ffs_vfsops.c revision 156896
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: head/sys/ufs/ffs/ffs_vfsops.c 156896 2006-03-19 21:09:19Z tegge $"); 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 71static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 72 73static int ffs_sbupdate(struct ufsmount *, int, int); 74static int ffs_reload(struct mount *, struct thread *); 75static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 76static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 77 ufs2_daddr_t); 78static void ffs_oldfscompat_write(struct fs *, struct ufsmount *); 79static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 80static vfs_init_t ffs_init; 81static vfs_uninit_t ffs_uninit; 82static vfs_extattrctl_t ffs_extattrctl; 83static vfs_cmount_t ffs_cmount; 84static vfs_unmount_t ffs_unmount; 85static vfs_mount_t ffs_mount; 86static vfs_statfs_t ffs_statfs; 87static vfs_fhtovp_t ffs_fhtovp; 88static vfs_vptofh_t ffs_vptofh; 89static vfs_sync_t ffs_sync; 90 91static 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 107VFS_SET(ufs_vfsops, ufs, 0); 108 109static b_strategy_t ffs_geom_strategy; 110static b_write_t ffs_bufwrite; 111 112static 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 119static const char *ffs_opts[] = { "acls", "async", "atime", "clusterr", 120 "clusterw", "exec", "errmsg", "export", "force", "from", "multilabel", 121 "snapshot", "suid", "suiddir", "symfollow", "sync", 122 "update", "union", NULL }; 123 124static int 125ffs_mount(struct mount *mp, struct thread *td) 126{ 127 struct vnode *devvp; 128 struct ufsmount *ump = 0; 129 struct fs *fs; 130 int error, flags; 131 mode_t accessmode; 132 struct nameidata ndp; 133 struct export_args export; 134 char *fspec; 135 136 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 137 return (EINVAL); 138 if (uma_inode == NULL) { 139 uma_inode = uma_zcreate("FFS inode", 140 sizeof(struct inode), NULL, NULL, NULL, NULL, 141 UMA_ALIGN_PTR, 0); 142 uma_ufs1 = uma_zcreate("FFS1 dinode", 143 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 144 UMA_ALIGN_PTR, 0); 145 uma_ufs2 = uma_zcreate("FFS2 dinode", 146 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 147 UMA_ALIGN_PTR, 0); 148 } 149 150 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 151 if (error) 152 return (error); 153 154 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 155 mp->mnt_flag |= MNT_ACLS; 156 157 if (vfs_getopt(mp->mnt_optnew, "async", NULL, NULL) == 0) 158 mp->mnt_flag |= MNT_ASYNC; 159 160 if (vfs_getopt(mp->mnt_optnew, "force", NULL, NULL) == 0) 161 mp->mnt_flag |= MNT_FORCE; 162 163 if (vfs_getopt(mp->mnt_optnew, "multilabel", NULL, NULL) == 0) 164 mp->mnt_flag |= MNT_MULTILABEL; 165 166 if (vfs_getopt(mp->mnt_optnew, "noasync", NULL, NULL) == 0) 167 mp->mnt_flag &= ~MNT_ASYNC; 168 169 if (vfs_getopt(mp->mnt_optnew, "noatime", NULL, NULL) == 0) 170 mp->mnt_flag |= MNT_NOATIME; 171 172 if (vfs_getopt(mp->mnt_optnew, "noclusterr", NULL, NULL) == 0) 173 mp->mnt_flag |= MNT_NOCLUSTERR; 174 175 if (vfs_getopt(mp->mnt_optnew, "noclusterw", NULL, NULL) == 0) 176 mp->mnt_flag |= MNT_NOCLUSTERW; 177 178 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) 179 mp->mnt_flag |= MNT_SNAPSHOT; 180 181 if (vfs_getopt(mp->mnt_optnew, "update", NULL, NULL) == 0) 182 mp->mnt_flag |= MNT_UPDATE; 183 184 export.ex_root = -2; /* DEFAULT_ROOTID */ 185 186 if (mp->mnt_flag & MNT_RDONLY) 187 export.ex_flags = MNT_EXRDONLY; 188 else 189 export.ex_flags = 0; 190 191 /* 192 * If updating, check whether changing from read-only to 193 * read/write; if there is no device name, that's all we do. 194 */ 195 if (mp->mnt_flag & MNT_UPDATE) { 196 ump = VFSTOUFS(mp); 197 fs = ump->um_fs; 198 devvp = ump->um_devvp; 199 if (fs->fs_ronly == 0 && 200 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 201 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 202 return (error); 203 /* 204 * Flush any dirty data. 205 */ 206 if ((error = ffs_sync(mp, MNT_WAIT, td)) != 0) { 207 vn_finished_write(mp); 208 return (error); 209 } 210 /* 211 * Check for and optionally get rid of files open 212 * for writing. 213 */ 214 flags = WRITECLOSE; 215 if (mp->mnt_flag & MNT_FORCE) 216 flags |= FORCECLOSE; 217 if (mp->mnt_flag & MNT_SOFTDEP) { 218 error = softdep_flushfiles(mp, flags, td); 219 } else { 220 error = ffs_flushfiles(mp, flags, td); 221 } 222 if (error) { 223 vn_finished_write(mp); 224 return (error); 225 } 226 if (fs->fs_pendingblocks != 0 || 227 fs->fs_pendinginodes != 0) { 228 printf("%s: %s: blocks %jd files %d\n", 229 fs->fs_fsmnt, "update error", 230 (intmax_t)fs->fs_pendingblocks, 231 fs->fs_pendinginodes); 232 fs->fs_pendingblocks = 0; 233 fs->fs_pendinginodes = 0; 234 } 235 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 236 fs->fs_clean = 1; 237 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 238 fs->fs_ronly = 0; 239 fs->fs_clean = 0; 240 vn_finished_write(mp); 241 return (error); 242 } 243 vn_finished_write(mp); 244 DROP_GIANT(); 245 g_topology_lock(); 246 g_access(ump->um_cp, 0, -1, 0); 247 g_topology_unlock(); 248 PICKUP_GIANT(); 249 fs->fs_ronly = 1; 250 mp->mnt_flag |= MNT_RDONLY; 251 } 252 if ((mp->mnt_flag & MNT_RELOAD) && 253 (error = ffs_reload(mp, td)) != 0) 254 return (error); 255 if (fs->fs_ronly && 256 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 257 /* 258 * If upgrade to read-write by non-root, then verify 259 * that user has necessary permissions on the device. 260 */ 261 if (suser(td)) { 262 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 263 if ((error = VOP_ACCESS(devvp, VREAD | VWRITE, 264 td->td_ucred, td)) != 0) { 265 VOP_UNLOCK(devvp, 0, td); 266 return (error); 267 } 268 VOP_UNLOCK(devvp, 0, td); 269 } 270 fs->fs_flags &= ~FS_UNCLEAN; 271 if (fs->fs_clean == 0) { 272 fs->fs_flags |= FS_UNCLEAN; 273 if ((mp->mnt_flag & MNT_FORCE) || 274 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 275 (fs->fs_flags & FS_DOSOFTDEP))) { 276 printf("WARNING: %s was not %s\n", 277 fs->fs_fsmnt, "properly dismounted"); 278 } else { 279 printf( 280"WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 281 fs->fs_fsmnt); 282 return (EPERM); 283 } 284 } 285 DROP_GIANT(); 286 g_topology_lock(); 287 /* 288 * If we're the root device, we may not have an E count 289 * yet, get it now. 290 */ 291 if (ump->um_cp->ace == 0) 292 error = g_access(ump->um_cp, 0, 1, 1); 293 else 294 error = g_access(ump->um_cp, 0, 1, 0); 295 g_topology_unlock(); 296 PICKUP_GIANT(); 297 if (error) 298 return (error); 299 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 300 return (error); 301 fs->fs_ronly = 0; 302 mp->mnt_flag &= ~MNT_RDONLY; 303 fs->fs_clean = 0; 304 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 305 vn_finished_write(mp); 306 return (error); 307 } 308 /* check to see if we need to start softdep */ 309 if ((fs->fs_flags & FS_DOSOFTDEP) && 310 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 311 vn_finished_write(mp); 312 return (error); 313 } 314 if (fs->fs_snapinum[0] != 0) 315 ffs_snapshot_mount(mp); 316 vn_finished_write(mp); 317 } 318 /* 319 * Soft updates is incompatible with "async", 320 * so if we are doing softupdates stop the user 321 * from setting the async flag in an update. 322 * Softdep_mount() clears it in an initial mount 323 * or ro->rw remount. 324 */ 325 if (mp->mnt_flag & MNT_SOFTDEP) 326 mp->mnt_flag &= ~MNT_ASYNC; 327 /* 328 * Keep MNT_ACLS flag if it is stored in superblock. 329 */ 330 if ((fs->fs_flags & FS_ACLS) != 0) 331 mp->mnt_flag |= MNT_ACLS; 332 /* 333 * If not updating name, process export requests. 334 */ 335 error = 0; 336 if (vfs_getopt(mp->mnt_optnew, "export", NULL, NULL) == 0) { 337 error = vfs_copyopt(mp->mnt_optnew, "export", 338 &export, sizeof export); 339 } 340 341 if (error == 0 && export.ex_flags != 0) 342 return (vfs_export(mp, &export)); 343 /* 344 * If this is a snapshot request, take the snapshot. 345 */ 346 if (mp->mnt_flag & MNT_SNAPSHOT) 347 return (ffs_snapshot(mp, fspec)); 348 } 349 350 /* 351 * Not an update, or updating the name: look up the name 352 * and verify that it refers to a sensible disk device. 353 */ 354 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 355 if ((error = namei(&ndp)) != 0) 356 return (error); 357 NDFREE(&ndp, NDF_ONLY_PNBUF); 358 devvp = ndp.ni_vp; 359 if (!vn_isdisk(devvp, &error)) { 360 vput(devvp); 361 return (error); 362 } 363 364 /* 365 * If mount by non-root, then verify that user has necessary 366 * permissions on the device. 367 */ 368 if (suser(td)) { 369 accessmode = VREAD; 370 if ((mp->mnt_flag & MNT_RDONLY) == 0) 371 accessmode |= VWRITE; 372 if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){ 373 vput(devvp); 374 return (error); 375 } 376 } 377 378 if (mp->mnt_flag & MNT_UPDATE) { 379 /* 380 * Update only 381 * 382 * If it's not the same vnode, or at least the same device 383 * then it's not correct. 384 */ 385 386 if (devvp->v_rdev != ump->um_devvp->v_rdev) 387 error = EINVAL; /* needs translation */ 388 vput(devvp); 389 if (error) 390 return (error); 391 } else { 392 /* 393 * New mount 394 * 395 * We need the name for the mount point (also used for 396 * "last mounted on") copied in. If an error occurs, 397 * the mount point is discarded by the upper level code. 398 * Note that vfs_mount() populates f_mntonname for us. 399 */ 400 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 401 vrele(devvp); 402 return (error); 403 } 404 } 405 vfs_mountedfrom(mp, fspec); 406 return (0); 407} 408 409/* 410 * Compatibility with old mount system call. 411 */ 412 413static int 414ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td) 415{ 416 struct ufs_args args; 417 int error; 418 419 if (data == NULL) 420 return (EINVAL); 421 error = copyin(data, &args, sizeof args); 422 if (error) 423 return (error); 424 425 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 426 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 427 error = kernel_mount(ma, flags); 428 429 return (error); 430} 431 432/* 433 * Reload all incore data for a filesystem (used after running fsck on 434 * the root filesystem and finding things to fix). The filesystem must 435 * be mounted read-only. 436 * 437 * Things to do to update the mount: 438 * 1) invalidate all cached meta-data. 439 * 2) re-read superblock from disk. 440 * 3) re-read summary information from disk. 441 * 4) invalidate all inactive vnodes. 442 * 5) invalidate all cached file data. 443 * 6) re-read inode data for all active vnodes. 444 */ 445static int 446ffs_reload(struct mount *mp, struct thread *td) 447{ 448 struct vnode *vp, *mvp, *devvp; 449 struct inode *ip; 450 void *space; 451 struct buf *bp; 452 struct fs *fs, *newfs; 453 struct ufsmount *ump; 454 ufs2_daddr_t sblockloc; 455 int i, blks, size, error; 456 int32_t *lp; 457 458 if ((mp->mnt_flag & MNT_RDONLY) == 0) 459 return (EINVAL); 460 ump = VFSTOUFS(mp); 461 /* 462 * Step 1: invalidate all cached meta-data. 463 */ 464 devvp = VFSTOUFS(mp)->um_devvp; 465 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 466 if (vinvalbuf(devvp, 0, td, 0, 0) != 0) 467 panic("ffs_reload: dirty1"); 468 VOP_UNLOCK(devvp, 0, td); 469 470 /* 471 * Step 2: re-read superblock from disk. 472 */ 473 fs = VFSTOUFS(mp)->um_fs; 474 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 475 NOCRED, &bp)) != 0) 476 return (error); 477 newfs = (struct fs *)bp->b_data; 478 if ((newfs->fs_magic != FS_UFS1_MAGIC && 479 newfs->fs_magic != FS_UFS2_MAGIC) || 480 newfs->fs_bsize > MAXBSIZE || 481 newfs->fs_bsize < sizeof(struct fs)) { 482 brelse(bp); 483 return (EIO); /* XXX needs translation */ 484 } 485 /* 486 * Copy pointer fields back into superblock before copying in XXX 487 * new superblock. These should really be in the ufsmount. XXX 488 * Note that important parameters (eg fs_ncg) are unchanged. 489 */ 490 newfs->fs_csp = fs->fs_csp; 491 newfs->fs_maxcluster = fs->fs_maxcluster; 492 newfs->fs_contigdirs = fs->fs_contigdirs; 493 newfs->fs_active = fs->fs_active; 494 /* The file system is still read-only. */ 495 newfs->fs_ronly = 1; 496 sblockloc = fs->fs_sblockloc; 497 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 498 brelse(bp); 499 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 500 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 501 UFS_LOCK(ump); 502 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 503 printf("%s: reload pending error: blocks %jd files %d\n", 504 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 505 fs->fs_pendinginodes); 506 fs->fs_pendingblocks = 0; 507 fs->fs_pendinginodes = 0; 508 } 509 UFS_UNLOCK(ump); 510 511 /* 512 * Step 3: re-read summary information from disk. 513 */ 514 blks = howmany(fs->fs_cssize, fs->fs_fsize); 515 space = fs->fs_csp; 516 for (i = 0; i < blks; i += fs->fs_frag) { 517 size = fs->fs_bsize; 518 if (i + fs->fs_frag > blks) 519 size = (blks - i) * fs->fs_fsize; 520 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 521 NOCRED, &bp); 522 if (error) 523 return (error); 524 bcopy(bp->b_data, space, (u_int)size); 525 space = (char *)space + size; 526 brelse(bp); 527 } 528 /* 529 * We no longer know anything about clusters per cylinder group. 530 */ 531 if (fs->fs_contigsumsize > 0) { 532 lp = fs->fs_maxcluster; 533 for (i = 0; i < fs->fs_ncg; i++) 534 *lp++ = fs->fs_contigsumsize; 535 } 536 537loop: 538 MNT_ILOCK(mp); 539 MNT_VNODE_FOREACH(vp, mp, mvp) { 540 VI_LOCK(vp); 541 if (vp->v_iflag & VI_DOOMED) { 542 VI_UNLOCK(vp); 543 continue; 544 } 545 MNT_IUNLOCK(mp); 546 /* 547 * Step 4: invalidate all cached file data. 548 */ 549 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 550 MNT_VNODE_FOREACH_ABORT(mp, mvp); 551 goto loop; 552 } 553 if (vinvalbuf(vp, 0, td, 0, 0)) 554 panic("ffs_reload: dirty2"); 555 /* 556 * Step 5: re-read inode data for all active vnodes. 557 */ 558 ip = VTOI(vp); 559 error = 560 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 561 (int)fs->fs_bsize, NOCRED, &bp); 562 if (error) { 563 VOP_UNLOCK(vp, 0, td); 564 vrele(vp); 565 MNT_VNODE_FOREACH_ABORT(mp, mvp); 566 return (error); 567 } 568 ffs_load_inode(bp, ip, fs, ip->i_number); 569 ip->i_effnlink = ip->i_nlink; 570 brelse(bp); 571 VOP_UNLOCK(vp, 0, td); 572 vrele(vp); 573 MNT_ILOCK(mp); 574 } 575 MNT_IUNLOCK(mp); 576 return (0); 577} 578 579/* 580 * Possible superblock locations ordered from most to least likely. 581 */ 582static int sblock_try[] = SBLOCKSEARCH; 583 584/* 585 * Common code for mount and mountroot 586 */ 587static int 588ffs_mountfs(devvp, mp, td) 589 struct vnode *devvp; 590 struct mount *mp; 591 struct thread *td; 592{ 593 struct ufsmount *ump; 594 struct buf *bp; 595 struct fs *fs; 596 struct cdev *dev; 597 void *space; 598 ufs2_daddr_t sblockloc; 599 int error, i, blks, size, ronly; 600 int32_t *lp; 601 struct ucred *cred; 602 struct g_consumer *cp; 603 604 dev = devvp->v_rdev; 605 cred = td ? td->td_ucred : NOCRED; 606 607 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 608 DROP_GIANT(); 609 g_topology_lock(); 610 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 611 612 /* 613 * If we are a root mount, drop the E flag so fsck can do its magic. 614 * We will pick it up again when we remount R/W. 615 */ 616 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 617 error = g_access(cp, 0, 0, -1); 618 g_topology_unlock(); 619 PICKUP_GIANT(); 620 VOP_UNLOCK(devvp, 0, td); 621 if (error) 622 return (error); 623 if (devvp->v_rdev->si_iosize_max != 0) 624 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 625 if (mp->mnt_iosize_max > MAXPHYS) 626 mp->mnt_iosize_max = MAXPHYS; 627 628 devvp->v_bufobj.bo_private = cp; 629 devvp->v_bufobj.bo_ops = &ffs_ops; 630 631 bp = NULL; 632 ump = NULL; 633 fs = NULL; 634 sblockloc = 0; 635 /* 636 * Try reading the superblock in each of its possible locations. 637 */ 638 for (i = 0; sblock_try[i] != -1; i++) { 639 if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE, 640 cred, &bp)) != 0) 641 goto out; 642 fs = (struct fs *)bp->b_data; 643 sblockloc = sblock_try[i]; 644 if ((fs->fs_magic == FS_UFS1_MAGIC || 645 (fs->fs_magic == FS_UFS2_MAGIC && 646 (fs->fs_sblockloc == sblockloc || 647 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 648 fs->fs_bsize <= MAXBSIZE && 649 fs->fs_bsize >= sizeof(struct fs)) 650 break; 651 brelse(bp); 652 bp = NULL; 653 } 654 if (sblock_try[i] == -1) { 655 error = EINVAL; /* XXX needs translation */ 656 goto out; 657 } 658 fs->fs_fmod = 0; 659 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 660 fs->fs_flags &= ~FS_UNCLEAN; 661 if (fs->fs_clean == 0) { 662 fs->fs_flags |= FS_UNCLEAN; 663 if (ronly || (mp->mnt_flag & MNT_FORCE) || 664 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 665 (fs->fs_flags & FS_DOSOFTDEP))) { 666 printf( 667"WARNING: %s was not properly dismounted\n", 668 fs->fs_fsmnt); 669 } else { 670 printf( 671"WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 672 fs->fs_fsmnt); 673 error = EPERM; 674 goto out; 675 } 676 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 677 (mp->mnt_flag & MNT_FORCE)) { 678 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 679 (intmax_t)fs->fs_pendingblocks, 680 fs->fs_pendinginodes); 681 fs->fs_pendingblocks = 0; 682 fs->fs_pendinginodes = 0; 683 } 684 } 685 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 686 printf("%s: mount pending error: blocks %jd files %d\n", 687 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 688 fs->fs_pendinginodes); 689 fs->fs_pendingblocks = 0; 690 fs->fs_pendinginodes = 0; 691 } 692 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 693 ump->um_cp = cp; 694 ump->um_bo = &devvp->v_bufobj; 695 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 696 if (fs->fs_magic == FS_UFS1_MAGIC) { 697 ump->um_fstype = UFS1; 698 ump->um_balloc = ffs_balloc_ufs1; 699 } else { 700 ump->um_fstype = UFS2; 701 ump->um_balloc = ffs_balloc_ufs2; 702 } 703 ump->um_blkatoff = ffs_blkatoff; 704 ump->um_truncate = ffs_truncate; 705 ump->um_update = ffs_update; 706 ump->um_valloc = ffs_valloc; 707 ump->um_vfree = ffs_vfree; 708 ump->um_ifree = ffs_ifree; 709 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 710 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 711 if (fs->fs_sbsize < SBLOCKSIZE) 712 bp->b_flags |= B_INVAL | B_NOCACHE; 713 brelse(bp); 714 bp = NULL; 715 fs = ump->um_fs; 716 ffs_oldfscompat_read(fs, ump, sblockloc); 717 fs->fs_ronly = ronly; 718 size = fs->fs_cssize; 719 blks = howmany(size, fs->fs_fsize); 720 if (fs->fs_contigsumsize > 0) 721 size += fs->fs_ncg * sizeof(int32_t); 722 size += fs->fs_ncg * sizeof(u_int8_t); 723 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 724 fs->fs_csp = space; 725 for (i = 0; i < blks; i += fs->fs_frag) { 726 size = fs->fs_bsize; 727 if (i + fs->fs_frag > blks) 728 size = (blks - i) * fs->fs_fsize; 729 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 730 cred, &bp)) != 0) { 731 free(fs->fs_csp, M_UFSMNT); 732 goto out; 733 } 734 bcopy(bp->b_data, space, (u_int)size); 735 space = (char *)space + size; 736 brelse(bp); 737 bp = NULL; 738 } 739 if (fs->fs_contigsumsize > 0) { 740 fs->fs_maxcluster = lp = space; 741 for (i = 0; i < fs->fs_ncg; i++) 742 *lp++ = fs->fs_contigsumsize; 743 space = lp; 744 } 745 size = fs->fs_ncg * sizeof(u_int8_t); 746 fs->fs_contigdirs = (u_int8_t *)space; 747 bzero(fs->fs_contigdirs, size); 748 fs->fs_active = NULL; 749 mp->mnt_data = (qaddr_t)ump; 750 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 751 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 752 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 753 vfs_getvfs(&mp->mnt_stat.f_fsid)) 754 vfs_getnewfsid(mp); 755 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 756 mp->mnt_flag |= MNT_LOCAL; 757 if ((fs->fs_flags & FS_MULTILABEL) != 0) 758#ifdef MAC 759 mp->mnt_flag |= MNT_MULTILABEL; 760#else 761 printf( 762"WARNING: %s: multilabel flag on fs but no MAC support\n", 763 fs->fs_fsmnt); 764#endif 765 if ((fs->fs_flags & FS_ACLS) != 0) 766#ifdef UFS_ACL 767 mp->mnt_flag |= MNT_ACLS; 768#else 769 printf( 770"WARNING: %s: ACLs flag on fs but no ACLs support\n", 771 fs->fs_fsmnt); 772#endif 773 ump->um_mountp = mp; 774 ump->um_dev = dev; 775 ump->um_devvp = devvp; 776 ump->um_nindir = fs->fs_nindir; 777 ump->um_bptrtodb = fs->fs_fsbtodb; 778 ump->um_seqinc = fs->fs_frag; 779 for (i = 0; i < MAXQUOTAS; i++) 780 ump->um_quotas[i] = NULLVP; 781#ifdef UFS_EXTATTR 782 ufs_extattr_uepm_init(&ump->um_extattr); 783#endif 784 /* 785 * Set FS local "last mounted on" information (NULL pad) 786 */ 787 bzero(fs->fs_fsmnt, MAXMNTLEN); 788 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 789 790 if( mp->mnt_flag & MNT_ROOTFS) { 791 /* 792 * Root mount; update timestamp in mount structure. 793 * this will be used by the common root mount code 794 * to update the system clock. 795 */ 796 mp->mnt_time = fs->fs_time; 797 } 798 799 if (ronly == 0) { 800 if ((fs->fs_flags & FS_DOSOFTDEP) && 801 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 802 free(fs->fs_csp, M_UFSMNT); 803 goto out; 804 } 805 if (fs->fs_snapinum[0] != 0) 806 ffs_snapshot_mount(mp); 807 fs->fs_fmod = 1; 808 fs->fs_clean = 0; 809 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 810 } 811 /* 812 * Initialize filesystem stat information in mount struct. 813 */ 814#ifdef UFS_EXTATTR 815#ifdef UFS_EXTATTR_AUTOSTART 816 /* 817 * 818 * Auto-starting does the following: 819 * - check for /.attribute in the fs, and extattr_start if so 820 * - for each file in .attribute, enable that file with 821 * an attribute of the same name. 822 * Not clear how to report errors -- probably eat them. 823 * This would all happen while the filesystem was busy/not 824 * available, so would effectively be "atomic". 825 */ 826 (void) ufs_extattr_autostart(mp, td); 827#endif /* !UFS_EXTATTR_AUTOSTART */ 828#endif /* !UFS_EXTATTR */ 829#ifndef QUOTA 830 mp->mnt_kern_flag |= MNTK_MPSAFE; 831#endif 832 return (0); 833out: 834 if (bp) 835 brelse(bp); 836 if (cp != NULL) { 837 DROP_GIANT(); 838 g_topology_lock(); 839 g_vfs_close(cp, td); 840 g_topology_unlock(); 841 PICKUP_GIANT(); 842 } 843 if (ump) { 844 mtx_destroy(UFS_MTX(ump)); 845 free(ump->um_fs, M_UFSMNT); 846 free(ump, M_UFSMNT); 847 mp->mnt_data = (qaddr_t)0; 848 } 849 return (error); 850} 851 852#include <sys/sysctl.h> 853static int bigcgs = 0; 854SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 855 856/* 857 * Sanity checks for loading old filesystem superblocks. 858 * See ffs_oldfscompat_write below for unwound actions. 859 * 860 * XXX - Parts get retired eventually. 861 * Unfortunately new bits get added. 862 */ 863static void 864ffs_oldfscompat_read(fs, ump, sblockloc) 865 struct fs *fs; 866 struct ufsmount *ump; 867 ufs2_daddr_t sblockloc; 868{ 869 off_t maxfilesize; 870 871 /* 872 * If not yet done, update fs_flags location and value of fs_sblockloc. 873 */ 874 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 875 fs->fs_flags = fs->fs_old_flags; 876 fs->fs_old_flags |= FS_FLAGS_UPDATED; 877 fs->fs_sblockloc = sblockloc; 878 } 879 /* 880 * If not yet done, update UFS1 superblock with new wider fields. 881 */ 882 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 883 fs->fs_maxbsize = fs->fs_bsize; 884 fs->fs_time = fs->fs_old_time; 885 fs->fs_size = fs->fs_old_size; 886 fs->fs_dsize = fs->fs_old_dsize; 887 fs->fs_csaddr = fs->fs_old_csaddr; 888 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 889 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 890 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 891 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 892 } 893 if (fs->fs_magic == FS_UFS1_MAGIC && 894 fs->fs_old_inodefmt < FS_44INODEFMT) { 895 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 896 fs->fs_qbmask = ~fs->fs_bmask; 897 fs->fs_qfmask = ~fs->fs_fmask; 898 } 899 if (fs->fs_magic == FS_UFS1_MAGIC) { 900 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 901 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 902 if (fs->fs_maxfilesize > maxfilesize) 903 fs->fs_maxfilesize = maxfilesize; 904 } 905 /* Compatibility for old filesystems */ 906 if (fs->fs_avgfilesize <= 0) 907 fs->fs_avgfilesize = AVFILESIZ; 908 if (fs->fs_avgfpdir <= 0) 909 fs->fs_avgfpdir = AFPDIR; 910 if (bigcgs) { 911 fs->fs_save_cgsize = fs->fs_cgsize; 912 fs->fs_cgsize = fs->fs_bsize; 913 } 914} 915 916/* 917 * Unwinding superblock updates for old filesystems. 918 * See ffs_oldfscompat_read above for details. 919 * 920 * XXX - Parts get retired eventually. 921 * Unfortunately new bits get added. 922 */ 923static void 924ffs_oldfscompat_write(fs, ump) 925 struct fs *fs; 926 struct ufsmount *ump; 927{ 928 929 /* 930 * Copy back UFS2 updated fields that UFS1 inspects. 931 */ 932 if (fs->fs_magic == FS_UFS1_MAGIC) { 933 fs->fs_old_time = fs->fs_time; 934 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 935 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 936 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 937 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 938 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 939 } 940 if (bigcgs) { 941 fs->fs_cgsize = fs->fs_save_cgsize; 942 fs->fs_save_cgsize = 0; 943 } 944} 945 946/* 947 * unmount system call 948 */ 949static int 950ffs_unmount(mp, mntflags, td) 951 struct mount *mp; 952 int mntflags; 953 struct thread *td; 954{ 955 struct ufsmount *ump = VFSTOUFS(mp); 956 struct fs *fs; 957 int error, flags; 958 959 flags = 0; 960 if (mntflags & MNT_FORCE) { 961 flags |= FORCECLOSE; 962 } 963#ifdef UFS_EXTATTR 964 if ((error = ufs_extattr_stop(mp, td))) { 965 if (error != EOPNOTSUPP) 966 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 967 error); 968 } else { 969 ufs_extattr_uepm_destroy(&ump->um_extattr); 970 } 971#endif 972 if (mp->mnt_flag & MNT_SOFTDEP) { 973 if ((error = softdep_flushfiles(mp, flags, td)) != 0) 974 return (error); 975 } else { 976 if ((error = ffs_flushfiles(mp, flags, td)) != 0) 977 return (error); 978 } 979 fs = ump->um_fs; 980 UFS_LOCK(ump); 981 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 982 printf("%s: unmount pending error: blocks %jd files %d\n", 983 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 984 fs->fs_pendinginodes); 985 fs->fs_pendingblocks = 0; 986 fs->fs_pendinginodes = 0; 987 } 988 UFS_UNLOCK(ump); 989 if (fs->fs_ronly == 0) { 990 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 991 error = ffs_sbupdate(ump, MNT_WAIT, 0); 992 if (error) { 993 fs->fs_clean = 0; 994 return (error); 995 } 996 } 997 DROP_GIANT(); 998 g_topology_lock(); 999 g_vfs_close(ump->um_cp, td); 1000 g_topology_unlock(); 1001 PICKUP_GIANT(); 1002 vrele(ump->um_devvp); 1003 mtx_destroy(UFS_MTX(ump)); 1004 free(fs->fs_csp, M_UFSMNT); 1005 free(fs, M_UFSMNT); 1006 free(ump, M_UFSMNT); 1007 mp->mnt_data = (qaddr_t)0; 1008 mp->mnt_flag &= ~MNT_LOCAL; 1009 return (error); 1010} 1011 1012/* 1013 * Flush out all the files in a filesystem. 1014 */ 1015int 1016ffs_flushfiles(mp, flags, td) 1017 struct mount *mp; 1018 int flags; 1019 struct thread *td; 1020{ 1021 struct ufsmount *ump; 1022 int error; 1023 1024 ump = VFSTOUFS(mp); 1025#ifdef QUOTA 1026 if (mp->mnt_flag & MNT_QUOTA) { 1027 int i; 1028 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1029 if (error) 1030 return (error); 1031 for (i = 0; i < MAXQUOTAS; i++) { 1032 if (ump->um_quotas[i] == NULLVP) 1033 continue; 1034 quotaoff(td, mp, i); 1035 } 1036 /* 1037 * Here we fall through to vflush again to ensure 1038 * that we have gotten rid of all the system vnodes. 1039 */ 1040 } 1041#endif 1042 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1043 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1044 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1045 return (error); 1046 ffs_snapshot_unmount(mp); 1047 flags |= FORCECLOSE; 1048 /* 1049 * Here we fall through to vflush again to ensure 1050 * that we have gotten rid of all the system vnodes. 1051 */ 1052 } 1053 /* 1054 * Flush all the files. 1055 */ 1056 if ((error = vflush(mp, 0, flags, td)) != 0) 1057 return (error); 1058 /* 1059 * Flush filesystem metadata. 1060 */ 1061 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td); 1062 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1063 VOP_UNLOCK(ump->um_devvp, 0, td); 1064 return (error); 1065} 1066 1067/* 1068 * Get filesystem statistics. 1069 */ 1070static int 1071ffs_statfs(mp, sbp, td) 1072 struct mount *mp; 1073 struct statfs *sbp; 1074 struct thread *td; 1075{ 1076 struct ufsmount *ump; 1077 struct fs *fs; 1078 1079 ump = VFSTOUFS(mp); 1080 fs = ump->um_fs; 1081 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1082 panic("ffs_statfs"); 1083 sbp->f_version = STATFS_VERSION; 1084 sbp->f_bsize = fs->fs_fsize; 1085 sbp->f_iosize = fs->fs_bsize; 1086 sbp->f_blocks = fs->fs_dsize; 1087 UFS_LOCK(ump); 1088 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1089 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1090 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1091 dbtofsb(fs, fs->fs_pendingblocks); 1092 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1093 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1094 UFS_UNLOCK(ump); 1095 sbp->f_namemax = NAME_MAX; 1096 return (0); 1097} 1098 1099/* 1100 * Go through the disk queues to initiate sandbagged IO; 1101 * go through the inodes to write those that have been modified; 1102 * initiate the writing of the super block if it has been modified. 1103 * 1104 * Note: we are always called with the filesystem marked `MPBUSY'. 1105 */ 1106static int 1107ffs_sync(mp, waitfor, td) 1108 struct mount *mp; 1109 int waitfor; 1110 struct thread *td; 1111{ 1112 struct vnode *mvp, *vp, *devvp; 1113 struct inode *ip; 1114 struct ufsmount *ump = VFSTOUFS(mp); 1115 struct fs *fs; 1116 int error, count, wait, lockreq, allerror = 0; 1117 int suspend; 1118 int suspended; 1119 int secondary_writes; 1120 int secondary_accwrites; 1121 int softdep_deps; 1122 int softdep_accdeps; 1123 struct bufobj *bo; 1124 1125 fs = ump->um_fs; 1126 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1127 printf("fs = %s\n", fs->fs_fsmnt); 1128 panic("ffs_sync: rofs mod"); 1129 } 1130 /* 1131 * Write back each (modified) inode. 1132 */ 1133 wait = 0; 1134 suspend = 0; 1135 suspended = 0; 1136 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1137 if (waitfor == MNT_SUSPEND) { 1138 suspend = 1; 1139 waitfor = MNT_WAIT; 1140 } 1141 if (waitfor == MNT_WAIT) { 1142 wait = 1; 1143 lockreq = LK_EXCLUSIVE; 1144 } 1145 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1146 MNT_ILOCK(mp); 1147loop: 1148 /* Grab snapshot of secondary write counts */ 1149 secondary_writes = mp->mnt_secondary_writes; 1150 secondary_accwrites = mp->mnt_secondary_accwrites; 1151 1152 /* Grab snapshot of softdep dependency counts */ 1153 MNT_IUNLOCK(mp); 1154 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1155 MNT_ILOCK(mp); 1156 1157 MNT_VNODE_FOREACH(vp, mp, mvp) { 1158 /* 1159 * Depend on the mntvnode_slock to keep things stable enough 1160 * for a quick test. Since there might be hundreds of 1161 * thousands of vnodes, we cannot afford even a subroutine 1162 * call unless there's a good chance that we have work to do. 1163 */ 1164 VI_LOCK(vp); 1165 if (vp->v_iflag & VI_DOOMED) { 1166 VI_UNLOCK(vp); 1167 continue; 1168 } 1169 ip = VTOI(vp); 1170 if (vp->v_type == VNON || ((ip->i_flag & 1171 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1172 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1173 VI_UNLOCK(vp); 1174 continue; 1175 } 1176 MNT_IUNLOCK(mp); 1177 if ((error = vget(vp, lockreq, td)) != 0) { 1178 MNT_ILOCK(mp); 1179 if (error == ENOENT || error == ENOLCK) { 1180 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1181 goto loop; 1182 } 1183 continue; 1184 } 1185 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1186 allerror = error; 1187 vput(vp); 1188 MNT_ILOCK(mp); 1189 } 1190 MNT_IUNLOCK(mp); 1191 /* 1192 * Force stale filesystem control information to be flushed. 1193 */ 1194 if (waitfor == MNT_WAIT) { 1195 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1196 allerror = error; 1197 /* Flushed work items may create new vnodes to clean */ 1198 if (allerror == 0 && count) { 1199 MNT_ILOCK(mp); 1200 goto loop; 1201 } 1202 } 1203#ifdef QUOTA 1204 qsync(mp); 1205#endif 1206 devvp = ump->um_devvp; 1207 VI_LOCK(devvp); 1208 bo = &devvp->v_bufobj; 1209 if (waitfor != MNT_LAZY && 1210 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1211 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td); 1212 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1213 allerror = error; 1214 VOP_UNLOCK(devvp, 0, td); 1215 if (allerror == 0 && waitfor == MNT_WAIT) { 1216 MNT_ILOCK(mp); 1217 goto loop; 1218 } 1219 } else if (suspend != 0) { 1220 if (softdep_check_suspend(mp, 1221 devvp, 1222 softdep_deps, 1223 softdep_accdeps, 1224 secondary_writes, 1225 secondary_accwrites) != 0) 1226 goto loop; /* More work needed */ 1227 mtx_assert(MNT_MTX(mp), MA_OWNED); 1228 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1229 MNT_IUNLOCK(mp); 1230 suspended = 1; 1231 } else 1232 VI_UNLOCK(devvp); 1233 /* 1234 * Write back modified superblock. 1235 */ 1236 if (fs->fs_fmod != 0 && 1237 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1238 allerror = error; 1239 return (allerror); 1240} 1241 1242int 1243ffs_vget(mp, ino, flags, vpp) 1244 struct mount *mp; 1245 ino_t ino; 1246 int flags; 1247 struct vnode **vpp; 1248{ 1249 struct fs *fs; 1250 struct inode *ip; 1251 struct ufsmount *ump; 1252 struct buf *bp; 1253 struct vnode *vp; 1254 struct cdev *dev; 1255 int error; 1256 1257 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1258 if (error || *vpp != NULL) 1259 return (error); 1260 1261 /* 1262 * We must promote to an exclusive lock for vnode creation. This 1263 * can happen if lookup is passed LOCKSHARED. 1264 */ 1265 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1266 flags &= ~LK_TYPE_MASK; 1267 flags |= LK_EXCLUSIVE; 1268 } 1269 1270 /* 1271 * We do not lock vnode creation as it is believed to be too 1272 * expensive for such rare case as simultaneous creation of vnode 1273 * for same ino by different processes. We just allow them to race 1274 * and check later to decide who wins. Let the race begin! 1275 */ 1276 1277 ump = VFSTOUFS(mp); 1278 dev = ump->um_dev; 1279 fs = ump->um_fs; 1280 1281 /* 1282 * If this MALLOC() is performed after the getnewvnode() 1283 * it might block, leaving a vnode with a NULL v_data to be 1284 * found by ffs_sync() if a sync happens to fire right then, 1285 * which will cause a panic because ffs_sync() blindly 1286 * dereferences vp->v_data (as well it should). 1287 */ 1288 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1289 1290 /* Allocate a new vnode/inode. */ 1291 if (fs->fs_magic == FS_UFS1_MAGIC) 1292 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1293 else 1294 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1295 if (error) { 1296 *vpp = NULL; 1297 uma_zfree(uma_inode, ip); 1298 return (error); 1299 } 1300 /* 1301 * FFS supports recursive and shared locking. 1302 */ 1303 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 1304 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 1305 vp->v_data = ip; 1306 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1307 ip->i_vnode = vp; 1308 ip->i_ump = ump; 1309 ip->i_fs = fs; 1310 ip->i_dev = dev; 1311 ip->i_number = ino; 1312#ifdef QUOTA 1313 { 1314 int i; 1315 for (i = 0; i < MAXQUOTAS; i++) 1316 ip->i_dquot[i] = NODQUOT; 1317 } 1318#endif 1319 1320 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1321 if (error || *vpp != NULL) 1322 return (error); 1323 1324 /* Read in the disk contents for the inode, copy into the inode. */ 1325 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1326 (int)fs->fs_bsize, NOCRED, &bp); 1327 if (error) { 1328 /* 1329 * The inode does not contain anything useful, so it would 1330 * be misleading to leave it on its hash chain. With mode 1331 * still zero, it will be unlinked and returned to the free 1332 * list by vput(). 1333 */ 1334 brelse(bp); 1335 vput(vp); 1336 *vpp = NULL; 1337 return (error); 1338 } 1339 if (ip->i_ump->um_fstype == UFS1) 1340 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1341 else 1342 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1343 ffs_load_inode(bp, ip, fs, ino); 1344 if (DOINGSOFTDEP(vp)) 1345 softdep_load_inodeblock(ip); 1346 else 1347 ip->i_effnlink = ip->i_nlink; 1348 bqrelse(bp); 1349 1350 /* 1351 * Initialize the vnode from the inode, check for aliases. 1352 * Note that the underlying vnode may have changed. 1353 */ 1354 if (ip->i_ump->um_fstype == UFS1) 1355 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1356 else 1357 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1358 if (error) { 1359 vput(vp); 1360 *vpp = NULL; 1361 return (error); 1362 } 1363 1364 /* 1365 * Finish inode initialization. 1366 */ 1367 1368 /* 1369 * Set up a generation number for this inode if it does not 1370 * already have one. This should only happen on old filesystems. 1371 */ 1372 if (ip->i_gen == 0) { 1373 ip->i_gen = arc4random() / 2 + 1; 1374 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1375 ip->i_flag |= IN_MODIFIED; 1376 DIP_SET(ip, i_gen, ip->i_gen); 1377 } 1378 } 1379 /* 1380 * Ensure that uid and gid are correct. This is a temporary 1381 * fix until fsck has been changed to do the update. 1382 */ 1383 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1384 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1385 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1386 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1387 } /* XXX */ 1388 1389#ifdef MAC 1390 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1391 /* 1392 * If this vnode is already allocated, and we're running 1393 * multi-label, attempt to perform a label association 1394 * from the extended attributes on the inode. 1395 */ 1396 error = mac_associate_vnode_extattr(mp, vp); 1397 if (error) { 1398 /* ufs_inactive will release ip->i_devvp ref. */ 1399 vput(vp); 1400 *vpp = NULL; 1401 return (error); 1402 } 1403 } 1404#endif 1405 1406 *vpp = vp; 1407 return (0); 1408} 1409 1410/* 1411 * File handle to vnode 1412 * 1413 * Have to be really careful about stale file handles: 1414 * - check that the inode number is valid 1415 * - call ffs_vget() to get the locked inode 1416 * - check for an unallocated inode (i_mode == 0) 1417 * - check that the given client host has export rights and return 1418 * those rights via. exflagsp and credanonp 1419 */ 1420static int 1421ffs_fhtovp(mp, fhp, vpp) 1422 struct mount *mp; 1423 struct fid *fhp; 1424 struct vnode **vpp; 1425{ 1426 struct ufid *ufhp; 1427 struct fs *fs; 1428 1429 ufhp = (struct ufid *)fhp; 1430 fs = VFSTOUFS(mp)->um_fs; 1431 if (ufhp->ufid_ino < ROOTINO || 1432 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1433 return (ESTALE); 1434 return (ufs_fhtovp(mp, ufhp, vpp)); 1435} 1436 1437/* 1438 * Vnode pointer to File handle 1439 */ 1440/* ARGSUSED */ 1441static int 1442ffs_vptofh(vp, fhp) 1443 struct vnode *vp; 1444 struct fid *fhp; 1445{ 1446 struct inode *ip; 1447 struct ufid *ufhp; 1448 1449 ip = VTOI(vp); 1450 ufhp = (struct ufid *)fhp; 1451 ufhp->ufid_len = sizeof(struct ufid); 1452 ufhp->ufid_ino = ip->i_number; 1453 ufhp->ufid_gen = ip->i_gen; 1454 return (0); 1455} 1456 1457/* 1458 * Initialize the filesystem. 1459 */ 1460static int 1461ffs_init(vfsp) 1462 struct vfsconf *vfsp; 1463{ 1464 1465 softdep_initialize(); 1466 return (ufs_init(vfsp)); 1467} 1468 1469/* 1470 * Undo the work of ffs_init(). 1471 */ 1472static int 1473ffs_uninit(vfsp) 1474 struct vfsconf *vfsp; 1475{ 1476 int ret; 1477 1478 ret = ufs_uninit(vfsp); 1479 softdep_uninitialize(); 1480 return (ret); 1481} 1482 1483/* 1484 * Write a superblock and associated information back to disk. 1485 */ 1486static int 1487ffs_sbupdate(mp, waitfor, suspended) 1488 struct ufsmount *mp; 1489 int waitfor; 1490 int suspended; 1491{ 1492 struct fs *fs = mp->um_fs; 1493 struct buf *sbbp; 1494 struct buf *bp; 1495 int blks; 1496 void *space; 1497 int i, size, error, allerror = 0; 1498 1499 if (fs->fs_ronly == 1 && 1500 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1501 (MNT_RDONLY | MNT_UPDATE)) 1502 panic("ffs_sbupdate: write read-only filesystem"); 1503 /* 1504 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1505 */ 1506 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1507 0, 0, 0); 1508 /* 1509 * First write back the summary information. 1510 */ 1511 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1512 space = fs->fs_csp; 1513 for (i = 0; i < blks; i += fs->fs_frag) { 1514 size = fs->fs_bsize; 1515 if (i + fs->fs_frag > blks) 1516 size = (blks - i) * fs->fs_fsize; 1517 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1518 size, 0, 0, 0); 1519 bcopy(space, bp->b_data, (u_int)size); 1520 space = (char *)space + size; 1521 if (suspended) 1522 bp->b_flags |= B_VALIDSUSPWRT; 1523 if (waitfor != MNT_WAIT) 1524 bawrite(bp); 1525 else if ((error = bwrite(bp)) != 0) 1526 allerror = error; 1527 } 1528 /* 1529 * Now write back the superblock itself. If any errors occurred 1530 * up to this point, then fail so that the superblock avoids 1531 * being written out as clean. 1532 */ 1533 if (allerror) { 1534 brelse(sbbp); 1535 return (allerror); 1536 } 1537 bp = sbbp; 1538 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1539 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1540 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1541 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1542 fs->fs_sblockloc = SBLOCK_UFS1; 1543 } 1544 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1545 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1546 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1547 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1548 fs->fs_sblockloc = SBLOCK_UFS2; 1549 } 1550 fs->fs_fmod = 0; 1551 fs->fs_time = time_second; 1552 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1553 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1554 if (suspended) 1555 bp->b_flags |= B_VALIDSUSPWRT; 1556 if (waitfor != MNT_WAIT) 1557 bawrite(bp); 1558 else if ((error = bwrite(bp)) != 0) 1559 allerror = error; 1560 return (allerror); 1561} 1562 1563static int 1564ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1565 int attrnamespace, const char *attrname, struct thread *td) 1566{ 1567 1568#ifdef UFS_EXTATTR 1569 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1570 attrname, td)); 1571#else 1572 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1573 attrname, td)); 1574#endif 1575} 1576 1577static void 1578ffs_ifree(struct ufsmount *ump, struct inode *ip) 1579{ 1580 1581 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1582 uma_zfree(uma_ufs1, ip->i_din1); 1583 else if (ip->i_din2 != NULL) 1584 uma_zfree(uma_ufs2, ip->i_din2); 1585 uma_zfree(uma_inode, ip); 1586} 1587 1588static int dobkgrdwrite = 1; 1589SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1590 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1591 1592/* 1593 * Complete a background write started from bwrite. 1594 */ 1595static void 1596ffs_backgroundwritedone(struct buf *bp) 1597{ 1598 struct bufobj *bufobj; 1599 struct buf *origbp; 1600 1601 /* 1602 * Find the original buffer that we are writing. 1603 */ 1604 bufobj = bp->b_bufobj; 1605 BO_LOCK(bufobj); 1606 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1607 panic("backgroundwritedone: lost buffer"); 1608 /* Grab an extra reference to be dropped by the bufdone() below. */ 1609 bufobj_wrefl(bufobj); 1610 BO_UNLOCK(bufobj); 1611 /* 1612 * Process dependencies then return any unfinished ones. 1613 */ 1614 if (LIST_FIRST(&bp->b_dep) != NULL) 1615 buf_complete(bp); 1616#ifdef SOFTUPDATES 1617 if (LIST_FIRST(&bp->b_dep) != NULL) 1618 softdep_move_dependencies(bp, origbp); 1619#endif 1620 /* 1621 * This buffer is marked B_NOCACHE so when it is released 1622 * by biodone it will be tossed. 1623 */ 1624 bp->b_flags |= B_NOCACHE; 1625 bp->b_flags &= ~B_CACHE; 1626 bufdone(bp); 1627 BO_LOCK(bufobj); 1628 /* 1629 * Clear the BV_BKGRDINPROG flag in the original buffer 1630 * and awaken it if it is waiting for the write to complete. 1631 * If BV_BKGRDINPROG is not set in the original buffer it must 1632 * have been released and re-instantiated - which is not legal. 1633 */ 1634 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1635 ("backgroundwritedone: lost buffer2")); 1636 origbp->b_vflags &= ~BV_BKGRDINPROG; 1637 if (origbp->b_vflags & BV_BKGRDWAIT) { 1638 origbp->b_vflags &= ~BV_BKGRDWAIT; 1639 wakeup(&origbp->b_xflags); 1640 } 1641 BO_UNLOCK(bufobj); 1642} 1643 1644 1645/* 1646 * Write, release buffer on completion. (Done by iodone 1647 * if async). Do not bother writing anything if the buffer 1648 * is invalid. 1649 * 1650 * Note that we set B_CACHE here, indicating that buffer is 1651 * fully valid and thus cacheable. This is true even of NFS 1652 * now so we set it generally. This could be set either here 1653 * or in biodone() since the I/O is synchronous. We put it 1654 * here. 1655 */ 1656static int 1657ffs_bufwrite(struct buf *bp) 1658{ 1659 int oldflags, s; 1660 struct buf *newbp; 1661 1662 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1663 if (bp->b_flags & B_INVAL) { 1664 brelse(bp); 1665 return (0); 1666 } 1667 1668 oldflags = bp->b_flags; 1669 1670 if (BUF_REFCNT(bp) == 0) 1671 panic("bufwrite: buffer is not busy???"); 1672 s = splbio(); 1673 /* 1674 * If a background write is already in progress, delay 1675 * writing this block if it is asynchronous. Otherwise 1676 * wait for the background write to complete. 1677 */ 1678 BO_LOCK(bp->b_bufobj); 1679 if (bp->b_vflags & BV_BKGRDINPROG) { 1680 if (bp->b_flags & B_ASYNC) { 1681 BO_UNLOCK(bp->b_bufobj); 1682 splx(s); 1683 bdwrite(bp); 1684 return (0); 1685 } 1686 bp->b_vflags |= BV_BKGRDWAIT; 1687 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1688 if (bp->b_vflags & BV_BKGRDINPROG) 1689 panic("bufwrite: still writing"); 1690 } 1691 BO_UNLOCK(bp->b_bufobj); 1692 1693 /* Mark the buffer clean */ 1694 bundirty(bp); 1695 1696 /* 1697 * If this buffer is marked for background writing and we 1698 * do not have to wait for it, make a copy and write the 1699 * copy so as to leave this buffer ready for further use. 1700 * 1701 * This optimization eats a lot of memory. If we have a page 1702 * or buffer shortfall we can't do it. 1703 */ 1704 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1705 (bp->b_flags & B_ASYNC) && 1706 !vm_page_count_severe() && 1707 !buf_dirty_count_severe()) { 1708 KASSERT(bp->b_iodone == NULL, 1709 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1710 1711 /* get a new block */ 1712 newbp = geteblk(bp->b_bufsize); 1713 1714 /* 1715 * set it to be identical to the old block. We have to 1716 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1717 * to avoid confusing the splay tree and gbincore(). 1718 */ 1719 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1720 newbp->b_lblkno = bp->b_lblkno; 1721 newbp->b_xflags |= BX_BKGRDMARKER; 1722 BO_LOCK(bp->b_bufobj); 1723 bp->b_vflags |= BV_BKGRDINPROG; 1724 bgetvp(bp->b_vp, newbp); 1725 BO_UNLOCK(bp->b_bufobj); 1726 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1727 newbp->b_blkno = bp->b_blkno; 1728 newbp->b_offset = bp->b_offset; 1729 newbp->b_iodone = ffs_backgroundwritedone; 1730 newbp->b_flags |= B_ASYNC; 1731 newbp->b_flags &= ~B_INVAL; 1732 1733#ifdef SOFTUPDATES 1734 /* move over the dependencies */ 1735 if (LIST_FIRST(&bp->b_dep) != NULL) 1736 softdep_move_dependencies(bp, newbp); 1737#endif 1738 1739 /* 1740 * Initiate write on the copy, release the original to 1741 * the B_LOCKED queue so that it cannot go away until 1742 * the background write completes. If not locked it could go 1743 * away and then be reconstituted while it was being written. 1744 * If the reconstituted buffer were written, we could end up 1745 * with two background copies being written at the same time. 1746 */ 1747 bqrelse(bp); 1748 bp = newbp; 1749 } 1750 1751 /* Let the normal bufwrite do the rest for us */ 1752 return (bufwrite(bp)); 1753} 1754 1755 1756static void 1757ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1758{ 1759 struct vnode *vp; 1760 int error; 1761 1762 vp = bo->__bo_vnode; 1763 if (bp->b_iocmd == BIO_WRITE) { 1764#ifdef SOFTUPDATES 1765 if (LIST_FIRST(&bp->b_dep) != NULL) 1766 buf_start(bp); 1767#endif 1768 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1769 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1770 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1771 panic("ffs_geom_strategy: bad I/O"); 1772 bp->b_flags &= ~B_VALIDSUSPWRT; 1773 if ((vp->v_vflag & VV_COPYONWRITE) && 1774 vp->v_rdev->si_snapdata != NULL && 1775 (error = (ffs_copyonwrite)(vp, bp)) != 0 && 1776 error != EOPNOTSUPP) { 1777 bp->b_error = error; 1778 bp->b_ioflags |= BIO_ERROR; 1779 bufdone(bp); 1780 return; 1781 } 1782 } 1783 g_vfs_strategy(bo, bp); 1784} 1785