xref: /illumos-gate/usr/src/uts/common/fs/dev/sdev_zvolops.c (revision aeac2d873b68a43f6650e0d0f021c02f5a653a21)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  * Copyright 2013 Joyent, Inc.  All rights reserved.
25  */
26 
27 /* vnode ops for the /dev/zvol directory */
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/sysmacros.h>
32 #include <sys/ddi.h>
33 #include <sys/sunndi.h>
34 #include <sys/sunldi.h>
35 #include <fs/fs_subr.h>
36 #include <sys/fs/dv_node.h>
37 #include <sys/fs/sdev_impl.h>
38 #include <sys/zfs_ioctl.h>
39 #include <sys/policy.h>
40 #include <sys/stat.h>
41 #include <sys/vfs_opreg.h>
42 
43 struct vnodeops	*devzvol_vnodeops;
44 static major_t devzvol_major;
45 static taskq_ent_t devzvol_zclist_task;
46 
47 static kmutex_t devzvol_mtx;
48 /* Below are protected by devzvol_mtx */
49 static boolean_t devzvol_isopen;
50 static boolean_t devzvol_zclist_task_running = B_FALSE;
51 static uint64_t devzvol_gen = 0;
52 static uint64_t devzvol_zclist;
53 static size_t devzvol_zclist_size;
54 static ldi_ident_t devzvol_li;
55 static ldi_handle_t devzvol_lh;
56 
57 /*
58  * we need to use ddi_mod* since fs/dev gets loaded early on in
59  * startup(), and linking fs/dev to fs/zfs would drag in a lot of
60  * other stuff (like drv/random) before the rest of the system is
61  * ready to go
62  */
63 ddi_modhandle_t zfs_mod;
64 int (*szcm)(char *);
65 int (*szn2m)(char *, minor_t *);
66 
67 int
68 sdev_zvol_create_minor(char *dsname)
69 {
70 	if (szcm == NULL)
71 		return (-1);
72 	return ((*szcm)(dsname));
73 }
74 
75 int
76 sdev_zvol_name2minor(char *dsname, minor_t *minor)
77 {
78 	if (szn2m == NULL)
79 		return (-1);
80 	return ((*szn2m)(dsname, minor));
81 }
82 
83 int
84 devzvol_open_zfs()
85 {
86 	int rc;
87 	dev_t dv;
88 
89 	devzvol_li = ldi_ident_from_anon();
90 	if (ldi_open_by_name("/dev/zfs", FREAD | FWRITE, kcred,
91 	    &devzvol_lh, devzvol_li))
92 		return (-1);
93 	if (zfs_mod == NULL && ((zfs_mod = ddi_modopen("fs/zfs",
94 	    KRTLD_MODE_FIRST, &rc)) == NULL)) {
95 		return (rc);
96 	}
97 	ASSERT(szcm == NULL && szn2m == NULL);
98 	if ((szcm = (int (*)(char *))
99 	    ddi_modsym(zfs_mod, "zvol_create_minor", &rc)) == NULL) {
100 		cmn_err(CE_WARN, "couldn't resolve zvol_create_minor");
101 		return (rc);
102 	}
103 	if ((szn2m = (int(*)(char *, minor_t *))
104 	    ddi_modsym(zfs_mod, "zvol_name2minor", &rc)) == NULL) {
105 		cmn_err(CE_WARN, "couldn't resolve zvol_name2minor");
106 		return (rc);
107 	}
108 	if (ldi_get_dev(devzvol_lh, &dv))
109 		return (-1);
110 	devzvol_major = getmajor(dv);
111 	return (0);
112 }
113 
114 void
115 devzvol_close_zfs()
116 {
117 	szcm = NULL;
118 	szn2m = NULL;
119 	(void) ldi_close(devzvol_lh, FREAD|FWRITE, kcred);
120 	ldi_ident_release(devzvol_li);
121 	if (zfs_mod != NULL) {
122 		(void) ddi_modclose(zfs_mod);
123 		zfs_mod = NULL;
124 	}
125 }
126 
127 int
128 devzvol_handle_ioctl(int cmd, zfs_cmd_t *zc, size_t *alloc_size)
129 {
130 	uint64_t cookie;
131 	int size = 8000;
132 	int unused;
133 	int rc;
134 
135 	if (cmd != ZFS_IOC_POOL_CONFIGS)
136 		mutex_enter(&devzvol_mtx);
137 	if (!devzvol_isopen) {
138 		if ((rc = devzvol_open_zfs()) == 0) {
139 			devzvol_isopen = B_TRUE;
140 		} else {
141 			if (cmd != ZFS_IOC_POOL_CONFIGS)
142 				mutex_exit(&devzvol_mtx);
143 			return (ENXIO);
144 		}
145 	}
146 	cookie = zc->zc_cookie;
147 again:
148 	zc->zc_nvlist_dst = (uint64_t)(intptr_t)kmem_alloc(size,
149 	    KM_SLEEP);
150 	zc->zc_nvlist_dst_size = size;
151 	rc = ldi_ioctl(devzvol_lh, cmd, (intptr_t)zc, FKIOCTL, kcred,
152 	    &unused);
153 	if (rc == ENOMEM) {
154 		int newsize;
155 		newsize = zc->zc_nvlist_dst_size;
156 		ASSERT(newsize > size);
157 		kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst, size);
158 		size = newsize;
159 		zc->zc_cookie = cookie;
160 		goto again;
161 	}
162 	if (alloc_size == NULL)
163 		kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst, size);
164 	else
165 		*alloc_size = size;
166 	if (cmd != ZFS_IOC_POOL_CONFIGS)
167 		mutex_exit(&devzvol_mtx);
168 	return (rc);
169 }
170 
171 /* figures out if the objset exists and returns its type */
172 int
173 devzvol_objset_check(char *dsname, dmu_objset_type_t *type)
174 {
175 	boolean_t	ispool;
176 	zfs_cmd_t	*zc;
177 	int rc;
178 
179 	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
180 	(void) strlcpy(zc->zc_name, dsname, MAXPATHLEN);
181 
182 	ispool = (strchr(dsname, '/') == NULL) ? B_TRUE : B_FALSE;
183 	if (!ispool && sdev_zvol_name2minor(dsname, NULL) == 0) {
184 		sdcmn_err13(("found cached minor node"));
185 		if (type)
186 			*type = DMU_OST_ZVOL;
187 		kmem_free(zc, sizeof (zfs_cmd_t));
188 		return (0);
189 	}
190 	rc = devzvol_handle_ioctl(ispool ? ZFS_IOC_POOL_STATS :
191 	    ZFS_IOC_OBJSET_STATS, zc, NULL);
192 	if (type && rc == 0)
193 		*type = (ispool) ? DMU_OST_ZFS :
194 		    zc->zc_objset_stats.dds_type;
195 	kmem_free(zc, sizeof (zfs_cmd_t));
196 	return (rc);
197 }
198 
199 /*
200  * Returns what the zfs dataset name should be, given the /dev/zvol
201  * path and an optional name (can be NULL).
202  *
203  * Note that if the name param is NULL, then path must be an
204  * actual dataset's directory and not one of the top-level
205  * /dev/zvol/{dsk,rdsk} dirs, as these do not correspond to a
206  * specific dataset.
207  */
208 char *
209 devzvol_make_dsname(const char *path, const char *name)
210 {
211 	char *dsname;
212 	const char *ptr;
213 	int dslen;
214 
215 	if (strcmp(path, ZVOL_DIR) == 0)
216 		return (NULL);
217 	if (name && (strcmp(name, ".") == 0 || strcmp(name, "..") == 0))
218 		return (NULL);
219 	ptr = path + strlen(ZVOL_DIR);
220 	if (strncmp(ptr, "/dsk", 4) == 0)
221 		ptr += strlen("/dsk");
222 	else if (strncmp(ptr, "/rdsk", 5) == 0)
223 		ptr += strlen("/rdsk");
224 	else
225 		return (NULL);
226 
227 	if (*ptr == '/')
228 		ptr++;
229 	else if (name == NULL)
230 		return (NULL);
231 
232 	dslen = strlen(ptr);
233 	if (dslen)
234 		dslen++;			/* plus null */
235 	if (name)
236 		dslen += strlen(name) + 1;	/* plus slash */
237 	dsname = kmem_zalloc(dslen, KM_SLEEP);
238 	if (*ptr) {
239 		(void) strlcpy(dsname, ptr, dslen);
240 		if (name)
241 			(void) strlcat(dsname, "/", dslen);
242 	}
243 	if (name)
244 		(void) strlcat(dsname, name, dslen);
245 	return (dsname);
246 }
247 
248 /*
249  * check if the zvol's sdev_node is still valid, which means make
250  * sure the zvol is still valid.  zvol minors aren't proactively
251  * destroyed when the zvol is destroyed, so we use a validator to clean
252  * these up (in other words, when such nodes are encountered during
253  * subsequent lookup() and readdir() operations) so that only valid
254  * nodes are returned.  The ordering between devname_lookup_func and
255  * devzvol_validate is a little inefficient in the case of invalid
256  * or stale nodes because devname_lookup_func calls
257  * devzvol_create_{dir, link}, then the validator says it's invalid,
258  * and then the node gets cleaned up.
259  */
260 int
261 devzvol_validate(struct sdev_node *dv)
262 {
263 	dmu_objset_type_t do_type;
264 	char *dsname;
265 	char *nm = dv->sdev_name;
266 	int rc;
267 
268 	sdcmn_err13(("validating ('%s' '%s')", dv->sdev_path, nm));
269 	/*
270 	 * validate only READY nodes; if someone is sitting on the
271 	 * directory of a dataset that just got destroyed we could
272 	 * get a zombie node which we just skip.
273 	 */
274 	if (dv->sdev_state != SDEV_READY) {
275 		sdcmn_err13(("skipping '%s'", nm));
276 		return (SDEV_VTOR_SKIP);
277 	}
278 
279 	if ((strcmp(dv->sdev_path, ZVOL_DIR "/dsk") == 0) ||
280 	    (strcmp(dv->sdev_path, ZVOL_DIR "/rdsk") == 0))
281 		return (SDEV_VTOR_VALID);
282 	dsname = devzvol_make_dsname(dv->sdev_path, NULL);
283 	if (dsname == NULL)
284 		return (SDEV_VTOR_INVALID);
285 
286 	rc = devzvol_objset_check(dsname, &do_type);
287 	sdcmn_err13(("  '%s' rc %d", dsname, rc));
288 	if (rc != 0) {
289 		kmem_free(dsname, strlen(dsname) + 1);
290 		return (SDEV_VTOR_INVALID);
291 	}
292 	sdcmn_err13(("  v_type %d do_type %d",
293 	    SDEVTOV(dv)->v_type, do_type));
294 	if ((SDEVTOV(dv)->v_type == VLNK && do_type != DMU_OST_ZVOL) ||
295 	    ((SDEVTOV(dv)->v_type == VBLK || SDEVTOV(dv)->v_type == VCHR) &&
296 	    do_type != DMU_OST_ZVOL) ||
297 	    (SDEVTOV(dv)->v_type == VDIR && do_type == DMU_OST_ZVOL)) {
298 		kmem_free(dsname, strlen(dsname) + 1);
299 		return (SDEV_VTOR_STALE);
300 	}
301 	if (SDEVTOV(dv)->v_type == VLNK) {
302 		char *ptr, *link;
303 		long val = 0;
304 		minor_t lminor, ominor;
305 
306 		rc = sdev_getlink(SDEVTOV(dv), &link);
307 		ASSERT(rc == 0);
308 
309 		ptr = strrchr(link, ':') + 1;
310 		rc = ddi_strtol(ptr, NULL, 10, &val);
311 		kmem_free(link, strlen(link) + 1);
312 		ASSERT(rc == 0 && val != 0);
313 		lminor = (minor_t)val;
314 		if (sdev_zvol_name2minor(dsname, &ominor) < 0 ||
315 		    ominor != lminor) {
316 			kmem_free(dsname, strlen(dsname) + 1);
317 			return (SDEV_VTOR_STALE);
318 		}
319 	}
320 	kmem_free(dsname, strlen(dsname) + 1);
321 	return (SDEV_VTOR_VALID);
322 }
323 
324 /*
325  * Taskq callback to update the devzvol_zclist.
326  *
327  * We need to defer this to the taskq to avoid it running with a user
328  * context that might be associated with some non-global zone, and thus
329  * not being able to list all of the pools on the entire system.
330  */
331 /*ARGSUSED*/
332 static void
333 devzvol_update_zclist_cb(void *arg)
334 {
335 	zfs_cmd_t	*zc;
336 	int		rc;
337 	size_t		size;
338 
339 	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
340 	mutex_enter(&devzvol_mtx);
341 	zc->zc_cookie = devzvol_gen;
342 
343 	rc = devzvol_handle_ioctl(ZFS_IOC_POOL_CONFIGS, zc, &size);
344 	switch (rc) {
345 		case 0:
346 			/* new generation */
347 			ASSERT(devzvol_gen != zc->zc_cookie);
348 			devzvol_gen = zc->zc_cookie;
349 			if (devzvol_zclist)
350 				kmem_free((void *)(uintptr_t)devzvol_zclist,
351 				    devzvol_zclist_size);
352 			devzvol_zclist = zc->zc_nvlist_dst;
353 			/* Keep the alloc'd size, not the nvlist size. */
354 			devzvol_zclist_size = size;
355 			break;
356 		default:
357 			/*
358 			 * Either there was no change in pool configuration
359 			 * since we last asked (rc == EEXIST) or we got a
360 			 * catastrophic error.
361 			 *
362 			 * Give up memory and exit.
363 			 */
364 			kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst,
365 			    size);
366 			break;
367 	}
368 
369 	VERIFY(devzvol_zclist_task_running == B_TRUE);
370 	devzvol_zclist_task_running = B_FALSE;
371 	mutex_exit(&devzvol_mtx);
372 
373 	kmem_free(zc, sizeof (zfs_cmd_t));
374 }
375 
376 static void
377 devzvol_update_zclist(void)
378 {
379 	mutex_enter(&devzvol_mtx);
380 	if (devzvol_zclist_task_running == B_TRUE) {
381 		mutex_exit(&devzvol_mtx);
382 		goto wait;
383 	}
384 
385 	devzvol_zclist_task_running = B_TRUE;
386 
387 	taskq_dispatch_ent(sdev_taskq, devzvol_update_zclist_cb, NULL, 0,
388 	    &devzvol_zclist_task);
389 
390 	mutex_exit(&devzvol_mtx);
391 
392 wait:
393 	taskq_wait(sdev_taskq);
394 }
395 
396 /*
397  * Creates sub-directories for each zpool as needed in response to a
398  * readdir on one of the /dev/zvol/{dsk,rdsk} directories.
399  */
400 void
401 devzvol_create_pool_dirs(struct vnode *dvp)
402 {
403 	nvlist_t *nv = NULL;
404 	nvpair_t *elem = NULL;
405 	int pools = 0;
406 	int rc;
407 
408 	sdcmn_err13(("devzvol_create_pool_dirs"));
409 
410 	devzvol_update_zclist();
411 
412 	mutex_enter(&devzvol_mtx);
413 
414 	rc = nvlist_unpack((char *)(uintptr_t)devzvol_zclist,
415 	    devzvol_zclist_size, &nv, 0);
416 	if (rc) {
417 		ASSERT(rc == 0);
418 		kmem_free((void *)(uintptr_t)devzvol_zclist,
419 		    devzvol_zclist_size);
420 		devzvol_gen = 0;
421 		devzvol_zclist = NULL;
422 		devzvol_zclist_size = 0;
423 		goto out;
424 	}
425 	mutex_exit(&devzvol_mtx);
426 	while ((elem = nvlist_next_nvpair(nv, elem)) != NULL) {
427 		struct vnode *vp;
428 		ASSERT(dvp->v_count > 0);
429 		rc = VOP_LOOKUP(dvp, nvpair_name(elem), &vp, NULL, 0,
430 		    NULL, kcred, NULL, 0, NULL);
431 		/* should either work, or not be visible from a zone */
432 		ASSERT(rc == 0 || rc == ENOENT);
433 		if (rc == 0)
434 			VN_RELE(vp);
435 		pools++;
436 	}
437 	nvlist_free(nv);
438 	mutex_enter(&devzvol_mtx);
439 	if (devzvol_isopen && pools == 0) {
440 		/* clean up so zfs can be unloaded */
441 		devzvol_close_zfs();
442 		devzvol_isopen = B_FALSE;
443 	}
444 out:
445 	mutex_exit(&devzvol_mtx);
446 }
447 
448 /*ARGSUSED3*/
449 static int
450 devzvol_create_dir(struct sdev_node *ddv, char *nm, void **arg,
451     cred_t *cred, void *whatever, char *whichever)
452 {
453 	timestruc_t now;
454 	struct vattr *vap = (struct vattr *)arg;
455 
456 	sdcmn_err13(("create_dir (%s) (%s) '%s'", ddv->sdev_name,
457 	    ddv->sdev_path, nm));
458 	ASSERT(strncmp(ddv->sdev_path, ZVOL_DIR,
459 	    strlen(ZVOL_DIR)) == 0);
460 	*vap = *sdev_getdefault_attr(VDIR);
461 	gethrestime(&now);
462 	vap->va_atime = now;
463 	vap->va_mtime = now;
464 	vap->va_ctime = now;
465 	return (0);
466 }
467 
468 /*ARGSUSED3*/
469 static int
470 devzvol_create_link(struct sdev_node *ddv, char *nm,
471     void **arg, cred_t *cred, void *whatever, char *whichever)
472 {
473 	minor_t minor;
474 	char *pathname = (char *)*arg;
475 	int rc;
476 	char *dsname;
477 	char *x;
478 	char str[MAXNAMELEN];
479 	sdcmn_err13(("create_link (%s) (%s) '%s'", ddv->sdev_name,
480 	    ddv->sdev_path, nm));
481 	dsname = devzvol_make_dsname(ddv->sdev_path, nm);
482 	rc = sdev_zvol_create_minor(dsname);
483 	if ((rc != 0 && rc != EEXIST && rc != EBUSY) ||
484 	    sdev_zvol_name2minor(dsname, &minor)) {
485 		sdcmn_err13(("devzvol_create_link %d", rc));
486 		kmem_free(dsname, strlen(dsname) + 1);
487 		return (-1);
488 	}
489 	kmem_free(dsname, strlen(dsname) + 1);
490 
491 	/*
492 	 * This is a valid zvol; create a symlink that points to the
493 	 * minor which was created under /devices/pseudo/zfs@0
494 	 */
495 	*pathname = '\0';
496 	for (x = ddv->sdev_path; x = strchr(x, '/'); x++)
497 		(void) strcat(pathname, "../");
498 	(void) snprintf(str, sizeof (str), ZVOL_PSEUDO_DEV "%u", minor);
499 	(void) strncat(pathname, str, MAXPATHLEN);
500 	if (strncmp(ddv->sdev_path, ZVOL_FULL_RDEV_DIR,
501 	    strlen(ZVOL_FULL_RDEV_DIR)) == 0)
502 		(void) strcat(pathname, ",raw");
503 	return (0);
504 }
505 
506 /* Clean zvol sdev_nodes that are no longer valid.  */
507 static void
508 devzvol_prunedir(struct sdev_node *ddv)
509 {
510 	struct sdev_node *dv;
511 
512 	ASSERT(RW_READ_HELD(&ddv->sdev_contents));
513 
514 	sdcmn_err13(("prunedir '%s'", ddv->sdev_name));
515 	ASSERT(strncmp(ddv->sdev_path, ZVOL_DIR, strlen(ZVOL_DIR)) == 0);
516 	if (rw_tryupgrade(&ddv->sdev_contents) == 0) {
517 		rw_exit(&ddv->sdev_contents);
518 		rw_enter(&ddv->sdev_contents, RW_WRITER);
519 	}
520 
521 	dv = SDEV_FIRST_ENTRY(ddv);
522 	while (dv) {
523 		sdcmn_err13(("sdev_name '%s'", dv->sdev_name));
524 
525 		switch (devzvol_validate(dv)) {
526 		case SDEV_VTOR_VALID:
527 		case SDEV_VTOR_SKIP:
528 			dv = SDEV_NEXT_ENTRY(ddv, dv);
529 			continue;
530 		case SDEV_VTOR_INVALID:
531 			sdcmn_err7(("prunedir: destroy invalid "
532 			    "node: %s\n", dv->sdev_name));
533 			break;
534 		}
535 
536 		if ((SDEVTOV(dv)->v_type == VDIR) &&
537 		    (sdev_cleandir(dv, NULL, 0) != 0)) {
538 			dv = SDEV_NEXT_ENTRY(ddv, dv);
539 			continue;
540 		}
541 		SDEV_HOLD(dv);
542 		/* remove the cache node */
543 		sdev_cache_update(ddv, &dv, dv->sdev_name,
544 		    SDEV_CACHE_DELETE);
545 		SDEV_RELE(dv);
546 		dv = SDEV_FIRST_ENTRY(ddv);
547 	}
548 	rw_downgrade(&ddv->sdev_contents);
549 }
550 
551 /*
552  * This function is used to create a dir or dev inside a zone's /dev when the
553  * zone has a zvol that is dynamically created within the zone (i.e. inside
554  * of a delegated dataset.  Since there is no /devices tree within a zone,
555  * we create the chr/blk devices directly inside the zone's /dev instead of
556  * making symlinks.
557  */
558 static int
559 devzvol_mk_ngz_node(struct sdev_node *parent, char *nm)
560 {
561 	struct vattr vattr;
562 	timestruc_t now;
563 	enum vtype expected_type = VDIR;
564 	dmu_objset_type_t do_type;
565 	struct sdev_node *dv = NULL;
566 	int res;
567 	char *dsname;
568 
569 	bzero(&vattr, sizeof (vattr));
570 	gethrestime(&now);
571 	vattr.va_mask = AT_TYPE|AT_MODE|AT_UID|AT_GID;
572 	vattr.va_uid = SDEV_UID_DEFAULT;
573 	vattr.va_gid = SDEV_GID_DEFAULT;
574 	vattr.va_type = VNON;
575 	vattr.va_atime = now;
576 	vattr.va_mtime = now;
577 	vattr.va_ctime = now;
578 
579 	if ((dsname = devzvol_make_dsname(parent->sdev_path, nm)) == NULL)
580 		return (ENOENT);
581 
582 	if (devzvol_objset_check(dsname, &do_type) != 0) {
583 		kmem_free(dsname, strlen(dsname) + 1);
584 		return (ENOENT);
585 	}
586 	if (do_type == DMU_OST_ZVOL)
587 		expected_type = VBLK;
588 
589 	if (expected_type == VDIR) {
590 		vattr.va_type = VDIR;
591 		vattr.va_mode = SDEV_DIRMODE_DEFAULT;
592 	} else {
593 		minor_t minor;
594 		dev_t devnum;
595 		int rc;
596 
597 		rc = sdev_zvol_create_minor(dsname);
598 		if ((rc != 0 && rc != EEXIST && rc != EBUSY) ||
599 		    sdev_zvol_name2minor(dsname, &minor)) {
600 			kmem_free(dsname, strlen(dsname) + 1);
601 			return (ENOENT);
602 		}
603 
604 		devnum = makedevice(devzvol_major, minor);
605 		vattr.va_rdev = devnum;
606 
607 		if (strstr(parent->sdev_path, "/rdsk/") != NULL)
608 			vattr.va_type = VCHR;
609 		else
610 			vattr.va_type = VBLK;
611 		vattr.va_mode = SDEV_DEVMODE_DEFAULT;
612 	}
613 	kmem_free(dsname, strlen(dsname) + 1);
614 
615 	rw_enter(&parent->sdev_contents, RW_WRITER);
616 
617 	res = sdev_mknode(parent, nm, &dv, &vattr,
618 	    NULL, NULL, kcred, SDEV_READY);
619 	rw_exit(&parent->sdev_contents);
620 	if (res != 0)
621 		return (ENOENT);
622 
623 	SDEV_RELE(dv);
624 	return (0);
625 }
626 
627 /*ARGSUSED*/
628 static int
629 devzvol_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
630     struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cred,
631     caller_context_t *ct, int *direntflags, pathname_t *realpnp)
632 {
633 	enum vtype expected_type = VDIR;
634 	struct sdev_node *parent = VTOSDEV(dvp);
635 	char *dsname;
636 	dmu_objset_type_t do_type;
637 	int error;
638 
639 	sdcmn_err13(("devzvol_lookup '%s' '%s'", parent->sdev_path, nm));
640 	*vpp = NULL;
641 	/* execute access is required to search the directory */
642 	if ((error = VOP_ACCESS(dvp, VEXEC, 0, cred, ct)) != 0)
643 		return (error);
644 
645 	rw_enter(&parent->sdev_contents, RW_READER);
646 	if (SDEV_IS_GLOBAL(parent)) {
647 		/*
648 		 * During iter_datasets, don't create GZ dev when running in
649 		 * NGZ.  We can't return ENOENT here since that could
650 		 * incorrectly trigger the creation of the dev from the
651 		 * recursive call through prof_filldir during iter_datasets.
652 		 */
653 		if (getzoneid() != GLOBAL_ZONEID) {
654 			rw_exit(&parent->sdev_contents);
655 			return (EPERM);
656 		}
657 	} else {
658 		int res;
659 
660 		rw_exit(&parent->sdev_contents);
661 
662 		/*
663 		 * If we're in the global zone and reach down into a non-global
664 		 * zone's /dev/zvol then this action could trigger the creation
665 		 * of all of the zvol devices for every zone into the non-global
666 		 * zone's /dev tree. This could be a big security hole. To
667 		 * prevent this, disallow the global zone from looking inside
668 		 * a non-global zones /dev/zvol. This behavior is similar to
669 		 * delegated datasets, which cannot be used by the global zone.
670 		 */
671 		if (getzoneid() == GLOBAL_ZONEID)
672 			return (EPERM);
673 
674 		res = prof_lookup(dvp, nm, vpp, cred);
675 
676 		/*
677 		 * We won't find a zvol that was dynamically created inside
678 		 * a NGZ, within a delegated dataset, in the zone's dev profile
679 		 * but prof_lookup will also find it via sdev_cache_lookup.
680 		 */
681 		if (res == ENOENT) {
682 			/*
683 			 * We have to create the sdev node for the dymamically
684 			 * created zvol.
685 			 */
686 			if (devzvol_mk_ngz_node(parent, nm) != 0)
687 				return (ENOENT);
688 			res = prof_lookup(dvp, nm, vpp, cred);
689 		}
690 
691 		return (res);
692 	}
693 
694 	dsname = devzvol_make_dsname(parent->sdev_path, nm);
695 	rw_exit(&parent->sdev_contents);
696 	sdcmn_err13(("rvp dsname %s", dsname ? dsname : "(null)"));
697 	if (dsname) {
698 		error = devzvol_objset_check(dsname, &do_type);
699 		if (error != 0) {
700 			error = ENOENT;
701 			goto out;
702 		}
703 		if (do_type == DMU_OST_ZVOL)
704 			expected_type = VLNK;
705 	}
706 	/*
707 	 * the callbacks expect:
708 	 *
709 	 * parent->sdev_path		   nm
710 	 * /dev/zvol			   {r}dsk
711 	 * /dev/zvol/{r}dsk		   <pool name>
712 	 * /dev/zvol/{r}dsk/<dataset name> <last ds component>
713 	 *
714 	 * sdev_name is always last path component of sdev_path
715 	 */
716 	if (expected_type == VDIR) {
717 		error = devname_lookup_func(parent, nm, vpp, cred,
718 		    devzvol_create_dir, SDEV_VATTR);
719 	} else {
720 		error = devname_lookup_func(parent, nm, vpp, cred,
721 		    devzvol_create_link, SDEV_VLINK);
722 	}
723 	sdcmn_err13(("devzvol_lookup %d %d", expected_type, error));
724 	ASSERT(error || ((*vpp)->v_type == expected_type));
725 out:
726 	if (dsname)
727 		kmem_free(dsname, strlen(dsname) + 1);
728 	sdcmn_err13(("devzvol_lookup %d", error));
729 	return (error);
730 }
731 
732 /*
733  * We allow create to find existing nodes
734  *	- if the node doesn't exist - EROFS
735  *	- creating an existing dir read-only succeeds, otherwise EISDIR
736  *	- exclusive creates fail - EEXIST
737  */
738 /*ARGSUSED2*/
739 static int
740 devzvol_create(struct vnode *dvp, char *nm, struct vattr *vap, vcexcl_t excl,
741     int mode, struct vnode **vpp, struct cred *cred, int flag,
742     caller_context_t *ct, vsecattr_t *vsecp)
743 {
744 	int error;
745 	struct vnode *vp;
746 
747 	*vpp = NULL;
748 
749 	error = devzvol_lookup(dvp, nm, &vp, NULL, 0, NULL, cred, ct, NULL,
750 	    NULL);
751 	if (error == 0) {
752 		if (excl == EXCL)
753 			error = EEXIST;
754 		else if (vp->v_type == VDIR && (mode & VWRITE))
755 			error = EISDIR;
756 		else
757 			error = VOP_ACCESS(vp, mode, 0, cred, ct);
758 
759 		if (error) {
760 			VN_RELE(vp);
761 		} else
762 			*vpp = vp;
763 	} else if (error == ENOENT) {
764 		error = EROFS;
765 	}
766 
767 	return (error);
768 }
769 
770 void sdev_iter_snapshots(struct vnode *dvp, char *name);
771 
772 void
773 sdev_iter_datasets(struct vnode *dvp, int arg, char *name)
774 {
775 	zfs_cmd_t	*zc;
776 	int rc;
777 
778 	sdcmn_err13(("iter name is '%s' (arg %x)", name, arg));
779 	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
780 	(void) strcpy(zc->zc_name, name);
781 
782 	while ((rc = devzvol_handle_ioctl(arg, zc, B_FALSE)) == 0) {
783 		struct vnode *vpp;
784 		char *ptr;
785 
786 		sdcmn_err13(("  name %s", zc->zc_name));
787 		if (strchr(zc->zc_name, '$') || strchr(zc->zc_name, '%'))
788 			goto skip;
789 		ptr = strrchr(zc->zc_name, '/') + 1;
790 		rc = devzvol_lookup(dvp, ptr, &vpp, NULL, 0, NULL,
791 		    kcred, NULL, NULL, NULL);
792 		if (rc == 0) {
793 			VN_RELE(vpp);
794 		} else if (rc == ENOENT) {
795 			goto skip;
796 		} else {
797 			/*
798 			 * EBUSY == problem with zvols's dmu holds?
799 			 * EPERM when in a NGZ and traversing up and out.
800 			 */
801 			goto skip;
802 		}
803 		if (arg == ZFS_IOC_DATASET_LIST_NEXT &&
804 		    zc->zc_objset_stats.dds_type != DMU_OST_ZFS)
805 			sdev_iter_snapshots(dvp, zc->zc_name);
806 skip:
807 		(void) strcpy(zc->zc_name, name);
808 	}
809 	kmem_free(zc, sizeof (zfs_cmd_t));
810 }
811 
812 void
813 sdev_iter_snapshots(struct vnode *dvp, char *name)
814 {
815 	sdev_iter_datasets(dvp, ZFS_IOC_SNAPSHOT_LIST_NEXT, name);
816 }
817 
818 /*ARGSUSED4*/
819 static int
820 devzvol_readdir(struct vnode *dvp, struct uio *uiop, struct cred *cred,
821     int *eofp, caller_context_t *ct_unused, int flags_unused)
822 {
823 	struct sdev_node *sdvp = VTOSDEV(dvp);
824 	char *ptr;
825 
826 	sdcmn_err13(("zv readdir of '%s' %s'", sdvp->sdev_path,
827 	    sdvp->sdev_name));
828 
829 	if (strcmp(sdvp->sdev_path, ZVOL_DIR) == 0) {
830 		struct vnode *vp;
831 
832 		rw_exit(&sdvp->sdev_contents);
833 		(void) devname_lookup_func(sdvp, "dsk", &vp, cred,
834 		    devzvol_create_dir, SDEV_VATTR);
835 		VN_RELE(vp);
836 		(void) devname_lookup_func(sdvp, "rdsk", &vp, cred,
837 		    devzvol_create_dir, SDEV_VATTR);
838 		VN_RELE(vp);
839 		rw_enter(&sdvp->sdev_contents, RW_READER);
840 		return (devname_readdir_func(dvp, uiop, cred, eofp, 0));
841 	}
842 	if (uiop->uio_offset == 0)
843 		devzvol_prunedir(sdvp);
844 	ptr = sdvp->sdev_path + strlen(ZVOL_DIR);
845 	if ((strcmp(ptr, "/dsk") == 0) || (strcmp(ptr, "/rdsk") == 0)) {
846 		rw_exit(&sdvp->sdev_contents);
847 		devzvol_create_pool_dirs(dvp);
848 		rw_enter(&sdvp->sdev_contents, RW_READER);
849 		return (devname_readdir_func(dvp, uiop, cred, eofp, 0));
850 	}
851 
852 	ptr = strchr(ptr + 1, '/');
853 	if (ptr == NULL)
854 		return (ENOENT);
855 	ptr++;
856 	rw_exit(&sdvp->sdev_contents);
857 	sdev_iter_datasets(dvp, ZFS_IOC_DATASET_LIST_NEXT, ptr);
858 	rw_enter(&sdvp->sdev_contents, RW_READER);
859 	return (devname_readdir_func(dvp, uiop, cred, eofp, 0));
860 }
861 
862 const fs_operation_def_t devzvol_vnodeops_tbl[] = {
863 	VOPNAME_READDIR,	{ .vop_readdir = devzvol_readdir },
864 	VOPNAME_LOOKUP,		{ .vop_lookup = devzvol_lookup },
865 	VOPNAME_CREATE,		{ .vop_create = devzvol_create },
866 	VOPNAME_RENAME,		{ .error = fs_nosys },
867 	VOPNAME_MKDIR,		{ .error = fs_nosys },
868 	VOPNAME_RMDIR,		{ .error = fs_nosys },
869 	VOPNAME_REMOVE,		{ .error = fs_nosys },
870 	VOPNAME_SYMLINK,	{ .error = fs_nosys },
871 	NULL,			NULL
872 };
873