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 73void 74zfs_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 84void 85zfs_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 104static int 105zfs_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 */ 120int 121zfs_onexit_fd_hold(int fd, minor_t *minorp) 122{ 123 file_t *fp, *tmpfp; 124 zfs_onexit_t *zo; 125 void *data; 126 int error; 127 128 fp = getf(fd); 129 if (fp == NULL) 130 return (SET_ERROR(EBADF)); 131 132 tmpfp = curthread->td_fpop; 133 curthread->td_fpop = fp; 134 error = devfs_get_cdevpriv(&data); 135 if (error == 0) 136 *minorp = (minor_t)(uintptr_t)data; 137 curthread->td_fpop = tmpfp; 138 if (error != 0) 139 return (error); 140 return (zfs_onexit_minor_to_state(*minorp, &zo)); 141} 142 143void 144zfs_onexit_fd_rele(int fd) 145{ 146 releasef(fd); 147} 148 149/* 150 * Add a callback to be invoked when the calling process exits. 151 */ 152int 153zfs_onexit_add_cb(minor_t minor, void (*func)(void *), void *data, 154 uint64_t *action_handle) 155{ 156 zfs_onexit_t *zo; 157 zfs_onexit_action_node_t *ap; 158 int error; 159 160 error = zfs_onexit_minor_to_state(minor, &zo); 161 if (error) 162 return (error); 163 164 ap = kmem_alloc(sizeof (zfs_onexit_action_node_t), KM_SLEEP); 165 list_link_init(&ap->za_link); 166 ap->za_func = func; 167 ap->za_data = data; 168 169 mutex_enter(&zo->zo_lock); 170 list_insert_tail(&zo->zo_actions, ap); 171 mutex_exit(&zo->zo_lock); 172 if (action_handle) 173 *action_handle = (uint64_t)(uintptr_t)ap; 174 175 return (0); 176} 177 178static zfs_onexit_action_node_t * 179zfs_onexit_find_cb(zfs_onexit_t *zo, uint64_t action_handle) 180{ 181 zfs_onexit_action_node_t *match; 182 zfs_onexit_action_node_t *ap; 183 list_t *l; 184 185 ASSERT(MUTEX_HELD(&zo->zo_lock)); 186 187 match = (zfs_onexit_action_node_t *)(uintptr_t)action_handle; 188 l = &zo->zo_actions; 189 for (ap = list_head(l); ap != NULL; ap = list_next(l, ap)) { 190 if (match == ap) 191 break; 192 } 193 return (ap); 194} 195 196/* 197 * Delete the callback, triggering it first if 'fire' is set. 198 */ 199int 200zfs_onexit_del_cb(minor_t minor, uint64_t action_handle, boolean_t fire) 201{ 202 zfs_onexit_t *zo; 203 zfs_onexit_action_node_t *ap; 204 int error; 205 206 error = zfs_onexit_minor_to_state(minor, &zo); 207 if (error) 208 return (error); 209 210 mutex_enter(&zo->zo_lock); 211 ap = zfs_onexit_find_cb(zo, action_handle); 212 if (ap != NULL) { 213 list_remove(&zo->zo_actions, ap); 214 mutex_exit(&zo->zo_lock); 215 if (fire) 216 ap->za_func(ap->za_data); 217 kmem_free(ap, sizeof (zfs_onexit_action_node_t)); 218 } else { 219 mutex_exit(&zo->zo_lock); 220 error = SET_ERROR(ENOENT); 221 } 222 223 return (error); 224} 225 226/* 227 * Return the data associated with this callback. This allows consumers 228 * of the cleanup-on-exit interfaces to stash kernel data across system 229 * calls, knowing that it will be cleaned up if the calling process exits. 230 */ 231int 232zfs_onexit_cb_data(minor_t minor, uint64_t action_handle, void **data) 233{ 234 zfs_onexit_t *zo; 235 zfs_onexit_action_node_t *ap; 236 int error; 237 238 *data = NULL; 239 240 error = zfs_onexit_minor_to_state(minor, &zo); 241 if (error) 242 return (error); 243 244 mutex_enter(&zo->zo_lock); 245 ap = zfs_onexit_find_cb(zo, action_handle); 246 if (ap != NULL) 247 *data = ap->za_data; 248 else 249 error = SET_ERROR(ENOENT); 250 mutex_exit(&zo->zo_lock); 251 252 return (error); 253} 254