Print this page
5045 use atomic_{inc,dec}_* instead of atomic_add_*
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zfs_vfsops.c
+++ new/usr/src/uts/common/fs/zfs/zfs_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]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 24 */
25 25
26 26 /* Portions Copyright 2010 Robert Milkowski */
27 27
28 28 #include <sys/types.h>
29 29 #include <sys/param.h>
30 30 #include <sys/systm.h>
31 31 #include <sys/sysmacros.h>
32 32 #include <sys/kmem.h>
33 33 #include <sys/pathname.h>
34 34 #include <sys/vnode.h>
35 35 #include <sys/vfs.h>
36 36 #include <sys/vfs_opreg.h>
37 37 #include <sys/mntent.h>
38 38 #include <sys/mount.h>
39 39 #include <sys/cmn_err.h>
40 40 #include "fs/fs_subr.h"
41 41 #include <sys/zfs_znode.h>
42 42 #include <sys/zfs_dir.h>
43 43 #include <sys/zil.h>
44 44 #include <sys/fs/zfs.h>
45 45 #include <sys/dmu.h>
46 46 #include <sys/dsl_prop.h>
47 47 #include <sys/dsl_dataset.h>
48 48 #include <sys/dsl_deleg.h>
49 49 #include <sys/spa.h>
50 50 #include <sys/zap.h>
51 51 #include <sys/sa.h>
52 52 #include <sys/sa_impl.h>
53 53 #include <sys/varargs.h>
54 54 #include <sys/policy.h>
55 55 #include <sys/atomic.h>
56 56 #include <sys/mkdev.h>
57 57 #include <sys/modctl.h>
58 58 #include <sys/refstr.h>
59 59 #include <sys/zfs_ioctl.h>
60 60 #include <sys/zfs_ctldir.h>
61 61 #include <sys/zfs_fuid.h>
62 62 #include <sys/bootconf.h>
63 63 #include <sys/sunddi.h>
64 64 #include <sys/dnlc.h>
65 65 #include <sys/dmu_objset.h>
66 66 #include <sys/spa_boot.h>
67 67 #include "zfs_comutil.h"
68 68
69 69 int zfsfstype;
70 70 vfsops_t *zfs_vfsops = NULL;
71 71 static major_t zfs_major;
72 72 static minor_t zfs_minor;
73 73 static kmutex_t zfs_dev_mtx;
74 74
75 75 extern int sys_shutdown;
76 76
77 77 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
78 78 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
79 79 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
80 80 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
81 81 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
82 82 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
83 83 static void zfs_freevfs(vfs_t *vfsp);
84 84
85 85 static const fs_operation_def_t zfs_vfsops_template[] = {
86 86 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
87 87 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
88 88 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
89 89 VFSNAME_ROOT, { .vfs_root = zfs_root },
90 90 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
91 91 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
92 92 VFSNAME_VGET, { .vfs_vget = zfs_vget },
93 93 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
94 94 NULL, NULL
95 95 };
96 96
97 97 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
98 98 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
99 99 NULL, NULL
100 100 };
101 101
102 102 /*
103 103 * We need to keep a count of active fs's.
104 104 * This is necessary to prevent our module
105 105 * from being unloaded after a umount -f
106 106 */
107 107 static uint32_t zfs_active_fs_count = 0;
108 108
109 109 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
110 110 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
111 111 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
112 112 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
113 113
114 114 /*
115 115 * MO_DEFAULT is not used since the default value is determined
116 116 * by the equivalent property.
117 117 */
118 118 static mntopt_t mntopts[] = {
119 119 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
120 120 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
121 121 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
122 122 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
123 123 };
124 124
125 125 static mntopts_t zfs_mntopts = {
126 126 sizeof (mntopts) / sizeof (mntopt_t),
127 127 mntopts
128 128 };
129 129
130 130 /*ARGSUSED*/
131 131 int
132 132 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
133 133 {
134 134 /*
135 135 * Data integrity is job one. We don't want a compromised kernel
136 136 * writing to the storage pool, so we never sync during panic.
137 137 */
138 138 if (panicstr)
139 139 return (0);
140 140
141 141 /*
142 142 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
143 143 * to sync metadata, which they would otherwise cache indefinitely.
144 144 * Semantically, the only requirement is that the sync be initiated.
145 145 * The DMU syncs out txgs frequently, so there's nothing to do.
146 146 */
147 147 if (flag & SYNC_ATTR)
148 148 return (0);
149 149
150 150 if (vfsp != NULL) {
151 151 /*
152 152 * Sync a specific filesystem.
153 153 */
154 154 zfsvfs_t *zfsvfs = vfsp->vfs_data;
155 155 dsl_pool_t *dp;
156 156
157 157 ZFS_ENTER(zfsvfs);
158 158 dp = dmu_objset_pool(zfsvfs->z_os);
159 159
160 160 /*
161 161 * If the system is shutting down, then skip any
162 162 * filesystems which may exist on a suspended pool.
163 163 */
164 164 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
165 165 ZFS_EXIT(zfsvfs);
166 166 return (0);
167 167 }
168 168
169 169 if (zfsvfs->z_log != NULL)
170 170 zil_commit(zfsvfs->z_log, 0);
171 171
172 172 ZFS_EXIT(zfsvfs);
173 173 } else {
174 174 /*
175 175 * Sync all ZFS filesystems. This is what happens when you
176 176 * run sync(1M). Unlike other filesystems, ZFS honors the
177 177 * request by waiting for all pools to commit all dirty data.
178 178 */
179 179 spa_sync_allpools();
180 180 }
181 181
182 182 return (0);
183 183 }
184 184
185 185 static int
186 186 zfs_create_unique_device(dev_t *dev)
187 187 {
188 188 major_t new_major;
189 189
190 190 do {
191 191 ASSERT3U(zfs_minor, <=, MAXMIN32);
192 192 minor_t start = zfs_minor;
193 193 do {
194 194 mutex_enter(&zfs_dev_mtx);
195 195 if (zfs_minor >= MAXMIN32) {
196 196 /*
197 197 * If we're still using the real major
198 198 * keep out of /dev/zfs and /dev/zvol minor
199 199 * number space. If we're using a getudev()'ed
200 200 * major number, we can use all of its minors.
201 201 */
202 202 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
203 203 zfs_minor = ZFS_MIN_MINOR;
204 204 else
205 205 zfs_minor = 0;
206 206 } else {
207 207 zfs_minor++;
208 208 }
209 209 *dev = makedevice(zfs_major, zfs_minor);
210 210 mutex_exit(&zfs_dev_mtx);
211 211 } while (vfs_devismounted(*dev) && zfs_minor != start);
212 212 if (zfs_minor == start) {
213 213 /*
214 214 * We are using all ~262,000 minor numbers for the
215 215 * current major number. Create a new major number.
216 216 */
217 217 if ((new_major = getudev()) == (major_t)-1) {
218 218 cmn_err(CE_WARN,
219 219 "zfs_mount: Can't get unique major "
220 220 "device number.");
221 221 return (-1);
222 222 }
223 223 mutex_enter(&zfs_dev_mtx);
224 224 zfs_major = new_major;
225 225 zfs_minor = 0;
226 226
227 227 mutex_exit(&zfs_dev_mtx);
228 228 } else {
229 229 break;
230 230 }
231 231 /* CONSTANTCONDITION */
232 232 } while (1);
233 233
234 234 return (0);
235 235 }
236 236
237 237 static void
238 238 atime_changed_cb(void *arg, uint64_t newval)
239 239 {
240 240 zfsvfs_t *zfsvfs = arg;
241 241
242 242 if (newval == TRUE) {
243 243 zfsvfs->z_atime = TRUE;
244 244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
245 245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
246 246 } else {
247 247 zfsvfs->z_atime = FALSE;
248 248 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
249 249 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
250 250 }
251 251 }
252 252
253 253 static void
254 254 xattr_changed_cb(void *arg, uint64_t newval)
255 255 {
256 256 zfsvfs_t *zfsvfs = arg;
257 257
258 258 if (newval == TRUE) {
259 259 /* XXX locking on vfs_flag? */
260 260 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
261 261 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
262 262 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
263 263 } else {
264 264 /* XXX locking on vfs_flag? */
265 265 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
266 266 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
267 267 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
268 268 }
269 269 }
270 270
271 271 static void
272 272 blksz_changed_cb(void *arg, uint64_t newval)
273 273 {
274 274 zfsvfs_t *zfsvfs = arg;
275 275
276 276 if (newval < SPA_MINBLOCKSIZE ||
277 277 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
278 278 newval = SPA_MAXBLOCKSIZE;
279 279
280 280 zfsvfs->z_max_blksz = newval;
281 281 zfsvfs->z_vfs->vfs_bsize = newval;
282 282 }
283 283
284 284 static void
285 285 readonly_changed_cb(void *arg, uint64_t newval)
286 286 {
287 287 zfsvfs_t *zfsvfs = arg;
288 288
289 289 if (newval) {
290 290 /* XXX locking on vfs_flag? */
291 291 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
292 292 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
293 293 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
294 294 } else {
295 295 /* XXX locking on vfs_flag? */
296 296 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
297 297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
298 298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
299 299 }
300 300 }
301 301
302 302 static void
303 303 devices_changed_cb(void *arg, uint64_t newval)
304 304 {
305 305 zfsvfs_t *zfsvfs = arg;
306 306
307 307 if (newval == FALSE) {
308 308 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
309 309 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
310 310 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
311 311 } else {
312 312 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
313 313 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
314 314 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
315 315 }
316 316 }
317 317
318 318 static void
319 319 setuid_changed_cb(void *arg, uint64_t newval)
320 320 {
321 321 zfsvfs_t *zfsvfs = arg;
322 322
323 323 if (newval == FALSE) {
324 324 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
325 325 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
326 326 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
327 327 } else {
328 328 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
329 329 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
330 330 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
331 331 }
332 332 }
333 333
334 334 static void
335 335 exec_changed_cb(void *arg, uint64_t newval)
336 336 {
337 337 zfsvfs_t *zfsvfs = arg;
338 338
339 339 if (newval == FALSE) {
340 340 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
341 341 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
342 342 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
343 343 } else {
344 344 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
345 345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
346 346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
347 347 }
348 348 }
349 349
350 350 /*
351 351 * The nbmand mount option can be changed at mount time.
352 352 * We can't allow it to be toggled on live file systems or incorrect
353 353 * behavior may be seen from cifs clients
354 354 *
355 355 * This property isn't registered via dsl_prop_register(), but this callback
356 356 * will be called when a file system is first mounted
357 357 */
358 358 static void
359 359 nbmand_changed_cb(void *arg, uint64_t newval)
360 360 {
361 361 zfsvfs_t *zfsvfs = arg;
362 362 if (newval == FALSE) {
363 363 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
364 364 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
365 365 } else {
366 366 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
367 367 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
368 368 }
369 369 }
370 370
371 371 static void
372 372 snapdir_changed_cb(void *arg, uint64_t newval)
373 373 {
374 374 zfsvfs_t *zfsvfs = arg;
375 375
376 376 zfsvfs->z_show_ctldir = newval;
377 377 }
378 378
379 379 static void
380 380 vscan_changed_cb(void *arg, uint64_t newval)
381 381 {
382 382 zfsvfs_t *zfsvfs = arg;
383 383
384 384 zfsvfs->z_vscan = newval;
385 385 }
386 386
387 387 static void
388 388 acl_mode_changed_cb(void *arg, uint64_t newval)
389 389 {
390 390 zfsvfs_t *zfsvfs = arg;
391 391
392 392 zfsvfs->z_acl_mode = newval;
393 393 }
394 394
395 395 static void
396 396 acl_inherit_changed_cb(void *arg, uint64_t newval)
397 397 {
398 398 zfsvfs_t *zfsvfs = arg;
399 399
400 400 zfsvfs->z_acl_inherit = newval;
401 401 }
402 402
403 403 static int
404 404 zfs_register_callbacks(vfs_t *vfsp)
405 405 {
406 406 struct dsl_dataset *ds = NULL;
407 407 objset_t *os = NULL;
408 408 zfsvfs_t *zfsvfs = NULL;
409 409 uint64_t nbmand;
410 410 boolean_t readonly = B_FALSE;
411 411 boolean_t do_readonly = B_FALSE;
412 412 boolean_t setuid = B_FALSE;
413 413 boolean_t do_setuid = B_FALSE;
414 414 boolean_t exec = B_FALSE;
415 415 boolean_t do_exec = B_FALSE;
416 416 boolean_t devices = B_FALSE;
417 417 boolean_t do_devices = B_FALSE;
418 418 boolean_t xattr = B_FALSE;
419 419 boolean_t do_xattr = B_FALSE;
420 420 boolean_t atime = B_FALSE;
421 421 boolean_t do_atime = B_FALSE;
422 422 int error = 0;
423 423
424 424 ASSERT(vfsp);
425 425 zfsvfs = vfsp->vfs_data;
426 426 ASSERT(zfsvfs);
427 427 os = zfsvfs->z_os;
428 428
429 429 /*
430 430 * The act of registering our callbacks will destroy any mount
431 431 * options we may have. In order to enable temporary overrides
432 432 * of mount options, we stash away the current values and
433 433 * restore them after we register the callbacks.
434 434 */
435 435 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
436 436 !spa_writeable(dmu_objset_spa(os))) {
437 437 readonly = B_TRUE;
438 438 do_readonly = B_TRUE;
439 439 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
440 440 readonly = B_FALSE;
441 441 do_readonly = B_TRUE;
442 442 }
443 443 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
444 444 devices = B_FALSE;
445 445 setuid = B_FALSE;
446 446 do_devices = B_TRUE;
447 447 do_setuid = B_TRUE;
448 448 } else {
449 449 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
450 450 devices = B_FALSE;
451 451 do_devices = B_TRUE;
452 452 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
453 453 devices = B_TRUE;
454 454 do_devices = B_TRUE;
455 455 }
456 456
457 457 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
458 458 setuid = B_FALSE;
459 459 do_setuid = B_TRUE;
460 460 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
461 461 setuid = B_TRUE;
462 462 do_setuid = B_TRUE;
463 463 }
464 464 }
465 465 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
466 466 exec = B_FALSE;
467 467 do_exec = B_TRUE;
468 468 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
469 469 exec = B_TRUE;
470 470 do_exec = B_TRUE;
471 471 }
472 472 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
473 473 xattr = B_FALSE;
474 474 do_xattr = B_TRUE;
475 475 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
476 476 xattr = B_TRUE;
477 477 do_xattr = B_TRUE;
478 478 }
479 479 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
480 480 atime = B_FALSE;
481 481 do_atime = B_TRUE;
482 482 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
483 483 atime = B_TRUE;
484 484 do_atime = B_TRUE;
485 485 }
486 486
487 487 /*
488 488 * nbmand is a special property. It can only be changed at
489 489 * mount time.
490 490 *
491 491 * This is weird, but it is documented to only be changeable
492 492 * at mount time.
493 493 */
494 494 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
495 495 nbmand = B_FALSE;
496 496 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
497 497 nbmand = B_TRUE;
498 498 } else {
499 499 char osname[MAXNAMELEN];
500 500
501 501 dmu_objset_name(os, osname);
502 502 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
503 503 NULL)) {
504 504 return (error);
505 505 }
506 506 }
507 507
508 508 /*
509 509 * Register property callbacks.
510 510 *
511 511 * It would probably be fine to just check for i/o error from
512 512 * the first prop_register(), but I guess I like to go
513 513 * overboard...
514 514 */
515 515 ds = dmu_objset_ds(os);
516 516 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
517 517 error = dsl_prop_register(ds,
518 518 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
519 519 error = error ? error : dsl_prop_register(ds,
520 520 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
521 521 error = error ? error : dsl_prop_register(ds,
522 522 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
523 523 error = error ? error : dsl_prop_register(ds,
524 524 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
525 525 error = error ? error : dsl_prop_register(ds,
526 526 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
527 527 error = error ? error : dsl_prop_register(ds,
528 528 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
529 529 error = error ? error : dsl_prop_register(ds,
530 530 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
531 531 error = error ? error : dsl_prop_register(ds,
532 532 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
533 533 error = error ? error : dsl_prop_register(ds,
534 534 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
535 535 error = error ? error : dsl_prop_register(ds,
536 536 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
537 537 zfsvfs);
538 538 error = error ? error : dsl_prop_register(ds,
539 539 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
540 540 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
541 541 if (error)
542 542 goto unregister;
543 543
544 544 /*
545 545 * Invoke our callbacks to restore temporary mount options.
546 546 */
547 547 if (do_readonly)
548 548 readonly_changed_cb(zfsvfs, readonly);
549 549 if (do_setuid)
550 550 setuid_changed_cb(zfsvfs, setuid);
551 551 if (do_exec)
552 552 exec_changed_cb(zfsvfs, exec);
553 553 if (do_devices)
554 554 devices_changed_cb(zfsvfs, devices);
555 555 if (do_xattr)
556 556 xattr_changed_cb(zfsvfs, xattr);
557 557 if (do_atime)
558 558 atime_changed_cb(zfsvfs, atime);
559 559
560 560 nbmand_changed_cb(zfsvfs, nbmand);
561 561
562 562 return (0);
563 563
564 564 unregister:
565 565 /*
566 566 * We may attempt to unregister some callbacks that are not
567 567 * registered, but this is OK; it will simply return ENOMSG,
568 568 * which we will ignore.
569 569 */
570 570 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
571 571 atime_changed_cb, zfsvfs);
572 572 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
573 573 xattr_changed_cb, zfsvfs);
574 574 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
575 575 blksz_changed_cb, zfsvfs);
576 576 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
577 577 readonly_changed_cb, zfsvfs);
578 578 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
579 579 devices_changed_cb, zfsvfs);
580 580 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
581 581 setuid_changed_cb, zfsvfs);
582 582 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
583 583 exec_changed_cb, zfsvfs);
584 584 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
585 585 snapdir_changed_cb, zfsvfs);
586 586 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
587 587 acl_mode_changed_cb, zfsvfs);
588 588 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
589 589 acl_inherit_changed_cb, zfsvfs);
590 590 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
591 591 vscan_changed_cb, zfsvfs);
592 592 return (error);
593 593 }
594 594
595 595 static int
596 596 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
597 597 uint64_t *userp, uint64_t *groupp)
598 598 {
599 599 /*
600 600 * Is it a valid type of object to track?
601 601 */
602 602 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
603 603 return (SET_ERROR(ENOENT));
604 604
605 605 /*
606 606 * If we have a NULL data pointer
607 607 * then assume the id's aren't changing and
608 608 * return EEXIST to the dmu to let it know to
609 609 * use the same ids
610 610 */
611 611 if (data == NULL)
612 612 return (SET_ERROR(EEXIST));
613 613
614 614 if (bonustype == DMU_OT_ZNODE) {
615 615 znode_phys_t *znp = data;
616 616 *userp = znp->zp_uid;
617 617 *groupp = znp->zp_gid;
618 618 } else {
619 619 int hdrsize;
620 620 sa_hdr_phys_t *sap = data;
621 621 sa_hdr_phys_t sa = *sap;
622 622 boolean_t swap = B_FALSE;
623 623
624 624 ASSERT(bonustype == DMU_OT_SA);
625 625
626 626 if (sa.sa_magic == 0) {
627 627 /*
628 628 * This should only happen for newly created
629 629 * files that haven't had the znode data filled
630 630 * in yet.
631 631 */
632 632 *userp = 0;
633 633 *groupp = 0;
634 634 return (0);
635 635 }
636 636 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
637 637 sa.sa_magic = SA_MAGIC;
638 638 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
639 639 swap = B_TRUE;
640 640 } else {
641 641 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
642 642 }
643 643
644 644 hdrsize = sa_hdrsize(&sa);
645 645 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
646 646 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
647 647 SA_UID_OFFSET));
648 648 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
649 649 SA_GID_OFFSET));
650 650 if (swap) {
651 651 *userp = BSWAP_64(*userp);
652 652 *groupp = BSWAP_64(*groupp);
653 653 }
654 654 }
655 655 return (0);
656 656 }
657 657
658 658 static void
659 659 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
660 660 char *domainbuf, int buflen, uid_t *ridp)
661 661 {
662 662 uint64_t fuid;
663 663 const char *domain;
664 664
665 665 fuid = strtonum(fuidstr, NULL);
666 666
667 667 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
668 668 if (domain)
669 669 (void) strlcpy(domainbuf, domain, buflen);
670 670 else
671 671 domainbuf[0] = '\0';
672 672 *ridp = FUID_RID(fuid);
673 673 }
674 674
675 675 static uint64_t
676 676 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
677 677 {
678 678 switch (type) {
679 679 case ZFS_PROP_USERUSED:
680 680 return (DMU_USERUSED_OBJECT);
681 681 case ZFS_PROP_GROUPUSED:
682 682 return (DMU_GROUPUSED_OBJECT);
683 683 case ZFS_PROP_USERQUOTA:
684 684 return (zfsvfs->z_userquota_obj);
685 685 case ZFS_PROP_GROUPQUOTA:
686 686 return (zfsvfs->z_groupquota_obj);
687 687 }
688 688 return (0);
689 689 }
690 690
691 691 int
692 692 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
693 693 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
694 694 {
695 695 int error;
696 696 zap_cursor_t zc;
697 697 zap_attribute_t za;
698 698 zfs_useracct_t *buf = vbuf;
699 699 uint64_t obj;
700 700
701 701 if (!dmu_objset_userspace_present(zfsvfs->z_os))
702 702 return (SET_ERROR(ENOTSUP));
703 703
704 704 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
705 705 if (obj == 0) {
706 706 *bufsizep = 0;
707 707 return (0);
708 708 }
709 709
710 710 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
711 711 (error = zap_cursor_retrieve(&zc, &za)) == 0;
712 712 zap_cursor_advance(&zc)) {
713 713 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
714 714 *bufsizep)
715 715 break;
716 716
717 717 fuidstr_to_sid(zfsvfs, za.za_name,
718 718 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
719 719
720 720 buf->zu_space = za.za_first_integer;
721 721 buf++;
722 722 }
723 723 if (error == ENOENT)
724 724 error = 0;
725 725
726 726 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
727 727 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
728 728 *cookiep = zap_cursor_serialize(&zc);
729 729 zap_cursor_fini(&zc);
730 730 return (error);
731 731 }
732 732
733 733 /*
734 734 * buf must be big enough (eg, 32 bytes)
735 735 */
736 736 static int
737 737 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
738 738 char *buf, boolean_t addok)
739 739 {
740 740 uint64_t fuid;
741 741 int domainid = 0;
742 742
743 743 if (domain && domain[0]) {
744 744 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
745 745 if (domainid == -1)
746 746 return (SET_ERROR(ENOENT));
747 747 }
748 748 fuid = FUID_ENCODE(domainid, rid);
749 749 (void) sprintf(buf, "%llx", (longlong_t)fuid);
750 750 return (0);
751 751 }
752 752
753 753 int
754 754 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
755 755 const char *domain, uint64_t rid, uint64_t *valp)
756 756 {
757 757 char buf[32];
758 758 int err;
759 759 uint64_t obj;
760 760
761 761 *valp = 0;
762 762
763 763 if (!dmu_objset_userspace_present(zfsvfs->z_os))
764 764 return (SET_ERROR(ENOTSUP));
765 765
766 766 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
767 767 if (obj == 0)
768 768 return (0);
769 769
770 770 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
771 771 if (err)
772 772 return (err);
773 773
774 774 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
775 775 if (err == ENOENT)
776 776 err = 0;
777 777 return (err);
778 778 }
779 779
780 780 int
781 781 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
782 782 const char *domain, uint64_t rid, uint64_t quota)
783 783 {
784 784 char buf[32];
785 785 int err;
786 786 dmu_tx_t *tx;
787 787 uint64_t *objp;
788 788 boolean_t fuid_dirtied;
789 789
790 790 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
791 791 return (SET_ERROR(EINVAL));
792 792
793 793 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
794 794 return (SET_ERROR(ENOTSUP));
795 795
796 796 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
797 797 &zfsvfs->z_groupquota_obj;
798 798
799 799 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
800 800 if (err)
801 801 return (err);
802 802 fuid_dirtied = zfsvfs->z_fuid_dirty;
803 803
804 804 tx = dmu_tx_create(zfsvfs->z_os);
805 805 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
806 806 if (*objp == 0) {
807 807 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
808 808 zfs_userquota_prop_prefixes[type]);
809 809 }
810 810 if (fuid_dirtied)
811 811 zfs_fuid_txhold(zfsvfs, tx);
812 812 err = dmu_tx_assign(tx, TXG_WAIT);
813 813 if (err) {
814 814 dmu_tx_abort(tx);
815 815 return (err);
816 816 }
817 817
818 818 mutex_enter(&zfsvfs->z_lock);
819 819 if (*objp == 0) {
820 820 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
821 821 DMU_OT_NONE, 0, tx);
822 822 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
823 823 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
824 824 }
825 825 mutex_exit(&zfsvfs->z_lock);
826 826
827 827 if (quota == 0) {
828 828 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
829 829 if (err == ENOENT)
830 830 err = 0;
831 831 } else {
832 832 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
833 833 }
834 834 ASSERT(err == 0);
835 835 if (fuid_dirtied)
836 836 zfs_fuid_sync(zfsvfs, tx);
837 837 dmu_tx_commit(tx);
838 838 return (err);
839 839 }
840 840
841 841 boolean_t
842 842 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
843 843 {
844 844 char buf[32];
845 845 uint64_t used, quota, usedobj, quotaobj;
846 846 int err;
847 847
848 848 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
849 849 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
850 850
851 851 if (quotaobj == 0 || zfsvfs->z_replay)
852 852 return (B_FALSE);
853 853
854 854 (void) sprintf(buf, "%llx", (longlong_t)fuid);
855 855 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
856 856 if (err != 0)
857 857 return (B_FALSE);
858 858
859 859 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
860 860 if (err != 0)
861 861 return (B_FALSE);
862 862 return (used >= quota);
863 863 }
864 864
865 865 boolean_t
866 866 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
867 867 {
868 868 uint64_t fuid;
869 869 uint64_t quotaobj;
870 870
871 871 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
872 872
873 873 fuid = isgroup ? zp->z_gid : zp->z_uid;
874 874
875 875 if (quotaobj == 0 || zfsvfs->z_replay)
876 876 return (B_FALSE);
877 877
878 878 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
879 879 }
880 880
881 881 int
882 882 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
883 883 {
884 884 objset_t *os;
885 885 zfsvfs_t *zfsvfs;
886 886 uint64_t zval;
887 887 int i, error;
888 888 uint64_t sa_obj;
889 889
890 890 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
891 891
892 892 /*
893 893 * We claim to always be readonly so we can open snapshots;
894 894 * other ZPL code will prevent us from writing to snapshots.
895 895 */
896 896 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
897 897 if (error) {
898 898 kmem_free(zfsvfs, sizeof (zfsvfs_t));
899 899 return (error);
900 900 }
901 901
902 902 /*
903 903 * Initialize the zfs-specific filesystem structure.
904 904 * Should probably make this a kmem cache, shuffle fields,
905 905 * and just bzero up to z_hold_mtx[].
906 906 */
907 907 zfsvfs->z_vfs = NULL;
908 908 zfsvfs->z_parent = zfsvfs;
909 909 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
910 910 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
911 911 zfsvfs->z_os = os;
912 912
913 913 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
914 914 if (error) {
915 915 goto out;
916 916 } else if (zfsvfs->z_version >
917 917 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
918 918 (void) printf("Can't mount a version %lld file system "
919 919 "on a version %lld pool\n. Pool must be upgraded to mount "
920 920 "this file system.", (u_longlong_t)zfsvfs->z_version,
921 921 (u_longlong_t)spa_version(dmu_objset_spa(os)));
922 922 error = SET_ERROR(ENOTSUP);
923 923 goto out;
924 924 }
925 925 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
926 926 goto out;
927 927 zfsvfs->z_norm = (int)zval;
928 928
929 929 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
930 930 goto out;
931 931 zfsvfs->z_utf8 = (zval != 0);
932 932
933 933 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
934 934 goto out;
935 935 zfsvfs->z_case = (uint_t)zval;
936 936
937 937 /*
938 938 * Fold case on file systems that are always or sometimes case
939 939 * insensitive.
940 940 */
941 941 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
942 942 zfsvfs->z_case == ZFS_CASE_MIXED)
943 943 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
944 944
945 945 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
946 946 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
947 947
948 948 if (zfsvfs->z_use_sa) {
949 949 /* should either have both of these objects or none */
950 950 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
951 951 &sa_obj);
952 952 if (error)
953 953 return (error);
954 954 } else {
955 955 /*
956 956 * Pre SA versions file systems should never touch
957 957 * either the attribute registration or layout objects.
958 958 */
959 959 sa_obj = 0;
960 960 }
961 961
962 962 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
963 963 &zfsvfs->z_attr_table);
964 964 if (error)
965 965 goto out;
966 966
967 967 if (zfsvfs->z_version >= ZPL_VERSION_SA)
968 968 sa_register_update_callback(os, zfs_sa_upgrade);
969 969
970 970 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
971 971 &zfsvfs->z_root);
972 972 if (error)
973 973 goto out;
974 974 ASSERT(zfsvfs->z_root != 0);
975 975
976 976 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
977 977 &zfsvfs->z_unlinkedobj);
978 978 if (error)
979 979 goto out;
980 980
981 981 error = zap_lookup(os, MASTER_NODE_OBJ,
982 982 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
983 983 8, 1, &zfsvfs->z_userquota_obj);
984 984 if (error && error != ENOENT)
985 985 goto out;
986 986
987 987 error = zap_lookup(os, MASTER_NODE_OBJ,
988 988 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
989 989 8, 1, &zfsvfs->z_groupquota_obj);
990 990 if (error && error != ENOENT)
991 991 goto out;
992 992
993 993 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
994 994 &zfsvfs->z_fuid_obj);
995 995 if (error && error != ENOENT)
996 996 goto out;
997 997
998 998 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
999 999 &zfsvfs->z_shares_dir);
1000 1000 if (error && error != ENOENT)
1001 1001 goto out;
1002 1002
1003 1003 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1004 1004 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1005 1005 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1006 1006 offsetof(znode_t, z_link_node));
1007 1007 rrw_init(&zfsvfs->z_teardown_lock, B_FALSE);
1008 1008 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1009 1009 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1010 1010 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1011 1011 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1012 1012
1013 1013 *zfvp = zfsvfs;
1014 1014 return (0);
1015 1015
1016 1016 out:
1017 1017 dmu_objset_disown(os, zfsvfs);
1018 1018 *zfvp = NULL;
1019 1019 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1020 1020 return (error);
1021 1021 }
1022 1022
1023 1023 static int
1024 1024 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1025 1025 {
1026 1026 int error;
1027 1027
1028 1028 error = zfs_register_callbacks(zfsvfs->z_vfs);
1029 1029 if (error)
1030 1030 return (error);
1031 1031
1032 1032 /*
1033 1033 * Set the objset user_ptr to track its zfsvfs.
1034 1034 */
1035 1035 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1036 1036 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1037 1037 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1038 1038
1039 1039 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1040 1040
1041 1041 /*
1042 1042 * If we are not mounting (ie: online recv), then we don't
1043 1043 * have to worry about replaying the log as we blocked all
1044 1044 * operations out since we closed the ZIL.
1045 1045 */
1046 1046 if (mounting) {
1047 1047 boolean_t readonly;
1048 1048
1049 1049 /*
1050 1050 * During replay we remove the read only flag to
1051 1051 * allow replays to succeed.
1052 1052 */
1053 1053 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1054 1054 if (readonly != 0)
1055 1055 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1056 1056 else
1057 1057 zfs_unlinked_drain(zfsvfs);
1058 1058
1059 1059 /*
1060 1060 * Parse and replay the intent log.
1061 1061 *
1062 1062 * Because of ziltest, this must be done after
1063 1063 * zfs_unlinked_drain(). (Further note: ziltest
1064 1064 * doesn't use readonly mounts, where
1065 1065 * zfs_unlinked_drain() isn't called.) This is because
1066 1066 * ziltest causes spa_sync() to think it's committed,
1067 1067 * but actually it is not, so the intent log contains
1068 1068 * many txg's worth of changes.
1069 1069 *
1070 1070 * In particular, if object N is in the unlinked set in
1071 1071 * the last txg to actually sync, then it could be
1072 1072 * actually freed in a later txg and then reallocated
1073 1073 * in a yet later txg. This would write a "create
1074 1074 * object N" record to the intent log. Normally, this
1075 1075 * would be fine because the spa_sync() would have
1076 1076 * written out the fact that object N is free, before
1077 1077 * we could write the "create object N" intent log
1078 1078 * record.
1079 1079 *
1080 1080 * But when we are in ziltest mode, we advance the "open
1081 1081 * txg" without actually spa_sync()-ing the changes to
1082 1082 * disk. So we would see that object N is still
1083 1083 * allocated and in the unlinked set, and there is an
1084 1084 * intent log record saying to allocate it.
1085 1085 */
1086 1086 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1087 1087 if (zil_replay_disable) {
1088 1088 zil_destroy(zfsvfs->z_log, B_FALSE);
1089 1089 } else {
1090 1090 zfsvfs->z_replay = B_TRUE;
1091 1091 zil_replay(zfsvfs->z_os, zfsvfs,
1092 1092 zfs_replay_vector);
1093 1093 zfsvfs->z_replay = B_FALSE;
1094 1094 }
1095 1095 }
1096 1096 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1097 1097 }
1098 1098
1099 1099 return (0);
1100 1100 }
1101 1101
1102 1102 void
1103 1103 zfsvfs_free(zfsvfs_t *zfsvfs)
1104 1104 {
1105 1105 int i;
1106 1106 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1107 1107
1108 1108 /*
1109 1109 * This is a barrier to prevent the filesystem from going away in
1110 1110 * zfs_znode_move() until we can safely ensure that the filesystem is
1111 1111 * not unmounted. We consider the filesystem valid before the barrier
1112 1112 * and invalid after the barrier.
1113 1113 */
1114 1114 rw_enter(&zfsvfs_lock, RW_READER);
1115 1115 rw_exit(&zfsvfs_lock);
1116 1116
1117 1117 zfs_fuid_destroy(zfsvfs);
1118 1118
1119 1119 mutex_destroy(&zfsvfs->z_znodes_lock);
1120 1120 mutex_destroy(&zfsvfs->z_lock);
1121 1121 list_destroy(&zfsvfs->z_all_znodes);
1122 1122 rrw_destroy(&zfsvfs->z_teardown_lock);
1123 1123 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1124 1124 rw_destroy(&zfsvfs->z_fuid_lock);
1125 1125 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1126 1126 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1127 1127 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1128 1128 }
1129 1129
1130 1130 static void
1131 1131 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1132 1132 {
1133 1133 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1134 1134 if (zfsvfs->z_vfs) {
1135 1135 if (zfsvfs->z_use_fuids) {
1136 1136 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1137 1137 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1138 1138 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1139 1139 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1140 1140 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1141 1141 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1142 1142 } else {
1143 1143 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1144 1144 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1145 1145 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1146 1146 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1147 1147 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1148 1148 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1149 1149 }
1150 1150 }
1151 1151 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1152 1152 }
1153 1153
1154 1154 static int
1155 1155 zfs_domount(vfs_t *vfsp, char *osname)
1156 1156 {
1157 1157 dev_t mount_dev;
1158 1158 uint64_t recordsize, fsid_guid;
1159 1159 int error = 0;
1160 1160 zfsvfs_t *zfsvfs;
1161 1161
1162 1162 ASSERT(vfsp);
1163 1163 ASSERT(osname);
1164 1164
1165 1165 error = zfsvfs_create(osname, &zfsvfs);
1166 1166 if (error)
1167 1167 return (error);
1168 1168 zfsvfs->z_vfs = vfsp;
1169 1169
1170 1170 /* Initialize the generic filesystem structure. */
1171 1171 vfsp->vfs_bcount = 0;
1172 1172 vfsp->vfs_data = NULL;
1173 1173
1174 1174 if (zfs_create_unique_device(&mount_dev) == -1) {
1175 1175 error = SET_ERROR(ENODEV);
1176 1176 goto out;
1177 1177 }
1178 1178 ASSERT(vfs_devismounted(mount_dev) == 0);
1179 1179
1180 1180 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1181 1181 NULL))
1182 1182 goto out;
1183 1183
1184 1184 vfsp->vfs_dev = mount_dev;
1185 1185 vfsp->vfs_fstype = zfsfstype;
1186 1186 vfsp->vfs_bsize = recordsize;
1187 1187 vfsp->vfs_flag |= VFS_NOTRUNC;
1188 1188 vfsp->vfs_data = zfsvfs;
1189 1189
1190 1190 /*
1191 1191 * The fsid is 64 bits, composed of an 8-bit fs type, which
1192 1192 * separates our fsid from any other filesystem types, and a
1193 1193 * 56-bit objset unique ID. The objset unique ID is unique to
1194 1194 * all objsets open on this system, provided by unique_create().
1195 1195 * The 8-bit fs type must be put in the low bits of fsid[1]
1196 1196 * because that's where other Solaris filesystems put it.
1197 1197 */
1198 1198 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1199 1199 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1200 1200 vfsp->vfs_fsid.val[0] = fsid_guid;
1201 1201 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1202 1202 zfsfstype & 0xFF;
1203 1203
1204 1204 /*
1205 1205 * Set features for file system.
1206 1206 */
1207 1207 zfs_set_fuid_feature(zfsvfs);
1208 1208 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1209 1209 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1210 1210 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1211 1211 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1212 1212 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1213 1213 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1214 1214 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1215 1215 }
1216 1216 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1217 1217
1218 1218 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1219 1219 uint64_t pval;
1220 1220
1221 1221 atime_changed_cb(zfsvfs, B_FALSE);
1222 1222 readonly_changed_cb(zfsvfs, B_TRUE);
1223 1223 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1224 1224 goto out;
1225 1225 xattr_changed_cb(zfsvfs, pval);
1226 1226 zfsvfs->z_issnap = B_TRUE;
1227 1227 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1228 1228
1229 1229 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1230 1230 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1231 1231 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1232 1232 } else {
|
↓ open down ↓ |
1232 lines elided |
↑ open up ↑ |
1233 1233 error = zfsvfs_setup(zfsvfs, B_TRUE);
1234 1234 }
1235 1235
1236 1236 if (!zfsvfs->z_issnap)
1237 1237 zfsctl_create(zfsvfs);
1238 1238 out:
1239 1239 if (error) {
1240 1240 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1241 1241 zfsvfs_free(zfsvfs);
1242 1242 } else {
1243 - atomic_add_32(&zfs_active_fs_count, 1);
1243 + atomic_inc_32(&zfs_active_fs_count);
1244 1244 }
1245 1245
1246 1246 return (error);
1247 1247 }
1248 1248
1249 1249 void
1250 1250 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1251 1251 {
1252 1252 objset_t *os = zfsvfs->z_os;
1253 1253 struct dsl_dataset *ds;
1254 1254
1255 1255 /*
1256 1256 * Unregister properties.
1257 1257 */
1258 1258 if (!dmu_objset_is_snapshot(os)) {
1259 1259 ds = dmu_objset_ds(os);
1260 1260 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1261 1261 zfsvfs) == 0);
1262 1262
1263 1263 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1264 1264 zfsvfs) == 0);
1265 1265
1266 1266 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1267 1267 zfsvfs) == 0);
1268 1268
1269 1269 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1270 1270 zfsvfs) == 0);
1271 1271
1272 1272 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1273 1273 zfsvfs) == 0);
1274 1274
1275 1275 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1276 1276 zfsvfs) == 0);
1277 1277
1278 1278 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1279 1279 zfsvfs) == 0);
1280 1280
1281 1281 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1282 1282 zfsvfs) == 0);
1283 1283
1284 1284 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1285 1285 zfsvfs) == 0);
1286 1286
1287 1287 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1288 1288 acl_inherit_changed_cb, zfsvfs) == 0);
1289 1289
1290 1290 VERIFY(dsl_prop_unregister(ds, "vscan",
1291 1291 vscan_changed_cb, zfsvfs) == 0);
1292 1292 }
1293 1293 }
1294 1294
1295 1295 /*
1296 1296 * Convert a decimal digit string to a uint64_t integer.
1297 1297 */
1298 1298 static int
1299 1299 str_to_uint64(char *str, uint64_t *objnum)
1300 1300 {
1301 1301 uint64_t num = 0;
1302 1302
1303 1303 while (*str) {
1304 1304 if (*str < '0' || *str > '9')
1305 1305 return (SET_ERROR(EINVAL));
1306 1306
1307 1307 num = num*10 + *str++ - '0';
1308 1308 }
1309 1309
1310 1310 *objnum = num;
1311 1311 return (0);
1312 1312 }
1313 1313
1314 1314 /*
1315 1315 * The boot path passed from the boot loader is in the form of
1316 1316 * "rootpool-name/root-filesystem-object-number'. Convert this
1317 1317 * string to a dataset name: "rootpool-name/root-filesystem-name".
1318 1318 */
1319 1319 static int
1320 1320 zfs_parse_bootfs(char *bpath, char *outpath)
1321 1321 {
1322 1322 char *slashp;
1323 1323 uint64_t objnum;
1324 1324 int error;
1325 1325
1326 1326 if (*bpath == 0 || *bpath == '/')
1327 1327 return (SET_ERROR(EINVAL));
1328 1328
1329 1329 (void) strcpy(outpath, bpath);
1330 1330
1331 1331 slashp = strchr(bpath, '/');
1332 1332
1333 1333 /* if no '/', just return the pool name */
1334 1334 if (slashp == NULL) {
1335 1335 return (0);
1336 1336 }
1337 1337
1338 1338 /* if not a number, just return the root dataset name */
1339 1339 if (str_to_uint64(slashp+1, &objnum)) {
1340 1340 return (0);
1341 1341 }
1342 1342
1343 1343 *slashp = '\0';
1344 1344 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1345 1345 *slashp = '/';
1346 1346
1347 1347 return (error);
1348 1348 }
1349 1349
1350 1350 /*
1351 1351 * Check that the hex label string is appropriate for the dataset being
1352 1352 * mounted into the global_zone proper.
1353 1353 *
1354 1354 * Return an error if the hex label string is not default or
1355 1355 * admin_low/admin_high. For admin_low labels, the corresponding
1356 1356 * dataset must be readonly.
1357 1357 */
1358 1358 int
1359 1359 zfs_check_global_label(const char *dsname, const char *hexsl)
1360 1360 {
1361 1361 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1362 1362 return (0);
1363 1363 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1364 1364 return (0);
1365 1365 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1366 1366 /* must be readonly */
1367 1367 uint64_t rdonly;
1368 1368
1369 1369 if (dsl_prop_get_integer(dsname,
1370 1370 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1371 1371 return (SET_ERROR(EACCES));
1372 1372 return (rdonly ? 0 : EACCES);
1373 1373 }
1374 1374 return (SET_ERROR(EACCES));
1375 1375 }
1376 1376
1377 1377 /*
1378 1378 * Determine whether the mount is allowed according to MAC check.
1379 1379 * by comparing (where appropriate) label of the dataset against
1380 1380 * the label of the zone being mounted into. If the dataset has
1381 1381 * no label, create one.
1382 1382 *
1383 1383 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1384 1384 */
1385 1385 static int
1386 1386 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1387 1387 {
1388 1388 int error, retv;
1389 1389 zone_t *mntzone = NULL;
1390 1390 ts_label_t *mnt_tsl;
1391 1391 bslabel_t *mnt_sl;
1392 1392 bslabel_t ds_sl;
1393 1393 char ds_hexsl[MAXNAMELEN];
1394 1394
1395 1395 retv = EACCES; /* assume the worst */
1396 1396
1397 1397 /*
1398 1398 * Start by getting the dataset label if it exists.
1399 1399 */
1400 1400 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1401 1401 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1402 1402 if (error)
1403 1403 return (SET_ERROR(EACCES));
1404 1404
1405 1405 /*
1406 1406 * If labeling is NOT enabled, then disallow the mount of datasets
1407 1407 * which have a non-default label already. No other label checks
1408 1408 * are needed.
1409 1409 */
1410 1410 if (!is_system_labeled()) {
1411 1411 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1412 1412 return (0);
1413 1413 return (SET_ERROR(EACCES));
1414 1414 }
1415 1415
1416 1416 /*
1417 1417 * Get the label of the mountpoint. If mounting into the global
1418 1418 * zone (i.e. mountpoint is not within an active zone and the
1419 1419 * zoned property is off), the label must be default or
1420 1420 * admin_low/admin_high only; no other checks are needed.
1421 1421 */
1422 1422 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1423 1423 if (mntzone->zone_id == GLOBAL_ZONEID) {
1424 1424 uint64_t zoned;
1425 1425
1426 1426 zone_rele(mntzone);
1427 1427
1428 1428 if (dsl_prop_get_integer(osname,
1429 1429 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1430 1430 return (SET_ERROR(EACCES));
1431 1431 if (!zoned)
1432 1432 return (zfs_check_global_label(osname, ds_hexsl));
1433 1433 else
1434 1434 /*
1435 1435 * This is the case of a zone dataset being mounted
1436 1436 * initially, before the zone has been fully created;
1437 1437 * allow this mount into global zone.
1438 1438 */
1439 1439 return (0);
1440 1440 }
1441 1441
1442 1442 mnt_tsl = mntzone->zone_slabel;
1443 1443 ASSERT(mnt_tsl != NULL);
1444 1444 label_hold(mnt_tsl);
1445 1445 mnt_sl = label2bslabel(mnt_tsl);
1446 1446
1447 1447 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1448 1448 /*
1449 1449 * The dataset doesn't have a real label, so fabricate one.
1450 1450 */
1451 1451 char *str = NULL;
1452 1452
1453 1453 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1454 1454 dsl_prop_set_string(osname,
1455 1455 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1456 1456 ZPROP_SRC_LOCAL, str) == 0)
1457 1457 retv = 0;
1458 1458 if (str != NULL)
1459 1459 kmem_free(str, strlen(str) + 1);
1460 1460 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1461 1461 /*
1462 1462 * Now compare labels to complete the MAC check. If the
1463 1463 * labels are equal then allow access. If the mountpoint
1464 1464 * label dominates the dataset label, allow readonly access.
1465 1465 * Otherwise, access is denied.
1466 1466 */
1467 1467 if (blequal(mnt_sl, &ds_sl))
1468 1468 retv = 0;
1469 1469 else if (bldominates(mnt_sl, &ds_sl)) {
1470 1470 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1471 1471 retv = 0;
1472 1472 }
1473 1473 }
1474 1474
1475 1475 label_rele(mnt_tsl);
1476 1476 zone_rele(mntzone);
1477 1477 return (retv);
1478 1478 }
1479 1479
1480 1480 static int
1481 1481 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1482 1482 {
1483 1483 int error = 0;
1484 1484 static int zfsrootdone = 0;
1485 1485 zfsvfs_t *zfsvfs = NULL;
1486 1486 znode_t *zp = NULL;
1487 1487 vnode_t *vp = NULL;
1488 1488 char *zfs_bootfs;
1489 1489 char *zfs_devid;
1490 1490
1491 1491 ASSERT(vfsp);
1492 1492
1493 1493 /*
1494 1494 * The filesystem that we mount as root is defined in the
1495 1495 * boot property "zfs-bootfs" with a format of
1496 1496 * "poolname/root-dataset-objnum".
1497 1497 */
1498 1498 if (why == ROOT_INIT) {
1499 1499 if (zfsrootdone++)
1500 1500 return (SET_ERROR(EBUSY));
1501 1501 /*
1502 1502 * the process of doing a spa_load will require the
1503 1503 * clock to be set before we could (for example) do
1504 1504 * something better by looking at the timestamp on
1505 1505 * an uberblock, so just set it to -1.
1506 1506 */
1507 1507 clkset(-1);
1508 1508
1509 1509 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1510 1510 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1511 1511 "bootfs name");
1512 1512 return (SET_ERROR(EINVAL));
1513 1513 }
1514 1514 zfs_devid = spa_get_bootprop("diskdevid");
1515 1515 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1516 1516 if (zfs_devid)
1517 1517 spa_free_bootprop(zfs_devid);
1518 1518 if (error) {
1519 1519 spa_free_bootprop(zfs_bootfs);
1520 1520 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1521 1521 error);
1522 1522 return (error);
1523 1523 }
1524 1524 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1525 1525 spa_free_bootprop(zfs_bootfs);
1526 1526 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1527 1527 error);
1528 1528 return (error);
1529 1529 }
1530 1530
1531 1531 spa_free_bootprop(zfs_bootfs);
1532 1532
1533 1533 if (error = vfs_lock(vfsp))
1534 1534 return (error);
1535 1535
1536 1536 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1537 1537 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1538 1538 goto out;
1539 1539 }
1540 1540
1541 1541 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1542 1542 ASSERT(zfsvfs);
1543 1543 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1544 1544 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1545 1545 goto out;
1546 1546 }
1547 1547
1548 1548 vp = ZTOV(zp);
1549 1549 mutex_enter(&vp->v_lock);
1550 1550 vp->v_flag |= VROOT;
1551 1551 mutex_exit(&vp->v_lock);
1552 1552 rootvp = vp;
1553 1553
1554 1554 /*
1555 1555 * Leave rootvp held. The root file system is never unmounted.
1556 1556 */
1557 1557
1558 1558 vfs_add((struct vnode *)0, vfsp,
1559 1559 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1560 1560 out:
1561 1561 vfs_unlock(vfsp);
1562 1562 return (error);
1563 1563 } else if (why == ROOT_REMOUNT) {
1564 1564 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1565 1565 vfsp->vfs_flag |= VFS_REMOUNT;
1566 1566
1567 1567 /* refresh mount options */
1568 1568 zfs_unregister_callbacks(vfsp->vfs_data);
1569 1569 return (zfs_register_callbacks(vfsp));
1570 1570
1571 1571 } else if (why == ROOT_UNMOUNT) {
1572 1572 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1573 1573 (void) zfs_sync(vfsp, 0, 0);
1574 1574 return (0);
1575 1575 }
1576 1576
1577 1577 /*
1578 1578 * if "why" is equal to anything else other than ROOT_INIT,
1579 1579 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1580 1580 */
1581 1581 return (SET_ERROR(ENOTSUP));
1582 1582 }
1583 1583
1584 1584 /*ARGSUSED*/
1585 1585 static int
1586 1586 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1587 1587 {
1588 1588 char *osname;
1589 1589 pathname_t spn;
1590 1590 int error = 0;
1591 1591 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1592 1592 UIO_SYSSPACE : UIO_USERSPACE;
1593 1593 int canwrite;
1594 1594
1595 1595 if (mvp->v_type != VDIR)
1596 1596 return (SET_ERROR(ENOTDIR));
1597 1597
1598 1598 mutex_enter(&mvp->v_lock);
1599 1599 if ((uap->flags & MS_REMOUNT) == 0 &&
1600 1600 (uap->flags & MS_OVERLAY) == 0 &&
1601 1601 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1602 1602 mutex_exit(&mvp->v_lock);
1603 1603 return (SET_ERROR(EBUSY));
1604 1604 }
1605 1605 mutex_exit(&mvp->v_lock);
1606 1606
1607 1607 /*
1608 1608 * ZFS does not support passing unparsed data in via MS_DATA.
1609 1609 * Users should use the MS_OPTIONSTR interface; this means
1610 1610 * that all option parsing is already done and the options struct
1611 1611 * can be interrogated.
1612 1612 */
1613 1613 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1614 1614 return (SET_ERROR(EINVAL));
1615 1615
1616 1616 /*
1617 1617 * Get the objset name (the "special" mount argument).
1618 1618 */
1619 1619 if (error = pn_get(uap->spec, fromspace, &spn))
1620 1620 return (error);
1621 1621
1622 1622 osname = spn.pn_path;
1623 1623
1624 1624 /*
1625 1625 * Check for mount privilege?
1626 1626 *
1627 1627 * If we don't have privilege then see if
1628 1628 * we have local permission to allow it
1629 1629 */
1630 1630 error = secpolicy_fs_mount(cr, mvp, vfsp);
1631 1631 if (error) {
1632 1632 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1633 1633 vattr_t vattr;
1634 1634
1635 1635 /*
1636 1636 * Make sure user is the owner of the mount point
1637 1637 * or has sufficient privileges.
1638 1638 */
1639 1639
1640 1640 vattr.va_mask = AT_UID;
1641 1641
1642 1642 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1643 1643 goto out;
1644 1644 }
1645 1645
1646 1646 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1647 1647 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1648 1648 goto out;
1649 1649 }
1650 1650 secpolicy_fs_mount_clearopts(cr, vfsp);
1651 1651 } else {
1652 1652 goto out;
1653 1653 }
1654 1654 }
1655 1655
1656 1656 /*
1657 1657 * Refuse to mount a filesystem if we are in a local zone and the
1658 1658 * dataset is not visible.
1659 1659 */
1660 1660 if (!INGLOBALZONE(curproc) &&
1661 1661 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1662 1662 error = SET_ERROR(EPERM);
1663 1663 goto out;
1664 1664 }
1665 1665
1666 1666 error = zfs_mount_label_policy(vfsp, osname);
1667 1667 if (error)
1668 1668 goto out;
1669 1669
1670 1670 /*
1671 1671 * When doing a remount, we simply refresh our temporary properties
1672 1672 * according to those options set in the current VFS options.
1673 1673 */
1674 1674 if (uap->flags & MS_REMOUNT) {
1675 1675 /* refresh mount options */
1676 1676 zfs_unregister_callbacks(vfsp->vfs_data);
1677 1677 error = zfs_register_callbacks(vfsp);
1678 1678 goto out;
1679 1679 }
1680 1680
1681 1681 error = zfs_domount(vfsp, osname);
1682 1682
1683 1683 /*
1684 1684 * Add an extra VFS_HOLD on our parent vfs so that it can't
1685 1685 * disappear due to a forced unmount.
1686 1686 */
1687 1687 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1688 1688 VFS_HOLD(mvp->v_vfsp);
1689 1689
1690 1690 out:
1691 1691 pn_free(&spn);
1692 1692 return (error);
1693 1693 }
1694 1694
1695 1695 static int
1696 1696 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1697 1697 {
1698 1698 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1699 1699 dev32_t d32;
1700 1700 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1701 1701
1702 1702 ZFS_ENTER(zfsvfs);
1703 1703
1704 1704 dmu_objset_space(zfsvfs->z_os,
1705 1705 &refdbytes, &availbytes, &usedobjs, &availobjs);
1706 1706
1707 1707 /*
1708 1708 * The underlying storage pool actually uses multiple block sizes.
1709 1709 * We report the fragsize as the smallest block size we support,
1710 1710 * and we report our blocksize as the filesystem's maximum blocksize.
1711 1711 */
1712 1712 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1713 1713 statp->f_bsize = zfsvfs->z_max_blksz;
1714 1714
1715 1715 /*
1716 1716 * The following report "total" blocks of various kinds in the
1717 1717 * file system, but reported in terms of f_frsize - the
1718 1718 * "fragment" size.
1719 1719 */
1720 1720
1721 1721 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1722 1722 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1723 1723 statp->f_bavail = statp->f_bfree; /* no root reservation */
1724 1724
1725 1725 /*
1726 1726 * statvfs() should really be called statufs(), because it assumes
1727 1727 * static metadata. ZFS doesn't preallocate files, so the best
1728 1728 * we can do is report the max that could possibly fit in f_files,
1729 1729 * and that minus the number actually used in f_ffree.
1730 1730 * For f_ffree, report the smaller of the number of object available
1731 1731 * and the number of blocks (each object will take at least a block).
1732 1732 */
1733 1733 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1734 1734 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1735 1735 statp->f_files = statp->f_ffree + usedobjs;
1736 1736
1737 1737 (void) cmpldev(&d32, vfsp->vfs_dev);
1738 1738 statp->f_fsid = d32;
1739 1739
1740 1740 /*
1741 1741 * We're a zfs filesystem.
1742 1742 */
1743 1743 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1744 1744
1745 1745 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1746 1746
1747 1747 statp->f_namemax = ZFS_MAXNAMELEN;
1748 1748
1749 1749 /*
1750 1750 * We have all of 32 characters to stuff a string here.
1751 1751 * Is there anything useful we could/should provide?
1752 1752 */
1753 1753 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1754 1754
1755 1755 ZFS_EXIT(zfsvfs);
1756 1756 return (0);
1757 1757 }
1758 1758
1759 1759 static int
1760 1760 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1761 1761 {
1762 1762 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1763 1763 znode_t *rootzp;
1764 1764 int error;
1765 1765
1766 1766 ZFS_ENTER(zfsvfs);
1767 1767
1768 1768 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1769 1769 if (error == 0)
1770 1770 *vpp = ZTOV(rootzp);
1771 1771
1772 1772 ZFS_EXIT(zfsvfs);
1773 1773 return (error);
1774 1774 }
1775 1775
1776 1776 /*
1777 1777 * Teardown the zfsvfs::z_os.
1778 1778 *
1779 1779 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1780 1780 * and 'z_teardown_inactive_lock' held.
1781 1781 */
1782 1782 static int
1783 1783 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1784 1784 {
1785 1785 znode_t *zp;
1786 1786
1787 1787 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1788 1788
1789 1789 if (!unmounting) {
1790 1790 /*
1791 1791 * We purge the parent filesystem's vfsp as the parent
1792 1792 * filesystem and all of its snapshots have their vnode's
1793 1793 * v_vfsp set to the parent's filesystem's vfsp. Note,
1794 1794 * 'z_parent' is self referential for non-snapshots.
1795 1795 */
1796 1796 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1797 1797 }
1798 1798
1799 1799 /*
1800 1800 * Close the zil. NB: Can't close the zil while zfs_inactive
1801 1801 * threads are blocked as zil_close can call zfs_inactive.
1802 1802 */
1803 1803 if (zfsvfs->z_log) {
1804 1804 zil_close(zfsvfs->z_log);
1805 1805 zfsvfs->z_log = NULL;
1806 1806 }
1807 1807
1808 1808 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1809 1809
1810 1810 /*
1811 1811 * If we are not unmounting (ie: online recv) and someone already
1812 1812 * unmounted this file system while we were doing the switcheroo,
1813 1813 * or a reopen of z_os failed then just bail out now.
1814 1814 */
1815 1815 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1816 1816 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1817 1817 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1818 1818 return (SET_ERROR(EIO));
1819 1819 }
1820 1820
1821 1821 /*
1822 1822 * At this point there are no vops active, and any new vops will
1823 1823 * fail with EIO since we have z_teardown_lock for writer (only
1824 1824 * relavent for forced unmount).
1825 1825 *
1826 1826 * Release all holds on dbufs.
1827 1827 */
1828 1828 mutex_enter(&zfsvfs->z_znodes_lock);
1829 1829 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1830 1830 zp = list_next(&zfsvfs->z_all_znodes, zp))
1831 1831 if (zp->z_sa_hdl) {
1832 1832 ASSERT(ZTOV(zp)->v_count > 0);
1833 1833 zfs_znode_dmu_fini(zp);
1834 1834 }
1835 1835 mutex_exit(&zfsvfs->z_znodes_lock);
1836 1836
1837 1837 /*
1838 1838 * If we are unmounting, set the unmounted flag and let new vops
1839 1839 * unblock. zfs_inactive will have the unmounted behavior, and all
1840 1840 * other vops will fail with EIO.
1841 1841 */
1842 1842 if (unmounting) {
1843 1843 zfsvfs->z_unmounted = B_TRUE;
1844 1844 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1845 1845 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1846 1846 }
1847 1847
1848 1848 /*
1849 1849 * z_os will be NULL if there was an error in attempting to reopen
1850 1850 * zfsvfs, so just return as the properties had already been
1851 1851 * unregistered and cached data had been evicted before.
1852 1852 */
1853 1853 if (zfsvfs->z_os == NULL)
1854 1854 return (0);
1855 1855
1856 1856 /*
1857 1857 * Unregister properties.
1858 1858 */
1859 1859 zfs_unregister_callbacks(zfsvfs);
1860 1860
1861 1861 /*
1862 1862 * Evict cached data
1863 1863 */
1864 1864 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1865 1865 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1866 1866 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1867 1867 dmu_objset_evict_dbufs(zfsvfs->z_os);
1868 1868
1869 1869 return (0);
1870 1870 }
1871 1871
1872 1872 /*ARGSUSED*/
1873 1873 static int
1874 1874 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1875 1875 {
1876 1876 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1877 1877 objset_t *os;
1878 1878 int ret;
1879 1879
1880 1880 ret = secpolicy_fs_unmount(cr, vfsp);
1881 1881 if (ret) {
1882 1882 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1883 1883 ZFS_DELEG_PERM_MOUNT, cr))
1884 1884 return (ret);
1885 1885 }
1886 1886
1887 1887 /*
1888 1888 * We purge the parent filesystem's vfsp as the parent filesystem
1889 1889 * and all of its snapshots have their vnode's v_vfsp set to the
1890 1890 * parent's filesystem's vfsp. Note, 'z_parent' is self
1891 1891 * referential for non-snapshots.
1892 1892 */
1893 1893 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1894 1894
1895 1895 /*
1896 1896 * Unmount any snapshots mounted under .zfs before unmounting the
1897 1897 * dataset itself.
1898 1898 */
1899 1899 if (zfsvfs->z_ctldir != NULL &&
1900 1900 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1901 1901 return (ret);
1902 1902 }
1903 1903
1904 1904 if (!(fflag & MS_FORCE)) {
1905 1905 /*
1906 1906 * Check the number of active vnodes in the file system.
1907 1907 * Our count is maintained in the vfs structure, but the
1908 1908 * number is off by 1 to indicate a hold on the vfs
1909 1909 * structure itself.
1910 1910 *
1911 1911 * The '.zfs' directory maintains a reference of its
1912 1912 * own, and any active references underneath are
1913 1913 * reflected in the vnode count.
1914 1914 */
1915 1915 if (zfsvfs->z_ctldir == NULL) {
1916 1916 if (vfsp->vfs_count > 1)
1917 1917 return (SET_ERROR(EBUSY));
1918 1918 } else {
1919 1919 if (vfsp->vfs_count > 2 ||
1920 1920 zfsvfs->z_ctldir->v_count > 1)
1921 1921 return (SET_ERROR(EBUSY));
1922 1922 }
1923 1923 }
1924 1924
1925 1925 vfsp->vfs_flag |= VFS_UNMOUNTED;
1926 1926
1927 1927 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1928 1928 os = zfsvfs->z_os;
1929 1929
1930 1930 /*
1931 1931 * z_os will be NULL if there was an error in
1932 1932 * attempting to reopen zfsvfs.
1933 1933 */
1934 1934 if (os != NULL) {
1935 1935 /*
1936 1936 * Unset the objset user_ptr.
1937 1937 */
1938 1938 mutex_enter(&os->os_user_ptr_lock);
1939 1939 dmu_objset_set_user(os, NULL);
1940 1940 mutex_exit(&os->os_user_ptr_lock);
1941 1941
1942 1942 /*
1943 1943 * Finally release the objset
1944 1944 */
1945 1945 dmu_objset_disown(os, zfsvfs);
1946 1946 }
1947 1947
1948 1948 /*
1949 1949 * We can now safely destroy the '.zfs' directory node.
1950 1950 */
1951 1951 if (zfsvfs->z_ctldir != NULL)
1952 1952 zfsctl_destroy(zfsvfs);
1953 1953
1954 1954 return (0);
1955 1955 }
1956 1956
1957 1957 static int
1958 1958 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1959 1959 {
1960 1960 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1961 1961 znode_t *zp;
1962 1962 uint64_t object = 0;
1963 1963 uint64_t fid_gen = 0;
1964 1964 uint64_t gen_mask;
1965 1965 uint64_t zp_gen;
1966 1966 int i, err;
1967 1967
1968 1968 *vpp = NULL;
1969 1969
1970 1970 ZFS_ENTER(zfsvfs);
1971 1971
1972 1972 if (fidp->fid_len == LONG_FID_LEN) {
1973 1973 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1974 1974 uint64_t objsetid = 0;
1975 1975 uint64_t setgen = 0;
1976 1976
1977 1977 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1978 1978 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1979 1979
1980 1980 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1981 1981 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1982 1982
1983 1983 ZFS_EXIT(zfsvfs);
1984 1984
1985 1985 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1986 1986 if (err)
1987 1987 return (SET_ERROR(EINVAL));
1988 1988 ZFS_ENTER(zfsvfs);
1989 1989 }
1990 1990
1991 1991 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1992 1992 zfid_short_t *zfid = (zfid_short_t *)fidp;
1993 1993
1994 1994 for (i = 0; i < sizeof (zfid->zf_object); i++)
1995 1995 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1996 1996
1997 1997 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1998 1998 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1999 1999 } else {
2000 2000 ZFS_EXIT(zfsvfs);
2001 2001 return (SET_ERROR(EINVAL));
2002 2002 }
2003 2003
2004 2004 /* A zero fid_gen means we are in the .zfs control directories */
2005 2005 if (fid_gen == 0 &&
2006 2006 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2007 2007 *vpp = zfsvfs->z_ctldir;
2008 2008 ASSERT(*vpp != NULL);
2009 2009 if (object == ZFSCTL_INO_SNAPDIR) {
2010 2010 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2011 2011 0, NULL, NULL, NULL, NULL, NULL) == 0);
2012 2012 } else {
2013 2013 VN_HOLD(*vpp);
2014 2014 }
2015 2015 ZFS_EXIT(zfsvfs);
2016 2016 return (0);
2017 2017 }
2018 2018
2019 2019 gen_mask = -1ULL >> (64 - 8 * i);
2020 2020
2021 2021 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2022 2022 if (err = zfs_zget(zfsvfs, object, &zp)) {
2023 2023 ZFS_EXIT(zfsvfs);
2024 2024 return (err);
2025 2025 }
2026 2026 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2027 2027 sizeof (uint64_t));
2028 2028 zp_gen = zp_gen & gen_mask;
2029 2029 if (zp_gen == 0)
2030 2030 zp_gen = 1;
2031 2031 if (zp->z_unlinked || zp_gen != fid_gen) {
2032 2032 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2033 2033 VN_RELE(ZTOV(zp));
2034 2034 ZFS_EXIT(zfsvfs);
2035 2035 return (SET_ERROR(EINVAL));
2036 2036 }
2037 2037
2038 2038 *vpp = ZTOV(zp);
2039 2039 ZFS_EXIT(zfsvfs);
2040 2040 return (0);
2041 2041 }
2042 2042
2043 2043 /*
2044 2044 * Block out VOPs and close zfsvfs_t::z_os
2045 2045 *
2046 2046 * Note, if successful, then we return with the 'z_teardown_lock' and
2047 2047 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2048 2048 * dataset and objset intact so that they can be atomically handed off during
2049 2049 * a subsequent rollback or recv operation and the resume thereafter.
2050 2050 */
2051 2051 int
2052 2052 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2053 2053 {
2054 2054 int error;
2055 2055
2056 2056 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2057 2057 return (error);
2058 2058
2059 2059 return (0);
2060 2060 }
2061 2061
2062 2062 /*
2063 2063 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2064 2064 * is an invariant across any of the operations that can be performed while the
2065 2065 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2066 2066 * are the same: the relevant objset and associated dataset are owned by
2067 2067 * zfsvfs, held, and long held on entry.
2068 2068 */
2069 2069 int
2070 2070 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2071 2071 {
2072 2072 int err;
2073 2073 znode_t *zp;
2074 2074 uint64_t sa_obj = 0;
2075 2075
2076 2076 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2077 2077 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2078 2078
2079 2079 /*
2080 2080 * We already own this, so just hold and rele it to update the
2081 2081 * objset_t, as the one we had before may have been evicted.
2082 2082 */
2083 2083 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2084 2084 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2085 2085 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2086 2086 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2087 2087
2088 2088 /*
2089 2089 * Make sure version hasn't changed
2090 2090 */
2091 2091
2092 2092 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2093 2093 &zfsvfs->z_version);
2094 2094
2095 2095 if (err)
2096 2096 goto bail;
2097 2097
2098 2098 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2099 2099 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2100 2100
2101 2101 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2102 2102 goto bail;
2103 2103
2104 2104 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2105 2105 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2106 2106 goto bail;
2107 2107
2108 2108 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2109 2109 sa_register_update_callback(zfsvfs->z_os,
2110 2110 zfs_sa_upgrade);
2111 2111
2112 2112 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2113 2113
2114 2114 zfs_set_fuid_feature(zfsvfs);
2115 2115
2116 2116 /*
2117 2117 * Attempt to re-establish all the active znodes with
2118 2118 * their dbufs. If a zfs_rezget() fails, then we'll let
2119 2119 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2120 2120 * when they try to use their znode.
2121 2121 */
2122 2122 mutex_enter(&zfsvfs->z_znodes_lock);
2123 2123 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2124 2124 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2125 2125 (void) zfs_rezget(zp);
2126 2126 }
2127 2127 mutex_exit(&zfsvfs->z_znodes_lock);
2128 2128
2129 2129 bail:
2130 2130 /* release the VOPs */
2131 2131 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2132 2132 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2133 2133
2134 2134 if (err) {
2135 2135 /*
2136 2136 * Since we couldn't setup the sa framework, try to force
2137 2137 * unmount this file system.
2138 2138 */
2139 2139 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2140 2140 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2141 2141 }
2142 2142 return (err);
2143 2143 }
2144 2144
2145 2145 static void
2146 2146 zfs_freevfs(vfs_t *vfsp)
2147 2147 {
2148 2148 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2149 2149
2150 2150 /*
|
↓ open down ↓ |
897 lines elided |
↑ open up ↑ |
2151 2151 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2152 2152 * from zfs_mount(). Release it here. If we came through
2153 2153 * zfs_mountroot() instead, we didn't grab an extra hold, so
2154 2154 * skip the VFS_RELE for rootvfs.
2155 2155 */
2156 2156 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2157 2157 VFS_RELE(zfsvfs->z_parent->z_vfs);
2158 2158
2159 2159 zfsvfs_free(zfsvfs);
2160 2160
2161 - atomic_add_32(&zfs_active_fs_count, -1);
2161 + atomic_dec_32(&zfs_active_fs_count);
2162 2162 }
2163 2163
2164 2164 /*
2165 2165 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2166 2166 * so we can't safely do any non-idempotent initialization here.
2167 2167 * Leave that to zfs_init() and zfs_fini(), which are called
2168 2168 * from the module's _init() and _fini() entry points.
2169 2169 */
2170 2170 /*ARGSUSED*/
2171 2171 static int
2172 2172 zfs_vfsinit(int fstype, char *name)
2173 2173 {
2174 2174 int error;
2175 2175
2176 2176 zfsfstype = fstype;
2177 2177
2178 2178 /*
2179 2179 * Setup vfsops and vnodeops tables.
2180 2180 */
2181 2181 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2182 2182 if (error != 0) {
2183 2183 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2184 2184 }
2185 2185
2186 2186 error = zfs_create_op_tables();
2187 2187 if (error) {
2188 2188 zfs_remove_op_tables();
2189 2189 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2190 2190 (void) vfs_freevfsops_by_type(zfsfstype);
2191 2191 return (error);
2192 2192 }
2193 2193
2194 2194 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2195 2195
2196 2196 /*
2197 2197 * Unique major number for all zfs mounts.
2198 2198 * If we run out of 32-bit minors, we'll getudev() another major.
2199 2199 */
2200 2200 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2201 2201 zfs_minor = ZFS_MIN_MINOR;
2202 2202
2203 2203 return (0);
2204 2204 }
2205 2205
2206 2206 void
2207 2207 zfs_init(void)
2208 2208 {
2209 2209 /*
2210 2210 * Initialize .zfs directory structures
2211 2211 */
2212 2212 zfsctl_init();
2213 2213
2214 2214 /*
2215 2215 * Initialize znode cache, vnode ops, etc...
2216 2216 */
2217 2217 zfs_znode_init();
2218 2218
2219 2219 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2220 2220 }
2221 2221
2222 2222 void
2223 2223 zfs_fini(void)
2224 2224 {
2225 2225 zfsctl_fini();
2226 2226 zfs_znode_fini();
2227 2227 }
2228 2228
2229 2229 int
2230 2230 zfs_busy(void)
2231 2231 {
2232 2232 return (zfs_active_fs_count != 0);
2233 2233 }
2234 2234
2235 2235 int
2236 2236 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2237 2237 {
2238 2238 int error;
2239 2239 objset_t *os = zfsvfs->z_os;
2240 2240 dmu_tx_t *tx;
2241 2241
2242 2242 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2243 2243 return (SET_ERROR(EINVAL));
2244 2244
2245 2245 if (newvers < zfsvfs->z_version)
2246 2246 return (SET_ERROR(EINVAL));
2247 2247
2248 2248 if (zfs_spa_version_map(newvers) >
2249 2249 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2250 2250 return (SET_ERROR(ENOTSUP));
2251 2251
2252 2252 tx = dmu_tx_create(os);
2253 2253 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2254 2254 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2255 2255 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2256 2256 ZFS_SA_ATTRS);
2257 2257 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2258 2258 }
2259 2259 error = dmu_tx_assign(tx, TXG_WAIT);
2260 2260 if (error) {
2261 2261 dmu_tx_abort(tx);
2262 2262 return (error);
2263 2263 }
2264 2264
2265 2265 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2266 2266 8, 1, &newvers, tx);
2267 2267
2268 2268 if (error) {
2269 2269 dmu_tx_commit(tx);
2270 2270 return (error);
2271 2271 }
2272 2272
2273 2273 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2274 2274 uint64_t sa_obj;
2275 2275
2276 2276 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2277 2277 SPA_VERSION_SA);
2278 2278 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2279 2279 DMU_OT_NONE, 0, tx);
2280 2280
2281 2281 error = zap_add(os, MASTER_NODE_OBJ,
2282 2282 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2283 2283 ASSERT0(error);
2284 2284
2285 2285 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2286 2286 sa_register_update_callback(os, zfs_sa_upgrade);
2287 2287 }
2288 2288
2289 2289 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2290 2290 "from %llu to %llu", zfsvfs->z_version, newvers);
2291 2291
2292 2292 dmu_tx_commit(tx);
2293 2293
2294 2294 zfsvfs->z_version = newvers;
2295 2295
2296 2296 zfs_set_fuid_feature(zfsvfs);
2297 2297
2298 2298 return (0);
2299 2299 }
2300 2300
2301 2301 /*
2302 2302 * Read a property stored within the master node.
2303 2303 */
2304 2304 int
2305 2305 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2306 2306 {
2307 2307 const char *pname;
2308 2308 int error = ENOENT;
2309 2309
2310 2310 /*
2311 2311 * Look up the file system's value for the property. For the
2312 2312 * version property, we look up a slightly different string.
2313 2313 */
2314 2314 if (prop == ZFS_PROP_VERSION)
2315 2315 pname = ZPL_VERSION_STR;
2316 2316 else
2317 2317 pname = zfs_prop_to_name(prop);
2318 2318
2319 2319 if (os != NULL)
2320 2320 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2321 2321
2322 2322 if (error == ENOENT) {
2323 2323 /* No value set, use the default value */
2324 2324 switch (prop) {
2325 2325 case ZFS_PROP_VERSION:
2326 2326 *value = ZPL_VERSION;
2327 2327 break;
2328 2328 case ZFS_PROP_NORMALIZE:
2329 2329 case ZFS_PROP_UTF8ONLY:
2330 2330 *value = 0;
2331 2331 break;
2332 2332 case ZFS_PROP_CASE:
2333 2333 *value = ZFS_CASE_SENSITIVE;
2334 2334 break;
2335 2335 default:
2336 2336 return (error);
2337 2337 }
2338 2338 error = 0;
2339 2339 }
2340 2340 return (error);
2341 2341 }
2342 2342
2343 2343 static vfsdef_t vfw = {
2344 2344 VFSDEF_VERSION,
2345 2345 MNTTYPE_ZFS,
2346 2346 zfs_vfsinit,
2347 2347 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2348 2348 VSW_XID|VSW_ZMOUNT,
2349 2349 &zfs_mntopts
2350 2350 };
2351 2351
2352 2352 struct modlfs zfs_modlfs = {
2353 2353 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2354 2354 };
|
↓ open down ↓ |
183 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX