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