1 /*- 2 * Copyright (c) 1989, 1991, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_mac.h" 36 #include "opt_quota.h" 37 #include "opt_ufs.h" 38 #include "opt_ffs.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/namei.h> 43 #include <sys/proc.h> 44 #include <sys/kernel.h> 45 #include <sys/mac.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 <ufs/ufs/extattr.h> 56 #include <ufs/ufs/quota.h> 57 #include <ufs/ufs/ufsmount.h> 58 #include <ufs/ufs/inode.h> 59 #include <ufs/ufs/ufs_extern.h> 60 61 #include <ufs/ffs/fs.h> 62 #include <ufs/ffs/ffs_extern.h> 63 64 #include <vm/vm.h> 65 #include <vm/uma.h> 66 #include <vm/vm_page.h> 67 68 #include <geom/geom.h> 69 #include <geom/geom_vfs.h> 70 71 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 72 73 static int ffs_sbupdate(struct ufsmount *, int); 74 static int ffs_reload(struct mount *, struct thread *); 75 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 76 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 77 ufs2_daddr_t); 78 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *); 79 static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 80 static vfs_init_t ffs_init; 81 static vfs_uninit_t ffs_uninit; 82 static vfs_extattrctl_t ffs_extattrctl; 83 static vfs_cmount_t ffs_cmount; 84 static vfs_unmount_t ffs_unmount; 85 static vfs_mount_t ffs_mount; 86 static vfs_statfs_t ffs_statfs; 87 static vfs_fhtovp_t ffs_fhtovp; 88 static vfs_vptofh_t ffs_vptofh; 89 static vfs_sync_t ffs_sync; 90 91 static struct vfsops ufs_vfsops = { 92 .vfs_extattrctl = ffs_extattrctl, 93 .vfs_fhtovp = ffs_fhtovp, 94 .vfs_init = ffs_init, 95 .vfs_mount = ffs_mount, 96 .vfs_cmount = ffs_cmount, 97 .vfs_quotactl = ufs_quotactl, 98 .vfs_root = ufs_root, 99 .vfs_statfs = ffs_statfs, 100 .vfs_sync = ffs_sync, 101 .vfs_uninit = ffs_uninit, 102 .vfs_unmount = ffs_unmount, 103 .vfs_vget = ffs_vget, 104 .vfs_vptofh = ffs_vptofh, 105 }; 106 107 VFS_SET(ufs_vfsops, ufs, 0); 108 109 static b_strategy_t ffs_geom_strategy; 110 static b_write_t ffs_bufwrite; 111 112 static struct buf_ops ffs_ops = { 113 .bop_name = "FFS", 114 .bop_write = ffs_bufwrite, 115 .bop_strategy = ffs_geom_strategy, 116 .bop_sync = bufsync, 117 }; 118 119 static const char *ffs_opts[] = { "from", "export", NULL }; 120 121 static int 122 ffs_mount(struct mount *mp, struct thread *td) 123 { 124 struct vnode *devvp; 125 struct ufsmount *ump = 0; 126 struct fs *fs; 127 int error, flags; 128 mode_t accessmode; 129 struct nameidata ndp; 130 struct export_args export; 131 char *fspec; 132 133 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 134 return (EINVAL); 135 if (uma_inode == NULL) { 136 uma_inode = uma_zcreate("FFS inode", 137 sizeof(struct inode), NULL, NULL, NULL, NULL, 138 UMA_ALIGN_PTR, 0); 139 uma_ufs1 = uma_zcreate("FFS1 dinode", 140 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 141 UMA_ALIGN_PTR, 0); 142 uma_ufs2 = uma_zcreate("FFS2 dinode", 143 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 144 UMA_ALIGN_PTR, 0); 145 } 146 147 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 148 if (error) 149 return (error); 150 151 /* 152 * If updating, check whether changing from read-only to 153 * read/write; if there is no device name, that's all we do. 154 */ 155 if (mp->mnt_flag & MNT_UPDATE) { 156 ump = VFSTOUFS(mp); 157 fs = ump->um_fs; 158 devvp = ump->um_devvp; 159 if (fs->fs_ronly == 0 && 160 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 161 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 162 return (error); 163 /* 164 * Flush any dirty data. 165 */ 166 if ((error = ffs_sync(mp, MNT_WAIT, td)) != 0) { 167 vn_finished_write(mp); 168 return (error); 169 } 170 /* 171 * Check for and optionally get rid of files open 172 * for writing. 173 */ 174 flags = WRITECLOSE; 175 if (mp->mnt_flag & MNT_FORCE) 176 flags |= FORCECLOSE; 177 if (mp->mnt_flag & MNT_SOFTDEP) { 178 error = softdep_flushfiles(mp, flags, td); 179 } else { 180 error = ffs_flushfiles(mp, flags, td); 181 } 182 if (error) { 183 vn_finished_write(mp); 184 return (error); 185 } 186 if (fs->fs_pendingblocks != 0 || 187 fs->fs_pendinginodes != 0) { 188 printf("%s: %s: blocks %jd files %d\n", 189 fs->fs_fsmnt, "update error", 190 (intmax_t)fs->fs_pendingblocks, 191 fs->fs_pendinginodes); 192 fs->fs_pendingblocks = 0; 193 fs->fs_pendinginodes = 0; 194 } 195 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 196 fs->fs_clean = 1; 197 if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) { 198 fs->fs_ronly = 0; 199 fs->fs_clean = 0; 200 vn_finished_write(mp); 201 return (error); 202 } 203 vn_finished_write(mp); 204 DROP_GIANT(); 205 g_topology_lock(); 206 g_access(ump->um_cp, 0, -1, 0); 207 g_topology_unlock(); 208 PICKUP_GIANT(); 209 fs->fs_ronly = 1; 210 mp->mnt_flag |= MNT_RDONLY; 211 } 212 if ((mp->mnt_flag & MNT_RELOAD) && 213 (error = ffs_reload(mp, td)) != 0) 214 return (error); 215 if (fs->fs_ronly && 216 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 217 /* 218 * If upgrade to read-write by non-root, then verify 219 * that user has necessary permissions on the device. 220 */ 221 if (suser(td)) { 222 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 223 if ((error = VOP_ACCESS(devvp, VREAD | VWRITE, 224 td->td_ucred, td)) != 0) { 225 VOP_UNLOCK(devvp, 0, td); 226 return (error); 227 } 228 VOP_UNLOCK(devvp, 0, td); 229 } 230 fs->fs_flags &= ~FS_UNCLEAN; 231 if (fs->fs_clean == 0) { 232 fs->fs_flags |= FS_UNCLEAN; 233 if ((mp->mnt_flag & MNT_FORCE) || 234 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 235 (fs->fs_flags & FS_DOSOFTDEP))) { 236 printf("WARNING: %s was not %s\n", 237 fs->fs_fsmnt, "properly dismounted"); 238 } else { 239 printf( 240 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 241 fs->fs_fsmnt); 242 return (EPERM); 243 } 244 } 245 DROP_GIANT(); 246 g_topology_lock(); 247 /* 248 * If we're the root device, we may not have an E count 249 * yet, get it now. 250 */ 251 if (ump->um_cp->ace == 0) 252 error = g_access(ump->um_cp, 0, 1, 1); 253 else 254 error = g_access(ump->um_cp, 0, 1, 0); 255 g_topology_unlock(); 256 PICKUP_GIANT(); 257 if (error) 258 return (error); 259 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 260 return (error); 261 fs->fs_ronly = 0; 262 mp->mnt_flag &= ~MNT_RDONLY; 263 fs->fs_clean = 0; 264 if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) { 265 vn_finished_write(mp); 266 return (error); 267 } 268 /* check to see if we need to start softdep */ 269 if ((fs->fs_flags & FS_DOSOFTDEP) && 270 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 271 vn_finished_write(mp); 272 return (error); 273 } 274 if (fs->fs_snapinum[0] != 0) 275 ffs_snapshot_mount(mp); 276 vn_finished_write(mp); 277 } 278 /* 279 * Soft updates is incompatible with "async", 280 * so if we are doing softupdates stop the user 281 * from setting the async flag in an update. 282 * Softdep_mount() clears it in an initial mount 283 * or ro->rw remount. 284 */ 285 if (mp->mnt_flag & MNT_SOFTDEP) 286 mp->mnt_flag &= ~MNT_ASYNC; 287 /* 288 * Keep MNT_ACLS flag if it is stored in superblock. 289 */ 290 if ((fs->fs_flags & FS_ACLS) != 0) 291 mp->mnt_flag |= MNT_ACLS; 292 /* 293 * If not updating name, process export requests. 294 */ 295 error = vfs_copyopt(mp->mnt_optnew, "export", &export, sizeof export); 296 if (error == 0 && export.ex_flags != 0) 297 return (vfs_export(mp, &export)); 298 /* 299 * If this is a snapshot request, take the snapshot. 300 */ 301 if (mp->mnt_flag & MNT_SNAPSHOT) 302 return (ffs_snapshot(mp, fspec)); 303 } 304 305 /* 306 * Not an update, or updating the name: look up the name 307 * and verify that it refers to a sensible disk device. 308 */ 309 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 310 if ((error = namei(&ndp)) != 0) 311 return (error); 312 NDFREE(&ndp, NDF_ONLY_PNBUF); 313 devvp = ndp.ni_vp; 314 if (!vn_isdisk(devvp, &error)) { 315 vput(devvp); 316 return (error); 317 } 318 319 /* 320 * If mount by non-root, then verify that user has necessary 321 * permissions on the device. 322 */ 323 if (suser(td)) { 324 accessmode = VREAD; 325 if ((mp->mnt_flag & MNT_RDONLY) == 0) 326 accessmode |= VWRITE; 327 if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){ 328 vput(devvp); 329 return (error); 330 } 331 } 332 333 if (mp->mnt_flag & MNT_UPDATE) { 334 /* 335 * Update only 336 * 337 * If it's not the same vnode, or at least the same device 338 * then it's not correct. 339 */ 340 341 if (devvp->v_rdev != ump->um_devvp->v_rdev) 342 error = EINVAL; /* needs translation */ 343 vput(devvp); 344 if (error) 345 return (error); 346 } else { 347 /* 348 * New mount 349 * 350 * We need the name for the mount point (also used for 351 * "last mounted on") copied in. If an error occurs, 352 * the mount point is discarded by the upper level code. 353 * Note that vfs_mount() populates f_mntonname for us. 354 */ 355 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 356 vrele(devvp); 357 return (error); 358 } 359 } 360 vfs_mountedfrom(mp, fspec); 361 return (0); 362 } 363 364 /* 365 * Compatibility with old mount system call. 366 */ 367 368 static int 369 ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td) 370 { 371 struct ufs_args args; 372 int error; 373 374 if (data == NULL) 375 return (EINVAL); 376 error = copyin(data, &args, sizeof args); 377 if (error) 378 return (error); 379 380 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 381 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 382 error = kernel_mount(ma, flags); 383 384 return (error); 385 } 386 387 /* 388 * Reload all incore data for a filesystem (used after running fsck on 389 * the root filesystem and finding things to fix). The filesystem must 390 * be mounted read-only. 391 * 392 * Things to do to update the mount: 393 * 1) invalidate all cached meta-data. 394 * 2) re-read superblock from disk. 395 * 3) re-read summary information from disk. 396 * 4) invalidate all inactive vnodes. 397 * 5) invalidate all cached file data. 398 * 6) re-read inode data for all active vnodes. 399 */ 400 static int 401 ffs_reload(struct mount *mp, struct thread *td) 402 { 403 struct vnode *vp, *nvp, *devvp; 404 struct inode *ip; 405 void *space; 406 struct buf *bp; 407 struct fs *fs, *newfs; 408 struct ufsmount *ump; 409 ufs2_daddr_t sblockloc; 410 int i, blks, size, error; 411 int32_t *lp; 412 413 if ((mp->mnt_flag & MNT_RDONLY) == 0) 414 return (EINVAL); 415 ump = VFSTOUFS(mp); 416 /* 417 * Step 1: invalidate all cached meta-data. 418 */ 419 devvp = VFSTOUFS(mp)->um_devvp; 420 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 421 if (vinvalbuf(devvp, 0, td, 0, 0) != 0) 422 panic("ffs_reload: dirty1"); 423 VOP_UNLOCK(devvp, 0, td); 424 425 /* 426 * Step 2: re-read superblock from disk. 427 */ 428 fs = VFSTOUFS(mp)->um_fs; 429 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 430 NOCRED, &bp)) != 0) 431 return (error); 432 newfs = (struct fs *)bp->b_data; 433 if ((newfs->fs_magic != FS_UFS1_MAGIC && 434 newfs->fs_magic != FS_UFS2_MAGIC) || 435 newfs->fs_bsize > MAXBSIZE || 436 newfs->fs_bsize < sizeof(struct fs)) { 437 brelse(bp); 438 return (EIO); /* XXX needs translation */ 439 } 440 /* 441 * Copy pointer fields back into superblock before copying in XXX 442 * new superblock. These should really be in the ufsmount. XXX 443 * Note that important parameters (eg fs_ncg) are unchanged. 444 */ 445 newfs->fs_csp = fs->fs_csp; 446 newfs->fs_maxcluster = fs->fs_maxcluster; 447 newfs->fs_contigdirs = fs->fs_contigdirs; 448 newfs->fs_active = fs->fs_active; 449 /* The file system is still read-only. */ 450 newfs->fs_ronly = 1; 451 sblockloc = fs->fs_sblockloc; 452 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 453 brelse(bp); 454 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 455 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 456 UFS_LOCK(ump); 457 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 458 printf("%s: reload pending error: blocks %jd files %d\n", 459 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 460 fs->fs_pendinginodes); 461 fs->fs_pendingblocks = 0; 462 fs->fs_pendinginodes = 0; 463 } 464 UFS_UNLOCK(ump); 465 466 /* 467 * Step 3: re-read summary information from disk. 468 */ 469 blks = howmany(fs->fs_cssize, fs->fs_fsize); 470 space = fs->fs_csp; 471 for (i = 0; i < blks; i += fs->fs_frag) { 472 size = fs->fs_bsize; 473 if (i + fs->fs_frag > blks) 474 size = (blks - i) * fs->fs_fsize; 475 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 476 NOCRED, &bp); 477 if (error) 478 return (error); 479 bcopy(bp->b_data, space, (u_int)size); 480 space = (char *)space + size; 481 brelse(bp); 482 } 483 /* 484 * We no longer know anything about clusters per cylinder group. 485 */ 486 if (fs->fs_contigsumsize > 0) { 487 lp = fs->fs_maxcluster; 488 for (i = 0; i < fs->fs_ncg; i++) 489 *lp++ = fs->fs_contigsumsize; 490 } 491 492 loop: 493 MNT_ILOCK(mp); 494 MNT_VNODE_FOREACH(vp, mp, nvp) { 495 VI_LOCK(vp); 496 if (vp->v_iflag & VI_DOOMED) { 497 VI_UNLOCK(vp); 498 continue; 499 } 500 MNT_IUNLOCK(mp); 501 /* 502 * Step 4: invalidate all cached file data. 503 */ 504 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 505 goto loop; 506 } 507 if (vinvalbuf(vp, 0, td, 0, 0)) 508 panic("ffs_reload: dirty2"); 509 /* 510 * Step 5: re-read inode data for all active vnodes. 511 */ 512 ip = VTOI(vp); 513 error = 514 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 515 (int)fs->fs_bsize, NOCRED, &bp); 516 if (error) { 517 VOP_UNLOCK(vp, 0, td); 518 vrele(vp); 519 return (error); 520 } 521 ffs_load_inode(bp, ip, fs, ip->i_number); 522 ip->i_effnlink = ip->i_nlink; 523 brelse(bp); 524 VOP_UNLOCK(vp, 0, td); 525 vrele(vp); 526 MNT_ILOCK(mp); 527 } 528 MNT_IUNLOCK(mp); 529 return (0); 530 } 531 532 /* 533 * Possible superblock locations ordered from most to least likely. 534 */ 535 static int sblock_try[] = SBLOCKSEARCH; 536 537 /* 538 * Common code for mount and mountroot 539 */ 540 static int 541 ffs_mountfs(devvp, mp, td) 542 struct vnode *devvp; 543 struct mount *mp; 544 struct thread *td; 545 { 546 struct ufsmount *ump; 547 struct buf *bp; 548 struct fs *fs; 549 struct cdev *dev; 550 void *space; 551 ufs2_daddr_t sblockloc; 552 int error, i, blks, size, ronly; 553 int32_t *lp; 554 struct ucred *cred; 555 struct g_consumer *cp; 556 557 dev = devvp->v_rdev; 558 cred = td ? td->td_ucred : NOCRED; 559 560 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 561 DROP_GIANT(); 562 g_topology_lock(); 563 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 564 565 /* 566 * If we are a root mount, drop the E flag so fsck can do its magic. 567 * We will pick it up again when we remount R/W. 568 */ 569 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 570 error = g_access(cp, 0, 0, -1); 571 g_topology_unlock(); 572 PICKUP_GIANT(); 573 VOP_UNLOCK(devvp, 0, td); 574 if (error) 575 return (error); 576 if (devvp->v_rdev->si_iosize_max != 0) 577 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 578 if (mp->mnt_iosize_max > MAXPHYS) 579 mp->mnt_iosize_max = MAXPHYS; 580 581 devvp->v_bufobj.bo_private = cp; 582 devvp->v_bufobj.bo_ops = &ffs_ops; 583 584 bp = NULL; 585 ump = NULL; 586 fs = NULL; 587 sblockloc = 0; 588 /* 589 * Try reading the superblock in each of its possible locations. 590 */ 591 for (i = 0; sblock_try[i] != -1; i++) { 592 if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE, 593 cred, &bp)) != 0) 594 goto out; 595 fs = (struct fs *)bp->b_data; 596 sblockloc = sblock_try[i]; 597 if ((fs->fs_magic == FS_UFS1_MAGIC || 598 (fs->fs_magic == FS_UFS2_MAGIC && 599 (fs->fs_sblockloc == sblockloc || 600 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 601 fs->fs_bsize <= MAXBSIZE && 602 fs->fs_bsize >= sizeof(struct fs)) 603 break; 604 brelse(bp); 605 bp = NULL; 606 } 607 if (sblock_try[i] == -1) { 608 error = EINVAL; /* XXX needs translation */ 609 goto out; 610 } 611 fs->fs_fmod = 0; 612 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 613 fs->fs_flags &= ~FS_UNCLEAN; 614 if (fs->fs_clean == 0) { 615 fs->fs_flags |= FS_UNCLEAN; 616 if (ronly || (mp->mnt_flag & MNT_FORCE) || 617 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 618 (fs->fs_flags & FS_DOSOFTDEP))) { 619 printf( 620 "WARNING: %s was not properly dismounted\n", 621 fs->fs_fsmnt); 622 } else { 623 printf( 624 "WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 625 fs->fs_fsmnt); 626 error = EPERM; 627 goto out; 628 } 629 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 630 (mp->mnt_flag & MNT_FORCE)) { 631 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 632 (intmax_t)fs->fs_pendingblocks, 633 fs->fs_pendinginodes); 634 fs->fs_pendingblocks = 0; 635 fs->fs_pendinginodes = 0; 636 } 637 } 638 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 639 printf("%s: mount pending error: blocks %jd files %d\n", 640 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 641 fs->fs_pendinginodes); 642 fs->fs_pendingblocks = 0; 643 fs->fs_pendinginodes = 0; 644 } 645 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 646 ump->um_cp = cp; 647 ump->um_bo = &devvp->v_bufobj; 648 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 649 if (fs->fs_magic == FS_UFS1_MAGIC) { 650 ump->um_fstype = UFS1; 651 ump->um_balloc = ffs_balloc_ufs1; 652 } else { 653 ump->um_fstype = UFS2; 654 ump->um_balloc = ffs_balloc_ufs2; 655 } 656 ump->um_blkatoff = ffs_blkatoff; 657 ump->um_truncate = ffs_truncate; 658 ump->um_update = ffs_update; 659 ump->um_valloc = ffs_valloc; 660 ump->um_vfree = ffs_vfree; 661 ump->um_ifree = ffs_ifree; 662 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 663 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 664 if (fs->fs_sbsize < SBLOCKSIZE) 665 bp->b_flags |= B_INVAL | B_NOCACHE; 666 brelse(bp); 667 bp = NULL; 668 fs = ump->um_fs; 669 ffs_oldfscompat_read(fs, ump, sblockloc); 670 fs->fs_ronly = ronly; 671 size = fs->fs_cssize; 672 blks = howmany(size, fs->fs_fsize); 673 if (fs->fs_contigsumsize > 0) 674 size += fs->fs_ncg * sizeof(int32_t); 675 size += fs->fs_ncg * sizeof(u_int8_t); 676 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 677 fs->fs_csp = space; 678 for (i = 0; i < blks; i += fs->fs_frag) { 679 size = fs->fs_bsize; 680 if (i + fs->fs_frag > blks) 681 size = (blks - i) * fs->fs_fsize; 682 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 683 cred, &bp)) != 0) { 684 free(fs->fs_csp, M_UFSMNT); 685 goto out; 686 } 687 bcopy(bp->b_data, space, (u_int)size); 688 space = (char *)space + size; 689 brelse(bp); 690 bp = NULL; 691 } 692 if (fs->fs_contigsumsize > 0) { 693 fs->fs_maxcluster = lp = space; 694 for (i = 0; i < fs->fs_ncg; i++) 695 *lp++ = fs->fs_contigsumsize; 696 space = lp; 697 } 698 size = fs->fs_ncg * sizeof(u_int8_t); 699 fs->fs_contigdirs = (u_int8_t *)space; 700 bzero(fs->fs_contigdirs, size); 701 fs->fs_active = NULL; 702 mp->mnt_data = (qaddr_t)ump; 703 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 704 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 705 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 706 vfs_getvfs(&mp->mnt_stat.f_fsid)) 707 vfs_getnewfsid(mp); 708 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 709 mp->mnt_flag |= MNT_LOCAL; 710 if ((fs->fs_flags & FS_MULTILABEL) != 0) 711 #ifdef MAC 712 mp->mnt_flag |= MNT_MULTILABEL; 713 #else 714 printf( 715 "WARNING: %s: multilabel flag on fs but no MAC support\n", 716 fs->fs_fsmnt); 717 #endif 718 if ((fs->fs_flags & FS_ACLS) != 0) 719 #ifdef UFS_ACL 720 mp->mnt_flag |= MNT_ACLS; 721 #else 722 printf( 723 "WARNING: %s: ACLs flag on fs but no ACLs support\n", 724 fs->fs_fsmnt); 725 #endif 726 ump->um_mountp = mp; 727 ump->um_dev = dev; 728 ump->um_devvp = devvp; 729 ump->um_nindir = fs->fs_nindir; 730 ump->um_bptrtodb = fs->fs_fsbtodb; 731 ump->um_seqinc = fs->fs_frag; 732 for (i = 0; i < MAXQUOTAS; i++) 733 ump->um_quotas[i] = NULLVP; 734 #ifdef UFS_EXTATTR 735 ufs_extattr_uepm_init(&ump->um_extattr); 736 #endif 737 /* 738 * Set FS local "last mounted on" information (NULL pad) 739 */ 740 bzero(fs->fs_fsmnt, MAXMNTLEN); 741 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 742 743 if( mp->mnt_flag & MNT_ROOTFS) { 744 /* 745 * Root mount; update timestamp in mount structure. 746 * this will be used by the common root mount code 747 * to update the system clock. 748 */ 749 mp->mnt_time = fs->fs_time; 750 } 751 752 if (ronly == 0) { 753 if ((fs->fs_flags & FS_DOSOFTDEP) && 754 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 755 free(fs->fs_csp, M_UFSMNT); 756 goto out; 757 } 758 if (fs->fs_snapinum[0] != 0) 759 ffs_snapshot_mount(mp); 760 fs->fs_fmod = 1; 761 fs->fs_clean = 0; 762 (void) ffs_sbupdate(ump, MNT_WAIT); 763 } 764 /* 765 * Initialize filesystem stat information in mount struct. 766 */ 767 #ifdef UFS_EXTATTR 768 #ifdef UFS_EXTATTR_AUTOSTART 769 /* 770 * 771 * Auto-starting does the following: 772 * - check for /.attribute in the fs, and extattr_start if so 773 * - for each file in .attribute, enable that file with 774 * an attribute of the same name. 775 * Not clear how to report errors -- probably eat them. 776 * This would all happen while the filesystem was busy/not 777 * available, so would effectively be "atomic". 778 */ 779 (void) ufs_extattr_autostart(mp, td); 780 #endif /* !UFS_EXTATTR_AUTOSTART */ 781 #endif /* !UFS_EXTATTR */ 782 #ifndef QUOTA 783 mp->mnt_kern_flag |= MNTK_MPSAFE; 784 #endif 785 return (0); 786 out: 787 if (bp) 788 brelse(bp); 789 if (cp != NULL) { 790 DROP_GIANT(); 791 g_topology_lock(); 792 g_vfs_close(cp, td); 793 g_topology_unlock(); 794 PICKUP_GIANT(); 795 } 796 if (ump) { 797 mtx_destroy(UFS_MTX(ump)); 798 free(ump->um_fs, M_UFSMNT); 799 free(ump, M_UFSMNT); 800 mp->mnt_data = (qaddr_t)0; 801 } 802 return (error); 803 } 804 805 #include <sys/sysctl.h> 806 static int bigcgs = 0; 807 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 808 809 /* 810 * Sanity checks for loading old filesystem superblocks. 811 * See ffs_oldfscompat_write below for unwound actions. 812 * 813 * XXX - Parts get retired eventually. 814 * Unfortunately new bits get added. 815 */ 816 static void 817 ffs_oldfscompat_read(fs, ump, sblockloc) 818 struct fs *fs; 819 struct ufsmount *ump; 820 ufs2_daddr_t sblockloc; 821 { 822 off_t maxfilesize; 823 824 /* 825 * If not yet done, update fs_flags location and value of fs_sblockloc. 826 */ 827 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 828 fs->fs_flags = fs->fs_old_flags; 829 fs->fs_old_flags |= FS_FLAGS_UPDATED; 830 fs->fs_sblockloc = sblockloc; 831 } 832 /* 833 * If not yet done, update UFS1 superblock with new wider fields. 834 */ 835 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 836 fs->fs_maxbsize = fs->fs_bsize; 837 fs->fs_time = fs->fs_old_time; 838 fs->fs_size = fs->fs_old_size; 839 fs->fs_dsize = fs->fs_old_dsize; 840 fs->fs_csaddr = fs->fs_old_csaddr; 841 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 842 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 843 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 844 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 845 } 846 if (fs->fs_magic == FS_UFS1_MAGIC && 847 fs->fs_old_inodefmt < FS_44INODEFMT) { 848 fs->fs_maxfilesize = (u_quad_t) 1LL << 39; 849 fs->fs_qbmask = ~fs->fs_bmask; 850 fs->fs_qfmask = ~fs->fs_fmask; 851 } 852 if (fs->fs_magic == FS_UFS1_MAGIC) { 853 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 854 maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1; 855 if (fs->fs_maxfilesize > maxfilesize) 856 fs->fs_maxfilesize = maxfilesize; 857 } 858 /* Compatibility for old filesystems */ 859 if (fs->fs_avgfilesize <= 0) 860 fs->fs_avgfilesize = AVFILESIZ; 861 if (fs->fs_avgfpdir <= 0) 862 fs->fs_avgfpdir = AFPDIR; 863 if (bigcgs) { 864 fs->fs_save_cgsize = fs->fs_cgsize; 865 fs->fs_cgsize = fs->fs_bsize; 866 } 867 } 868 869 /* 870 * Unwinding superblock updates for old filesystems. 871 * See ffs_oldfscompat_read above for details. 872 * 873 * XXX - Parts get retired eventually. 874 * Unfortunately new bits get added. 875 */ 876 static void 877 ffs_oldfscompat_write(fs, ump) 878 struct fs *fs; 879 struct ufsmount *ump; 880 { 881 882 /* 883 * Copy back UFS2 updated fields that UFS1 inspects. 884 */ 885 if (fs->fs_magic == FS_UFS1_MAGIC) { 886 fs->fs_old_time = fs->fs_time; 887 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 888 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 889 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 890 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 891 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 892 } 893 if (bigcgs) { 894 fs->fs_cgsize = fs->fs_save_cgsize; 895 fs->fs_save_cgsize = 0; 896 } 897 } 898 899 /* 900 * unmount system call 901 */ 902 static int 903 ffs_unmount(mp, mntflags, td) 904 struct mount *mp; 905 int mntflags; 906 struct thread *td; 907 { 908 struct ufsmount *ump = VFSTOUFS(mp); 909 struct fs *fs; 910 int error, flags; 911 912 flags = 0; 913 if (mntflags & MNT_FORCE) { 914 flags |= FORCECLOSE; 915 } 916 #ifdef UFS_EXTATTR 917 if ((error = ufs_extattr_stop(mp, td))) { 918 if (error != EOPNOTSUPP) 919 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 920 error); 921 } else { 922 ufs_extattr_uepm_destroy(&ump->um_extattr); 923 } 924 #endif 925 if (mp->mnt_flag & MNT_SOFTDEP) { 926 if ((error = softdep_flushfiles(mp, flags, td)) != 0) 927 return (error); 928 } else { 929 if ((error = ffs_flushfiles(mp, flags, td)) != 0) 930 return (error); 931 } 932 fs = ump->um_fs; 933 UFS_LOCK(ump); 934 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 935 printf("%s: unmount pending error: blocks %jd files %d\n", 936 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 937 fs->fs_pendinginodes); 938 fs->fs_pendingblocks = 0; 939 fs->fs_pendinginodes = 0; 940 } 941 UFS_UNLOCK(ump); 942 if (fs->fs_ronly == 0) { 943 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 944 error = ffs_sbupdate(ump, MNT_WAIT); 945 if (error) { 946 fs->fs_clean = 0; 947 return (error); 948 } 949 } 950 DROP_GIANT(); 951 g_topology_lock(); 952 g_vfs_close(ump->um_cp, td); 953 g_topology_unlock(); 954 PICKUP_GIANT(); 955 vrele(ump->um_devvp); 956 mtx_destroy(UFS_MTX(ump)); 957 free(fs->fs_csp, M_UFSMNT); 958 free(fs, M_UFSMNT); 959 free(ump, M_UFSMNT); 960 mp->mnt_data = (qaddr_t)0; 961 mp->mnt_flag &= ~MNT_LOCAL; 962 return (error); 963 } 964 965 /* 966 * Flush out all the files in a filesystem. 967 */ 968 int 969 ffs_flushfiles(mp, flags, td) 970 struct mount *mp; 971 int flags; 972 struct thread *td; 973 { 974 struct ufsmount *ump; 975 int error; 976 977 ump = VFSTOUFS(mp); 978 #ifdef QUOTA 979 if (mp->mnt_flag & MNT_QUOTA) { 980 int i; 981 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 982 if (error) 983 return (error); 984 for (i = 0; i < MAXQUOTAS; i++) { 985 if (ump->um_quotas[i] == NULLVP) 986 continue; 987 quotaoff(td, mp, i); 988 } 989 /* 990 * Here we fall through to vflush again to ensure 991 * that we have gotten rid of all the system vnodes. 992 */ 993 } 994 #endif 995 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 996 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 997 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 998 return (error); 999 ffs_snapshot_unmount(mp); 1000 /* 1001 * Here we fall through to vflush again to ensure 1002 * that we have gotten rid of all the system vnodes. 1003 */ 1004 } 1005 /* 1006 * Flush all the files. 1007 */ 1008 if ((error = vflush(mp, 0, flags, td)) != 0) 1009 return (error); 1010 /* 1011 * Flush filesystem metadata. 1012 */ 1013 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td); 1014 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1015 VOP_UNLOCK(ump->um_devvp, 0, td); 1016 return (error); 1017 } 1018 1019 /* 1020 * Get filesystem statistics. 1021 */ 1022 static int 1023 ffs_statfs(mp, sbp, td) 1024 struct mount *mp; 1025 struct statfs *sbp; 1026 struct thread *td; 1027 { 1028 struct ufsmount *ump; 1029 struct fs *fs; 1030 1031 ump = VFSTOUFS(mp); 1032 fs = ump->um_fs; 1033 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1034 panic("ffs_statfs"); 1035 sbp->f_version = STATFS_VERSION; 1036 sbp->f_bsize = fs->fs_fsize; 1037 sbp->f_iosize = fs->fs_bsize; 1038 sbp->f_blocks = fs->fs_dsize; 1039 UFS_LOCK(ump); 1040 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1041 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1042 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1043 dbtofsb(fs, fs->fs_pendingblocks); 1044 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1045 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1046 UFS_UNLOCK(ump); 1047 sbp->f_namemax = NAME_MAX; 1048 return (0); 1049 } 1050 1051 /* 1052 * Go through the disk queues to initiate sandbagged IO; 1053 * go through the inodes to write those that have been modified; 1054 * initiate the writing of the super block if it has been modified. 1055 * 1056 * Note: we are always called with the filesystem marked `MPBUSY'. 1057 */ 1058 static int 1059 ffs_sync(mp, waitfor, td) 1060 struct mount *mp; 1061 int waitfor; 1062 struct thread *td; 1063 { 1064 struct vnode *nvp, *vp, *devvp; 1065 struct inode *ip; 1066 struct ufsmount *ump = VFSTOUFS(mp); 1067 struct fs *fs; 1068 int error, count, wait, lockreq, allerror = 0; 1069 struct bufobj *bo; 1070 1071 fs = ump->um_fs; 1072 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1073 printf("fs = %s\n", fs->fs_fsmnt); 1074 panic("ffs_sync: rofs mod"); 1075 } 1076 /* 1077 * Write back each (modified) inode. 1078 */ 1079 wait = 0; 1080 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1081 if (waitfor == MNT_WAIT) { 1082 wait = 1; 1083 lockreq = LK_EXCLUSIVE; 1084 } 1085 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1086 MNT_ILOCK(mp); 1087 loop: 1088 MNT_VNODE_FOREACH(vp, mp, nvp) { 1089 /* 1090 * Depend on the mntvnode_slock to keep things stable enough 1091 * for a quick test. Since there might be hundreds of 1092 * thousands of vnodes, we cannot afford even a subroutine 1093 * call unless there's a good chance that we have work to do. 1094 */ 1095 VI_LOCK(vp); 1096 if (vp->v_iflag & VI_DOOMED) { 1097 VI_UNLOCK(vp); 1098 continue; 1099 } 1100 ip = VTOI(vp); 1101 if (vp->v_type == VNON || ((ip->i_flag & 1102 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1103 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1104 VI_UNLOCK(vp); 1105 continue; 1106 } 1107 MNT_IUNLOCK(mp); 1108 if ((error = vget(vp, lockreq, td)) != 0) { 1109 MNT_ILOCK(mp); 1110 if (error == ENOENT || error == ENOLCK) 1111 goto loop; 1112 continue; 1113 } 1114 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1115 allerror = error; 1116 vput(vp); 1117 MNT_ILOCK(mp); 1118 } 1119 MNT_IUNLOCK(mp); 1120 /* 1121 * Force stale filesystem control information to be flushed. 1122 */ 1123 if (waitfor == MNT_WAIT) { 1124 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1125 allerror = error; 1126 /* Flushed work items may create new vnodes to clean */ 1127 if (allerror == 0 && count) { 1128 MNT_ILOCK(mp); 1129 goto loop; 1130 } 1131 } 1132 #ifdef QUOTA 1133 qsync(mp); 1134 #endif 1135 devvp = ump->um_devvp; 1136 VI_LOCK(devvp); 1137 bo = &devvp->v_bufobj; 1138 if (waitfor != MNT_LAZY && 1139 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1140 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td); 1141 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1142 allerror = error; 1143 VOP_UNLOCK(devvp, 0, td); 1144 if (allerror == 0 && waitfor == MNT_WAIT) { 1145 MNT_ILOCK(mp); 1146 goto loop; 1147 } 1148 } else 1149 VI_UNLOCK(devvp); 1150 /* 1151 * Write back modified superblock. 1152 */ 1153 if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0) 1154 allerror = error; 1155 return (allerror); 1156 } 1157 1158 int 1159 ffs_vget(mp, ino, flags, vpp) 1160 struct mount *mp; 1161 ino_t ino; 1162 int flags; 1163 struct vnode **vpp; 1164 { 1165 struct fs *fs; 1166 struct inode *ip; 1167 struct ufsmount *ump; 1168 struct buf *bp; 1169 struct vnode *vp; 1170 struct cdev *dev; 1171 int error; 1172 1173 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1174 if (error || *vpp != NULL) 1175 return (error); 1176 1177 /* 1178 * We must promote to an exclusive lock for vnode creation. This 1179 * can happen if lookup is passed LOCKSHARED. 1180 */ 1181 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1182 flags &= ~LK_TYPE_MASK; 1183 flags |= LK_EXCLUSIVE; 1184 } 1185 1186 /* 1187 * We do not lock vnode creation as it is believed to be too 1188 * expensive for such rare case as simultaneous creation of vnode 1189 * for same ino by different processes. We just allow them to race 1190 * and check later to decide who wins. Let the race begin! 1191 */ 1192 1193 ump = VFSTOUFS(mp); 1194 dev = ump->um_dev; 1195 fs = ump->um_fs; 1196 1197 /* 1198 * If this MALLOC() is performed after the getnewvnode() 1199 * it might block, leaving a vnode with a NULL v_data to be 1200 * found by ffs_sync() if a sync happens to fire right then, 1201 * which will cause a panic because ffs_sync() blindly 1202 * dereferences vp->v_data (as well it should). 1203 */ 1204 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1205 1206 /* Allocate a new vnode/inode. */ 1207 if (fs->fs_magic == FS_UFS1_MAGIC) 1208 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1209 else 1210 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1211 if (error) { 1212 *vpp = NULL; 1213 uma_zfree(uma_inode, ip); 1214 return (error); 1215 } 1216 /* 1217 * FFS supports recursive and shared locking. 1218 */ 1219 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 1220 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 1221 vp->v_data = ip; 1222 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1223 ip->i_vnode = vp; 1224 ip->i_ump = ump; 1225 ip->i_fs = fs; 1226 ip->i_dev = dev; 1227 ip->i_number = ino; 1228 #ifdef QUOTA 1229 { 1230 int i; 1231 for (i = 0; i < MAXQUOTAS; i++) 1232 ip->i_dquot[i] = NODQUOT; 1233 } 1234 #endif 1235 1236 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1237 if (error || *vpp != NULL) 1238 return (error); 1239 1240 /* Read in the disk contents for the inode, copy into the inode. */ 1241 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1242 (int)fs->fs_bsize, NOCRED, &bp); 1243 if (error) { 1244 /* 1245 * The inode does not contain anything useful, so it would 1246 * be misleading to leave it on its hash chain. With mode 1247 * still zero, it will be unlinked and returned to the free 1248 * list by vput(). 1249 */ 1250 brelse(bp); 1251 vput(vp); 1252 *vpp = NULL; 1253 return (error); 1254 } 1255 if (ip->i_ump->um_fstype == UFS1) 1256 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1257 else 1258 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1259 ffs_load_inode(bp, ip, fs, ino); 1260 if (DOINGSOFTDEP(vp)) 1261 softdep_load_inodeblock(ip); 1262 else 1263 ip->i_effnlink = ip->i_nlink; 1264 bqrelse(bp); 1265 1266 /* 1267 * Initialize the vnode from the inode, check for aliases. 1268 * Note that the underlying vnode may have changed. 1269 */ 1270 if (ip->i_ump->um_fstype == UFS1) 1271 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1272 else 1273 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1274 if (error) { 1275 vput(vp); 1276 *vpp = NULL; 1277 return (error); 1278 } 1279 1280 /* 1281 * Finish inode initialization. 1282 */ 1283 1284 /* 1285 * Set up a generation number for this inode if it does not 1286 * already have one. This should only happen on old filesystems. 1287 */ 1288 if (ip->i_gen == 0) { 1289 ip->i_gen = arc4random() / 2 + 1; 1290 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1291 ip->i_flag |= IN_MODIFIED; 1292 DIP_SET(ip, i_gen, ip->i_gen); 1293 } 1294 } 1295 /* 1296 * Ensure that uid and gid are correct. This is a temporary 1297 * fix until fsck has been changed to do the update. 1298 */ 1299 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1300 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1301 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1302 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1303 } /* XXX */ 1304 1305 #ifdef MAC 1306 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1307 /* 1308 * If this vnode is already allocated, and we're running 1309 * multi-label, attempt to perform a label association 1310 * from the extended attributes on the inode. 1311 */ 1312 error = mac_associate_vnode_extattr(mp, vp); 1313 if (error) { 1314 /* ufs_inactive will release ip->i_devvp ref. */ 1315 vput(vp); 1316 *vpp = NULL; 1317 return (error); 1318 } 1319 } 1320 #endif 1321 1322 *vpp = vp; 1323 return (0); 1324 } 1325 1326 /* 1327 * File handle to vnode 1328 * 1329 * Have to be really careful about stale file handles: 1330 * - check that the inode number is valid 1331 * - call ffs_vget() to get the locked inode 1332 * - check for an unallocated inode (i_mode == 0) 1333 * - check that the given client host has export rights and return 1334 * those rights via. exflagsp and credanonp 1335 */ 1336 static int 1337 ffs_fhtovp(mp, fhp, vpp) 1338 struct mount *mp; 1339 struct fid *fhp; 1340 struct vnode **vpp; 1341 { 1342 struct ufid *ufhp; 1343 struct fs *fs; 1344 1345 ufhp = (struct ufid *)fhp; 1346 fs = VFSTOUFS(mp)->um_fs; 1347 if (ufhp->ufid_ino < ROOTINO || 1348 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1349 return (ESTALE); 1350 return (ufs_fhtovp(mp, ufhp, vpp)); 1351 } 1352 1353 /* 1354 * Vnode pointer to File handle 1355 */ 1356 /* ARGSUSED */ 1357 static int 1358 ffs_vptofh(vp, fhp) 1359 struct vnode *vp; 1360 struct fid *fhp; 1361 { 1362 struct inode *ip; 1363 struct ufid *ufhp; 1364 1365 ip = VTOI(vp); 1366 ufhp = (struct ufid *)fhp; 1367 ufhp->ufid_len = sizeof(struct ufid); 1368 ufhp->ufid_ino = ip->i_number; 1369 ufhp->ufid_gen = ip->i_gen; 1370 return (0); 1371 } 1372 1373 /* 1374 * Initialize the filesystem. 1375 */ 1376 static int 1377 ffs_init(vfsp) 1378 struct vfsconf *vfsp; 1379 { 1380 1381 softdep_initialize(); 1382 return (ufs_init(vfsp)); 1383 } 1384 1385 /* 1386 * Undo the work of ffs_init(). 1387 */ 1388 static int 1389 ffs_uninit(vfsp) 1390 struct vfsconf *vfsp; 1391 { 1392 int ret; 1393 1394 ret = ufs_uninit(vfsp); 1395 softdep_uninitialize(); 1396 return (ret); 1397 } 1398 1399 /* 1400 * Write a superblock and associated information back to disk. 1401 */ 1402 static int 1403 ffs_sbupdate(mp, waitfor) 1404 struct ufsmount *mp; 1405 int waitfor; 1406 { 1407 struct fs *fs = mp->um_fs; 1408 struct buf *sbbp; 1409 struct buf *bp; 1410 int blks; 1411 void *space; 1412 int i, size, error, allerror = 0; 1413 1414 if (fs->fs_ronly == 1 && 1415 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1416 (MNT_RDONLY | MNT_UPDATE)) 1417 panic("ffs_sbupdate: write read-only filesystem"); 1418 /* 1419 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1420 */ 1421 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1422 0, 0, 0); 1423 /* 1424 * First write back the summary information. 1425 */ 1426 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1427 space = fs->fs_csp; 1428 for (i = 0; i < blks; i += fs->fs_frag) { 1429 size = fs->fs_bsize; 1430 if (i + fs->fs_frag > blks) 1431 size = (blks - i) * fs->fs_fsize; 1432 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1433 size, 0, 0, 0); 1434 bcopy(space, bp->b_data, (u_int)size); 1435 space = (char *)space + size; 1436 if (waitfor != MNT_WAIT) 1437 bawrite(bp); 1438 else if ((error = bwrite(bp)) != 0) 1439 allerror = error; 1440 } 1441 /* 1442 * Now write back the superblock itself. If any errors occurred 1443 * up to this point, then fail so that the superblock avoids 1444 * being written out as clean. 1445 */ 1446 if (allerror) { 1447 brelse(sbbp); 1448 return (allerror); 1449 } 1450 bp = sbbp; 1451 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1452 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1453 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1454 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1455 fs->fs_sblockloc = SBLOCK_UFS1; 1456 } 1457 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1458 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1459 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1460 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1461 fs->fs_sblockloc = SBLOCK_UFS2; 1462 } 1463 fs->fs_fmod = 0; 1464 fs->fs_time = time_second; 1465 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1466 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1467 if (waitfor != MNT_WAIT) 1468 bawrite(bp); 1469 else if ((error = bwrite(bp)) != 0) 1470 allerror = error; 1471 return (allerror); 1472 } 1473 1474 static int 1475 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1476 int attrnamespace, const char *attrname, struct thread *td) 1477 { 1478 1479 #ifdef UFS_EXTATTR 1480 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1481 attrname, td)); 1482 #else 1483 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1484 attrname, td)); 1485 #endif 1486 } 1487 1488 static void 1489 ffs_ifree(struct ufsmount *ump, struct inode *ip) 1490 { 1491 1492 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1493 uma_zfree(uma_ufs1, ip->i_din1); 1494 else if (ip->i_din2 != NULL) 1495 uma_zfree(uma_ufs2, ip->i_din2); 1496 uma_zfree(uma_inode, ip); 1497 } 1498 1499 static int dobkgrdwrite = 1; 1500 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1501 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1502 1503 /* 1504 * Complete a background write started from bwrite. 1505 */ 1506 static void 1507 ffs_backgroundwritedone(struct buf *bp) 1508 { 1509 struct bufobj *bufobj; 1510 struct buf *origbp; 1511 1512 /* 1513 * Find the original buffer that we are writing. 1514 */ 1515 bufobj = bp->b_bufobj; 1516 BO_LOCK(bufobj); 1517 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1518 panic("backgroundwritedone: lost buffer"); 1519 /* Grab an extra reference to be dropped by the bufdone() below. */ 1520 bufobj_wrefl(bufobj); 1521 BO_UNLOCK(bufobj); 1522 /* 1523 * Process dependencies then return any unfinished ones. 1524 */ 1525 if (LIST_FIRST(&bp->b_dep) != NULL) 1526 buf_complete(bp); 1527 #ifdef SOFTUPDATES 1528 if (LIST_FIRST(&bp->b_dep) != NULL) 1529 softdep_move_dependencies(bp, origbp); 1530 #endif 1531 /* 1532 * This buffer is marked B_NOCACHE so when it is released 1533 * by biodone it will be tossed. 1534 */ 1535 bp->b_flags |= B_NOCACHE; 1536 bp->b_flags &= ~B_CACHE; 1537 bufdone(bp); 1538 BO_LOCK(bufobj); 1539 /* 1540 * Clear the BV_BKGRDINPROG flag in the original buffer 1541 * and awaken it if it is waiting for the write to complete. 1542 * If BV_BKGRDINPROG is not set in the original buffer it must 1543 * have been released and re-instantiated - which is not legal. 1544 */ 1545 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1546 ("backgroundwritedone: lost buffer2")); 1547 origbp->b_vflags &= ~BV_BKGRDINPROG; 1548 if (origbp->b_vflags & BV_BKGRDWAIT) { 1549 origbp->b_vflags &= ~BV_BKGRDWAIT; 1550 wakeup(&origbp->b_xflags); 1551 } 1552 BO_UNLOCK(bufobj); 1553 } 1554 1555 1556 /* 1557 * Write, release buffer on completion. (Done by iodone 1558 * if async). Do not bother writing anything if the buffer 1559 * is invalid. 1560 * 1561 * Note that we set B_CACHE here, indicating that buffer is 1562 * fully valid and thus cacheable. This is true even of NFS 1563 * now so we set it generally. This could be set either here 1564 * or in biodone() since the I/O is synchronous. We put it 1565 * here. 1566 */ 1567 static int 1568 ffs_bufwrite(struct buf *bp) 1569 { 1570 int oldflags, s; 1571 struct buf *newbp; 1572 1573 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1574 if (bp->b_flags & B_INVAL) { 1575 brelse(bp); 1576 return (0); 1577 } 1578 1579 oldflags = bp->b_flags; 1580 1581 if (BUF_REFCNT(bp) == 0) 1582 panic("bufwrite: buffer is not busy???"); 1583 s = splbio(); 1584 /* 1585 * If a background write is already in progress, delay 1586 * writing this block if it is asynchronous. Otherwise 1587 * wait for the background write to complete. 1588 */ 1589 BO_LOCK(bp->b_bufobj); 1590 if (bp->b_vflags & BV_BKGRDINPROG) { 1591 if (bp->b_flags & B_ASYNC) { 1592 BO_UNLOCK(bp->b_bufobj); 1593 splx(s); 1594 bdwrite(bp); 1595 return (0); 1596 } 1597 bp->b_vflags |= BV_BKGRDWAIT; 1598 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1599 if (bp->b_vflags & BV_BKGRDINPROG) 1600 panic("bufwrite: still writing"); 1601 } 1602 BO_UNLOCK(bp->b_bufobj); 1603 1604 /* Mark the buffer clean */ 1605 bundirty(bp); 1606 1607 /* 1608 * If this buffer is marked for background writing and we 1609 * do not have to wait for it, make a copy and write the 1610 * copy so as to leave this buffer ready for further use. 1611 * 1612 * This optimization eats a lot of memory. If we have a page 1613 * or buffer shortfall we can't do it. 1614 */ 1615 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1616 (bp->b_flags & B_ASYNC) && 1617 !vm_page_count_severe() && 1618 !buf_dirty_count_severe()) { 1619 KASSERT(bp->b_iodone == NULL, 1620 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1621 1622 /* get a new block */ 1623 newbp = geteblk(bp->b_bufsize); 1624 1625 /* 1626 * set it to be identical to the old block. We have to 1627 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1628 * to avoid confusing the splay tree and gbincore(). 1629 */ 1630 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1631 newbp->b_lblkno = bp->b_lblkno; 1632 newbp->b_xflags |= BX_BKGRDMARKER; 1633 BO_LOCK(bp->b_bufobj); 1634 bp->b_vflags |= BV_BKGRDINPROG; 1635 bgetvp(bp->b_vp, newbp); 1636 BO_UNLOCK(bp->b_bufobj); 1637 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1638 newbp->b_blkno = bp->b_blkno; 1639 newbp->b_offset = bp->b_offset; 1640 newbp->b_iodone = ffs_backgroundwritedone; 1641 newbp->b_flags |= B_ASYNC; 1642 newbp->b_flags &= ~B_INVAL; 1643 1644 #ifdef SOFTUPDATES 1645 /* move over the dependencies */ 1646 if (LIST_FIRST(&bp->b_dep) != NULL) 1647 softdep_move_dependencies(bp, newbp); 1648 #endif 1649 1650 /* 1651 * Initiate write on the copy, release the original to 1652 * the B_LOCKED queue so that it cannot go away until 1653 * the background write completes. If not locked it could go 1654 * away and then be reconstituted while it was being written. 1655 * If the reconstituted buffer were written, we could end up 1656 * with two background copies being written at the same time. 1657 */ 1658 bqrelse(bp); 1659 bp = newbp; 1660 } 1661 1662 /* Let the normal bufwrite do the rest for us */ 1663 bufwrite(bp); 1664 1665 return (0); 1666 } 1667 1668 1669 static void 1670 ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1671 { 1672 struct vnode *vp; 1673 int error; 1674 1675 vp = bo->__bo_vnode; 1676 if (bp->b_iocmd == BIO_WRITE) { 1677 #ifdef SOFTUPDATES 1678 if (LIST_FIRST(&bp->b_dep) != NULL) 1679 buf_start(bp); 1680 #endif 1681 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1682 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1683 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1684 panic("ffs_geom_strategy: bad I/O"); 1685 bp->b_flags &= ~B_VALIDSUSPWRT; 1686 if ((vp->v_vflag & VV_COPYONWRITE) && 1687 vp->v_rdev->si_snapdata != NULL && 1688 (error = (ffs_copyonwrite)(vp, bp)) != 0 && 1689 error != EOPNOTSUPP) { 1690 bp->b_error = error; 1691 bp->b_ioflags |= BIO_ERROR; 1692 bufdone(bp); 1693 return; 1694 } 1695 } 1696 g_vfs_strategy(bo, bp); 1697 } 1698