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