ffs_vfsops.c revision 306175
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: stable/10/sys/ufs/ffs/ffs_vfsops.c 306175 2016-09-22 10:42:40Z kib $"); 34 35#include "opt_quota.h" 36#include "opt_ufs.h" 37#include "opt_ffs.h" 38#include "opt_ddb.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/namei.h> 43#include <sys/priv.h> 44#include <sys/proc.h> 45#include <sys/taskqueue.h> 46#include <sys/kernel.h> 47#include <sys/vnode.h> 48#include <sys/mount.h> 49#include <sys/bio.h> 50#include <sys/buf.h> 51#include <sys/conf.h> 52#include <sys/fcntl.h> 53#include <sys/ioccom.h> 54#include <sys/malloc.h> 55#include <sys/mutex.h> 56#include <sys/rwlock.h> 57 58#include <security/mac/mac_framework.h> 59 60#include <ufs/ufs/extattr.h> 61#include <ufs/ufs/gjournal.h> 62#include <ufs/ufs/quota.h> 63#include <ufs/ufs/ufsmount.h> 64#include <ufs/ufs/inode.h> 65#include <ufs/ufs/ufs_extern.h> 66 67#include <ufs/ffs/fs.h> 68#include <ufs/ffs/ffs_extern.h> 69 70#include <vm/vm.h> 71#include <vm/uma.h> 72#include <vm/vm_page.h> 73 74#include <geom/geom.h> 75#include <geom/geom_vfs.h> 76 77#include <ddb/ddb.h> 78 79static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; 80 81static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); 82static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, 83 ufs2_daddr_t); 84static void ffs_ifree(struct ufsmount *ump, struct inode *ip); 85static int ffs_sync_lazy(struct mount *mp); 86 87static vfs_init_t ffs_init; 88static vfs_uninit_t ffs_uninit; 89static vfs_extattrctl_t ffs_extattrctl; 90static vfs_cmount_t ffs_cmount; 91static vfs_unmount_t ffs_unmount; 92static vfs_mount_t ffs_mount; 93static vfs_statfs_t ffs_statfs; 94static vfs_fhtovp_t ffs_fhtovp; 95static vfs_sync_t ffs_sync; 96 97static struct vfsops ufs_vfsops = { 98 .vfs_extattrctl = ffs_extattrctl, 99 .vfs_fhtovp = ffs_fhtovp, 100 .vfs_init = ffs_init, 101 .vfs_mount = ffs_mount, 102 .vfs_cmount = ffs_cmount, 103 .vfs_quotactl = ufs_quotactl, 104 .vfs_root = ufs_root, 105 .vfs_statfs = ffs_statfs, 106 .vfs_sync = ffs_sync, 107 .vfs_uninit = ffs_uninit, 108 .vfs_unmount = ffs_unmount, 109 .vfs_vget = ffs_vget, 110 .vfs_susp_clean = process_deferred_inactive, 111}; 112 113VFS_SET(ufs_vfsops, ufs, 0); 114MODULE_VERSION(ufs, 1); 115 116static b_strategy_t ffs_geom_strategy; 117static b_write_t ffs_bufwrite; 118 119static struct buf_ops ffs_ops = { 120 .bop_name = "FFS", 121 .bop_write = ffs_bufwrite, 122 .bop_strategy = ffs_geom_strategy, 123 .bop_sync = bufsync, 124#ifdef NO_FFS_SNAPSHOT 125 .bop_bdflush = bufbdflush, 126#else 127 .bop_bdflush = ffs_bdflush, 128#endif 129}; 130 131/* 132 * Note that userquota and groupquota options are not currently used 133 * by UFS/FFS code and generally mount(8) does not pass those options 134 * from userland, but they can be passed by loader(8) via 135 * vfs.root.mountfrom.options. 136 */ 137static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", 138 "noclusterw", "noexec", "export", "force", "from", "groupquota", 139 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir", 140 "nosymfollow", "sync", "union", "userquota", NULL }; 141 142static int 143ffs_mount(struct mount *mp) 144{ 145 struct vnode *devvp; 146 struct thread *td; 147 struct ufsmount *ump = NULL; 148 struct fs *fs; 149 pid_t fsckpid = 0; 150 int error, flags; 151 uint64_t mntorflags; 152 accmode_t accmode; 153 struct nameidata ndp; 154 char *fspec; 155 156 td = curthread; 157 if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) 158 return (EINVAL); 159 if (uma_inode == NULL) { 160 uma_inode = uma_zcreate("FFS inode", 161 sizeof(struct inode), NULL, NULL, NULL, NULL, 162 UMA_ALIGN_PTR, 0); 163 uma_ufs1 = uma_zcreate("FFS1 dinode", 164 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, 165 UMA_ALIGN_PTR, 0); 166 uma_ufs2 = uma_zcreate("FFS2 dinode", 167 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, 168 UMA_ALIGN_PTR, 0); 169 } 170 171 vfs_deleteopt(mp->mnt_optnew, "groupquota"); 172 vfs_deleteopt(mp->mnt_optnew, "userquota"); 173 174 fspec = vfs_getopts(mp->mnt_optnew, "from", &error); 175 if (error) 176 return (error); 177 178 mntorflags = 0; 179 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) 180 mntorflags |= MNT_ACLS; 181 182 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { 183 mntorflags |= MNT_SNAPSHOT; 184 /* 185 * Once we have set the MNT_SNAPSHOT flag, do not 186 * persist "snapshot" in the options list. 187 */ 188 vfs_deleteopt(mp->mnt_optnew, "snapshot"); 189 vfs_deleteopt(mp->mnt_opt, "snapshot"); 190 } 191 192 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 && 193 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { 194 /* 195 * Once we have set the restricted PID, do not 196 * persist "fsckpid" in the options list. 197 */ 198 vfs_deleteopt(mp->mnt_optnew, "fsckpid"); 199 vfs_deleteopt(mp->mnt_opt, "fsckpid"); 200 if (mp->mnt_flag & MNT_UPDATE) { 201 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && 202 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 203 vfs_mount_error(mp, 204 "Checker enable: Must be read-only"); 205 return (EINVAL); 206 } 207 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { 208 vfs_mount_error(mp, 209 "Checker enable: Must be read-only"); 210 return (EINVAL); 211 } 212 /* Set to -1 if we are done */ 213 if (fsckpid == 0) 214 fsckpid = -1; 215 } 216 217 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { 218 if (mntorflags & MNT_ACLS) { 219 vfs_mount_error(mp, 220 "\"acls\" and \"nfsv4acls\" options " 221 "are mutually exclusive"); 222 return (EINVAL); 223 } 224 mntorflags |= MNT_NFS4ACLS; 225 } 226 227 MNT_ILOCK(mp); 228 mp->mnt_flag |= mntorflags; 229 MNT_IUNLOCK(mp); 230 /* 231 * If updating, check whether changing from read-only to 232 * read/write; if there is no device name, that's all we do. 233 */ 234 if (mp->mnt_flag & MNT_UPDATE) { 235 ump = VFSTOUFS(mp); 236 fs = ump->um_fs; 237 devvp = ump->um_devvp; 238 if (fsckpid == -1 && ump->um_fsckpid > 0) { 239 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || 240 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) 241 return (error); 242 DROP_GIANT(); 243 g_topology_lock(); 244 /* 245 * Return to normal read-only mode. 246 */ 247 error = g_access(ump->um_cp, 0, -1, 0); 248 g_topology_unlock(); 249 PICKUP_GIANT(); 250 ump->um_fsckpid = 0; 251 } 252 if (fs->fs_ronly == 0 && 253 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 254 /* 255 * Flush any dirty data and suspend filesystem. 256 */ 257 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 258 return (error); 259 error = vfs_write_suspend_umnt(mp); 260 if (error != 0) 261 return (error); 262 /* 263 * Check for and optionally get rid of files open 264 * for writing. 265 */ 266 flags = WRITECLOSE; 267 if (mp->mnt_flag & MNT_FORCE) 268 flags |= FORCECLOSE; 269 if (MOUNTEDSOFTDEP(mp)) { 270 error = softdep_flushfiles(mp, flags, td); 271 } else { 272 error = ffs_flushfiles(mp, flags, td); 273 } 274 if (error) { 275 vfs_write_resume(mp, 0); 276 return (error); 277 } 278 if (fs->fs_pendingblocks != 0 || 279 fs->fs_pendinginodes != 0) { 280 printf("WARNING: %s Update error: blocks %jd " 281 "files %d\n", fs->fs_fsmnt, 282 (intmax_t)fs->fs_pendingblocks, 283 fs->fs_pendinginodes); 284 fs->fs_pendingblocks = 0; 285 fs->fs_pendinginodes = 0; 286 } 287 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) 288 fs->fs_clean = 1; 289 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 290 fs->fs_ronly = 0; 291 fs->fs_clean = 0; 292 vfs_write_resume(mp, 0); 293 return (error); 294 } 295 if (MOUNTEDSOFTDEP(mp)) 296 softdep_unmount(mp); 297 DROP_GIANT(); 298 g_topology_lock(); 299 /* 300 * Drop our write and exclusive access. 301 */ 302 g_access(ump->um_cp, 0, -1, -1); 303 g_topology_unlock(); 304 PICKUP_GIANT(); 305 fs->fs_ronly = 1; 306 MNT_ILOCK(mp); 307 mp->mnt_flag |= MNT_RDONLY; 308 MNT_IUNLOCK(mp); 309 /* 310 * Allow the writers to note that filesystem 311 * is ro now. 312 */ 313 vfs_write_resume(mp, 0); 314 } 315 if ((mp->mnt_flag & MNT_RELOAD) && 316 (error = ffs_reload(mp, td, 0)) != 0) 317 return (error); 318 if (fs->fs_ronly && 319 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { 320 /* 321 * If we are running a checker, do not allow upgrade. 322 */ 323 if (ump->um_fsckpid > 0) { 324 vfs_mount_error(mp, 325 "Active checker, cannot upgrade to write"); 326 return (EINVAL); 327 } 328 /* 329 * If upgrade to read-write by non-root, then verify 330 * that user has necessary permissions on the device. 331 */ 332 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 333 error = VOP_ACCESS(devvp, VREAD | VWRITE, 334 td->td_ucred, td); 335 if (error) 336 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 337 if (error) { 338 VOP_UNLOCK(devvp, 0); 339 return (error); 340 } 341 VOP_UNLOCK(devvp, 0); 342 fs->fs_flags &= ~FS_UNCLEAN; 343 if (fs->fs_clean == 0) { 344 fs->fs_flags |= FS_UNCLEAN; 345 if ((mp->mnt_flag & MNT_FORCE) || 346 ((fs->fs_flags & 347 (FS_SUJ | FS_NEEDSFSCK)) == 0 && 348 (fs->fs_flags & FS_DOSOFTDEP))) { 349 printf("WARNING: %s was not properly " 350 "dismounted\n", fs->fs_fsmnt); 351 } else { 352 vfs_mount_error(mp, 353 "R/W mount of %s denied. %s.%s", 354 fs->fs_fsmnt, 355 "Filesystem is not clean - run fsck", 356 (fs->fs_flags & FS_SUJ) == 0 ? "" : 357 " Forced mount will invalidate" 358 " journal contents"); 359 return (EPERM); 360 } 361 } 362 DROP_GIANT(); 363 g_topology_lock(); 364 /* 365 * Request exclusive write access. 366 */ 367 error = g_access(ump->um_cp, 0, 1, 1); 368 g_topology_unlock(); 369 PICKUP_GIANT(); 370 if (error) 371 return (error); 372 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) 373 return (error); 374 fs->fs_ronly = 0; 375 MNT_ILOCK(mp); 376 mp->mnt_flag &= ~MNT_RDONLY; 377 MNT_IUNLOCK(mp); 378 fs->fs_mtime = time_second; 379 /* check to see if we need to start softdep */ 380 if ((fs->fs_flags & FS_DOSOFTDEP) && 381 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ 382 vn_finished_write(mp); 383 return (error); 384 } 385 fs->fs_clean = 0; 386 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { 387 vn_finished_write(mp); 388 return (error); 389 } 390 if (fs->fs_snapinum[0] != 0) 391 ffs_snapshot_mount(mp); 392 vn_finished_write(mp); 393 } 394 /* 395 * Soft updates is incompatible with "async", 396 * so if we are doing softupdates stop the user 397 * from setting the async flag in an update. 398 * Softdep_mount() clears it in an initial mount 399 * or ro->rw remount. 400 */ 401 if (MOUNTEDSOFTDEP(mp)) { 402 /* XXX: Reset too late ? */ 403 MNT_ILOCK(mp); 404 mp->mnt_flag &= ~MNT_ASYNC; 405 MNT_IUNLOCK(mp); 406 } 407 /* 408 * Keep MNT_ACLS flag if it is stored in superblock. 409 */ 410 if ((fs->fs_flags & FS_ACLS) != 0) { 411 /* XXX: Set too late ? */ 412 MNT_ILOCK(mp); 413 mp->mnt_flag |= MNT_ACLS; 414 MNT_IUNLOCK(mp); 415 } 416 417 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 418 /* XXX: Set too late ? */ 419 MNT_ILOCK(mp); 420 mp->mnt_flag |= MNT_NFS4ACLS; 421 MNT_IUNLOCK(mp); 422 } 423 /* 424 * If this is a request from fsck to clean up the filesystem, 425 * then allow the specified pid to proceed. 426 */ 427 if (fsckpid > 0) { 428 if (ump->um_fsckpid != 0) { 429 vfs_mount_error(mp, 430 "Active checker already running on %s", 431 fs->fs_fsmnt); 432 return (EINVAL); 433 } 434 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 435 ("soft updates enabled on read-only file system")); 436 DROP_GIANT(); 437 g_topology_lock(); 438 /* 439 * Request write access. 440 */ 441 error = g_access(ump->um_cp, 0, 1, 0); 442 g_topology_unlock(); 443 PICKUP_GIANT(); 444 if (error) { 445 vfs_mount_error(mp, 446 "Checker activation failed on %s", 447 fs->fs_fsmnt); 448 return (error); 449 } 450 ump->um_fsckpid = fsckpid; 451 if (fs->fs_snapinum[0] != 0) 452 ffs_snapshot_mount(mp); 453 fs->fs_mtime = time_second; 454 fs->fs_fmod = 1; 455 fs->fs_clean = 0; 456 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 457 } 458 459 /* 460 * If this is a snapshot request, take the snapshot. 461 */ 462 if (mp->mnt_flag & MNT_SNAPSHOT) 463 return (ffs_snapshot(mp, fspec)); 464 } 465 466 /* 467 * Not an update, or updating the name: look up the name 468 * and verify that it refers to a sensible disk device. 469 */ 470 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); 471 if ((error = namei(&ndp)) != 0) 472 return (error); 473 NDFREE(&ndp, NDF_ONLY_PNBUF); 474 devvp = ndp.ni_vp; 475 if (!vn_isdisk(devvp, &error)) { 476 vput(devvp); 477 return (error); 478 } 479 480 /* 481 * If mount by non-root, then verify that user has necessary 482 * permissions on the device. 483 */ 484 accmode = VREAD; 485 if ((mp->mnt_flag & MNT_RDONLY) == 0) 486 accmode |= VWRITE; 487 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); 488 if (error) 489 error = priv_check(td, PRIV_VFS_MOUNT_PERM); 490 if (error) { 491 vput(devvp); 492 return (error); 493 } 494 495 if (mp->mnt_flag & MNT_UPDATE) { 496 /* 497 * Update only 498 * 499 * If it's not the same vnode, or at least the same device 500 * then it's not correct. 501 */ 502 503 if (devvp->v_rdev != ump->um_devvp->v_rdev) 504 error = EINVAL; /* needs translation */ 505 vput(devvp); 506 if (error) 507 return (error); 508 } else { 509 /* 510 * New mount 511 * 512 * We need the name for the mount point (also used for 513 * "last mounted on") copied in. If an error occurs, 514 * the mount point is discarded by the upper level code. 515 * Note that vfs_mount_alloc() populates f_mntonname for us. 516 */ 517 if ((error = ffs_mountfs(devvp, mp, td)) != 0) { 518 vrele(devvp); 519 return (error); 520 } 521 if (fsckpid > 0) { 522 KASSERT(MOUNTEDSOFTDEP(mp) == 0, 523 ("soft updates enabled on read-only file system")); 524 ump = VFSTOUFS(mp); 525 fs = ump->um_fs; 526 DROP_GIANT(); 527 g_topology_lock(); 528 /* 529 * Request write access. 530 */ 531 error = g_access(ump->um_cp, 0, 1, 0); 532 g_topology_unlock(); 533 PICKUP_GIANT(); 534 if (error) { 535 printf("WARNING: %s: Checker activation " 536 "failed\n", fs->fs_fsmnt); 537 } else { 538 ump->um_fsckpid = fsckpid; 539 if (fs->fs_snapinum[0] != 0) 540 ffs_snapshot_mount(mp); 541 fs->fs_mtime = time_second; 542 fs->fs_clean = 0; 543 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 544 } 545 } 546 } 547 vfs_mountedfrom(mp, fspec); 548 return (0); 549} 550 551/* 552 * Compatibility with old mount system call. 553 */ 554 555static int 556ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) 557{ 558 struct ufs_args args; 559 struct export_args exp; 560 int error; 561 562 if (data == NULL) 563 return (EINVAL); 564 error = copyin(data, &args, sizeof args); 565 if (error) 566 return (error); 567 vfs_oexport_conv(&args.export, &exp); 568 569 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); 570 ma = mount_arg(ma, "export", &exp, sizeof(exp)); 571 error = kernel_mount(ma, flags); 572 573 return (error); 574} 575 576/* 577 * Reload all incore data for a filesystem (used after running fsck on 578 * the root filesystem and finding things to fix). If the 'force' flag 579 * is 0, the filesystem must be mounted read-only. 580 * 581 * Things to do to update the mount: 582 * 1) invalidate all cached meta-data. 583 * 2) re-read superblock from disk. 584 * 3) re-read summary information from disk. 585 * 4) invalidate all inactive vnodes. 586 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary 587 * writers, if requested. 588 * 6) invalidate all cached file data. 589 * 7) re-read inode data for all active vnodes. 590 */ 591int 592ffs_reload(struct mount *mp, struct thread *td, int flags) 593{ 594 struct vnode *vp, *mvp, *devvp; 595 struct inode *ip; 596 void *space; 597 struct buf *bp; 598 struct fs *fs, *newfs; 599 struct ufsmount *ump; 600 ufs2_daddr_t sblockloc; 601 int i, blks, size, error; 602 int32_t *lp; 603 604 ump = VFSTOUFS(mp); 605 606 MNT_ILOCK(mp); 607 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { 608 MNT_IUNLOCK(mp); 609 return (EINVAL); 610 } 611 MNT_IUNLOCK(mp); 612 613 /* 614 * Step 1: invalidate all cached meta-data. 615 */ 616 devvp = VFSTOUFS(mp)->um_devvp; 617 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 618 if (vinvalbuf(devvp, 0, 0, 0) != 0) 619 panic("ffs_reload: dirty1"); 620 VOP_UNLOCK(devvp, 0); 621 622 /* 623 * Step 2: re-read superblock from disk. 624 */ 625 fs = VFSTOUFS(mp)->um_fs; 626 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, 627 NOCRED, &bp)) != 0) 628 return (error); 629 newfs = (struct fs *)bp->b_data; 630 if ((newfs->fs_magic != FS_UFS1_MAGIC && 631 newfs->fs_magic != FS_UFS2_MAGIC) || 632 newfs->fs_bsize > MAXBSIZE || 633 newfs->fs_bsize < sizeof(struct fs)) { 634 brelse(bp); 635 return (EIO); /* XXX needs translation */ 636 } 637 /* 638 * Copy pointer fields back into superblock before copying in XXX 639 * new superblock. These should really be in the ufsmount. XXX 640 * Note that important parameters (eg fs_ncg) are unchanged. 641 */ 642 newfs->fs_csp = fs->fs_csp; 643 newfs->fs_maxcluster = fs->fs_maxcluster; 644 newfs->fs_contigdirs = fs->fs_contigdirs; 645 newfs->fs_active = fs->fs_active; 646 newfs->fs_ronly = fs->fs_ronly; 647 sblockloc = fs->fs_sblockloc; 648 bcopy(newfs, fs, (u_int)fs->fs_sbsize); 649 brelse(bp); 650 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 651 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); 652 UFS_LOCK(ump); 653 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 654 printf("WARNING: %s: reload pending error: blocks %jd " 655 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 656 fs->fs_pendinginodes); 657 fs->fs_pendingblocks = 0; 658 fs->fs_pendinginodes = 0; 659 } 660 UFS_UNLOCK(ump); 661 662 /* 663 * Step 3: re-read summary information from disk. 664 */ 665 size = fs->fs_cssize; 666 blks = howmany(size, fs->fs_fsize); 667 if (fs->fs_contigsumsize > 0) 668 size += fs->fs_ncg * sizeof(int32_t); 669 size += fs->fs_ncg * sizeof(u_int8_t); 670 free(fs->fs_csp, M_UFSMNT); 671 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 672 fs->fs_csp = space; 673 for (i = 0; i < blks; i += fs->fs_frag) { 674 size = fs->fs_bsize; 675 if (i + fs->fs_frag > blks) 676 size = (blks - i) * fs->fs_fsize; 677 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 678 NOCRED, &bp); 679 if (error) 680 return (error); 681 bcopy(bp->b_data, space, (u_int)size); 682 space = (char *)space + size; 683 brelse(bp); 684 } 685 /* 686 * We no longer know anything about clusters per cylinder group. 687 */ 688 if (fs->fs_contigsumsize > 0) { 689 fs->fs_maxcluster = lp = space; 690 for (i = 0; i < fs->fs_ncg; i++) 691 *lp++ = fs->fs_contigsumsize; 692 space = lp; 693 } 694 size = fs->fs_ncg * sizeof(u_int8_t); 695 fs->fs_contigdirs = (u_int8_t *)space; 696 bzero(fs->fs_contigdirs, size); 697 if ((flags & FFSR_UNSUSPEND) != 0) { 698 MNT_ILOCK(mp); 699 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 700 wakeup(&mp->mnt_flag); 701 MNT_IUNLOCK(mp); 702 } 703 704loop: 705 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 706 /* 707 * Skip syncer vnode. 708 */ 709 if (vp->v_type == VNON) { 710 VI_UNLOCK(vp); 711 continue; 712 } 713 /* 714 * Step 4: invalidate all cached file data. 715 */ 716 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 717 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 718 goto loop; 719 } 720 if (vinvalbuf(vp, 0, 0, 0)) 721 panic("ffs_reload: dirty2"); 722 /* 723 * Step 5: re-read inode data for all active vnodes. 724 */ 725 ip = VTOI(vp); 726 error = 727 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), 728 (int)fs->fs_bsize, NOCRED, &bp); 729 if (error) { 730 VOP_UNLOCK(vp, 0); 731 vrele(vp); 732 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 733 return (error); 734 } 735 ffs_load_inode(bp, ip, fs, ip->i_number); 736 ip->i_effnlink = ip->i_nlink; 737 brelse(bp); 738 VOP_UNLOCK(vp, 0); 739 vrele(vp); 740 } 741 return (0); 742} 743 744/* 745 * Possible superblock locations ordered from most to least likely. 746 */ 747static int sblock_try[] = SBLOCKSEARCH; 748 749/* 750 * Common code for mount and mountroot 751 */ 752static int 753ffs_mountfs(devvp, mp, td) 754 struct vnode *devvp; 755 struct mount *mp; 756 struct thread *td; 757{ 758 struct ufsmount *ump; 759 struct buf *bp; 760 struct fs *fs; 761 struct cdev *dev; 762 void *space; 763 ufs2_daddr_t sblockloc; 764 int error, i, blks, size, ronly; 765 int32_t *lp; 766 struct ucred *cred; 767 struct g_consumer *cp; 768 struct mount *nmp; 769 770 bp = NULL; 771 ump = NULL; 772 cred = td ? td->td_ucred : NOCRED; 773 ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 774 775 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); 776 dev = devvp->v_rdev; 777 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, 778 (uintptr_t)mp) == 0) { 779 VOP_UNLOCK(devvp, 0); 780 return (EBUSY); 781 } 782 DROP_GIANT(); 783 g_topology_lock(); 784 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); 785 g_topology_unlock(); 786 PICKUP_GIANT(); 787 if (error != 0) { 788 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 789 VOP_UNLOCK(devvp, 0); 790 return (error); 791 } 792 dev_ref(dev); 793 devvp->v_bufobj.bo_ops = &ffs_ops; 794 VOP_UNLOCK(devvp, 0); 795 if (dev->si_iosize_max != 0) 796 mp->mnt_iosize_max = dev->si_iosize_max; 797 if (mp->mnt_iosize_max > MAXPHYS) 798 mp->mnt_iosize_max = MAXPHYS; 799 800 fs = NULL; 801 sblockloc = 0; 802 /* 803 * Try reading the superblock in each of its possible locations. 804 */ 805 for (i = 0; sblock_try[i] != -1; i++) { 806 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { 807 error = EINVAL; 808 vfs_mount_error(mp, 809 "Invalid sectorsize %d for superblock size %d", 810 cp->provider->sectorsize, SBLOCKSIZE); 811 goto out; 812 } 813 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE, 814 cred, &bp)) != 0) 815 goto out; 816 fs = (struct fs *)bp->b_data; 817 sblockloc = sblock_try[i]; 818 if ((fs->fs_magic == FS_UFS1_MAGIC || 819 (fs->fs_magic == FS_UFS2_MAGIC && 820 (fs->fs_sblockloc == sblockloc || 821 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) && 822 fs->fs_bsize <= MAXBSIZE && 823 fs->fs_bsize >= sizeof(struct fs)) 824 break; 825 brelse(bp); 826 bp = NULL; 827 } 828 if (sblock_try[i] == -1) { 829 error = EINVAL; /* XXX needs translation */ 830 goto out; 831 } 832 fs->fs_fmod = 0; 833 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */ 834 fs->fs_flags &= ~FS_UNCLEAN; 835 if (fs->fs_clean == 0) { 836 fs->fs_flags |= FS_UNCLEAN; 837 if (ronly || (mp->mnt_flag & MNT_FORCE) || 838 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && 839 (fs->fs_flags & FS_DOSOFTDEP))) { 840 printf("WARNING: %s was not properly dismounted\n", 841 fs->fs_fsmnt); 842 } else { 843 vfs_mount_error(mp, "R/W mount of %s denied. %s%s", 844 fs->fs_fsmnt, "Filesystem is not clean - run fsck.", 845 (fs->fs_flags & FS_SUJ) == 0 ? "" : 846 " Forced mount will invalidate journal contents"); 847 error = EPERM; 848 goto out; 849 } 850 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && 851 (mp->mnt_flag & MNT_FORCE)) { 852 printf("WARNING: %s: lost blocks %jd files %d\n", 853 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 854 fs->fs_pendinginodes); 855 fs->fs_pendingblocks = 0; 856 fs->fs_pendinginodes = 0; 857 } 858 } 859 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 860 printf("WARNING: %s: mount pending error: blocks %jd " 861 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 862 fs->fs_pendinginodes); 863 fs->fs_pendingblocks = 0; 864 fs->fs_pendinginodes = 0; 865 } 866 if ((fs->fs_flags & FS_GJOURNAL) != 0) { 867#ifdef UFS_GJOURNAL 868 /* 869 * Get journal provider name. 870 */ 871 size = 1024; 872 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK); 873 if (g_io_getattr("GJOURNAL::provider", cp, &size, 874 mp->mnt_gjprovider) == 0) { 875 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size, 876 M_UFSMNT, M_WAITOK); 877 MNT_ILOCK(mp); 878 mp->mnt_flag |= MNT_GJOURNAL; 879 MNT_IUNLOCK(mp); 880 } else { 881 printf("WARNING: %s: GJOURNAL flag on fs " 882 "but no gjournal provider below\n", 883 mp->mnt_stat.f_mntonname); 884 free(mp->mnt_gjprovider, M_UFSMNT); 885 mp->mnt_gjprovider = NULL; 886 } 887#else 888 printf("WARNING: %s: GJOURNAL flag on fs but no " 889 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); 890#endif 891 } else { 892 mp->mnt_gjprovider = NULL; 893 } 894 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); 895 ump->um_cp = cp; 896 ump->um_bo = &devvp->v_bufobj; 897 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK); 898 if (fs->fs_magic == FS_UFS1_MAGIC) { 899 ump->um_fstype = UFS1; 900 ump->um_balloc = ffs_balloc_ufs1; 901 } else { 902 ump->um_fstype = UFS2; 903 ump->um_balloc = ffs_balloc_ufs2; 904 } 905 ump->um_blkatoff = ffs_blkatoff; 906 ump->um_truncate = ffs_truncate; 907 ump->um_update = ffs_update; 908 ump->um_valloc = ffs_valloc; 909 ump->um_vfree = ffs_vfree; 910 ump->um_ifree = ffs_ifree; 911 ump->um_rdonly = ffs_rdonly; 912 ump->um_snapgone = ffs_snapgone; 913 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); 914 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize); 915 if (fs->fs_sbsize < SBLOCKSIZE) 916 bp->b_flags |= B_INVAL | B_NOCACHE; 917 brelse(bp); 918 bp = NULL; 919 fs = ump->um_fs; 920 ffs_oldfscompat_read(fs, ump, sblockloc); 921 fs->fs_ronly = ronly; 922 size = fs->fs_cssize; 923 blks = howmany(size, fs->fs_fsize); 924 if (fs->fs_contigsumsize > 0) 925 size += fs->fs_ncg * sizeof(int32_t); 926 size += fs->fs_ncg * sizeof(u_int8_t); 927 space = malloc((u_long)size, M_UFSMNT, M_WAITOK); 928 fs->fs_csp = space; 929 for (i = 0; i < blks; i += fs->fs_frag) { 930 size = fs->fs_bsize; 931 if (i + fs->fs_frag > blks) 932 size = (blks - i) * fs->fs_fsize; 933 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, 934 cred, &bp)) != 0) { 935 free(fs->fs_csp, M_UFSMNT); 936 goto out; 937 } 938 bcopy(bp->b_data, space, (u_int)size); 939 space = (char *)space + size; 940 brelse(bp); 941 bp = NULL; 942 } 943 if (fs->fs_contigsumsize > 0) { 944 fs->fs_maxcluster = lp = space; 945 for (i = 0; i < fs->fs_ncg; i++) 946 *lp++ = fs->fs_contigsumsize; 947 space = lp; 948 } 949 size = fs->fs_ncg * sizeof(u_int8_t); 950 fs->fs_contigdirs = (u_int8_t *)space; 951 bzero(fs->fs_contigdirs, size); 952 fs->fs_active = NULL; 953 mp->mnt_data = ump; 954 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; 955 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; 956 nmp = NULL; 957 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || 958 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { 959 if (nmp) 960 vfs_rel(nmp); 961 vfs_getnewfsid(mp); 962 } 963 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; 964 MNT_ILOCK(mp); 965 mp->mnt_flag |= MNT_LOCAL; 966 MNT_IUNLOCK(mp); 967 if ((fs->fs_flags & FS_MULTILABEL) != 0) { 968#ifdef MAC 969 MNT_ILOCK(mp); 970 mp->mnt_flag |= MNT_MULTILABEL; 971 MNT_IUNLOCK(mp); 972#else 973 printf("WARNING: %s: multilabel flag on fs but " 974 "no MAC support\n", mp->mnt_stat.f_mntonname); 975#endif 976 } 977 if ((fs->fs_flags & FS_ACLS) != 0) { 978#ifdef UFS_ACL 979 MNT_ILOCK(mp); 980 981 if (mp->mnt_flag & MNT_NFS4ACLS) 982 printf("WARNING: %s: ACLs flag on fs conflicts with " 983 "\"nfsv4acls\" mount option; option ignored\n", 984 mp->mnt_stat.f_mntonname); 985 mp->mnt_flag &= ~MNT_NFS4ACLS; 986 mp->mnt_flag |= MNT_ACLS; 987 988 MNT_IUNLOCK(mp); 989#else 990 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", 991 mp->mnt_stat.f_mntonname); 992#endif 993 } 994 if ((fs->fs_flags & FS_NFS4ACLS) != 0) { 995#ifdef UFS_ACL 996 MNT_ILOCK(mp); 997 998 if (mp->mnt_flag & MNT_ACLS) 999 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " 1000 "with \"acls\" mount option; option ignored\n", 1001 mp->mnt_stat.f_mntonname); 1002 mp->mnt_flag &= ~MNT_ACLS; 1003 mp->mnt_flag |= MNT_NFS4ACLS; 1004 1005 MNT_IUNLOCK(mp); 1006#else 1007 printf("WARNING: %s: NFSv4 ACLs flag on fs but no " 1008 "ACLs support\n", mp->mnt_stat.f_mntonname); 1009#endif 1010 } 1011 if ((fs->fs_flags & FS_TRIM) != 0) { 1012 size = sizeof(int); 1013 if (g_io_getattr("GEOM::candelete", cp, &size, 1014 &ump->um_candelete) == 0) { 1015 if (!ump->um_candelete) 1016 printf("WARNING: %s: TRIM flag on fs but disk " 1017 "does not support TRIM\n", 1018 mp->mnt_stat.f_mntonname); 1019 } else { 1020 printf("WARNING: %s: TRIM flag on fs but disk does " 1021 "not confirm that it supports TRIM\n", 1022 mp->mnt_stat.f_mntonname); 1023 ump->um_candelete = 0; 1024 } 1025 if (ump->um_candelete) { 1026 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, 1027 taskqueue_thread_enqueue, &ump->um_trim_tq); 1028 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, 1029 "%s trim", mp->mnt_stat.f_mntonname); 1030 } 1031 } 1032 1033 ump->um_mountp = mp; 1034 ump->um_dev = dev; 1035 ump->um_devvp = devvp; 1036 ump->um_nindir = fs->fs_nindir; 1037 ump->um_bptrtodb = fs->fs_fsbtodb; 1038 ump->um_seqinc = fs->fs_frag; 1039 for (i = 0; i < MAXQUOTAS; i++) 1040 ump->um_quotas[i] = NULLVP; 1041#ifdef UFS_EXTATTR 1042 ufs_extattr_uepm_init(&ump->um_extattr); 1043#endif 1044 /* 1045 * Set FS local "last mounted on" information (NULL pad) 1046 */ 1047 bzero(fs->fs_fsmnt, MAXMNTLEN); 1048 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); 1049 mp->mnt_stat.f_iosize = fs->fs_bsize; 1050 1051 if (mp->mnt_flag & MNT_ROOTFS) { 1052 /* 1053 * Root mount; update timestamp in mount structure. 1054 * this will be used by the common root mount code 1055 * to update the system clock. 1056 */ 1057 mp->mnt_time = fs->fs_time; 1058 } 1059 1060 if (ronly == 0) { 1061 fs->fs_mtime = time_second; 1062 if ((fs->fs_flags & FS_DOSOFTDEP) && 1063 (error = softdep_mount(devvp, mp, fs, cred)) != 0) { 1064 free(fs->fs_csp, M_UFSMNT); 1065 ffs_flushfiles(mp, FORCECLOSE, td); 1066 goto out; 1067 } 1068 if (fs->fs_snapinum[0] != 0) 1069 ffs_snapshot_mount(mp); 1070 fs->fs_fmod = 1; 1071 fs->fs_clean = 0; 1072 (void) ffs_sbupdate(ump, MNT_WAIT, 0); 1073 } 1074 /* 1075 * Initialize filesystem state information in mount struct. 1076 */ 1077 MNT_ILOCK(mp); 1078 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | 1079 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; 1080 MNT_IUNLOCK(mp); 1081#ifdef UFS_EXTATTR 1082#ifdef UFS_EXTATTR_AUTOSTART 1083 /* 1084 * 1085 * Auto-starting does the following: 1086 * - check for /.attribute in the fs, and extattr_start if so 1087 * - for each file in .attribute, enable that file with 1088 * an attribute of the same name. 1089 * Not clear how to report errors -- probably eat them. 1090 * This would all happen while the filesystem was busy/not 1091 * available, so would effectively be "atomic". 1092 */ 1093 (void) ufs_extattr_autostart(mp, td); 1094#endif /* !UFS_EXTATTR_AUTOSTART */ 1095#endif /* !UFS_EXTATTR */ 1096 return (0); 1097out: 1098 if (bp) 1099 brelse(bp); 1100 if (cp != NULL) { 1101 DROP_GIANT(); 1102 g_topology_lock(); 1103 g_vfs_close(cp); 1104 g_topology_unlock(); 1105 PICKUP_GIANT(); 1106 } 1107 if (ump) { 1108 mtx_destroy(UFS_MTX(ump)); 1109 if (mp->mnt_gjprovider != NULL) { 1110 free(mp->mnt_gjprovider, M_UFSMNT); 1111 mp->mnt_gjprovider = NULL; 1112 } 1113 free(ump->um_fs, M_UFSMNT); 1114 free(ump, M_UFSMNT); 1115 mp->mnt_data = NULL; 1116 } 1117 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); 1118 dev_rel(dev); 1119 return (error); 1120} 1121 1122#include <sys/sysctl.h> 1123static int bigcgs = 0; 1124SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); 1125 1126/* 1127 * Sanity checks for loading old filesystem superblocks. 1128 * See ffs_oldfscompat_write below for unwound actions. 1129 * 1130 * XXX - Parts get retired eventually. 1131 * Unfortunately new bits get added. 1132 */ 1133static void 1134ffs_oldfscompat_read(fs, ump, sblockloc) 1135 struct fs *fs; 1136 struct ufsmount *ump; 1137 ufs2_daddr_t sblockloc; 1138{ 1139 off_t maxfilesize; 1140 1141 /* 1142 * If not yet done, update fs_flags location and value of fs_sblockloc. 1143 */ 1144 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { 1145 fs->fs_flags = fs->fs_old_flags; 1146 fs->fs_old_flags |= FS_FLAGS_UPDATED; 1147 fs->fs_sblockloc = sblockloc; 1148 } 1149 /* 1150 * If not yet done, update UFS1 superblock with new wider fields. 1151 */ 1152 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { 1153 fs->fs_maxbsize = fs->fs_bsize; 1154 fs->fs_time = fs->fs_old_time; 1155 fs->fs_size = fs->fs_old_size; 1156 fs->fs_dsize = fs->fs_old_dsize; 1157 fs->fs_csaddr = fs->fs_old_csaddr; 1158 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; 1159 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; 1160 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; 1161 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; 1162 } 1163 if (fs->fs_magic == FS_UFS1_MAGIC && 1164 fs->fs_old_inodefmt < FS_44INODEFMT) { 1165 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; 1166 fs->fs_qbmask = ~fs->fs_bmask; 1167 fs->fs_qfmask = ~fs->fs_fmask; 1168 } 1169 if (fs->fs_magic == FS_UFS1_MAGIC) { 1170 ump->um_savedmaxfilesize = fs->fs_maxfilesize; 1171 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; 1172 if (fs->fs_maxfilesize > maxfilesize) 1173 fs->fs_maxfilesize = maxfilesize; 1174 } 1175 /* Compatibility for old filesystems */ 1176 if (fs->fs_avgfilesize <= 0) 1177 fs->fs_avgfilesize = AVFILESIZ; 1178 if (fs->fs_avgfpdir <= 0) 1179 fs->fs_avgfpdir = AFPDIR; 1180 if (bigcgs) { 1181 fs->fs_save_cgsize = fs->fs_cgsize; 1182 fs->fs_cgsize = fs->fs_bsize; 1183 } 1184} 1185 1186/* 1187 * Unwinding superblock updates for old filesystems. 1188 * See ffs_oldfscompat_read above for details. 1189 * 1190 * XXX - Parts get retired eventually. 1191 * Unfortunately new bits get added. 1192 */ 1193void 1194ffs_oldfscompat_write(fs, ump) 1195 struct fs *fs; 1196 struct ufsmount *ump; 1197{ 1198 1199 /* 1200 * Copy back UFS2 updated fields that UFS1 inspects. 1201 */ 1202 if (fs->fs_magic == FS_UFS1_MAGIC) { 1203 fs->fs_old_time = fs->fs_time; 1204 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; 1205 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; 1206 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; 1207 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; 1208 fs->fs_maxfilesize = ump->um_savedmaxfilesize; 1209 } 1210 if (bigcgs) { 1211 fs->fs_cgsize = fs->fs_save_cgsize; 1212 fs->fs_save_cgsize = 0; 1213 } 1214} 1215 1216/* 1217 * unmount system call 1218 */ 1219static int 1220ffs_unmount(mp, mntflags) 1221 struct mount *mp; 1222 int mntflags; 1223{ 1224 struct thread *td; 1225 struct ufsmount *ump = VFSTOUFS(mp); 1226 struct fs *fs; 1227 int error, flags, susp; 1228#ifdef UFS_EXTATTR 1229 int e_restart; 1230#endif 1231 1232 flags = 0; 1233 td = curthread; 1234 fs = ump->um_fs; 1235 susp = 0; 1236 if (mntflags & MNT_FORCE) { 1237 flags |= FORCECLOSE; 1238 susp = fs->fs_ronly == 0; 1239 } 1240#ifdef UFS_EXTATTR 1241 if ((error = ufs_extattr_stop(mp, td))) { 1242 if (error != EOPNOTSUPP) 1243 printf("WARNING: unmount %s: ufs_extattr_stop " 1244 "returned errno %d\n", mp->mnt_stat.f_mntonname, 1245 error); 1246 e_restart = 0; 1247 } else { 1248 ufs_extattr_uepm_destroy(&ump->um_extattr); 1249 e_restart = 1; 1250 } 1251#endif 1252 if (susp) { 1253 error = vfs_write_suspend_umnt(mp); 1254 if (error != 0) 1255 goto fail1; 1256 } 1257 if (MOUNTEDSOFTDEP(mp)) 1258 error = softdep_flushfiles(mp, flags, td); 1259 else 1260 error = ffs_flushfiles(mp, flags, td); 1261 if (error != 0 && error != ENXIO) 1262 goto fail; 1263 1264 UFS_LOCK(ump); 1265 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { 1266 printf("WARNING: unmount %s: pending error: blocks %jd " 1267 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, 1268 fs->fs_pendinginodes); 1269 fs->fs_pendingblocks = 0; 1270 fs->fs_pendinginodes = 0; 1271 } 1272 UFS_UNLOCK(ump); 1273 if (MOUNTEDSOFTDEP(mp)) 1274 softdep_unmount(mp); 1275 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { 1276 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; 1277 error = ffs_sbupdate(ump, MNT_WAIT, 0); 1278 if (error && error != ENXIO) { 1279 fs->fs_clean = 0; 1280 goto fail; 1281 } 1282 } 1283 if (susp) 1284 vfs_write_resume(mp, VR_START_WRITE); 1285 if (ump->um_trim_tq != NULL) { 1286 while (ump->um_trim_inflight != 0) 1287 pause("ufsutr", hz); 1288 taskqueue_drain_all(ump->um_trim_tq); 1289 taskqueue_free(ump->um_trim_tq); 1290 } 1291 DROP_GIANT(); 1292 g_topology_lock(); 1293 if (ump->um_fsckpid > 0) { 1294 /* 1295 * Return to normal read-only mode. 1296 */ 1297 error = g_access(ump->um_cp, 0, -1, 0); 1298 ump->um_fsckpid = 0; 1299 } 1300 g_vfs_close(ump->um_cp); 1301 g_topology_unlock(); 1302 PICKUP_GIANT(); 1303 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); 1304 vrele(ump->um_devvp); 1305 dev_rel(ump->um_dev); 1306 mtx_destroy(UFS_MTX(ump)); 1307 if (mp->mnt_gjprovider != NULL) { 1308 free(mp->mnt_gjprovider, M_UFSMNT); 1309 mp->mnt_gjprovider = NULL; 1310 } 1311 free(fs->fs_csp, M_UFSMNT); 1312 free(fs, M_UFSMNT); 1313 free(ump, M_UFSMNT); 1314 mp->mnt_data = NULL; 1315 MNT_ILOCK(mp); 1316 mp->mnt_flag &= ~MNT_LOCAL; 1317 MNT_IUNLOCK(mp); 1318 return (error); 1319 1320fail: 1321 if (susp) 1322 vfs_write_resume(mp, VR_START_WRITE); 1323fail1: 1324#ifdef UFS_EXTATTR 1325 if (e_restart) { 1326 ufs_extattr_uepm_init(&ump->um_extattr); 1327#ifdef UFS_EXTATTR_AUTOSTART 1328 (void) ufs_extattr_autostart(mp, td); 1329#endif 1330 } 1331#endif 1332 1333 return (error); 1334} 1335 1336/* 1337 * Flush out all the files in a filesystem. 1338 */ 1339int 1340ffs_flushfiles(mp, flags, td) 1341 struct mount *mp; 1342 int flags; 1343 struct thread *td; 1344{ 1345 struct ufsmount *ump; 1346 int qerror, error; 1347 1348 ump = VFSTOUFS(mp); 1349 qerror = 0; 1350#ifdef QUOTA 1351 if (mp->mnt_flag & MNT_QUOTA) { 1352 int i; 1353 error = vflush(mp, 0, SKIPSYSTEM|flags, td); 1354 if (error) 1355 return (error); 1356 for (i = 0; i < MAXQUOTAS; i++) { 1357 error = quotaoff(td, mp, i); 1358 if (error != 0) { 1359 if ((flags & EARLYFLUSH) == 0) 1360 return (error); 1361 else 1362 qerror = error; 1363 } 1364 } 1365 1366 /* 1367 * Here we fall through to vflush again to ensure that 1368 * we have gotten rid of all the system vnodes, unless 1369 * quotas must not be closed. 1370 */ 1371 } 1372#endif 1373 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); 1374 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { 1375 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) 1376 return (error); 1377 ffs_snapshot_unmount(mp); 1378 flags |= FORCECLOSE; 1379 /* 1380 * Here we fall through to vflush again to ensure 1381 * that we have gotten rid of all the system vnodes. 1382 */ 1383 } 1384 1385 /* 1386 * Do not close system files if quotas were not closed, to be 1387 * able to sync the remaining dquots. The freeblks softupdate 1388 * workitems might hold a reference on a dquot, preventing 1389 * quotaoff() from completing. Next round of 1390 * softdep_flushworklist() iteration should process the 1391 * blockers, allowing the next run of quotaoff() to finally 1392 * flush held dquots. 1393 * 1394 * Otherwise, flush all the files. 1395 */ 1396 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) 1397 return (error); 1398 1399 /* 1400 * Flush filesystem metadata. 1401 */ 1402 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); 1403 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); 1404 VOP_UNLOCK(ump->um_devvp, 0); 1405 return (error); 1406} 1407 1408/* 1409 * Get filesystem statistics. 1410 */ 1411static int 1412ffs_statfs(mp, sbp) 1413 struct mount *mp; 1414 struct statfs *sbp; 1415{ 1416 struct ufsmount *ump; 1417 struct fs *fs; 1418 1419 ump = VFSTOUFS(mp); 1420 fs = ump->um_fs; 1421 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) 1422 panic("ffs_statfs"); 1423 sbp->f_version = STATFS_VERSION; 1424 sbp->f_bsize = fs->fs_fsize; 1425 sbp->f_iosize = fs->fs_bsize; 1426 sbp->f_blocks = fs->fs_dsize; 1427 UFS_LOCK(ump); 1428 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + 1429 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); 1430 sbp->f_bavail = freespace(fs, fs->fs_minfree) + 1431 dbtofsb(fs, fs->fs_pendingblocks); 1432 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO; 1433 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; 1434 UFS_UNLOCK(ump); 1435 sbp->f_namemax = NAME_MAX; 1436 return (0); 1437} 1438 1439static bool 1440sync_doupdate(struct inode *ip) 1441{ 1442 1443 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | 1444 IN_UPDATE)) != 0); 1445} 1446 1447/* 1448 * For a lazy sync, we only care about access times, quotas and the 1449 * superblock. Other filesystem changes are already converted to 1450 * cylinder group blocks or inode blocks updates and are written to 1451 * disk by syncer. 1452 */ 1453static int 1454ffs_sync_lazy(mp) 1455 struct mount *mp; 1456{ 1457 struct vnode *mvp, *vp; 1458 struct inode *ip; 1459 struct thread *td; 1460 int allerror, error; 1461 1462 allerror = 0; 1463 td = curthread; 1464 if ((mp->mnt_flag & MNT_NOATIME) != 0) 1465 goto qupdate; 1466 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { 1467 if (vp->v_type == VNON) { 1468 VI_UNLOCK(vp); 1469 continue; 1470 } 1471 ip = VTOI(vp); 1472 1473 /* 1474 * The IN_ACCESS flag is converted to IN_MODIFIED by 1475 * ufs_close() and ufs_getattr() by the calls to 1476 * ufs_itimes_locked(), without subsequent UFS_UPDATE(). 1477 * Test also all the other timestamp flags too, to pick up 1478 * any other cases that could be missed. 1479 */ 1480 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { 1481 VI_UNLOCK(vp); 1482 continue; 1483 } 1484 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, 1485 td)) != 0) 1486 continue; 1487 if (sync_doupdate(ip)) 1488 error = ffs_update(vp, 0); 1489 if (error != 0) 1490 allerror = error; 1491 vput(vp); 1492 } 1493 1494qupdate: 1495#ifdef QUOTA 1496 qsync(mp); 1497#endif 1498 1499 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && 1500 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) 1501 allerror = error; 1502 return (allerror); 1503} 1504 1505/* 1506 * Go through the disk queues to initiate sandbagged IO; 1507 * go through the inodes to write those that have been modified; 1508 * initiate the writing of the super block if it has been modified. 1509 * 1510 * Note: we are always called with the filesystem marked busy using 1511 * vfs_busy(). 1512 */ 1513static int 1514ffs_sync(mp, waitfor) 1515 struct mount *mp; 1516 int waitfor; 1517{ 1518 struct vnode *mvp, *vp, *devvp; 1519 struct thread *td; 1520 struct inode *ip; 1521 struct ufsmount *ump = VFSTOUFS(mp); 1522 struct fs *fs; 1523 int error, count, lockreq, allerror = 0; 1524 int suspend; 1525 int suspended; 1526 int secondary_writes; 1527 int secondary_accwrites; 1528 int softdep_deps; 1529 int softdep_accdeps; 1530 struct bufobj *bo; 1531 1532 suspend = 0; 1533 suspended = 0; 1534 td = curthread; 1535 fs = ump->um_fs; 1536 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) 1537 panic("%s: ffs_sync: modification on read-only filesystem", 1538 fs->fs_fsmnt); 1539 if (waitfor == MNT_LAZY) { 1540 if (!rebooting) 1541 return (ffs_sync_lazy(mp)); 1542 waitfor = MNT_NOWAIT; 1543 } 1544 1545 /* 1546 * Write back each (modified) inode. 1547 */ 1548 lockreq = LK_EXCLUSIVE | LK_NOWAIT; 1549 if (waitfor == MNT_SUSPEND) { 1550 suspend = 1; 1551 waitfor = MNT_WAIT; 1552 } 1553 if (waitfor == MNT_WAIT) 1554 lockreq = LK_EXCLUSIVE; 1555 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; 1556loop: 1557 /* Grab snapshot of secondary write counts */ 1558 MNT_ILOCK(mp); 1559 secondary_writes = mp->mnt_secondary_writes; 1560 secondary_accwrites = mp->mnt_secondary_accwrites; 1561 MNT_IUNLOCK(mp); 1562 1563 /* Grab snapshot of softdep dependency counts */ 1564 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); 1565 1566 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1567 /* 1568 * Depend on the vnode interlock to keep things stable enough 1569 * for a quick test. Since there might be hundreds of 1570 * thousands of vnodes, we cannot afford even a subroutine 1571 * call unless there's a good chance that we have work to do. 1572 */ 1573 if (vp->v_type == VNON) { 1574 VI_UNLOCK(vp); 1575 continue; 1576 } 1577 ip = VTOI(vp); 1578 if ((ip->i_flag & 1579 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && 1580 vp->v_bufobj.bo_dirty.bv_cnt == 0) { 1581 VI_UNLOCK(vp); 1582 continue; 1583 } 1584 if ((error = vget(vp, lockreq, td)) != 0) { 1585 if (error == ENOENT || error == ENOLCK) { 1586 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1587 goto loop; 1588 } 1589 continue; 1590 } 1591 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0) 1592 allerror = error; 1593 vput(vp); 1594 } 1595 /* 1596 * Force stale filesystem control information to be flushed. 1597 */ 1598 if (waitfor == MNT_WAIT || rebooting) { 1599 if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) 1600 allerror = error; 1601 /* Flushed work items may create new vnodes to clean */ 1602 if (allerror == 0 && count) 1603 goto loop; 1604 } 1605#ifdef QUOTA 1606 qsync(mp); 1607#endif 1608 1609 devvp = ump->um_devvp; 1610 bo = &devvp->v_bufobj; 1611 BO_LOCK(bo); 1612 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { 1613 BO_UNLOCK(bo); 1614 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); 1615 error = VOP_FSYNC(devvp, waitfor, td); 1616 VOP_UNLOCK(devvp, 0); 1617 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) 1618 error = ffs_sbupdate(ump, waitfor, 0); 1619 if (error != 0) 1620 allerror = error; 1621 if (allerror == 0 && waitfor == MNT_WAIT) 1622 goto loop; 1623 } else if (suspend != 0) { 1624 if (softdep_check_suspend(mp, 1625 devvp, 1626 softdep_deps, 1627 softdep_accdeps, 1628 secondary_writes, 1629 secondary_accwrites) != 0) { 1630 MNT_IUNLOCK(mp); 1631 goto loop; /* More work needed */ 1632 } 1633 mtx_assert(MNT_MTX(mp), MA_OWNED); 1634 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 1635 MNT_IUNLOCK(mp); 1636 suspended = 1; 1637 } else 1638 BO_UNLOCK(bo); 1639 /* 1640 * Write back modified superblock. 1641 */ 1642 if (fs->fs_fmod != 0 && 1643 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) 1644 allerror = error; 1645 return (allerror); 1646} 1647 1648int 1649ffs_vget(mp, ino, flags, vpp) 1650 struct mount *mp; 1651 ino_t ino; 1652 int flags; 1653 struct vnode **vpp; 1654{ 1655 return (ffs_vgetf(mp, ino, flags, vpp, 0)); 1656} 1657 1658int 1659ffs_vgetf(mp, ino, flags, vpp, ffs_flags) 1660 struct mount *mp; 1661 ino_t ino; 1662 int flags; 1663 struct vnode **vpp; 1664 int ffs_flags; 1665{ 1666 struct fs *fs; 1667 struct inode *ip; 1668 struct ufsmount *ump; 1669 struct buf *bp; 1670 struct vnode *vp; 1671 struct cdev *dev; 1672 int error; 1673 1674 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); 1675 if (error || *vpp != NULL) 1676 return (error); 1677 1678 /* 1679 * We must promote to an exclusive lock for vnode creation. This 1680 * can happen if lookup is passed LOCKSHARED. 1681 */ 1682 if ((flags & LK_TYPE_MASK) == LK_SHARED) { 1683 flags &= ~LK_TYPE_MASK; 1684 flags |= LK_EXCLUSIVE; 1685 } 1686 1687 /* 1688 * We do not lock vnode creation as it is believed to be too 1689 * expensive for such rare case as simultaneous creation of vnode 1690 * for same ino by different processes. We just allow them to race 1691 * and check later to decide who wins. Let the race begin! 1692 */ 1693 1694 ump = VFSTOUFS(mp); 1695 dev = ump->um_dev; 1696 fs = ump->um_fs; 1697 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); 1698 1699 /* Allocate a new vnode/inode. */ 1700 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? 1701 &ffs_vnodeops1 : &ffs_vnodeops2, &vp); 1702 if (error) { 1703 *vpp = NULL; 1704 uma_zfree(uma_inode, ip); 1705 return (error); 1706 } 1707 /* 1708 * FFS supports recursive locking. 1709 */ 1710 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 1711 VN_LOCK_AREC(vp); 1712 vp->v_data = ip; 1713 vp->v_bufobj.bo_bsize = fs->fs_bsize; 1714 ip->i_vnode = vp; 1715 ip->i_ump = ump; 1716 ip->i_fs = fs; 1717 ip->i_dev = dev; 1718 ip->i_number = ino; 1719 ip->i_ea_refs = 0; 1720 ip->i_nextclustercg = -1; 1721#ifdef QUOTA 1722 { 1723 int i; 1724 for (i = 0; i < MAXQUOTAS; i++) 1725 ip->i_dquot[i] = NODQUOT; 1726 } 1727#endif 1728 1729 if (ffs_flags & FFSV_FORCEINSMQ) 1730 vp->v_vflag |= VV_FORCEINSMQ; 1731 error = insmntque(vp, mp); 1732 if (error != 0) { 1733 uma_zfree(uma_inode, ip); 1734 *vpp = NULL; 1735 return (error); 1736 } 1737 vp->v_vflag &= ~VV_FORCEINSMQ; 1738 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); 1739 if (error || *vpp != NULL) 1740 return (error); 1741 1742 /* Read in the disk contents for the inode, copy into the inode. */ 1743 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), 1744 (int)fs->fs_bsize, NOCRED, &bp); 1745 if (error) { 1746 /* 1747 * The inode does not contain anything useful, so it would 1748 * be misleading to leave it on its hash chain. With mode 1749 * still zero, it will be unlinked and returned to the free 1750 * list by vput(). 1751 */ 1752 brelse(bp); 1753 vput(vp); 1754 *vpp = NULL; 1755 return (error); 1756 } 1757 if (ip->i_ump->um_fstype == UFS1) 1758 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); 1759 else 1760 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); 1761 ffs_load_inode(bp, ip, fs, ino); 1762 if (DOINGSOFTDEP(vp)) 1763 softdep_load_inodeblock(ip); 1764 else 1765 ip->i_effnlink = ip->i_nlink; 1766 bqrelse(bp); 1767 1768 /* 1769 * Initialize the vnode from the inode, check for aliases. 1770 * Note that the underlying vnode may have changed. 1771 */ 1772 if (ip->i_ump->um_fstype == UFS1) 1773 error = ufs_vinit(mp, &ffs_fifoops1, &vp); 1774 else 1775 error = ufs_vinit(mp, &ffs_fifoops2, &vp); 1776 if (error) { 1777 vput(vp); 1778 *vpp = NULL; 1779 return (error); 1780 } 1781 1782 /* 1783 * Finish inode initialization. 1784 */ 1785 if (vp->v_type != VFIFO) { 1786 /* FFS supports shared locking for all files except fifos. */ 1787 VN_LOCK_ASHARE(vp); 1788 } 1789 1790 /* 1791 * Set up a generation number for this inode if it does not 1792 * already have one. This should only happen on old filesystems. 1793 */ 1794 if (ip->i_gen == 0) { 1795 ip->i_gen = arc4random() / 2 + 1; 1796 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { 1797 ip->i_flag |= IN_MODIFIED; 1798 DIP_SET(ip, i_gen, ip->i_gen); 1799 } 1800 } 1801#ifdef MAC 1802 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { 1803 /* 1804 * If this vnode is already allocated, and we're running 1805 * multi-label, attempt to perform a label association 1806 * from the extended attributes on the inode. 1807 */ 1808 error = mac_vnode_associate_extattr(mp, vp); 1809 if (error) { 1810 /* ufs_inactive will release ip->i_devvp ref. */ 1811 vput(vp); 1812 *vpp = NULL; 1813 return (error); 1814 } 1815 } 1816#endif 1817 1818 *vpp = vp; 1819 return (0); 1820} 1821 1822/* 1823 * File handle to vnode 1824 * 1825 * Have to be really careful about stale file handles: 1826 * - check that the inode number is valid 1827 * - call ffs_vget() to get the locked inode 1828 * - check for an unallocated inode (i_mode == 0) 1829 * - check that the given client host has export rights and return 1830 * those rights via. exflagsp and credanonp 1831 */ 1832static int 1833ffs_fhtovp(mp, fhp, flags, vpp) 1834 struct mount *mp; 1835 struct fid *fhp; 1836 int flags; 1837 struct vnode **vpp; 1838{ 1839 struct ufid *ufhp; 1840 struct fs *fs; 1841 1842 ufhp = (struct ufid *)fhp; 1843 fs = VFSTOUFS(mp)->um_fs; 1844 if (ufhp->ufid_ino < ROOTINO || 1845 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg) 1846 return (ESTALE); 1847 return (ufs_fhtovp(mp, ufhp, flags, vpp)); 1848} 1849 1850/* 1851 * Initialize the filesystem. 1852 */ 1853static int 1854ffs_init(vfsp) 1855 struct vfsconf *vfsp; 1856{ 1857 1858 ffs_susp_initialize(); 1859 softdep_initialize(); 1860 return (ufs_init(vfsp)); 1861} 1862 1863/* 1864 * Undo the work of ffs_init(). 1865 */ 1866static int 1867ffs_uninit(vfsp) 1868 struct vfsconf *vfsp; 1869{ 1870 int ret; 1871 1872 ret = ufs_uninit(vfsp); 1873 softdep_uninitialize(); 1874 ffs_susp_uninitialize(); 1875 return (ret); 1876} 1877 1878/* 1879 * Write a superblock and associated information back to disk. 1880 */ 1881int 1882ffs_sbupdate(ump, waitfor, suspended) 1883 struct ufsmount *ump; 1884 int waitfor; 1885 int suspended; 1886{ 1887 struct fs *fs = ump->um_fs; 1888 struct buf *sbbp; 1889 struct buf *bp; 1890 int blks; 1891 void *space; 1892 int i, size, error, allerror = 0; 1893 1894 if (fs->fs_ronly == 1 && 1895 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != 1896 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) 1897 panic("ffs_sbupdate: write read-only filesystem"); 1898 /* 1899 * We use the superblock's buf to serialize calls to ffs_sbupdate(). 1900 */ 1901 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), 1902 (int)fs->fs_sbsize, 0, 0, 0); 1903 /* 1904 * First write back the summary information. 1905 */ 1906 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1907 space = fs->fs_csp; 1908 for (i = 0; i < blks; i += fs->fs_frag) { 1909 size = fs->fs_bsize; 1910 if (i + fs->fs_frag > blks) 1911 size = (blks - i) * fs->fs_fsize; 1912 bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i), 1913 size, 0, 0, 0); 1914 bcopy(space, bp->b_data, (u_int)size); 1915 space = (char *)space + size; 1916 if (suspended) 1917 bp->b_flags |= B_VALIDSUSPWRT; 1918 if (waitfor != MNT_WAIT) 1919 bawrite(bp); 1920 else if ((error = bwrite(bp)) != 0) 1921 allerror = error; 1922 } 1923 /* 1924 * Now write back the superblock itself. If any errors occurred 1925 * up to this point, then fail so that the superblock avoids 1926 * being written out as clean. 1927 */ 1928 if (allerror) { 1929 brelse(sbbp); 1930 return (allerror); 1931 } 1932 bp = sbbp; 1933 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && 1934 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1935 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 1936 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); 1937 fs->fs_sblockloc = SBLOCK_UFS1; 1938 } 1939 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && 1940 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) { 1941 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", 1942 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); 1943 fs->fs_sblockloc = SBLOCK_UFS2; 1944 } 1945 fs->fs_fmod = 0; 1946 fs->fs_time = time_second; 1947 if (MOUNTEDSOFTDEP(ump->um_mountp)) 1948 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); 1949 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); 1950 ffs_oldfscompat_write((struct fs *)bp->b_data, ump); 1951 if (suspended) 1952 bp->b_flags |= B_VALIDSUSPWRT; 1953 if (waitfor != MNT_WAIT) 1954 bawrite(bp); 1955 else if ((error = bwrite(bp)) != 0) 1956 allerror = error; 1957 return (allerror); 1958} 1959 1960static int 1961ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, 1962 int attrnamespace, const char *attrname) 1963{ 1964 1965#ifdef UFS_EXTATTR 1966 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, 1967 attrname)); 1968#else 1969 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, 1970 attrname)); 1971#endif 1972} 1973 1974static void 1975ffs_ifree(struct ufsmount *ump, struct inode *ip) 1976{ 1977 1978 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) 1979 uma_zfree(uma_ufs1, ip->i_din1); 1980 else if (ip->i_din2 != NULL) 1981 uma_zfree(uma_ufs2, ip->i_din2); 1982 uma_zfree(uma_inode, ip); 1983} 1984 1985static int dobkgrdwrite = 1; 1986SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, 1987 "Do background writes (honoring the BV_BKGRDWRITE flag)?"); 1988 1989/* 1990 * Complete a background write started from bwrite. 1991 */ 1992static void 1993ffs_backgroundwritedone(struct buf *bp) 1994{ 1995 struct bufobj *bufobj; 1996 struct buf *origbp; 1997 1998 /* 1999 * Find the original buffer that we are writing. 2000 */ 2001 bufobj = bp->b_bufobj; 2002 BO_LOCK(bufobj); 2003 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) 2004 panic("backgroundwritedone: lost buffer"); 2005 2006 /* 2007 * We should mark the cylinder group buffer origbp as 2008 * dirty, to not loose the failed write. 2009 */ 2010 if ((bp->b_ioflags & BIO_ERROR) != 0) 2011 origbp->b_vflags |= BV_BKGRDERR; 2012 BO_UNLOCK(bufobj); 2013 /* 2014 * Process dependencies then return any unfinished ones. 2015 */ 2016 pbrelvp(bp); 2017 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) 2018 buf_complete(bp); 2019#ifdef SOFTUPDATES 2020 if (!LIST_EMPTY(&bp->b_dep)) 2021 softdep_move_dependencies(bp, origbp); 2022#endif 2023 /* 2024 * This buffer is marked B_NOCACHE so when it is released 2025 * by biodone it will be tossed. 2026 */ 2027 bp->b_flags |= B_NOCACHE; 2028 bp->b_flags &= ~B_CACHE; 2029 2030 /* 2031 * Prevent brelse() from trying to keep and re-dirtying bp on 2032 * errors. It causes b_bufobj dereference in 2033 * bdirty()/reassignbuf(), and b_bufobj was cleared in 2034 * pbrelvp() above. 2035 */ 2036 if ((bp->b_ioflags & BIO_ERROR) != 0) 2037 bp->b_flags |= B_INVAL; 2038 bufdone(bp); 2039 BO_LOCK(bufobj); 2040 /* 2041 * Clear the BV_BKGRDINPROG flag in the original buffer 2042 * and awaken it if it is waiting for the write to complete. 2043 * If BV_BKGRDINPROG is not set in the original buffer it must 2044 * have been released and re-instantiated - which is not legal. 2045 */ 2046 KASSERT((origbp->b_vflags & BV_BKGRDINPROG), 2047 ("backgroundwritedone: lost buffer2")); 2048 origbp->b_vflags &= ~BV_BKGRDINPROG; 2049 if (origbp->b_vflags & BV_BKGRDWAIT) { 2050 origbp->b_vflags &= ~BV_BKGRDWAIT; 2051 wakeup(&origbp->b_xflags); 2052 } 2053 BO_UNLOCK(bufobj); 2054} 2055 2056 2057/* 2058 * Write, release buffer on completion. (Done by iodone 2059 * if async). Do not bother writing anything if the buffer 2060 * is invalid. 2061 * 2062 * Note that we set B_CACHE here, indicating that buffer is 2063 * fully valid and thus cacheable. This is true even of NFS 2064 * now so we set it generally. This could be set either here 2065 * or in biodone() since the I/O is synchronous. We put it 2066 * here. 2067 */ 2068static int 2069ffs_bufwrite(struct buf *bp) 2070{ 2071 struct buf *newbp; 2072 2073 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 2074 if (bp->b_flags & B_INVAL) { 2075 brelse(bp); 2076 return (0); 2077 } 2078 2079 if (!BUF_ISLOCKED(bp)) 2080 panic("bufwrite: buffer is not busy???"); 2081 /* 2082 * If a background write is already in progress, delay 2083 * writing this block if it is asynchronous. Otherwise 2084 * wait for the background write to complete. 2085 */ 2086 BO_LOCK(bp->b_bufobj); 2087 if (bp->b_vflags & BV_BKGRDINPROG) { 2088 if (bp->b_flags & B_ASYNC) { 2089 BO_UNLOCK(bp->b_bufobj); 2090 bdwrite(bp); 2091 return (0); 2092 } 2093 bp->b_vflags |= BV_BKGRDWAIT; 2094 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, 2095 "bwrbg", 0); 2096 if (bp->b_vflags & BV_BKGRDINPROG) 2097 panic("bufwrite: still writing"); 2098 } 2099 bp->b_vflags &= ~BV_BKGRDERR; 2100 BO_UNLOCK(bp->b_bufobj); 2101 2102 /* 2103 * If this buffer is marked for background writing and we 2104 * do not have to wait for it, make a copy and write the 2105 * copy so as to leave this buffer ready for further use. 2106 * 2107 * This optimization eats a lot of memory. If we have a page 2108 * or buffer shortfall we can't do it. 2109 */ 2110 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && 2111 (bp->b_flags & B_ASYNC) && 2112 !vm_page_count_severe() && 2113 !buf_dirty_count_severe()) { 2114 KASSERT(bp->b_iodone == NULL, 2115 ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); 2116 2117 /* get a new block */ 2118 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); 2119 if (newbp == NULL) 2120 goto normal_write; 2121 2122 KASSERT((bp->b_flags & B_UNMAPPED) == 0, ("Unmapped cg")); 2123 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); 2124 BO_LOCK(bp->b_bufobj); 2125 bp->b_vflags |= BV_BKGRDINPROG; 2126 BO_UNLOCK(bp->b_bufobj); 2127 newbp->b_xflags |= BX_BKGRDMARKER; 2128 newbp->b_lblkno = bp->b_lblkno; 2129 newbp->b_blkno = bp->b_blkno; 2130 newbp->b_offset = bp->b_offset; 2131 newbp->b_iodone = ffs_backgroundwritedone; 2132 newbp->b_flags |= B_ASYNC; 2133 newbp->b_flags &= ~B_INVAL; 2134 pbgetvp(bp->b_vp, newbp); 2135 2136#ifdef SOFTUPDATES 2137 /* 2138 * Move over the dependencies. If there are rollbacks, 2139 * leave the parent buffer dirtied as it will need to 2140 * be written again. 2141 */ 2142 if (LIST_EMPTY(&bp->b_dep) || 2143 softdep_move_dependencies(bp, newbp) == 0) 2144 bundirty(bp); 2145#else 2146 bundirty(bp); 2147#endif 2148 2149 /* 2150 * Initiate write on the copy, release the original. The 2151 * BKGRDINPROG flag prevents it from going away until 2152 * the background write completes. 2153 */ 2154 bqrelse(bp); 2155 bp = newbp; 2156 } else 2157 /* Mark the buffer clean */ 2158 bundirty(bp); 2159 2160 2161 /* Let the normal bufwrite do the rest for us */ 2162normal_write: 2163 return (bufwrite(bp)); 2164} 2165 2166 2167static void 2168ffs_geom_strategy(struct bufobj *bo, struct buf *bp) 2169{ 2170 struct vnode *vp; 2171 int error; 2172 struct buf *tbp; 2173 int nocopy; 2174 2175 vp = bo->__bo_vnode; 2176 if (bp->b_iocmd == BIO_WRITE) { 2177 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && 2178 bp->b_vp != NULL && bp->b_vp->v_mount != NULL && 2179 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) 2180 panic("ffs_geom_strategy: bad I/O"); 2181 nocopy = bp->b_flags & B_NOCOPY; 2182 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); 2183 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && 2184 vp->v_rdev->si_snapdata != NULL) { 2185 if ((bp->b_flags & B_CLUSTER) != 0) { 2186 runningbufwakeup(bp); 2187 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2188 b_cluster.cluster_entry) { 2189 error = ffs_copyonwrite(vp, tbp); 2190 if (error != 0 && 2191 error != EOPNOTSUPP) { 2192 bp->b_error = error; 2193 bp->b_ioflags |= BIO_ERROR; 2194 bufdone(bp); 2195 return; 2196 } 2197 } 2198 bp->b_runningbufspace = bp->b_bufsize; 2199 atomic_add_long(&runningbufspace, 2200 bp->b_runningbufspace); 2201 } else { 2202 error = ffs_copyonwrite(vp, bp); 2203 if (error != 0 && error != EOPNOTSUPP) { 2204 bp->b_error = error; 2205 bp->b_ioflags |= BIO_ERROR; 2206 bufdone(bp); 2207 return; 2208 } 2209 } 2210 } 2211#ifdef SOFTUPDATES 2212 if ((bp->b_flags & B_CLUSTER) != 0) { 2213 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, 2214 b_cluster.cluster_entry) { 2215 if (!LIST_EMPTY(&tbp->b_dep)) 2216 buf_start(tbp); 2217 } 2218 } else { 2219 if (!LIST_EMPTY(&bp->b_dep)) 2220 buf_start(bp); 2221 } 2222 2223#endif 2224 } 2225 g_vfs_strategy(bo, bp); 2226} 2227 2228int 2229ffs_own_mount(const struct mount *mp) 2230{ 2231 2232 if (mp->mnt_op == &ufs_vfsops) 2233 return (1); 2234 return (0); 2235} 2236 2237#ifdef DDB 2238#ifdef SOFTUPDATES 2239 2240/* defined in ffs_softdep.c */ 2241extern void db_print_ffs(struct ufsmount *ump); 2242 2243DB_SHOW_COMMAND(ffs, db_show_ffs) 2244{ 2245 struct mount *mp; 2246 struct ufsmount *ump; 2247 2248 if (have_addr) { 2249 ump = VFSTOUFS((struct mount *)addr); 2250 db_print_ffs(ump); 2251 return; 2252 } 2253 2254 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2255 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) 2256 db_print_ffs(VFSTOUFS(mp)); 2257 } 2258} 2259 2260#endif /* SOFTUPDATES */ 2261#endif /* DDB */ 2262