1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ 27 /* All Rights Reserved */ 28 29 /* 30 * University Copyright- Copyright (c) 1982, 1986, 1988 31 * The Regents of the University of California 32 * All Rights Reserved 33 * 34 * University Acknowledgment- Portions of this document are derived from 35 * software developed by the University of California, Berkeley, and its 36 * contributors. 37 */ 38 39 40 #pragma ident "%Z%%M% %I% %E% SMI" 41 42 #include <sys/types.h> 43 #include <sys/t_lock.h> 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/bitmap.h> 47 #include <sys/sysmacros.h> 48 #include <sys/kmem.h> 49 #include <sys/signal.h> 50 #include <sys/user.h> 51 #include <sys/proc.h> 52 #include <sys/disp.h> 53 #include <sys/buf.h> 54 #include <sys/pathname.h> 55 #include <sys/vfs.h> 56 #include <sys/vfs_opreg.h> 57 #include <sys/vnode.h> 58 #include <sys/file.h> 59 #include <sys/atomic.h> 60 #include <sys/uio.h> 61 #include <sys/dkio.h> 62 #include <sys/cred.h> 63 #include <sys/conf.h> 64 #include <sys/dnlc.h> 65 #include <sys/kstat.h> 66 #include <sys/acl.h> 67 #include <sys/fs/ufs_fsdir.h> 68 #include <sys/fs/ufs_fs.h> 69 #include <sys/fs/ufs_inode.h> 70 #include <sys/fs/ufs_mount.h> 71 #include <sys/fs/ufs_acl.h> 72 #include <sys/fs/ufs_panic.h> 73 #include <sys/fs/ufs_bio.h> 74 #include <sys/fs/ufs_quota.h> 75 #include <sys/fs/ufs_log.h> 76 #undef NFS 77 #include <sys/statvfs.h> 78 #include <sys/mount.h> 79 #include <sys/mntent.h> 80 #include <sys/swap.h> 81 #include <sys/errno.h> 82 #include <sys/debug.h> 83 #include "fs/fs_subr.h" 84 #include <sys/cmn_err.h> 85 #include <sys/dnlc.h> 86 #include <sys/fssnap_if.h> 87 #include <sys/sunddi.h> 88 #include <sys/bootconf.h> 89 #include <sys/policy.h> 90 #include <sys/zone.h> 91 92 /* 93 * This is the loadable module wrapper. 94 */ 95 #include <sys/modctl.h> 96 97 int ufsfstype; 98 vfsops_t *ufs_vfsops; 99 static int ufsinit(int, char *); 100 static int mountfs(); 101 extern int highbit(); 102 extern struct instats ins; 103 extern struct vnode *common_specvp(struct vnode *vp); 104 extern vfs_t EIO_vfs; 105 106 struct dquot *dquot, *dquotNDQUOT; 107 108 /* 109 * Cylinder group summary information handling tunable. 110 * This defines when these deltas get logged. 111 * If the number of cylinders in the file system is over the 112 * tunable then we log csum updates. Otherwise the updates are only 113 * done for performance on unmount. After a panic they can be 114 * quickly constructed during mounting. See ufs_construct_si() 115 * called from ufs_getsummaryinfo(). 116 * 117 * This performance feature can of course be disabled by setting 118 * ufs_ncg_log to 0, and fully enabled by setting it to 0xffffffff. 119 */ 120 #define UFS_LOG_NCG_DEFAULT 10000 121 uint32_t ufs_ncg_log = UFS_LOG_NCG_DEFAULT; 122 123 /* 124 * ufs_clean_root indicates whether the root fs went down cleanly 125 */ 126 static int ufs_clean_root = 0; 127 128 /* 129 * UFS Mount options table 130 */ 131 static char *intr_cancel[] = { MNTOPT_NOINTR, NULL }; 132 static char *nointr_cancel[] = { MNTOPT_INTR, NULL }; 133 static char *forcedirectio_cancel[] = { MNTOPT_NOFORCEDIRECTIO, NULL }; 134 static char *noforcedirectio_cancel[] = { MNTOPT_FORCEDIRECTIO, NULL }; 135 static char *largefiles_cancel[] = { MNTOPT_NOLARGEFILES, NULL }; 136 static char *nolargefiles_cancel[] = { MNTOPT_LARGEFILES, NULL }; 137 static char *logging_cancel[] = { MNTOPT_NOLOGGING, NULL }; 138 static char *nologging_cancel[] = { MNTOPT_LOGGING, NULL }; 139 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL }; 140 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL }; 141 static char *quota_cancel[] = { MNTOPT_NOQUOTA, NULL }; 142 static char *noquota_cancel[] = { MNTOPT_QUOTA, NULL }; 143 static char *dfratime_cancel[] = { MNTOPT_NODFRATIME, NULL }; 144 static char *nodfratime_cancel[] = { MNTOPT_DFRATIME, NULL }; 145 146 static mntopt_t mntopts[] = { 147 /* 148 * option name cancel option default arg flags 149 * ufs arg flag 150 */ 151 { MNTOPT_INTR, intr_cancel, NULL, MO_DEFAULT, 152 (void *)0 }, 153 { MNTOPT_NOINTR, nointr_cancel, NULL, 0, 154 (void *)UFSMNT_NOINTR }, 155 { MNTOPT_SYNCDIR, NULL, NULL, 0, 156 (void *)UFSMNT_SYNCDIR }, 157 { MNTOPT_FORCEDIRECTIO, forcedirectio_cancel, NULL, 0, 158 (void *)UFSMNT_FORCEDIRECTIO }, 159 { MNTOPT_NOFORCEDIRECTIO, noforcedirectio_cancel, NULL, 0, 160 (void *)UFSMNT_NOFORCEDIRECTIO }, 161 { MNTOPT_NOSETSEC, NULL, NULL, 0, 162 (void *)UFSMNT_NOSETSEC }, 163 { MNTOPT_LARGEFILES, largefiles_cancel, NULL, MO_DEFAULT, 164 (void *)UFSMNT_LARGEFILES }, 165 { MNTOPT_NOLARGEFILES, nolargefiles_cancel, NULL, 0, 166 (void *)0 }, 167 { MNTOPT_LOGGING, logging_cancel, NULL, MO_TAG, 168 (void *)UFSMNT_LOGGING }, 169 { MNTOPT_NOLOGGING, nologging_cancel, NULL, 170 MO_NODISPLAY|MO_DEFAULT|MO_TAG, (void *)0 }, 171 { MNTOPT_QUOTA, quota_cancel, NULL, MO_IGNORE, 172 (void *)0 }, 173 { MNTOPT_NOQUOTA, noquota_cancel, NULL, 174 MO_NODISPLAY|MO_DEFAULT, (void *)0 }, 175 { MNTOPT_GLOBAL, NULL, NULL, 0, 176 (void *)0 }, 177 { MNTOPT_XATTR, xattr_cancel, NULL, MO_DEFAULT, 178 (void *)0 }, 179 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, 180 (void *)0 }, 181 { MNTOPT_NOATIME, NULL, NULL, 0, 182 (void *)UFSMNT_NOATIME }, 183 { MNTOPT_DFRATIME, dfratime_cancel, NULL, 0, 184 (void *)0 }, 185 { MNTOPT_NODFRATIME, nodfratime_cancel, NULL, 186 MO_NODISPLAY|MO_DEFAULT, (void *)UFSMNT_NODFRATIME }, 187 { MNTOPT_ONERROR, NULL, UFSMNT_ONERROR_PANIC_STR, 188 MO_DEFAULT|MO_HASVALUE, (void *)0 }, 189 }; 190 191 static mntopts_t ufs_mntopts = { 192 sizeof (mntopts) / sizeof (mntopt_t), 193 mntopts 194 }; 195 196 static vfsdef_t vfw = { 197 VFSDEF_VERSION, 198 "ufs", 199 ufsinit, 200 VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS, 201 &ufs_mntopts 202 }; 203 204 /* 205 * Module linkage information for the kernel. 206 */ 207 extern struct mod_ops mod_fsops; 208 209 static struct modlfs modlfs = { 210 &mod_fsops, "filesystem for ufs", &vfw 211 }; 212 213 static struct modlinkage modlinkage = { 214 MODREV_1, (void *)&modlfs, NULL 215 }; 216 217 /* 218 * An attempt has been made to make this module unloadable. In order to 219 * test it, we need a system in which the root fs is NOT ufs. THIS HAS NOT 220 * BEEN DONE 221 */ 222 223 extern kstat_t *ufs_inode_kstat; 224 extern uint_t ufs_lockfs_key; 225 extern void ufs_lockfs_tsd_destructor(void *); 226 extern uint_t bypass_snapshot_throttle_key; 227 228 int 229 _init(void) 230 { 231 /* 232 * Create an index into the per thread array so that any thread doing 233 * VOP will have a lockfs mark on it. 234 */ 235 tsd_create(&ufs_lockfs_key, ufs_lockfs_tsd_destructor); 236 tsd_create(&bypass_snapshot_throttle_key, NULL); 237 return (mod_install(&modlinkage)); 238 } 239 240 int 241 _fini(void) 242 { 243 return (EBUSY); 244 } 245 246 int 247 _info(struct modinfo *modinfop) 248 { 249 return (mod_info(&modlinkage, modinfop)); 250 } 251 252 extern struct vnode *makespecvp(dev_t dev, vtype_t type); 253 254 extern kmutex_t ufs_scan_lock; 255 256 static int mountfs(struct vfs *, enum whymountroot, struct vnode *, char *, 257 struct cred *, int, void *, int); 258 259 260 static int 261 ufs_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap, 262 struct cred *cr) 263 264 { 265 char *data = uap->dataptr; 266 int datalen = uap->datalen; 267 dev_t dev; 268 struct vnode *bvp; 269 struct pathname dpn; 270 int error; 271 enum whymountroot why = ROOT_INIT; 272 struct ufs_args args; 273 int oflag, aflag; 274 int fromspace = (uap->flags & MS_SYSSPACE) ? 275 UIO_SYSSPACE : UIO_USERSPACE; 276 277 if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0) 278 return (error); 279 280 if (mvp->v_type != VDIR) 281 return (ENOTDIR); 282 283 mutex_enter(&mvp->v_lock); 284 if ((uap->flags & MS_REMOUNT) == 0 && 285 (uap->flags & MS_OVERLAY) == 0 && 286 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 287 mutex_exit(&mvp->v_lock); 288 return (EBUSY); 289 } 290 mutex_exit(&mvp->v_lock); 291 292 /* 293 * Get arguments 294 */ 295 bzero(&args, sizeof (args)); 296 if ((uap->flags & MS_DATA) && data != NULL && datalen != 0) { 297 int copy_result = 0; 298 299 if (datalen > sizeof (args)) 300 return (EINVAL); 301 if (uap->flags & MS_SYSSPACE) 302 bcopy(data, &args, datalen); 303 else 304 copy_result = copyin(data, &args, datalen); 305 if (copy_result) 306 return (EFAULT); 307 datalen = sizeof (struct ufs_args); 308 } else { 309 datalen = 0; 310 } 311 /* 312 * Read in the mount point pathname 313 * (so we can record the directory the file system was last mounted on). 314 */ 315 if (error = pn_get(uap->dir, fromspace, &dpn)) 316 return (error); 317 318 /* 319 * Resolve path name of special file being mounted. 320 */ 321 if (error = lookupname(uap->spec, fromspace, FOLLOW, NULL, &bvp)) { 322 pn_free(&dpn); 323 return (error); 324 } 325 if (bvp->v_type != VBLK) { 326 VN_RELE(bvp); 327 pn_free(&dpn); 328 return (ENOTBLK); 329 } 330 dev = bvp->v_rdev; 331 if (getmajor(dev) >= devcnt) { 332 pn_free(&dpn); 333 VN_RELE(bvp); 334 return (ENXIO); 335 } 336 if (uap->flags & MS_REMOUNT) 337 why = ROOT_REMOUNT; 338 339 /* 340 * In SunCluster, requests to a global device are satisfied by 341 * a local device. We substitute the global pxfs node with a 342 * local spec node here. 343 */ 344 if (IS_PXFSVP(bvp)) { 345 VN_RELE(bvp); 346 bvp = makespecvp(dev, VBLK); 347 } 348 349 /* 350 * Open block device mounted on. We need this to 351 * check whether the caller has sufficient rights to 352 * access the device in question. 353 * When bio is fixed for vnodes this can all be vnode 354 * operations. 355 */ 356 if ((vfsp->vfs_flag & VFS_RDONLY) != 0 || 357 (uap->flags & MS_RDONLY) != 0) { 358 oflag = FREAD; 359 aflag = VREAD; 360 } else { 361 oflag = FREAD | FWRITE; 362 aflag = VREAD | VWRITE; 363 } 364 if ((error = VOP_ACCESS(bvp, aflag, 0, cr)) != 0 || 365 (error = secpolicy_spec_open(cr, bvp, oflag)) != 0) { 366 pn_free(&dpn); 367 VN_RELE(bvp); 368 return (error); 369 } 370 371 /* 372 * Ensure that this device isn't already mounted or in progress on a 373 * mount unless this is a REMOUNT request or we are told to suppress 374 * mount checks. Global mounts require special handling. 375 */ 376 if ((uap->flags & MS_NOCHECK) == 0) { 377 if ((uap->flags & MS_GLOBAL) == 0 && 378 vfs_devmounting(dev, vfsp)) { 379 pn_free(&dpn); 380 VN_RELE(bvp); 381 return (EBUSY); 382 } 383 if (vfs_devismounted(dev)) { 384 if ((uap->flags & MS_REMOUNT) == 0) { 385 pn_free(&dpn); 386 VN_RELE(bvp); 387 return (EBUSY); 388 } 389 } 390 } 391 392 /* 393 * If the device is a tape, mount it read only 394 */ 395 if (devopsp[getmajor(dev)]->devo_cb_ops->cb_flag & D_TAPE) { 396 vfsp->vfs_flag |= VFS_RDONLY; 397 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 398 } 399 if (uap->flags & MS_RDONLY) 400 vfsp->vfs_flag |= VFS_RDONLY; 401 402 /* 403 * Mount the filesystem, free the device vnode on error. 404 */ 405 error = mountfs(vfsp, why, bvp, dpn.pn_path, cr, 0, &args, datalen); 406 pn_free(&dpn); 407 if (error) { 408 VN_RELE(bvp); 409 } 410 return (error); 411 } 412 /* 413 * Mount root file system. 414 * "why" is ROOT_INIT on initial call ROOT_REMOUNT if called to 415 * remount the root file system, and ROOT_UNMOUNT if called to 416 * unmount the root (e.g., as part of a system shutdown). 417 * 418 * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP 419 * operation, goes along with auto-configuration. A mechanism should be 420 * provided by which machine-INdependent code in the kernel can say "get me the 421 * right root file system" and "get me the right initial swap area", and have 422 * that done in what may well be a machine-dependent fashion. 423 * Unfortunately, it is also file-system-type dependent (NFS gets it via 424 * bootparams calls, UFS gets it from various and sundry machine-dependent 425 * mechanisms, as SPECFS does for swap). 426 */ 427 static int 428 ufs_mountroot(struct vfs *vfsp, enum whymountroot why) 429 { 430 struct fs *fsp; 431 int error; 432 static int ufsrootdone = 0; 433 dev_t rootdev; 434 struct vnode *vp; 435 struct vnode *devvp = 0; 436 int ovflags; 437 int doclkset; 438 ufsvfs_t *ufsvfsp; 439 440 if (why == ROOT_INIT) { 441 if (ufsrootdone++) 442 return (EBUSY); 443 rootdev = getrootdev(); 444 if (rootdev == (dev_t)NODEV) 445 return (ENODEV); 446 vfsp->vfs_dev = rootdev; 447 vfsp->vfs_flag |= VFS_RDONLY; 448 } else if (why == ROOT_REMOUNT) { 449 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp; 450 (void) dnlc_purge_vfsp(vfsp, 0); 451 vp = common_specvp(vp); 452 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_INVAL, CRED()); 453 (void) bfinval(vfsp->vfs_dev, 0); 454 fsp = getfs(vfsp); 455 456 ovflags = vfsp->vfs_flag; 457 vfsp->vfs_flag &= ~VFS_RDONLY; 458 vfsp->vfs_flag |= VFS_REMOUNT; 459 rootdev = vfsp->vfs_dev; 460 } else if (why == ROOT_UNMOUNT) { 461 if (vfs_lock(vfsp) == 0) { 462 (void) ufs_flush(vfsp); 463 /* 464 * Mark the log as fully rolled 465 */ 466 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data; 467 fsp = ufsvfsp->vfs_fs; 468 if (TRANS_ISTRANS(ufsvfsp) && 469 !TRANS_ISERROR(ufsvfsp) && 470 (fsp->fs_rolled == FS_NEED_ROLL)) { 471 ml_unit_t *ul = ufsvfsp->vfs_log; 472 473 error = ufs_putsummaryinfo(ul->un_dev, 474 ufsvfsp, fsp); 475 if (error == 0) { 476 fsp->fs_rolled = FS_ALL_ROLLED; 477 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp); 478 } 479 } 480 vfs_unlock(vfsp); 481 } else { 482 ufs_update(0); 483 } 484 485 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp; 486 (void) VOP_CLOSE(vp, FREAD|FWRITE, 1, 487 (offset_t)0, CRED()); 488 return (0); 489 } 490 error = vfs_lock(vfsp); 491 if (error) 492 return (error); 493 494 devvp = makespecvp(rootdev, VBLK); 495 496 /* If RO media, don't call clkset() (see below) */ 497 doclkset = 1; 498 if (why == ROOT_INIT) { 499 error = VOP_OPEN(&devvp, FREAD|FWRITE, CRED()); 500 if (error == 0) { 501 (void) VOP_CLOSE(devvp, FREAD|FWRITE, 1, 502 (offset_t)0, CRED()); 503 } else { 504 doclkset = 0; 505 } 506 } 507 508 error = mountfs(vfsp, why, devvp, "/", CRED(), 1, NULL, 0); 509 /* 510 * XXX - assumes root device is not indirect, because we don't set 511 * rootvp. Is rootvp used for anything? If so, make another arg 512 * to mountfs. 513 */ 514 if (error) { 515 vfs_unlock(vfsp); 516 if (why == ROOT_REMOUNT) 517 vfsp->vfs_flag = ovflags; 518 if (rootvp) { 519 VN_RELE(rootvp); 520 rootvp = (struct vnode *)0; 521 } 522 VN_RELE(devvp); 523 return (error); 524 } 525 if (why == ROOT_INIT) 526 vfs_add((struct vnode *)0, vfsp, 527 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 528 vfs_unlock(vfsp); 529 fsp = getfs(vfsp); 530 clkset(doclkset ? fsp->fs_time : -1); 531 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data; 532 if (ufsvfsp->vfs_log) { 533 vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0); 534 } 535 return (0); 536 } 537 538 static int 539 remountfs(struct vfs *vfsp, dev_t dev, void *raw_argsp, int args_len) 540 { 541 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 542 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs; 543 struct buf *bp = ufsvfsp->vfs_bufp; 544 struct fs *fsp = (struct fs *)bp->b_un.b_addr; 545 struct fs *fspt; 546 struct buf *tpt = 0; 547 int error = 0; 548 int flags = 0; 549 550 if (args_len == sizeof (struct ufs_args) && raw_argsp) 551 flags = ((struct ufs_args *)raw_argsp)->flags; 552 553 /* cannot remount to RDONLY */ 554 if (vfsp->vfs_flag & VFS_RDONLY) 555 return (ENOTSUP); 556 557 /* whoops, wrong dev */ 558 if (vfsp->vfs_dev != dev) 559 return (EINVAL); 560 561 /* 562 * synchronize w/ufs ioctls 563 */ 564 mutex_enter(&ulp->ul_lock); 565 atomic_add_long(&ufs_quiesce_pend, 1); 566 567 /* 568 * reset options 569 */ 570 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR; 571 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR; 572 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC; 573 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME; 574 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime) 575 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME; 576 else /* dfratime, default behavior */ 577 ufsvfsp->vfs_dfritime |= UFS_DFRATIME; 578 if (flags & UFSMNT_FORCEDIRECTIO) 579 ufsvfsp->vfs_forcedirectio = 1; 580 else /* default is no direct I/O */ 581 ufsvfsp->vfs_forcedirectio = 0; 582 ufsvfsp->vfs_iotstamp = lbolt; 583 584 /* 585 * set largefiles flag in ufsvfs equal to the 586 * value passed in by the mount command. If 587 * it is "nolargefiles", and the flag is set 588 * in the superblock, the mount fails. 589 */ 590 if (!(flags & UFSMNT_LARGEFILES)) { /* "nolargefiles" */ 591 if (fsp->fs_flags & FSLARGEFILES) { 592 error = EFBIG; 593 goto remounterr; 594 } 595 ufsvfsp->vfs_lfflags &= ~UFS_LARGEFILES; 596 } else /* "largefiles" */ 597 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES; 598 /* 599 * read/write to read/write; all done 600 */ 601 if (fsp->fs_ronly == 0) 602 goto remounterr; 603 604 /* 605 * fix-on-panic assumes RO->RW remount implies system-critical fs 606 * if it is shortly after boot; so, don't attempt to lock and fix 607 * (unless the user explicitly asked for another action on error) 608 * XXX UFSMNT_ONERROR_RDONLY rather than UFSMNT_ONERROR_PANIC 609 */ 610 #define BOOT_TIME_LIMIT (180*hz) 611 if (!(flags & UFSMNT_ONERROR_FLGMASK) && lbolt < BOOT_TIME_LIMIT) { 612 cmn_err(CE_WARN, "%s is required to be mounted onerror=%s", 613 ufsvfsp->vfs_fs->fs_fsmnt, UFSMNT_ONERROR_PANIC_STR); 614 flags |= UFSMNT_ONERROR_PANIC; 615 } 616 617 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0) 618 goto remounterr; 619 620 /* 621 * quiesce the file system 622 */ 623 error = ufs_quiesce(ulp); 624 if (error) 625 goto remounterr; 626 627 tpt = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, SBLOCK, SBSIZE); 628 if (tpt->b_flags & B_ERROR) { 629 error = EIO; 630 goto remounterr; 631 } 632 fspt = (struct fs *)tpt->b_un.b_addr; 633 if (((fspt->fs_magic != FS_MAGIC) && 634 (fspt->fs_magic != MTB_UFS_MAGIC)) || 635 (fspt->fs_magic == FS_MAGIC && 636 (fspt->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && 637 fspt->fs_version != UFS_VERSION_MIN)) || 638 (fspt->fs_magic == MTB_UFS_MAGIC && 639 (fspt->fs_version > MTB_UFS_VERSION_1 || 640 fspt->fs_version < MTB_UFS_VERSION_MIN)) || 641 fspt->fs_bsize > MAXBSIZE || fspt->fs_frag > MAXFRAG || 642 fspt->fs_bsize < sizeof (struct fs) || fspt->fs_bsize < PAGESIZE) { 643 tpt->b_flags |= B_STALE | B_AGE; 644 error = EINVAL; 645 goto remounterr; 646 } 647 648 if (ufsvfsp->vfs_log && (ufsvfsp->vfs_log->un_flags & LDL_NOROLL)) { 649 ufsvfsp->vfs_log->un_flags &= ~LDL_NOROLL; 650 logmap_start_roll(ufsvfsp->vfs_log); 651 } 652 653 if (TRANS_ISERROR(ufsvfsp)) 654 goto remounterr; 655 TRANS_DOMATAMAP(ufsvfsp); 656 657 if ((fspt->fs_state + fspt->fs_time == FSOKAY) && 658 fspt->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp)) { 659 ufsvfsp->vfs_log = NULL; 660 ufsvfsp->vfs_domatamap = 0; 661 error = ENOSPC; 662 goto remounterr; 663 } 664 665 if (fspt->fs_state + fspt->fs_time == FSOKAY && 666 (fspt->fs_clean == FSCLEAN || 667 fspt->fs_clean == FSSTABLE || 668 fspt->fs_clean == FSLOG)) { 669 670 /* 671 * Ensure that ufs_getsummaryinfo doesn't reconstruct 672 * the summary info. 673 */ 674 error = ufs_getsummaryinfo(vfsp->vfs_dev, ufsvfsp, fspt); 675 if (error) 676 goto remounterr; 677 678 /* preserve mount name */ 679 (void) strncpy(fspt->fs_fsmnt, fsp->fs_fsmnt, MAXMNTLEN); 680 /* free the old cg space */ 681 kmem_free(fsp->fs_u.fs_csp, fsp->fs_cssize); 682 /* switch in the new superblock */ 683 fspt->fs_rolled = FS_NEED_ROLL; 684 bcopy(tpt->b_un.b_addr, bp->b_un.b_addr, fspt->fs_sbsize); 685 686 fsp->fs_clean = FSSTABLE; 687 } /* superblock updated in memory */ 688 tpt->b_flags |= B_STALE | B_AGE; 689 brelse(tpt); 690 tpt = 0; 691 692 if (fsp->fs_clean != FSSTABLE) { 693 error = ENOSPC; 694 goto remounterr; 695 } 696 697 698 if (TRANS_ISTRANS(ufsvfsp)) { 699 fsp->fs_clean = FSLOG; 700 ufsvfsp->vfs_dio = 0; 701 } else 702 if (ufsvfsp->vfs_dio) 703 fsp->fs_clean = FSSUSPEND; 704 705 TRANS_MATA_MOUNT(ufsvfsp); 706 707 fsp->fs_fmod = 0; 708 fsp->fs_ronly = 0; 709 710 atomic_add_long(&ufs_quiesce_pend, -1); 711 cv_broadcast(&ulp->ul_cv); 712 mutex_exit(&ulp->ul_lock); 713 714 if (TRANS_ISTRANS(ufsvfsp)) { 715 716 /* 717 * start the delete thread 718 */ 719 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp); 720 721 /* 722 * start the reclaim thread 723 */ 724 if (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) { 725 fsp->fs_reclaim &= ~FS_RECLAIM; 726 fsp->fs_reclaim |= FS_RECLAIMING; 727 ufs_thread_start(&ufsvfsp->vfs_reclaim, 728 ufs_thread_reclaim, vfsp); 729 } 730 } 731 732 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT); 733 734 return (0); 735 736 remounterr: 737 if (tpt) 738 brelse(tpt); 739 atomic_add_long(&ufs_quiesce_pend, -1); 740 cv_broadcast(&ulp->ul_cv); 741 mutex_exit(&ulp->ul_lock); 742 return (error); 743 } 744 745 /* 746 * If the device maxtransfer size is not available, we use ufs_maxmaxphys 747 * along with the system value for maxphys to determine the value for 748 * maxtransfer. 749 */ 750 int ufs_maxmaxphys = (1024 * 1024); 751 752 #include <sys/ddi.h> /* for delay(9f) */ 753 754 int ufs_mount_error_delay = 20; /* default to 20ms */ 755 int ufs_mount_timeout = 60000; /* default to 1 minute */ 756 757 static int 758 mountfs(struct vfs *vfsp, enum whymountroot why, struct vnode *devvp, 759 char *path, cred_t *cr, int isroot, void *raw_argsp, int args_len) 760 { 761 dev_t dev = devvp->v_rdev; 762 struct fs *fsp; 763 struct ufsvfs *ufsvfsp = 0; 764 struct buf *bp = 0; 765 struct buf *tp = 0; 766 struct dk_cinfo ci; 767 int error = 0; 768 size_t len; 769 int needclose = 0; 770 int needtrans = 0; 771 struct inode *rip; 772 struct vnode *rvp = NULL; 773 int flags = 0; 774 kmutex_t *ihm; 775 int elapsed; 776 int status; 777 extern int maxphys; 778 779 if (args_len == sizeof (struct ufs_args) && raw_argsp) 780 flags = ((struct ufs_args *)raw_argsp)->flags; 781 782 ASSERT(vfs_lock_held(vfsp)); 783 784 if (why == ROOT_INIT) { 785 /* 786 * Open block device mounted on. 787 * When bio is fixed for vnodes this can all be vnode 788 * operations. 789 */ 790 error = VOP_OPEN(&devvp, 791 (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, cr); 792 if (error) 793 goto out; 794 needclose = 1; 795 796 /* 797 * Refuse to go any further if this 798 * device is being used for swapping. 799 */ 800 if (IS_SWAPVP(devvp)) { 801 error = EBUSY; 802 goto out; 803 } 804 } 805 806 /* 807 * check for dev already mounted on 808 */ 809 if (vfsp->vfs_flag & VFS_REMOUNT) { 810 error = remountfs(vfsp, dev, raw_argsp, args_len); 811 if (error == 0) 812 VN_RELE(devvp); 813 return (error); 814 } 815 816 ASSERT(devvp != 0); 817 818 /* 819 * Flush back any dirty pages on the block device to 820 * try and keep the buffer cache in sync with the page 821 * cache if someone is trying to use block devices when 822 * they really should be using the raw device. 823 */ 824 (void) VOP_PUTPAGE(common_specvp(devvp), (offset_t)0, 825 (size_t)0, B_INVAL, cr); 826 827 /* 828 * read in superblock 829 */ 830 ufsvfsp = kmem_zalloc(sizeof (struct ufsvfs), KM_SLEEP); 831 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE); 832 if (tp->b_flags & B_ERROR) 833 goto out; 834 fsp = (struct fs *)tp->b_un.b_addr; 835 836 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) { 837 cmn_err(CE_NOTE, 838 "mount: not a UFS magic number (0x%x)", fsp->fs_magic); 839 error = EINVAL; 840 goto out; 841 } 842 843 if ((fsp->fs_magic == FS_MAGIC) && 844 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && 845 fsp->fs_version != UFS_VERSION_MIN)) { 846 cmn_err(CE_NOTE, 847 "mount: unrecognized version of UFS on-disk format: %d", 848 fsp->fs_version); 849 error = EINVAL; 850 goto out; 851 } 852 853 if ((fsp->fs_magic == MTB_UFS_MAGIC) && 854 (fsp->fs_version > MTB_UFS_VERSION_1 || 855 fsp->fs_version < MTB_UFS_VERSION_MIN)) { 856 cmn_err(CE_NOTE, 857 "mount: unrecognized version of UFS on-disk format: %d", 858 fsp->fs_version); 859 error = EINVAL; 860 goto out; 861 } 862 863 #ifndef _LP64 864 if (fsp->fs_magic == MTB_UFS_MAGIC) { 865 /* 866 * Find the size of the device in sectors. If the 867 * the size in sectors is greater than INT_MAX, it's 868 * a multi-terabyte file system, which can't be 869 * mounted by a 32-bit kernel. We can't use the 870 * fsbtodb() macro in the next line because the macro 871 * casts the intermediate values to daddr_t, which is 872 * a 32-bit quantity in a 32-bit kernel. Here we 873 * really do need the intermediate values to be held 874 * in 64-bit quantities because we're checking for 875 * overflow of a 32-bit field. 876 */ 877 if ((((diskaddr_t)(fsp->fs_size)) << fsp->fs_fsbtodb) 878 > INT_MAX) { 879 cmn_err(CE_NOTE, 880 "mount: multi-terabyte UFS cannot be" 881 " mounted by a 32-bit kernel"); 882 error = EINVAL; 883 goto out; 884 } 885 886 } 887 #endif 888 889 if (fsp->fs_bsize > MAXBSIZE || fsp->fs_frag > MAXFRAG || 890 fsp->fs_bsize < sizeof (struct fs) || fsp->fs_bsize < PAGESIZE) { 891 error = EINVAL; /* also needs translation */ 892 goto out; 893 } 894 895 /* 896 * Allocate VFS private data. 897 */ 898 vfsp->vfs_bcount = 0; 899 vfsp->vfs_data = (caddr_t)ufsvfsp; 900 vfsp->vfs_fstype = ufsfstype; 901 vfsp->vfs_dev = dev; 902 vfsp->vfs_flag |= VFS_NOTRUNC; 903 vfs_make_fsid(&vfsp->vfs_fsid, dev, ufsfstype); 904 ufsvfsp->vfs_devvp = devvp; 905 906 /* 907 * Cross-link with vfs and add to instance list. 908 */ 909 ufsvfsp->vfs_vfs = vfsp; 910 ufs_vfs_add(ufsvfsp); 911 912 ufsvfsp->vfs_dev = dev; 913 ufsvfsp->vfs_bufp = tp; 914 915 ufsvfsp->vfs_dirsize = INODESIZE + (4 * ALLOCSIZE) + fsp->fs_fsize; 916 ufsvfsp->vfs_minfrags = (int)((int64_t)fsp->fs_dsize * 917 fsp->fs_minfree / 100); 918 /* 919 * if mount allows largefiles, indicate so in ufsvfs 920 */ 921 if (flags & UFSMNT_LARGEFILES) 922 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES; 923 /* 924 * Initialize threads 925 */ 926 ufs_delete_init(ufsvfsp, 1); 927 ufs_thread_init(&ufsvfsp->vfs_reclaim, 0); 928 929 /* 930 * Chicken and egg problem. The superblock may have deltas 931 * in the log. So after the log is scanned we reread the 932 * superblock. We guarantee that the fields needed to 933 * scan the log will not be in the log. 934 */ 935 if (fsp->fs_logbno && fsp->fs_clean == FSLOG && 936 (fsp->fs_state + fsp->fs_time == FSOKAY)) { 937 error = lufs_snarf(ufsvfsp, fsp, (vfsp->vfs_flag & VFS_RDONLY)); 938 if (error) { 939 /* 940 * Allow a ro mount to continue even if the 941 * log cannot be processed - yet. 942 */ 943 if (!(vfsp->vfs_flag & VFS_RDONLY)) { 944 cmn_err(CE_WARN, "Error accessing ufs " 945 "log for %s; Please run fsck(1M)", 946 path); 947 goto out; 948 } 949 } 950 tp->b_flags |= (B_AGE | B_STALE); 951 brelse(tp); 952 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE); 953 fsp = (struct fs *)tp->b_un.b_addr; 954 ufsvfsp->vfs_bufp = tp; 955 if (tp->b_flags & B_ERROR) 956 goto out; 957 } 958 959 /* 960 * Set logging mounted flag used by lockfs 961 */ 962 ufsvfsp->vfs_validfs = UT_MOUNTED; 963 964 /* 965 * Copy the super block into a buffer in its native size. 966 * Use ngeteblk to allocate the buffer 967 */ 968 bp = ngeteblk(fsp->fs_bsize); 969 ufsvfsp->vfs_bufp = bp; 970 bp->b_edev = dev; 971 bp->b_dev = cmpdev(dev); 972 bp->b_blkno = SBLOCK; 973 bp->b_bcount = fsp->fs_sbsize; 974 bcopy(tp->b_un.b_addr, bp->b_un.b_addr, fsp->fs_sbsize); 975 tp->b_flags |= B_STALE | B_AGE; 976 brelse(tp); 977 tp = 0; 978 979 fsp = (struct fs *)bp->b_un.b_addr; 980 /* 981 * Mount fails if superblock flag indicates presence of large 982 * files and filesystem is attempted to be mounted 'nolargefiles'. 983 * The exception is for a read only mount of root, which we 984 * always want to succeed, so fsck can fix potential problems. 985 * The assumption is that we will remount root at some point, 986 * and the remount will enforce the mount option. 987 */ 988 if (!(isroot & (vfsp->vfs_flag & VFS_RDONLY)) && 989 (fsp->fs_flags & FSLARGEFILES) && 990 !(flags & UFSMNT_LARGEFILES)) { 991 error = EFBIG; 992 goto out; 993 } 994 995 if (vfsp->vfs_flag & VFS_RDONLY) { 996 fsp->fs_ronly = 1; 997 fsp->fs_fmod = 0; 998 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) && 999 ((fsp->fs_clean == FSCLEAN) || 1000 (fsp->fs_clean == FSSTABLE) || 1001 (fsp->fs_clean == FSLOG))) { 1002 if (isroot) { 1003 if (fsp->fs_clean == FSLOG) { 1004 if (fsp->fs_rolled == FS_ALL_ROLLED) { 1005 ufs_clean_root = 1; 1006 } 1007 } else { 1008 ufs_clean_root = 1; 1009 } 1010 } 1011 fsp->fs_clean = FSSTABLE; 1012 } else { 1013 fsp->fs_clean = FSBAD; 1014 } 1015 } else { 1016 1017 fsp->fs_fmod = 0; 1018 fsp->fs_ronly = 0; 1019 1020 TRANS_DOMATAMAP(ufsvfsp); 1021 1022 if ((TRANS_ISERROR(ufsvfsp)) || 1023 (((fsp->fs_state + fsp->fs_time) == FSOKAY) && 1024 fsp->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp))) { 1025 ufsvfsp->vfs_log = NULL; 1026 ufsvfsp->vfs_domatamap = 0; 1027 error = ENOSPC; 1028 goto out; 1029 } 1030 1031 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) && 1032 (fsp->fs_clean == FSCLEAN || 1033 fsp->fs_clean == FSSTABLE || 1034 fsp->fs_clean == FSLOG)) 1035 fsp->fs_clean = FSSTABLE; 1036 else { 1037 if (isroot) { 1038 /* 1039 * allow root partition to be mounted even 1040 * when fs_state is not ok 1041 * will be fixed later by a remount root 1042 */ 1043 fsp->fs_clean = FSBAD; 1044 ufsvfsp->vfs_log = NULL; 1045 ufsvfsp->vfs_domatamap = 0; 1046 } else { 1047 error = ENOSPC; 1048 goto out; 1049 } 1050 } 1051 1052 if (fsp->fs_clean == FSSTABLE && TRANS_ISTRANS(ufsvfsp)) 1053 fsp->fs_clean = FSLOG; 1054 } 1055 TRANS_MATA_MOUNT(ufsvfsp); 1056 needtrans = 1; 1057 1058 vfsp->vfs_bsize = fsp->fs_bsize; 1059 1060 /* 1061 * Read in summary info 1062 */ 1063 if (error = ufs_getsummaryinfo(dev, ufsvfsp, fsp)) 1064 goto out; 1065 1066 /* 1067 * lastwhinetime is set to zero rather than lbolt, so that after 1068 * mounting if the filesystem is found to be full, then immediately the 1069 * "file system message" will be logged. 1070 */ 1071 ufsvfsp->vfs_lastwhinetime = 0L; 1072 1073 1074 mutex_init(&ufsvfsp->vfs_lock, NULL, MUTEX_DEFAULT, NULL); 1075 (void) copystr(path, fsp->fs_fsmnt, sizeof (fsp->fs_fsmnt) - 1, &len); 1076 bzero(fsp->fs_fsmnt + len, sizeof (fsp->fs_fsmnt) - len); 1077 1078 /* 1079 * Sanity checks for old file systems 1080 */ 1081 if (fsp->fs_postblformat == FS_42POSTBLFMT) 1082 ufsvfsp->vfs_nrpos = 8; 1083 else 1084 ufsvfsp->vfs_nrpos = fsp->fs_nrpos; 1085 1086 /* 1087 * Initialize lockfs structure to support file system locking 1088 */ 1089 bzero(&ufsvfsp->vfs_ulockfs.ul_lockfs, 1090 sizeof (struct lockfs)); 1091 ufsvfsp->vfs_ulockfs.ul_fs_lock = ULOCKFS_ULOCK; 1092 mutex_init(&ufsvfsp->vfs_ulockfs.ul_lock, NULL, 1093 MUTEX_DEFAULT, NULL); 1094 cv_init(&ufsvfsp->vfs_ulockfs.ul_cv, NULL, CV_DEFAULT, NULL); 1095 1096 /* 1097 * We don't need to grab vfs_dqrwlock for this ufs_iget() call. 1098 * We are in the process of mounting the file system so there 1099 * is no need to grab the quota lock. If a quota applies to the 1100 * root inode, then it will be updated when quotas are enabled. 1101 * 1102 * However, we have an ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock)) 1103 * in getinoquota() that we want to keep so grab it anyway. 1104 */ 1105 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 1106 1107 error = ufs_iget_alloced(vfsp, UFSROOTINO, &rip, cr); 1108 1109 rw_exit(&ufsvfsp->vfs_dqrwlock); 1110 1111 if (error) 1112 goto out; 1113 1114 /* 1115 * make sure root inode is a directory. Returning ENOTDIR might 1116 * be confused with the mount point not being a directory, so 1117 * we use EIO instead. 1118 */ 1119 if ((rip->i_mode & IFMT) != IFDIR) { 1120 /* 1121 * Mark this inode as subject for cleanup 1122 * to avoid stray inodes in the cache. 1123 */ 1124 rvp = ITOV(rip); 1125 error = EIO; 1126 goto out; 1127 } 1128 1129 rvp = ITOV(rip); 1130 mutex_enter(&rvp->v_lock); 1131 rvp->v_flag |= VROOT; 1132 mutex_exit(&rvp->v_lock); 1133 ufsvfsp->vfs_root = rvp; 1134 /* The buffer for the root inode does not contain a valid b_vp */ 1135 (void) bfinval(dev, 0); 1136 1137 /* options */ 1138 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC; 1139 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR; 1140 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR; 1141 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME; 1142 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime) 1143 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME; 1144 else /* dfratime, default behavior */ 1145 ufsvfsp->vfs_dfritime |= UFS_DFRATIME; 1146 if (flags & UFSMNT_FORCEDIRECTIO) 1147 ufsvfsp->vfs_forcedirectio = 1; 1148 else if (flags & UFSMNT_NOFORCEDIRECTIO) 1149 ufsvfsp->vfs_forcedirectio = 0; 1150 ufsvfsp->vfs_iotstamp = lbolt; 1151 1152 ufsvfsp->vfs_nindiroffset = fsp->fs_nindir - 1; 1153 ufsvfsp->vfs_nindirshift = highbit(ufsvfsp->vfs_nindiroffset); 1154 ufsvfsp->vfs_ioclustsz = fsp->fs_bsize * fsp->fs_maxcontig; 1155 1156 if (cdev_ioctl(dev, DKIOCINFO, (intptr_t)&ci, 1157 FKIOCTL|FNATIVE|FREAD, CRED(), &status) == 0) { 1158 ufsvfsp->vfs_iotransz = ci.dki_maxtransfer * DEV_BSIZE; 1159 } else { 1160 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys); 1161 } 1162 1163 if (ufsvfsp->vfs_iotransz <= 0) { 1164 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys); 1165 } 1166 1167 /* 1168 * When logging, used to reserve log space for writes and truncs 1169 */ 1170 ufsvfsp->vfs_avgbfree = fsp->fs_cstotal.cs_nbfree / fsp->fs_ncg; 1171 1172 /* 1173 * Determine whether to log cylinder group summary info. 1174 */ 1175 ufsvfsp->vfs_nolog_si = (fsp->fs_ncg < ufs_ncg_log); 1176 1177 if (TRANS_ISTRANS(ufsvfsp)) { 1178 /* 1179 * start the delete thread 1180 */ 1181 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp); 1182 1183 /* 1184 * start reclaim thread if the filesystem was not mounted 1185 * read only. 1186 */ 1187 if (!fsp->fs_ronly && (fsp->fs_reclaim & 1188 (FS_RECLAIM|FS_RECLAIMING))) { 1189 fsp->fs_reclaim &= ~FS_RECLAIM; 1190 fsp->fs_reclaim |= FS_RECLAIMING; 1191 ufs_thread_start(&ufsvfsp->vfs_reclaim, 1192 ufs_thread_reclaim, vfsp); 1193 } 1194 1195 /* Mark the fs as unrolled */ 1196 fsp->fs_rolled = FS_NEED_ROLL; 1197 } else if (!fsp->fs_ronly && (fsp->fs_reclaim & 1198 (FS_RECLAIM|FS_RECLAIMING))) { 1199 /* 1200 * If a file system that is mounted nologging, after 1201 * having previously been mounted logging, becomes 1202 * unmounted whilst the reclaim thread is in the throes 1203 * of reclaiming open/deleted inodes, a subsequent mount 1204 * of such a file system with logging disabled could lead 1205 * to inodes becoming lost. So, start reclaim now, even 1206 * though logging was disabled for the previous mount, to 1207 * tidy things up. 1208 */ 1209 fsp->fs_reclaim &= ~FS_RECLAIM; 1210 fsp->fs_reclaim |= FS_RECLAIMING; 1211 ufs_thread_start(&ufsvfsp->vfs_reclaim, 1212 ufs_thread_reclaim, vfsp); 1213 } 1214 1215 if (!fsp->fs_ronly) { 1216 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT); 1217 if (error = geterror(ufsvfsp->vfs_bufp)) 1218 goto out; 1219 } 1220 1221 /* fix-on-panic initialization */ 1222 if (isroot && !(flags & UFSMNT_ONERROR_FLGMASK)) 1223 flags |= UFSMNT_ONERROR_PANIC; /* XXX ..._RDONLY */ 1224 1225 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0) 1226 goto out; 1227 1228 if (why == ROOT_INIT && isroot) 1229 rootvp = devvp; 1230 1231 return (0); 1232 out: 1233 if (error == 0) 1234 error = EIO; 1235 if (rvp) { 1236 /* the following sequence is similar to ufs_unmount() */ 1237 1238 /* 1239 * There's a problem that ufs_iget() puts inodes into 1240 * the inode cache before it returns them. If someone 1241 * traverses that cache and gets a reference to our 1242 * inode, there's a chance they'll still be using it 1243 * after we've destroyed it. This is a hard race to 1244 * hit, but it's happened (putting in a medium delay 1245 * here, and a large delay in ufs_scan_inodes() for 1246 * inodes on the device we're bailing out on, makes 1247 * the race easy to demonstrate). The symptom is some 1248 * other part of UFS faulting on bad inode contents, 1249 * or when grabbing one of the locks inside the inode, 1250 * etc. The usual victim is ufs_scan_inodes() or 1251 * someone called by it. 1252 */ 1253 1254 /* 1255 * First, isolate it so that no new references can be 1256 * gotten via the inode cache. 1257 */ 1258 ihm = &ih_lock[INOHASH(UFSROOTINO)]; 1259 mutex_enter(ihm); 1260 remque(rip); 1261 mutex_exit(ihm); 1262 1263 /* 1264 * Now wait for all outstanding references except our 1265 * own to drain. This could, in theory, take forever, 1266 * so don't wait *too* long. If we time out, mark 1267 * it stale and leak it, so we don't hit the problem 1268 * described above. 1269 * 1270 * Note that v_count is an int, which means we can read 1271 * it in one operation. Thus, there's no need to lock 1272 * around our tests. 1273 */ 1274 elapsed = 0; 1275 while ((rvp->v_count > 1) && (elapsed < ufs_mount_timeout)) { 1276 delay(ufs_mount_error_delay * drv_usectohz(1000)); 1277 elapsed += ufs_mount_error_delay; 1278 } 1279 1280 if (rvp->v_count > 1) { 1281 mutex_enter(&rip->i_tlock); 1282 rip->i_flag |= ISTALE; 1283 mutex_exit(&rip->i_tlock); 1284 cmn_err(CE_WARN, 1285 "Timed out while cleaning up after failed mount of %s", 1286 path); 1287 } else { 1288 1289 /* 1290 * Now we're the only one with a handle left, so tear 1291 * it down the rest of the way. 1292 */ 1293 if (ufs_rmidle(rip)) 1294 VN_RELE(rvp); 1295 ufs_si_del(rip); 1296 rip->i_ufsvfs = NULL; 1297 rvp->v_vfsp = NULL; 1298 rvp->v_type = VBAD; 1299 VN_RELE(rvp); 1300 } 1301 } 1302 if (needtrans) { 1303 TRANS_MATA_UMOUNT(ufsvfsp); 1304 } 1305 if (ufsvfsp) { 1306 ufs_vfs_remove(ufsvfsp); 1307 ufs_thread_exit(&ufsvfsp->vfs_delete); 1308 ufs_thread_exit(&ufsvfsp->vfs_reclaim); 1309 mutex_destroy(&ufsvfsp->vfs_lock); 1310 if (ufsvfsp->vfs_log) { 1311 lufs_unsnarf(ufsvfsp); 1312 } 1313 kmem_free(ufsvfsp, sizeof (struct ufsvfs)); 1314 } 1315 if (bp) { 1316 bp->b_flags |= (B_STALE|B_AGE); 1317 brelse(bp); 1318 } 1319 if (tp) { 1320 tp->b_flags |= (B_STALE|B_AGE); 1321 brelse(tp); 1322 } 1323 if (needclose) { 1324 (void) VOP_CLOSE(devvp, (vfsp->vfs_flag & VFS_RDONLY) ? 1325 FREAD : FREAD|FWRITE, 1, (offset_t)0, cr); 1326 bflush(dev); 1327 (void) bfinval(dev, 1); 1328 } 1329 return (error); 1330 } 1331 1332 /* 1333 * vfs operations 1334 */ 1335 static int 1336 ufs_unmount(struct vfs *vfsp, int fflag, struct cred *cr) 1337 { 1338 dev_t dev = vfsp->vfs_dev; 1339 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1340 struct fs *fs = ufsvfsp->vfs_fs; 1341 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs; 1342 struct vnode *bvp, *vp; 1343 struct buf *bp; 1344 struct inode *ip, *inext, *rip; 1345 union ihead *ih; 1346 int error, flag, i; 1347 struct lockfs lockfs; 1348 int poll_events = POLLPRI; 1349 extern struct pollhead ufs_pollhd; 1350 refstr_t *mountpoint; 1351 1352 ASSERT(vfs_lock_held(vfsp)); 1353 1354 if (secpolicy_fs_unmount(cr, vfsp) != 0) 1355 return (EPERM); 1356 /* 1357 * Forced unmount is now supported through the 1358 * lockfs protocol. 1359 */ 1360 if (fflag & MS_FORCE) { 1361 /* 1362 * Mark the filesystem as being unmounted now in 1363 * case of a forcible umount before we take any 1364 * locks inside UFS to prevent racing with a VFS_VGET() 1365 * request. Throw these VFS_VGET() requests away for 1366 * the duration of the forcible umount so they won't 1367 * use stale or even freed data later on when we're done. 1368 * It may happen that the VFS has had a additional hold 1369 * placed on it by someone other than UFS and thus will 1370 * not get freed immediately once we're done with the 1371 * umount by dounmount() - use VFS_UNMOUNTED to inform 1372 * users of this still-alive VFS that its corresponding 1373 * filesystem being gone so they can detect that and error 1374 * out. 1375 */ 1376 vfsp->vfs_flag |= VFS_UNMOUNTED; 1377 1378 ufs_thread_suspend(&ufsvfsp->vfs_delete); 1379 mutex_enter(&ulp->ul_lock); 1380 /* 1381 * If file system is already hard locked, 1382 * unmount the file system, otherwise 1383 * hard lock it before unmounting. 1384 */ 1385 if (!ULOCKFS_IS_HLOCK(ulp)) { 1386 atomic_add_long(&ufs_quiesce_pend, 1); 1387 lockfs.lf_lock = LOCKFS_HLOCK; 1388 lockfs.lf_flags = 0; 1389 lockfs.lf_key = ulp->ul_lockfs.lf_key + 1; 1390 lockfs.lf_comlen = 0; 1391 lockfs.lf_comment = NULL; 1392 ufs_freeze(ulp, &lockfs); 1393 ULOCKFS_SET_BUSY(ulp); 1394 LOCKFS_SET_BUSY(&ulp->ul_lockfs); 1395 (void) ufs_quiesce(ulp); 1396 (void) ufs_flush(vfsp); 1397 (void) ufs_thaw(vfsp, ufsvfsp, ulp); 1398 atomic_add_long(&ufs_quiesce_pend, -1); 1399 ULOCKFS_CLR_BUSY(ulp); 1400 LOCKFS_CLR_BUSY(&ulp->ul_lockfs); 1401 poll_events |= POLLERR; 1402 pollwakeup(&ufs_pollhd, poll_events); 1403 } 1404 ufs_thread_continue(&ufsvfsp->vfs_delete); 1405 mutex_exit(&ulp->ul_lock); 1406 } 1407 1408 /* let all types of writes go through */ 1409 ufsvfsp->vfs_iotstamp = lbolt; 1410 1411 /* coordinate with global hlock thread */ 1412 if (TRANS_ISTRANS(ufsvfsp) && (ufsvfsp->vfs_validfs == UT_HLOCKING)) { 1413 /* 1414 * last possibility for a forced umount to fail hence clear 1415 * VFS_UNMOUNTED if appropriate. 1416 */ 1417 if (fflag & MS_FORCE) 1418 vfsp->vfs_flag &= ~VFS_UNMOUNTED; 1419 return (EAGAIN); 1420 } 1421 1422 ufsvfsp->vfs_validfs = UT_UNMOUNTED; 1423 1424 /* kill the reclaim thread */ 1425 ufs_thread_exit(&ufsvfsp->vfs_reclaim); 1426 1427 /* suspend the delete thread */ 1428 ufs_thread_suspend(&ufsvfsp->vfs_delete); 1429 1430 /* 1431 * drain the delete and idle queues 1432 */ 1433 ufs_delete_drain(vfsp, -1, 1); 1434 ufs_idle_drain(vfsp); 1435 1436 /* 1437 * use the lockfs protocol to prevent new ops from starting 1438 * a forcible umount can not fail beyond this point as 1439 * we hard-locked the filesystem and drained all current consumers 1440 * before. 1441 */ 1442 mutex_enter(&ulp->ul_lock); 1443 1444 /* 1445 * if the file system is busy; return EBUSY 1446 */ 1447 if (ulp->ul_vnops_cnt || ulp->ul_falloc_cnt || ULOCKFS_IS_SLOCK(ulp)) { 1448 error = EBUSY; 1449 goto out; 1450 } 1451 1452 /* 1453 * if this is not a forced unmount (!hard/error locked), then 1454 * get rid of every inode except the root and quota inodes 1455 * also, commit any outstanding transactions 1456 */ 1457 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) 1458 if (error = ufs_flush(vfsp)) 1459 goto out; 1460 1461 /* 1462 * ignore inodes in the cache if fs is hard locked or error locked 1463 */ 1464 rip = VTOI(ufsvfsp->vfs_root); 1465 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) { 1466 /* 1467 * Otherwise, only the quota and root inodes are in the cache. 1468 * 1469 * Avoid racing with ufs_update() and ufs_sync(). 1470 */ 1471 mutex_enter(&ufs_scan_lock); 1472 1473 for (i = 0, ih = ihead; i < inohsz; i++, ih++) { 1474 mutex_enter(&ih_lock[i]); 1475 for (ip = ih->ih_chain[0]; 1476 ip != (struct inode *)ih; 1477 ip = ip->i_forw) { 1478 if (ip->i_ufsvfs != ufsvfsp) 1479 continue; 1480 if (ip == ufsvfsp->vfs_qinod) 1481 continue; 1482 if (ip == rip && ITOV(ip)->v_count == 1) 1483 continue; 1484 mutex_exit(&ih_lock[i]); 1485 mutex_exit(&ufs_scan_lock); 1486 error = EBUSY; 1487 goto out; 1488 } 1489 mutex_exit(&ih_lock[i]); 1490 } 1491 mutex_exit(&ufs_scan_lock); 1492 } 1493 1494 /* 1495 * if a snapshot exists and this is a forced unmount, then delete 1496 * the snapshot. Otherwise return EBUSY. This will insure the 1497 * snapshot always belongs to a valid file system. 1498 */ 1499 if (ufsvfsp->vfs_snapshot) { 1500 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) { 1501 (void) fssnap_delete(&ufsvfsp->vfs_snapshot); 1502 } else { 1503 error = EBUSY; 1504 goto out; 1505 } 1506 } 1507 1508 /* 1509 * Close the quota file and invalidate anything left in the quota 1510 * cache for this file system. Pass kcred to allow all quota 1511 * manipulations. 1512 */ 1513 (void) closedq(ufsvfsp, kcred); 1514 invalidatedq(ufsvfsp); 1515 /* 1516 * drain the delete and idle queues 1517 */ 1518 ufs_delete_drain(vfsp, -1, 0); 1519 ufs_idle_drain(vfsp); 1520 1521 /* 1522 * discard the inodes for this fs (including root, shadow, and quota) 1523 */ 1524 for (i = 0, ih = ihead; i < inohsz; i++, ih++) { 1525 mutex_enter(&ih_lock[i]); 1526 for (inext = 0, ip = ih->ih_chain[0]; 1527 ip != (struct inode *)ih; 1528 ip = inext) { 1529 inext = ip->i_forw; 1530 if (ip->i_ufsvfs != ufsvfsp) 1531 continue; 1532 1533 /* 1534 * We've found the inode in the cache and as we 1535 * hold the hash mutex the inode can not 1536 * disappear from underneath us. 1537 * We also know it must have at least a vnode 1538 * reference count of 1. 1539 * We perform an additional VN_HOLD so the VN_RELE 1540 * in case we take the inode off the idle queue 1541 * can not be the last one. 1542 * It is safe to grab the writer contents lock here 1543 * to prevent a race with ufs_iinactive() putting 1544 * inodes into the idle queue while we operate on 1545 * this inode. 1546 */ 1547 rw_enter(&ip->i_contents, RW_WRITER); 1548 1549 vp = ITOV(ip); 1550 VN_HOLD(vp) 1551 remque(ip); 1552 if (ufs_rmidle(ip)) 1553 VN_RELE(vp); 1554 ufs_si_del(ip); 1555 /* 1556 * rip->i_ufsvfsp is needed by bflush() 1557 */ 1558 if (ip != rip) 1559 ip->i_ufsvfs = NULL; 1560 /* 1561 * Set vnode's vfsops to dummy ops, which return 1562 * EIO. This is needed to forced unmounts to work 1563 * with lofs/nfs properly. 1564 */ 1565 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) 1566 vp->v_vfsp = &EIO_vfs; 1567 else 1568 vp->v_vfsp = NULL; 1569 vp->v_type = VBAD; 1570 1571 rw_exit(&ip->i_contents); 1572 1573 VN_RELE(vp); 1574 } 1575 mutex_exit(&ih_lock[i]); 1576 } 1577 ufs_si_cache_flush(dev); 1578 1579 /* 1580 * kill the delete thread and drain the idle queue 1581 */ 1582 ufs_thread_exit(&ufsvfsp->vfs_delete); 1583 ufs_idle_drain(vfsp); 1584 1585 bp = ufsvfsp->vfs_bufp; 1586 bvp = ufsvfsp->vfs_devvp; 1587 flag = !fs->fs_ronly; 1588 if (flag) { 1589 bflush(dev); 1590 if (fs->fs_clean != FSBAD) { 1591 if (fs->fs_clean == FSSTABLE) 1592 fs->fs_clean = FSCLEAN; 1593 fs->fs_reclaim &= ~FS_RECLAIM; 1594 } 1595 if (TRANS_ISTRANS(ufsvfsp) && 1596 !TRANS_ISERROR(ufsvfsp) && 1597 !ULOCKFS_IS_HLOCK(ulp) && 1598 (fs->fs_rolled == FS_NEED_ROLL)) { 1599 /* 1600 * ufs_flush() above has flushed the last Moby. 1601 * This is needed to ensure the following superblock 1602 * update really is the last metadata update 1603 */ 1604 error = ufs_putsummaryinfo(dev, ufsvfsp, fs); 1605 if (error == 0) { 1606 fs->fs_rolled = FS_ALL_ROLLED; 1607 } 1608 } 1609 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UNMOUNT); 1610 /* 1611 * push this last transaction 1612 */ 1613 curthread->t_flag |= T_DONTBLOCK; 1614 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE, 1615 error); 1616 if (!error) 1617 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UNMOUNT, 1618 TOP_COMMIT_SIZE); 1619 curthread->t_flag &= ~T_DONTBLOCK; 1620 } 1621 1622 TRANS_MATA_UMOUNT(ufsvfsp); 1623 lufs_unsnarf(ufsvfsp); /* Release the in-memory structs */ 1624 ufsfx_unmount(ufsvfsp); /* fix-on-panic bookkeeping */ 1625 kmem_free(fs->fs_u.fs_csp, fs->fs_cssize); 1626 1627 bp->b_flags |= B_STALE|B_AGE; 1628 ufsvfsp->vfs_bufp = NULL; /* don't point at freed buf */ 1629 brelse(bp); /* free the superblock buf */ 1630 1631 (void) VOP_PUTPAGE(common_specvp(bvp), (offset_t)0, (size_t)0, 1632 B_INVAL, cr); 1633 (void) VOP_CLOSE(bvp, flag, 1, (offset_t)0, cr); 1634 bflush(dev); 1635 (void) bfinval(dev, 1); 1636 VN_RELE(bvp); 1637 1638 /* 1639 * It is now safe to NULL out the ufsvfs pointer and discard 1640 * the root inode. 1641 */ 1642 rip->i_ufsvfs = NULL; 1643 VN_RELE(ITOV(rip)); 1644 1645 /* free up lockfs comment structure, if any */ 1646 if (ulp->ul_lockfs.lf_comlen && ulp->ul_lockfs.lf_comment) 1647 kmem_free(ulp->ul_lockfs.lf_comment, ulp->ul_lockfs.lf_comlen); 1648 1649 /* 1650 * Remove from instance list. 1651 */ 1652 ufs_vfs_remove(ufsvfsp); 1653 1654 /* 1655 * For a forcible unmount, threads may be asleep in 1656 * ufs_lockfs_begin/ufs_check_lockfs. These threads will need 1657 * the ufsvfs structure so we don't free it, yet. ufs_update 1658 * will free it up after awhile. 1659 */ 1660 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) { 1661 extern kmutex_t ufsvfs_mutex; 1662 extern struct ufsvfs *ufsvfslist; 1663 1664 mutex_enter(&ufsvfs_mutex); 1665 ufsvfsp->vfs_dontblock = 1; 1666 ufsvfsp->vfs_next = ufsvfslist; 1667 ufsvfslist = ufsvfsp; 1668 mutex_exit(&ufsvfs_mutex); 1669 /* wakeup any suspended threads */ 1670 cv_broadcast(&ulp->ul_cv); 1671 mutex_exit(&ulp->ul_lock); 1672 } else { 1673 mutex_destroy(&ufsvfsp->vfs_lock); 1674 kmem_free(ufsvfsp, sizeof (struct ufsvfs)); 1675 } 1676 1677 /* 1678 * Now mark the filesystem as unmounted since we're done with it. 1679 */ 1680 vfsp->vfs_flag |= VFS_UNMOUNTED; 1681 1682 return (0); 1683 out: 1684 /* open the fs to new ops */ 1685 cv_broadcast(&ulp->ul_cv); 1686 mutex_exit(&ulp->ul_lock); 1687 1688 if (TRANS_ISTRANS(ufsvfsp)) { 1689 /* allow the delete thread to continue */ 1690 ufs_thread_continue(&ufsvfsp->vfs_delete); 1691 /* restart the reclaim thread */ 1692 ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim, 1693 vfsp); 1694 /* coordinate with global hlock thread */ 1695 ufsvfsp->vfs_validfs = UT_MOUNTED; 1696 /* check for trans errors during umount */ 1697 ufs_trans_onerror(); 1698 1699 /* 1700 * if we have a seperate /usr it will never unmount 1701 * when halting. In order to not re-read all the 1702 * cylinder group summary info on mounting after 1703 * reboot the logging of summary info is re-enabled 1704 * and the super block written out. 1705 */ 1706 mountpoint = vfs_getmntpoint(vfsp); 1707 if ((fs->fs_si == FS_SI_OK) && 1708 (strcmp("/usr", refstr_value(mountpoint)) == 0)) { 1709 ufsvfsp->vfs_nolog_si = 0; 1710 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp); 1711 } 1712 refstr_rele(mountpoint); 1713 } 1714 1715 return (error); 1716 } 1717 1718 static int 1719 ufs_root(struct vfs *vfsp, struct vnode **vpp) 1720 { 1721 struct ufsvfs *ufsvfsp; 1722 struct vnode *vp; 1723 1724 if (!vfsp) 1725 return (EIO); 1726 1727 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1728 if (!ufsvfsp || !ufsvfsp->vfs_root) 1729 return (EIO); /* forced unmount */ 1730 1731 vp = ufsvfsp->vfs_root; 1732 VN_HOLD(vp); 1733 *vpp = vp; 1734 return (0); 1735 } 1736 1737 /* 1738 * Get file system statistics. 1739 */ 1740 static int 1741 ufs_statvfs(struct vfs *vfsp, struct statvfs64 *sp) 1742 { 1743 struct fs *fsp; 1744 struct ufsvfs *ufsvfsp; 1745 int blk, i; 1746 long max_avail, used; 1747 dev32_t d32; 1748 1749 if (vfsp->vfs_flag & VFS_UNMOUNTED) 1750 return (EIO); 1751 1752 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1753 fsp = ufsvfsp->vfs_fs; 1754 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) 1755 return (EINVAL); 1756 if (fsp->fs_magic == FS_MAGIC && 1757 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 && 1758 fsp->fs_version != UFS_VERSION_MIN)) 1759 return (EINVAL); 1760 if (fsp->fs_magic == MTB_UFS_MAGIC && 1761 (fsp->fs_version > MTB_UFS_VERSION_1 || 1762 fsp->fs_version < MTB_UFS_VERSION_MIN)) 1763 return (EINVAL); 1764 1765 /* 1766 * get the basic numbers 1767 */ 1768 (void) bzero(sp, sizeof (*sp)); 1769 1770 sp->f_bsize = fsp->fs_bsize; 1771 sp->f_frsize = fsp->fs_fsize; 1772 sp->f_blocks = (fsblkcnt64_t)fsp->fs_dsize; 1773 sp->f_bfree = (fsblkcnt64_t)fsp->fs_cstotal.cs_nbfree * fsp->fs_frag + 1774 fsp->fs_cstotal.cs_nffree; 1775 1776 sp->f_files = (fsfilcnt64_t)fsp->fs_ncg * fsp->fs_ipg; 1777 sp->f_ffree = (fsfilcnt64_t)fsp->fs_cstotal.cs_nifree; 1778 1779 /* 1780 * Adjust the numbers based on things waiting to be deleted. 1781 * modifies f_bfree and f_ffree. Afterwards, everything we 1782 * come up with will be self-consistent. By definition, this 1783 * is a point-in-time snapshot, so the fact that the delete 1784 * thread's probably already invalidated the results is not a 1785 * problem. Note that if the delete thread is ever extended to 1786 * non-logging ufs, this adjustment must always be made. 1787 */ 1788 if (TRANS_ISTRANS(ufsvfsp)) 1789 ufs_delete_adjust_stats(ufsvfsp, sp); 1790 1791 /* 1792 * avail = MAX(max_avail - used, 0) 1793 */ 1794 max_avail = fsp->fs_dsize - ufsvfsp->vfs_minfrags; 1795 1796 used = (fsp->fs_dsize - sp->f_bfree); 1797 1798 if (max_avail > used) 1799 sp->f_bavail = (fsblkcnt64_t)max_avail - used; 1800 else 1801 sp->f_bavail = (fsblkcnt64_t)0; 1802 1803 sp->f_favail = sp->f_ffree; 1804 (void) cmpldev(&d32, vfsp->vfs_dev); 1805 sp->f_fsid = d32; 1806 (void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name); 1807 sp->f_flag = vf_to_stf(vfsp->vfs_flag); 1808 1809 /* keep coordinated with ufs_l_pathconf() */ 1810 sp->f_namemax = MAXNAMLEN; 1811 1812 if (fsp->fs_cpc == 0) { 1813 bzero(sp->f_fstr, 14); 1814 return (0); 1815 } 1816 blk = fsp->fs_spc * fsp->fs_cpc / NSPF(fsp); 1817 for (i = 0; i < blk; i += fsp->fs_frag) /* CSTYLED */ 1818 /* void */; 1819 i -= fsp->fs_frag; 1820 blk = i / fsp->fs_frag; 1821 bcopy(&(fs_rotbl(fsp)[blk]), sp->f_fstr, 14); 1822 return (0); 1823 } 1824 1825 /* 1826 * Flush any pending I/O to file system vfsp. 1827 * The ufs_update() routine will only flush *all* ufs files. 1828 * If vfsp is non-NULL, only sync this ufs (in preparation 1829 * for a umount). 1830 */ 1831 /*ARGSUSED*/ 1832 static int 1833 ufs_sync(struct vfs *vfsp, short flag, struct cred *cr) 1834 { 1835 struct ufsvfs *ufsvfsp; 1836 struct fs *fs; 1837 int cheap = flag & SYNC_ATTR; 1838 int error; 1839 1840 /* 1841 * SYNC_CLOSE means we're rebooting. Toss everything 1842 * on the idle queue so we don't have to slog through 1843 * a bunch of uninteresting inodes over and over again. 1844 */ 1845 if (flag & SYNC_CLOSE) 1846 ufs_idle_drain(NULL); 1847 1848 if (vfsp == NULL) { 1849 ufs_update(flag); 1850 return (0); 1851 } 1852 1853 /* Flush a single ufs */ 1854 if (!vfs_matchops(vfsp, ufs_vfsops) || vfs_lock(vfsp) != 0) 1855 return (0); 1856 1857 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1858 if (!ufsvfsp) 1859 return (EIO); 1860 fs = ufsvfsp->vfs_fs; 1861 mutex_enter(&ufsvfsp->vfs_lock); 1862 1863 if (ufsvfsp->vfs_dio && 1864 fs->fs_ronly == 0 && 1865 fs->fs_clean != FSBAD && 1866 fs->fs_clean != FSLOG) { 1867 /* turn off fast-io on unmount, so no fsck needed (4029401) */ 1868 ufsvfsp->vfs_dio = 0; 1869 fs->fs_clean = FSACTIVE; 1870 fs->fs_fmod = 1; 1871 } 1872 1873 /* Write back modified superblock */ 1874 if (fs->fs_fmod == 0) { 1875 mutex_exit(&ufsvfsp->vfs_lock); 1876 } else { 1877 if (fs->fs_ronly != 0) { 1878 mutex_exit(&ufsvfsp->vfs_lock); 1879 vfs_unlock(vfsp); 1880 return (ufs_fault(ufsvfsp->vfs_root, 1881 "fs = %s update: ro fs mod\n", 1882 fs->fs_fsmnt)); 1883 } 1884 fs->fs_fmod = 0; 1885 mutex_exit(&ufsvfsp->vfs_lock); 1886 1887 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UPDATE); 1888 } 1889 vfs_unlock(vfsp); 1890 1891 /* 1892 * Avoid racing with ufs_update() and ufs_unmount(). 1893 * 1894 */ 1895 mutex_enter(&ufs_scan_lock); 1896 1897 (void) ufs_scan_inodes(1, ufs_sync_inode, 1898 (void *)(uintptr_t)cheap, ufsvfsp); 1899 1900 mutex_exit(&ufs_scan_lock); 1901 1902 bflush((dev_t)vfsp->vfs_dev); 1903 1904 /* 1905 * commit any outstanding async transactions 1906 */ 1907 curthread->t_flag |= T_DONTBLOCK; 1908 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE, error); 1909 if (!error) { 1910 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UPDATE, 1911 TOP_COMMIT_SIZE); 1912 } 1913 curthread->t_flag &= ~T_DONTBLOCK; 1914 1915 return (0); 1916 } 1917 1918 1919 void 1920 sbupdate(struct vfs *vfsp) 1921 { 1922 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1923 struct fs *fs = ufsvfsp->vfs_fs; 1924 struct buf *bp; 1925 int blks; 1926 caddr_t space; 1927 int i; 1928 size_t size; 1929 1930 /* 1931 * for ulockfs processing, limit the superblock writes 1932 */ 1933 if ((ufsvfsp->vfs_ulockfs.ul_sbowner) && 1934 (curthread != ufsvfsp->vfs_ulockfs.ul_sbowner)) { 1935 /* process later */ 1936 fs->fs_fmod = 1; 1937 return; 1938 } 1939 ULOCKFS_SET_MOD((&ufsvfsp->vfs_ulockfs)); 1940 1941 if (TRANS_ISTRANS(ufsvfsp)) { 1942 mutex_enter(&ufsvfsp->vfs_lock); 1943 ufs_sbwrite(ufsvfsp); 1944 mutex_exit(&ufsvfsp->vfs_lock); 1945 return; 1946 } 1947 1948 blks = howmany(fs->fs_cssize, fs->fs_fsize); 1949 space = (caddr_t)fs->fs_u.fs_csp; 1950 for (i = 0; i < blks; i += fs->fs_frag) { 1951 size = fs->fs_bsize; 1952 if (i + fs->fs_frag > blks) 1953 size = (blks - i) * fs->fs_fsize; 1954 bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev, 1955 (daddr_t)(fsbtodb(fs, fs->fs_csaddr + i)), 1956 fs->fs_bsize); 1957 bcopy(space, bp->b_un.b_addr, size); 1958 space += size; 1959 bp->b_bcount = size; 1960 UFS_BRWRITE(ufsvfsp, bp); 1961 } 1962 mutex_enter(&ufsvfsp->vfs_lock); 1963 ufs_sbwrite(ufsvfsp); 1964 mutex_exit(&ufsvfsp->vfs_lock); 1965 } 1966 1967 int ufs_vget_idle_count = 2; /* Number of inodes to idle each time */ 1968 static int 1969 ufs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp) 1970 { 1971 int error = 0; 1972 struct ufid *ufid; 1973 struct inode *ip; 1974 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 1975 struct ulockfs *ulp; 1976 1977 /* 1978 * Check for unmounted filesystem. 1979 */ 1980 if (vfsp->vfs_flag & VFS_UNMOUNTED) { 1981 error = EIO; 1982 goto errout; 1983 } 1984 1985 /* 1986 * Keep the idle queue from getting too long by 1987 * idling an inode before attempting to allocate another. 1988 * This operation must be performed before entering 1989 * lockfs or a transaction. 1990 */ 1991 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat) 1992 if ((curthread->t_flag & T_DONTBLOCK) == 0) { 1993 ins.in_vidles.value.ul += ufs_vget_idle_count; 1994 ufs_idle_some(ufs_vget_idle_count); 1995 } 1996 1997 ufid = (struct ufid *)fidp; 1998 1999 if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_VGET_MASK)) 2000 goto errout; 2001 2002 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 2003 2004 error = ufs_iget(vfsp, ufid->ufid_ino, &ip, CRED()); 2005 2006 rw_exit(&ufsvfsp->vfs_dqrwlock); 2007 2008 ufs_lockfs_end(ulp); 2009 2010 if (error) 2011 goto errout; 2012 2013 /* 2014 * Check if the inode has been deleted or freed or is in transient state 2015 * since the last VFS_VGET() request for it, release it and don't return 2016 * it to the caller, presumably NFS, as it's no longer valid. 2017 */ 2018 if (ip->i_gen != ufid->ufid_gen || ip->i_mode == 0 || 2019 (ip->i_flag & IDEL)) { 2020 VN_RELE(ITOV(ip)); 2021 error = EINVAL; 2022 goto errout; 2023 } 2024 2025 *vpp = ITOV(ip); 2026 return (0); 2027 2028 errout: 2029 *vpp = NULL; 2030 return (error); 2031 } 2032 2033 static int 2034 ufsinit(int fstype, char *name) 2035 { 2036 static const fs_operation_def_t ufs_vfsops_template[] = { 2037 VFSNAME_MOUNT, { .vfs_mount = ufs_mount }, 2038 VFSNAME_UNMOUNT, { .vfs_unmount = ufs_unmount }, 2039 VFSNAME_ROOT, { .vfs_root = ufs_root }, 2040 VFSNAME_STATVFS, { .vfs_statvfs = ufs_statvfs }, 2041 VFSNAME_SYNC, { .vfs_sync = ufs_sync }, 2042 VFSNAME_VGET, { .vfs_vget = ufs_vget }, 2043 VFSNAME_MOUNTROOT, { .vfs_mountroot = ufs_mountroot }, 2044 NULL, NULL 2045 }; 2046 int error; 2047 2048 ufsfstype = fstype; 2049 2050 error = vfs_setfsops(fstype, ufs_vfsops_template, &ufs_vfsops); 2051 if (error != 0) { 2052 cmn_err(CE_WARN, "ufsinit: bad vfs ops template"); 2053 return (error); 2054 } 2055 2056 error = vn_make_ops(name, ufs_vnodeops_template, &ufs_vnodeops); 2057 if (error != 0) { 2058 (void) vfs_freevfsops_by_type(fstype); 2059 cmn_err(CE_WARN, "ufsinit: bad vnode ops template"); 2060 return (error); 2061 } 2062 2063 ufs_iinit(); 2064 return (0); 2065 } 2066 2067 #ifdef __sparc 2068 2069 /* 2070 * Mounting a mirrored SVM volume is only supported on ufs, 2071 * this is special-case boot code to support that configuration. 2072 * At this point, we have booted and mounted root on a 2073 * single component of the mirror. Complete the boot 2074 * by configuring SVM and converting the root to the 2075 * dev_t of the mirrored root device. This dev_t conversion 2076 * only works because the underlying device doesn't change. 2077 */ 2078 int 2079 ufs_remountroot(struct vfs *vfsp) 2080 { 2081 struct ufsvfs *ufsvfsp; 2082 struct ulockfs *ulp; 2083 dev_t new_rootdev; 2084 dev_t old_rootdev; 2085 struct vnode *old_rootvp; 2086 struct vnode *new_rootvp; 2087 int error, sberror = 0; 2088 struct inode *ip; 2089 union ihead *ih; 2090 struct buf *bp; 2091 int i; 2092 2093 old_rootdev = rootdev; 2094 old_rootvp = rootvp; 2095 2096 new_rootdev = getrootdev(); 2097 if (new_rootdev == (dev_t)NODEV) { 2098 return (ENODEV); 2099 } 2100 2101 new_rootvp = makespecvp(new_rootdev, VBLK); 2102 2103 error = VOP_OPEN(&new_rootvp, 2104 (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, CRED()); 2105 if (error) { 2106 cmn_err(CE_CONT, 2107 "Cannot open mirrored root device, error %d\n", error); 2108 return (error); 2109 } 2110 2111 if (vfs_lock(vfsp) != 0) { 2112 return (EBUSY); 2113 } 2114 2115 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data; 2116 ulp = &ufsvfsp->vfs_ulockfs; 2117 2118 mutex_enter(&ulp->ul_lock); 2119 atomic_add_long(&ufs_quiesce_pend, 1); 2120 2121 (void) ufs_quiesce(ulp); 2122 (void) ufs_flush(vfsp); 2123 2124 /* 2125 * Convert root vfs to new dev_t, including vfs hash 2126 * table and fs id. 2127 */ 2128 vfs_root_redev(vfsp, new_rootdev, ufsfstype); 2129 2130 ufsvfsp->vfs_devvp = new_rootvp; 2131 ufsvfsp->vfs_dev = new_rootdev; 2132 2133 bp = ufsvfsp->vfs_bufp; 2134 bp->b_edev = new_rootdev; 2135 bp->b_dev = cmpdev(new_rootdev); 2136 2137 /* 2138 * The buffer for the root inode does not contain a valid b_vp 2139 */ 2140 (void) bfinval(new_rootdev, 0); 2141 2142 /* 2143 * Here we hand-craft inodes with old root device 2144 * references to refer to the new device instead. 2145 */ 2146 mutex_enter(&ufs_scan_lock); 2147 2148 for (i = 0, ih = ihead; i < inohsz; i++, ih++) { 2149 mutex_enter(&ih_lock[i]); 2150 for (ip = ih->ih_chain[0]; 2151 ip != (struct inode *)ih; 2152 ip = ip->i_forw) { 2153 if (ip->i_ufsvfs != ufsvfsp) 2154 continue; 2155 if (ip == ufsvfsp->vfs_qinod) 2156 continue; 2157 if (ip->i_dev == old_rootdev) { 2158 ip->i_dev = new_rootdev; 2159 } 2160 2161 if (ip->i_devvp == old_rootvp) { 2162 ip->i_devvp = new_rootvp; 2163 } 2164 } 2165 mutex_exit(&ih_lock[i]); 2166 } 2167 2168 mutex_exit(&ufs_scan_lock); 2169 2170 /* 2171 * Make Sure logging structures are using the new device 2172 * if logging is enabled. Also start any logging thread that 2173 * needs to write to the device and couldn't earlier. 2174 */ 2175 if (ufsvfsp->vfs_log) { 2176 buf_t *bp, *tbp; 2177 ml_unit_t *ul = ufsvfsp->vfs_log; 2178 struct fs *fsp = ufsvfsp->vfs_fs; 2179 2180 /* 2181 * Update the main logging structure. 2182 */ 2183 ul->un_dev = new_rootdev; 2184 2185 /* 2186 * Get a new bp for the on disk structures. 2187 */ 2188 bp = ul->un_bp; 2189 tbp = ngeteblk(dbtob(LS_SECTORS)); 2190 tbp->b_edev = new_rootdev; 2191 tbp->b_dev = cmpdev(new_rootdev); 2192 tbp->b_blkno = bp->b_blkno; 2193 bcopy(bp->b_un.b_addr, tbp->b_un.b_addr, DEV_BSIZE); 2194 bcopy(bp->b_un.b_addr, tbp->b_un.b_addr + DEV_BSIZE, DEV_BSIZE); 2195 bp->b_flags |= (B_STALE | B_AGE); 2196 brelse(bp); 2197 ul->un_bp = tbp; 2198 2199 /* 2200 * Allocate new circular buffers. 2201 */ 2202 alloc_rdbuf(&ul->un_rdbuf, MAPBLOCKSIZE, MAPBLOCKSIZE); 2203 alloc_wrbuf(&ul->un_wrbuf, ldl_bufsize(ul)); 2204 2205 /* 2206 * Clear the noroll bit which indicates that logging 2207 * can't roll the log yet and start the logmap roll thread 2208 * unless the filesystem is still read-only in which case 2209 * remountfs() will do it when going to read-write. 2210 */ 2211 ASSERT(ul->un_flags & LDL_NOROLL); 2212 2213 if (!fsp->fs_ronly) { 2214 ul->un_flags &= ~LDL_NOROLL; 2215 logmap_start_roll(ul); 2216 } 2217 2218 /* 2219 * Start the reclaim thread if needed. 2220 */ 2221 if (!fsp->fs_ronly && (fsp->fs_reclaim & 2222 (FS_RECLAIM|FS_RECLAIMING))) { 2223 fsp->fs_reclaim &= ~FS_RECLAIM; 2224 fsp->fs_reclaim |= FS_RECLAIMING; 2225 ufs_thread_start(&ufsvfsp->vfs_reclaim, 2226 ufs_thread_reclaim, vfsp); 2227 TRANS_SBWRITE(ufsvfsp, TOP_SBUPDATE_UPDATE); 2228 if (sberror = geterror(ufsvfsp->vfs_bufp)) { 2229 refstr_t *mntpt; 2230 mntpt = vfs_getmntpoint(vfsp); 2231 cmn_err(CE_WARN, 2232 "Remountroot failed to update Reclaim" 2233 "state for filesystem %s " 2234 "Error writing SuperBlock %d", 2235 refstr_value(mntpt), error); 2236 refstr_rele(mntpt); 2237 } 2238 } 2239 } 2240 2241 rootdev = new_rootdev; 2242 rootvp = new_rootvp; 2243 2244 atomic_add_long(&ufs_quiesce_pend, -1); 2245 cv_broadcast(&ulp->ul_cv); 2246 mutex_exit(&ulp->ul_lock); 2247 2248 vfs_unlock(vfsp); 2249 2250 error = VOP_CLOSE(old_rootvp, FREAD, 1, (offset_t)0, CRED()); 2251 if (error) { 2252 cmn_err(CE_CONT, 2253 "close of root device component failed, error %d\n", 2254 error); 2255 } 2256 VN_RELE(old_rootvp); 2257 2258 return (sberror ? sberror : error); 2259 } 2260 2261 #endif /* __sparc */ 2262