ffs_vfsops.c revision 149358
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 149358 2005-08-21 22:06:41Z ssouhlal $"); 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); 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[] = { "from", "export", NULL }; 120 121static int 122ffs_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, 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 vrele(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 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 328 if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){ 329 vput(devvp); 330 return (error); 331 } 332 VOP_UNLOCK(devvp, 0, td); 333 } 334 335 if (mp->mnt_flag & MNT_UPDATE) { 336 /* 337 * Update only 338 * 339 * If it's not the same vnode, or at least the same device 340 * then it's not correct. 341 */ 342 343 if (devvp->v_rdev != ump->um_devvp->v_rdev) 344 error = EINVAL; /* needs translation */ 345 vrele(devvp); 346 if (error) 347 return (error); 348 } else { 349 /* 350 * New mount 351 * 352 * We need the name for the mount point (also used for 353 * "last mounted on") copied in. If an error occurs, 354 * the mount point is discarded by the upper level code. 355 * Note that vfs_mount() populates f_mntonname for us. 356 */ 357 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 358 vrele(devvp); 359 return (error); 360 } 361 } 362 vfs_mountedfrom(mp, fspec); 363 return (0); 364} 365 366/* 367 * Compatibility with old mount system call. 368 */ 369 370static int 371ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td) 372{ 373 struct ufs_args args; 374 int error; 375 376 if (data == NULL) 377 return (EINVAL); 378 error = copyin(data, &args, sizeof args); 379 if (error) 380 return (error); 381 382 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 383 ma = mount_arg(ma, "export", &args.export, sizeof args.export); 384 error = kernel_mount(ma, flags); 385 386 return (error); 387} 388 389/* 390 * Reload all incore data for a filesystem (used after running fsck on 391 * the root filesystem and finding things to fix). The filesystem must 392 * be mounted read-only. 393 * 394 * Things to do to update the mount: 395 * 1) invalidate all cached meta-data. 396 * 2) re-read superblock from disk. 397 * 3) re-read summary information from disk. 398 * 4) invalidate all inactive vnodes. 399 * 5) invalidate all cached file data. 400 * 6) re-read inode data for all active vnodes. 401 */ 402static int 403ffs_reload(struct mount *mp, struct thread *td) 404{ 405 struct vnode *vp, *nvp, *devvp; 406 struct inode *ip; 407 void *space; 408 struct buf *bp; 409 struct fs *fs, *newfs; 410 struct ufsmount *ump; 411 ufs2_daddr_t sblockloc; 412 int i, blks, size, error; 413 int32_t *lp; 414 415 if ((mp->mnt_flag & MNT_RDONLY) == 0) 416 return (EINVAL); 417 ump = VFSTOUFS(mp); 418 /* 419 * Step 1: invalidate all cached meta-data. 420 */ 421 devvp = VFSTOUFS(mp)->um_devvp; 422 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td); 423 if (vinvalbuf(devvp, 0, td, 0, 0) != 0) 424 panic("ffs_reload: dirty1"); 425 VOP_UNLOCK(devvp, 0, td); 426 427 /* 428 * Step 2: re-read superblock from disk. 429 */ 430 fs = VFSTOUFS(mp)->um_fs; 431 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 432 NOCRED, &bp)) != 0) 433 return (error); 434 newfs = (struct fs *)bp->b_data; 435 if ((newfs->fs_magic != FS_UFS1_MAGIC && 436 newfs->fs_magic != FS_UFS2_MAGIC) || 437 newfs->fs_bsize > MAXBSIZE || 438 newfs->fs_bsize < sizeof(struct fs)) { 439 brelse(bp); 440 return (EIO); /* XXX needs translation */ 441 } 442 /* 443 * Copy pointer fields back into superblock before copying in XXX 444 * new superblock. These should really be in the ufsmount. XXX 445 * Note that important parameters (eg fs_ncg) are unchanged. 446 */ 447 newfs->fs_csp = fs->fs_csp; 448 newfs->fs_maxcluster = fs->fs_maxcluster; 449 newfs->fs_contigdirs = fs->fs_contigdirs; 450 newfs->fs_active = fs->fs_active; 451 /* The file system is still read-only. */ 452 newfs->fs_ronly = 1; 453 sblockloc = fs->fs_sblockloc; 454 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 455 brelse(bp); 456 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 457 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 458 UFS_LOCK(ump); 459 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 460 printf("%s: reload pending error: blocks %jd files %d\n", 461 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 462 fs->fs_pendinginodes); 463 fs->fs_pendingblocks = 0; 464 fs->fs_pendinginodes = 0; 465 } 466 UFS_UNLOCK(ump); 467 468 /* 469 * Step 3: re-read summary information from disk. 470 */ 471 blks = howmany(fs->fs_cssize, fs->fs_fsize); 472 space = fs->fs_csp; 473 for (i = 0; i < blks; i += fs->fs_frag) { 474 size = fs->fs_bsize; 475 if (i + fs->fs_frag > blks) 476 size = (blks - i) * fs->fs_fsize; 477 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 478 NOCRED, &bp); 479 if (error) 480 return (error); 481 bcopy(bp->b_data, space, (u_int)size); 482 space = (char *)space + size; 483 brelse(bp); 484 } 485 /* 486 * We no longer know anything about clusters per cylinder group. 487 */ 488 if (fs->fs_contigsumsize > 0) { 489 lp = fs->fs_maxcluster; 490 for (i = 0; i < fs->fs_ncg; i++) 491 *lp++ = fs->fs_contigsumsize; 492 } 493 494loop: 495 MNT_ILOCK(mp); 496 MNT_VNODE_FOREACH(vp, mp, nvp) { 497 VI_LOCK(vp); 498 if (vp->v_iflag & VI_DOOMED) { 499 VI_UNLOCK(vp); 500 continue; 501 } 502 MNT_IUNLOCK(mp); 503 /* 504 * Step 4: invalidate all cached file data. 505 */ 506 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 507 goto loop; 508 } 509 if (vinvalbuf(vp, 0, td, 0, 0)) 510 panic("ffs_reload: dirty2"); 511 /* 512 * Step 5: re-read inode data for all active vnodes. 513 */ 514 ip = VTOI(vp); 515 error = 516 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 517 (int)fs->fs_bsize, NOCRED, &bp); 518 if (error) { 519 VOP_UNLOCK(vp, 0, td); 520 vrele(vp); 521 return (error); 522 } 523 ffs_load_inode(bp, ip, fs, ip->i_number); 524 ip->i_effnlink = ip->i_nlink; 525 brelse(bp); 526 VOP_UNLOCK(vp, 0, td); 527 vrele(vp); 528 MNT_ILOCK(mp); 529 } 530 MNT_IUNLOCK(mp); 531 return (0); 532} 533 534/* 535 * Possible superblock locations ordered from most to least likely. 536 */ 537static int sblock_try[] = SBLOCKSEARCH; 538 539/* 540 * Common code for mount and mountroot 541 */ 542static int 543ffs_mountfs(devvp, mp, td) 544 struct vnode *devvp; 545 struct mount *mp; 546 struct thread *td; 547{ 548 struct ufsmount *ump; 549 struct buf *bp; 550 struct fs *fs; 551 struct cdev *dev; 552 void *space; 553 ufs2_daddr_t sblockloc; 554 int error, i, blks, size, ronly; 555 int32_t *lp; 556 struct ucred *cred; 557 struct g_consumer *cp; 558 559 dev = devvp->v_rdev; 560 cred = td ? td->td_ucred : NOCRED; 561 562 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 563 DROP_GIANT(); 564 g_topology_lock(); 565 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 566 567 /* 568 * If we are a root mount, drop the E flag so fsck can do its magic. 569 * We will pick it up again when we remount R/W. 570 */ 571 if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS)) 572 error = g_access(cp, 0, 0, -1); 573 g_topology_unlock(); 574 PICKUP_GIANT(); 575 VOP_UNLOCK(devvp, 0, td); 576 if (error) 577 return (error); 578 if (devvp->v_rdev->si_iosize_max != 0) 579 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max; 580 if (mp->mnt_iosize_max > MAXPHYS) 581 mp->mnt_iosize_max = MAXPHYS; 582 583 devvp->v_bufobj.bo_private = cp; 584 devvp->v_bufobj.bo_ops = &ffs_ops; 585 586 bp = NULL; 587 ump = NULL; 588 fs = NULL; 589 sblockloc = 0; 590 /* 591 * Try reading the superblock in each of its possible locations. 592 */ 593 for (i = 0; sblock_try[i] != -1; i++) { 594 if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE, 595 cred, &bp)) != 0) 596 goto out; 597 fs = (struct fs *)bp->b_data; 598 sblockloc = sblock_try[i]; 599 if ((fs->fs_magic == FS_UFS1_MAGIC || 600 (fs->fs_magic == FS_UFS2_MAGIC && 601 (fs->fs_sblockloc == sblockloc || 602 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 603 fs->fs_bsize <= MAXBSIZE && 604 fs->fs_bsize >= sizeof(struct fs)) 605 break; 606 brelse(bp); 607 bp = NULL; 608 } 609 if (sblock_try[i] == -1) { 610 error = EINVAL; /* XXX needs translation */ 611 goto out; 612 } 613 fs->fs_fmod = 0; 614 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 615 fs->fs_flags &= ~FS_UNCLEAN; 616 if (fs->fs_clean == 0) { 617 fs->fs_flags |= FS_UNCLEAN; 618 if (ronly || (mp->mnt_flag & MNT_FORCE) || 619 ((fs->fs_flags & FS_NEEDSFSCK) == 0 && 620 (fs->fs_flags & FS_DOSOFTDEP))) { 621 printf( 622"WARNING: %s was not properly dismounted\n", 623 fs->fs_fsmnt); 624 } else { 625 printf( 626"WARNING: R/W mount of %s denied. Filesystem is not clean - run fsck\n", 627 fs->fs_fsmnt); 628 error = EPERM; 629 goto out; 630 } 631 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 632 (mp->mnt_flag & MNT_FORCE)) { 633 printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt, 634 (intmax_t)fs->fs_pendingblocks, 635 fs->fs_pendinginodes); 636 fs->fs_pendingblocks = 0; 637 fs->fs_pendinginodes = 0; 638 } 639 } 640 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 641 printf("%s: mount pending error: blocks %jd files %d\n", 642 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 643 fs->fs_pendinginodes); 644 fs->fs_pendingblocks = 0; 645 fs->fs_pendinginodes = 0; 646 } 647 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 648 ump->um_cp = cp; 649 ump->um_bo = &devvp->v_bufobj; 650 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 651 if (fs->fs_magic == FS_UFS1_MAGIC) { 652 ump->um_fstype = UFS1; 653 ump->um_balloc = ffs_balloc_ufs1; 654 } else { 655 ump->um_fstype = UFS2; 656 ump->um_balloc = ffs_balloc_ufs2; 657 } 658 ump->um_blkatoff = ffs_blkatoff; 659 ump->um_truncate = ffs_truncate; 660 ump->um_update = ffs_update; 661 ump->um_valloc = ffs_valloc; 662 ump->um_vfree = ffs_vfree; 663 ump->um_ifree = ffs_ifree; 664 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 665 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 666 if (fs->fs_sbsize < SBLOCKSIZE) 667 bp->b_flags |= B_INVAL | B_NOCACHE; 668 brelse(bp); 669 bp = NULL; 670 fs = ump->um_fs; 671 ffs_oldfscompat_read(fs, ump, sblockloc); 672 fs->fs_ronly = ronly; 673 size = fs->fs_cssize; 674 blks = howmany(size, fs->fs_fsize); 675 if (fs->fs_contigsumsize > 0) 676 size += fs->fs_ncg * sizeof(int32_t); 677 size += fs->fs_ncg * sizeof(u_int8_t); 678 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 679 fs->fs_csp = space; 680 for (i = 0; i < blks; i += fs->fs_frag) { 681 size = fs->fs_bsize; 682 if (i + fs->fs_frag > blks) 683 size = (blks - i) * fs->fs_fsize; 684 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 685 cred, &bp)) != 0) { 686 free(fs->fs_csp, M_UFSMNT); 687 goto out; 688 } 689 bcopy(bp->b_data, space, (u_int)size); 690 space = (char *)space + size; 691 brelse(bp); 692 bp = NULL; 693 } 694 if (fs->fs_contigsumsize > 0) { 695 fs->fs_maxcluster = lp = space; 696 for (i = 0; i < fs->fs_ncg; i++) 697 *lp++ = fs->fs_contigsumsize; 698 space = lp; 699 } 700 size = fs->fs_ncg * sizeof(u_int8_t); 701 fs->fs_contigdirs = (u_int8_t *)space; 702 bzero(fs->fs_contigdirs, size); 703 fs->fs_active = NULL; 704 mp->mnt_data = (qaddr_t)ump; 705 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 706 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 707 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 708 vfs_getvfs(&mp->mnt_stat.f_fsid)) 709 vfs_getnewfsid(mp); 710 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 711 mp->mnt_flag |= MNT_LOCAL; 712 if ((fs->fs_flags & FS_MULTILABEL) != 0) 713#ifdef MAC 714 mp->mnt_flag |= MNT_MULTILABEL; 715#else 716 printf( 717"WARNING: %s: multilabel flag on fs but no MAC support\n", 718 fs->fs_fsmnt); 719#endif 720 if ((fs->fs_flags & FS_ACLS) != 0) 721#ifdef UFS_ACL 722 mp->mnt_flag |= MNT_ACLS; 723#else 724 printf( 725"WARNING: %s: ACLs flag on fs but no ACLs support\n", 726 fs->fs_fsmnt); 727#endif 728 ump->um_mountp = mp; 729 ump->um_dev = dev; 730 ump->um_devvp = devvp; 731 ump->um_nindir = fs->fs_nindir; 732 ump->um_bptrtodb = fs->fs_fsbtodb; 733 ump->um_seqinc = fs->fs_frag; 734 for (i = 0; i < MAXQUOTAS; i++) 735 ump->um_quotas[i] = NULLVP; 736#ifdef UFS_EXTATTR 737 ufs_extattr_uepm_init(&ump->um_extattr); 738#endif 739 /* 740 * Set FS local "last mounted on" information (NULL pad) 741 */ 742 bzero(fs->fs_fsmnt, MAXMNTLEN); 743 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 744 745 if( mp->mnt_flag & MNT_ROOTFS) { 746 /* 747 * Root mount; update timestamp in mount structure. 748 * this will be used by the common root mount code 749 * to update the system clock. 750 */ 751 mp->mnt_time = fs->fs_time; 752 } 753 754 if (ronly == 0) { 755 if ((fs->fs_flags & FS_DOSOFTDEP) && 756 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 757 free(fs->fs_csp, M_UFSMNT); 758 goto out; 759 } 760 if (fs->fs_snapinum[0] != 0) 761 ffs_snapshot_mount(mp); 762 fs->fs_fmod = 1; 763 fs->fs_clean = 0; 764 (void) ffs_sbupdate(ump, MNT_WAIT); 765 } 766 /* 767 * Initialize filesystem stat information in mount struct. 768 */ 769#ifdef UFS_EXTATTR 770#ifdef UFS_EXTATTR_AUTOSTART 771 /* 772 * 773 * Auto-starting does the following: 774 * - check for /.attribute in the fs, and extattr_start if so 775 * - for each file in .attribute, enable that file with 776 * an attribute of the same name. 777 * Not clear how to report errors -- probably eat them. 778 * This would all happen while the filesystem was busy/not 779 * available, so would effectively be "atomic". 780 */ 781 (void) ufs_extattr_autostart(mp, td); 782#endif /* !UFS_EXTATTR_AUTOSTART */ 783#endif /* !UFS_EXTATTR */ 784#ifndef QUOTA 785 mp->mnt_kern_flag |= MNTK_MPSAFE; 786#endif 787 return (0); 788out: 789 if (bp) 790 brelse(bp); 791 if (cp != NULL) { 792 DROP_GIANT(); 793 g_topology_lock(); 794 g_vfs_close(cp, td); 795 g_topology_unlock(); 796 PICKUP_GIANT(); 797 } 798 if (ump) { 799 mtx_destroy(UFS_MTX(ump)); 800 free(ump->um_fs, M_UFSMNT); 801 free(ump, M_UFSMNT); 802 mp->mnt_data = (qaddr_t)0; 803 } 804 return (error); 805} 806 807#include <sys/sysctl.h> 808static int bigcgs = 0; 809SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 810 811/* 812 * Sanity checks for loading old filesystem superblocks. 813 * See ffs_oldfscompat_write below for unwound actions. 814 * 815 * XXX - Parts get retired eventually. 816 * Unfortunately new bits get added. 817 */ 818static void 819ffs_oldfscompat_read(fs, ump, sblockloc) 820 struct fs *fs; 821 struct ufsmount *ump; 822 ufs2_daddr_t sblockloc; 823{ 824 off_t maxfilesize; 825 826 /* 827 * If not yet done, update fs_flags location and value of fs_sblockloc. 828 */ 829 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 830 fs->fs_flags = fs->fs_old_flags; 831 fs->fs_old_flags |= FS_FLAGS_UPDATED; 832 fs->fs_sblockloc = sblockloc; 833 } 834 /* 835 * If not yet done, update UFS1 superblock with new wider fields. 836 */ 837 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 838 fs->fs_maxbsize = fs->fs_bsize; 839 fs->fs_time = fs->fs_old_time; 840 fs->fs_size = fs->fs_old_size; 841 fs->fs_dsize = fs->fs_old_dsize; 842 fs->fs_csaddr = fs->fs_old_csaddr; 843 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 844 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 845 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 846 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 847 } 848 if (fs->fs_magic == FS_UFS1_MAGIC && 849 fs->fs_old_inodefmt < FS_44INODEFMT) { 850 fs->fs_maxfilesize = (u_quad_t) 1LL << 39; 851 fs->fs_qbmask = ~fs->fs_bmask; 852 fs->fs_qfmask = ~fs->fs_fmask; 853 } 854 if (fs->fs_magic == FS_UFS1_MAGIC) { 855 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 856 maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1; 857 if (fs->fs_maxfilesize > maxfilesize) 858 fs->fs_maxfilesize = maxfilesize; 859 } 860 /* Compatibility for old filesystems */ 861 if (fs->fs_avgfilesize <= 0) 862 fs->fs_avgfilesize = AVFILESIZ; 863 if (fs->fs_avgfpdir <= 0) 864 fs->fs_avgfpdir = AFPDIR; 865 if (bigcgs) { 866 fs->fs_save_cgsize = fs->fs_cgsize; 867 fs->fs_cgsize = fs->fs_bsize; 868 } 869} 870 871/* 872 * Unwinding superblock updates for old filesystems. 873 * See ffs_oldfscompat_read above for details. 874 * 875 * XXX - Parts get retired eventually. 876 * Unfortunately new bits get added. 877 */ 878static void 879ffs_oldfscompat_write(fs, ump) 880 struct fs *fs; 881 struct ufsmount *ump; 882{ 883 884 /* 885 * Copy back UFS2 updated fields that UFS1 inspects. 886 */ 887 if (fs->fs_magic == FS_UFS1_MAGIC) { 888 fs->fs_old_time = fs->fs_time; 889 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 890 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 891 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 892 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 893 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 894 } 895 if (bigcgs) { 896 fs->fs_cgsize = fs->fs_save_cgsize; 897 fs->fs_save_cgsize = 0; 898 } 899} 900 901/* 902 * unmount system call 903 */ 904static int 905ffs_unmount(mp, mntflags, td) 906 struct mount *mp; 907 int mntflags; 908 struct thread *td; 909{ 910 struct ufsmount *ump = VFSTOUFS(mp); 911 struct fs *fs; 912 int error, flags; 913 914 flags = 0; 915 if (mntflags & MNT_FORCE) { 916 flags |= FORCECLOSE; 917 } 918#ifdef UFS_EXTATTR 919 if ((error = ufs_extattr_stop(mp, td))) { 920 if (error != EOPNOTSUPP) 921 printf("ffs_unmount: ufs_extattr_stop returned %d\n", 922 error); 923 } else { 924 ufs_extattr_uepm_destroy(&ump->um_extattr); 925 } 926#endif 927 if (mp->mnt_flag & MNT_SOFTDEP) { 928 if ((error = softdep_flushfiles(mp, flags, td)) != 0) 929 return (error); 930 } else { 931 if ((error = ffs_flushfiles(mp, flags, td)) != 0) 932 return (error); 933 } 934 fs = ump->um_fs; 935 UFS_LOCK(ump); 936 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 937 printf("%s: unmount pending error: blocks %jd files %d\n", 938 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 939 fs->fs_pendinginodes); 940 fs->fs_pendingblocks = 0; 941 fs->fs_pendinginodes = 0; 942 } 943 UFS_UNLOCK(ump); 944 if (fs->fs_ronly == 0) { 945 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 946 error = ffs_sbupdate(ump, MNT_WAIT); 947 if (error) { 948 fs->fs_clean = 0; 949 return (error); 950 } 951 } 952 DROP_GIANT(); 953 g_topology_lock(); 954 g_vfs_close(ump->um_cp, td); 955 g_topology_unlock(); 956 PICKUP_GIANT(); 957 vrele(ump->um_devvp); 958 mtx_destroy(UFS_MTX(ump)); 959 free(fs->fs_csp, M_UFSMNT); 960 free(fs, M_UFSMNT); 961 free(ump, M_UFSMNT); 962 mp->mnt_data = (qaddr_t)0; 963 mp->mnt_flag &= ~MNT_LOCAL; 964 return (error); 965} 966 967/* 968 * Flush out all the files in a filesystem. 969 */ 970int 971ffs_flushfiles(mp, flags, td) 972 struct mount *mp; 973 int flags; 974 struct thread *td; 975{ 976 struct ufsmount *ump; 977 int error; 978 979 ump = VFSTOUFS(mp); 980#ifdef QUOTA 981 if (mp->mnt_flag & MNT_QUOTA) { 982 int i; 983 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 984 if (error) 985 return (error); 986 for (i = 0; i < MAXQUOTAS; i++) { 987 if (ump->um_quotas[i] == NULLVP) 988 continue; 989 quotaoff(td, mp, i); 990 } 991 /* 992 * Here we fall through to vflush again to ensure 993 * that we have gotten rid of all the system vnodes. 994 */ 995 } 996#endif 997 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 998 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 999 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1000 return (error); 1001 ffs_snapshot_unmount(mp); 1002 /* 1003 * Here we fall through to vflush again to ensure 1004 * that we have gotten rid of all the system vnodes. 1005 */ 1006 } 1007 /* 1008 * Flush all the files. 1009 */ 1010 if ((error = vflush(mp, 0, flags, td)) != 0) 1011 return (error); 1012 /* 1013 * Flush filesystem metadata. 1014 */ 1015 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td); 1016 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1017 VOP_UNLOCK(ump->um_devvp, 0, td); 1018 return (error); 1019} 1020 1021/* 1022 * Get filesystem statistics. 1023 */ 1024static int 1025ffs_statfs(mp, sbp, td) 1026 struct mount *mp; 1027 struct statfs *sbp; 1028 struct thread *td; 1029{ 1030 struct ufsmount *ump; 1031 struct fs *fs; 1032 1033 ump = VFSTOUFS(mp); 1034 fs = ump->um_fs; 1035 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1036 panic("ffs_statfs"); 1037 sbp->f_version = STATFS_VERSION; 1038 sbp->f_bsize = fs->fs_fsize; 1039 sbp->f_iosize = fs->fs_bsize; 1040 sbp->f_blocks = fs->fs_dsize; 1041 UFS_LOCK(ump); 1042 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1043 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1044 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1045 dbtofsb(fs, fs->fs_pendingblocks); 1046 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1047 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1048 UFS_UNLOCK(ump); 1049 sbp->f_namemax = NAME_MAX; 1050 return (0); 1051} 1052 1053/* 1054 * Go through the disk queues to initiate sandbagged IO; 1055 * go through the inodes to write those that have been modified; 1056 * initiate the writing of the super block if it has been modified. 1057 * 1058 * Note: we are always called with the filesystem marked `MPBUSY'. 1059 */ 1060static int 1061ffs_sync(mp, waitfor, td) 1062 struct mount *mp; 1063 int waitfor; 1064 struct thread *td; 1065{ 1066 struct vnode *nvp, *vp, *devvp; 1067 struct inode *ip; 1068 struct ufsmount *ump = VFSTOUFS(mp); 1069 struct fs *fs; 1070 int error, count, wait, lockreq, allerror = 0; 1071 struct bufobj *bo; 1072 1073 fs = ump->um_fs; 1074 if (fs->fs_fmod != 0 && fs->fs_ronly != 0) { /* XXX */ 1075 printf("fs = %s\n", fs->fs_fsmnt); 1076 panic("ffs_sync: rofs mod"); 1077 } 1078 /* 1079 * Write back each (modified) inode. 1080 */ 1081 wait = 0; 1082 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1083 if (waitfor == MNT_WAIT) { 1084 wait = 1; 1085 lockreq = LK_EXCLUSIVE; 1086 } 1087 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1088 MNT_ILOCK(mp); 1089loop: 1090 MNT_VNODE_FOREACH(vp, mp, nvp) { 1091 /* 1092 * Depend on the mntvnode_slock to keep things stable enough 1093 * for a quick test. Since there might be hundreds of 1094 * thousands of vnodes, we cannot afford even a subroutine 1095 * call unless there's a good chance that we have work to do. 1096 */ 1097 VI_LOCK(vp); 1098 if (vp->v_iflag & VI_DOOMED) { 1099 VI_UNLOCK(vp); 1100 continue; 1101 } 1102 ip = VTOI(vp); 1103 if (vp->v_type == VNON || ((ip->i_flag & 1104 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1105 vp->v_bufobj.bo_dirty.bv_cnt == 0)) { 1106 VI_UNLOCK(vp); 1107 continue; 1108 } 1109 MNT_IUNLOCK(mp); 1110 if ((error = vget(vp, lockreq, td)) != 0) { 1111 MNT_ILOCK(mp); 1112 if (error == ENOENT || error == ENOLCK) 1113 goto loop; 1114 continue; 1115 } 1116 if ((error = ffs_syncvnode(vp, waitfor)) != 0) 1117 allerror = error; 1118 vput(vp); 1119 MNT_ILOCK(mp); 1120 } 1121 MNT_IUNLOCK(mp); 1122 /* 1123 * Force stale filesystem control information to be flushed. 1124 */ 1125 if (waitfor == MNT_WAIT) { 1126 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1127 allerror = error; 1128 /* Flushed work items may create new vnodes to clean */ 1129 if (allerror == 0 && count) { 1130 MNT_ILOCK(mp); 1131 goto loop; 1132 } 1133 } 1134#ifdef QUOTA 1135 qsync(mp); 1136#endif 1137 devvp = ump->um_devvp; 1138 VI_LOCK(devvp); 1139 bo = &devvp->v_bufobj; 1140 if (waitfor != MNT_LAZY && 1141 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { 1142 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td); 1143 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0) 1144 allerror = error; 1145 VOP_UNLOCK(devvp, 0, td); 1146 if (allerror == 0 && waitfor == MNT_WAIT) { 1147 MNT_ILOCK(mp); 1148 goto loop; 1149 } 1150 } else 1151 VI_UNLOCK(devvp); 1152 /* 1153 * Write back modified superblock. 1154 */ 1155 if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0) 1156 allerror = error; 1157 return (allerror); 1158} 1159 1160int 1161ffs_vget(mp, ino, flags, vpp) 1162 struct mount *mp; 1163 ino_t ino; 1164 int flags; 1165 struct vnode **vpp; 1166{ 1167 struct fs *fs; 1168 struct inode *ip; 1169 struct ufsmount *ump; 1170 struct buf *bp; 1171 struct vnode *vp; 1172 struct cdev *dev; 1173 int error; 1174 1175 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1176 if (error || *vpp != NULL) 1177 return (error); 1178 1179 /* 1180 * We must promote to an exclusive lock for vnode creation. This 1181 * can happen if lookup is passed LOCKSHARED. 1182 */ 1183 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1184 flags &= ~LK_TYPE_MASK; 1185 flags |= LK_EXCLUSIVE; 1186 } 1187 1188 /* 1189 * We do not lock vnode creation as it is believed to be too 1190 * expensive for such rare case as simultaneous creation of vnode 1191 * for same ino by different processes. We just allow them to race 1192 * and check later to decide who wins. Let the race begin! 1193 */ 1194 1195 ump = VFSTOUFS(mp); 1196 dev = ump->um_dev; 1197 fs = ump->um_fs; 1198 1199 /* 1200 * If this MALLOC() is performed after the getnewvnode() 1201 * it might block, leaving a vnode with a NULL v_data to be 1202 * found by ffs_sync() if a sync happens to fire right then, 1203 * which will cause a panic because ffs_sync() blindly 1204 * dereferences vp->v_data (as well it should). 1205 */ 1206 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1207 1208 /* Allocate a new vnode/inode. */ 1209 if (fs->fs_magic == FS_UFS1_MAGIC) 1210 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp); 1211 else 1212 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp); 1213 if (error) { 1214 *vpp = NULL; 1215 uma_zfree(uma_inode, ip); 1216 return (error); 1217 } 1218 /* 1219 * FFS supports recursive and shared locking. 1220 */ 1221 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 1222 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 1223 vp->v_data = ip; 1224 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1225 ip->i_vnode = vp; 1226 ip->i_ump = ump; 1227 ip->i_fs = fs; 1228 ip->i_dev = dev; 1229 ip->i_number = ino; 1230#ifdef QUOTA 1231 { 1232 int i; 1233 for (i = 0; i < MAXQUOTAS; i++) 1234 ip->i_dquot[i] = NODQUOT; 1235 } 1236#endif 1237 1238 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1239 if (error || *vpp != NULL) 1240 return (error); 1241 1242 /* Read in the disk contents for the inode, copy into the inode. */ 1243 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1244 (int)fs->fs_bsize, NOCRED, &bp); 1245 if (error) { 1246 /* 1247 * The inode does not contain anything useful, so it would 1248 * be misleading to leave it on its hash chain. With mode 1249 * still zero, it will be unlinked and returned to the free 1250 * list by vput(). 1251 */ 1252 brelse(bp); 1253 vput(vp); 1254 *vpp = NULL; 1255 return (error); 1256 } 1257 if (ip->i_ump->um_fstype == UFS1) 1258 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1259 else 1260 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1261 ffs_load_inode(bp, ip, fs, ino); 1262 if (DOINGSOFTDEP(vp)) 1263 softdep_load_inodeblock(ip); 1264 else 1265 ip->i_effnlink = ip->i_nlink; 1266 bqrelse(bp); 1267 1268 /* 1269 * Initialize the vnode from the inode, check for aliases. 1270 * Note that the underlying vnode may have changed. 1271 */ 1272 if (ip->i_ump->um_fstype == UFS1) 1273 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1274 else 1275 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1276 if (error) { 1277 vput(vp); 1278 *vpp = NULL; 1279 return (error); 1280 } 1281 1282 /* 1283 * Finish inode initialization. 1284 */ 1285 1286 /* 1287 * Set up a generation number for this inode if it does not 1288 * already have one. This should only happen on old filesystems. 1289 */ 1290 if (ip->i_gen == 0) { 1291 ip->i_gen = arc4random() / 2 + 1; 1292 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1293 ip->i_flag |= IN_MODIFIED; 1294 DIP_SET(ip, i_gen, ip->i_gen); 1295 } 1296 } 1297 /* 1298 * Ensure that uid and gid are correct. This is a temporary 1299 * fix until fsck has been changed to do the update. 1300 */ 1301 if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */ 1302 fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */ 1303 ip->i_uid = ip->i_din1->di_ouid; /* XXX */ 1304 ip->i_gid = ip->i_din1->di_ogid; /* XXX */ 1305 } /* XXX */ 1306 1307#ifdef MAC 1308 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1309 /* 1310 * If this vnode is already allocated, and we're running 1311 * multi-label, attempt to perform a label association 1312 * from the extended attributes on the inode. 1313 */ 1314 error = mac_associate_vnode_extattr(mp, vp); 1315 if (error) { 1316 /* ufs_inactive will release ip->i_devvp ref. */ 1317 vput(vp); 1318 *vpp = NULL; 1319 return (error); 1320 } 1321 } 1322#endif 1323 1324 *vpp = vp; 1325 return (0); 1326} 1327 1328/* 1329 * File handle to vnode 1330 * 1331 * Have to be really careful about stale file handles: 1332 * - check that the inode number is valid 1333 * - call ffs_vget() to get the locked inode 1334 * - check for an unallocated inode (i_mode == 0) 1335 * - check that the given client host has export rights and return 1336 * those rights via. exflagsp and credanonp 1337 */ 1338static int 1339ffs_fhtovp(mp, fhp, vpp) 1340 struct mount *mp; 1341 struct fid *fhp; 1342 struct vnode **vpp; 1343{ 1344 struct ufid *ufhp; 1345 struct fs *fs; 1346 1347 ufhp = (struct ufid *)fhp; 1348 fs = VFSTOUFS(mp)->um_fs; 1349 if (ufhp->ufid_ino < ROOTINO || 1350 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1351 return (ESTALE); 1352 return (ufs_fhtovp(mp, ufhp, vpp)); 1353} 1354 1355/* 1356 * Vnode pointer to File handle 1357 */ 1358/* ARGSUSED */ 1359static int 1360ffs_vptofh(vp, fhp) 1361 struct vnode *vp; 1362 struct fid *fhp; 1363{ 1364 struct inode *ip; 1365 struct ufid *ufhp; 1366 1367 ip = VTOI(vp); 1368 ufhp = (struct ufid *)fhp; 1369 ufhp->ufid_len = sizeof(struct ufid); 1370 ufhp->ufid_ino = ip->i_number; 1371 ufhp->ufid_gen = ip->i_gen; 1372 return (0); 1373} 1374 1375/* 1376 * Initialize the filesystem. 1377 */ 1378static int 1379ffs_init(vfsp) 1380 struct vfsconf *vfsp; 1381{ 1382 1383 softdep_initialize(); 1384 return (ufs_init(vfsp)); 1385} 1386 1387/* 1388 * Undo the work of ffs_init(). 1389 */ 1390static int 1391ffs_uninit(vfsp) 1392 struct vfsconf *vfsp; 1393{ 1394 int ret; 1395 1396 ret = ufs_uninit(vfsp); 1397 softdep_uninitialize(); 1398 return (ret); 1399} 1400 1401/* 1402 * Write a superblock and associated information back to disk. 1403 */ 1404static int 1405ffs_sbupdate(mp, waitfor) 1406 struct ufsmount *mp; 1407 int waitfor; 1408{ 1409 struct fs *fs = mp->um_fs; 1410 struct buf *sbbp; 1411 struct buf *bp; 1412 int blks; 1413 void *space; 1414 int i, size, error, allerror = 0; 1415 1416 if (fs->fs_ronly == 1 && 1417 (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1418 (MNT_RDONLY | MNT_UPDATE)) 1419 panic("ffs_sbupdate: write read-only filesystem"); 1420 /* 1421 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1422 */ 1423 sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 1424 0, 0, 0); 1425 /* 1426 * First write back the summary information. 1427 */ 1428 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1429 space = fs->fs_csp; 1430 for (i = 0; i < blks; i += fs->fs_frag) { 1431 size = fs->fs_bsize; 1432 if (i + fs->fs_frag > blks) 1433 size = (blks - i) * fs->fs_fsize; 1434 bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1435 size, 0, 0, 0); 1436 bcopy(space, bp->b_data, (u_int)size); 1437 space = (char *)space + size; 1438 if (waitfor != MNT_WAIT) 1439 bawrite(bp); 1440 else if ((error = bwrite(bp)) != 0) 1441 allerror = error; 1442 } 1443 /* 1444 * Now write back the superblock itself. If any errors occurred 1445 * up to this point, then fail so that the superblock avoids 1446 * being written out as clean. 1447 */ 1448 if (allerror) { 1449 brelse(sbbp); 1450 return (allerror); 1451 } 1452 bp = sbbp; 1453 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1454 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1455 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1456 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1457 fs->fs_sblockloc = SBLOCK_UFS1; 1458 } 1459 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1460 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1461 printf("%s: correcting fs_sblockloc from %jd to %d\n", 1462 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1463 fs->fs_sblockloc = SBLOCK_UFS2; 1464 } 1465 fs->fs_fmod = 0; 1466 fs->fs_time = time_second; 1467 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1468 ffs_oldfscompat_write((struct fs *)bp->b_data, mp); 1469 if (waitfor != MNT_WAIT) 1470 bawrite(bp); 1471 else if ((error = bwrite(bp)) != 0) 1472 allerror = error; 1473 return (allerror); 1474} 1475 1476static int 1477ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1478 int attrnamespace, const char *attrname, struct thread *td) 1479{ 1480 1481#ifdef UFS_EXTATTR 1482 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1483 attrname, td)); 1484#else 1485 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1486 attrname, td)); 1487#endif 1488} 1489 1490static void 1491ffs_ifree(struct ufsmount *ump, struct inode *ip) 1492{ 1493 1494 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1495 uma_zfree(uma_ufs1, ip->i_din1); 1496 else if (ip->i_din2 != NULL) 1497 uma_zfree(uma_ufs2, ip->i_din2); 1498 uma_zfree(uma_inode, ip); 1499} 1500 1501static int dobkgrdwrite = 1; 1502SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1503 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1504 1505/* 1506 * Complete a background write started from bwrite. 1507 */ 1508static void 1509ffs_backgroundwritedone(struct buf *bp) 1510{ 1511 struct bufobj *bufobj; 1512 struct buf *origbp; 1513 1514 /* 1515 * Find the original buffer that we are writing. 1516 */ 1517 bufobj = bp->b_bufobj; 1518 BO_LOCK(bufobj); 1519 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 1520 panic("backgroundwritedone: lost buffer"); 1521 /* Grab an extra reference to be dropped by the bufdone() below. */ 1522 bufobj_wrefl(bufobj); 1523 BO_UNLOCK(bufobj); 1524 /* 1525 * Process dependencies then return any unfinished ones. 1526 */ 1527 if (LIST_FIRST(&bp->b_dep) != NULL) 1528 buf_complete(bp); 1529#ifdef SOFTUPDATES 1530 if (LIST_FIRST(&bp->b_dep) != NULL) 1531 softdep_move_dependencies(bp, origbp); 1532#endif 1533 /* 1534 * This buffer is marked B_NOCACHE so when it is released 1535 * by biodone it will be tossed. 1536 */ 1537 bp->b_flags |= B_NOCACHE; 1538 bp->b_flags &= ~B_CACHE; 1539 bufdone(bp); 1540 BO_LOCK(bufobj); 1541 /* 1542 * Clear the BV_BKGRDINPROG flag in the original buffer 1543 * and awaken it if it is waiting for the write to complete. 1544 * If BV_BKGRDINPROG is not set in the original buffer it must 1545 * have been released and re-instantiated - which is not legal. 1546 */ 1547 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 1548 ("backgroundwritedone: lost buffer2")); 1549 origbp->b_vflags &= ~BV_BKGRDINPROG; 1550 if (origbp->b_vflags & BV_BKGRDWAIT) { 1551 origbp->b_vflags &= ~BV_BKGRDWAIT; 1552 wakeup(&origbp->b_xflags); 1553 } 1554 BO_UNLOCK(bufobj); 1555} 1556 1557 1558/* 1559 * Write, release buffer on completion. (Done by iodone 1560 * if async). Do not bother writing anything if the buffer 1561 * is invalid. 1562 * 1563 * Note that we set B_CACHE here, indicating that buffer is 1564 * fully valid and thus cacheable. This is true even of NFS 1565 * now so we set it generally. This could be set either here 1566 * or in biodone() since the I/O is synchronous. We put it 1567 * here. 1568 */ 1569static int 1570ffs_bufwrite(struct buf *bp) 1571{ 1572 int oldflags, s; 1573 struct buf *newbp; 1574 1575 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1576 if (bp->b_flags & B_INVAL) { 1577 brelse(bp); 1578 return (0); 1579 } 1580 1581 oldflags = bp->b_flags; 1582 1583 if (BUF_REFCNT(bp) == 0) 1584 panic("bufwrite: buffer is not busy???"); 1585 s = splbio(); 1586 /* 1587 * If a background write is already in progress, delay 1588 * writing this block if it is asynchronous. Otherwise 1589 * wait for the background write to complete. 1590 */ 1591 BO_LOCK(bp->b_bufobj); 1592 if (bp->b_vflags & BV_BKGRDINPROG) { 1593 if (bp->b_flags & B_ASYNC) { 1594 BO_UNLOCK(bp->b_bufobj); 1595 splx(s); 1596 bdwrite(bp); 1597 return (0); 1598 } 1599 bp->b_vflags |= BV_BKGRDWAIT; 1600 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0); 1601 if (bp->b_vflags & BV_BKGRDINPROG) 1602 panic("bufwrite: still writing"); 1603 } 1604 BO_UNLOCK(bp->b_bufobj); 1605 1606 /* Mark the buffer clean */ 1607 bundirty(bp); 1608 1609 /* 1610 * If this buffer is marked for background writing and we 1611 * do not have to wait for it, make a copy and write the 1612 * copy so as to leave this buffer ready for further use. 1613 * 1614 * This optimization eats a lot of memory. If we have a page 1615 * or buffer shortfall we can't do it. 1616 */ 1617 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 1618 (bp->b_flags & B_ASYNC) && 1619 !vm_page_count_severe() && 1620 !buf_dirty_count_severe()) { 1621 KASSERT(bp->b_iodone == NULL, 1622 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 1623 1624 /* get a new block */ 1625 newbp = geteblk(bp->b_bufsize); 1626 1627 /* 1628 * set it to be identical to the old block. We have to 1629 * set b_lblkno and BKGRDMARKER before calling bgetvp() 1630 * to avoid confusing the splay tree and gbincore(). 1631 */ 1632 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 1633 newbp->b_lblkno = bp->b_lblkno; 1634 newbp->b_xflags |= BX_BKGRDMARKER; 1635 BO_LOCK(bp->b_bufobj); 1636 bp->b_vflags |= BV_BKGRDINPROG; 1637 bgetvp(bp->b_vp, newbp); 1638 BO_UNLOCK(bp->b_bufobj); 1639 newbp->b_bufobj = &bp->b_vp->v_bufobj; 1640 newbp->b_blkno = bp->b_blkno; 1641 newbp->b_offset = bp->b_offset; 1642 newbp->b_iodone = ffs_backgroundwritedone; 1643 newbp->b_flags |= B_ASYNC; 1644 newbp->b_flags &= ~B_INVAL; 1645 1646#ifdef SOFTUPDATES 1647 /* move over the dependencies */ 1648 if (LIST_FIRST(&bp->b_dep) != NULL) 1649 softdep_move_dependencies(bp, newbp); 1650#endif 1651 1652 /* 1653 * Initiate write on the copy, release the original to 1654 * the B_LOCKED queue so that it cannot go away until 1655 * the background write completes. If not locked it could go 1656 * away and then be reconstituted while it was being written. 1657 * If the reconstituted buffer were written, we could end up 1658 * with two background copies being written at the same time. 1659 */ 1660 bqrelse(bp); 1661 bp = newbp; 1662 } 1663 1664 /* Let the normal bufwrite do the rest for us */ 1665 bufwrite(bp); 1666 1667 return (0); 1668} 1669 1670 1671static void 1672ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 1673{ 1674 struct vnode *vp; 1675 int error; 1676 1677 vp = bo->__bo_vnode; 1678 if (bp->b_iocmd == BIO_WRITE) { 1679#ifdef SOFTUPDATES 1680 if (LIST_FIRST(&bp->b_dep) != NULL) 1681 buf_start(bp); 1682#endif 1683 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 1684 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 1685 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1686 panic("ffs_geom_strategy: bad I/O"); 1687 bp->b_flags &= ~B_VALIDSUSPWRT; 1688 if ((vp->v_vflag & VV_COPYONWRITE) && 1689 vp->v_rdev->si_snapdata != NULL && 1690 (error = (ffs_copyonwrite)(vp, bp)) != 0 && 1691 error != EOPNOTSUPP) { 1692 bp->b_error = error; 1693 bp->b_ioflags |= BIO_ERROR; 1694 bufdone(bp); 1695 return; 1696 } 1697 } 1698 g_vfs_strategy(bo, bp); 1699} 1700