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 2009 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 #include <sys/types.h>
40 #include <sys/t_lock.h>
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/bitmap.h>
44 #include <sys/sysmacros.h>
45 #include <sys/kmem.h>
46 #include <sys/signal.h>
47 #include <sys/user.h>
48 #include <sys/proc.h>
49 #include <sys/disp.h>
50 #include <sys/buf.h>
51 #include <sys/pathname.h>
52 #include <sys/vfs.h>
53 #include <sys/vfs_opreg.h>
54 #include <sys/vnode.h>
55 #include <sys/file.h>
56 #include <sys/atomic.h>
57 #include <sys/uio.h>
58 #include <sys/dkio.h>
59 #include <sys/cred.h>
60 #include <sys/conf.h>
61 #include <sys/dnlc.h>
62 #include <sys/kstat.h>
63 #include <sys/acl.h>
64 #include <sys/fs/ufs_fsdir.h>
65 #include <sys/fs/ufs_fs.h>
66 #include <sys/fs/ufs_inode.h>
67 #include <sys/fs/ufs_mount.h>
68 #include <sys/fs/ufs_acl.h>
69 #include <sys/fs/ufs_panic.h>
70 #include <sys/fs/ufs_bio.h>
71 #include <sys/fs/ufs_quota.h>
72 #include <sys/fs/ufs_log.h>
73 #undef NFS
74 #include <sys/statvfs.h>
75 #include <sys/mount.h>
76 #include <sys/mntent.h>
77 #include <sys/swap.h>
78 #include <sys/errno.h>
79 #include <sys/debug.h>
80 #include "fs/fs_subr.h"
81 #include <sys/cmn_err.h>
82 #include <sys/dnlc.h>
83 #include <sys/fssnap_if.h>
84 #include <sys/sunddi.h>
85 #include <sys/bootconf.h>
86 #include <sys/policy.h>
87 #include <sys/zone.h>
88
89 /*
90 * This is the loadable module wrapper.
91 */
92 #include <sys/modctl.h>
93
94 int ufsfstype;
95 vfsops_t *ufs_vfsops;
96 static int ufsinit(int, char *);
97 static int mountfs();
98 extern int highbit();
99 extern struct instats ins;
100 extern struct vnode *common_specvp(struct vnode *vp);
101 extern vfs_t EIO_vfs;
102
103 struct dquot *dquot, *dquotNDQUOT;
104
105 /*
106 * Cylinder group summary information handling tunable.
107 * This defines when these deltas get logged.
108 * If the number of cylinders in the file system is over the
109 * tunable then we log csum updates. Otherwise the updates are only
110 * done for performance on unmount. After a panic they can be
111 * quickly constructed during mounting. See ufs_construct_si()
112 * called from ufs_getsummaryinfo().
113 *
114 * This performance feature can of course be disabled by setting
115 * ufs_ncg_log to 0, and fully enabled by setting it to 0xffffffff.
116 */
117 #define UFS_LOG_NCG_DEFAULT 10000
118 uint32_t ufs_ncg_log = UFS_LOG_NCG_DEFAULT;
119
120 /*
121 * ufs_clean_root indicates whether the root fs went down cleanly
122 */
123 static int ufs_clean_root = 0;
124
125 /*
126 * UFS Mount options table
127 */
128 static char *intr_cancel[] = { MNTOPT_NOINTR, NULL };
129 static char *nointr_cancel[] = { MNTOPT_INTR, NULL };
130 static char *forcedirectio_cancel[] = { MNTOPT_NOFORCEDIRECTIO, NULL };
131 static char *noforcedirectio_cancel[] = { MNTOPT_FORCEDIRECTIO, NULL };
132 static char *largefiles_cancel[] = { MNTOPT_NOLARGEFILES, NULL };
133 static char *nolargefiles_cancel[] = { MNTOPT_LARGEFILES, NULL };
134 static char *logging_cancel[] = { MNTOPT_NOLOGGING, NULL };
135 static char *nologging_cancel[] = { MNTOPT_LOGGING, NULL };
136 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
137 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
138 static char *quota_cancel[] = { MNTOPT_NOQUOTA, NULL };
139 static char *noquota_cancel[] = { MNTOPT_QUOTA, NULL };
140 static char *dfratime_cancel[] = { MNTOPT_NODFRATIME, NULL };
141 static char *nodfratime_cancel[] = { MNTOPT_DFRATIME, NULL };
142
143 static mntopt_t mntopts[] = {
144 /*
145 * option name cancel option default arg flags
146 * ufs arg flag
147 */
148 { MNTOPT_INTR, intr_cancel, NULL, MO_DEFAULT,
149 (void *)0 },
150 { MNTOPT_NOINTR, nointr_cancel, NULL, 0,
151 (void *)UFSMNT_NOINTR },
152 { MNTOPT_SYNCDIR, NULL, NULL, 0,
153 (void *)UFSMNT_SYNCDIR },
154 { MNTOPT_FORCEDIRECTIO, forcedirectio_cancel, NULL, 0,
155 (void *)UFSMNT_FORCEDIRECTIO },
156 { MNTOPT_NOFORCEDIRECTIO, noforcedirectio_cancel, NULL, 0,
157 (void *)UFSMNT_NOFORCEDIRECTIO },
158 { MNTOPT_NOSETSEC, NULL, NULL, 0,
159 (void *)UFSMNT_NOSETSEC },
160 { MNTOPT_LARGEFILES, largefiles_cancel, NULL, MO_DEFAULT,
161 (void *)UFSMNT_LARGEFILES },
162 { MNTOPT_NOLARGEFILES, nolargefiles_cancel, NULL, 0,
163 (void *)0 },
164 { MNTOPT_LOGGING, logging_cancel, NULL, MO_TAG,
165 (void *)UFSMNT_LOGGING },
166 { MNTOPT_NOLOGGING, nologging_cancel, NULL,
167 MO_NODISPLAY|MO_DEFAULT|MO_TAG, (void *)0 },
168 { MNTOPT_QUOTA, quota_cancel, NULL, MO_IGNORE,
169 (void *)0 },
170 { MNTOPT_NOQUOTA, noquota_cancel, NULL,
171 MO_NODISPLAY|MO_DEFAULT, (void *)0 },
172 { MNTOPT_GLOBAL, NULL, NULL, 0,
173 (void *)0 },
174 { MNTOPT_XATTR, xattr_cancel, NULL, MO_DEFAULT,
175 (void *)0 },
176 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0,
177 (void *)0 },
178 { MNTOPT_NOATIME, NULL, NULL, 0,
179 (void *)UFSMNT_NOATIME },
180 { MNTOPT_DFRATIME, dfratime_cancel, NULL, 0,
181 (void *)0 },
182 { MNTOPT_NODFRATIME, nodfratime_cancel, NULL,
183 MO_NODISPLAY|MO_DEFAULT, (void *)UFSMNT_NODFRATIME },
184 { MNTOPT_ONERROR, NULL, UFSMNT_ONERROR_PANIC_STR,
185 MO_DEFAULT|MO_HASVALUE, (void *)0 },
186 };
187
188 static mntopts_t ufs_mntopts = {
189 sizeof (mntopts) / sizeof (mntopt_t),
190 mntopts
191 };
192
193 static vfsdef_t vfw = {
194 VFSDEF_VERSION,
195 "ufs",
196 ufsinit,
197 VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_CANLOFI,
198 &ufs_mntopts
199 };
200
201 /*
202 * Module linkage information for the kernel.
203 */
204 extern struct mod_ops mod_fsops;
205
206 static struct modlfs modlfs = {
207 &mod_fsops, "filesystem for ufs", &vfw
208 };
209
210 static struct modlinkage modlinkage = {
211 MODREV_1, { (void *)&modlfs, NULL }
212 };
213
214 /*
215 * An attempt has been made to make this module unloadable. In order to
216 * test it, we need a system in which the root fs is NOT ufs. THIS HAS NOT
217 * BEEN DONE
218 */
219
220 extern kstat_t *ufs_inode_kstat;
221 extern uint_t ufs_lockfs_key;
222 extern void ufs_lockfs_tsd_destructor(void *);
223 extern uint_t bypass_snapshot_throttle_key;
224
225 int
226 _init(void)
227 {
228 /*
229 * Create an index into the per thread array so that any thread doing
230 * VOP will have a lockfs mark on it.
231 */
232 tsd_create(&ufs_lockfs_key, ufs_lockfs_tsd_destructor);
233 tsd_create(&bypass_snapshot_throttle_key, NULL);
234 return (mod_install(&modlinkage));
235 }
236
237 int
238 _fini(void)
239 {
240 return (EBUSY);
241 }
242
243 int
244 _info(struct modinfo *modinfop)
245 {
246 return (mod_info(&modlinkage, modinfop));
247 }
248
249 extern struct vnode *makespecvp(dev_t dev, vtype_t type);
250
251 extern kmutex_t ufs_scan_lock;
252
253 static int mountfs(struct vfs *, enum whymountroot, struct vnode *, char *,
254 struct cred *, int, void *, int);
255
256
257 static int
258 ufs_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap,
259 struct cred *cr)
260
261 {
262 char *data = uap->dataptr;
263 int datalen = uap->datalen;
264 dev_t dev;
265 struct vnode *lvp = NULL;
266 struct vnode *svp = NULL;
267 struct pathname dpn;
268 int error;
269 enum whymountroot why = ROOT_INIT;
270 struct ufs_args args;
271 int oflag, aflag;
272 int fromspace = (uap->flags & MS_SYSSPACE) ?
273 UIO_SYSSPACE : UIO_USERSPACE;
274
275 if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
276 return (error);
277
278 if (mvp->v_type != VDIR)
279 return (ENOTDIR);
280
281 mutex_enter(&mvp->v_lock);
282 if ((uap->flags & MS_REMOUNT) == 0 &&
283 (uap->flags & MS_OVERLAY) == 0 &&
284 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
285 mutex_exit(&mvp->v_lock);
286 return (EBUSY);
287 }
288 mutex_exit(&mvp->v_lock);
289
290 /*
291 * Get arguments
292 */
293 bzero(&args, sizeof (args));
294 if ((uap->flags & MS_DATA) && data != NULL && datalen != 0) {
295 int copy_result = 0;
296
297 if (datalen > sizeof (args))
298 return (EINVAL);
299 if (uap->flags & MS_SYSSPACE)
300 bcopy(data, &args, datalen);
301 else
302 copy_result = copyin(data, &args, datalen);
303 if (copy_result)
304 return (EFAULT);
305 datalen = sizeof (struct ufs_args);
306 } else {
307 datalen = 0;
308 }
309
310 if ((vfsp->vfs_flag & VFS_RDONLY) != 0 ||
311 (uap->flags & MS_RDONLY) != 0) {
312 oflag = FREAD;
313 aflag = VREAD;
314 } else {
315 oflag = FREAD | FWRITE;
316 aflag = VREAD | VWRITE;
317 }
318
319 /*
320 * Read in the mount point pathname
321 * (so we can record the directory the file system was last mounted on).
322 */
323 if (error = pn_get(uap->dir, fromspace, &dpn))
324 return (error);
325
326 /*
327 * Resolve path name of special file being mounted.
328 */
329 if (error = lookupname(uap->spec, fromspace, FOLLOW, NULL, &svp)) {
330 pn_free(&dpn);
331 return (error);
332 }
333
334 error = vfs_get_lofi(vfsp, &lvp);
335
336 if (error > 0) {
337 VN_RELE(svp);
338 pn_free(&dpn);
339 return (error);
340 } else if (error == 0) {
341 dev = lvp->v_rdev;
342
343 if (getmajor(dev) >= devcnt) {
344 error = ENXIO;
345 goto out;
346 }
347 } else {
348 dev = svp->v_rdev;
349
350 if (svp->v_type != VBLK) {
351 VN_RELE(svp);
352 pn_free(&dpn);
353 return (ENOTBLK);
354 }
355
356 if (getmajor(dev) >= devcnt) {
357 error = ENXIO;
358 goto out;
359 }
360
361 /*
362 * In SunCluster, requests to a global device are
363 * satisfied by a local device. We substitute the global
364 * pxfs node with a local spec node here.
365 */
366 if (IS_PXFSVP(svp)) {
367 ASSERT(lvp == NULL);
368 VN_RELE(svp);
369 svp = makespecvp(dev, VBLK);
370 }
371
372 if ((error = secpolicy_spec_open(cr, svp, oflag)) != 0) {
373 VN_RELE(svp);
374 pn_free(&dpn);
375 return (error);
376 }
377 }
378
379 if (uap->flags & MS_REMOUNT)
380 why = ROOT_REMOUNT;
381
382 /*
383 * Open device/file mounted on. We need this to check whether
384 * the caller has sufficient rights to access the resource in
385 * question. When bio is fixed for vnodes this can all be vnode
386 * operations.
387 */
388 if ((error = VOP_ACCESS(svp, aflag, 0, cr, NULL)) != 0)
389 goto out;
390
391 /*
392 * Ensure that this device isn't already mounted or in progress on a
393 * mount unless this is a REMOUNT request or we are told to suppress
394 * mount checks. Global mounts require special handling.
395 */
396 if ((uap->flags & MS_NOCHECK) == 0) {
397 if ((uap->flags & MS_GLOBAL) == 0 &&
398 vfs_devmounting(dev, vfsp)) {
399 error = EBUSY;
400 goto out;
401 }
402 if (vfs_devismounted(dev)) {
403 if ((uap->flags & MS_REMOUNT) == 0) {
404 error = EBUSY;
405 goto out;
406 }
407 }
408 }
409
410 /*
411 * If the device is a tape, mount it read only
412 */
413 if (devopsp[getmajor(dev)]->devo_cb_ops->cb_flag & D_TAPE) {
414 vfsp->vfs_flag |= VFS_RDONLY;
415 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
416 }
417 if (uap->flags & MS_RDONLY)
418 vfsp->vfs_flag |= VFS_RDONLY;
419
420 /*
421 * Mount the filesystem, free the device vnode on error.
422 */
423 error = mountfs(vfsp, why, lvp != NULL ? lvp : svp,
424 dpn.pn_path, cr, 0, &args, datalen);
425
426 if (error == 0) {
427 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS);
428
429 /*
430 * If lofi, drop our reference to the original file.
431 */
432 if (lvp != NULL)
433 VN_RELE(svp);
434 }
435
436 out:
437 pn_free(&dpn);
438
439 if (error) {
440 if (lvp != NULL)
441 VN_RELE(lvp);
442 if (svp != NULL)
443 VN_RELE(svp);
444 }
445 return (error);
446 }
447
448 /*
449 * Mount root file system.
450 * "why" is ROOT_INIT on initial call ROOT_REMOUNT if called to
451 * remount the root file system, and ROOT_UNMOUNT if called to
452 * unmount the root (e.g., as part of a system shutdown).
453 *
454 * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP
455 * operation, goes along with auto-configuration. A mechanism should be
456 * provided by which machine-INdependent code in the kernel can say "get me the
457 * right root file system" and "get me the right initial swap area", and have
458 * that done in what may well be a machine-dependent fashion.
459 * Unfortunately, it is also file-system-type dependent (NFS gets it via
460 * bootparams calls, UFS gets it from various and sundry machine-dependent
461 * mechanisms, as SPECFS does for swap).
462 */
463 static int
464 ufs_mountroot(struct vfs *vfsp, enum whymountroot why)
465 {
466 struct fs *fsp;
467 int error;
468 static int ufsrootdone = 0;
469 dev_t rootdev;
470 struct vnode *vp;
471 struct vnode *devvp = 0;
472 int ovflags;
473 int doclkset;
474 ufsvfs_t *ufsvfsp;
475
476 if (why == ROOT_INIT) {
477 if (ufsrootdone++)
478 return (EBUSY);
479 rootdev = getrootdev();
480 if (rootdev == (dev_t)NODEV)
481 return (ENODEV);
482 vfsp->vfs_dev = rootdev;
483 vfsp->vfs_flag |= VFS_RDONLY;
484 } else if (why == ROOT_REMOUNT) {
485 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
486 (void) dnlc_purge_vfsp(vfsp, 0);
487 vp = common_specvp(vp);
488 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_INVAL,
489 CRED(), NULL);
490 (void) bfinval(vfsp->vfs_dev, 0);
491 fsp = getfs(vfsp);
492
493 ovflags = vfsp->vfs_flag;
494 vfsp->vfs_flag &= ~VFS_RDONLY;
495 vfsp->vfs_flag |= VFS_REMOUNT;
496 rootdev = vfsp->vfs_dev;
497 } else if (why == ROOT_UNMOUNT) {
498 if (vfs_lock(vfsp) == 0) {
499 (void) ufs_flush(vfsp);
500 /*
501 * Mark the log as fully rolled
502 */
503 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
504 fsp = ufsvfsp->vfs_fs;
505 if (TRANS_ISTRANS(ufsvfsp) &&
506 !TRANS_ISERROR(ufsvfsp) &&
507 (fsp->fs_rolled == FS_NEED_ROLL)) {
508 ml_unit_t *ul = ufsvfsp->vfs_log;
509
510 error = ufs_putsummaryinfo(ul->un_dev,
511 ufsvfsp, fsp);
512 if (error == 0) {
513 fsp->fs_rolled = FS_ALL_ROLLED;
514 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
515 }
516 }
517 vfs_unlock(vfsp);
518 } else {
519 ufs_update(0);
520 }
521
522 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
523 (void) VOP_CLOSE(vp, FREAD|FWRITE, 1,
524 (offset_t)0, CRED(), NULL);
525 return (0);
526 }
527 error = vfs_lock(vfsp);
528 if (error)
529 return (error);
530
531 devvp = makespecvp(rootdev, VBLK);
532
533 /* If RO media, don't call clkset() (see below) */
534 doclkset = 1;
535 if (why == ROOT_INIT) {
536 error = VOP_OPEN(&devvp, FREAD|FWRITE, CRED(), NULL);
537 if (error == 0) {
538 (void) VOP_CLOSE(devvp, FREAD|FWRITE, 1,
539 (offset_t)0, CRED(), NULL);
540 } else {
541 doclkset = 0;
542 }
543 }
544
545 error = mountfs(vfsp, why, devvp, "/", CRED(), 1, NULL, 0);
546 /*
547 * XXX - assumes root device is not indirect, because we don't set
548 * rootvp. Is rootvp used for anything? If so, make another arg
549 * to mountfs.
550 */
551 if (error) {
552 vfs_unlock(vfsp);
553 if (why == ROOT_REMOUNT)
554 vfsp->vfs_flag = ovflags;
555 if (rootvp) {
556 VN_RELE(rootvp);
557 rootvp = (struct vnode *)0;
558 }
559 VN_RELE(devvp);
560 return (error);
561 }
562 if (why == ROOT_INIT)
563 vfs_add((struct vnode *)0, vfsp,
564 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
565 vfs_unlock(vfsp);
566 fsp = getfs(vfsp);
567 clkset(doclkset ? fsp->fs_time : -1);
568 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
569 if (ufsvfsp->vfs_log) {
570 vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
571 }
572 return (0);
573 }
574
575 static int
576 remountfs(struct vfs *vfsp, dev_t dev, void *raw_argsp, int args_len)
577 {
578 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
579 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs;
580 struct buf *bp = ufsvfsp->vfs_bufp;
581 struct fs *fsp = (struct fs *)bp->b_un.b_addr;
582 struct fs *fspt;
583 struct buf *tpt = 0;
584 int error = 0;
585 int flags = 0;
586
587 if (args_len == sizeof (struct ufs_args) && raw_argsp)
588 flags = ((struct ufs_args *)raw_argsp)->flags;
589
590 /* cannot remount to RDONLY */
591 if (vfsp->vfs_flag & VFS_RDONLY)
592 return (ENOTSUP);
593
594 /* whoops, wrong dev */
595 if (vfsp->vfs_dev != dev)
596 return (EINVAL);
597
598 /*
599 * synchronize w/ufs ioctls
600 */
601 mutex_enter(&ulp->ul_lock);
602 atomic_inc_ulong(&ufs_quiesce_pend);
603
604 /*
605 * reset options
606 */
607 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR;
608 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
609 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
610 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
611 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
612 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
613 else /* dfratime, default behavior */
614 ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
615 if (flags & UFSMNT_FORCEDIRECTIO)
616 ufsvfsp->vfs_forcedirectio = 1;
617 else /* default is no direct I/O */
618 ufsvfsp->vfs_forcedirectio = 0;
619 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
620
621 /*
622 * set largefiles flag in ufsvfs equal to the
623 * value passed in by the mount command. If
624 * it is "nolargefiles", and the flag is set
625 * in the superblock, the mount fails.
626 */
627 if (!(flags & UFSMNT_LARGEFILES)) { /* "nolargefiles" */
628 if (fsp->fs_flags & FSLARGEFILES) {
629 error = EFBIG;
630 goto remounterr;
631 }
632 ufsvfsp->vfs_lfflags &= ~UFS_LARGEFILES;
633 } else /* "largefiles" */
634 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
635 /*
636 * read/write to read/write; all done
637 */
638 if (fsp->fs_ronly == 0)
639 goto remounterr;
640
641 /*
642 * fix-on-panic assumes RO->RW remount implies system-critical fs
643 * if it is shortly after boot; so, don't attempt to lock and fix
644 * (unless the user explicitly asked for another action on error)
645 * XXX UFSMNT_ONERROR_RDONLY rather than UFSMNT_ONERROR_PANIC
646 */
647 #define BOOT_TIME_LIMIT (180*hz)
648 if (!(flags & UFSMNT_ONERROR_FLGMASK) &&
649 ddi_get_lbolt() < BOOT_TIME_LIMIT) {
650 cmn_err(CE_WARN, "%s is required to be mounted onerror=%s",
651 ufsvfsp->vfs_fs->fs_fsmnt, UFSMNT_ONERROR_PANIC_STR);
652 flags |= UFSMNT_ONERROR_PANIC;
653 }
654
655 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
656 goto remounterr;
657
658 /*
659 * quiesce the file system
660 */
661 error = ufs_quiesce(ulp);
662 if (error)
663 goto remounterr;
664
665 tpt = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, SBLOCK, SBSIZE);
666 if (tpt->b_flags & B_ERROR) {
667 error = EIO;
668 goto remounterr;
669 }
670 fspt = (struct fs *)tpt->b_un.b_addr;
671 if (((fspt->fs_magic != FS_MAGIC) &&
672 (fspt->fs_magic != MTB_UFS_MAGIC)) ||
673 (fspt->fs_magic == FS_MAGIC &&
674 (fspt->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
675 fspt->fs_version != UFS_VERSION_MIN)) ||
676 (fspt->fs_magic == MTB_UFS_MAGIC &&
677 (fspt->fs_version > MTB_UFS_VERSION_1 ||
678 fspt->fs_version < MTB_UFS_VERSION_MIN)) ||
679 fspt->fs_bsize > MAXBSIZE || fspt->fs_frag > MAXFRAG ||
680 fspt->fs_bsize < sizeof (struct fs) || fspt->fs_bsize < PAGESIZE) {
681 tpt->b_flags |= B_STALE | B_AGE;
682 error = EINVAL;
683 goto remounterr;
684 }
685
686 if (ufsvfsp->vfs_log && (ufsvfsp->vfs_log->un_flags & LDL_NOROLL)) {
687 ufsvfsp->vfs_log->un_flags &= ~LDL_NOROLL;
688 logmap_start_roll(ufsvfsp->vfs_log);
689 }
690
691 if (TRANS_ISERROR(ufsvfsp))
692 goto remounterr;
693 TRANS_DOMATAMAP(ufsvfsp);
694
695 if ((fspt->fs_state + fspt->fs_time == FSOKAY) &&
696 fspt->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp)) {
697 ufsvfsp->vfs_log = NULL;
698 ufsvfsp->vfs_domatamap = 0;
699 error = ENOSPC;
700 goto remounterr;
701 }
702
703 if (fspt->fs_state + fspt->fs_time == FSOKAY &&
704 (fspt->fs_clean == FSCLEAN ||
705 fspt->fs_clean == FSSTABLE ||
706 fspt->fs_clean == FSLOG)) {
707
708 /*
709 * Ensure that ufs_getsummaryinfo doesn't reconstruct
710 * the summary info.
711 */
712 error = ufs_getsummaryinfo(vfsp->vfs_dev, ufsvfsp, fspt);
713 if (error)
714 goto remounterr;
715
716 /* preserve mount name */
717 (void) strncpy(fspt->fs_fsmnt, fsp->fs_fsmnt, MAXMNTLEN);
718 /* free the old cg space */
719 kmem_free(fsp->fs_u.fs_csp, fsp->fs_cssize);
720 /* switch in the new superblock */
721 fspt->fs_rolled = FS_NEED_ROLL;
722 bcopy(tpt->b_un.b_addr, bp->b_un.b_addr, fspt->fs_sbsize);
723
724 fsp->fs_clean = FSSTABLE;
725 } /* superblock updated in memory */
726 tpt->b_flags |= B_STALE | B_AGE;
727 brelse(tpt);
728 tpt = 0;
729
730 if (fsp->fs_clean != FSSTABLE) {
731 error = ENOSPC;
732 goto remounterr;
733 }
734
735
736 if (TRANS_ISTRANS(ufsvfsp)) {
737 fsp->fs_clean = FSLOG;
738 ufsvfsp->vfs_dio = 0;
739 } else
740 if (ufsvfsp->vfs_dio)
741 fsp->fs_clean = FSSUSPEND;
742
743 TRANS_MATA_MOUNT(ufsvfsp);
744
745 fsp->fs_fmod = 0;
746 fsp->fs_ronly = 0;
747
748 atomic_dec_ulong(&ufs_quiesce_pend);
749 cv_broadcast(&ulp->ul_cv);
750 mutex_exit(&ulp->ul_lock);
751
752 if (TRANS_ISTRANS(ufsvfsp)) {
753
754 /*
755 * start the delete thread
756 */
757 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
758
759 /*
760 * start the reclaim thread
761 */
762 if (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
763 fsp->fs_reclaim &= ~FS_RECLAIM;
764 fsp->fs_reclaim |= FS_RECLAIMING;
765 ufs_thread_start(&ufsvfsp->vfs_reclaim,
766 ufs_thread_reclaim, vfsp);
767 }
768 }
769
770 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
771
772 return (0);
773
774 remounterr:
775 if (tpt)
776 brelse(tpt);
777 atomic_dec_ulong(&ufs_quiesce_pend);
778 cv_broadcast(&ulp->ul_cv);
779 mutex_exit(&ulp->ul_lock);
780 return (error);
781 }
782
783 /*
784 * If the device maxtransfer size is not available, we use ufs_maxmaxphys
785 * along with the system value for maxphys to determine the value for
786 * maxtransfer.
787 */
788 int ufs_maxmaxphys = (1024 * 1024);
789
790 #include <sys/ddi.h> /* for delay(9f) */
791
792 int ufs_mount_error_delay = 20; /* default to 20ms */
793 int ufs_mount_timeout = 60000; /* default to 1 minute */
794
795 static int
796 mountfs(struct vfs *vfsp, enum whymountroot why, struct vnode *devvp,
797 char *path, cred_t *cr, int isroot, void *raw_argsp, int args_len)
798 {
799 dev_t dev = devvp->v_rdev;
800 struct fs *fsp;
801 struct ufsvfs *ufsvfsp = 0;
802 struct buf *bp = 0;
803 struct buf *tp = 0;
804 struct dk_cinfo ci;
805 int error = 0;
806 size_t len;
807 int needclose = 0;
808 int needtrans = 0;
809 struct inode *rip;
810 struct vnode *rvp = NULL;
811 int flags = 0;
812 kmutex_t *ihm;
813 int elapsed;
814 int status;
815 extern int maxphys;
816
817 if (args_len == sizeof (struct ufs_args) && raw_argsp)
818 flags = ((struct ufs_args *)raw_argsp)->flags;
819
820 ASSERT(vfs_lock_held(vfsp));
821
822 if (why == ROOT_INIT) {
823 /*
824 * Open block device mounted on.
825 * When bio is fixed for vnodes this can all be vnode
826 * operations.
827 */
828 error = VOP_OPEN(&devvp,
829 (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE,
830 cr, NULL);
831 if (error)
832 goto out;
833 needclose = 1;
834
835 /*
836 * Refuse to go any further if this
837 * device is being used for swapping.
838 */
839 if (IS_SWAPVP(devvp)) {
840 error = EBUSY;
841 goto out;
842 }
843 }
844
845 /*
846 * check for dev already mounted on
847 */
848 if (vfsp->vfs_flag & VFS_REMOUNT) {
849 error = remountfs(vfsp, dev, raw_argsp, args_len);
850 if (error == 0)
851 VN_RELE(devvp);
852 return (error);
853 }
854
855 ASSERT(devvp != 0);
856
857 /*
858 * Flush back any dirty pages on the block device to
859 * try and keep the buffer cache in sync with the page
860 * cache if someone is trying to use block devices when
861 * they really should be using the raw device.
862 */
863 (void) VOP_PUTPAGE(common_specvp(devvp), (offset_t)0,
864 (size_t)0, B_INVAL, cr, NULL);
865
866 /*
867 * read in superblock
868 */
869 ufsvfsp = kmem_zalloc(sizeof (struct ufsvfs), KM_SLEEP);
870 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
871 if (tp->b_flags & B_ERROR)
872 goto out;
873 fsp = (struct fs *)tp->b_un.b_addr;
874
875 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) {
876 cmn_err(CE_NOTE,
877 "mount: not a UFS magic number (0x%x)", fsp->fs_magic);
878 error = EINVAL;
879 goto out;
880 }
881
882 if ((fsp->fs_magic == FS_MAGIC) &&
883 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
884 fsp->fs_version != UFS_VERSION_MIN)) {
885 cmn_err(CE_NOTE,
886 "mount: unrecognized version of UFS on-disk format: %d",
887 fsp->fs_version);
888 error = EINVAL;
889 goto out;
890 }
891
892 if ((fsp->fs_magic == MTB_UFS_MAGIC) &&
893 (fsp->fs_version > MTB_UFS_VERSION_1 ||
894 fsp->fs_version < MTB_UFS_VERSION_MIN)) {
895 cmn_err(CE_NOTE,
896 "mount: unrecognized version of UFS on-disk format: %d",
897 fsp->fs_version);
898 error = EINVAL;
899 goto out;
900 }
901
902 #ifndef _LP64
903 if (fsp->fs_magic == MTB_UFS_MAGIC) {
904 /*
905 * Find the size of the device in sectors. If the
906 * the size in sectors is greater than INT_MAX, it's
907 * a multi-terabyte file system, which can't be
908 * mounted by a 32-bit kernel. We can't use the
909 * fsbtodb() macro in the next line because the macro
910 * casts the intermediate values to daddr_t, which is
911 * a 32-bit quantity in a 32-bit kernel. Here we
912 * really do need the intermediate values to be held
913 * in 64-bit quantities because we're checking for
914 * overflow of a 32-bit field.
915 */
916 if ((((diskaddr_t)(fsp->fs_size)) << fsp->fs_fsbtodb)
917 > INT_MAX) {
918 cmn_err(CE_NOTE,
919 "mount: multi-terabyte UFS cannot be"
920 " mounted by a 32-bit kernel");
921 error = EINVAL;
922 goto out;
923 }
924
925 }
926 #endif
927
928 if (fsp->fs_bsize > MAXBSIZE || fsp->fs_frag > MAXFRAG ||
929 fsp->fs_bsize < sizeof (struct fs) || fsp->fs_bsize < PAGESIZE) {
930 error = EINVAL; /* also needs translation */
931 goto out;
932 }
933
934 /*
935 * Allocate VFS private data.
936 */
937 vfsp->vfs_bcount = 0;
938 vfsp->vfs_data = (caddr_t)ufsvfsp;
939 vfsp->vfs_fstype = ufsfstype;
940 vfsp->vfs_dev = dev;
941 vfsp->vfs_flag |= VFS_NOTRUNC;
942 vfs_make_fsid(&vfsp->vfs_fsid, dev, ufsfstype);
943 ufsvfsp->vfs_devvp = devvp;
944
945 /*
946 * Cross-link with vfs and add to instance list.
947 */
948 ufsvfsp->vfs_vfs = vfsp;
949 ufs_vfs_add(ufsvfsp);
950
951 ufsvfsp->vfs_dev = dev;
952 ufsvfsp->vfs_bufp = tp;
953
954 ufsvfsp->vfs_dirsize = INODESIZE + (4 * ALLOCSIZE) + fsp->fs_fsize;
955 ufsvfsp->vfs_minfrags =
956 (int)((int64_t)fsp->fs_dsize * fsp->fs_minfree / 100);
957 /*
958 * if mount allows largefiles, indicate so in ufsvfs
959 */
960 if (flags & UFSMNT_LARGEFILES)
961 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
962 /*
963 * Initialize threads
964 */
965 ufs_delete_init(ufsvfsp, 1);
966 ufs_thread_init(&ufsvfsp->vfs_reclaim, 0);
967
968 /*
969 * Chicken and egg problem. The superblock may have deltas
970 * in the log. So after the log is scanned we reread the
971 * superblock. We guarantee that the fields needed to
972 * scan the log will not be in the log.
973 */
974 if (fsp->fs_logbno && fsp->fs_clean == FSLOG &&
975 (fsp->fs_state + fsp->fs_time == FSOKAY)) {
976 error = lufs_snarf(ufsvfsp, fsp, (vfsp->vfs_flag & VFS_RDONLY));
977 if (error) {
978 /*
979 * Allow a ro mount to continue even if the
980 * log cannot be processed - yet.
981 */
982 if (!(vfsp->vfs_flag & VFS_RDONLY)) {
983 cmn_err(CE_WARN, "Error accessing ufs "
984 "log for %s; Please run fsck(1M)", path);
985 goto out;
986 }
987 }
988 tp->b_flags |= (B_AGE | B_STALE);
989 brelse(tp);
990 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
991 fsp = (struct fs *)tp->b_un.b_addr;
992 ufsvfsp->vfs_bufp = tp;
993 if (tp->b_flags & B_ERROR)
994 goto out;
995 }
996
997 /*
998 * Set logging mounted flag used by lockfs
999 */
1000 ufsvfsp->vfs_validfs = UT_MOUNTED;
1001
1002 /*
1003 * Copy the super block into a buffer in its native size.
1004 * Use ngeteblk to allocate the buffer
1005 */
1006 bp = ngeteblk(fsp->fs_bsize);
1007 ufsvfsp->vfs_bufp = bp;
1008 bp->b_edev = dev;
1009 bp->b_dev = cmpdev(dev);
1010 bp->b_blkno = SBLOCK;
1011 bp->b_bcount = fsp->fs_sbsize;
1012 bcopy(tp->b_un.b_addr, bp->b_un.b_addr, fsp->fs_sbsize);
1013 tp->b_flags |= B_STALE | B_AGE;
1014 brelse(tp);
1015 tp = 0;
1016
1017 fsp = (struct fs *)bp->b_un.b_addr;
1018 /*
1019 * Mount fails if superblock flag indicates presence of large
1020 * files and filesystem is attempted to be mounted 'nolargefiles'.
1021 * The exception is for a read only mount of root, which we
1022 * always want to succeed, so fsck can fix potential problems.
1023 * The assumption is that we will remount root at some point,
1024 * and the remount will enforce the mount option.
1025 */
1026 if (!(isroot & (vfsp->vfs_flag & VFS_RDONLY)) &&
1027 (fsp->fs_flags & FSLARGEFILES) &&
1028 !(flags & UFSMNT_LARGEFILES)) {
1029 error = EFBIG;
1030 goto out;
1031 }
1032
1033 if (vfsp->vfs_flag & VFS_RDONLY) {
1034 fsp->fs_ronly = 1;
1035 fsp->fs_fmod = 0;
1036 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1037 ((fsp->fs_clean == FSCLEAN) ||
1038 (fsp->fs_clean == FSSTABLE) ||
1039 (fsp->fs_clean == FSLOG))) {
1040 if (isroot) {
1041 if (fsp->fs_clean == FSLOG) {
1042 if (fsp->fs_rolled == FS_ALL_ROLLED) {
1043 ufs_clean_root = 1;
1044 }
1045 } else {
1046 ufs_clean_root = 1;
1047 }
1048 }
1049 fsp->fs_clean = FSSTABLE;
1050 } else {
1051 fsp->fs_clean = FSBAD;
1052 }
1053 } else {
1054
1055 fsp->fs_fmod = 0;
1056 fsp->fs_ronly = 0;
1057
1058 TRANS_DOMATAMAP(ufsvfsp);
1059
1060 if ((TRANS_ISERROR(ufsvfsp)) ||
1061 (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1062 fsp->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp))) {
1063 ufsvfsp->vfs_log = NULL;
1064 ufsvfsp->vfs_domatamap = 0;
1065 error = ENOSPC;
1066 goto out;
1067 }
1068
1069 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1070 (fsp->fs_clean == FSCLEAN ||
1071 fsp->fs_clean == FSSTABLE ||
1072 fsp->fs_clean == FSLOG))
1073 fsp->fs_clean = FSSTABLE;
1074 else {
1075 if (isroot) {
1076 /*
1077 * allow root partition to be mounted even
1078 * when fs_state is not ok
1079 * will be fixed later by a remount root
1080 */
1081 fsp->fs_clean = FSBAD;
1082 ufsvfsp->vfs_log = NULL;
1083 ufsvfsp->vfs_domatamap = 0;
1084 } else {
1085 error = ENOSPC;
1086 goto out;
1087 }
1088 }
1089
1090 if (fsp->fs_clean == FSSTABLE && TRANS_ISTRANS(ufsvfsp))
1091 fsp->fs_clean = FSLOG;
1092 }
1093 TRANS_MATA_MOUNT(ufsvfsp);
1094 needtrans = 1;
1095
1096 vfsp->vfs_bsize = fsp->fs_bsize;
1097
1098 /*
1099 * Read in summary info
1100 */
1101 if (error = ufs_getsummaryinfo(dev, ufsvfsp, fsp))
1102 goto out;
1103
1104 /*
1105 * lastwhinetime is set to zero rather than lbolt, so that after
1106 * mounting if the filesystem is found to be full, then immediately the
1107 * "file system message" will be logged.
1108 */
1109 ufsvfsp->vfs_lastwhinetime = 0L;
1110
1111
1112 mutex_init(&ufsvfsp->vfs_lock, NULL, MUTEX_DEFAULT, NULL);
1113 (void) copystr(path, fsp->fs_fsmnt, sizeof (fsp->fs_fsmnt) - 1, &len);
1114 bzero(fsp->fs_fsmnt + len, sizeof (fsp->fs_fsmnt) - len);
1115
1116 /*
1117 * Sanity checks for old file systems
1118 */
1119 if (fsp->fs_postblformat == FS_42POSTBLFMT)
1120 ufsvfsp->vfs_nrpos = 8;
1121 else
1122 ufsvfsp->vfs_nrpos = fsp->fs_nrpos;
1123
1124 /*
1125 * Initialize lockfs structure to support file system locking
1126 */
1127 bzero(&ufsvfsp->vfs_ulockfs.ul_lockfs,
1128 sizeof (struct lockfs));
1129 ufsvfsp->vfs_ulockfs.ul_fs_lock = ULOCKFS_ULOCK;
1130 mutex_init(&ufsvfsp->vfs_ulockfs.ul_lock, NULL,
1131 MUTEX_DEFAULT, NULL);
1132 cv_init(&ufsvfsp->vfs_ulockfs.ul_cv, NULL, CV_DEFAULT, NULL);
1133
1134 /*
1135 * We don't need to grab vfs_dqrwlock for this ufs_iget() call.
1136 * We are in the process of mounting the file system so there
1137 * is no need to grab the quota lock. If a quota applies to the
1138 * root inode, then it will be updated when quotas are enabled.
1139 *
1140 * However, we have an ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock))
1141 * in getinoquota() that we want to keep so grab it anyway.
1142 */
1143 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1144
1145 error = ufs_iget_alloced(vfsp, UFSROOTINO, &rip, cr);
1146
1147 rw_exit(&ufsvfsp->vfs_dqrwlock);
1148
1149 if (error)
1150 goto out;
1151
1152 /*
1153 * make sure root inode is a directory. Returning ENOTDIR might
1154 * be confused with the mount point not being a directory, so
1155 * we use EIO instead.
1156 */
1157 if ((rip->i_mode & IFMT) != IFDIR) {
1158 /*
1159 * Mark this inode as subject for cleanup
1160 * to avoid stray inodes in the cache.
1161 */
1162 rvp = ITOV(rip);
1163 error = EIO;
1164 goto out;
1165 }
1166
1167 rvp = ITOV(rip);
1168 mutex_enter(&rvp->v_lock);
1169 rvp->v_flag |= VROOT;
1170 mutex_exit(&rvp->v_lock);
1171 ufsvfsp->vfs_root = rvp;
1172 /* The buffer for the root inode does not contain a valid b_vp */
1173 (void) bfinval(dev, 0);
1174
1175 /* options */
1176 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
1177 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR;
1178 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
1179 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
1180 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
1181 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
1182 else /* dfratime, default behavior */
1183 ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
1184 if (flags & UFSMNT_FORCEDIRECTIO)
1185 ufsvfsp->vfs_forcedirectio = 1;
1186 else if (flags & UFSMNT_NOFORCEDIRECTIO)
1187 ufsvfsp->vfs_forcedirectio = 0;
1188 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1189
1190 ufsvfsp->vfs_nindiroffset = fsp->fs_nindir - 1;
1191 ufsvfsp->vfs_nindirshift = highbit(ufsvfsp->vfs_nindiroffset);
1192 ufsvfsp->vfs_ioclustsz = fsp->fs_bsize * fsp->fs_maxcontig;
1193
1194 if (cdev_ioctl(dev, DKIOCINFO, (intptr_t)&ci,
1195 FKIOCTL|FNATIVE|FREAD, CRED(), &status) == 0) {
1196 ufsvfsp->vfs_iotransz = ci.dki_maxtransfer * DEV_BSIZE;
1197 } else {
1198 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1199 }
1200
1201 if (ufsvfsp->vfs_iotransz <= 0) {
1202 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1203 }
1204
1205 /*
1206 * When logging, used to reserve log space for writes and truncs
1207 */
1208 ufsvfsp->vfs_avgbfree = fsp->fs_cstotal.cs_nbfree / fsp->fs_ncg;
1209
1210 /*
1211 * Determine whether to log cylinder group summary info.
1212 */
1213 ufsvfsp->vfs_nolog_si = (fsp->fs_ncg < ufs_ncg_log);
1214
1215 if (TRANS_ISTRANS(ufsvfsp)) {
1216 /*
1217 * start the delete thread
1218 */
1219 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
1220
1221 /*
1222 * start reclaim thread if the filesystem was not mounted
1223 * read only.
1224 */
1225 if (!fsp->fs_ronly && (fsp->fs_reclaim &
1226 (FS_RECLAIM|FS_RECLAIMING))) {
1227 fsp->fs_reclaim &= ~FS_RECLAIM;
1228 fsp->fs_reclaim |= FS_RECLAIMING;
1229 ufs_thread_start(&ufsvfsp->vfs_reclaim,
1230 ufs_thread_reclaim, vfsp);
1231 }
1232
1233 /* Mark the fs as unrolled */
1234 fsp->fs_rolled = FS_NEED_ROLL;
1235 } else if (!fsp->fs_ronly && (fsp->fs_reclaim &
1236 (FS_RECLAIM|FS_RECLAIMING))) {
1237 /*
1238 * If a file system that is mounted nologging, after
1239 * having previously been mounted logging, becomes
1240 * unmounted whilst the reclaim thread is in the throes
1241 * of reclaiming open/deleted inodes, a subsequent mount
1242 * of such a file system with logging disabled could lead
1243 * to inodes becoming lost. So, start reclaim now, even
1244 * though logging was disabled for the previous mount, to
1245 * tidy things up.
1246 */
1247 fsp->fs_reclaim &= ~FS_RECLAIM;
1248 fsp->fs_reclaim |= FS_RECLAIMING;
1249 ufs_thread_start(&ufsvfsp->vfs_reclaim,
1250 ufs_thread_reclaim, vfsp);
1251 }
1252
1253 if (!fsp->fs_ronly) {
1254 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
1255 if (error = geterror(ufsvfsp->vfs_bufp))
1256 goto out;
1257 }
1258
1259 /* fix-on-panic initialization */
1260 if (isroot && !(flags & UFSMNT_ONERROR_FLGMASK))
1261 flags |= UFSMNT_ONERROR_PANIC; /* XXX ..._RDONLY */
1262
1263 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
1264 goto out;
1265
1266 if (why == ROOT_INIT && isroot)
1267 rootvp = devvp;
1268
1269 return (0);
1270 out:
1271 if (error == 0)
1272 error = EIO;
1273 if (rvp) {
1274 /* the following sequence is similar to ufs_unmount() */
1275
1276 /*
1277 * There's a problem that ufs_iget() puts inodes into
1278 * the inode cache before it returns them. If someone
1279 * traverses that cache and gets a reference to our
1280 * inode, there's a chance they'll still be using it
1281 * after we've destroyed it. This is a hard race to
1282 * hit, but it's happened (putting in a medium delay
1283 * here, and a large delay in ufs_scan_inodes() for
1284 * inodes on the device we're bailing out on, makes
1285 * the race easy to demonstrate). The symptom is some
1286 * other part of UFS faulting on bad inode contents,
1287 * or when grabbing one of the locks inside the inode,
1288 * etc. The usual victim is ufs_scan_inodes() or
1289 * someone called by it.
1290 */
1291
1292 /*
1293 * First, isolate it so that no new references can be
1294 * gotten via the inode cache.
1295 */
1296 ihm = &ih_lock[INOHASH(UFSROOTINO)];
1297 mutex_enter(ihm);
1298 remque(rip);
1299 mutex_exit(ihm);
1300
1301 /*
1302 * Now wait for all outstanding references except our
1303 * own to drain. This could, in theory, take forever,
1304 * so don't wait *too* long. If we time out, mark
1305 * it stale and leak it, so we don't hit the problem
1306 * described above.
1307 *
1308 * Note that v_count is an int, which means we can read
1309 * it in one operation. Thus, there's no need to lock
1310 * around our tests.
1311 */
1312 elapsed = 0;
1313 while ((rvp->v_count > 1) && (elapsed < ufs_mount_timeout)) {
1314 delay(ufs_mount_error_delay * drv_usectohz(1000));
1315 elapsed += ufs_mount_error_delay;
1316 }
1317
1318 if (rvp->v_count > 1) {
1319 mutex_enter(&rip->i_tlock);
1320 rip->i_flag |= ISTALE;
1321 mutex_exit(&rip->i_tlock);
1322 cmn_err(CE_WARN,
1323 "Timed out while cleaning up after "
1324 "failed mount of %s", path);
1325 } else {
1326
1327 /*
1328 * Now we're the only one with a handle left, so tear
1329 * it down the rest of the way.
1330 */
1331 if (ufs_rmidle(rip))
1332 VN_RELE(rvp);
1333 ufs_si_del(rip);
1334 rip->i_ufsvfs = NULL;
1335 rvp->v_vfsp = NULL;
1336 rvp->v_type = VBAD;
1337 VN_RELE(rvp);
1338 }
1339 }
1340 if (needtrans) {
1341 TRANS_MATA_UMOUNT(ufsvfsp);
1342 }
1343 if (ufsvfsp) {
1344 ufs_vfs_remove(ufsvfsp);
1345 ufs_thread_exit(&ufsvfsp->vfs_delete);
1346 ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1347 mutex_destroy(&ufsvfsp->vfs_lock);
1348 if (ufsvfsp->vfs_log) {
1349 lufs_unsnarf(ufsvfsp);
1350 }
1351 kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1352 }
1353 if (bp) {
1354 bp->b_flags |= (B_STALE|B_AGE);
1355 brelse(bp);
1356 }
1357 if (tp) {
1358 tp->b_flags |= (B_STALE|B_AGE);
1359 brelse(tp);
1360 }
1361 if (needclose) {
1362 (void) VOP_CLOSE(devvp, (vfsp->vfs_flag & VFS_RDONLY) ?
1363 FREAD : FREAD|FWRITE, 1, (offset_t)0, cr, NULL);
1364 bflush(dev);
1365 (void) bfinval(dev, 1);
1366 }
1367 return (error);
1368 }
1369
1370 /*
1371 * vfs operations
1372 */
1373 static int
1374 ufs_unmount(struct vfs *vfsp, int fflag, struct cred *cr)
1375 {
1376 dev_t dev = vfsp->vfs_dev;
1377 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1378 struct fs *fs = ufsvfsp->vfs_fs;
1379 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs;
1380 struct vnode *bvp, *vp;
1381 struct buf *bp;
1382 struct inode *ip, *inext, *rip;
1383 union ihead *ih;
1384 int error, flag, i;
1385 struct lockfs lockfs;
1386 int poll_events = POLLPRI;
1387 extern struct pollhead ufs_pollhd;
1388 refstr_t *mountpoint;
1389
1390 ASSERT(vfs_lock_held(vfsp));
1391
1392 if (secpolicy_fs_unmount(cr, vfsp) != 0)
1393 return (EPERM);
1394 /*
1395 * Forced unmount is now supported through the
1396 * lockfs protocol.
1397 */
1398 if (fflag & MS_FORCE) {
1399 /*
1400 * Mark the filesystem as being unmounted now in
1401 * case of a forcible umount before we take any
1402 * locks inside UFS to prevent racing with a VFS_VGET()
1403 * request. Throw these VFS_VGET() requests away for
1404 * the duration of the forcible umount so they won't
1405 * use stale or even freed data later on when we're done.
1406 * It may happen that the VFS has had a additional hold
1407 * placed on it by someone other than UFS and thus will
1408 * not get freed immediately once we're done with the
1409 * umount by dounmount() - use VFS_UNMOUNTED to inform
1410 * users of this still-alive VFS that its corresponding
1411 * filesystem being gone so they can detect that and error
1412 * out.
1413 */
1414 vfsp->vfs_flag |= VFS_UNMOUNTED;
1415
1416 ufs_thread_suspend(&ufsvfsp->vfs_delete);
1417 mutex_enter(&ulp->ul_lock);
1418 /*
1419 * If file system is already hard locked,
1420 * unmount the file system, otherwise
1421 * hard lock it before unmounting.
1422 */
1423 if (!ULOCKFS_IS_HLOCK(ulp)) {
1424 atomic_inc_ulong(&ufs_quiesce_pend);
1425 lockfs.lf_lock = LOCKFS_HLOCK;
1426 lockfs.lf_flags = 0;
1427 lockfs.lf_key = ulp->ul_lockfs.lf_key + 1;
1428 lockfs.lf_comlen = 0;
1429 lockfs.lf_comment = NULL;
1430 ufs_freeze(ulp, &lockfs);
1431 ULOCKFS_SET_BUSY(ulp);
1432 LOCKFS_SET_BUSY(&ulp->ul_lockfs);
1433 (void) ufs_quiesce(ulp);
1434 (void) ufs_flush(vfsp);
1435 (void) ufs_thaw(vfsp, ufsvfsp, ulp);
1436 atomic_dec_ulong(&ufs_quiesce_pend);
1437 ULOCKFS_CLR_BUSY(ulp);
1438 LOCKFS_CLR_BUSY(&ulp->ul_lockfs);
1439 poll_events |= POLLERR;
1440 pollwakeup(&ufs_pollhd, poll_events);
1441 }
1442 ufs_thread_continue(&ufsvfsp->vfs_delete);
1443 mutex_exit(&ulp->ul_lock);
1444 }
1445
1446 /* let all types of writes go through */
1447 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1448
1449 /* coordinate with global hlock thread */
1450 if (TRANS_ISTRANS(ufsvfsp) && (ufsvfsp->vfs_validfs == UT_HLOCKING)) {
1451 /*
1452 * last possibility for a forced umount to fail hence clear
1453 * VFS_UNMOUNTED if appropriate.
1454 */
1455 if (fflag & MS_FORCE)
1456 vfsp->vfs_flag &= ~VFS_UNMOUNTED;
1457 return (EAGAIN);
1458 }
1459
1460 ufsvfsp->vfs_validfs = UT_UNMOUNTED;
1461
1462 /* kill the reclaim thread */
1463 ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1464
1465 /* suspend the delete thread */
1466 ufs_thread_suspend(&ufsvfsp->vfs_delete);
1467
1468 /*
1469 * drain the delete and idle queues
1470 */
1471 ufs_delete_drain(vfsp, -1, 1);
1472 ufs_idle_drain(vfsp);
1473
1474 /*
1475 * use the lockfs protocol to prevent new ops from starting
1476 * a forcible umount can not fail beyond this point as
1477 * we hard-locked the filesystem and drained all current consumers
1478 * before.
1479 */
1480 mutex_enter(&ulp->ul_lock);
1481
1482 /*
1483 * if the file system is busy; return EBUSY
1484 */
1485 if (ulp->ul_vnops_cnt || ulp->ul_falloc_cnt || ULOCKFS_IS_SLOCK(ulp)) {
1486 error = EBUSY;
1487 goto out;
1488 }
1489
1490 /*
1491 * if this is not a forced unmount (!hard/error locked), then
1492 * get rid of every inode except the root and quota inodes
1493 * also, commit any outstanding transactions
1494 */
1495 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp))
1496 if (error = ufs_flush(vfsp))
1497 goto out;
1498
1499 /*
1500 * ignore inodes in the cache if fs is hard locked or error locked
1501 */
1502 rip = VTOI(ufsvfsp->vfs_root);
1503 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) {
1504 /*
1505 * Otherwise, only the quota and root inodes are in the cache.
1506 *
1507 * Avoid racing with ufs_update() and ufs_sync().
1508 */
1509 mutex_enter(&ufs_scan_lock);
1510
1511 for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1512 mutex_enter(&ih_lock[i]);
1513 for (ip = ih->ih_chain[0];
1514 ip != (struct inode *)ih;
1515 ip = ip->i_forw) {
1516 if (ip->i_ufsvfs != ufsvfsp)
1517 continue;
1518 if (ip == ufsvfsp->vfs_qinod)
1519 continue;
1520 if (ip == rip && ITOV(ip)->v_count == 1)
1521 continue;
1522 mutex_exit(&ih_lock[i]);
1523 mutex_exit(&ufs_scan_lock);
1524 error = EBUSY;
1525 goto out;
1526 }
1527 mutex_exit(&ih_lock[i]);
1528 }
1529 mutex_exit(&ufs_scan_lock);
1530 }
1531
1532 /*
1533 * if a snapshot exists and this is a forced unmount, then delete
1534 * the snapshot. Otherwise return EBUSY. This will insure the
1535 * snapshot always belongs to a valid file system.
1536 */
1537 if (ufsvfsp->vfs_snapshot) {
1538 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1539 (void) fssnap_delete(&ufsvfsp->vfs_snapshot);
1540 } else {
1541 error = EBUSY;
1542 goto out;
1543 }
1544 }
1545
1546 /*
1547 * Close the quota file and invalidate anything left in the quota
1548 * cache for this file system. Pass kcred to allow all quota
1549 * manipulations.
1550 */
1551 (void) closedq(ufsvfsp, kcred);
1552 invalidatedq(ufsvfsp);
1553 /*
1554 * drain the delete and idle queues
1555 */
1556 ufs_delete_drain(vfsp, -1, 0);
1557 ufs_idle_drain(vfsp);
1558
1559 /*
1560 * discard the inodes for this fs (including root, shadow, and quota)
1561 */
1562 for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1563 mutex_enter(&ih_lock[i]);
1564 for (inext = 0, ip = ih->ih_chain[0];
1565 ip != (struct inode *)ih;
1566 ip = inext) {
1567 inext = ip->i_forw;
1568 if (ip->i_ufsvfs != ufsvfsp)
1569 continue;
1570
1571 /*
1572 * We've found the inode in the cache and as we
1573 * hold the hash mutex the inode can not
1574 * disappear from underneath us.
1575 * We also know it must have at least a vnode
1576 * reference count of 1.
1577 * We perform an additional VN_HOLD so the VN_RELE
1578 * in case we take the inode off the idle queue
1579 * can not be the last one.
1580 * It is safe to grab the writer contents lock here
1581 * to prevent a race with ufs_iinactive() putting
1582 * inodes into the idle queue while we operate on
1583 * this inode.
1584 */
1585 rw_enter(&ip->i_contents, RW_WRITER);
1586
1587 vp = ITOV(ip);
1588 VN_HOLD(vp)
1589 remque(ip);
1590 if (ufs_rmidle(ip))
1591 VN_RELE(vp);
1592 ufs_si_del(ip);
1593 /*
1594 * rip->i_ufsvfsp is needed by bflush()
1595 */
1596 if (ip != rip)
1597 ip->i_ufsvfs = NULL;
1598 /*
1599 * Set vnode's vfsops to dummy ops, which return
1600 * EIO. This is needed to forced unmounts to work
1601 * with lofs/nfs properly.
1602 */
1603 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp))
1604 vp->v_vfsp = &EIO_vfs;
1605 else
1606 vp->v_vfsp = NULL;
1607 vp->v_type = VBAD;
1608
1609 rw_exit(&ip->i_contents);
1610
1611 VN_RELE(vp);
1612 }
1613 mutex_exit(&ih_lock[i]);
1614 }
1615 ufs_si_cache_flush(dev);
1616
1617 /*
1618 * kill the delete thread and drain the idle queue
1619 */
1620 ufs_thread_exit(&ufsvfsp->vfs_delete);
1621 ufs_idle_drain(vfsp);
1622
1623 bp = ufsvfsp->vfs_bufp;
1624 bvp = ufsvfsp->vfs_devvp;
1625 flag = !fs->fs_ronly;
1626 if (flag) {
1627 bflush(dev);
1628 if (fs->fs_clean != FSBAD) {
1629 if (fs->fs_clean == FSSTABLE)
1630 fs->fs_clean = FSCLEAN;
1631 fs->fs_reclaim &= ~FS_RECLAIM;
1632 }
1633 if (TRANS_ISTRANS(ufsvfsp) &&
1634 !TRANS_ISERROR(ufsvfsp) &&
1635 !ULOCKFS_IS_HLOCK(ulp) &&
1636 (fs->fs_rolled == FS_NEED_ROLL)) {
1637 /*
1638 * ufs_flush() above has flushed the last Moby.
1639 * This is needed to ensure the following superblock
1640 * update really is the last metadata update
1641 */
1642 error = ufs_putsummaryinfo(dev, ufsvfsp, fs);
1643 if (error == 0) {
1644 fs->fs_rolled = FS_ALL_ROLLED;
1645 }
1646 }
1647 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UNMOUNT);
1648 /*
1649 * push this last transaction
1650 */
1651 curthread->t_flag |= T_DONTBLOCK;
1652 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE,
1653 error);
1654 if (!error)
1655 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UNMOUNT,
1656 TOP_COMMIT_SIZE);
1657 curthread->t_flag &= ~T_DONTBLOCK;
1658 }
1659
1660 TRANS_MATA_UMOUNT(ufsvfsp);
1661 lufs_unsnarf(ufsvfsp); /* Release the in-memory structs */
1662 ufsfx_unmount(ufsvfsp); /* fix-on-panic bookkeeping */
1663 kmem_free(fs->fs_u.fs_csp, fs->fs_cssize);
1664
1665 bp->b_flags |= B_STALE|B_AGE;
1666 ufsvfsp->vfs_bufp = NULL; /* don't point at freed buf */
1667 brelse(bp); /* free the superblock buf */
1668
1669 (void) VOP_PUTPAGE(common_specvp(bvp), (offset_t)0, (size_t)0,
1670 B_INVAL, cr, NULL);
1671 (void) VOP_CLOSE(bvp, flag, 1, (offset_t)0, cr, NULL);
1672 bflush(dev);
1673 (void) bfinval(dev, 1);
1674 VN_RELE(bvp);
1675
1676 /*
1677 * It is now safe to NULL out the ufsvfs pointer and discard
1678 * the root inode.
1679 */
1680 rip->i_ufsvfs = NULL;
1681 VN_RELE(ITOV(rip));
1682
1683 /* free up lockfs comment structure, if any */
1684 if (ulp->ul_lockfs.lf_comlen && ulp->ul_lockfs.lf_comment)
1685 kmem_free(ulp->ul_lockfs.lf_comment, ulp->ul_lockfs.lf_comlen);
1686
1687 /*
1688 * Remove from instance list.
1689 */
1690 ufs_vfs_remove(ufsvfsp);
1691
1692 /*
1693 * For a forcible unmount, threads may be asleep in
1694 * ufs_lockfs_begin/ufs_check_lockfs. These threads will need
1695 * the ufsvfs structure so we don't free it, yet. ufs_update
1696 * will free it up after awhile.
1697 */
1698 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1699 extern kmutex_t ufsvfs_mutex;
1700 extern struct ufsvfs *ufsvfslist;
1701
1702 mutex_enter(&ufsvfs_mutex);
1703 ufsvfsp->vfs_dontblock = 1;
1704 ufsvfsp->vfs_next = ufsvfslist;
1705 ufsvfslist = ufsvfsp;
1706 mutex_exit(&ufsvfs_mutex);
1707 /* wakeup any suspended threads */
1708 cv_broadcast(&ulp->ul_cv);
1709 mutex_exit(&ulp->ul_lock);
1710 } else {
1711 mutex_destroy(&ufsvfsp->vfs_lock);
1712 kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1713 }
1714
1715 /*
1716 * Now mark the filesystem as unmounted since we're done with it.
1717 */
1718 vfsp->vfs_flag |= VFS_UNMOUNTED;
1719
1720 return (0);
1721 out:
1722 /* open the fs to new ops */
1723 cv_broadcast(&ulp->ul_cv);
1724 mutex_exit(&ulp->ul_lock);
1725
1726 if (TRANS_ISTRANS(ufsvfsp)) {
1727 /* allow the delete thread to continue */
1728 ufs_thread_continue(&ufsvfsp->vfs_delete);
1729 /* restart the reclaim thread */
1730 ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim,
1731 vfsp);
1732 /* coordinate with global hlock thread */
1733 ufsvfsp->vfs_validfs = UT_MOUNTED;
1734 /* check for trans errors during umount */
1735 ufs_trans_onerror();
1736
1737 /*
1738 * if we have a separate /usr it will never unmount
1739 * when halting. In order to not re-read all the
1740 * cylinder group summary info on mounting after
1741 * reboot the logging of summary info is re-enabled
1742 * and the super block written out.
1743 */
1744 mountpoint = vfs_getmntpoint(vfsp);
1745 if ((fs->fs_si == FS_SI_OK) &&
1746 (strcmp("/usr", refstr_value(mountpoint)) == 0)) {
1747 ufsvfsp->vfs_nolog_si = 0;
1748 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
1749 }
1750 refstr_rele(mountpoint);
1751 }
1752
1753 return (error);
1754 }
1755
1756 static int
1757 ufs_root(struct vfs *vfsp, struct vnode **vpp)
1758 {
1759 struct ufsvfs *ufsvfsp;
1760 struct vnode *vp;
1761
1762 if (!vfsp)
1763 return (EIO);
1764
1765 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1766 if (!ufsvfsp || !ufsvfsp->vfs_root)
1767 return (EIO); /* forced unmount */
1768
1769 vp = ufsvfsp->vfs_root;
1770 VN_HOLD(vp);
1771 *vpp = vp;
1772 return (0);
1773 }
1774
1775 /*
1776 * Get file system statistics.
1777 */
1778 static int
1779 ufs_statvfs(struct vfs *vfsp, struct statvfs64 *sp)
1780 {
1781 struct fs *fsp;
1782 struct ufsvfs *ufsvfsp;
1783 int blk, i;
1784 long max_avail, used;
1785 dev32_t d32;
1786
1787 if (vfsp->vfs_flag & VFS_UNMOUNTED)
1788 return (EIO);
1789
1790 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1791 fsp = ufsvfsp->vfs_fs;
1792 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC))
1793 return (EINVAL);
1794 if (fsp->fs_magic == FS_MAGIC &&
1795 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
1796 fsp->fs_version != UFS_VERSION_MIN))
1797 return (EINVAL);
1798 if (fsp->fs_magic == MTB_UFS_MAGIC &&
1799 (fsp->fs_version > MTB_UFS_VERSION_1 ||
1800 fsp->fs_version < MTB_UFS_VERSION_MIN))
1801 return (EINVAL);
1802
1803 /*
1804 * get the basic numbers
1805 */
1806 (void) bzero(sp, sizeof (*sp));
1807
1808 sp->f_bsize = fsp->fs_bsize;
1809 sp->f_frsize = fsp->fs_fsize;
1810 sp->f_blocks = (fsblkcnt64_t)fsp->fs_dsize;
1811 sp->f_bfree = (fsblkcnt64_t)fsp->fs_cstotal.cs_nbfree * fsp->fs_frag +
1812 fsp->fs_cstotal.cs_nffree;
1813
1814 sp->f_files = (fsfilcnt64_t)fsp->fs_ncg * fsp->fs_ipg;
1815 sp->f_ffree = (fsfilcnt64_t)fsp->fs_cstotal.cs_nifree;
1816
1817 /*
1818 * Adjust the numbers based on things waiting to be deleted.
1819 * modifies f_bfree and f_ffree. Afterwards, everything we
1820 * come up with will be self-consistent. By definition, this
1821 * is a point-in-time snapshot, so the fact that the delete
1822 * thread's probably already invalidated the results is not a
1823 * problem. Note that if the delete thread is ever extended to
1824 * non-logging ufs, this adjustment must always be made.
1825 */
1826 if (TRANS_ISTRANS(ufsvfsp))
1827 ufs_delete_adjust_stats(ufsvfsp, sp);
1828
1829 /*
1830 * avail = MAX(max_avail - used, 0)
1831 */
1832 max_avail = fsp->fs_dsize - ufsvfsp->vfs_minfrags;
1833
1834 used = (fsp->fs_dsize - sp->f_bfree);
1835
1836 if (max_avail > used)
1837 sp->f_bavail = (fsblkcnt64_t)max_avail - used;
1838 else
1839 sp->f_bavail = (fsblkcnt64_t)0;
1840
1841 sp->f_favail = sp->f_ffree;
1842 (void) cmpldev(&d32, vfsp->vfs_dev);
1843 sp->f_fsid = d32;
1844 (void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1845 sp->f_flag = vf_to_stf(vfsp->vfs_flag);
1846
1847 /* keep coordinated with ufs_l_pathconf() */
1848 sp->f_namemax = MAXNAMLEN;
1849
1850 if (fsp->fs_cpc == 0) {
1851 bzero(sp->f_fstr, 14);
1852 return (0);
1853 }
1854 blk = fsp->fs_spc * fsp->fs_cpc / NSPF(fsp);
1855 for (i = 0; i < blk; i += fsp->fs_frag) /* CSTYLED */
1856 /* void */;
1857 i -= fsp->fs_frag;
1858 blk = i / fsp->fs_frag;
1859 bcopy(&(fs_rotbl(fsp)[blk]), sp->f_fstr, 14);
1860 return (0);
1861 }
1862
1863 /*
1864 * Flush any pending I/O to file system vfsp.
1865 * The ufs_update() routine will only flush *all* ufs files.
1866 * If vfsp is non-NULL, only sync this ufs (in preparation
1867 * for a umount).
1868 */
1869 /*ARGSUSED*/
1870 static int
1871 ufs_sync(struct vfs *vfsp, short flag, struct cred *cr)
1872 {
1873 struct ufsvfs *ufsvfsp;
1874 struct fs *fs;
1875 int cheap = flag & SYNC_ATTR;
1876 int error;
1877
1878 /*
1879 * SYNC_CLOSE means we're rebooting. Toss everything
1880 * on the idle queue so we don't have to slog through
1881 * a bunch of uninteresting inodes over and over again.
1882 */
1883 if (flag & SYNC_CLOSE)
1884 ufs_idle_drain(NULL);
1885
1886 if (vfsp == NULL) {
1887 ufs_update(flag);
1888 return (0);
1889 }
1890
1891 /* Flush a single ufs */
1892 if (!vfs_matchops(vfsp, ufs_vfsops) || vfs_lock(vfsp) != 0)
1893 return (0);
1894
1895 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1896 if (!ufsvfsp)
1897 return (EIO);
1898 fs = ufsvfsp->vfs_fs;
1899 mutex_enter(&ufsvfsp->vfs_lock);
1900
1901 if (ufsvfsp->vfs_dio &&
1902 fs->fs_ronly == 0 &&
1903 fs->fs_clean != FSBAD &&
1904 fs->fs_clean != FSLOG) {
1905 /* turn off fast-io on unmount, so no fsck needed (4029401) */
1906 ufsvfsp->vfs_dio = 0;
1907 fs->fs_clean = FSACTIVE;
1908 fs->fs_fmod = 1;
1909 }
1910
1911 /* Write back modified superblock */
1912 if (fs->fs_fmod == 0) {
1913 mutex_exit(&ufsvfsp->vfs_lock);
1914 } else {
1915 if (fs->fs_ronly != 0) {
1916 mutex_exit(&ufsvfsp->vfs_lock);
1917 vfs_unlock(vfsp);
1918 return (ufs_fault(ufsvfsp->vfs_root,
1919 "fs = %s update: ro fs mod\n", fs->fs_fsmnt));
1920 }
1921 fs->fs_fmod = 0;
1922 mutex_exit(&ufsvfsp->vfs_lock);
1923
1924 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UPDATE);
1925 }
1926 vfs_unlock(vfsp);
1927
1928 /*
1929 * Avoid racing with ufs_update() and ufs_unmount().
1930 *
1931 */
1932 mutex_enter(&ufs_scan_lock);
1933
1934 (void) ufs_scan_inodes(1, ufs_sync_inode,
1935 (void *)(uintptr_t)cheap, ufsvfsp);
1936
1937 mutex_exit(&ufs_scan_lock);
1938
1939 bflush((dev_t)vfsp->vfs_dev);
1940
1941 /*
1942 * commit any outstanding async transactions
1943 */
1944 curthread->t_flag |= T_DONTBLOCK;
1945 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE, error);
1946 if (!error) {
1947 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UPDATE,
1948 TOP_COMMIT_SIZE);
1949 }
1950 curthread->t_flag &= ~T_DONTBLOCK;
1951
1952 return (0);
1953 }
1954
1955
1956 void
1957 sbupdate(struct vfs *vfsp)
1958 {
1959 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1960 struct fs *fs = ufsvfsp->vfs_fs;
1961 struct buf *bp;
1962 int blks;
1963 caddr_t space;
1964 int i;
1965 size_t size;
1966
1967 /*
1968 * for ulockfs processing, limit the superblock writes
1969 */
1970 if ((ufsvfsp->vfs_ulockfs.ul_sbowner) &&
1971 (curthread != ufsvfsp->vfs_ulockfs.ul_sbowner)) {
1972 /* process later */
1973 fs->fs_fmod = 1;
1974 return;
1975 }
1976 ULOCKFS_SET_MOD((&ufsvfsp->vfs_ulockfs));
1977
1978 if (TRANS_ISTRANS(ufsvfsp)) {
1979 mutex_enter(&ufsvfsp->vfs_lock);
1980 ufs_sbwrite(ufsvfsp);
1981 mutex_exit(&ufsvfsp->vfs_lock);
1982 return;
1983 }
1984
1985 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1986 space = (caddr_t)fs->fs_u.fs_csp;
1987 for (i = 0; i < blks; i += fs->fs_frag) {
1988 size = fs->fs_bsize;
1989 if (i + fs->fs_frag > blks)
1990 size = (blks - i) * fs->fs_fsize;
1991 bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev,
1992 (daddr_t)(fsbtodb(fs, fs->fs_csaddr + i)),
1993 fs->fs_bsize);
1994 bcopy(space, bp->b_un.b_addr, size);
1995 space += size;
1996 bp->b_bcount = size;
1997 UFS_BRWRITE(ufsvfsp, bp);
1998 }
1999 mutex_enter(&ufsvfsp->vfs_lock);
2000 ufs_sbwrite(ufsvfsp);
2001 mutex_exit(&ufsvfsp->vfs_lock);
2002 }
2003
2004 int ufs_vget_idle_count = 2; /* Number of inodes to idle each time */
2005 static int
2006 ufs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
2007 {
2008 int error = 0;
2009 struct ufid *ufid;
2010 struct inode *ip;
2011 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
2012 struct ulockfs *ulp;
2013
2014 /*
2015 * Check for unmounted filesystem.
2016 */
2017 if (vfsp->vfs_flag & VFS_UNMOUNTED) {
2018 error = EIO;
2019 goto errout;
2020 }
2021
2022 /*
2023 * Keep the idle queue from getting too long by
2024 * idling an inode before attempting to allocate another.
2025 * This operation must be performed before entering
2026 * lockfs or a transaction.
2027 */
2028 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2029 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2030 ins.in_vidles.value.ul += ufs_vget_idle_count;
2031 ufs_idle_some(ufs_vget_idle_count);
2032 }
2033
2034 ufid = (struct ufid *)fidp;
2035
2036 if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_VGET_MASK))
2037 goto errout;
2038
2039 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2040
2041 error = ufs_iget(vfsp, ufid->ufid_ino, &ip, CRED());
2042
2043 rw_exit(&ufsvfsp->vfs_dqrwlock);
2044
2045 ufs_lockfs_end(ulp);
2046
2047 if (error)
2048 goto errout;
2049
2050 /*
2051 * Check if the inode has been deleted or freed or is in transient state
2052 * since the last VFS_VGET() request for it, release it and don't return
2053 * it to the caller, presumably NFS, as it's no longer valid.
2054 */
2055 if (ip->i_gen != ufid->ufid_gen || ip->i_mode == 0 ||
2056 (ip->i_nlink <= 0)) {
2057 VN_RELE(ITOV(ip));
2058 error = EINVAL;
2059 goto errout;
2060 }
2061
2062 *vpp = ITOV(ip);
2063 return (0);
2064
2065 errout:
2066 *vpp = NULL;
2067 return (error);
2068 }
2069
2070 static int
2071 ufsinit(int fstype, char *name)
2072 {
2073 static const fs_operation_def_t ufs_vfsops_template[] = {
2074 { VFSNAME_MOUNT, { .vfs_mount = ufs_mount } },
2075 { VFSNAME_UNMOUNT, { .vfs_unmount = ufs_unmount } },
2076 { VFSNAME_ROOT, { .vfs_root = ufs_root } },
2077 { VFSNAME_STATVFS, { .vfs_statvfs = ufs_statvfs } },
2078 { VFSNAME_SYNC, { .vfs_sync = ufs_sync } },
2079 { VFSNAME_VGET, { .vfs_vget = ufs_vget } },
2080 { VFSNAME_MOUNTROOT, { .vfs_mountroot = ufs_mountroot } },
2081 { NULL, { NULL } }
2082 };
2083 int error;
2084
2085 ufsfstype = fstype;
2086
2087 error = vfs_setfsops(fstype, ufs_vfsops_template, &ufs_vfsops);
2088 if (error != 0) {
2089 cmn_err(CE_WARN, "ufsinit: bad vfs ops template");
2090 return (error);
2091 }
2092
2093 error = vn_make_ops(name, ufs_vnodeops_template, &ufs_vnodeops);
2094 if (error != 0) {
2095 (void) vfs_freevfsops_by_type(fstype);
2096 cmn_err(CE_WARN, "ufsinit: bad vnode ops template");
2097 return (error);
2098 }
2099
2100 ufs_iinit();
2101 return (0);
2102 }
2103
2104 #ifdef __sparc
2105
2106 /*
2107 * Mounting a mirrored SVM volume is only supported on ufs,
2108 * this is special-case boot code to support that configuration.
2109 * At this point, we have booted and mounted root on a
2110 * single component of the mirror. Complete the boot
2111 * by configuring SVM and converting the root to the
2112 * dev_t of the mirrored root device. This dev_t conversion
2113 * only works because the underlying device doesn't change.
2114 */
2115 int
2116 ufs_remountroot(struct vfs *vfsp)
2117 {
2118 struct ufsvfs *ufsvfsp;
2119 struct ulockfs *ulp;
2120 dev_t new_rootdev;
2121 dev_t old_rootdev;
2122 struct vnode *old_rootvp;
2123 struct vnode *new_rootvp;
2124 int error, sberror = 0;
2125 struct inode *ip;
2126 union ihead *ih;
2127 struct buf *bp;
2128 int i;
2129
2130 old_rootdev = rootdev;
2131 old_rootvp = rootvp;
2132
2133 new_rootdev = getrootdev();
2134 if (new_rootdev == (dev_t)NODEV) {
2135 return (ENODEV);
2136 }
2137
2138 new_rootvp = makespecvp(new_rootdev, VBLK);
2139
2140 error = VOP_OPEN(&new_rootvp,
2141 (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE, CRED(), NULL);
2142 if (error) {
2143 cmn_err(CE_CONT,
2144 "Cannot open mirrored root device, error %d\n", error);
2145 return (error);
2146 }
2147
2148 if (vfs_lock(vfsp) != 0) {
2149 return (EBUSY);
2150 }
2151
2152 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
2153 ulp = &ufsvfsp->vfs_ulockfs;
2154
2155 mutex_enter(&ulp->ul_lock);
2156 atomic_inc_ulong(&ufs_quiesce_pend);
2157
2158 (void) ufs_quiesce(ulp);
2159 (void) ufs_flush(vfsp);
2160
2161 /*
2162 * Convert root vfs to new dev_t, including vfs hash
2163 * table and fs id.
2164 */
2165 vfs_root_redev(vfsp, new_rootdev, ufsfstype);
2166
2167 ufsvfsp->vfs_devvp = new_rootvp;
2168 ufsvfsp->vfs_dev = new_rootdev;
2169
2170 bp = ufsvfsp->vfs_bufp;
2171 bp->b_edev = new_rootdev;
2172 bp->b_dev = cmpdev(new_rootdev);
2173
2174 /*
2175 * The buffer for the root inode does not contain a valid b_vp
2176 */
2177 (void) bfinval(new_rootdev, 0);
2178
2179 /*
2180 * Here we hand-craft inodes with old root device
2181 * references to refer to the new device instead.
2182 */
2183 mutex_enter(&ufs_scan_lock);
2184
2185 for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
2186 mutex_enter(&ih_lock[i]);
2187 for (ip = ih->ih_chain[0];
2188 ip != (struct inode *)ih;
2189 ip = ip->i_forw) {
2190 if (ip->i_ufsvfs != ufsvfsp)
2191 continue;
2192 if (ip == ufsvfsp->vfs_qinod)
2193 continue;
2194 if (ip->i_dev == old_rootdev) {
2195 ip->i_dev = new_rootdev;
2196 }
2197
2198 if (ip->i_devvp == old_rootvp) {
2199 ip->i_devvp = new_rootvp;
2200 }
2201 }
2202 mutex_exit(&ih_lock[i]);
2203 }
2204
2205 mutex_exit(&ufs_scan_lock);
2206
2207 /*
2208 * Make Sure logging structures are using the new device
2209 * if logging is enabled. Also start any logging thread that
2210 * needs to write to the device and couldn't earlier.
2211 */
2212 if (ufsvfsp->vfs_log) {
2213 buf_t *bp, *tbp;
2214 ml_unit_t *ul = ufsvfsp->vfs_log;
2215 struct fs *fsp = ufsvfsp->vfs_fs;
2216
2217 /*
2218 * Update the main logging structure.
2219 */
2220 ul->un_dev = new_rootdev;
2221
2222 /*
2223 * Get a new bp for the on disk structures.
2224 */
2225 bp = ul->un_bp;
2226 tbp = ngeteblk(dbtob(LS_SECTORS));
2227 tbp->b_edev = new_rootdev;
2228 tbp->b_dev = cmpdev(new_rootdev);
2229 tbp->b_blkno = bp->b_blkno;
2230 bcopy(bp->b_un.b_addr, tbp->b_un.b_addr, DEV_BSIZE);
2231 bcopy(bp->b_un.b_addr, tbp->b_un.b_addr + DEV_BSIZE, DEV_BSIZE);
2232 bp->b_flags |= (B_STALE | B_AGE);
2233 brelse(bp);
2234 ul->un_bp = tbp;
2235
2236 /*
2237 * Allocate new circular buffers.
2238 */
2239 alloc_rdbuf(&ul->un_rdbuf, MAPBLOCKSIZE, MAPBLOCKSIZE);
2240 alloc_wrbuf(&ul->un_wrbuf, ldl_bufsize(ul));
2241
2242 /*
2243 * Clear the noroll bit which indicates that logging
2244 * can't roll the log yet and start the logmap roll thread
2245 * unless the filesystem is still read-only in which case
2246 * remountfs() will do it when going to read-write.
2247 */
2248 ASSERT(ul->un_flags & LDL_NOROLL);
2249
2250 if (!fsp->fs_ronly) {
2251 ul->un_flags &= ~LDL_NOROLL;
2252 logmap_start_roll(ul);
2253 }
2254
2255 /*
2256 * Start the reclaim thread if needed.
2257 */
2258 if (!fsp->fs_ronly && (fsp->fs_reclaim &
2259 (FS_RECLAIM|FS_RECLAIMING))) {
2260 fsp->fs_reclaim &= ~FS_RECLAIM;
2261 fsp->fs_reclaim |= FS_RECLAIMING;
2262 ufs_thread_start(&ufsvfsp->vfs_reclaim,
2263 ufs_thread_reclaim, vfsp);
2264 TRANS_SBWRITE(ufsvfsp, TOP_SBUPDATE_UPDATE);
2265 if (sberror = geterror(ufsvfsp->vfs_bufp)) {
2266 refstr_t *mntpt;
2267 mntpt = vfs_getmntpoint(vfsp);
2268 cmn_err(CE_WARN,
2269 "Remountroot failed to update Reclaim"
2270 "state for filesystem %s "
2271 "Error writing SuperBlock %d",
2272 refstr_value(mntpt), error);
2273 refstr_rele(mntpt);
2274 }
2275 }
2276 }
2277
2278 rootdev = new_rootdev;
2279 rootvp = new_rootvp;
2280
2281 atomic_dec_ulong(&ufs_quiesce_pend);
2282 cv_broadcast(&ulp->ul_cv);
2283 mutex_exit(&ulp->ul_lock);
2284
2285 vfs_unlock(vfsp);
2286
2287 error = VOP_CLOSE(old_rootvp, FREAD, 1, (offset_t)0, CRED(), NULL);
2288 if (error) {
2289 cmn_err(CE_CONT,
2290 "close of root device component failed, error %d\n",
2291 error);
2292 }
2293 VN_RELE(old_rootvp);
2294
2295 return (sberror ? sberror : error);
2296 }
2297
2298 #endif /* __sparc */