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