]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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, Version 1.0 only | |
6 | * (the "License"). You may not use this file except in compliance | |
7 | * with the License. | |
8 | * | |
9 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
10 | * or http://www.opensolaris.org/os/licensing. | |
11 | * See the License for the specific language governing permissions | |
12 | * and limitations under the License. | |
13 | * | |
14 | * When distributing Covered Code, include this CDDL HEADER in each | |
15 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
16 | * If applicable, add the following below this CDDL HEADER, with the | |
17 | * fields enclosed by brackets "[]" replaced with your own identifying | |
18 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
19 | * | |
20 | * CDDL HEADER END | |
21 | */ | |
22 | /* | |
23 | * Copyright 2008 Sun Microsystems, Inc. All rights reserved. | |
24 | * Use is subject to license terms. | |
25 | */ | |
26 | ||
27 | #include <stdio.h> | |
28 | #include <stddef.h> | |
29 | #include <stdlib.h> | |
30 | #include <string.h> | |
31 | #include <signal.h> | |
32 | #include <limits.h> | |
33 | #include <errno.h> | |
34 | #include <poll.h> | |
35 | #include <pthread.h> | |
36 | #include <unistd.h> | |
37 | #include <sys/debug.h> | |
38 | #include <sys/socket.h> | |
39 | #include <sys/stat.h> | |
40 | #include <sys/types.h> | |
41 | #include <sys/un.h> | |
42 | #include <sys/list.h> | |
43 | #include <sys/cred.h> | |
44 | ||
45 | #include <sys/dmu_ctl.h> | |
46 | #include <sys/dmu_ctl_impl.h> | |
47 | ||
48 | static dctl_sock_info_t ctl_sock = { | |
49 | .dsi_mtx = PTHREAD_MUTEX_INITIALIZER, | |
50 | .dsi_fd = -1 | |
51 | }; | |
52 | ||
53 | static int dctl_create_socket_common(); | |
54 | ||
55 | /* | |
56 | * Routines from zfs_ioctl.c | |
57 | */ | |
58 | extern int zfs_ioctl_init(); | |
59 | extern int zfs_ioctl_fini(); | |
60 | extern int zfsdev_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, | |
61 | int *rvalp); | |
62 | ||
63 | /* | |
64 | * We can't simply put the client file descriptor in wthr_info_t because we | |
65 | * have no way of accessing it from the DMU code without extensive | |
66 | * modifications. | |
67 | * | |
68 | * Therefore each worker thread will have it's own global thread-specific | |
69 | * client_fd variable. | |
70 | */ | |
71 | static __thread int client_fd = -1; | |
72 | ||
73 | int dctls_copyin(const void *src, void *dest, size_t size) | |
74 | { | |
75 | dctl_cmd_t cmd; | |
76 | ||
77 | VERIFY(client_fd >= 0); | |
78 | ||
79 | cmd.dcmd_msg = DCTL_COPYIN; | |
80 | cmd.u.dcmd_copy.ptr = (uintptr_t) src; | |
81 | cmd.u.dcmd_copy.size = size; | |
82 | ||
83 | if (dctl_send_msg(client_fd, &cmd) != 0) | |
84 | return EFAULT; | |
85 | ||
86 | if (dctl_read_data(client_fd, dest, size) != 0) | |
87 | return EFAULT; | |
88 | ||
89 | return 0; | |
90 | } | |
91 | ||
92 | int dctls_copyinstr(const char *from, char *to, size_t max, size_t *len) | |
93 | { | |
94 | dctl_cmd_t msg; | |
95 | size_t copied; | |
96 | ||
97 | VERIFY(client_fd >= 0); | |
98 | ||
99 | if (max == 0) | |
100 | return ENAMETOOLONG; | |
101 | if (max < 0) | |
102 | return EFAULT; | |
103 | ||
104 | msg.dcmd_msg = DCTL_COPYINSTR; | |
105 | msg.u.dcmd_copy.ptr = (uintptr_t) from; | |
106 | msg.u.dcmd_copy.size = max; | |
107 | ||
108 | if (dctl_send_msg(client_fd, &msg) != 0) | |
109 | return EFAULT; | |
110 | ||
111 | if (dctl_read_msg(client_fd, &msg) != 0) | |
112 | return EFAULT; | |
113 | ||
114 | if (msg.dcmd_msg != DCTL_GEN_REPLY) | |
115 | return EFAULT; | |
116 | ||
117 | copied = msg.u.dcmd_reply.size; | |
118 | ||
119 | if (copied >= max) | |
120 | return EFAULT; | |
121 | ||
122 | if (copied > 0) | |
123 | if (dctl_read_data(client_fd, to, copied) != 0) | |
124 | return EFAULT; | |
125 | ||
126 | to[copied] = '\0'; | |
127 | ||
128 | if (len != NULL) | |
129 | *len = copied + 1; | |
130 | ||
131 | return msg.u.dcmd_reply.rc; | |
132 | } | |
133 | ||
134 | int dctls_copyout(const void *src, void *dest, size_t size) | |
135 | { | |
136 | dctl_cmd_t cmd; | |
137 | ||
138 | VERIFY(client_fd >= 0); | |
139 | ||
140 | cmd.dcmd_msg = DCTL_COPYOUT; | |
141 | cmd.u.dcmd_copy.ptr = (uintptr_t) dest; | |
142 | cmd.u.dcmd_copy.size = size; | |
143 | ||
144 | if (dctl_send_msg(client_fd, &cmd) != 0) | |
145 | return EFAULT; | |
146 | ||
147 | if (dctl_send_data(client_fd, src, size) != 0) | |
148 | return EFAULT; | |
149 | ||
150 | return 0; | |
151 | } | |
152 | ||
153 | int dctls_fd_read(int fd, void *buf, ssize_t len, ssize_t *residp) | |
154 | { | |
155 | dctl_cmd_t msg; | |
156 | uint64_t dsize; | |
157 | int error; | |
158 | ||
159 | VERIFY(client_fd >= 0); | |
160 | ||
161 | msg.dcmd_msg = DCTL_FD_READ; | |
162 | msg.u.dcmd_fd_io.fd = fd; | |
163 | msg.u.dcmd_fd_io.size = len; | |
164 | ||
165 | if ((error = dctl_send_msg(client_fd, &msg)) != 0) | |
166 | return error; | |
167 | ||
168 | if ((error = dctl_read_msg(client_fd, &msg)) != 0) | |
169 | return error; | |
170 | ||
171 | if (msg.dcmd_msg != DCTL_GEN_REPLY) | |
172 | return EIO; | |
173 | ||
174 | if (msg.u.dcmd_reply.rc != 0) | |
175 | return msg.u.dcmd_reply.rc; | |
176 | ||
177 | dsize = msg.u.dcmd_reply.size; | |
178 | ||
179 | if (dsize > 0) | |
180 | error = dctl_read_data(client_fd, buf, dsize); | |
181 | ||
182 | *residp = len - dsize; | |
183 | ||
184 | return error; | |
185 | } | |
186 | ||
187 | int dctls_fd_write(int fd, const void *src, ssize_t len) | |
188 | { | |
189 | dctl_cmd_t msg; | |
190 | int error; | |
191 | ||
192 | VERIFY(client_fd >= 0); | |
193 | ||
194 | msg.dcmd_msg = DCTL_FD_WRITE; | |
195 | msg.u.dcmd_fd_io.fd = fd; | |
196 | msg.u.dcmd_fd_io.size = len; | |
197 | ||
198 | error = dctl_send_msg(client_fd, &msg); | |
199 | ||
200 | if (!error) | |
201 | error = dctl_send_data(client_fd, src, len); | |
202 | ||
203 | if (!error) | |
204 | error = dctl_read_msg(client_fd, &msg); | |
205 | ||
206 | if (error) | |
207 | return error; | |
208 | ||
209 | if (msg.dcmd_msg != DCTL_GEN_REPLY) | |
210 | return EIO; | |
211 | ||
212 | if (msg.u.dcmd_reply.rc != 0) | |
213 | return msg.u.dcmd_reply.rc; | |
214 | ||
215 | /* | |
216 | * We have to do this because the original upstream code | |
217 | * does not check if residp == len. | |
218 | */ | |
219 | if (msg.u.dcmd_reply.size != len) | |
220 | return EIO; | |
221 | ||
222 | return 0; | |
223 | } | |
224 | ||
225 | /* Handle a new connection */ | |
226 | static void dctl_handle_conn(int sock_fd) | |
227 | { | |
228 | dctl_cmd_t cmd; | |
229 | dev_t dev = { 0 }; | |
230 | int rc; | |
231 | ||
232 | client_fd = sock_fd; | |
233 | ||
234 | while (dctl_read_msg(sock_fd, &cmd) == 0) { | |
235 | if (cmd.dcmd_msg != DCTL_IOCTL) { | |
236 | fprintf(stderr, "%s(): unexpected message type.\n", | |
237 | __func__); | |
238 | break; | |
239 | } | |
240 | ||
241 | rc = zfsdev_ioctl(dev, cmd.u.dcmd_ioctl.cmd, | |
242 | (intptr_t) cmd.u.dcmd_ioctl.arg, 0, NULL, NULL); | |
243 | ||
244 | cmd.dcmd_msg = DCTL_IOCTL_REPLY; | |
245 | cmd.u.dcmd_reply.rc = rc; | |
246 | ||
247 | if (dctl_send_msg(sock_fd, &cmd) != 0) | |
248 | break; | |
249 | } | |
250 | close(sock_fd); | |
251 | ||
252 | client_fd = -1; | |
253 | } | |
254 | ||
255 | /* Main worker thread loop */ | |
256 | static void *dctl_thread(void *arg) | |
257 | { | |
258 | wthr_info_t *thr = arg; | |
259 | struct pollfd fds[1]; | |
260 | ||
261 | fds[0].events = POLLIN; | |
262 | ||
263 | pthread_mutex_lock(&ctl_sock.dsi_mtx); | |
264 | ||
265 | while (!thr->wthr_exit) { | |
266 | /* Clean-up dead threads */ | |
267 | dctl_thr_join(); | |
268 | ||
269 | /* The file descriptor might change in the thread lifetime */ | |
270 | fds[0].fd = ctl_sock.dsi_fd; | |
271 | ||
272 | /* Poll socket with 1-second timeout */ | |
273 | int rc = poll(fds, 1, 1000); | |
274 | if (rc == 0 || (rc == -1 && errno == EINTR)) | |
275 | continue; | |
276 | ||
277 | /* Recheck the exit flag */ | |
278 | if (thr->wthr_exit) | |
279 | break; | |
280 | ||
281 | if (rc == -1) { | |
282 | /* Unknown error, let's try to recreate the socket */ | |
283 | close(ctl_sock.dsi_fd); | |
284 | ctl_sock.dsi_fd = -1; | |
285 | ||
286 | if (dctl_create_socket_common() != 0) | |
287 | break; | |
288 | ||
289 | continue; | |
290 | } | |
291 | ASSERT(rc == 1); | |
292 | ||
293 | short rev = fds[0].revents; | |
294 | if (rev == 0) | |
295 | continue; | |
296 | ASSERT(rev == POLLIN); | |
297 | ||
298 | /* | |
299 | * At this point there should be a connection ready to be | |
300 | * accepted. | |
301 | */ | |
302 | int client_fd = accept(ctl_sock.dsi_fd, NULL, NULL); | |
303 | /* Many possible errors here, we'll just retry */ | |
304 | if (client_fd == -1) | |
305 | continue; | |
306 | ||
307 | /* | |
308 | * Now lets handle the request. This can take a very | |
309 | * long time (hours even), so we'll let other threads | |
310 | * handle new connections. | |
311 | */ | |
312 | pthread_mutex_unlock(&ctl_sock.dsi_mtx); | |
313 | ||
314 | dctl_thr_rebalance(thr, B_FALSE); | |
315 | dctl_handle_conn(client_fd); | |
316 | dctl_thr_rebalance(thr, B_TRUE); | |
317 | ||
318 | pthread_mutex_lock(&ctl_sock.dsi_mtx); | |
319 | } | |
320 | pthread_mutex_unlock(&ctl_sock.dsi_mtx); | |
321 | ||
322 | dctl_thr_die(thr); | |
323 | ||
324 | return NULL; | |
325 | } | |
326 | ||
327 | static int dctl_create_socket_common() | |
328 | { | |
329 | dctl_sock_info_t *s = &ctl_sock; | |
330 | size_t size; | |
331 | int error; | |
332 | ||
333 | ASSERT(s->dsi_fd == -1); | |
334 | ||
335 | /* | |
336 | * Unlink old socket, in case it exists. | |
337 | * We don't care about errors here. | |
338 | */ | |
339 | unlink(s->dsi_path); | |
340 | ||
341 | /* Create the socket */ | |
342 | s->dsi_fd = socket(PF_UNIX, SOCK_STREAM, 0); | |
343 | if (s->dsi_fd == -1) { | |
344 | error = errno; | |
345 | perror("socket"); | |
346 | return error; | |
347 | } | |
348 | ||
349 | s->dsi_addr.sun_family = AF_UNIX; | |
350 | ||
351 | size = sizeof(s->dsi_addr.sun_path) - 1; | |
352 | strncpy(s->dsi_addr.sun_path, s->dsi_path, size); | |
353 | ||
354 | s->dsi_addr.sun_path[size] = '\0'; | |
355 | ||
356 | if (bind(s->dsi_fd, (struct sockaddr *) &s->dsi_addr, | |
357 | sizeof(s->dsi_addr)) != 0) { | |
358 | error = errno; | |
359 | perror("bind"); | |
360 | return error; | |
361 | } | |
362 | ||
363 | if (listen(s->dsi_fd, LISTEN_BACKLOG) != 0) { | |
364 | error = errno; | |
365 | perror("listen"); | |
366 | unlink(s->dsi_path); | |
367 | return error; | |
368 | } | |
369 | ||
370 | return 0; | |
371 | } | |
372 | ||
373 | static int dctl_create_socket(const char *cfg_dir) | |
374 | { | |
375 | int error; | |
376 | dctl_sock_info_t *s = &ctl_sock; | |
377 | ||
378 | ASSERT(s->dsi_path == NULL); | |
379 | ASSERT(s->dsi_fd == -1); | |
380 | ||
381 | int pathsize = strlen(cfg_dir) + strlen(SOCKNAME) + 2; | |
382 | if (pathsize > sizeof(s->dsi_addr.sun_path)) | |
383 | return ENAMETOOLONG; | |
384 | ||
385 | s->dsi_path = malloc(pathsize); | |
386 | if (s->dsi_path == NULL) | |
387 | return ENOMEM; | |
388 | ||
389 | strcpy(s->dsi_path, cfg_dir); | |
390 | strcat(s->dsi_path, "/" SOCKNAME); | |
391 | ||
392 | /* | |
393 | * For convenience, create the directory in case it doesn't exist. | |
394 | * We don't care about errors here. | |
395 | */ | |
396 | mkdir(cfg_dir, 0770); | |
397 | ||
398 | error = dctl_create_socket_common(); | |
399 | ||
400 | if (error) { | |
401 | free(s->dsi_path); | |
402 | ||
403 | if (s->dsi_fd != -1) { | |
404 | close(s->dsi_fd); | |
405 | s->dsi_fd = -1; | |
406 | } | |
407 | } | |
408 | ||
409 | return error; | |
410 | } | |
411 | ||
412 | static void dctl_destroy_socket() | |
413 | { | |
414 | dctl_sock_info_t *s = &ctl_sock; | |
415 | ||
416 | ASSERT(s->dsi_path != NULL); | |
417 | ASSERT(s->dsi_fd != -1); | |
418 | ||
419 | close(s->dsi_fd); | |
420 | s->dsi_fd = -1; | |
421 | ||
422 | unlink(s->dsi_path); | |
423 | free(s->dsi_path); | |
424 | } | |
425 | ||
426 | /* | |
427 | * Initialize the DMU userspace control interface. | |
428 | * This should be called after kernel_init(). | |
429 | * | |
430 | * Note that only very rarely we have more than a couple of simultaneous | |
431 | * lzfs/lzpool connections. Since the thread pool grows automatically when all | |
432 | * threads are busy, a good value for min_thr and max_free_thr is 2. | |
433 | */ | |
434 | int dctl_server_init(const char *cfg_dir, int min_thr, int max_free_thr) | |
435 | { | |
436 | int error; | |
437 | ||
438 | ASSERT(min_thr > 0); | |
439 | ASSERT(max_free_thr >= min_thr); | |
440 | ||
441 | error = zfs_ioctl_init(); | |
442 | if (error) | |
443 | return error; | |
444 | ||
445 | error = dctl_create_socket(cfg_dir); | |
446 | if (error) { | |
447 | (void) zfs_ioctl_fini(); | |
448 | return error; | |
449 | } | |
450 | ||
451 | error = dctl_thr_pool_create(min_thr, max_free_thr, dctl_thread); | |
452 | if (error) { | |
453 | (void) zfs_ioctl_fini(); | |
454 | dctl_destroy_socket(); | |
455 | return error; | |
456 | } | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | /* | |
462 | * Terminate control interface. | |
463 | * This should be called after closing all objsets, but before calling | |
464 | * kernel_fini(). | |
465 | * May return EBUSY if the SPA is busy. | |
466 | * | |
467 | * Thread pool destruction can take a while due to poll() | |
468 | * timeout or due to a thread being busy (e.g. a backup is being taken). | |
469 | */ | |
470 | int dctl_server_fini() | |
471 | { | |
472 | dctl_thr_pool_stop(); | |
473 | dctl_destroy_socket(); | |
474 | ||
475 | return zfs_ioctl_fini(); | |
476 | } |