module/zfs/zfs_onexit.c

   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/sunddi.h>
  31 #include <sys/zfs_ioctl.h>
  32 #include <sys/zfs_onexit.h>
  33 #include <sys/zvol.h>
  34
  35 /*
  36  * ZFS kernel routines may add/delete callback routines to be invoked
  37  * upon process exit (triggered via the close operation from the /dev/zfs
  38  * driver).
  39  *
  40  * These cleanup callbacks are intended to allow for the accumulation
  41  * of kernel state across multiple ioctls.  User processes participate
  42  * simply by opening ZFS_DEV. This causes the ZFS driver to do create
  43  * some private data for the file descriptor and generating a unique
  44  * minor number. The process then passes along that file descriptor to
  45  * 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
  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
  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
 105 zfs_onexit_minor_to_state(minor_t minor, zfs_onexit_t **zo)
 106 {
 107         *zo = zfsdev_get_state(minor, ZST_ONEXIT);
 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
 121 zfs_onexit_fd_hold(int fd, minor_t *minorp)
 122 {
 123         file_t *fp;
 124         zfs_onexit_t *zo;
 125         int error;
 126
 127         fp = getf(fd);
 128         if (fp == NULL)
 129                 return (SET_ERROR(EBADF));
 130
 131         error = zfsdev_getminor(fp->f_file, minorp);
 132         if (error == 0)
 133                 error = zfs_onexit_minor_to_state(*minorp, &zo);
 134
 135         if (error)
 136                 zfs_onexit_fd_rele(fd);
 137
 138         return (error);
 139 }
 140
 141 void
 142 zfs_onexit_fd_rele(int fd)
 143 {
 144         releasef(fd);
 145 }
 146
 147 /*
 148  * Add a callback to be invoked when the calling process exits.
 149  */
 150 int
 151 zfs_onexit_add_cb(minor_t minor, void (*func)(void *), void *data,
 152     uint64_t *action_handle)
 153 {
 154         zfs_onexit_t *zo;
 155         zfs_onexit_action_node_t *ap;
 156         int error;
 157
 158         error = zfs_onexit_minor_to_state(minor, &zo);
 159         if (error)
 160                 return (error);
 161
 162         ap = kmem_alloc(sizeof (zfs_onexit_action_node_t), KM_SLEEP);
 163         list_link_init(&ap->za_link);
 164         ap->za_func = func;
 165         ap->za_data = data;
 166
 167         mutex_enter(&zo->zo_lock);
 168         list_insert_tail(&zo->zo_actions, ap);
 169         mutex_exit(&zo->zo_lock);
 170         if (action_handle)
 171                 *action_handle = (uint64_t)(uintptr_t)ap;
 172
 173         return (0);
 174 }
 175
 176 static zfs_onexit_action_node_t *
 177 zfs_onexit_find_cb(zfs_onexit_t *zo, uint64_t action_handle)
 178 {
 179         zfs_onexit_action_node_t *match;
 180         zfs_onexit_action_node_t *ap;
 181         list_t *l;
 182
 183         ASSERT(MUTEX_HELD(&zo->zo_lock));
 184
 185         match = (zfs_onexit_action_node_t *)(uintptr_t)action_handle;
 186         l = &zo->zo_actions;
 187         for (ap = list_head(l); ap != NULL; ap = list_next(l, ap)) {
 188                 if (match == ap)
 189                         break;
 190         }
 191         return (ap);
 192 }
 193
 194 /*
 195  * Delete the callback, triggering it first if 'fire' is set.
 196  */
 197 int
 198 zfs_onexit_del_cb(minor_t minor, uint64_t action_handle, boolean_t fire)
 199 {
 200         zfs_onexit_t *zo;
 201         zfs_onexit_action_node_t *ap;
 202         int error;
 203
 204         error = zfs_onexit_minor_to_state(minor, &zo);
 205         if (error)
 206                 return (error);
 207
 208         mutex_enter(&zo->zo_lock);
 209         ap = zfs_onexit_find_cb(zo, action_handle);
 210         if (ap != NULL) {
 211                 list_remove(&zo->zo_actions, ap);
 212                 mutex_exit(&zo->zo_lock);
 213                 if (fire)
 214                         ap->za_func(ap->za_data);
 215                 kmem_free(ap, sizeof (zfs_onexit_action_node_t));
 216         } else {
 217                 mutex_exit(&zo->zo_lock);
 218                 error = SET_ERROR(ENOENT);
 219         }
 220
 221         return (error);
 222 }
 223
 224 /*
 225  * Return the data associated with this callback.  This allows consumers
 226  * of the cleanup-on-exit interfaces to stash kernel data across system
 227  * calls, knowing that it will be cleaned up if the calling process exits.
 228  */
 229 int
 230 zfs_onexit_cb_data(minor_t minor, uint64_t action_handle, void **data)
 231 {
 232         zfs_onexit_t *zo;
 233         zfs_onexit_action_node_t *ap;
 234         int error;
 235
 236         *data = NULL;
 237
 238         error = zfs_onexit_minor_to_state(minor, &zo);
 239         if (error)
 240                 return (error);
 241
 242         mutex_enter(&zo->zo_lock);
 243         ap = zfs_onexit_find_cb(zo, action_handle);
 244         if (ap != NULL)
 245                 *data = ap->za_data;
 246         else
 247                 error = SET_ERROR(ENOENT);
 248         mutex_exit(&zo->zo_lock);
 249
 250         return (error);
 251 }