Print this page
re #13613 rb4516 Tunables needs volatile keyword
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/ufs/ufs_vfsops.c
+++ new/usr/src/uts/common/fs/ufs/ufs_vfsops.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
|
↓ open down ↓ |
17 lines elided |
↑ open up ↑ |
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 24 * Use is subject to license terms.
25 25 * Copyright 2016 Nexenta Systems, Inc.
26 26 * Copyright (c) 2017 by Delphix. All rights reserved.
27 27 */
28 +/*
29 + * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
30 + */
28 31
29 32 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
30 33 /* All Rights Reserved */
31 34
32 35 /*
33 36 * University Copyright- Copyright (c) 1982, 1986, 1988
34 37 * The Regents of the University of California
35 38 * All Rights Reserved
36 39 *
37 40 * University Acknowledgment- Portions of this document are derived from
38 41 * software developed by the University of California, Berkeley, and its
39 42 * contributors.
40 43 */
41 44
42 45 #include <sys/types.h>
43 46 #include <sys/t_lock.h>
44 47 #include <sys/param.h>
45 48 #include <sys/systm.h>
46 49 #include <sys/bitmap.h>
47 50 #include <sys/sysmacros.h>
48 51 #include <sys/kmem.h>
49 52 #include <sys/signal.h>
50 53 #include <sys/user.h>
51 54 #include <sys/proc.h>
52 55 #include <sys/disp.h>
53 56 #include <sys/buf.h>
54 57 #include <sys/pathname.h>
55 58 #include <sys/vfs.h>
56 59 #include <sys/vfs_opreg.h>
57 60 #include <sys/vnode.h>
58 61 #include <sys/file.h>
59 62 #include <sys/atomic.h>
60 63 #include <sys/uio.h>
61 64 #include <sys/dkio.h>
62 65 #include <sys/cred.h>
63 66 #include <sys/conf.h>
64 67 #include <sys/dnlc.h>
65 68 #include <sys/kstat.h>
66 69 #include <sys/acl.h>
67 70 #include <sys/fs/ufs_fsdir.h>
68 71 #include <sys/fs/ufs_fs.h>
69 72 #include <sys/fs/ufs_inode.h>
70 73 #include <sys/fs/ufs_mount.h>
71 74 #include <sys/fs/ufs_acl.h>
72 75 #include <sys/fs/ufs_panic.h>
73 76 #include <sys/fs/ufs_bio.h>
74 77 #include <sys/fs/ufs_quota.h>
75 78 #include <sys/fs/ufs_log.h>
76 79 #undef NFS
77 80 #include <sys/statvfs.h>
78 81 #include <sys/mount.h>
79 82 #include <sys/mntent.h>
80 83 #include <sys/swap.h>
81 84 #include <sys/errno.h>
82 85 #include <sys/debug.h>
83 86 #include "fs/fs_subr.h"
84 87 #include <sys/cmn_err.h>
85 88 #include <sys/dnlc.h>
86 89 #include <sys/fssnap_if.h>
87 90 #include <sys/sunddi.h>
88 91 #include <sys/bootconf.h>
89 92 #include <sys/policy.h>
90 93 #include <sys/zone.h>
91 94
92 95 /*
93 96 * This is the loadable module wrapper.
94 97 */
95 98 #include <sys/modctl.h>
96 99
97 100 int ufsfstype;
98 101 vfsops_t *ufs_vfsops;
99 102 static int ufsinit(int, char *);
100 103 static int mountfs();
101 104 extern int highbit();
102 105 extern struct instats ins;
103 106 extern struct vnode *common_specvp(struct vnode *vp);
104 107 extern vfs_t EIO_vfs;
105 108
106 109 struct dquot *dquot, *dquotNDQUOT;
107 110
108 111 /*
109 112 * Cylinder group summary information handling tunable.
110 113 * This defines when these deltas get logged.
111 114 * If the number of cylinders in the file system is over the
112 115 * tunable then we log csum updates. Otherwise the updates are only
113 116 * done for performance on unmount. After a panic they can be
114 117 * quickly constructed during mounting. See ufs_construct_si()
115 118 * called from ufs_getsummaryinfo().
116 119 *
117 120 * This performance feature can of course be disabled by setting
118 121 * ufs_ncg_log to 0, and fully enabled by setting it to 0xffffffff.
119 122 */
120 123 #define UFS_LOG_NCG_DEFAULT 10000
121 124 uint32_t ufs_ncg_log = UFS_LOG_NCG_DEFAULT;
122 125
123 126 /*
124 127 * ufs_clean_root indicates whether the root fs went down cleanly
125 128 */
126 129 static int ufs_clean_root = 0;
127 130
128 131 /*
129 132 * UFS Mount options table
130 133 */
131 134 static char *intr_cancel[] = { MNTOPT_NOINTR, NULL };
132 135 static char *nointr_cancel[] = { MNTOPT_INTR, NULL };
133 136 static char *forcedirectio_cancel[] = { MNTOPT_NOFORCEDIRECTIO, NULL };
134 137 static char *noforcedirectio_cancel[] = { MNTOPT_FORCEDIRECTIO, NULL };
135 138 static char *largefiles_cancel[] = { MNTOPT_NOLARGEFILES, NULL };
136 139 static char *nolargefiles_cancel[] = { MNTOPT_LARGEFILES, NULL };
137 140 static char *logging_cancel[] = { MNTOPT_NOLOGGING, NULL };
138 141 static char *nologging_cancel[] = { MNTOPT_LOGGING, NULL };
139 142 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
140 143 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
141 144 static char *quota_cancel[] = { MNTOPT_NOQUOTA, NULL };
142 145 static char *noquota_cancel[] = { MNTOPT_QUOTA, NULL };
143 146 static char *dfratime_cancel[] = { MNTOPT_NODFRATIME, NULL };
144 147 static char *nodfratime_cancel[] = { MNTOPT_DFRATIME, NULL };
145 148
146 149 static mntopt_t mntopts[] = {
147 150 /*
148 151 * option name cancel option default arg flags
149 152 * ufs arg flag
150 153 */
151 154 { MNTOPT_INTR, intr_cancel, NULL, MO_DEFAULT,
152 155 (void *)0 },
153 156 { MNTOPT_NOINTR, nointr_cancel, NULL, 0,
154 157 (void *)UFSMNT_NOINTR },
155 158 { MNTOPT_SYNCDIR, NULL, NULL, 0,
156 159 (void *)UFSMNT_SYNCDIR },
157 160 { MNTOPT_FORCEDIRECTIO, forcedirectio_cancel, NULL, 0,
158 161 (void *)UFSMNT_FORCEDIRECTIO },
159 162 { MNTOPT_NOFORCEDIRECTIO, noforcedirectio_cancel, NULL, 0,
160 163 (void *)UFSMNT_NOFORCEDIRECTIO },
161 164 { MNTOPT_NOSETSEC, NULL, NULL, 0,
162 165 (void *)UFSMNT_NOSETSEC },
163 166 { MNTOPT_LARGEFILES, largefiles_cancel, NULL, MO_DEFAULT,
164 167 (void *)UFSMNT_LARGEFILES },
165 168 { MNTOPT_NOLARGEFILES, nolargefiles_cancel, NULL, 0,
166 169 (void *)0 },
167 170 { MNTOPT_LOGGING, logging_cancel, NULL, MO_TAG,
168 171 (void *)UFSMNT_LOGGING },
169 172 { MNTOPT_NOLOGGING, nologging_cancel, NULL,
170 173 MO_NODISPLAY|MO_DEFAULT|MO_TAG, (void *)0 },
171 174 { MNTOPT_QUOTA, quota_cancel, NULL, MO_IGNORE,
172 175 (void *)0 },
173 176 { MNTOPT_NOQUOTA, noquota_cancel, NULL,
174 177 MO_NODISPLAY|MO_DEFAULT, (void *)0 },
175 178 { MNTOPT_GLOBAL, NULL, NULL, 0,
176 179 (void *)0 },
177 180 { MNTOPT_XATTR, xattr_cancel, NULL, MO_DEFAULT,
178 181 (void *)0 },
179 182 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0,
180 183 (void *)0 },
181 184 { MNTOPT_NOATIME, NULL, NULL, 0,
182 185 (void *)UFSMNT_NOATIME },
183 186 { MNTOPT_DFRATIME, dfratime_cancel, NULL, 0,
184 187 (void *)0 },
185 188 { MNTOPT_NODFRATIME, nodfratime_cancel, NULL,
186 189 MO_NODISPLAY|MO_DEFAULT, (void *)UFSMNT_NODFRATIME },
187 190 { MNTOPT_ONERROR, NULL, UFSMNT_ONERROR_PANIC_STR,
188 191 MO_DEFAULT|MO_HASVALUE, (void *)0 },
189 192 };
190 193
191 194 static mntopts_t ufs_mntopts = {
192 195 sizeof (mntopts) / sizeof (mntopt_t),
193 196 mntopts
194 197 };
195 198
196 199 static vfsdef_t vfw = {
197 200 VFSDEF_VERSION,
198 201 "ufs",
199 202 ufsinit,
200 203 VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_CANLOFI|VSW_MOUNTDEV,
201 204 &ufs_mntopts
202 205 };
203 206
204 207 /*
205 208 * Module linkage information for the kernel.
206 209 */
207 210 extern struct mod_ops mod_fsops;
208 211
209 212 static struct modlfs modlfs = {
210 213 &mod_fsops, "filesystem for ufs", &vfw
211 214 };
212 215
213 216 static struct modlinkage modlinkage = {
214 217 MODREV_1, (void *)&modlfs, NULL
215 218 };
216 219
217 220 /*
218 221 * An attempt has been made to make this module unloadable. In order to
219 222 * test it, we need a system in which the root fs is NOT ufs. THIS HAS NOT
220 223 * BEEN DONE
221 224 */
222 225
223 226 extern kstat_t *ufs_inode_kstat;
224 227 extern uint_t ufs_lockfs_key;
225 228 extern void ufs_lockfs_tsd_destructor(void *);
226 229 extern uint_t bypass_snapshot_throttle_key;
227 230
228 231 int
229 232 _init(void)
230 233 {
231 234 /*
232 235 * Create an index into the per thread array so that any thread doing
233 236 * VOP will have a lockfs mark on it.
234 237 */
235 238 tsd_create(&ufs_lockfs_key, ufs_lockfs_tsd_destructor);
236 239 tsd_create(&bypass_snapshot_throttle_key, NULL);
237 240 return (mod_install(&modlinkage));
238 241 }
239 242
240 243 int
241 244 _fini(void)
242 245 {
243 246 return (EBUSY);
244 247 }
245 248
246 249 int
247 250 _info(struct modinfo *modinfop)
248 251 {
249 252 return (mod_info(&modlinkage, modinfop));
250 253 }
251 254
252 255 extern struct vnode *makespecvp(dev_t dev, vtype_t type);
253 256
254 257 extern kmutex_t ufs_scan_lock;
255 258
256 259 static int mountfs(struct vfs *, enum whymountroot, struct vnode *, char *,
257 260 struct cred *, int, void *, int);
258 261
259 262
260 263 static int
261 264 ufs_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap,
262 265 struct cred *cr)
263 266 {
264 267 char *data = uap->dataptr;
265 268 int datalen = uap->datalen;
266 269 dev_t dev;
267 270 struct vnode *lvp = NULL;
268 271 struct vnode *svp = NULL;
269 272 struct pathname dpn;
270 273 int error;
271 274 enum whymountroot why = ROOT_INIT;
272 275 struct ufs_args args;
273 276 int oflag, aflag;
274 277 int fromspace = (uap->flags & MS_SYSSPACE) ?
275 278 UIO_SYSSPACE : UIO_USERSPACE;
276 279
277 280 if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
278 281 return (error);
279 282
280 283 if (mvp->v_type != VDIR)
281 284 return (ENOTDIR);
282 285
283 286 mutex_enter(&mvp->v_lock);
284 287 if ((uap->flags & MS_REMOUNT) == 0 &&
285 288 (uap->flags & MS_OVERLAY) == 0 &&
286 289 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
287 290 mutex_exit(&mvp->v_lock);
288 291 return (EBUSY);
289 292 }
290 293 mutex_exit(&mvp->v_lock);
291 294
292 295 /*
293 296 * Get arguments
294 297 */
295 298 bzero(&args, sizeof (args));
296 299 if ((uap->flags & MS_DATA) && data != NULL && datalen != 0) {
297 300 int copy_result = 0;
298 301
299 302 if (datalen > sizeof (args))
300 303 return (EINVAL);
301 304 if (uap->flags & MS_SYSSPACE)
302 305 bcopy(data, &args, datalen);
303 306 else
304 307 copy_result = copyin(data, &args, datalen);
305 308 if (copy_result)
306 309 return (EFAULT);
307 310 datalen = sizeof (struct ufs_args);
308 311 } else {
309 312 datalen = 0;
310 313 }
311 314
312 315 if ((vfsp->vfs_flag & VFS_RDONLY) != 0 ||
313 316 (uap->flags & MS_RDONLY) != 0) {
314 317 oflag = FREAD;
315 318 aflag = VREAD;
316 319 } else {
317 320 oflag = FREAD | FWRITE;
318 321 aflag = VREAD | VWRITE;
319 322 }
320 323
321 324 /*
322 325 * Read in the mount point pathname
323 326 * (so we can record the directory the file system was last mounted on).
324 327 */
325 328 if (error = pn_get(uap->dir, fromspace, &dpn))
326 329 return (error);
327 330
328 331 /*
329 332 * Resolve path name of special file being mounted.
330 333 */
331 334 if (error = lookupname(uap->spec, fromspace, FOLLOW, NULL, &svp)) {
332 335 pn_free(&dpn);
333 336 return (error);
334 337 }
335 338
336 339 error = vfs_get_lofi(vfsp, &lvp);
337 340
338 341 if (error > 0) {
339 342 VN_RELE(svp);
340 343 pn_free(&dpn);
341 344 return (error);
342 345 } else if (error == 0) {
343 346 dev = lvp->v_rdev;
344 347
345 348 if (getmajor(dev) >= devcnt) {
346 349 error = ENXIO;
347 350 goto out;
348 351 }
349 352 } else {
350 353 dev = svp->v_rdev;
351 354
352 355 if (svp->v_type != VBLK) {
353 356 VN_RELE(svp);
354 357 pn_free(&dpn);
355 358 return (ENOTBLK);
356 359 }
357 360
358 361 if (getmajor(dev) >= devcnt) {
359 362 error = ENXIO;
360 363 goto out;
361 364 }
362 365
363 366 /*
364 367 * In SunCluster, requests to a global device are
365 368 * satisfied by a local device. We substitute the global
366 369 * pxfs node with a local spec node here.
367 370 */
368 371 if (IS_PXFSVP(svp)) {
369 372 ASSERT(lvp == NULL);
370 373 VN_RELE(svp);
371 374 svp = makespecvp(dev, VBLK);
372 375 }
373 376
374 377 if ((error = secpolicy_spec_open(cr, svp, oflag)) != 0) {
375 378 VN_RELE(svp);
376 379 pn_free(&dpn);
377 380 return (error);
378 381 }
379 382 }
380 383
381 384 if (uap->flags & MS_REMOUNT)
382 385 why = ROOT_REMOUNT;
383 386
384 387 /*
385 388 * Open device/file mounted on. We need this to check whether
386 389 * the caller has sufficient rights to access the resource in
387 390 * question. When bio is fixed for vnodes this can all be vnode
388 391 * operations.
389 392 */
390 393 if ((error = VOP_ACCESS(svp, aflag, 0, cr, NULL)) != 0)
391 394 goto out;
392 395
393 396 /*
394 397 * Ensure that this device isn't already mounted or in progress on a
395 398 * mount unless this is a REMOUNT request or we are told to suppress
396 399 * mount checks. Global mounts require special handling.
397 400 */
398 401 if ((uap->flags & MS_NOCHECK) == 0) {
399 402 if ((uap->flags & MS_GLOBAL) == 0 &&
400 403 vfs_devmounting(dev, vfsp)) {
401 404 error = EBUSY;
402 405 goto out;
403 406 }
404 407 if (vfs_devismounted(dev)) {
405 408 if ((uap->flags & MS_REMOUNT) == 0) {
406 409 error = EBUSY;
407 410 goto out;
408 411 }
409 412 }
410 413 }
411 414
412 415 /*
413 416 * If the device is a tape, mount it read only
414 417 */
415 418 if (devopsp[getmajor(dev)]->devo_cb_ops->cb_flag & D_TAPE) {
416 419 vfsp->vfs_flag |= VFS_RDONLY;
417 420 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
418 421 }
419 422 if (uap->flags & MS_RDONLY)
420 423 vfsp->vfs_flag |= VFS_RDONLY;
421 424
422 425 /*
423 426 * Mount the filesystem, free the device vnode on error.
424 427 */
425 428 error = mountfs(vfsp, why, lvp != NULL ? lvp : svp,
426 429 dpn.pn_path, cr, 0, &args, datalen);
427 430
428 431 if (error == 0) {
429 432 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS);
430 433
431 434 /*
432 435 * If lofi, drop our reference to the original file.
433 436 */
434 437 if (lvp != NULL)
435 438 VN_RELE(svp);
436 439 }
437 440
438 441 out:
439 442 pn_free(&dpn);
440 443
441 444 if (error) {
442 445 if (lvp != NULL)
443 446 VN_RELE(lvp);
444 447 if (svp != NULL)
445 448 VN_RELE(svp);
446 449 }
447 450 return (error);
448 451 }
449 452
450 453 /*
451 454 * Mount root file system.
452 455 * "why" is ROOT_INIT on initial call ROOT_REMOUNT if called to
453 456 * remount the root file system, and ROOT_UNMOUNT if called to
454 457 * unmount the root (e.g., as part of a system shutdown).
455 458 *
456 459 * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP
457 460 * operation, goes along with auto-configuration. A mechanism should be
458 461 * provided by which machine-INdependent code in the kernel can say "get me the
459 462 * right root file system" and "get me the right initial swap area", and have
460 463 * that done in what may well be a machine-dependent fashion.
461 464 * Unfortunately, it is also file-system-type dependent (NFS gets it via
462 465 * bootparams calls, UFS gets it from various and sundry machine-dependent
463 466 * mechanisms, as SPECFS does for swap).
464 467 */
465 468 static int
466 469 ufs_mountroot(struct vfs *vfsp, enum whymountroot why)
467 470 {
468 471 struct fs *fsp;
469 472 int error;
470 473 static int ufsrootdone = 0;
471 474 dev_t rootdev;
472 475 struct vnode *vp;
473 476 struct vnode *devvp = 0;
474 477 int ovflags;
475 478 int doclkset;
476 479 ufsvfs_t *ufsvfsp;
477 480
478 481 if (why == ROOT_INIT) {
479 482 if (ufsrootdone++)
480 483 return (EBUSY);
481 484 rootdev = getrootdev();
482 485 if (rootdev == (dev_t)NODEV)
483 486 return (ENODEV);
484 487 vfsp->vfs_dev = rootdev;
485 488 vfsp->vfs_flag |= VFS_RDONLY;
486 489 } else if (why == ROOT_REMOUNT) {
487 490 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
488 491 (void) dnlc_purge_vfsp(vfsp, 0);
489 492 vp = common_specvp(vp);
490 493 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_INVAL,
491 494 CRED(), NULL);
492 495 (void) bfinval(vfsp->vfs_dev, 0);
493 496 fsp = getfs(vfsp);
494 497
495 498 ovflags = vfsp->vfs_flag;
496 499 vfsp->vfs_flag &= ~VFS_RDONLY;
497 500 vfsp->vfs_flag |= VFS_REMOUNT;
498 501 rootdev = vfsp->vfs_dev;
499 502 } else if (why == ROOT_UNMOUNT) {
500 503 if (vfs_lock(vfsp) == 0) {
501 504 (void) ufs_flush(vfsp);
502 505 /*
503 506 * Mark the log as fully rolled
504 507 */
505 508 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
506 509 fsp = ufsvfsp->vfs_fs;
507 510 if (TRANS_ISTRANS(ufsvfsp) &&
508 511 !TRANS_ISERROR(ufsvfsp) &&
509 512 (fsp->fs_rolled == FS_NEED_ROLL)) {
510 513 ml_unit_t *ul = ufsvfsp->vfs_log;
511 514
512 515 error = ufs_putsummaryinfo(ul->un_dev,
513 516 ufsvfsp, fsp);
514 517 if (error == 0) {
515 518 fsp->fs_rolled = FS_ALL_ROLLED;
516 519 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
517 520 }
518 521 }
519 522 vfs_unlock(vfsp);
520 523 } else {
521 524 ufs_update(0);
522 525 }
523 526
524 527 vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
525 528 (void) VOP_CLOSE(vp, FREAD|FWRITE, 1,
526 529 (offset_t)0, CRED(), NULL);
527 530 return (0);
528 531 }
529 532 error = vfs_lock(vfsp);
530 533 if (error)
531 534 return (error);
532 535
533 536 devvp = makespecvp(rootdev, VBLK);
534 537
535 538 /* If RO media, don't call clkset() (see below) */
536 539 doclkset = 1;
537 540 if (why == ROOT_INIT) {
538 541 error = VOP_OPEN(&devvp, FREAD|FWRITE, CRED(), NULL);
539 542 if (error == 0) {
540 543 (void) VOP_CLOSE(devvp, FREAD|FWRITE, 1,
541 544 (offset_t)0, CRED(), NULL);
542 545 } else {
543 546 doclkset = 0;
544 547 }
545 548 }
546 549
547 550 error = mountfs(vfsp, why, devvp, "/", CRED(), 1, NULL, 0);
548 551 /*
549 552 * XXX - assumes root device is not indirect, because we don't set
550 553 * rootvp. Is rootvp used for anything? If so, make another arg
551 554 * to mountfs.
552 555 */
553 556 if (error) {
554 557 vfs_unlock(vfsp);
555 558 if (why == ROOT_REMOUNT)
556 559 vfsp->vfs_flag = ovflags;
557 560 if (rootvp) {
558 561 VN_RELE(rootvp);
559 562 rootvp = (struct vnode *)0;
560 563 }
561 564 VN_RELE(devvp);
562 565 return (error);
563 566 }
564 567 if (why == ROOT_INIT)
565 568 vfs_add((struct vnode *)0, vfsp,
566 569 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
567 570 vfs_unlock(vfsp);
568 571 fsp = getfs(vfsp);
569 572 clkset(doclkset ? fsp->fs_time : -1);
570 573 ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
571 574 if (ufsvfsp->vfs_log) {
572 575 vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
573 576 }
574 577 return (0);
575 578 }
576 579
577 580 static int
578 581 remountfs(struct vfs *vfsp, dev_t dev, void *raw_argsp, int args_len)
579 582 {
580 583 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
581 584 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs;
582 585 struct buf *bp = ufsvfsp->vfs_bufp;
583 586 struct fs *fsp = (struct fs *)bp->b_un.b_addr;
584 587 struct fs *fspt;
585 588 struct buf *tpt = 0;
586 589 int error = 0;
587 590 int flags = 0;
588 591
589 592 if (args_len == sizeof (struct ufs_args) && raw_argsp)
590 593 flags = ((struct ufs_args *)raw_argsp)->flags;
591 594
592 595 /* cannot remount to RDONLY */
593 596 if (vfsp->vfs_flag & VFS_RDONLY)
594 597 return (ENOTSUP);
595 598
596 599 /* whoops, wrong dev */
597 600 if (vfsp->vfs_dev != dev)
598 601 return (EINVAL);
599 602
600 603 /*
601 604 * synchronize w/ufs ioctls
602 605 */
603 606 mutex_enter(&ulp->ul_lock);
604 607 atomic_inc_ulong(&ufs_quiesce_pend);
605 608
606 609 /*
607 610 * reset options
608 611 */
609 612 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR;
610 613 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
611 614 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
612 615 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
613 616 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
614 617 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
615 618 else /* dfratime, default behavior */
616 619 ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
617 620 if (flags & UFSMNT_FORCEDIRECTIO)
618 621 ufsvfsp->vfs_forcedirectio = 1;
619 622 else /* default is no direct I/O */
620 623 ufsvfsp->vfs_forcedirectio = 0;
621 624 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
622 625
623 626 /*
624 627 * set largefiles flag in ufsvfs equal to the
625 628 * value passed in by the mount command. If
626 629 * it is "nolargefiles", and the flag is set
627 630 * in the superblock, the mount fails.
628 631 */
629 632 if (!(flags & UFSMNT_LARGEFILES)) { /* "nolargefiles" */
630 633 if (fsp->fs_flags & FSLARGEFILES) {
631 634 error = EFBIG;
632 635 goto remounterr;
633 636 }
634 637 ufsvfsp->vfs_lfflags &= ~UFS_LARGEFILES;
635 638 } else /* "largefiles" */
636 639 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
637 640 /*
638 641 * read/write to read/write; all done
639 642 */
640 643 if (fsp->fs_ronly == 0)
641 644 goto remounterr;
642 645
643 646 /*
644 647 * fix-on-panic assumes RO->RW remount implies system-critical fs
645 648 * if it is shortly after boot; so, don't attempt to lock and fix
646 649 * (unless the user explicitly asked for another action on error)
647 650 * XXX UFSMNT_ONERROR_RDONLY rather than UFSMNT_ONERROR_PANIC
648 651 */
649 652 #define BOOT_TIME_LIMIT (180*hz)
650 653 if (!(flags & UFSMNT_ONERROR_FLGMASK) &&
651 654 ddi_get_lbolt() < BOOT_TIME_LIMIT) {
652 655 cmn_err(CE_WARN, "%s is required to be mounted onerror=%s",
653 656 ufsvfsp->vfs_fs->fs_fsmnt, UFSMNT_ONERROR_PANIC_STR);
654 657 flags |= UFSMNT_ONERROR_PANIC;
655 658 }
656 659
657 660 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
658 661 goto remounterr;
659 662
660 663 /*
661 664 * quiesce the file system
662 665 */
663 666 error = ufs_quiesce(ulp);
664 667 if (error)
665 668 goto remounterr;
666 669
667 670 tpt = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, SBLOCK, SBSIZE);
668 671 if (tpt->b_flags & B_ERROR) {
669 672 error = EIO;
670 673 goto remounterr;
671 674 }
672 675 fspt = (struct fs *)tpt->b_un.b_addr;
673 676 if (((fspt->fs_magic != FS_MAGIC) &&
674 677 (fspt->fs_magic != MTB_UFS_MAGIC)) ||
675 678 (fspt->fs_magic == FS_MAGIC &&
676 679 (fspt->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
677 680 fspt->fs_version != UFS_VERSION_MIN)) ||
678 681 (fspt->fs_magic == MTB_UFS_MAGIC &&
679 682 (fspt->fs_version > MTB_UFS_VERSION_1 ||
680 683 fspt->fs_version < MTB_UFS_VERSION_MIN)) ||
681 684 fspt->fs_bsize > MAXBSIZE || fspt->fs_frag > MAXFRAG ||
682 685 fspt->fs_bsize < sizeof (struct fs) || fspt->fs_bsize < PAGESIZE) {
683 686 tpt->b_flags |= B_STALE | B_AGE;
684 687 error = EINVAL;
685 688 goto remounterr;
686 689 }
687 690
688 691 if (ufsvfsp->vfs_log && (ufsvfsp->vfs_log->un_flags & LDL_NOROLL)) {
689 692 ufsvfsp->vfs_log->un_flags &= ~LDL_NOROLL;
690 693 logmap_start_roll(ufsvfsp->vfs_log);
691 694 }
692 695
693 696 if (TRANS_ISERROR(ufsvfsp))
694 697 goto remounterr;
695 698 TRANS_DOMATAMAP(ufsvfsp);
696 699
697 700 if ((fspt->fs_state + fspt->fs_time == FSOKAY) &&
698 701 fspt->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp)) {
699 702 ufsvfsp->vfs_log = NULL;
700 703 ufsvfsp->vfs_domatamap = 0;
701 704 error = ENOSPC;
702 705 goto remounterr;
703 706 }
704 707
705 708 if (fspt->fs_state + fspt->fs_time == FSOKAY &&
706 709 (fspt->fs_clean == FSCLEAN ||
707 710 fspt->fs_clean == FSSTABLE ||
708 711 fspt->fs_clean == FSLOG)) {
709 712
710 713 /*
711 714 * Ensure that ufs_getsummaryinfo doesn't reconstruct
712 715 * the summary info.
713 716 */
714 717 error = ufs_getsummaryinfo(vfsp->vfs_dev, ufsvfsp, fspt);
715 718 if (error)
716 719 goto remounterr;
717 720
718 721 /* preserve mount name */
719 722 (void) strncpy(fspt->fs_fsmnt, fsp->fs_fsmnt, MAXMNTLEN);
720 723 /* free the old cg space */
721 724 kmem_free(fsp->fs_u.fs_csp, fsp->fs_cssize);
722 725 /* switch in the new superblock */
723 726 fspt->fs_rolled = FS_NEED_ROLL;
724 727 bcopy(tpt->b_un.b_addr, bp->b_un.b_addr, fspt->fs_sbsize);
725 728
726 729 fsp->fs_clean = FSSTABLE;
727 730 } /* superblock updated in memory */
728 731 tpt->b_flags |= B_STALE | B_AGE;
729 732 brelse(tpt);
730 733 tpt = 0;
731 734
732 735 if (fsp->fs_clean != FSSTABLE) {
733 736 error = ENOSPC;
734 737 goto remounterr;
735 738 }
736 739
737 740
738 741 if (TRANS_ISTRANS(ufsvfsp)) {
739 742 fsp->fs_clean = FSLOG;
740 743 ufsvfsp->vfs_dio = 0;
741 744 } else
742 745 if (ufsvfsp->vfs_dio)
743 746 fsp->fs_clean = FSSUSPEND;
744 747
745 748 TRANS_MATA_MOUNT(ufsvfsp);
746 749
747 750 fsp->fs_fmod = 0;
748 751 fsp->fs_ronly = 0;
749 752
750 753 atomic_dec_ulong(&ufs_quiesce_pend);
751 754 cv_broadcast(&ulp->ul_cv);
752 755 mutex_exit(&ulp->ul_lock);
753 756
754 757 if (TRANS_ISTRANS(ufsvfsp)) {
755 758
756 759 /*
757 760 * start the delete thread
758 761 */
759 762 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
760 763
761 764 /*
762 765 * start the reclaim thread
763 766 */
764 767 if (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
765 768 fsp->fs_reclaim &= ~FS_RECLAIM;
766 769 fsp->fs_reclaim |= FS_RECLAIMING;
767 770 ufs_thread_start(&ufsvfsp->vfs_reclaim,
768 771 ufs_thread_reclaim, vfsp);
769 772 }
770 773 }
771 774
772 775 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
773 776
774 777 return (0);
775 778
776 779 remounterr:
777 780 if (tpt)
778 781 brelse(tpt);
779 782 atomic_dec_ulong(&ufs_quiesce_pend);
780 783 cv_broadcast(&ulp->ul_cv);
781 784 mutex_exit(&ulp->ul_lock);
782 785 return (error);
783 786 }
784 787
785 788 /*
786 789 * If the device maxtransfer size is not available, we use ufs_maxmaxphys
787 790 * along with the system value for maxphys to determine the value for
788 791 * maxtransfer.
789 792 */
790 793 int ufs_maxmaxphys = (1024 * 1024);
791 794
792 795 #include <sys/ddi.h> /* for delay(9f) */
793 796
794 797 int ufs_mount_error_delay = 20; /* default to 20ms */
795 798 int ufs_mount_timeout = 60000; /* default to 1 minute */
796 799
797 800 static int
798 801 mountfs(struct vfs *vfsp, enum whymountroot why, struct vnode *devvp,
799 802 char *path, cred_t *cr, int isroot, void *raw_argsp, int args_len)
800 803 {
801 804 dev_t dev = devvp->v_rdev;
802 805 struct fs *fsp;
803 806 struct ufsvfs *ufsvfsp = 0;
804 807 struct buf *bp = 0;
805 808 struct buf *tp = 0;
806 809 struct dk_cinfo ci;
|
↓ open down ↓ |
769 lines elided |
↑ open up ↑ |
807 810 int error = 0;
808 811 size_t len;
809 812 int needclose = 0;
810 813 int needtrans = 0;
811 814 struct inode *rip;
812 815 struct vnode *rvp = NULL;
813 816 int flags = 0;
814 817 kmutex_t *ihm;
815 818 int elapsed;
816 819 int status;
817 - extern int maxphys;
818 820
819 821 if (args_len == sizeof (struct ufs_args) && raw_argsp)
820 822 flags = ((struct ufs_args *)raw_argsp)->flags;
821 823
822 824 ASSERT(vfs_lock_held(vfsp));
823 825
824 826 if (why == ROOT_INIT) {
825 827 /*
826 828 * Open block device mounted on.
827 829 * When bio is fixed for vnodes this can all be vnode
828 830 * operations.
829 831 */
830 832 error = VOP_OPEN(&devvp,
831 833 (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE,
832 834 cr, NULL);
833 835 if (error)
834 836 goto out;
835 837 needclose = 1;
836 838
837 839 /*
838 840 * Refuse to go any further if this
839 841 * device is being used for swapping.
840 842 */
841 843 if (IS_SWAPVP(devvp)) {
842 844 error = EBUSY;
843 845 goto out;
844 846 }
845 847 }
846 848
847 849 /*
848 850 * check for dev already mounted on
849 851 */
850 852 if (vfsp->vfs_flag & VFS_REMOUNT) {
851 853 error = remountfs(vfsp, dev, raw_argsp, args_len);
852 854 if (error == 0)
853 855 VN_RELE(devvp);
854 856 return (error);
855 857 }
856 858
857 859 ASSERT(devvp != 0);
858 860
859 861 /*
860 862 * Flush back any dirty pages on the block device to
861 863 * try and keep the buffer cache in sync with the page
862 864 * cache if someone is trying to use block devices when
863 865 * they really should be using the raw device.
864 866 */
865 867 (void) VOP_PUTPAGE(common_specvp(devvp), (offset_t)0,
866 868 (size_t)0, B_INVAL, cr, NULL);
867 869
868 870 /*
869 871 * read in superblock
870 872 */
871 873 ufsvfsp = kmem_zalloc(sizeof (struct ufsvfs), KM_SLEEP);
872 874 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
873 875 if (tp->b_flags & B_ERROR)
874 876 goto out;
875 877 fsp = (struct fs *)tp->b_un.b_addr;
876 878
877 879 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) {
878 880 cmn_err(CE_NOTE,
879 881 "mount: not a UFS magic number (0x%x)", fsp->fs_magic);
880 882 error = EINVAL;
881 883 goto out;
882 884 }
883 885
884 886 if ((fsp->fs_magic == FS_MAGIC) &&
885 887 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
886 888 fsp->fs_version != UFS_VERSION_MIN)) {
887 889 cmn_err(CE_NOTE,
888 890 "mount: unrecognized version of UFS on-disk format: %d",
889 891 fsp->fs_version);
890 892 error = EINVAL;
891 893 goto out;
892 894 }
893 895
894 896 if ((fsp->fs_magic == MTB_UFS_MAGIC) &&
895 897 (fsp->fs_version > MTB_UFS_VERSION_1 ||
896 898 fsp->fs_version < MTB_UFS_VERSION_MIN)) {
897 899 cmn_err(CE_NOTE,
898 900 "mount: unrecognized version of UFS on-disk format: %d",
899 901 fsp->fs_version);
900 902 error = EINVAL;
901 903 goto out;
902 904 }
903 905
904 906 #ifndef _LP64
905 907 if (fsp->fs_magic == MTB_UFS_MAGIC) {
906 908 /*
907 909 * Find the size of the device in sectors. If the
908 910 * the size in sectors is greater than INT_MAX, it's
909 911 * a multi-terabyte file system, which can't be
910 912 * mounted by a 32-bit kernel. We can't use the
911 913 * fsbtodb() macro in the next line because the macro
912 914 * casts the intermediate values to daddr_t, which is
913 915 * a 32-bit quantity in a 32-bit kernel. Here we
914 916 * really do need the intermediate values to be held
915 917 * in 64-bit quantities because we're checking for
916 918 * overflow of a 32-bit field.
917 919 */
918 920 if ((((diskaddr_t)(fsp->fs_size)) << fsp->fs_fsbtodb)
919 921 > INT_MAX) {
920 922 cmn_err(CE_NOTE,
921 923 "mount: multi-terabyte UFS cannot be"
922 924 " mounted by a 32-bit kernel");
923 925 error = EINVAL;
924 926 goto out;
925 927 }
926 928
927 929 }
928 930 #endif
929 931
930 932 if (fsp->fs_bsize > MAXBSIZE || fsp->fs_frag > MAXFRAG ||
931 933 fsp->fs_bsize < sizeof (struct fs) || fsp->fs_bsize < PAGESIZE) {
932 934 error = EINVAL; /* also needs translation */
933 935 goto out;
934 936 }
935 937
936 938 /*
937 939 * Allocate VFS private data.
938 940 */
939 941 vfsp->vfs_bcount = 0;
940 942 vfsp->vfs_data = (caddr_t)ufsvfsp;
941 943 vfsp->vfs_fstype = ufsfstype;
942 944 vfsp->vfs_dev = dev;
943 945 vfsp->vfs_flag |= VFS_NOTRUNC;
944 946 vfs_make_fsid(&vfsp->vfs_fsid, dev, ufsfstype);
945 947 ufsvfsp->vfs_devvp = devvp;
946 948
947 949 /*
948 950 * Cross-link with vfs and add to instance list.
949 951 */
950 952 ufsvfsp->vfs_vfs = vfsp;
951 953 ufs_vfs_add(ufsvfsp);
952 954
953 955 ufsvfsp->vfs_dev = dev;
954 956 ufsvfsp->vfs_bufp = tp;
955 957
956 958 ufsvfsp->vfs_dirsize = INODESIZE + (4 * ALLOCSIZE) + fsp->fs_fsize;
957 959 ufsvfsp->vfs_minfrags =
958 960 (int)((int64_t)fsp->fs_dsize * fsp->fs_minfree / 100);
959 961 /*
960 962 * if mount allows largefiles, indicate so in ufsvfs
961 963 */
962 964 if (flags & UFSMNT_LARGEFILES)
963 965 ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
964 966 /*
965 967 * Initialize threads
966 968 */
967 969 ufs_delete_init(ufsvfsp, 1);
968 970 ufs_thread_init(&ufsvfsp->vfs_reclaim, 0);
969 971
970 972 /*
971 973 * Chicken and egg problem. The superblock may have deltas
972 974 * in the log. So after the log is scanned we reread the
973 975 * superblock. We guarantee that the fields needed to
974 976 * scan the log will not be in the log.
975 977 */
976 978 if (fsp->fs_logbno && fsp->fs_clean == FSLOG &&
977 979 (fsp->fs_state + fsp->fs_time == FSOKAY)) {
978 980 error = lufs_snarf(ufsvfsp, fsp, (vfsp->vfs_flag & VFS_RDONLY));
979 981 if (error) {
980 982 /*
981 983 * Allow a ro mount to continue even if the
982 984 * log cannot be processed - yet.
983 985 */
984 986 if (!(vfsp->vfs_flag & VFS_RDONLY)) {
985 987 cmn_err(CE_WARN, "Error accessing ufs "
986 988 "log for %s; Please run fsck(1M)", path);
987 989 goto out;
988 990 }
989 991 }
990 992 tp->b_flags |= (B_AGE | B_STALE);
991 993 brelse(tp);
992 994 tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
993 995 fsp = (struct fs *)tp->b_un.b_addr;
994 996 ufsvfsp->vfs_bufp = tp;
995 997 if (tp->b_flags & B_ERROR)
996 998 goto out;
997 999 }
998 1000
999 1001 /*
1000 1002 * Set logging mounted flag used by lockfs
1001 1003 */
1002 1004 ufsvfsp->vfs_validfs = UT_MOUNTED;
1003 1005
1004 1006 /*
1005 1007 * Copy the super block into a buffer in its native size.
1006 1008 * Use ngeteblk to allocate the buffer
1007 1009 */
1008 1010 bp = ngeteblk(fsp->fs_bsize);
1009 1011 ufsvfsp->vfs_bufp = bp;
1010 1012 bp->b_edev = dev;
1011 1013 bp->b_dev = cmpdev(dev);
1012 1014 bp->b_blkno = SBLOCK;
1013 1015 bp->b_bcount = fsp->fs_sbsize;
1014 1016 bcopy(tp->b_un.b_addr, bp->b_un.b_addr, fsp->fs_sbsize);
1015 1017 tp->b_flags |= B_STALE | B_AGE;
1016 1018 brelse(tp);
1017 1019 tp = 0;
1018 1020
1019 1021 fsp = (struct fs *)bp->b_un.b_addr;
1020 1022 /*
1021 1023 * Mount fails if superblock flag indicates presence of large
1022 1024 * files and filesystem is attempted to be mounted 'nolargefiles'.
1023 1025 * The exception is for a read only mount of root, which we
1024 1026 * always want to succeed, so fsck can fix potential problems.
1025 1027 * The assumption is that we will remount root at some point,
1026 1028 * and the remount will enforce the mount option.
1027 1029 */
1028 1030 if (!(isroot & (vfsp->vfs_flag & VFS_RDONLY)) &&
1029 1031 (fsp->fs_flags & FSLARGEFILES) &&
1030 1032 !(flags & UFSMNT_LARGEFILES)) {
1031 1033 error = EFBIG;
1032 1034 goto out;
1033 1035 }
1034 1036
1035 1037 if (vfsp->vfs_flag & VFS_RDONLY) {
1036 1038 fsp->fs_ronly = 1;
1037 1039 fsp->fs_fmod = 0;
1038 1040 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1039 1041 ((fsp->fs_clean == FSCLEAN) ||
1040 1042 (fsp->fs_clean == FSSTABLE) ||
1041 1043 (fsp->fs_clean == FSLOG))) {
1042 1044 if (isroot) {
1043 1045 if (fsp->fs_clean == FSLOG) {
1044 1046 if (fsp->fs_rolled == FS_ALL_ROLLED) {
1045 1047 ufs_clean_root = 1;
1046 1048 }
1047 1049 } else {
1048 1050 ufs_clean_root = 1;
1049 1051 }
1050 1052 }
1051 1053 fsp->fs_clean = FSSTABLE;
1052 1054 } else {
1053 1055 fsp->fs_clean = FSBAD;
1054 1056 }
1055 1057 } else {
1056 1058
1057 1059 fsp->fs_fmod = 0;
1058 1060 fsp->fs_ronly = 0;
1059 1061
1060 1062 TRANS_DOMATAMAP(ufsvfsp);
1061 1063
1062 1064 if ((TRANS_ISERROR(ufsvfsp)) ||
1063 1065 (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1064 1066 fsp->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp))) {
1065 1067 ufsvfsp->vfs_log = NULL;
1066 1068 ufsvfsp->vfs_domatamap = 0;
1067 1069 error = ENOSPC;
1068 1070 goto out;
1069 1071 }
1070 1072
1071 1073 if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1072 1074 (fsp->fs_clean == FSCLEAN ||
1073 1075 fsp->fs_clean == FSSTABLE ||
1074 1076 fsp->fs_clean == FSLOG))
1075 1077 fsp->fs_clean = FSSTABLE;
1076 1078 else {
1077 1079 if (isroot) {
1078 1080 /*
1079 1081 * allow root partition to be mounted even
1080 1082 * when fs_state is not ok
1081 1083 * will be fixed later by a remount root
1082 1084 */
1083 1085 fsp->fs_clean = FSBAD;
1084 1086 ufsvfsp->vfs_log = NULL;
1085 1087 ufsvfsp->vfs_domatamap = 0;
1086 1088 } else {
1087 1089 error = ENOSPC;
1088 1090 goto out;
1089 1091 }
1090 1092 }
1091 1093
1092 1094 if (fsp->fs_clean == FSSTABLE && TRANS_ISTRANS(ufsvfsp))
1093 1095 fsp->fs_clean = FSLOG;
1094 1096 }
1095 1097 TRANS_MATA_MOUNT(ufsvfsp);
1096 1098 needtrans = 1;
1097 1099
1098 1100 vfsp->vfs_bsize = fsp->fs_bsize;
1099 1101
1100 1102 /*
1101 1103 * Read in summary info
1102 1104 */
1103 1105 if (error = ufs_getsummaryinfo(dev, ufsvfsp, fsp))
1104 1106 goto out;
1105 1107
1106 1108 /*
1107 1109 * lastwhinetime is set to zero rather than lbolt, so that after
1108 1110 * mounting if the filesystem is found to be full, then immediately the
1109 1111 * "file system message" will be logged.
1110 1112 */
1111 1113 ufsvfsp->vfs_lastwhinetime = 0L;
1112 1114
1113 1115
1114 1116 mutex_init(&ufsvfsp->vfs_lock, NULL, MUTEX_DEFAULT, NULL);
1115 1117 (void) copystr(path, fsp->fs_fsmnt, sizeof (fsp->fs_fsmnt) - 1, &len);
1116 1118 bzero(fsp->fs_fsmnt + len, sizeof (fsp->fs_fsmnt) - len);
1117 1119
1118 1120 /*
1119 1121 * Sanity checks for old file systems
1120 1122 */
1121 1123 if (fsp->fs_postblformat == FS_42POSTBLFMT)
1122 1124 ufsvfsp->vfs_nrpos = 8;
1123 1125 else
1124 1126 ufsvfsp->vfs_nrpos = fsp->fs_nrpos;
1125 1127
1126 1128 /*
1127 1129 * Initialize lockfs structure to support file system locking
1128 1130 */
1129 1131 bzero(&ufsvfsp->vfs_ulockfs.ul_lockfs,
1130 1132 sizeof (struct lockfs));
1131 1133 ufsvfsp->vfs_ulockfs.ul_fs_lock = ULOCKFS_ULOCK;
1132 1134 mutex_init(&ufsvfsp->vfs_ulockfs.ul_lock, NULL,
1133 1135 MUTEX_DEFAULT, NULL);
1134 1136 cv_init(&ufsvfsp->vfs_ulockfs.ul_cv, NULL, CV_DEFAULT, NULL);
1135 1137
1136 1138 /*
1137 1139 * We don't need to grab vfs_dqrwlock for this ufs_iget() call.
1138 1140 * We are in the process of mounting the file system so there
1139 1141 * is no need to grab the quota lock. If a quota applies to the
1140 1142 * root inode, then it will be updated when quotas are enabled.
1141 1143 *
1142 1144 * However, we have an ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock))
1143 1145 * in getinoquota() that we want to keep so grab it anyway.
1144 1146 */
1145 1147 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1146 1148
1147 1149 error = ufs_iget_alloced(vfsp, UFSROOTINO, &rip, cr);
1148 1150
1149 1151 rw_exit(&ufsvfsp->vfs_dqrwlock);
1150 1152
1151 1153 if (error)
1152 1154 goto out;
1153 1155
1154 1156 /*
1155 1157 * make sure root inode is a directory. Returning ENOTDIR might
1156 1158 * be confused with the mount point not being a directory, so
1157 1159 * we use EIO instead.
1158 1160 */
1159 1161 if ((rip->i_mode & IFMT) != IFDIR) {
1160 1162 /*
1161 1163 * Mark this inode as subject for cleanup
1162 1164 * to avoid stray inodes in the cache.
1163 1165 */
1164 1166 rvp = ITOV(rip);
1165 1167 error = EIO;
1166 1168 goto out;
1167 1169 }
1168 1170
1169 1171 rvp = ITOV(rip);
1170 1172 mutex_enter(&rvp->v_lock);
1171 1173 rvp->v_flag |= VROOT;
1172 1174 mutex_exit(&rvp->v_lock);
1173 1175 ufsvfsp->vfs_root = rvp;
1174 1176 /* The buffer for the root inode does not contain a valid b_vp */
1175 1177 (void) bfinval(dev, 0);
1176 1178
1177 1179 /* options */
1178 1180 ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
1179 1181 ufsvfsp->vfs_nointr = flags & UFSMNT_NOINTR;
1180 1182 ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
1181 1183 ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
1182 1184 if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
1183 1185 ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
1184 1186 else /* dfratime, default behavior */
1185 1187 ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
1186 1188 if (flags & UFSMNT_FORCEDIRECTIO)
1187 1189 ufsvfsp->vfs_forcedirectio = 1;
1188 1190 else if (flags & UFSMNT_NOFORCEDIRECTIO)
1189 1191 ufsvfsp->vfs_forcedirectio = 0;
1190 1192 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1191 1193
1192 1194 ufsvfsp->vfs_nindiroffset = fsp->fs_nindir - 1;
1193 1195 ufsvfsp->vfs_nindirshift = highbit(ufsvfsp->vfs_nindiroffset);
1194 1196 ufsvfsp->vfs_ioclustsz = fsp->fs_bsize * fsp->fs_maxcontig;
1195 1197
1196 1198 if (cdev_ioctl(dev, DKIOCINFO, (intptr_t)&ci,
1197 1199 FKIOCTL|FNATIVE|FREAD, CRED(), &status) == 0) {
1198 1200 ufsvfsp->vfs_iotransz = ci.dki_maxtransfer * DEV_BSIZE;
1199 1201 } else {
1200 1202 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1201 1203 }
1202 1204
1203 1205 if (ufsvfsp->vfs_iotransz <= 0) {
1204 1206 ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1205 1207 }
1206 1208
1207 1209 /*
1208 1210 * When logging, used to reserve log space for writes and truncs
1209 1211 */
1210 1212 ufsvfsp->vfs_avgbfree = fsp->fs_cstotal.cs_nbfree / fsp->fs_ncg;
1211 1213
1212 1214 /*
1213 1215 * Determine whether to log cylinder group summary info.
1214 1216 */
1215 1217 ufsvfsp->vfs_nolog_si = (fsp->fs_ncg < ufs_ncg_log);
1216 1218
1217 1219 if (TRANS_ISTRANS(ufsvfsp)) {
1218 1220 /*
1219 1221 * start the delete thread
1220 1222 */
1221 1223 ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
1222 1224
1223 1225 /*
1224 1226 * start reclaim thread if the filesystem was not mounted
1225 1227 * read only.
1226 1228 */
1227 1229 if (!fsp->fs_ronly && (fsp->fs_reclaim &
1228 1230 (FS_RECLAIM|FS_RECLAIMING))) {
1229 1231 fsp->fs_reclaim &= ~FS_RECLAIM;
1230 1232 fsp->fs_reclaim |= FS_RECLAIMING;
1231 1233 ufs_thread_start(&ufsvfsp->vfs_reclaim,
1232 1234 ufs_thread_reclaim, vfsp);
1233 1235 }
1234 1236
1235 1237 /* Mark the fs as unrolled */
1236 1238 fsp->fs_rolled = FS_NEED_ROLL;
1237 1239 } else if (!fsp->fs_ronly && (fsp->fs_reclaim &
1238 1240 (FS_RECLAIM|FS_RECLAIMING))) {
1239 1241 /*
1240 1242 * If a file system that is mounted nologging, after
1241 1243 * having previously been mounted logging, becomes
1242 1244 * unmounted whilst the reclaim thread is in the throes
1243 1245 * of reclaiming open/deleted inodes, a subsequent mount
1244 1246 * of such a file system with logging disabled could lead
1245 1247 * to inodes becoming lost. So, start reclaim now, even
1246 1248 * though logging was disabled for the previous mount, to
1247 1249 * tidy things up.
1248 1250 */
1249 1251 fsp->fs_reclaim &= ~FS_RECLAIM;
1250 1252 fsp->fs_reclaim |= FS_RECLAIMING;
1251 1253 ufs_thread_start(&ufsvfsp->vfs_reclaim,
1252 1254 ufs_thread_reclaim, vfsp);
1253 1255 }
1254 1256
1255 1257 if (!fsp->fs_ronly) {
1256 1258 TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
1257 1259 if (error = geterror(ufsvfsp->vfs_bufp))
1258 1260 goto out;
1259 1261 }
1260 1262
1261 1263 /* fix-on-panic initialization */
1262 1264 if (isroot && !(flags & UFSMNT_ONERROR_FLGMASK))
1263 1265 flags |= UFSMNT_ONERROR_PANIC; /* XXX ..._RDONLY */
1264 1266
1265 1267 if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
1266 1268 goto out;
1267 1269
1268 1270 if (why == ROOT_INIT && isroot)
1269 1271 rootvp = devvp;
1270 1272
1271 1273 return (0);
1272 1274 out:
1273 1275 if (error == 0)
1274 1276 error = EIO;
1275 1277 if (rvp) {
1276 1278 /* the following sequence is similar to ufs_unmount() */
1277 1279
1278 1280 /*
1279 1281 * There's a problem that ufs_iget() puts inodes into
1280 1282 * the inode cache before it returns them. If someone
1281 1283 * traverses that cache and gets a reference to our
1282 1284 * inode, there's a chance they'll still be using it
1283 1285 * after we've destroyed it. This is a hard race to
1284 1286 * hit, but it's happened (putting in a medium delay
1285 1287 * here, and a large delay in ufs_scan_inodes() for
1286 1288 * inodes on the device we're bailing out on, makes
1287 1289 * the race easy to demonstrate). The symptom is some
1288 1290 * other part of UFS faulting on bad inode contents,
1289 1291 * or when grabbing one of the locks inside the inode,
1290 1292 * etc. The usual victim is ufs_scan_inodes() or
1291 1293 * someone called by it.
1292 1294 */
1293 1295
1294 1296 /*
1295 1297 * First, isolate it so that no new references can be
1296 1298 * gotten via the inode cache.
1297 1299 */
1298 1300 ihm = &ih_lock[INOHASH(UFSROOTINO)];
1299 1301 mutex_enter(ihm);
1300 1302 remque(rip);
1301 1303 mutex_exit(ihm);
1302 1304
1303 1305 /*
1304 1306 * Now wait for all outstanding references except our
1305 1307 * own to drain. This could, in theory, take forever,
1306 1308 * so don't wait *too* long. If we time out, mark
1307 1309 * it stale and leak it, so we don't hit the problem
1308 1310 * described above.
1309 1311 *
1310 1312 * Note that v_count is an int, which means we can read
1311 1313 * it in one operation. Thus, there's no need to lock
1312 1314 * around our tests.
1313 1315 */
1314 1316 elapsed = 0;
1315 1317 while ((rvp->v_count > 1) && (elapsed < ufs_mount_timeout)) {
1316 1318 delay(ufs_mount_error_delay * drv_usectohz(1000));
1317 1319 elapsed += ufs_mount_error_delay;
1318 1320 }
1319 1321
1320 1322 if (rvp->v_count > 1) {
1321 1323 mutex_enter(&rip->i_tlock);
1322 1324 rip->i_flag |= ISTALE;
1323 1325 mutex_exit(&rip->i_tlock);
1324 1326 cmn_err(CE_WARN,
1325 1327 "Timed out while cleaning up after "
1326 1328 "failed mount of %s", path);
1327 1329 } else {
1328 1330
1329 1331 /*
1330 1332 * Now we're the only one with a handle left, so tear
1331 1333 * it down the rest of the way.
1332 1334 */
1333 1335 if (ufs_rmidle(rip))
1334 1336 VN_RELE(rvp);
1335 1337 ufs_si_del(rip);
1336 1338 rip->i_ufsvfs = NULL;
1337 1339 rvp->v_vfsp = NULL;
1338 1340 rvp->v_type = VBAD;
1339 1341 VN_RELE(rvp);
1340 1342 }
1341 1343 }
1342 1344 if (needtrans) {
1343 1345 TRANS_MATA_UMOUNT(ufsvfsp);
1344 1346 }
1345 1347 if (ufsvfsp) {
1346 1348 ufs_vfs_remove(ufsvfsp);
1347 1349 ufs_thread_exit(&ufsvfsp->vfs_delete);
1348 1350 ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1349 1351 mutex_destroy(&ufsvfsp->vfs_lock);
1350 1352 if (ufsvfsp->vfs_log) {
1351 1353 lufs_unsnarf(ufsvfsp);
1352 1354 }
1353 1355 kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1354 1356 }
1355 1357 if (bp) {
1356 1358 bp->b_flags |= (B_STALE|B_AGE);
1357 1359 brelse(bp);
1358 1360 }
1359 1361 if (tp) {
1360 1362 tp->b_flags |= (B_STALE|B_AGE);
1361 1363 brelse(tp);
1362 1364 }
1363 1365 if (needclose) {
1364 1366 (void) VOP_CLOSE(devvp, (vfsp->vfs_flag & VFS_RDONLY) ?
1365 1367 FREAD : FREAD|FWRITE, 1, (offset_t)0, cr, NULL);
1366 1368 bflush(dev);
1367 1369 (void) bfinval(dev, 1);
1368 1370 }
1369 1371 return (error);
1370 1372 }
1371 1373
1372 1374 /*
1373 1375 * vfs operations
1374 1376 */
1375 1377 static int
1376 1378 ufs_unmount(struct vfs *vfsp, int fflag, struct cred *cr)
1377 1379 {
1378 1380 dev_t dev = vfsp->vfs_dev;
1379 1381 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1380 1382 struct fs *fs = ufsvfsp->vfs_fs;
1381 1383 struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs;
1382 1384 struct vnode *bvp, *vp;
1383 1385 struct buf *bp;
1384 1386 struct inode *ip, *inext, *rip;
1385 1387 union ihead *ih;
1386 1388 int error, flag, i;
1387 1389 struct lockfs lockfs;
1388 1390 int poll_events = POLLPRI;
1389 1391 extern struct pollhead ufs_pollhd;
1390 1392 refstr_t *mountpoint;
1391 1393
1392 1394 ASSERT(vfs_lock_held(vfsp));
1393 1395
1394 1396 if (secpolicy_fs_unmount(cr, vfsp) != 0)
1395 1397 return (EPERM);
1396 1398 /*
1397 1399 * Forced unmount is now supported through the
1398 1400 * lockfs protocol.
1399 1401 */
1400 1402 if (fflag & MS_FORCE) {
1401 1403 /*
1402 1404 * Mark the filesystem as being unmounted now in
1403 1405 * case of a forcible umount before we take any
1404 1406 * locks inside UFS to prevent racing with a VFS_VGET()
1405 1407 * request. Throw these VFS_VGET() requests away for
1406 1408 * the duration of the forcible umount so they won't
1407 1409 * use stale or even freed data later on when we're done.
1408 1410 * It may happen that the VFS has had a additional hold
1409 1411 * placed on it by someone other than UFS and thus will
1410 1412 * not get freed immediately once we're done with the
1411 1413 * umount by dounmount() - use VFS_UNMOUNTED to inform
1412 1414 * users of this still-alive VFS that its corresponding
1413 1415 * filesystem being gone so they can detect that and error
1414 1416 * out.
1415 1417 */
1416 1418 vfsp->vfs_flag |= VFS_UNMOUNTED;
1417 1419
1418 1420 ufs_thread_suspend(&ufsvfsp->vfs_delete);
1419 1421 mutex_enter(&ulp->ul_lock);
1420 1422 /*
1421 1423 * If file system is already hard locked,
1422 1424 * unmount the file system, otherwise
1423 1425 * hard lock it before unmounting.
1424 1426 */
1425 1427 if (!ULOCKFS_IS_HLOCK(ulp)) {
1426 1428 atomic_inc_ulong(&ufs_quiesce_pend);
1427 1429 lockfs.lf_lock = LOCKFS_HLOCK;
1428 1430 lockfs.lf_flags = 0;
1429 1431 lockfs.lf_key = ulp->ul_lockfs.lf_key + 1;
1430 1432 lockfs.lf_comlen = 0;
1431 1433 lockfs.lf_comment = NULL;
1432 1434 ufs_freeze(ulp, &lockfs);
1433 1435 ULOCKFS_SET_BUSY(ulp);
1434 1436 LOCKFS_SET_BUSY(&ulp->ul_lockfs);
1435 1437 (void) ufs_quiesce(ulp);
1436 1438 (void) ufs_flush(vfsp);
1437 1439 (void) ufs_thaw(vfsp, ufsvfsp, ulp);
1438 1440 atomic_dec_ulong(&ufs_quiesce_pend);
1439 1441 ULOCKFS_CLR_BUSY(ulp);
1440 1442 LOCKFS_CLR_BUSY(&ulp->ul_lockfs);
1441 1443 poll_events |= POLLERR;
1442 1444 pollwakeup(&ufs_pollhd, poll_events);
1443 1445 }
1444 1446 ufs_thread_continue(&ufsvfsp->vfs_delete);
1445 1447 mutex_exit(&ulp->ul_lock);
1446 1448 }
1447 1449
1448 1450 /* let all types of writes go through */
1449 1451 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1450 1452
1451 1453 /* coordinate with global hlock thread */
1452 1454 if (TRANS_ISTRANS(ufsvfsp) && (ufsvfsp->vfs_validfs == UT_HLOCKING)) {
1453 1455 /*
1454 1456 * last possibility for a forced umount to fail hence clear
1455 1457 * VFS_UNMOUNTED if appropriate.
1456 1458 */
1457 1459 if (fflag & MS_FORCE)
1458 1460 vfsp->vfs_flag &= ~VFS_UNMOUNTED;
1459 1461 return (EAGAIN);
1460 1462 }
1461 1463
1462 1464 ufsvfsp->vfs_validfs = UT_UNMOUNTED;
1463 1465
1464 1466 /* kill the reclaim thread */
1465 1467 ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1466 1468
1467 1469 /* suspend the delete thread */
1468 1470 ufs_thread_suspend(&ufsvfsp->vfs_delete);
1469 1471
1470 1472 /*
1471 1473 * drain the delete and idle queues
1472 1474 */
1473 1475 ufs_delete_drain(vfsp, -1, 1);
1474 1476 ufs_idle_drain(vfsp);
1475 1477
1476 1478 /*
1477 1479 * use the lockfs protocol to prevent new ops from starting
1478 1480 * a forcible umount can not fail beyond this point as
1479 1481 * we hard-locked the filesystem and drained all current consumers
1480 1482 * before.
1481 1483 */
1482 1484 mutex_enter(&ulp->ul_lock);
1483 1485
1484 1486 /*
1485 1487 * if the file system is busy; return EBUSY
1486 1488 */
1487 1489 if (ulp->ul_vnops_cnt || ulp->ul_falloc_cnt || ULOCKFS_IS_SLOCK(ulp)) {
1488 1490 error = EBUSY;
1489 1491 goto out;
1490 1492 }
1491 1493
1492 1494 /*
1493 1495 * if this is not a forced unmount (!hard/error locked), then
1494 1496 * get rid of every inode except the root and quota inodes
1495 1497 * also, commit any outstanding transactions
1496 1498 */
1497 1499 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp))
1498 1500 if (error = ufs_flush(vfsp))
1499 1501 goto out;
1500 1502
1501 1503 /*
1502 1504 * ignore inodes in the cache if fs is hard locked or error locked
1503 1505 */
1504 1506 rip = VTOI(ufsvfsp->vfs_root);
1505 1507 if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) {
1506 1508 /*
1507 1509 * Otherwise, only the quota and root inodes are in the cache.
1508 1510 *
1509 1511 * Avoid racing with ufs_update() and ufs_sync().
1510 1512 */
1511 1513 mutex_enter(&ufs_scan_lock);
1512 1514
1513 1515 for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1514 1516 mutex_enter(&ih_lock[i]);
1515 1517 for (ip = ih->ih_chain[0];
1516 1518 ip != (struct inode *)ih;
1517 1519 ip = ip->i_forw) {
1518 1520 if (ip->i_ufsvfs != ufsvfsp)
1519 1521 continue;
1520 1522 if (ip == ufsvfsp->vfs_qinod)
1521 1523 continue;
1522 1524 if (ip == rip && ITOV(ip)->v_count == 1)
1523 1525 continue;
1524 1526 mutex_exit(&ih_lock[i]);
1525 1527 mutex_exit(&ufs_scan_lock);
1526 1528 error = EBUSY;
1527 1529 goto out;
1528 1530 }
1529 1531 mutex_exit(&ih_lock[i]);
1530 1532 }
1531 1533 mutex_exit(&ufs_scan_lock);
1532 1534 }
1533 1535
1534 1536 /*
1535 1537 * if a snapshot exists and this is a forced unmount, then delete
1536 1538 * the snapshot. Otherwise return EBUSY. This will insure the
1537 1539 * snapshot always belongs to a valid file system.
1538 1540 */
1539 1541 if (ufsvfsp->vfs_snapshot) {
1540 1542 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1541 1543 (void) fssnap_delete(&ufsvfsp->vfs_snapshot);
1542 1544 } else {
1543 1545 error = EBUSY;
1544 1546 goto out;
1545 1547 }
1546 1548 }
1547 1549
1548 1550 /*
1549 1551 * Close the quota file and invalidate anything left in the quota
1550 1552 * cache for this file system. Pass kcred to allow all quota
1551 1553 * manipulations.
1552 1554 */
1553 1555 (void) closedq(ufsvfsp, kcred);
1554 1556 invalidatedq(ufsvfsp);
1555 1557 /*
1556 1558 * drain the delete and idle queues
1557 1559 */
1558 1560 ufs_delete_drain(vfsp, -1, 0);
1559 1561 ufs_idle_drain(vfsp);
1560 1562
1561 1563 /*
1562 1564 * discard the inodes for this fs (including root, shadow, and quota)
1563 1565 */
1564 1566 for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1565 1567 mutex_enter(&ih_lock[i]);
1566 1568 for (inext = 0, ip = ih->ih_chain[0];
1567 1569 ip != (struct inode *)ih;
1568 1570 ip = inext) {
1569 1571 inext = ip->i_forw;
1570 1572 if (ip->i_ufsvfs != ufsvfsp)
1571 1573 continue;
1572 1574
1573 1575 /*
1574 1576 * We've found the inode in the cache and as we
1575 1577 * hold the hash mutex the inode can not
1576 1578 * disappear from underneath us.
1577 1579 * We also know it must have at least a vnode
1578 1580 * reference count of 1.
1579 1581 * We perform an additional VN_HOLD so the VN_RELE
1580 1582 * in case we take the inode off the idle queue
1581 1583 * can not be the last one.
1582 1584 * It is safe to grab the writer contents lock here
1583 1585 * to prevent a race with ufs_iinactive() putting
1584 1586 * inodes into the idle queue while we operate on
1585 1587 * this inode.
1586 1588 */
1587 1589 rw_enter(&ip->i_contents, RW_WRITER);
1588 1590
1589 1591 vp = ITOV(ip);
1590 1592 VN_HOLD(vp)
1591 1593 remque(ip);
1592 1594 if (ufs_rmidle(ip))
1593 1595 VN_RELE(vp);
1594 1596 ufs_si_del(ip);
1595 1597 /*
1596 1598 * rip->i_ufsvfsp is needed by bflush()
1597 1599 */
1598 1600 if (ip != rip)
1599 1601 ip->i_ufsvfs = NULL;
1600 1602 /*
1601 1603 * Set vnode's vfsops to dummy ops, which return
1602 1604 * EIO. This is needed to forced unmounts to work
1603 1605 * with lofs/nfs properly.
1604 1606 */
1605 1607 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp))
1606 1608 vp->v_vfsp = &EIO_vfs;
1607 1609 else
1608 1610 vp->v_vfsp = NULL;
1609 1611 vp->v_type = VBAD;
1610 1612
1611 1613 rw_exit(&ip->i_contents);
1612 1614
1613 1615 VN_RELE(vp);
1614 1616 }
1615 1617 mutex_exit(&ih_lock[i]);
1616 1618 }
1617 1619 ufs_si_cache_flush(dev);
1618 1620
1619 1621 /*
1620 1622 * kill the delete thread and drain the idle queue
1621 1623 */
1622 1624 ufs_thread_exit(&ufsvfsp->vfs_delete);
1623 1625 ufs_idle_drain(vfsp);
1624 1626
1625 1627 bp = ufsvfsp->vfs_bufp;
1626 1628 bvp = ufsvfsp->vfs_devvp;
1627 1629 flag = !fs->fs_ronly;
1628 1630 if (flag) {
1629 1631 bflush(dev);
1630 1632 if (fs->fs_clean != FSBAD) {
1631 1633 if (fs->fs_clean == FSSTABLE)
1632 1634 fs->fs_clean = FSCLEAN;
1633 1635 fs->fs_reclaim &= ~FS_RECLAIM;
1634 1636 }
1635 1637 if (TRANS_ISTRANS(ufsvfsp) &&
1636 1638 !TRANS_ISERROR(ufsvfsp) &&
1637 1639 !ULOCKFS_IS_HLOCK(ulp) &&
1638 1640 (fs->fs_rolled == FS_NEED_ROLL)) {
1639 1641 /*
1640 1642 * ufs_flush() above has flushed the last Moby.
1641 1643 * This is needed to ensure the following superblock
1642 1644 * update really is the last metadata update
1643 1645 */
1644 1646 error = ufs_putsummaryinfo(dev, ufsvfsp, fs);
1645 1647 if (error == 0) {
1646 1648 fs->fs_rolled = FS_ALL_ROLLED;
1647 1649 }
1648 1650 }
1649 1651 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UNMOUNT);
1650 1652 /*
1651 1653 * push this last transaction
1652 1654 */
1653 1655 curthread->t_flag |= T_DONTBLOCK;
1654 1656 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE,
1655 1657 error);
1656 1658 if (!error)
1657 1659 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UNMOUNT,
1658 1660 TOP_COMMIT_SIZE);
1659 1661 curthread->t_flag &= ~T_DONTBLOCK;
1660 1662 }
1661 1663
1662 1664 TRANS_MATA_UMOUNT(ufsvfsp);
1663 1665 lufs_unsnarf(ufsvfsp); /* Release the in-memory structs */
1664 1666 ufsfx_unmount(ufsvfsp); /* fix-on-panic bookkeeping */
1665 1667 kmem_free(fs->fs_u.fs_csp, fs->fs_cssize);
1666 1668
1667 1669 bp->b_flags |= B_STALE|B_AGE;
1668 1670 ufsvfsp->vfs_bufp = NULL; /* don't point at freed buf */
1669 1671 brelse(bp); /* free the superblock buf */
1670 1672
1671 1673 (void) VOP_PUTPAGE(common_specvp(bvp), (offset_t)0, (size_t)0,
1672 1674 B_INVAL, cr, NULL);
1673 1675 (void) VOP_CLOSE(bvp, flag, 1, (offset_t)0, cr, NULL);
1674 1676 bflush(dev);
1675 1677 (void) bfinval(dev, 1);
1676 1678 VN_RELE(bvp);
1677 1679
1678 1680 /*
1679 1681 * It is now safe to NULL out the ufsvfs pointer and discard
1680 1682 * the root inode.
1681 1683 */
1682 1684 rip->i_ufsvfs = NULL;
1683 1685 VN_RELE(ITOV(rip));
1684 1686
1685 1687 /* free up lockfs comment structure, if any */
1686 1688 if (ulp->ul_lockfs.lf_comlen && ulp->ul_lockfs.lf_comment)
1687 1689 kmem_free(ulp->ul_lockfs.lf_comment, ulp->ul_lockfs.lf_comlen);
1688 1690
1689 1691 /*
1690 1692 * Remove from instance list.
1691 1693 */
1692 1694 ufs_vfs_remove(ufsvfsp);
1693 1695
1694 1696 /*
1695 1697 * For a forcible unmount, threads may be asleep in
1696 1698 * ufs_lockfs_begin/ufs_check_lockfs. These threads will need
1697 1699 * the ufsvfs structure so we don't free it, yet. ufs_update
1698 1700 * will free it up after awhile.
1699 1701 */
1700 1702 if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1701 1703 extern kmutex_t ufsvfs_mutex;
1702 1704 extern struct ufsvfs *ufsvfslist;
1703 1705
1704 1706 mutex_enter(&ufsvfs_mutex);
1705 1707 ufsvfsp->vfs_dontblock = 1;
1706 1708 ufsvfsp->vfs_next = ufsvfslist;
1707 1709 ufsvfslist = ufsvfsp;
1708 1710 mutex_exit(&ufsvfs_mutex);
1709 1711 /* wakeup any suspended threads */
1710 1712 cv_broadcast(&ulp->ul_cv);
1711 1713 mutex_exit(&ulp->ul_lock);
1712 1714 } else {
1713 1715 mutex_destroy(&ufsvfsp->vfs_lock);
1714 1716 kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1715 1717 }
1716 1718
1717 1719 /*
1718 1720 * Now mark the filesystem as unmounted since we're done with it.
1719 1721 */
1720 1722 vfsp->vfs_flag |= VFS_UNMOUNTED;
1721 1723
1722 1724 return (0);
1723 1725 out:
1724 1726 /* open the fs to new ops */
1725 1727 cv_broadcast(&ulp->ul_cv);
1726 1728 mutex_exit(&ulp->ul_lock);
1727 1729
1728 1730 if (TRANS_ISTRANS(ufsvfsp)) {
1729 1731 /* allow the delete thread to continue */
1730 1732 ufs_thread_continue(&ufsvfsp->vfs_delete);
1731 1733 /* restart the reclaim thread */
1732 1734 ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim,
1733 1735 vfsp);
1734 1736 /* coordinate with global hlock thread */
1735 1737 ufsvfsp->vfs_validfs = UT_MOUNTED;
1736 1738 /* check for trans errors during umount */
1737 1739 ufs_trans_onerror();
1738 1740
1739 1741 /*
1740 1742 * if we have a separate /usr it will never unmount
1741 1743 * when halting. In order to not re-read all the
1742 1744 * cylinder group summary info on mounting after
1743 1745 * reboot the logging of summary info is re-enabled
1744 1746 * and the super block written out.
1745 1747 */
1746 1748 mountpoint = vfs_getmntpoint(vfsp);
1747 1749 if ((fs->fs_si == FS_SI_OK) &&
1748 1750 (strcmp("/usr", refstr_value(mountpoint)) == 0)) {
1749 1751 ufsvfsp->vfs_nolog_si = 0;
1750 1752 UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
1751 1753 }
1752 1754 refstr_rele(mountpoint);
1753 1755 }
1754 1756
1755 1757 return (error);
1756 1758 }
1757 1759
1758 1760 static int
1759 1761 ufs_root(struct vfs *vfsp, struct vnode **vpp)
1760 1762 {
1761 1763 struct ufsvfs *ufsvfsp;
1762 1764 struct vnode *vp;
1763 1765
1764 1766 if (!vfsp)
1765 1767 return (EIO);
1766 1768
1767 1769 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1768 1770 if (!ufsvfsp || !ufsvfsp->vfs_root)
1769 1771 return (EIO); /* forced unmount */
1770 1772
1771 1773 vp = ufsvfsp->vfs_root;
1772 1774 VN_HOLD(vp);
1773 1775 *vpp = vp;
1774 1776 return (0);
1775 1777 }
1776 1778
1777 1779 /*
1778 1780 * Get file system statistics.
1779 1781 */
1780 1782 static int
1781 1783 ufs_statvfs(struct vfs *vfsp, struct statvfs64 *sp)
1782 1784 {
1783 1785 struct fs *fsp;
1784 1786 struct ufsvfs *ufsvfsp;
1785 1787 int blk, i;
1786 1788 long max_avail, used;
1787 1789 dev32_t d32;
1788 1790
1789 1791 if (vfsp->vfs_flag & VFS_UNMOUNTED)
1790 1792 return (EIO);
1791 1793
1792 1794 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1793 1795 fsp = ufsvfsp->vfs_fs;
1794 1796 if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC))
1795 1797 return (EINVAL);
1796 1798 if (fsp->fs_magic == FS_MAGIC &&
1797 1799 (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
1798 1800 fsp->fs_version != UFS_VERSION_MIN))
1799 1801 return (EINVAL);
1800 1802 if (fsp->fs_magic == MTB_UFS_MAGIC &&
1801 1803 (fsp->fs_version > MTB_UFS_VERSION_1 ||
1802 1804 fsp->fs_version < MTB_UFS_VERSION_MIN))
1803 1805 return (EINVAL);
1804 1806
1805 1807 /*
1806 1808 * get the basic numbers
1807 1809 */
1808 1810 (void) bzero(sp, sizeof (*sp));
1809 1811
1810 1812 sp->f_bsize = fsp->fs_bsize;
1811 1813 sp->f_frsize = fsp->fs_fsize;
1812 1814 sp->f_blocks = (fsblkcnt64_t)fsp->fs_dsize;
1813 1815 sp->f_bfree = (fsblkcnt64_t)fsp->fs_cstotal.cs_nbfree * fsp->fs_frag +
1814 1816 fsp->fs_cstotal.cs_nffree;
1815 1817
1816 1818 sp->f_files = (fsfilcnt64_t)fsp->fs_ncg * fsp->fs_ipg;
1817 1819 sp->f_ffree = (fsfilcnt64_t)fsp->fs_cstotal.cs_nifree;
1818 1820
1819 1821 /*
1820 1822 * Adjust the numbers based on things waiting to be deleted.
1821 1823 * modifies f_bfree and f_ffree. Afterwards, everything we
1822 1824 * come up with will be self-consistent. By definition, this
1823 1825 * is a point-in-time snapshot, so the fact that the delete
1824 1826 * thread's probably already invalidated the results is not a
1825 1827 * problem. Note that if the delete thread is ever extended to
1826 1828 * non-logging ufs, this adjustment must always be made.
1827 1829 */
1828 1830 if (TRANS_ISTRANS(ufsvfsp))
1829 1831 ufs_delete_adjust_stats(ufsvfsp, sp);
1830 1832
1831 1833 /*
1832 1834 * avail = MAX(max_avail - used, 0)
1833 1835 */
1834 1836 max_avail = fsp->fs_dsize - ufsvfsp->vfs_minfrags;
1835 1837
1836 1838 used = (fsp->fs_dsize - sp->f_bfree);
1837 1839
1838 1840 if (max_avail > used)
1839 1841 sp->f_bavail = (fsblkcnt64_t)max_avail - used;
1840 1842 else
1841 1843 sp->f_bavail = (fsblkcnt64_t)0;
1842 1844
1843 1845 sp->f_favail = sp->f_ffree;
1844 1846 (void) cmpldev(&d32, vfsp->vfs_dev);
1845 1847 sp->f_fsid = d32;
1846 1848 (void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1847 1849 sp->f_flag = vf_to_stf(vfsp->vfs_flag);
1848 1850
1849 1851 /* keep coordinated with ufs_l_pathconf() */
1850 1852 sp->f_namemax = MAXNAMLEN;
1851 1853
1852 1854 if (fsp->fs_cpc == 0) {
1853 1855 bzero(sp->f_fstr, 14);
1854 1856 return (0);
1855 1857 }
1856 1858 blk = fsp->fs_spc * fsp->fs_cpc / NSPF(fsp);
1857 1859 for (i = 0; i < blk; i += fsp->fs_frag) /* CSTYLED */
1858 1860 /* void */;
1859 1861 i -= fsp->fs_frag;
1860 1862 blk = i / fsp->fs_frag;
1861 1863 bcopy(&(fs_rotbl(fsp)[blk]), sp->f_fstr, 14);
1862 1864 return (0);
1863 1865 }
1864 1866
1865 1867 /*
1866 1868 * Flush any pending I/O to file system vfsp.
1867 1869 * The ufs_update() routine will only flush *all* ufs files.
1868 1870 * If vfsp is non-NULL, only sync this ufs (in preparation
1869 1871 * for a umount).
1870 1872 */
1871 1873 /*ARGSUSED*/
1872 1874 static int
1873 1875 ufs_sync(struct vfs *vfsp, short flag, struct cred *cr)
1874 1876 {
1875 1877 struct ufsvfs *ufsvfsp;
1876 1878 struct fs *fs;
1877 1879 int cheap = flag & SYNC_ATTR;
1878 1880 int error;
1879 1881
1880 1882 /*
1881 1883 * SYNC_CLOSE means we're rebooting. Toss everything
1882 1884 * on the idle queue so we don't have to slog through
1883 1885 * a bunch of uninteresting inodes over and over again.
1884 1886 */
1885 1887 if (flag & SYNC_CLOSE)
1886 1888 ufs_idle_drain(NULL);
1887 1889
1888 1890 if (vfsp == NULL) {
1889 1891 ufs_update(flag);
1890 1892 return (0);
1891 1893 }
1892 1894
1893 1895 /* Flush a single ufs */
1894 1896 if (!vfs_matchops(vfsp, ufs_vfsops) || vfs_lock(vfsp) != 0)
1895 1897 return (0);
1896 1898
1897 1899 ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1898 1900 if (!ufsvfsp)
1899 1901 return (EIO);
1900 1902 fs = ufsvfsp->vfs_fs;
1901 1903 mutex_enter(&ufsvfsp->vfs_lock);
1902 1904
1903 1905 if (ufsvfsp->vfs_dio &&
1904 1906 fs->fs_ronly == 0 &&
1905 1907 fs->fs_clean != FSBAD &&
1906 1908 fs->fs_clean != FSLOG) {
1907 1909 /* turn off fast-io on unmount, so no fsck needed (4029401) */
1908 1910 ufsvfsp->vfs_dio = 0;
1909 1911 fs->fs_clean = FSACTIVE;
1910 1912 fs->fs_fmod = 1;
1911 1913 }
1912 1914
1913 1915 /* Write back modified superblock */
1914 1916 if (fs->fs_fmod == 0) {
1915 1917 mutex_exit(&ufsvfsp->vfs_lock);
1916 1918 } else {
1917 1919 if (fs->fs_ronly != 0) {
1918 1920 mutex_exit(&ufsvfsp->vfs_lock);
1919 1921 vfs_unlock(vfsp);
1920 1922 return (ufs_fault(ufsvfsp->vfs_root,
1921 1923 "fs = %s update: ro fs mod\n", fs->fs_fsmnt));
1922 1924 }
1923 1925 fs->fs_fmod = 0;
1924 1926 mutex_exit(&ufsvfsp->vfs_lock);
1925 1927
1926 1928 TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UPDATE);
1927 1929 }
1928 1930 vfs_unlock(vfsp);
1929 1931
1930 1932 /*
1931 1933 * Avoid racing with ufs_update() and ufs_unmount().
1932 1934 *
1933 1935 */
1934 1936 mutex_enter(&ufs_scan_lock);
1935 1937
1936 1938 (void) ufs_scan_inodes(1, ufs_sync_inode,
1937 1939 (void *)(uintptr_t)cheap, ufsvfsp);
1938 1940
1939 1941 mutex_exit(&ufs_scan_lock);
1940 1942
1941 1943 bflush((dev_t)vfsp->vfs_dev);
1942 1944
1943 1945 /*
1944 1946 * commit any outstanding async transactions
1945 1947 */
1946 1948 curthread->t_flag |= T_DONTBLOCK;
1947 1949 TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE, error);
1948 1950 if (!error) {
1949 1951 TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UPDATE,
1950 1952 TOP_COMMIT_SIZE);
1951 1953 }
1952 1954 curthread->t_flag &= ~T_DONTBLOCK;
1953 1955
1954 1956 return (0);
1955 1957 }
1956 1958
1957 1959
1958 1960 void
1959 1961 sbupdate(struct vfs *vfsp)
1960 1962 {
1961 1963 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1962 1964 struct fs *fs = ufsvfsp->vfs_fs;
1963 1965 struct buf *bp;
1964 1966 int blks;
1965 1967 caddr_t space;
1966 1968 int i;
1967 1969 size_t size;
1968 1970
1969 1971 /*
1970 1972 * for ulockfs processing, limit the superblock writes
1971 1973 */
1972 1974 if ((ufsvfsp->vfs_ulockfs.ul_sbowner) &&
1973 1975 (curthread != ufsvfsp->vfs_ulockfs.ul_sbowner)) {
1974 1976 /* process later */
1975 1977 fs->fs_fmod = 1;
1976 1978 return;
1977 1979 }
1978 1980 ULOCKFS_SET_MOD((&ufsvfsp->vfs_ulockfs));
1979 1981
1980 1982 if (TRANS_ISTRANS(ufsvfsp)) {
1981 1983 mutex_enter(&ufsvfsp->vfs_lock);
1982 1984 ufs_sbwrite(ufsvfsp);
1983 1985 mutex_exit(&ufsvfsp->vfs_lock);
1984 1986 return;
1985 1987 }
1986 1988
1987 1989 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1988 1990 space = (caddr_t)fs->fs_u.fs_csp;
1989 1991 for (i = 0; i < blks; i += fs->fs_frag) {
1990 1992 size = fs->fs_bsize;
1991 1993 if (i + fs->fs_frag > blks)
1992 1994 size = (blks - i) * fs->fs_fsize;
1993 1995 bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev,
1994 1996 (daddr_t)(fsbtodb(fs, fs->fs_csaddr + i)),
1995 1997 fs->fs_bsize);
1996 1998 bcopy(space, bp->b_un.b_addr, size);
1997 1999 space += size;
1998 2000 bp->b_bcount = size;
1999 2001 UFS_BRWRITE(ufsvfsp, bp);
2000 2002 }
2001 2003 mutex_enter(&ufsvfsp->vfs_lock);
2002 2004 ufs_sbwrite(ufsvfsp);
2003 2005 mutex_exit(&ufsvfsp->vfs_lock);
2004 2006 }
2005 2007
2006 2008 int ufs_vget_idle_count = 2; /* Number of inodes to idle each time */
2007 2009 static int
2008 2010 ufs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
2009 2011 {
2010 2012 int error = 0;
2011 2013 struct ufid *ufid;
2012 2014 struct inode *ip;
2013 2015 struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
2014 2016 struct ulockfs *ulp;
2015 2017
2016 2018 /*
2017 2019 * Check for unmounted filesystem.
2018 2020 */
2019 2021 if (vfsp->vfs_flag & VFS_UNMOUNTED) {
2020 2022 error = EIO;
2021 2023 goto errout;
2022 2024 }
2023 2025
2024 2026 /*
2025 2027 * Keep the idle queue from getting too long by
2026 2028 * idling an inode before attempting to allocate another.
2027 2029 * This operation must be performed before entering
2028 2030 * lockfs or a transaction.
2029 2031 */
2030 2032 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2031 2033 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2032 2034 ins.in_vidles.value.ul += ufs_vget_idle_count;
2033 2035 ufs_idle_some(ufs_vget_idle_count);
2034 2036 }
2035 2037
2036 2038 ufid = (struct ufid *)fidp;
2037 2039
2038 2040 if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_VGET_MASK))
2039 2041 goto errout;
2040 2042
2041 2043 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2042 2044
2043 2045 error = ufs_iget(vfsp, ufid->ufid_ino, &ip, CRED());
2044 2046
2045 2047 rw_exit(&ufsvfsp->vfs_dqrwlock);
2046 2048
2047 2049 ufs_lockfs_end(ulp);
2048 2050
2049 2051 if (error)
2050 2052 goto errout;
2051 2053
2052 2054 /*
2053 2055 * Check if the inode has been deleted or freed or is in transient state
2054 2056 * since the last VFS_VGET() request for it, release it and don't return
2055 2057 * it to the caller, presumably NFS, as it's no longer valid.
2056 2058 */
2057 2059 if (ip->i_gen != ufid->ufid_gen || ip->i_mode == 0 ||
2058 2060 (ip->i_nlink <= 0)) {
2059 2061 VN_RELE(ITOV(ip));
2060 2062 error = EINVAL;
2061 2063 goto errout;
2062 2064 }
2063 2065
2064 2066 *vpp = ITOV(ip);
2065 2067 return (0);
2066 2068
2067 2069 errout:
2068 2070 *vpp = NULL;
2069 2071 return (error);
2070 2072 }
2071 2073
2072 2074 static int
2073 2075 ufsinit(int fstype, char *name)
2074 2076 {
2075 2077 static const fs_operation_def_t ufs_vfsops_template[] = {
2076 2078 VFSNAME_MOUNT, { .vfs_mount = ufs_mount },
2077 2079 VFSNAME_UNMOUNT, { .vfs_unmount = ufs_unmount },
2078 2080 VFSNAME_ROOT, { .vfs_root = ufs_root },
2079 2081 VFSNAME_STATVFS, { .vfs_statvfs = ufs_statvfs },
2080 2082 VFSNAME_SYNC, { .vfs_sync = ufs_sync },
2081 2083 VFSNAME_VGET, { .vfs_vget = ufs_vget },
2082 2084 VFSNAME_MOUNTROOT, { .vfs_mountroot = ufs_mountroot },
2083 2085 NULL, NULL
2084 2086 };
2085 2087 int error;
2086 2088
2087 2089 ufsfstype = fstype;
2088 2090
2089 2091 error = vfs_setfsops(fstype, ufs_vfsops_template, &ufs_vfsops);
2090 2092 if (error != 0) {
2091 2093 cmn_err(CE_WARN, "ufsinit: bad vfs ops template");
2092 2094 return (error);
2093 2095 }
2094 2096
2095 2097 error = vn_make_ops(name, ufs_vnodeops_template, &ufs_vnodeops);
2096 2098 if (error != 0) {
2097 2099 (void) vfs_freevfsops_by_type(fstype);
2098 2100 cmn_err(CE_WARN, "ufsinit: bad vnode ops template");
2099 2101 return (error);
2100 2102 }
2101 2103
2102 2104 ufs_iinit();
2103 2105 return (0);
2104 2106 }
|
↓ open down ↓ |
1277 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX