devvp, 0, 0, 0) != 0) 504 panic("ffs_reload: dirty1"); 505 VOP_UNLOCK(devvp, 0); 506 507 /* 508 * Step 2: re-read superblock from disk. 509 */ 510 fs = VFSTOUFS(mp)->um_fs; 511 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 512 NOCRED, &bp)) != 0) 513 return (error); 514 newfs = (struct fs *)bp->b_data; 515 if ((newfs->fs_magic != FS_UFS1_MAGIC && 516 newfs->fs_magic != FS_UFS2_MAGIC) || 517 newfs->fs_bsize > MAXBSIZE || 518 newfs->fs_bsize < sizeof(struct fs)) { 519 brelse(bp); 520 return (EIO); /* XXX needs translation */ 521 } 522 /* 523 * Copy pointer fields back into superblock before copying in XXX 524 * new superblock. These should really be in the ufsmount. XXX 525 * Note that important parameters (eg fs_ncg) are unchanged. 526 */ 527 newfs->fs_csp = fs->fs_csp; 528 newfs->fs_maxcluster = fs->fs_maxcluster; 529 newfs->fs_contigdirs = fs->fs_contigdirs; 530 newfs->fs_active = fs->fs_active; 531 /* The file system is still read-only. */ 532 newfs->fs_ronly = 1; 533 sblockloc = fs->fs_sblockloc; 534 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 535 brelse(bp); 536 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 537 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 538 UFS_LOCK(ump); 539 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 540 printf("%s: reload pending error: blocks %jd files %d\n", 541 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 542 fs->fs_pendinginodes); 543 fs->fs_pendingblocks = 0; 544 fs->fs_pendinginodes = 0; 545 } 546 UFS_UNLOCK(ump); 547 548 /* 549 * Step 3: re-read summary information from disk. 550 */ 551 blks = howmany(fs->fs_cssize, fs->fs_fsize); 552 space = fs->fs_csp; 553 for (i = 0; i < blks; i += fs->fs_frag) { 554 size = fs->fs_bsize; 555 if (i + fs->fs_frag > blks) 556 size = (blks - i) * fs->fs_fsize; 557 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 558 NOCRED, &bp); 559 if (error) 560 return (error); 561 bcopy(bp->b_data, space, (u_int)size); 562 space = (char *)space + size; 563 brelse(bp); 564 } 565 /* 566 * We no longer know anything about clusters per cylinder group. 567 */ 568 if (fs->fs_contigsumsize > 0) { 569 lp = fs->fs_maxcluster; 570 for (i = 0; i < fs->fs_ncg; i++) 571 *lp++ = fs->fs_contigsumsize; 572 } 573 574 loop: 575 MNT_ILOCK(mp); 576 MNT_VNODE_FOREACH(vp, mp, mvp) { 577 VI_LOCK(vp); 578 if (vp->v_iflag & VI_DOOMED) { 579 VI_UNLOCK(vp); 580 continue; 581 } 582 MNT_IUNLOCK(mp); 583 /* 584 * Step 4: invalidate all cached file data. 585 */ 586 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 587 MNT_VNODE_FOREACH_ABORT(mp, mvp); 588 goto loop; 589 } 590 if (vinvalbuf(vp, 0, 0, 0)) 591 panic("ffs_reload: dirty2"); 592 /* 593 * Step 5: re-read inode data for all active vnodes. 594 */ 595 ip = VTOI(vp); 596 error = 597 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 598 (int)fs->fs_bsize, NOCRED, &bp); 599 if (error) { 600 VOP_UNLOCK(vp, 0); 601 vrele(vp); 602 MNT_VNODE_FOREACH_ABORT(mp, mvp); 603 return (error); 604 } 605 ffs_load_inode(bp, ip, fs, ip->i_number); 606 ip->i_effnlink = ip->i_nlink; 607 brelse(bp); 608 VOP_UNLOCK(vp, 0); 609 vrele(vp); 610 MNT_ILOCK(mp); 611 } 612 MNT_IUNLOCK(mp); 613 return (0); 614 } 615 616 /* 617 * Possible superblock locations ordered from most to least likely. 618 */ 619 static int sblock_try[] = SBLOCKSEARCH; 620 621 /* 622 * Common code for mount and mountroot 623 */ 624 static int 625 ffs_mountfs(devvp, mp, td) 626 struct vnode *devvp; 627 struct mount *mp; 628 struct thread *td; 629 { 630 struct ufsmount *ump; 631 struct buf *bp; 632 struct fs *fs; 633 struct cdev *dev; 634 void *space; 635 ufs2_daddr_t sblockloc; 636 int error, i, blks, size, ronly; 637 int32_t *lp; 638 struct ucred *cred; 639 struct g_consumer *cp; 640 struct mount *nmp; 641 642 bp = NULL; 643 ump = NULL; 644 cred = td ? td->td_ucred : NOCRED; 645 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 646 647 dev = devvp->v_rdev; 648 dev_ref(dev); 649 DROP_GIANT(); 650 g_topology_lock(); 651 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 652 653 /* 654 * If we are a root mount, drop the E flag so fsck can do its magic. 655 * We will pick it up again when we remount R/W. 656 */ 657 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 658 error = g_access(cp, 0, 0, -1); 659 g_topology_unlock(); 660 PICKUP_GIANT(); 661 VOP_UNLOCK(devvp, 0); 662 if (error) 663 goto out; 664 if (devvp->v_rdev->si_iosize_max != 0) 665 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 666 if (mp->mnt_iosize_max > MAXPHYS) 667 mp->mnt_iosize_max = MAXPHYS; 668 669 devvp->v_bufobj.bo_private = cp; 670 devvp->v_bufobj.bo_ops = &ffs_ops; 671 672 fs = NULL; 673 sblockloc = 0; 674 /* 675 * Try reading the superblock in each of its possible locations. 676 */ 677 for (i = 0; sblock_try[i] != -1; i++) { 678 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 679 error = EINVAL; 680 vfs_mount_error(mp, 681 "Invalid sectorsize %d for superblock size %d", 682 cp->provider->sectorsize, SBLOCKSIZE); 683 goto out; 684 } 685 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE, 686 cred, &bp)) != 0) 687 goto out; 688 fs = (struct fs *)bp->b_data; 689 sblockloc = sblock_try[i]; 690 if ((fs->fs_magic == FS_UFS1_MAGIC || 691 (fs->fs_magic == FS_UFS2_MAGIC && 692 (fs->fs_sblockloc == sblockloc || 693 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 694 fs->fs_bsize <= MAXBSIZE && 695 fs->fs_bsize >= sizeof(struct fs)) 696 break; 697 brelse(bp); 698 bp = NULL; 699 } 700 if (sblock_try[i] == -1) { 701 error = EINVAL; /* XXX needs translation */ 702 goto out; 703 } 704 fs->fs_fmod = 0; 705 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 706 fs->fs_flags &= ~FS_UNCLEAN; 707 if (fs->fs_clean == 0) { 708 fs->fs_flags |= FS_UNCLEAN; 709 if (ronly || (mp->mnt_flag & MNT_FORCE) || 710 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 711 (fs->fs_flags & FS_DOSOFTDEP))) { 712 printf( 713 "WARNING: %s was not properly dismounted\n", 714 fs->fs_fsmnt); 715 } else { 716 printf( 717 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 718 fs->fs_fsmnt); 719 error = EPERM; 720 goto out; 721 } 722 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 723 (mp->mnt_flag & MNT_FORCE)) { 724 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 725 (intmax_t)fs->fs_pendingblocks, 726 fs->fs_pendinginodes); 727 fs->fs_pendingblocks = 0; 728 fs->fs_pendinginodes = 0; 729 } 730 } 731 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 732 printf("%s: mount pending error: blocks %jd files %d\n", 733 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 734 fs->fs_pendinginodes); 735 fs->fs_pendingblocks = 0; 736 fs->fs_pendinginodes = 0; 737 } 738 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 739 #ifdef UFS_GJOURNAL 740 /* 741 * Get journal provider name. 742 */ 743 size = 1024; 744 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK); 745 if (g_io_getattr("GJOURNAL::provider", cp, &size, 746 mp->mnt_gjprovider) == 0) { 747 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size, 748 M_UFSMNT, M_WAITOK); 749 MNT_ILOCK(mp); 750 mp->mnt_flag |= MNT_GJOURNAL; 751 MNT_IUNLOCK(mp); 752 } else { 753 printf( 754 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n", 755 mp->mnt_stat.f_mntonname); 756 free(mp->mnt_gjprovider, M_UFSMNT); 757 mp->mnt_gjprovider = NULL; 758 } 759 #else 760 printf( 761 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n", 762 mp->mnt_stat.f_mntonname); 763 #endif 764 } else { 765 mp->mnt_gjprovider = NULL; 766 } 767 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 768 ump->um_cp = cp; 769 ump->um_bo = &devvp->v_bufobj; 770 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 771 if (fs->fs_magic == FS_UFS1_MAGIC) { 772 ump->um_fstype = UFS1; 773 ump->um_balloc = ffs_balloc_ufs1; 774 } else { 775 ump->um_fstype = UFS2; 776 ump->um_balloc = ffs_balloc_ufs2; 777 } 778 ump->um_blkatoff = ffs_blkatoff; 779 ump->um_truncate = ffs_truncate; 780 ump->um_update = ffs_update; 781 ump->um_valloc = ffs_valloc; 782 ump->um_vfree = ffs_vfree; 783 ump->um_ifree = ffs_ifree; 784 ump->um_rdonly = ffs_rdonly; 785 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 786 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 787 if (fs->fs_sbsize < SBLOCKSIZE) 788 bp->b_flags |= B_INVAL | B_NOCACHE; 789 brelse(bp); 790 bp = NULL; 791 fs = ump->um_fs; 792 ffs_oldfscompat_read(fs, ump, sblockloc); 793 fs->fs_ronly = ronly; 794 size = fs->fs_cssize; 795 blks = howmany(size, fs->fs_fsize); 796 if (fs->fs_contigsumsize > 0) 797 size += fs->fs_ncg * sizeof(int32_t); 798 size += fs->fs_ncg * sizeof(u_int8_t); 799 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 800 fs->fs_csp = space; 801 for (i = 0; i < blks; i += fs->fs_frag) { 802 size = fs->fs_bsize; 803 if (i + fs->fs_frag > blks) 804 size = (blks - i) * fs->fs_fsize; 805 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 806 cred, &bp)) != 0) { 807 free(fs->fs_csp, M_UFSMNT); 808 goto out; 809 } 810 bcopy(bp->b_data, space, (u_int)size); 811 space = (char *)space + size; 812 brelse(bp); 813 bp = NULL; 814 } 815 if (fs->fs_contigsumsize > 0) { 816 fs->fs_maxcluster = lp = space; 817 for (i = 0; i < fs->fs_ncg; i++) 818 *lp++ = fs->fs_contigsumsize; 819 space = lp; 820 } 821 size = fs->fs_ncg * sizeof(u_int8_t); 822 fs->fs_contigdirs = (u_int8_t *)space; 823 bzero(fs->fs_contigdirs, size); 824 fs->fs_active = NULL; 825 mp->mnt_data = ump; 826 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 827 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 828 nmp = NULL; 829 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 830 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 831 if (nmp) 832 vfs_rel(nmp); 833 vfs_getnewfsid(mp); 834 } 835 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 836 MNT_ILOCK(mp); 837 mp->mnt_flag |= MNT_LOCAL; 838 MNT_IUNLOCK(mp); 839 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 840 #ifdef MAC 841 MNT_ILOCK(mp); 842 mp->mnt_flag |= MNT_MULTILABEL; 843 MNT_IUNLOCK(mp); 844 #else 845 printf( 846 "WARNING: %s: multilabel flag on fs but no MAC support\n", 847 mp->mnt_stat.f_mntonname); 848 #endif 849 } 850 if ((fs->fs_flags & FS_ACLS) != 0) { 851 #ifdef UFS_ACL 852 MNT_ILOCK(mp); 853 854 if (mp->mnt_flag & MNT_NFS4ACLS) 855 printf("WARNING: ACLs flag on fs conflicts with " 856 "\"nfsv4acls\" mount option; option ignored\n"); 857 mp->mnt_flag &= ~MNT_NFS4ACLS; 858 mp->mnt_flag |= MNT_ACLS; 859 860 MNT_IUNLOCK(mp); 861 #else 862 printf( 863 "WARNING: %s: ACLs flag on fs but no ACLs support\n", 864 mp->mnt_stat.f_mntonname); 865 #endif 866 } 867 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 868 #ifdef UFS_ACL 869 MNT_ILOCK(mp); 870 871 if (mp->mnt_flag & MNT_ACLS) 872 printf("WARNING: NFSv4 ACLs flag on fs conflicts with " 873 "\"acls\" mount option; option ignored\n"); 874 mp->mnt_flag &= ~MNT_ACLS; 875 mp->mnt_flag |= MNT_NFS4ACLS; 876 877 MNT_IUNLOCK(mp); 878 #else 879 printf( 880 "WARNING: %s: NFSv4 ACLs flag on fs but no ACLs support\n", 881 mp->mnt_stat.f_mntonname); 882 #endif 883 } 884 885 ump->um_mountp = mp; 886 ump->um_dev = dev; 887 ump->um_devvp = devvp; 888 ump->um_nindir = fs->fs_nindir; 889 ump->um_bptrtodb = fs->fs_fsbtodb; 890 ump->um_seqinc = fs->fs_frag; 891 for (i = 0; i < MAXQUOTAS; i++) 892 ump->um_quotas[i] = NULLVP; 893 #ifdef UFS_EXTATTR 894 ufs_extattr_uepm_init(&ump->um_extattr); 895 #endif 896 /* 897 * Set FS local "last mounted on" information (NULL pad) 898 */ 899 bzero(fs->fs_fsmnt, MAXMNTLEN); 900 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 901 902 if( mp->mnt_flag & MNT_ROOTFS) { 903 /* 904 * Root mount; update timestamp in mount structure. 905 * this will be used by the common root mount code 906 * to update the system clock. 907 */ 908 mp->mnt_time = fs->fs_time; 909 } 910 911 if (ronly == 0) { 912 if ((fs->fs_flags & FS_DOSOFTDEP) && 913 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 914 free(fs->fs_csp, M_UFSMNT); 915 goto out; 916 } 917 if (fs->fs_snapinum[0] != 0) 918 ffs_snapshot_mount(mp); 919 fs->fs_fmod = 1; 920 fs->fs_clean = 0; 921 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 922 } 923 /* 924 * Initialize filesystem stat information in mount struct. 925 */ 926 MNT_ILOCK(mp); 927 mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED | 928 MNTK_EXTENDED_SHARED; 929 MNT_IUNLOCK(mp); 930 #ifdef UFS_EXTATTR 931 #ifdef UFS_EXTATTR_AUTOSTART 932 /* 933 * 934 * Auto-starting does the following: 935 * - check for /.attribute in the fs, and extattr_start if so 936 * - for each file in .attribute, enable that file with 937 * an attribute of the same name. 938 * Not clear how to report errors -- probably eat them. 939 * This would all happen while the filesystem was busy/not 940 * available, so would effectively be "atomic". 941 */ 942 mp->mnt_stat.f_iosize = fs->fs_bsize; 943 (void) ufs_extattr_autostart(mp, td); 944 #endif /* !UFS_EXTATTR_AUTOSTART */ 945 #endif /* !UFS_EXTATTR */ 946 return (0); 947 out: 948 if (bp) 949 brelse(bp); 950 if (cp != NULL) { 951 DROP_GIANT(); 952 g_topology_lock(); 953 g_vfs_close(cp); 954 g_topology_unlock(); 955 PICKUP_GIANT(); 956 } 957 if (ump) { 958 mtx_destroy(UFS_MTX(ump)); 959 if (mp->mnt_gjprovider != NULL) { 960 free(mp->mnt_gjprovider, M_UFSMNT); 961 mp->mnt_gjprovider = NULL; 962 } 963 free(ump->um_fs, M_UFSMNT); 964 free(ump, M_UFSMNT); 965 mp->mnt_data = NULL; 966 } 967 dev_rel(dev); 968 return (error); 969 } 970 971 #include <sys/sysctl.h> 972 static int bigcgs = 0; 973 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 974 975 /* 976 * Sanity checks for loading old filesystem superblocks. 977 * See ffs_oldfscompat_write below for unwound actions. 978 * 979 * XXX - Parts get retired eventually. 980 * Unfortunately new bits get added. 981 */ 982 static void 983 ffs_oldfscompat_read(fs, ump, sblockloc) 984 struct fs *fs; 985 struct ufsmount *ump; 986 ufs2_daddr_t sblockloc; 987 { 988 off_t maxfilesize; 989 990 /* 991 * If not yet done, update fs_flags location and value of fs_sblockloc. 992 */ 993 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 994 fs->fs_flags = fs->fs_old_flags; 995 fs->fs_old_flags |= FS_FLAGS_UPDATED; 996 fs->fs_sblockloc = sblockloc; 997 } 998 /* 999 * If not yet done, update UFS1 superblock with new wider fields. 1000 */ 1001 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1002 fs->fs_maxbsize = fs->fs_bsize; 1003 fs->fs_time = fs->fs_old_time; 1004 fs->fs_size = fs->fs_old_size; 1005 fs->fs_dsize = fs->fs_old_dsize; 1006 fs->fs_csaddr = fs->fs_old_csaddr; 1007 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1008 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1009 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1010 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1011 } 1012 if (fs->fs_magic == FS_UFS1_MAGIC && 1013 fs->fs_old_inodefmt < FS_44INODEFMT) { 1014 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1015 fs->fs_qbmask = ~fs->fs_bmask; 1016 fs->fs_qfmask = ~fs->fs_fmask; 1017 } 1018 if (fs->fs_magic == FS_UFS1_MAGIC) { 1019 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1020 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1021 if (fs->fs_maxfilesize > maxfilesize) 1022 fs->fs_maxfilesize = maxfilesize; 1023 } 1024 /* Compatibility for old filesystems */ 1025 if (fs->fs_avgfilesize <= 0) 1026 fs->fs_avgfilesize = AVFILESIZ; 1027 if (fs->fs_avgfpdir <= 0) 1028 fs->fs_avgfpdir = AFPDIR; 1029 if (bigcgs) { 1030 fs->fs_save_cgsize = fs->fs_cgsize; 1031 fs->fs_cgsize = fs->fs_bsize; 1032 } 1033 } 1034 1035 /* 1036 * Unwinding superblock updates for old filesystems. 1037 * See ffs_oldfscompat_read above for details. 1038 * 1039 * XXX - Parts get retired eventually. 1040 * Unfortunately new bits get added. 1041 */ 1042 static void 1043 ffs_oldfscompat_write(fs, ump) 1044 struct fs *fs; 1045 struct ufsmount *ump; 1046 { 1047 1048 /* 1049 * Copy back UFS2 updated fields that UFS1 inspects. 1050 */ 1051 if (fs->fs_magic == FS_UFS1_MAGIC) { 1052 fs->fs_old_time = fs->fs_time; 1053 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1054 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1055 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1056 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1057 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1058 } 1059 if (bigcgs) { 1060 fs->fs_cgsize = fs->fs_save_cgsize; 1061 fs->fs_save_cgsize = 0; 1062 } 1063 } 1064 1065 /* 1066 * unmount system call 1067 */ 1068 static int 1069 ffs_unmount(mp, mntflags) 1070 struct mount *mp; 1071 int mntflags; 1072 { 1073 struct thread *td; 1074 struct ufsmount *ump = VFSTOUFS(mp); 1075 struct fs *fs; 1076 int error, flags, susp; 1077 #ifdef UFS_EXTATTR 1078 int e_restart; 1079 #endif 1080 1081 flags = 0; 1082 td = curthread; 1083 fs = ump->um_fs; 1084 if (mntflags & MNT_FORCE) { 1085 flags |= FORCECLOSE; 1086 susp = fs->fs_ronly != 0; 1087 } else 1088 susp = 0; 1089 #ifdef UFS_EXTATTR 1090 if ((error = ufs_extattr_stop(mp, td))) { 1091 if (error != EOPNOTSUPP) 1092 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 1093 error); 1094 e_restart = 0; 1095 } else { 1096 ufs_extattr_uepm_destroy(&ump->um_extattr); 1097 e_restart = 1; 1098 } 1099 #endif 1100 if (susp) { 1101 /* 1102 * dounmount already called vn_start_write(). 1103 */ 1104 for (;;) { 1105 vn_finished_write(mp); 1106 if ((error = vfs_write_suspend(mp)) != 0) 1107 return (error); 1108 MNT_ILOCK(mp); 1109 if (mp->mnt_kern_flag & MNTK_SUSPENDED) { 1110 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | 1111 MNTK_SUSPEND2); 1112 wakeup(&mp->mnt_flag); 1113 MNT_IUNLOCK(mp); 1114 td->td_pflags |= TDP_IGNSUSP; 1115 break; 1116 } 1117 MNT_IUNLOCK(mp); 1118 vn_start_write(NULL, &mp, V_WAIT); 1119 } 1120 } 1121 if (mp->mnt_flag & MNT_SOFTDEP) 1122 error = softdep_flushfiles(mp, flags, td); 1123 else 1124 error = ffs_flushfiles(mp, flags, td); 1125 if (error != 0 && error != ENXIO) 1126 goto fail; 1127 1128 UFS_LOCK(ump); 1129 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1130 printf("%s: unmount pending error: blocks %jd files %d\n", 1131 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1132 fs->fs_pendinginodes); 1133 fs->fs_pendingblocks = 0; 1134 fs->fs_pendinginodes = 0; 1135 } 1136 UFS_UNLOCK(ump); 1137 if (fs->fs_ronly == 0) { 1138 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1139 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1140 if (error && error != ENXIO) { 1141 fs->fs_clean = 0; 1142 goto fail; 1143 } 1144 } 1145 if (susp) { 1146 vfs_write_resume(mp); 1147 vn_start_write(NULL, &mp, V_WAIT); 1148 } 1149 DROP_GIANT(); 1150 g_topology_lock(); 1151 g_vfs_close(ump->um_cp); 1152 g_topology_unlock(); 1153 PICKUP_GIANT(); 1154 vrele(ump->um_devvp); 1155 dev_rel(ump->um_dev); 1156 mtx_destroy(UFS_MTX(ump)); 1157 if (mp->mnt_gjprovider != NULL) { 1158 free(mp->mnt_gjprovider, M_UFSMNT); 1159 mp->mnt_gjprovider = NULL; 1160 } 1161 free(fs->fs_csp, M_UFSMNT); 1162 free(fs, M_UFSMNT); 1163 free(ump, M_UFSMNT); 1164 mp->mnt_data = NULL; 1165 MNT_ILOCK(mp); 1166 mp->mnt_flag &= ~MNT_LOCAL; 1167 MNT_IUNLOCK(mp); 1168 return (error); 1169 1170 fail: 1171 if (susp) { 1172 vfs_write_resume(mp); 1173 vn_start_write(NULL, &mp, V_WAIT); 1174 } 1175 #ifdef UFS_EXTATTR 1176 if (e_restart) { 1177 ufs_extattr_uepm_init(&ump->um_extattr); 1178 #ifdef UFS_EXTATTR_AUTOSTART 1179 (void) ufs_extattr_autostart(mp, td); 1180 #endif 1181 } 1182 #endif 1183 1184 return (error); 1185 } 1186 1187 /* 1188 * Flush out all the files in a filesystem. 1189 */ 1190 int 1191 ffs_flushfiles(mp, flags, td) 1192 struct mount *mp; 1193 int flags; 1194 struct thread *td; 1195 { 1196 struct ufsmount *ump; 1197 int error; 1198 1199 ump = VFSTOUFS(mp); 1200 #ifdef QUOTA 1201 if (mp->mnt_flag & MNT_QUOTA) { 1202 int i; 1203 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1204 if (error) 1205 return (error); 1206 for (i = 0; i < MAXQUOTAS; i++) { 1207 quotaoff(td, mp, i); 1208 } 1209 /* 1210 * Here we fall through to vflush again to ensure 1211 * that we have gotten rid of all the system vnodes. 1212 */ 1213 } 1214 #endif 1215 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1216 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1217 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1218 return (error); 1219 ffs_snapshot_unmount(mp); 1220 flags |= FORCECLOSE; 1221 /* 1222 * Here we fall through to vflush again to ensure 1223 * that we have gotten rid of all the system vnodes. 1224 */ 1225 } 1226 /* 1227 * Flush all the files. 1228 */ 1229 if ((error = vflush(mp, 0, flags, td)) != 0) 1230 return (error); 1231 /* 1232 * Flush filesystem metadata. 1233 */ 1234 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1235 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1236 VOP_UNLOCK(ump->um_devvp, 0); 1237 return (error); 1238 } 1239 1240 /* 1241 * Get filesystem statistics. 1242 */ 1243 static int 1244 ffs_statfs(mp, sbp) 1245 struct mount *mp; 1246 struct statfs *sbp; 1247 { 1248 struct ufsmount *ump; 1249 struct fs *fs; 1250 1251 ump = VFSTOUFS(mp); 1252 fs = ump->um_fs; 1253 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1254 panic("ffs_statfs"); 1255 sbp->f_version = STATFS_VERSION; 1256 sbp->f_bsize = fs->fs_fsize; 1257 sbp->f_iosize = fs->fs_bsize; 1258 sbp->f_blocks = fs->fs_dsize; 1259 UFS_LOCK(ump); 1260 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1261 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1262 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1263 dbtofsb(fs, fs->fs_pendingblocks); 1264 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1265 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1266 UFS_UNLOCK(ump); 1267 sbp->f_namemax = NAME_MAX; 1268 return (0); 1269 } 1270 1271 /* 1272 * Go through the disk queues to initiate sandbagged IO; 1273 * go through the inodes to write those that have been modified; 1274 * initiate the writing of the super block if it has been modified. 1275 * 1276 * Note: we are always called with the filesystem marked `MPBUSY'. 1277 */ 1278 static int 1279 ffs_sync(mp, waitfor) 1280 struct mount *mp; 1281 int waitfor; 1282 { 1283 struct vnode *mvp, *vp, *devvp; 1284 struct thread *td; 1285 struct inode *ip; 1286 struct ufsmount *ump = VFSTOUFS(mp); 1287 struct fs *fs; 1288 int error, count, wait, lockreq, allerror = 0; 1289 int suspend; 1290 int suspended; 1291 int secondary_writes; 1292 int secondary_accwrites; 1293 int softdep_deps; 1294 int softdep_accdeps; 1295 struct bufobj *bo; 1296 1297 td = curthread; 1298 fs = ump->um_fs; 1299 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1300 printf("fs = %s\n", fs->fs_fsmnt); 1301 panic("ffs_sync: rofs mod"); 1302 } 1303 /* 1304 * Write back each (modified) inode. 1305 */ 1306 wait = 0; 1307 suspend = 0; 1308 suspended = 0; 1309 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1310 if (waitfor == MNT_SUSPEND) { 1311 suspend = 1; 1312 waitfor = MNT_WAIT; 1313 } 1314 if (waitfor == MNT_WAIT) { 1315 wait = 1; 1316 lockreq = LK_EXCLUSIVE; 1317 } 1318 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1319 MNT_ILOCK(mp); 1320 loop: 1321 /* Grab snapshot of secondary write counts */ 1322 secondary_writes = mp->mnt_secondary_writes; 1323 secondary_accwrites = mp->mnt_secondary_accwrites; 1324 1325 /* Grab snapshot of softdep dependency counts */ 1326 MNT_IUNLOCK(mp); 1327 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1328 MNT_ILOCK(mp); 1329 1330 MNT_VNODE_FOREACH(vp, mp, mvp) { 1331 /* 1332 * Depend on the mntvnode_slock to keep things stable enough 1333 * for a quick test. Since there might be hundreds of 1334 * thousands of vnodes, we cannot afford even a subroutine 1335 * call unless there's a good chance that we have work to do. 1336 */ 1337 VI_LOCK(vp); 1338 if (vp->v_iflag & VI_DOOMED) { 1339 VI_UNLOCK(vp); 1340 continue; 1341 } 1342 ip = VTOI(vp); 1343 if (vp->v_type == VNON || ((ip->i_flag & 1344 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1345 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1346 VI_UNLOCK(vp); 1347 continue; 1348 } 1349 MNT_IUNLOCK(mp); 1350 if ((error = vget(vp, lockreq, td)) != 0) { 1351 MNT_ILOCK(mp); 1352 if (error == ENOENT || error == ENOLCK) { 1353 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1354 goto loop; 1355 } 1356 continue; 1357 } 1358 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1359 allerror = error; 1360 vput(vp); 1361 MNT_ILOCK(mp); 1362 } 1363 MNT_IUNLOCK(mp); 1364 /* 1365 * Force stale filesystem control information to be flushed. 1366 */ 1367 if (waitfor == MNT_WAIT) { 1368 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1369 allerror = error; 1370 /* Flushed work items may create new vnodes to clean */ 1371 if (allerror == 0 && count) { 1372 MNT_ILOCK(mp); 1373 goto loop; 1374 } 1375 } 1376 #ifdef QUOTA 1377 qsync(mp); 1378 #endif 1379 devvp = ump->um_devvp; 1380 bo = &devvp->v_bufobj; 1381 BO_LOCK(bo); 1382 if (waitfor != MNT_LAZY && 1383 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1384 BO_UNLOCK(bo); 1385 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1386 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1387 allerror = error; 1388 VOP_UNLOCK(devvp, 0); 1389 if (allerror == 0 && waitfor == MNT_WAIT) { 1390 MNT_ILOCK(mp); 1391 goto loop; 1392 } 1393 } else if (suspend != 0) { 1394 if (softdep_check_suspend(mp, 1395 devvp, 1396 softdep_deps, 1397 softdep_accdeps, 1398 secondary_writes, 1399 secondary_accwrites) != 0) 1400 goto loop; /* More work needed */ 1401 mtx_assert(MNT_MTX(mp), MA_OWNED); 1402 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1403 MNT_IUNLOCK(mp); 1404 suspended = 1; 1405 } else 1406 BO_UNLOCK(bo); 1407 /* 1408 * Write back modified superblock. 1409 */ 1410 if (fs->fs_fmod != 0 && 1411 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1412 allerror = error; 1413 return (allerror); 1414 } 1415 1416 int 1417 ffs_vget(mp, ino, flags, vpp) 1418 struct mount *mp; 1419 ino_t ino; 1420 int flags; 1421 struct vnode **vpp; 1422 { 1423 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1424 } 1425 1426 int 1427 ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1428 struct mount *mp; 1429 ino_t ino; 1430 int flags; 1431 struct vnode **vpp; 1432 int ffs_flags; 1433 { 1434 struct fs *fs; 1435 struct inode *ip; 1436 struct ufsmount *ump; 1437 struct buf *bp; 1438 struct vnode *vp; 1439 struct cdev *dev; 1440 int error; 1441 1442 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1443 if (error || *vpp != NULL) 1444 return (error); 1445 1446 /* 1447 * We must promote to an exclusive lock for vnode creation. This 1448 * can happen if lookup is passed LOCKSHARED. 1449 */ 1450 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1451 flags &= ~LK_TYPE_MASK; 1452 flags |= LK_EXCLUSIVE; 1453 } 1454 1455 /* 1456 * We do not lock vnode creation as it is believed to be too 1457 * expensive for such rare case as simultaneous creation of vnode 1458 * for same ino by different processes. We just allow them to race 1459 * and check later to decide who wins. Let the race begin! 1460 */ 1461 1462 ump = VFSTOUFS(mp); 1463 dev = ump->um_dev; 1464 fs = ump->um_fs; 1465 1466 /* 1467 * If this malloc() is performed after the getnewvnode() 1468 * it might block, leaving a vnode with a NULL v_data to be 1469 * found by ffs_sync() if a sync happens to fire right then, 1470 * which will cause a panic because ffs_sync() blindly 1471 * dereferences vp->v_data (as well it should). 1472 */ 1473 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1474 1475 /* Allocate a new vnode/inode. */ 1476 if (fs->fs_magic == FS_UFS1_MAGIC) 1477 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1478 else 1479 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1480 if (error) { 1481 *vpp = NULL; 1482 uma_zfree(uma_inode, ip); 1483 return (error); 1484 } 1485 /* 1486 * FFS supports recursive locking. 1487 */ 1488 VN_LOCK_AREC(vp); 1489 vp->v_data = ip; 1490 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1491 ip->i_vnode = vp; 1492 ip->i_ump = ump; 1493 ip->i_fs = fs; 1494 ip->i_dev = dev; 1495 ip->i_number = ino; 1496 ip->i_ea_refs = 0; 1497 #ifdef QUOTA 1498 { 1499 int i; 1500 for (i = 0; i < MAXQUOTAS; i++) 1501 ip->i_dquot[i] = NODQUOT; 1502 } 1503 #endif 1504 1505 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1506 if (ffs_flags & FFSV_FORCEINSMQ) 1507 vp->v_vflag |= VV_FORCEINSMQ; 1508 error = insmntque(vp, mp); 1509 if (error != 0) { 1510 uma_zfree(uma_inode, ip); 1511 *vpp = NULL; 1512 return (error); 1513 } 1514 vp->v_vflag &= ~VV_FORCEINSMQ; 1515 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1516 if (error || *vpp != NULL) 1517 return (error); 1518 1519 /* Read in the disk contents for the inode, copy into the inode. */ 1520 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1521 (int)fs->fs_bsize, NOCRED, &bp); 1522 if (error) { 1523 /* 1524 * The inode does not contain anything useful, so it would 1525 * be misleading to leave it on its hash chain. With mode 1526 * still zero, it will be unlinked and returned to the free 1527 * list by vput(). 1528 */ 1529 brelse(bp); 1530 vput(vp); 1531 *vpp = NULL; 1532 return (error); 1533 } 1534 if (ip->i_ump->um_fstype == UFS1) 1535 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1536 else 1537 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1538 ffs_load_inode(bp, ip, fs, ino); 1539 if (DOINGSOFTDEP(vp)) 1540 softdep_load_inodeblock(ip); 1541 else 1542 ip->i_effnlink = ip->i_nlink; 1543 bqrelse(bp); 1544 1545 /* 1546 * Initialize the vnode from the inode, check for aliases. 1547 * Note that the underlying vnode may have changed. 1548 */ 1549 if (ip->i_ump->um_fstype == UFS1) 1550 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1551 else 1552 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1553 if (error) { 1554 vput(vp); 1555 *vpp = NULL; 1556 return (error); 1557 } 1558 1559 /* 1560 * Finish inode initialization. 1561 */ 1562 if (vp->v_type != VFIFO) { 1563 /* FFS supports shared locking for all files except fifos. */ 1564 VN_LOCK_ASHARE(vp); 1565 } 1566 1567 /* 1568 * Set up a generation number for this inode if it does not 1569 * already have one. This should only happen on old filesystems. 1570 */ 1571 if (ip->i_gen == 0) { 1572 ip->i_gen = arc4random() / 2 + 1; 1573 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1574 ip->i_flag |= IN_MODIFIED; 1575 DIP_SET(ip, i_gen, ip->i_gen); 1576 } 1577 } 1578 /* 1579 * Ensure that uid and gid are correct. This is a temporary 1580 * fix until fsck has been changed to do the update. 1581 */ 1582 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1583 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1584 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1585 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1586 } /* XXX */ 1587 1588 #ifdef MAC 1589 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1590 /* 1591 * If this vnode is already allocated, and we're running 1592 * multi-label, attempt to perform a label association 1593 * from the extended attributes on the inode. 1594 */ 1595 error = mac_vnode_associate_extattr(mp, vp); 1596 if (error) { 1597 /* ufs_inactive will release ip->i_devvp ref. */ 1598 vput(vp); 1599 *vpp = NULL; 1600 return (error); 1601 } 1602 } 1603 #endif 1604 1605 *vpp = vp; 1606 return (0); 1607 } 1608 1609 /* 1610 * File handle to vnode 1611 * 1612 * Have to be really careful about stale file handles: 1613 * - check that the inode number is valid 1614 * - call ffs_vget() to get the locked inode 1615 * - check for an unallocated inode (i_mode == 0) 1616 * - check that the given client host has export rights and return 1617 * those rights via. exflagsp and credanonp 1618 */ 1619 static int 1620 ffs_fhtovp(mp, fhp, vpp) 1621 struct mount *mp; 1622 struct fid *fhp; 1623 struct vnode **vpp; 1624 { 1625 struct ufid *ufhp; 1626 struct fs *fs; 1627 1628 ufhp = (struct ufid *)fhp; 1629 fs = VFSTOUFS(mp)->um_fs; 1630 if (ufhp->ufid_ino < ROOTINO || 1631 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1632 return (ESTALE); 1633 return (ufs_fhtovp(mp, ufhp, vpp)); 1634 } 1635 1636 /* 1637 * Initialize the filesystem. 1638 */ 1639 static int 1640 ffs_init(vfsp) 1641 struct vfsconf *vfsp; 1642 { 1643 1644 softdep_initialize(); 1645 return (ufs_init(vfsp)); 1646 } 1647 1648 /* 1649 * Undo the work of ffs_init(). 1650 */ 1651 static int 1652 ffs_uninit(vfsp) 1653 struct vfsconf *vfsp; 1654 { 1655 int ret; 1656 1657 ret = ufs_uninit(vfsp); 1658 softdep_uninitialize(); 1659 return (ret); 1660 } 1661 1662 /* 1663 * Write a superblock and associated information back to disk. 1664 */ 1665 int 1666 ffs_sbupdate(mp, waitfor, suspended) 1667 struct ufsmount *mp; 1668 int waitfor; 1669 int suspended; 1670 { 1671 struct fs *fs = mp->um_fs; 1672 struct buf *sbbp; 1673 struct buf *bp; 1674 int blks; 1675 void *space; 1676 int i, size, error, allerror = 0; 1677 1678 if (fs->fs_ronly == 1 && 1679 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1680 (MNT_RDONLY | MNT_UPDATE)) 1681 panic("ffs_sbupdate: write read-only filesystem"); 1682 /* 1683 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1684 */ 1685 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1686 0, 0, 0); 1687 /* 1688 * First write back the summary information. 1689 */ 1690 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1691 space = fs->fs_csp; 1692 for (i = 0; i < blks; i += fs->fs_frag) { 1693 size = fs->fs_bsize; 1694 if (i + fs->fs_frag > blks) 1695 size = (blks - i) * fs->fs_fsize; 1696 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1697 size, 0, 0, 0); 1698 bcopy(space, bp->b_data, (u_int)size); 1699 space = (char *)space + size; 1700 if (suspended) 1701 bp->b_flags |= B_VALIDSUSPWRT; 1702 if (waitfor != MNT_WAIT) 1703 bawrite(bp); 1704 else if ((error = bwrite(bp)) != 0) 1705 allerror = error; 1706 } 1707 /* 1708 * Now write back the superblock itself. If any errors occurred 1709 * up to this point, then fail so that the superblock avoids 1710 * being written out as clean. 1711 */ 1712 if (allerror) { 1713 brelse(sbbp); 1714 return (allerror); 1715 } 1716 bp = sbbp; 1717 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1718 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1719 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1720 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1721 fs->fs_sblockloc = SBLOCK_UFS1; 1722 } 1723 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 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_UFS2); 1727 fs->fs_sblockloc = SBLOCK_UFS2; 1728 } 1729 fs->fs_fmod = 0; 1730 fs->fs_time = time_second; 1731 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1732 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1733 if (suspended) 1734 bp->b_flags |= B_VALIDSUSPWRT; 1735 if (waitfor != MNT_WAIT) 1736 bawrite(bp); 1737 else if ((error = bwrite(bp)) != 0) 1738 allerror = error; 1739 return (allerror); 1740 } 1741 1742 static int 1743 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1744 int attrnamespace, const char *attrname) 1745 { 1746 1747 #ifdef UFS_EXTATTR 1748 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1749 attrname)); 1750 #else 1751 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1752 attrname)); 1753 #endif 1754 } 1755 1756 static void 1757 ffs_ifree(struct ufsmount *ump, struct inode *ip) 1758 { 1759 1760 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1761 uma_zfree(uma_ufs1, ip->i_din1); 1762 else if (ip->i_din2 != NULL) 1763 uma_zfree(uma_ufs2, ip->i_din2); 1764 uma_zfree(uma_inode, ip); 1765 } 1766 1767 static int dobkgrdwrite = 1; 1768 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1769 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1770 1771 /* 1772 * Complete a background write started from bwrite. 1773 */ 1774 static void 1775 ffs_backgroundwritedone(struct buf *bp) 1776 { 1777 struct bufobj *bufobj; 1778 struct buf *origbp; 1779 1780 /* 1781 * Find the original buffer that we are writing. 1782 */ 1783 bufobj = bp->b_bufobj; 1784 BO_LOCK(bufobj); 1785 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1786 panic("backgroundwritedone: lost buffer"); 1787 /* Grab an extra reference to be dropped by the bufdone() below. */ 1788 bufobj_wrefl(bufobj); 1789 BO_UNLOCK(bufobj); 1790 /* 1791 * Process dependencies then return any unfinished ones. 1792 */ 1793 if (!LIST_EMPTY(&bp->b_dep)) 1794 buf_complete(bp); 1795 #ifdef SOFTUPDATES 1796 if (!LIST_EMPTY(&bp->b_dep)) 1797 softdep_move_dependencies(bp, origbp); 1798 #endif 1799 /* 1800 * This buffer is marked B_NOCACHE so when it is released 1801 * by biodone it will be tossed. 1802 */ 1803 bp->b_flags |= B_NOCACHE; 1804 bp->b_flags &= ~B_CACHE; 1805 bufdone(bp); 1806 BO_LOCK(bufobj); 1807 /* 1808 * Clear the BV_BKGRDINPROG flag in the original buffer 1809 * and awaken it if it is waiting for the write to complete. 1810 * If BV_BKGRDINPROG is not set in the original buffer it must 1811 * have been released and re-instantiated - which is not legal. 1812 */ 1813 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1814 ("backgroundwritedone: lost buffer2")); 1815 origbp->b_vflags &= ~BV_BKGRDINPROG; 1816 if (origbp->b_vflags & BV_BKGRDWAIT) { 1817 origbp->b_vflags &= ~BV_BKGRDWAIT; 1818 wakeup(&origbp->b_xflags); 1819 } 1820 BO_UNLOCK(bufobj); 1821 } 1822 1823 1824 /* 1825 * Write, release buffer on completion. (Done by iodone 1826 * if async). Do not bother writing anything if the buffer 1827 * is invalid. 1828 * 1829 * Note that we set B_CACHE here, indicating that buffer is 1830 * fully valid and thus cacheable. This is true even of NFS 1831 * now so we set it generally. This could be set either here 1832 * or in biodone() since the I/O is synchronous. We put it 1833 * here. 1834 */ 1835 static int 1836 ffs_bufwrite(struct buf *bp) 1837 { 1838 int oldflags, s; 1839 struct buf *newbp; 1840 1841 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1842 if (bp->b_flags & B_INVAL) { 1843 brelse(bp); 1844 return (0); 1845 } 1846 1847 oldflags = bp->b_flags; 1848 1849 if (!BUF_ISLOCKED(bp)) 1850 panic("bufwrite: buffer is not busy???"); 1851 s = splbio(); 1852 /* 1853 * If a background write is already in progress, delay 1854 * writing this block if it is asynchronous. Otherwise 1855 * wait for the background write to complete. 1856 */ 1857 BO_LOCK(bp->b_bufobj); 1858 if (bp->b_vflags & BV_BKGRDINPROG) { 1859 if (bp->b_flags & B_ASYNC) { 1860 BO_UNLOCK(bp->b_bufobj); 1861 splx(s); 1862 bdwrite(bp); 1863 return (0); 1864 } 1865 bp->b_vflags |= BV_BKGRDWAIT; 1866 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1867 if (bp->b_vflags & BV_BKGRDINPROG) 1868 panic("bufwrite: still writing"); 1869 } 1870 BO_UNLOCK(bp->b_bufobj); 1871 1872 /* Mark the buffer clean */ 1873 bundirty(bp); 1874 1875 /* 1876 * If this buffer is marked for background writing and we 1877 * do not have to wait for it, make a copy and write the 1878 * copy so as to leave this buffer ready for further use. 1879 * 1880 * This optimization eats a lot of memory. If we have a page 1881 * or buffer shortfall we can't do it. 1882 */ 1883 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1884 (bp->b_flags & B_ASYNC) && 1885 !vm_page_count_severe() && 1886 !buf_dirty_count_severe()) { 1887 KASSERT(bp->b_iodone == NULL, 1888 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1889 1890 /* get a new block */ 1891 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 1892 if (newbp == NULL) 1893 goto normal_write; 1894 1895 /* 1896 * set it to be identical to the old block. We have to 1897 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1898 * to avoid confusing the splay tree and gbincore(). 1899 */ 1900 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1901 newbp->b_lblkno = bp->b_lblkno; 1902 newbp->b_xflags |= BX_BKGRDMARKER; 1903 BO_LOCK(bp->b_bufobj); 1904 bp->b_vflags |= BV_BKGRDINPROG; 1905 bgetvp(bp->b_vp, newbp); 1906 BO_UNLOCK(bp->b_bufobj); 1907 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1908 newbp->b_blkno = bp->b_blkno; 1909 newbp->b_offset = bp->b_offset; 1910 newbp->b_iodone = ffs_backgroundwritedone; 1911 newbp->b_flags |= B_ASYNC; 1912 newbp->b_flags &= ~B_INVAL; 1913 1914 #ifdef SOFTUPDATES 1915 /* move over the dependencies */ 1916 if (!LIST_EMPTY(&bp->b_dep)) 1917 softdep_move_dependencies(bp, newbp); 1918 #endif 1919 1920 /* 1921 * Initiate write on the copy, release the original to 1922 * the B_LOCKED queue so that it cannot go away until 1923 * the background write completes. If not locked it could go 1924 * away and then be reconstituted while it was being written. 1925 * If the reconstituted buffer were written, we could end up 1926 * with two background copies being written at the same time. 1927 */ 1928 bqrelse(bp); 1929 bp = newbp; 1930 } 1931 1932 /* Let the normal bufwrite do the rest for us */ 1933 normal_write: 1934 return (bufwrite(bp)); 1935 } 1936 1937 1938 static void 1939 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1940 { 1941 struct vnode *vp; 1942 int error; 1943 struct buf *tbp; 1944 1945 vp = bo->__bo_vnode; 1946 if (bp->b_iocmd == BIO_WRITE) { 1947 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1948 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1949 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1950 panic("ffs_geom_strategy: bad I/O"); 1951 bp->b_flags &= ~B_VALIDSUSPWRT; 1952 if ((vp->v_vflag & VV_COPYONWRITE) && 1953 vp->v_rdev->si_snapdata != NULL) { 1954 if ((bp->b_flags & B_CLUSTER) != 0) { 1955 runningbufwakeup(bp); 1956 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1957 b_cluster.cluster_entry) { 1958 error = ffs_copyonwrite(vp, tbp); 1959 if (error != 0 && 1960 error != EOPNOTSUPP) { 1961 bp->b_error = error; 1962 bp->b_ioflags |= BIO_ERROR; 1963 bufdone(bp); 1964 return; 1965 } 1966 } 1967 bp->b_runningbufspace = bp->b_bufsize; 1968 atomic_add_long(&runningbufspace, 1969 bp->b_runningbufspace); 1970 } else { 1971 error = ffs_copyonwrite(vp, bp); 1972 if (error != 0 && error != EOPNOTSUPP) { 1973 bp->b_error = error; 1974 bp->b_ioflags |= BIO_ERROR; 1975 bufdone(bp); 1976 return; 1977 } 1978 } 1979 } 1980 #ifdef SOFTUPDATES 1981 if ((bp->b_flags & B_CLUSTER) != 0) { 1982 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 1983 b_cluster.cluster_entry) { 1984 if (!LIST_EMPTY(&tbp->b_dep)) 1985 buf_start(tbp); 1986 } 1987 } else { 1988 if (!LIST_EMPTY(&bp->b_dep)) 1989 buf_start(bp); 1990 } 1991 1992 #endif 1993 } 1994 g_vfs_strategy(bo, bp); 1995 } 1996 1997 #ifdef DDB 1998 1999 static void 2000 db_print_ffs(struct ufsmount *ump) 2001 { 2002 db_printf("mp %p %s devvp %p fs %p su_wl %d su_wl_in %d su_deps %d " 2003 "su_req %d\n", 2004 ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname, 2005 ump->um_devvp, ump->um_fs, ump->softdep_on_worklist, 2006 ump->softdep_on_worklist_inprogress, ump->softdep_deps, 2007 ump->softdep_req); 2008 } 2009 2010 DB_SHOW_COMMAND(ffs, db_show_ffs) 2011 { 2012 struct mount *mp; 2013 struct ufsmount *ump; 2014 2015 if (have_addr) { 2016 ump = VFSTOUFS((struct mount *)addr); 2017 db_print_ffs(ump); 2018 return; 2019 } 2020 2021 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2022 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2023 db_print_ffs(VFSTOUFS(mp)); 2024 } 2025 } 2026 2027 #endif /* DDB */ 2028