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 (c) 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 */
25
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/errno.h>
29 #include <sys/kmem.h>
30 #include <sys/conf.h>
31 #include <sys/sunddi.h>
32 #include <sys/zfs_ioctl.h>
33 #include <sys/zfs_onexit.h>
34 #include <sys/zvol.h>
35
36 /*
37 * ZFS kernel routines may add/delete callback routines to be invoked
38 * upon process exit (triggered via the close operation from the /dev/zfs
39 * driver).
40 *
41 * These cleanup callbacks are intended to allow for the accumulation
42 * of kernel state across multiple ioctls. User processes participate
43 * by opening ZFS_DEV with O_EXCL. This causes the ZFS driver to do a
44 * clone-open, generating a unique minor number. The process then passes
45 * along that file descriptor to each ioctl that might have a cleanup operation.
46 *
47 * Consumers of the onexit routines should call zfs_onexit_fd_hold() early
48 * on to validate the given fd and add a reference to its file table entry.
49 * This allows the consumer to do its work and then add a callback, knowing
50 * that zfs_onexit_add_cb() won't fail with EBADF. When finished, consumers
51 * should call zfs_onexit_fd_rele().
52 *
53 * A simple example is zfs_ioc_recv(), where we might create an AVL tree
54 * with dataset/GUID mappings and then reuse that tree on subsequent
55 * zfs_ioc_recv() calls.
56 *
57 * On the first zfs_ioc_recv() call, dmu_recv_stream() will kmem_alloc()
58 * the AVL tree and pass it along with a callback function to
59 * zfs_onexit_add_cb(). The zfs_onexit_add_cb() routine will register the
60 * callback and return an action handle.
61 *
62 * The action handle is then passed from user space to subsequent
63 * zfs_ioc_recv() calls, so that dmu_recv_stream() can fetch its AVL tree
64 * by calling zfs_onexit_cb_data() with the device minor number and
65 * action handle.
66 *
67 * If the user process exits abnormally, the callback is invoked implicitly
68 * as part of the driver close operation. Once the user space process is
69 * finished with the accumulated kernel state, it can also just call close(2)
70 * on the cleanup fd to trigger the cleanup callback.
71 */
72
73 void
zfs_onexit_init(zfs_onexit_t ** zop)74 zfs_onexit_init(zfs_onexit_t **zop)
75 {
76 zfs_onexit_t *zo;
77
78 zo = *zop = kmem_zalloc(sizeof (zfs_onexit_t), KM_SLEEP);
79 mutex_init(&zo->zo_lock, NULL, MUTEX_DEFAULT, NULL);
80 list_create(&zo->zo_actions, sizeof (zfs_onexit_action_node_t),
81 offsetof(zfs_onexit_action_node_t, za_link));
82 }
83
84 void
zfs_onexit_destroy(zfs_onexit_t * zo)85 zfs_onexit_destroy(zfs_onexit_t *zo)
86 {
87 zfs_onexit_action_node_t *ap;
88
89 mutex_enter(&zo->zo_lock);
90 while ((ap = list_head(&zo->zo_actions)) != NULL) {
91 list_remove(&zo->zo_actions, ap);
92 mutex_exit(&zo->zo_lock);
93 ap->za_func(ap->za_data);
94 kmem_free(ap, sizeof (zfs_onexit_action_node_t));
95 mutex_enter(&zo->zo_lock);
96 }
97 mutex_exit(&zo->zo_lock);
98
99 list_destroy(&zo->zo_actions);
100 mutex_destroy(&zo->zo_lock);
101 kmem_free(zo, sizeof (zfs_onexit_t));
102 }
103
104 static int
zfs_onexit_minor_to_state(minor_t minor,zfs_onexit_t ** zo)105 zfs_onexit_minor_to_state(minor_t minor, zfs_onexit_t **zo)
106 {
107 *zo = zfsdev_get_soft_state(minor, ZSST_CTLDEV);
108 if (*zo == NULL)
109 return (SET_ERROR(EBADF));
110
111 return (0);
112 }
113
114 /*
115 * Consumers might need to operate by minor number instead of fd, since
116 * they might be running in another thread (e.g. txg_sync_thread). Callers
117 * of this function must call zfs_onexit_fd_rele() when they're finished
118 * using the minor number.
119 */
120 int
zfs_onexit_fd_hold(int fd,minor_t * minorp)121 zfs_onexit_fd_hold(int fd, minor_t *minorp)
122 {
123 file_t *fp;
124 zfs_onexit_t *zo;
125
126 #ifdef __FreeBSD__
127 file_t *tmpfp;
128 cap_rights_t rights;
129 void *data;
130 int error;
131
132 fp = getf(fd, cap_rights_init(&rights));
133 if (fp == NULL)
134 return (SET_ERROR(EBADF));
135
136 tmpfp = curthread->td_fpop;
137 curthread->td_fpop = fp;
138 error = devfs_get_cdevpriv(&data);
139 if (error == 0)
140 *minorp = (minor_t)(uintptr_t)data;
141 curthread->td_fpop = tmpfp;
142 if (error != 0)
143 return (SET_ERROR(EBADF));
144 #else
145 fp = getf(fd);
146 if (fp == NULL)
147 return (SET_ERROR(EBADF));
148
149 ASSERT(strcmp(fp->f_ops->fo_name, "zfs") == 0);
150 *minorp = minor((dev_t)(uintptr_t)fp->f_data);
151 #endif
152
153 return (zfs_onexit_minor_to_state(*minorp, &zo));
154 }
155
156 void
zfs_onexit_fd_rele(int fd)157 zfs_onexit_fd_rele(int fd)
158 {
159 releasef(fd);
160 }
161
162 /*
163 * Add a callback to be invoked when the calling process exits.
164 */
165 int
zfs_onexit_add_cb(minor_t minor,void (* func)(void *),void * data,uint64_t * action_handle)166 zfs_onexit_add_cb(minor_t minor, void (*func)(void *), void *data,
167 uint64_t *action_handle)
168 {
169 zfs_onexit_t *zo;
170 zfs_onexit_action_node_t *ap;
171 int error;
172
173 error = zfs_onexit_minor_to_state(minor, &zo);
174 if (error)
175 return (error);
176
177 ap = kmem_alloc(sizeof (zfs_onexit_action_node_t), KM_SLEEP);
178 list_link_init(&ap->za_link);
179 ap->za_func = func;
180 ap->za_data = data;
181
182 mutex_enter(&zo->zo_lock);
183 list_insert_tail(&zo->zo_actions, ap);
184 mutex_exit(&zo->zo_lock);
185 if (action_handle)
186 *action_handle = (uint64_t)(uintptr_t)ap;
187
188 return (0);
189 }
190
191 static zfs_onexit_action_node_t *
zfs_onexit_find_cb(zfs_onexit_t * zo,uint64_t action_handle)192 zfs_onexit_find_cb(zfs_onexit_t *zo, uint64_t action_handle)
193 {
194 zfs_onexit_action_node_t *match;
195 zfs_onexit_action_node_t *ap;
196 list_t *l;
197
198 ASSERT(MUTEX_HELD(&zo->zo_lock));
199
200 match = (zfs_onexit_action_node_t *)(uintptr_t)action_handle;
201 l = &zo->zo_actions;
202 for (ap = list_head(l); ap != NULL; ap = list_next(l, ap)) {
203 if (match == ap)
204 break;
205 }
206 return (ap);
207 }
208
209 /*
210 * Delete the callback, triggering it first if 'fire' is set.
211 */
212 int
zfs_onexit_del_cb(minor_t minor,uint64_t action_handle,boolean_t fire)213 zfs_onexit_del_cb(minor_t minor, uint64_t action_handle, boolean_t fire)
214 {
215 zfs_onexit_t *zo;
216 zfs_onexit_action_node_t *ap;
217 int error;
218
219 error = zfs_onexit_minor_to_state(minor, &zo);
220 if (error)
221 return (error);
222
223 mutex_enter(&zo->zo_lock);
224 ap = zfs_onexit_find_cb(zo, action_handle);
225 if (ap != NULL) {
226 list_remove(&zo->zo_actions, ap);
227 mutex_exit(&zo->zo_lock);
228 if (fire)
229 ap->za_func(ap->za_data);
230 kmem_free(ap, sizeof (zfs_onexit_action_node_t));
231 } else {
232 mutex_exit(&zo->zo_lock);
233 error = SET_ERROR(ENOENT);
234 }
235
236 return (error);
237 }
238
239 /*
240 * Return the data associated with this callback. This allows consumers
241 * of the cleanup-on-exit interfaces to stash kernel data across system
242 * calls, knowing that it will be cleaned up if the calling process exits.
243 */
244 int
zfs_onexit_cb_data(minor_t minor,uint64_t action_handle,void ** data)245 zfs_onexit_cb_data(minor_t minor, uint64_t action_handle, void **data)
246 {
247 zfs_onexit_t *zo;
248 zfs_onexit_action_node_t *ap;
249 int error;
250
251 *data = NULL;
252
253 error = zfs_onexit_minor_to_state(minor, &zo);
254 if (error)
255 return (error);
256
257 mutex_enter(&zo->zo_lock);
258 ap = zfs_onexit_find_cb(zo, action_handle);
259 if (ap != NULL)
260 *data = ap->za_data;
261 else
262 error = SET_ERROR(ENOENT);
263 mutex_exit(&zo->zo_lock);
264
265 return (error);
266 }
267