Commit | Line | Data |
---|---|---|
602adf40 YS |
1 | /* |
2 | rbd.c -- Export ceph rados objects as a Linux block device | |
3 | ||
4 | ||
5 | based on drivers/block/osdblk.c: | |
6 | ||
7 | Copyright 2009 Red Hat, Inc. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; see the file COPYING. If not, write to | |
20 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | ||
22 | ||
23 | ||
dfc5606d | 24 | For usage instructions, please refer to: |
602adf40 | 25 | |
dfc5606d | 26 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
27 | |
28 | */ | |
29 | ||
30 | #include <linux/ceph/libceph.h> | |
31 | #include <linux/ceph/osd_client.h> | |
32 | #include <linux/ceph/mon_client.h> | |
33 | #include <linux/ceph/decode.h> | |
59c2be1e | 34 | #include <linux/parser.h> |
602adf40 YS |
35 | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/device.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/blkdev.h> | |
41 | ||
42 | #include "rbd_types.h" | |
43 | ||
aafb230e AE |
44 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
45 | ||
593a9e7b AE |
46 | /* |
47 | * The basic unit of block I/O is a sector. It is interpreted in a | |
48 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
49 | * universally 512 bytes. These symbols are just slightly more | |
50 | * meaningful than the bare numbers they represent. | |
51 | */ | |
52 | #define SECTOR_SHIFT 9 | |
53 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
54 | ||
df111be6 AE |
55 | /* It might be useful to have this defined elsewhere too */ |
56 | ||
57 | #define U64_MAX ((u64) (~0ULL)) | |
58 | ||
f0f8cef5 AE |
59 | #define RBD_DRV_NAME "rbd" |
60 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
61 | |
62 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
63 | ||
d4b125e9 AE |
64 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
65 | #define RBD_MAX_SNAP_NAME_LEN \ | |
66 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
67 | ||
35d489f9 | 68 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
69 | #define RBD_MAX_OPT_LEN 1024 |
70 | ||
71 | #define RBD_SNAP_HEAD_NAME "-" | |
72 | ||
1e130199 AE |
73 | #define RBD_IMAGE_ID_LEN_MAX 64 |
74 | #define RBD_OBJ_PREFIX_LEN_MAX 64 | |
589d30e0 | 75 | |
d889140c AE |
76 | /* Feature bits */ |
77 | ||
78 | #define RBD_FEATURE_LAYERING 1 | |
79 | ||
80 | /* Features supported by this (client software) implementation. */ | |
81 | ||
82 | #define RBD_FEATURES_ALL (0) | |
83 | ||
81a89793 AE |
84 | /* |
85 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
86 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
87 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
88 | * enough to hold all possible device names. | |
89 | */ | |
602adf40 | 90 | #define DEV_NAME_LEN 32 |
81a89793 | 91 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 | 92 | |
cc0538b6 | 93 | #define RBD_READ_ONLY_DEFAULT false |
59c2be1e | 94 | |
602adf40 YS |
95 | /* |
96 | * block device image metadata (in-memory version) | |
97 | */ | |
98 | struct rbd_image_header { | |
f84344f3 | 99 | /* These four fields never change for a given rbd image */ |
849b4260 | 100 | char *object_prefix; |
34b13184 | 101 | u64 features; |
602adf40 YS |
102 | __u8 obj_order; |
103 | __u8 crypt_type; | |
104 | __u8 comp_type; | |
602adf40 | 105 | |
f84344f3 AE |
106 | /* The remaining fields need to be updated occasionally */ |
107 | u64 image_size; | |
108 | struct ceph_snap_context *snapc; | |
602adf40 YS |
109 | char *snap_names; |
110 | u64 *snap_sizes; | |
59c2be1e YS |
111 | |
112 | u64 obj_version; | |
113 | }; | |
114 | ||
115 | struct rbd_options { | |
cc0538b6 | 116 | bool read_only; |
602adf40 YS |
117 | }; |
118 | ||
119 | /* | |
f0f8cef5 | 120 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
121 | */ |
122 | struct rbd_client { | |
123 | struct ceph_client *client; | |
124 | struct kref kref; | |
125 | struct list_head node; | |
126 | }; | |
127 | ||
128 | /* | |
f0f8cef5 | 129 | * a request completion status |
602adf40 | 130 | */ |
1fec7093 YS |
131 | struct rbd_req_status { |
132 | int done; | |
133 | int rc; | |
134 | u64 bytes; | |
135 | }; | |
136 | ||
137 | /* | |
138 | * a collection of requests | |
139 | */ | |
140 | struct rbd_req_coll { | |
141 | int total; | |
142 | int num_done; | |
143 | struct kref kref; | |
144 | struct rbd_req_status status[0]; | |
602adf40 YS |
145 | }; |
146 | ||
f0f8cef5 AE |
147 | /* |
148 | * a single io request | |
149 | */ | |
150 | struct rbd_request { | |
151 | struct request *rq; /* blk layer request */ | |
152 | struct bio *bio; /* cloned bio */ | |
153 | struct page **pages; /* list of used pages */ | |
154 | u64 len; | |
155 | int coll_index; | |
156 | struct rbd_req_coll *coll; | |
157 | }; | |
158 | ||
dfc5606d YS |
159 | struct rbd_snap { |
160 | struct device dev; | |
161 | const char *name; | |
3591538f | 162 | u64 size; |
dfc5606d YS |
163 | struct list_head node; |
164 | u64 id; | |
34b13184 | 165 | u64 features; |
dfc5606d YS |
166 | }; |
167 | ||
f84344f3 AE |
168 | struct rbd_mapping { |
169 | char *snap_name; | |
170 | u64 snap_id; | |
99c1f08f | 171 | u64 size; |
34b13184 | 172 | u64 features; |
f84344f3 AE |
173 | bool snap_exists; |
174 | bool read_only; | |
175 | }; | |
176 | ||
602adf40 YS |
177 | /* |
178 | * a single device | |
179 | */ | |
180 | struct rbd_device { | |
de71a297 | 181 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
182 | |
183 | int major; /* blkdev assigned major */ | |
184 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 185 | |
a30b71b9 | 186 | u32 image_format; /* Either 1 or 2 */ |
f8c38929 | 187 | struct rbd_options rbd_opts; |
602adf40 YS |
188 | struct rbd_client *rbd_client; |
189 | ||
190 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
191 | ||
192 | spinlock_t lock; /* queue lock */ | |
193 | ||
194 | struct rbd_image_header header; | |
589d30e0 AE |
195 | char *image_id; |
196 | size_t image_id_len; | |
0bed54dc AE |
197 | char *image_name; |
198 | size_t image_name_len; | |
199 | char *header_name; | |
d22f76e7 | 200 | char *pool_name; |
9bb2f334 | 201 | int pool_id; |
602adf40 | 202 | |
59c2be1e YS |
203 | struct ceph_osd_event *watch_event; |
204 | struct ceph_osd_request *watch_request; | |
205 | ||
c666601a JD |
206 | /* protects updating the header */ |
207 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
208 | |
209 | struct rbd_mapping mapping; | |
602adf40 YS |
210 | |
211 | struct list_head node; | |
dfc5606d YS |
212 | |
213 | /* list of snapshots */ | |
214 | struct list_head snaps; | |
215 | ||
216 | /* sysfs related */ | |
217 | struct device dev; | |
218 | }; | |
219 | ||
602adf40 | 220 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 221 | |
602adf40 | 222 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
223 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
224 | ||
432b8587 AE |
225 | static LIST_HEAD(rbd_client_list); /* clients */ |
226 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 227 | |
304f6808 AE |
228 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
229 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev); | |
230 | ||
dfc5606d | 231 | static void rbd_dev_release(struct device *dev); |
14e7085d | 232 | static void __rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 233 | |
f0f8cef5 AE |
234 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
235 | size_t count); | |
236 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
237 | size_t count); | |
238 | ||
239 | static struct bus_attribute rbd_bus_attrs[] = { | |
240 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
241 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
242 | __ATTR_NULL | |
243 | }; | |
244 | ||
245 | static struct bus_type rbd_bus_type = { | |
246 | .name = "rbd", | |
247 | .bus_attrs = rbd_bus_attrs, | |
248 | }; | |
249 | ||
250 | static void rbd_root_dev_release(struct device *dev) | |
251 | { | |
252 | } | |
253 | ||
254 | static struct device rbd_root_dev = { | |
255 | .init_name = "rbd", | |
256 | .release = rbd_root_dev_release, | |
257 | }; | |
258 | ||
aafb230e AE |
259 | #ifdef RBD_DEBUG |
260 | #define rbd_assert(expr) \ | |
261 | if (unlikely(!(expr))) { \ | |
262 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
263 | "at line %d:\n\n" \ | |
264 | "\trbd_assert(%s);\n\n", \ | |
265 | __func__, __LINE__, #expr); \ | |
266 | BUG(); \ | |
267 | } | |
268 | #else /* !RBD_DEBUG */ | |
269 | # define rbd_assert(expr) ((void) 0) | |
270 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 271 | |
dfc5606d YS |
272 | static struct device *rbd_get_dev(struct rbd_device *rbd_dev) |
273 | { | |
274 | return get_device(&rbd_dev->dev); | |
275 | } | |
276 | ||
277 | static void rbd_put_dev(struct rbd_device *rbd_dev) | |
278 | { | |
279 | put_device(&rbd_dev->dev); | |
280 | } | |
602adf40 | 281 | |
117973fb AE |
282 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
283 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 284 | |
602adf40 YS |
285 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
286 | { | |
f0f8cef5 | 287 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
602adf40 | 288 | |
f84344f3 | 289 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
290 | return -EROFS; |
291 | ||
340c7a2b | 292 | rbd_get_dev(rbd_dev); |
f84344f3 | 293 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
340c7a2b | 294 | |
602adf40 YS |
295 | return 0; |
296 | } | |
297 | ||
dfc5606d YS |
298 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
299 | { | |
300 | struct rbd_device *rbd_dev = disk->private_data; | |
301 | ||
302 | rbd_put_dev(rbd_dev); | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
602adf40 YS |
307 | static const struct block_device_operations rbd_bd_ops = { |
308 | .owner = THIS_MODULE, | |
309 | .open = rbd_open, | |
dfc5606d | 310 | .release = rbd_release, |
602adf40 YS |
311 | }; |
312 | ||
313 | /* | |
314 | * Initialize an rbd client instance. | |
43ae4701 | 315 | * We own *ceph_opts. |
602adf40 | 316 | */ |
f8c38929 | 317 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
318 | { |
319 | struct rbd_client *rbdc; | |
320 | int ret = -ENOMEM; | |
321 | ||
322 | dout("rbd_client_create\n"); | |
323 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); | |
324 | if (!rbdc) | |
325 | goto out_opt; | |
326 | ||
327 | kref_init(&rbdc->kref); | |
328 | INIT_LIST_HEAD(&rbdc->node); | |
329 | ||
bc534d86 AE |
330 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
331 | ||
43ae4701 | 332 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 333 | if (IS_ERR(rbdc->client)) |
bc534d86 | 334 | goto out_mutex; |
43ae4701 | 335 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
336 | |
337 | ret = ceph_open_session(rbdc->client); | |
338 | if (ret < 0) | |
339 | goto out_err; | |
340 | ||
432b8587 | 341 | spin_lock(&rbd_client_list_lock); |
602adf40 | 342 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 343 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 344 | |
bc534d86 AE |
345 | mutex_unlock(&ctl_mutex); |
346 | ||
602adf40 YS |
347 | dout("rbd_client_create created %p\n", rbdc); |
348 | return rbdc; | |
349 | ||
350 | out_err: | |
351 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
352 | out_mutex: |
353 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
354 | kfree(rbdc); |
355 | out_opt: | |
43ae4701 AE |
356 | if (ceph_opts) |
357 | ceph_destroy_options(ceph_opts); | |
28f259b7 | 358 | return ERR_PTR(ret); |
602adf40 YS |
359 | } |
360 | ||
361 | /* | |
1f7ba331 AE |
362 | * Find a ceph client with specific addr and configuration. If |
363 | * found, bump its reference count. | |
602adf40 | 364 | */ |
1f7ba331 | 365 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
366 | { |
367 | struct rbd_client *client_node; | |
1f7ba331 | 368 | bool found = false; |
602adf40 | 369 | |
43ae4701 | 370 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
371 | return NULL; |
372 | ||
1f7ba331 AE |
373 | spin_lock(&rbd_client_list_lock); |
374 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
375 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
376 | kref_get(&client_node->kref); | |
377 | found = true; | |
378 | break; | |
379 | } | |
380 | } | |
381 | spin_unlock(&rbd_client_list_lock); | |
382 | ||
383 | return found ? client_node : NULL; | |
602adf40 YS |
384 | } |
385 | ||
59c2be1e YS |
386 | /* |
387 | * mount options | |
388 | */ | |
389 | enum { | |
59c2be1e YS |
390 | Opt_last_int, |
391 | /* int args above */ | |
392 | Opt_last_string, | |
393 | /* string args above */ | |
cc0538b6 AE |
394 | Opt_read_only, |
395 | Opt_read_write, | |
396 | /* Boolean args above */ | |
397 | Opt_last_bool, | |
59c2be1e YS |
398 | }; |
399 | ||
43ae4701 | 400 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
401 | /* int args above */ |
402 | /* string args above */ | |
be466c1c | 403 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
404 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
405 | {Opt_read_write, "read_write"}, | |
406 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
407 | /* Boolean args above */ | |
59c2be1e YS |
408 | {-1, NULL} |
409 | }; | |
410 | ||
411 | static int parse_rbd_opts_token(char *c, void *private) | |
412 | { | |
43ae4701 | 413 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
414 | substring_t argstr[MAX_OPT_ARGS]; |
415 | int token, intval, ret; | |
416 | ||
43ae4701 | 417 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
418 | if (token < 0) |
419 | return -EINVAL; | |
420 | ||
421 | if (token < Opt_last_int) { | |
422 | ret = match_int(&argstr[0], &intval); | |
423 | if (ret < 0) { | |
424 | pr_err("bad mount option arg (not int) " | |
425 | "at '%s'\n", c); | |
426 | return ret; | |
427 | } | |
428 | dout("got int token %d val %d\n", token, intval); | |
429 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
430 | dout("got string token %d val %s\n", token, | |
431 | argstr[0].from); | |
cc0538b6 AE |
432 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
433 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
434 | } else { |
435 | dout("got token %d\n", token); | |
436 | } | |
437 | ||
438 | switch (token) { | |
cc0538b6 AE |
439 | case Opt_read_only: |
440 | rbd_opts->read_only = true; | |
441 | break; | |
442 | case Opt_read_write: | |
443 | rbd_opts->read_only = false; | |
444 | break; | |
59c2be1e | 445 | default: |
aafb230e AE |
446 | rbd_assert(false); |
447 | break; | |
59c2be1e YS |
448 | } |
449 | return 0; | |
450 | } | |
451 | ||
602adf40 YS |
452 | /* |
453 | * Get a ceph client with specific addr and configuration, if one does | |
454 | * not exist create it. | |
455 | */ | |
f8c38929 AE |
456 | static int rbd_get_client(struct rbd_device *rbd_dev, const char *mon_addr, |
457 | size_t mon_addr_len, char *options) | |
602adf40 | 458 | { |
f8c38929 | 459 | struct rbd_options *rbd_opts = &rbd_dev->rbd_opts; |
43ae4701 | 460 | struct ceph_options *ceph_opts; |
f8c38929 | 461 | struct rbd_client *rbdc; |
59c2be1e | 462 | |
cc0538b6 | 463 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; |
602adf40 | 464 | |
43ae4701 AE |
465 | ceph_opts = ceph_parse_options(options, mon_addr, |
466 | mon_addr + mon_addr_len, | |
467 | parse_rbd_opts_token, rbd_opts); | |
f8c38929 AE |
468 | if (IS_ERR(ceph_opts)) |
469 | return PTR_ERR(ceph_opts); | |
602adf40 | 470 | |
1f7ba331 | 471 | rbdc = rbd_client_find(ceph_opts); |
602adf40 | 472 | if (rbdc) { |
602adf40 | 473 | /* using an existing client */ |
43ae4701 | 474 | ceph_destroy_options(ceph_opts); |
f8c38929 AE |
475 | } else { |
476 | rbdc = rbd_client_create(ceph_opts); | |
477 | if (IS_ERR(rbdc)) | |
478 | return PTR_ERR(rbdc); | |
602adf40 | 479 | } |
f8c38929 | 480 | rbd_dev->rbd_client = rbdc; |
602adf40 | 481 | |
f8c38929 | 482 | return 0; |
602adf40 YS |
483 | } |
484 | ||
485 | /* | |
486 | * Destroy ceph client | |
d23a4b3f | 487 | * |
432b8587 | 488 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
489 | */ |
490 | static void rbd_client_release(struct kref *kref) | |
491 | { | |
492 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
493 | ||
494 | dout("rbd_release_client %p\n", rbdc); | |
cd9d9f5d | 495 | spin_lock(&rbd_client_list_lock); |
602adf40 | 496 | list_del(&rbdc->node); |
cd9d9f5d | 497 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
498 | |
499 | ceph_destroy_client(rbdc->client); | |
500 | kfree(rbdc); | |
501 | } | |
502 | ||
503 | /* | |
504 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
505 | * it. | |
506 | */ | |
507 | static void rbd_put_client(struct rbd_device *rbd_dev) | |
508 | { | |
509 | kref_put(&rbd_dev->rbd_client->kref, rbd_client_release); | |
510 | rbd_dev->rbd_client = NULL; | |
602adf40 YS |
511 | } |
512 | ||
1fec7093 YS |
513 | /* |
514 | * Destroy requests collection | |
515 | */ | |
516 | static void rbd_coll_release(struct kref *kref) | |
517 | { | |
518 | struct rbd_req_coll *coll = | |
519 | container_of(kref, struct rbd_req_coll, kref); | |
520 | ||
521 | dout("rbd_coll_release %p\n", coll); | |
522 | kfree(coll); | |
523 | } | |
602adf40 | 524 | |
a30b71b9 AE |
525 | static bool rbd_image_format_valid(u32 image_format) |
526 | { | |
527 | return image_format == 1 || image_format == 2; | |
528 | } | |
529 | ||
8e94af8e AE |
530 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
531 | { | |
103a150f AE |
532 | size_t size; |
533 | u32 snap_count; | |
534 | ||
535 | /* The header has to start with the magic rbd header text */ | |
536 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
537 | return false; | |
538 | ||
db2388b6 AE |
539 | /* The bio layer requires at least sector-sized I/O */ |
540 | ||
541 | if (ondisk->options.order < SECTOR_SHIFT) | |
542 | return false; | |
543 | ||
544 | /* If we use u64 in a few spots we may be able to loosen this */ | |
545 | ||
546 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
547 | return false; | |
548 | ||
103a150f AE |
549 | /* |
550 | * The size of a snapshot header has to fit in a size_t, and | |
551 | * that limits the number of snapshots. | |
552 | */ | |
553 | snap_count = le32_to_cpu(ondisk->snap_count); | |
554 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
555 | if (snap_count > size / sizeof (__le64)) | |
556 | return false; | |
557 | ||
558 | /* | |
559 | * Not only that, but the size of the entire the snapshot | |
560 | * header must also be representable in a size_t. | |
561 | */ | |
562 | size -= snap_count * sizeof (__le64); | |
563 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
564 | return false; | |
565 | ||
566 | return true; | |
8e94af8e AE |
567 | } |
568 | ||
602adf40 YS |
569 | /* |
570 | * Create a new header structure, translate header format from the on-disk | |
571 | * header. | |
572 | */ | |
573 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 574 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 575 | { |
ccece235 | 576 | u32 snap_count; |
58c17b0e | 577 | size_t len; |
d2bb24e5 | 578 | size_t size; |
621901d6 | 579 | u32 i; |
602adf40 | 580 | |
6a52325f AE |
581 | memset(header, 0, sizeof (*header)); |
582 | ||
103a150f AE |
583 | snap_count = le32_to_cpu(ondisk->snap_count); |
584 | ||
58c17b0e AE |
585 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
586 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 587 | if (!header->object_prefix) |
602adf40 | 588 | return -ENOMEM; |
58c17b0e AE |
589 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
590 | header->object_prefix[len] = '\0'; | |
00f1f36f | 591 | |
602adf40 | 592 | if (snap_count) { |
f785cc1d AE |
593 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
594 | ||
621901d6 AE |
595 | /* Save a copy of the snapshot names */ |
596 | ||
f785cc1d AE |
597 | if (snap_names_len > (u64) SIZE_MAX) |
598 | return -EIO; | |
599 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 600 | if (!header->snap_names) |
6a52325f | 601 | goto out_err; |
f785cc1d AE |
602 | /* |
603 | * Note that rbd_dev_v1_header_read() guarantees | |
604 | * the ondisk buffer we're working with has | |
605 | * snap_names_len bytes beyond the end of the | |
606 | * snapshot id array, this memcpy() is safe. | |
607 | */ | |
608 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
609 | snap_names_len); | |
6a52325f | 610 | |
621901d6 AE |
611 | /* Record each snapshot's size */ |
612 | ||
d2bb24e5 AE |
613 | size = snap_count * sizeof (*header->snap_sizes); |
614 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 615 | if (!header->snap_sizes) |
6a52325f | 616 | goto out_err; |
621901d6 AE |
617 | for (i = 0; i < snap_count; i++) |
618 | header->snap_sizes[i] = | |
619 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 620 | } else { |
ccece235 | 621 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
622 | header->snap_names = NULL; |
623 | header->snap_sizes = NULL; | |
624 | } | |
849b4260 | 625 | |
34b13184 | 626 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
627 | header->obj_order = ondisk->options.order; |
628 | header->crypt_type = ondisk->options.crypt_type; | |
629 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 630 | |
621901d6 AE |
631 | /* Allocate and fill in the snapshot context */ |
632 | ||
f84344f3 | 633 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
634 | size = sizeof (struct ceph_snap_context); |
635 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
636 | header->snapc = kzalloc(size, GFP_KERNEL); | |
637 | if (!header->snapc) | |
638 | goto out_err; | |
602adf40 YS |
639 | |
640 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 641 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 642 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
643 | for (i = 0; i < snap_count; i++) |
644 | header->snapc->snaps[i] = | |
645 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
646 | |
647 | return 0; | |
648 | ||
6a52325f | 649 | out_err: |
849b4260 | 650 | kfree(header->snap_sizes); |
ccece235 | 651 | header->snap_sizes = NULL; |
602adf40 | 652 | kfree(header->snap_names); |
ccece235 | 653 | header->snap_names = NULL; |
6a52325f AE |
654 | kfree(header->object_prefix); |
655 | header->object_prefix = NULL; | |
ccece235 | 656 | |
00f1f36f | 657 | return -ENOMEM; |
602adf40 YS |
658 | } |
659 | ||
8836b995 | 660 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 | 661 | { |
602adf40 | 662 | |
e86924a8 | 663 | struct rbd_snap *snap; |
602adf40 | 664 | |
e86924a8 AE |
665 | list_for_each_entry(snap, &rbd_dev->snaps, node) { |
666 | if (!strcmp(snap_name, snap->name)) { | |
667 | rbd_dev->mapping.snap_id = snap->id; | |
668 | rbd_dev->mapping.size = snap->size; | |
34b13184 | 669 | rbd_dev->mapping.features = snap->features; |
602adf40 | 670 | |
e86924a8 | 671 | return 0; |
00f1f36f | 672 | } |
00f1f36f | 673 | } |
e86924a8 | 674 | |
00f1f36f | 675 | return -ENOENT; |
602adf40 YS |
676 | } |
677 | ||
5ed16177 | 678 | static int rbd_dev_set_mapping(struct rbd_device *rbd_dev, char *snap_name) |
602adf40 | 679 | { |
78dc447d | 680 | int ret; |
602adf40 | 681 | |
4e1105a2 | 682 | if (!memcmp(snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 683 | sizeof (RBD_SNAP_HEAD_NAME))) { |
f84344f3 | 684 | rbd_dev->mapping.snap_id = CEPH_NOSNAP; |
99c1f08f | 685 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 686 | rbd_dev->mapping.features = rbd_dev->header.features; |
f84344f3 AE |
687 | rbd_dev->mapping.snap_exists = false; |
688 | rbd_dev->mapping.read_only = rbd_dev->rbd_opts.read_only; | |
e86924a8 | 689 | ret = 0; |
602adf40 | 690 | } else { |
8836b995 | 691 | ret = snap_by_name(rbd_dev, snap_name); |
602adf40 YS |
692 | if (ret < 0) |
693 | goto done; | |
f84344f3 AE |
694 | rbd_dev->mapping.snap_exists = true; |
695 | rbd_dev->mapping.read_only = true; | |
602adf40 | 696 | } |
4e1105a2 | 697 | rbd_dev->mapping.snap_name = snap_name; |
602adf40 | 698 | done: |
602adf40 YS |
699 | return ret; |
700 | } | |
701 | ||
702 | static void rbd_header_free(struct rbd_image_header *header) | |
703 | { | |
849b4260 | 704 | kfree(header->object_prefix); |
d78fd7ae | 705 | header->object_prefix = NULL; |
602adf40 | 706 | kfree(header->snap_sizes); |
d78fd7ae | 707 | header->snap_sizes = NULL; |
849b4260 | 708 | kfree(header->snap_names); |
d78fd7ae | 709 | header->snap_names = NULL; |
d1d25646 | 710 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 711 | header->snapc = NULL; |
602adf40 YS |
712 | } |
713 | ||
65ccfe21 | 714 | static char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 715 | { |
65ccfe21 AE |
716 | char *name; |
717 | u64 segment; | |
718 | int ret; | |
602adf40 | 719 | |
65ccfe21 AE |
720 | name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO); |
721 | if (!name) | |
722 | return NULL; | |
723 | segment = offset >> rbd_dev->header.obj_order; | |
724 | ret = snprintf(name, RBD_MAX_SEG_NAME_LEN, "%s.%012llx", | |
725 | rbd_dev->header.object_prefix, segment); | |
726 | if (ret < 0 || ret >= RBD_MAX_SEG_NAME_LEN) { | |
727 | pr_err("error formatting segment name for #%llu (%d)\n", | |
728 | segment, ret); | |
729 | kfree(name); | |
730 | name = NULL; | |
731 | } | |
602adf40 | 732 | |
65ccfe21 AE |
733 | return name; |
734 | } | |
602adf40 | 735 | |
65ccfe21 AE |
736 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
737 | { | |
738 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 739 | |
65ccfe21 AE |
740 | return offset & (segment_size - 1); |
741 | } | |
742 | ||
743 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
744 | u64 offset, u64 length) | |
745 | { | |
746 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
747 | ||
748 | offset &= segment_size - 1; | |
749 | ||
aafb230e | 750 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
751 | if (offset + length > segment_size) |
752 | length = segment_size - offset; | |
753 | ||
754 | return length; | |
602adf40 YS |
755 | } |
756 | ||
1fec7093 YS |
757 | static int rbd_get_num_segments(struct rbd_image_header *header, |
758 | u64 ofs, u64 len) | |
759 | { | |
df111be6 AE |
760 | u64 start_seg; |
761 | u64 end_seg; | |
762 | ||
763 | if (!len) | |
764 | return 0; | |
765 | if (len - 1 > U64_MAX - ofs) | |
766 | return -ERANGE; | |
767 | ||
768 | start_seg = ofs >> header->obj_order; | |
769 | end_seg = (ofs + len - 1) >> header->obj_order; | |
770 | ||
1fec7093 YS |
771 | return end_seg - start_seg + 1; |
772 | } | |
773 | ||
029bcbd8 JD |
774 | /* |
775 | * returns the size of an object in the image | |
776 | */ | |
777 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
778 | { | |
779 | return 1 << header->obj_order; | |
780 | } | |
781 | ||
602adf40 YS |
782 | /* |
783 | * bio helpers | |
784 | */ | |
785 | ||
786 | static void bio_chain_put(struct bio *chain) | |
787 | { | |
788 | struct bio *tmp; | |
789 | ||
790 | while (chain) { | |
791 | tmp = chain; | |
792 | chain = chain->bi_next; | |
793 | bio_put(tmp); | |
794 | } | |
795 | } | |
796 | ||
797 | /* | |
798 | * zeros a bio chain, starting at specific offset | |
799 | */ | |
800 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
801 | { | |
802 | struct bio_vec *bv; | |
803 | unsigned long flags; | |
804 | void *buf; | |
805 | int i; | |
806 | int pos = 0; | |
807 | ||
808 | while (chain) { | |
809 | bio_for_each_segment(bv, chain, i) { | |
810 | if (pos + bv->bv_len > start_ofs) { | |
811 | int remainder = max(start_ofs - pos, 0); | |
812 | buf = bvec_kmap_irq(bv, &flags); | |
813 | memset(buf + remainder, 0, | |
814 | bv->bv_len - remainder); | |
85b5aaa6 | 815 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
816 | } |
817 | pos += bv->bv_len; | |
818 | } | |
819 | ||
820 | chain = chain->bi_next; | |
821 | } | |
822 | } | |
823 | ||
824 | /* | |
825 | * bio_chain_clone - clone a chain of bios up to a certain length. | |
826 | * might return a bio_pair that will need to be released. | |
827 | */ | |
828 | static struct bio *bio_chain_clone(struct bio **old, struct bio **next, | |
829 | struct bio_pair **bp, | |
830 | int len, gfp_t gfpmask) | |
831 | { | |
542582fc AE |
832 | struct bio *old_chain = *old; |
833 | struct bio *new_chain = NULL; | |
834 | struct bio *tail; | |
602adf40 YS |
835 | int total = 0; |
836 | ||
837 | if (*bp) { | |
838 | bio_pair_release(*bp); | |
839 | *bp = NULL; | |
840 | } | |
841 | ||
842 | while (old_chain && (total < len)) { | |
542582fc AE |
843 | struct bio *tmp; |
844 | ||
602adf40 YS |
845 | tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs); |
846 | if (!tmp) | |
847 | goto err_out; | |
542582fc | 848 | gfpmask &= ~__GFP_WAIT; /* can't wait after the first */ |
602adf40 YS |
849 | |
850 | if (total + old_chain->bi_size > len) { | |
851 | struct bio_pair *bp; | |
852 | ||
853 | /* | |
854 | * this split can only happen with a single paged bio, | |
855 | * split_bio will BUG_ON if this is not the case | |
856 | */ | |
857 | dout("bio_chain_clone split! total=%d remaining=%d" | |
bd919d45 AE |
858 | "bi_size=%u\n", |
859 | total, len - total, old_chain->bi_size); | |
602adf40 YS |
860 | |
861 | /* split the bio. We'll release it either in the next | |
862 | call, or it will have to be released outside */ | |
593a9e7b | 863 | bp = bio_split(old_chain, (len - total) / SECTOR_SIZE); |
602adf40 YS |
864 | if (!bp) |
865 | goto err_out; | |
866 | ||
867 | __bio_clone(tmp, &bp->bio1); | |
868 | ||
869 | *next = &bp->bio2; | |
870 | } else { | |
871 | __bio_clone(tmp, old_chain); | |
872 | *next = old_chain->bi_next; | |
873 | } | |
874 | ||
875 | tmp->bi_bdev = NULL; | |
602adf40 | 876 | tmp->bi_next = NULL; |
542582fc | 877 | if (new_chain) |
602adf40 | 878 | tail->bi_next = tmp; |
542582fc AE |
879 | else |
880 | new_chain = tmp; | |
881 | tail = tmp; | |
602adf40 YS |
882 | old_chain = old_chain->bi_next; |
883 | ||
884 | total += tmp->bi_size; | |
885 | } | |
886 | ||
aafb230e | 887 | rbd_assert(total == len); |
602adf40 | 888 | |
602adf40 YS |
889 | *old = old_chain; |
890 | ||
891 | return new_chain; | |
892 | ||
893 | err_out: | |
894 | dout("bio_chain_clone with err\n"); | |
895 | bio_chain_put(new_chain); | |
896 | return NULL; | |
897 | } | |
898 | ||
899 | /* | |
900 | * helpers for osd request op vectors. | |
901 | */ | |
57cfc106 AE |
902 | static struct ceph_osd_req_op *rbd_create_rw_ops(int num_ops, |
903 | int opcode, u32 payload_len) | |
602adf40 | 904 | { |
57cfc106 AE |
905 | struct ceph_osd_req_op *ops; |
906 | ||
907 | ops = kzalloc(sizeof (*ops) * (num_ops + 1), GFP_NOIO); | |
908 | if (!ops) | |
909 | return NULL; | |
910 | ||
911 | ops[0].op = opcode; | |
912 | ||
602adf40 YS |
913 | /* |
914 | * op extent offset and length will be set later on | |
915 | * in calc_raw_layout() | |
916 | */ | |
57cfc106 AE |
917 | ops[0].payload_len = payload_len; |
918 | ||
919 | return ops; | |
602adf40 YS |
920 | } |
921 | ||
922 | static void rbd_destroy_ops(struct ceph_osd_req_op *ops) | |
923 | { | |
924 | kfree(ops); | |
925 | } | |
926 | ||
1fec7093 YS |
927 | static void rbd_coll_end_req_index(struct request *rq, |
928 | struct rbd_req_coll *coll, | |
929 | int index, | |
930 | int ret, u64 len) | |
931 | { | |
932 | struct request_queue *q; | |
933 | int min, max, i; | |
934 | ||
bd919d45 AE |
935 | dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n", |
936 | coll, index, ret, (unsigned long long) len); | |
1fec7093 YS |
937 | |
938 | if (!rq) | |
939 | return; | |
940 | ||
941 | if (!coll) { | |
942 | blk_end_request(rq, ret, len); | |
943 | return; | |
944 | } | |
945 | ||
946 | q = rq->q; | |
947 | ||
948 | spin_lock_irq(q->queue_lock); | |
949 | coll->status[index].done = 1; | |
950 | coll->status[index].rc = ret; | |
951 | coll->status[index].bytes = len; | |
952 | max = min = coll->num_done; | |
953 | while (max < coll->total && coll->status[max].done) | |
954 | max++; | |
955 | ||
956 | for (i = min; i<max; i++) { | |
957 | __blk_end_request(rq, coll->status[i].rc, | |
958 | coll->status[i].bytes); | |
959 | coll->num_done++; | |
960 | kref_put(&coll->kref, rbd_coll_release); | |
961 | } | |
962 | spin_unlock_irq(q->queue_lock); | |
963 | } | |
964 | ||
965 | static void rbd_coll_end_req(struct rbd_request *req, | |
966 | int ret, u64 len) | |
967 | { | |
968 | rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len); | |
969 | } | |
970 | ||
602adf40 YS |
971 | /* |
972 | * Send ceph osd request | |
973 | */ | |
974 | static int rbd_do_request(struct request *rq, | |
0ce1a794 | 975 | struct rbd_device *rbd_dev, |
602adf40 YS |
976 | struct ceph_snap_context *snapc, |
977 | u64 snapid, | |
aded07ea | 978 | const char *object_name, u64 ofs, u64 len, |
602adf40 YS |
979 | struct bio *bio, |
980 | struct page **pages, | |
981 | int num_pages, | |
982 | int flags, | |
983 | struct ceph_osd_req_op *ops, | |
1fec7093 YS |
984 | struct rbd_req_coll *coll, |
985 | int coll_index, | |
602adf40 | 986 | void (*rbd_cb)(struct ceph_osd_request *req, |
59c2be1e YS |
987 | struct ceph_msg *msg), |
988 | struct ceph_osd_request **linger_req, | |
989 | u64 *ver) | |
602adf40 YS |
990 | { |
991 | struct ceph_osd_request *req; | |
992 | struct ceph_file_layout *layout; | |
993 | int ret; | |
994 | u64 bno; | |
995 | struct timespec mtime = CURRENT_TIME; | |
996 | struct rbd_request *req_data; | |
997 | struct ceph_osd_request_head *reqhead; | |
1dbb4399 | 998 | struct ceph_osd_client *osdc; |
602adf40 | 999 | |
602adf40 | 1000 | req_data = kzalloc(sizeof(*req_data), GFP_NOIO); |
1fec7093 YS |
1001 | if (!req_data) { |
1002 | if (coll) | |
1003 | rbd_coll_end_req_index(rq, coll, coll_index, | |
1004 | -ENOMEM, len); | |
1005 | return -ENOMEM; | |
1006 | } | |
1007 | ||
1008 | if (coll) { | |
1009 | req_data->coll = coll; | |
1010 | req_data->coll_index = coll_index; | |
1011 | } | |
602adf40 | 1012 | |
bd919d45 AE |
1013 | dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name, |
1014 | (unsigned long long) ofs, (unsigned long long) len); | |
602adf40 | 1015 | |
0ce1a794 | 1016 | osdc = &rbd_dev->rbd_client->client->osdc; |
1dbb4399 AE |
1017 | req = ceph_osdc_alloc_request(osdc, flags, snapc, ops, |
1018 | false, GFP_NOIO, pages, bio); | |
4ad12621 | 1019 | if (!req) { |
4ad12621 | 1020 | ret = -ENOMEM; |
602adf40 YS |
1021 | goto done_pages; |
1022 | } | |
1023 | ||
1024 | req->r_callback = rbd_cb; | |
1025 | ||
1026 | req_data->rq = rq; | |
1027 | req_data->bio = bio; | |
1028 | req_data->pages = pages; | |
1029 | req_data->len = len; | |
1030 | ||
1031 | req->r_priv = req_data; | |
1032 | ||
1033 | reqhead = req->r_request->front.iov_base; | |
1034 | reqhead->snapid = cpu_to_le64(CEPH_NOSNAP); | |
1035 | ||
aded07ea | 1036 | strncpy(req->r_oid, object_name, sizeof(req->r_oid)); |
602adf40 YS |
1037 | req->r_oid_len = strlen(req->r_oid); |
1038 | ||
1039 | layout = &req->r_file_layout; | |
1040 | memset(layout, 0, sizeof(*layout)); | |
1041 | layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
1042 | layout->fl_stripe_count = cpu_to_le32(1); | |
1043 | layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
0ce1a794 | 1044 | layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id); |
6cae3717 SW |
1045 | ret = ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno, |
1046 | req, ops); | |
1047 | rbd_assert(ret == 0); | |
602adf40 YS |
1048 | |
1049 | ceph_osdc_build_request(req, ofs, &len, | |
1050 | ops, | |
1051 | snapc, | |
1052 | &mtime, | |
1053 | req->r_oid, req->r_oid_len); | |
602adf40 | 1054 | |
59c2be1e | 1055 | if (linger_req) { |
1dbb4399 | 1056 | ceph_osdc_set_request_linger(osdc, req); |
59c2be1e YS |
1057 | *linger_req = req; |
1058 | } | |
1059 | ||
1dbb4399 | 1060 | ret = ceph_osdc_start_request(osdc, req, false); |
602adf40 YS |
1061 | if (ret < 0) |
1062 | goto done_err; | |
1063 | ||
1064 | if (!rbd_cb) { | |
1dbb4399 | 1065 | ret = ceph_osdc_wait_request(osdc, req); |
59c2be1e YS |
1066 | if (ver) |
1067 | *ver = le64_to_cpu(req->r_reassert_version.version); | |
bd919d45 AE |
1068 | dout("reassert_ver=%llu\n", |
1069 | (unsigned long long) | |
1070 | le64_to_cpu(req->r_reassert_version.version)); | |
602adf40 YS |
1071 | ceph_osdc_put_request(req); |
1072 | } | |
1073 | return ret; | |
1074 | ||
1075 | done_err: | |
1076 | bio_chain_put(req_data->bio); | |
1077 | ceph_osdc_put_request(req); | |
1078 | done_pages: | |
1fec7093 | 1079 | rbd_coll_end_req(req_data, ret, len); |
602adf40 | 1080 | kfree(req_data); |
602adf40 YS |
1081 | return ret; |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Ceph osd op callback | |
1086 | */ | |
1087 | static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) | |
1088 | { | |
1089 | struct rbd_request *req_data = req->r_priv; | |
1090 | struct ceph_osd_reply_head *replyhead; | |
1091 | struct ceph_osd_op *op; | |
1092 | __s32 rc; | |
1093 | u64 bytes; | |
1094 | int read_op; | |
1095 | ||
1096 | /* parse reply */ | |
1097 | replyhead = msg->front.iov_base; | |
1098 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
1099 | op = (void *)(replyhead + 1); | |
1100 | rc = le32_to_cpu(replyhead->result); | |
1101 | bytes = le64_to_cpu(op->extent.length); | |
895cfcc8 | 1102 | read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ); |
602adf40 | 1103 | |
bd919d45 AE |
1104 | dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n", |
1105 | (unsigned long long) bytes, read_op, (int) rc); | |
602adf40 YS |
1106 | |
1107 | if (rc == -ENOENT && read_op) { | |
1108 | zero_bio_chain(req_data->bio, 0); | |
1109 | rc = 0; | |
1110 | } else if (rc == 0 && read_op && bytes < req_data->len) { | |
1111 | zero_bio_chain(req_data->bio, bytes); | |
1112 | bytes = req_data->len; | |
1113 | } | |
1114 | ||
1fec7093 | 1115 | rbd_coll_end_req(req_data, rc, bytes); |
602adf40 YS |
1116 | |
1117 | if (req_data->bio) | |
1118 | bio_chain_put(req_data->bio); | |
1119 | ||
1120 | ceph_osdc_put_request(req); | |
1121 | kfree(req_data); | |
1122 | } | |
1123 | ||
59c2be1e YS |
1124 | static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) |
1125 | { | |
1126 | ceph_osdc_put_request(req); | |
1127 | } | |
1128 | ||
602adf40 YS |
1129 | /* |
1130 | * Do a synchronous ceph osd operation | |
1131 | */ | |
0ce1a794 | 1132 | static int rbd_req_sync_op(struct rbd_device *rbd_dev, |
602adf40 YS |
1133 | struct ceph_snap_context *snapc, |
1134 | u64 snapid, | |
602adf40 | 1135 | int flags, |
913d2fdc | 1136 | struct ceph_osd_req_op *ops, |
aded07ea | 1137 | const char *object_name, |
f8d4de6e AE |
1138 | u64 ofs, u64 inbound_size, |
1139 | char *inbound, | |
59c2be1e YS |
1140 | struct ceph_osd_request **linger_req, |
1141 | u64 *ver) | |
602adf40 YS |
1142 | { |
1143 | int ret; | |
1144 | struct page **pages; | |
1145 | int num_pages; | |
913d2fdc | 1146 | |
aafb230e | 1147 | rbd_assert(ops != NULL); |
602adf40 | 1148 | |
f8d4de6e | 1149 | num_pages = calc_pages_for(ofs, inbound_size); |
602adf40 | 1150 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
b8d0638a DC |
1151 | if (IS_ERR(pages)) |
1152 | return PTR_ERR(pages); | |
602adf40 | 1153 | |
0ce1a794 | 1154 | ret = rbd_do_request(NULL, rbd_dev, snapc, snapid, |
f8d4de6e | 1155 | object_name, ofs, inbound_size, NULL, |
602adf40 YS |
1156 | pages, num_pages, |
1157 | flags, | |
1158 | ops, | |
1fec7093 | 1159 | NULL, 0, |
59c2be1e YS |
1160 | NULL, |
1161 | linger_req, ver); | |
602adf40 | 1162 | if (ret < 0) |
913d2fdc | 1163 | goto done; |
602adf40 | 1164 | |
f8d4de6e AE |
1165 | if ((flags & CEPH_OSD_FLAG_READ) && inbound) |
1166 | ret = ceph_copy_from_page_vector(pages, inbound, ofs, ret); | |
602adf40 | 1167 | |
602adf40 YS |
1168 | done: |
1169 | ceph_release_page_vector(pages, num_pages); | |
1170 | return ret; | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Do an asynchronous ceph osd operation | |
1175 | */ | |
1176 | static int rbd_do_op(struct request *rq, | |
0ce1a794 | 1177 | struct rbd_device *rbd_dev, |
602adf40 | 1178 | struct ceph_snap_context *snapc, |
602adf40 | 1179 | u64 ofs, u64 len, |
1fec7093 YS |
1180 | struct bio *bio, |
1181 | struct rbd_req_coll *coll, | |
1182 | int coll_index) | |
602adf40 YS |
1183 | { |
1184 | char *seg_name; | |
1185 | u64 seg_ofs; | |
1186 | u64 seg_len; | |
1187 | int ret; | |
1188 | struct ceph_osd_req_op *ops; | |
1189 | u32 payload_len; | |
ff2e4bb5 AE |
1190 | int opcode; |
1191 | int flags; | |
4634246d | 1192 | u64 snapid; |
602adf40 | 1193 | |
65ccfe21 | 1194 | seg_name = rbd_segment_name(rbd_dev, ofs); |
602adf40 YS |
1195 | if (!seg_name) |
1196 | return -ENOMEM; | |
65ccfe21 AE |
1197 | seg_len = rbd_segment_length(rbd_dev, ofs, len); |
1198 | seg_ofs = rbd_segment_offset(rbd_dev, ofs); | |
602adf40 | 1199 | |
ff2e4bb5 AE |
1200 | if (rq_data_dir(rq) == WRITE) { |
1201 | opcode = CEPH_OSD_OP_WRITE; | |
1202 | flags = CEPH_OSD_FLAG_WRITE|CEPH_OSD_FLAG_ONDISK; | |
4634246d | 1203 | snapid = CEPH_NOSNAP; |
ff2e4bb5 AE |
1204 | payload_len = seg_len; |
1205 | } else { | |
1206 | opcode = CEPH_OSD_OP_READ; | |
1207 | flags = CEPH_OSD_FLAG_READ; | |
4634246d AE |
1208 | snapc = NULL; |
1209 | snapid = rbd_dev->mapping.snap_id; | |
ff2e4bb5 AE |
1210 | payload_len = 0; |
1211 | } | |
602adf40 | 1212 | |
57cfc106 AE |
1213 | ret = -ENOMEM; |
1214 | ops = rbd_create_rw_ops(1, opcode, payload_len); | |
1215 | if (!ops) | |
602adf40 YS |
1216 | goto done; |
1217 | ||
1218 | /* we've taken care of segment sizes earlier when we | |
1219 | cloned the bios. We should never have a segment | |
1220 | truncated at this point */ | |
aafb230e | 1221 | rbd_assert(seg_len == len); |
602adf40 YS |
1222 | |
1223 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1224 | seg_name, seg_ofs, seg_len, | |
1225 | bio, | |
1226 | NULL, 0, | |
1227 | flags, | |
1228 | ops, | |
1fec7093 | 1229 | coll, coll_index, |
59c2be1e | 1230 | rbd_req_cb, 0, NULL); |
11f77002 SW |
1231 | |
1232 | rbd_destroy_ops(ops); | |
602adf40 YS |
1233 | done: |
1234 | kfree(seg_name); | |
1235 | return ret; | |
1236 | } | |
1237 | ||
602adf40 YS |
1238 | /* |
1239 | * Request sync osd read | |
1240 | */ | |
0ce1a794 | 1241 | static int rbd_req_sync_read(struct rbd_device *rbd_dev, |
602adf40 | 1242 | u64 snapid, |
aded07ea | 1243 | const char *object_name, |
602adf40 | 1244 | u64 ofs, u64 len, |
59c2be1e YS |
1245 | char *buf, |
1246 | u64 *ver) | |
602adf40 | 1247 | { |
913d2fdc AE |
1248 | struct ceph_osd_req_op *ops; |
1249 | int ret; | |
1250 | ||
1251 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_READ, 0); | |
1252 | if (!ops) | |
1253 | return -ENOMEM; | |
1254 | ||
1255 | ret = rbd_req_sync_op(rbd_dev, NULL, | |
b06e6a6b | 1256 | snapid, |
602adf40 | 1257 | CEPH_OSD_FLAG_READ, |
913d2fdc AE |
1258 | ops, object_name, ofs, len, buf, NULL, ver); |
1259 | rbd_destroy_ops(ops); | |
1260 | ||
1261 | return ret; | |
602adf40 YS |
1262 | } |
1263 | ||
1264 | /* | |
59c2be1e YS |
1265 | * Request sync osd watch |
1266 | */ | |
0ce1a794 | 1267 | static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev, |
59c2be1e | 1268 | u64 ver, |
7f0a24d8 | 1269 | u64 notify_id) |
59c2be1e YS |
1270 | { |
1271 | struct ceph_osd_req_op *ops; | |
11f77002 SW |
1272 | int ret; |
1273 | ||
57cfc106 AE |
1274 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY_ACK, 0); |
1275 | if (!ops) | |
1276 | return -ENOMEM; | |
59c2be1e | 1277 | |
a71b891b | 1278 | ops[0].watch.ver = cpu_to_le64(ver); |
59c2be1e YS |
1279 | ops[0].watch.cookie = notify_id; |
1280 | ops[0].watch.flag = 0; | |
1281 | ||
0ce1a794 | 1282 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
7f0a24d8 | 1283 | rbd_dev->header_name, 0, 0, NULL, |
ad4f232f | 1284 | NULL, 0, |
59c2be1e YS |
1285 | CEPH_OSD_FLAG_READ, |
1286 | ops, | |
1fec7093 | 1287 | NULL, 0, |
59c2be1e YS |
1288 | rbd_simple_req_cb, 0, NULL); |
1289 | ||
1290 | rbd_destroy_ops(ops); | |
1291 | return ret; | |
1292 | } | |
1293 | ||
1294 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1295 | { | |
0ce1a794 | 1296 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
a71b891b | 1297 | u64 hver; |
13143d2d SW |
1298 | int rc; |
1299 | ||
0ce1a794 | 1300 | if (!rbd_dev) |
59c2be1e YS |
1301 | return; |
1302 | ||
bd919d45 AE |
1303 | dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n", |
1304 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1305 | (unsigned int) opcode); | |
117973fb | 1306 | rc = rbd_dev_refresh(rbd_dev, &hver); |
13143d2d | 1307 | if (rc) |
f0f8cef5 | 1308 | pr_warning(RBD_DRV_NAME "%d got notification but failed to " |
0ce1a794 | 1309 | " update snaps: %d\n", rbd_dev->major, rc); |
59c2be1e | 1310 | |
7f0a24d8 | 1311 | rbd_req_sync_notify_ack(rbd_dev, hver, notify_id); |
59c2be1e YS |
1312 | } |
1313 | ||
1314 | /* | |
1315 | * Request sync osd watch | |
1316 | */ | |
0e6f322d | 1317 | static int rbd_req_sync_watch(struct rbd_device *rbd_dev) |
59c2be1e YS |
1318 | { |
1319 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1320 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
57cfc106 | 1321 | int ret; |
59c2be1e | 1322 | |
57cfc106 AE |
1323 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1324 | if (!ops) | |
1325 | return -ENOMEM; | |
59c2be1e YS |
1326 | |
1327 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, | |
0ce1a794 | 1328 | (void *)rbd_dev, &rbd_dev->watch_event); |
59c2be1e YS |
1329 | if (ret < 0) |
1330 | goto fail; | |
1331 | ||
0e6f322d | 1332 | ops[0].watch.ver = cpu_to_le64(rbd_dev->header.obj_version); |
0ce1a794 | 1333 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
59c2be1e YS |
1334 | ops[0].watch.flag = 1; |
1335 | ||
0ce1a794 | 1336 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e | 1337 | CEPH_NOSNAP, |
59c2be1e YS |
1338 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1339 | ops, | |
0e6f322d AE |
1340 | rbd_dev->header_name, |
1341 | 0, 0, NULL, | |
0ce1a794 | 1342 | &rbd_dev->watch_request, NULL); |
59c2be1e YS |
1343 | |
1344 | if (ret < 0) | |
1345 | goto fail_event; | |
1346 | ||
1347 | rbd_destroy_ops(ops); | |
1348 | return 0; | |
1349 | ||
1350 | fail_event: | |
0ce1a794 AE |
1351 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1352 | rbd_dev->watch_event = NULL; | |
59c2be1e YS |
1353 | fail: |
1354 | rbd_destroy_ops(ops); | |
1355 | return ret; | |
1356 | } | |
1357 | ||
79e3057c YS |
1358 | /* |
1359 | * Request sync osd unwatch | |
1360 | */ | |
070c633f | 1361 | static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev) |
79e3057c YS |
1362 | { |
1363 | struct ceph_osd_req_op *ops; | |
57cfc106 | 1364 | int ret; |
79e3057c | 1365 | |
57cfc106 AE |
1366 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1367 | if (!ops) | |
1368 | return -ENOMEM; | |
79e3057c YS |
1369 | |
1370 | ops[0].watch.ver = 0; | |
0ce1a794 | 1371 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
79e3057c YS |
1372 | ops[0].watch.flag = 0; |
1373 | ||
0ce1a794 | 1374 | ret = rbd_req_sync_op(rbd_dev, NULL, |
79e3057c | 1375 | CEPH_NOSNAP, |
79e3057c YS |
1376 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1377 | ops, | |
070c633f AE |
1378 | rbd_dev->header_name, |
1379 | 0, 0, NULL, NULL, NULL); | |
1380 | ||
79e3057c YS |
1381 | |
1382 | rbd_destroy_ops(ops); | |
0ce1a794 AE |
1383 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1384 | rbd_dev->watch_event = NULL; | |
79e3057c YS |
1385 | return ret; |
1386 | } | |
1387 | ||
602adf40 | 1388 | /* |
3cb4a687 | 1389 | * Synchronous osd object method call |
602adf40 | 1390 | */ |
0ce1a794 | 1391 | static int rbd_req_sync_exec(struct rbd_device *rbd_dev, |
aded07ea AE |
1392 | const char *object_name, |
1393 | const char *class_name, | |
1394 | const char *method_name, | |
3cb4a687 AE |
1395 | const char *outbound, |
1396 | size_t outbound_size, | |
f8d4de6e AE |
1397 | char *inbound, |
1398 | size_t inbound_size, | |
3cb4a687 | 1399 | int flags, |
59c2be1e | 1400 | u64 *ver) |
602adf40 YS |
1401 | { |
1402 | struct ceph_osd_req_op *ops; | |
aded07ea AE |
1403 | int class_name_len = strlen(class_name); |
1404 | int method_name_len = strlen(method_name); | |
3cb4a687 | 1405 | int payload_size; |
57cfc106 AE |
1406 | int ret; |
1407 | ||
3cb4a687 AE |
1408 | /* |
1409 | * Any input parameters required by the method we're calling | |
1410 | * will be sent along with the class and method names as | |
1411 | * part of the message payload. That data and its size are | |
1412 | * supplied via the indata and indata_len fields (named from | |
1413 | * the perspective of the server side) in the OSD request | |
1414 | * operation. | |
1415 | */ | |
1416 | payload_size = class_name_len + method_name_len + outbound_size; | |
1417 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_CALL, payload_size); | |
57cfc106 AE |
1418 | if (!ops) |
1419 | return -ENOMEM; | |
602adf40 | 1420 | |
aded07ea AE |
1421 | ops[0].cls.class_name = class_name; |
1422 | ops[0].cls.class_len = (__u8) class_name_len; | |
1423 | ops[0].cls.method_name = method_name; | |
1424 | ops[0].cls.method_len = (__u8) method_name_len; | |
602adf40 | 1425 | ops[0].cls.argc = 0; |
3cb4a687 AE |
1426 | ops[0].cls.indata = outbound; |
1427 | ops[0].cls.indata_len = outbound_size; | |
602adf40 | 1428 | |
0ce1a794 | 1429 | ret = rbd_req_sync_op(rbd_dev, NULL, |
602adf40 | 1430 | CEPH_NOSNAP, |
3cb4a687 | 1431 | flags, ops, |
f8d4de6e AE |
1432 | object_name, 0, inbound_size, inbound, |
1433 | NULL, ver); | |
602adf40 YS |
1434 | |
1435 | rbd_destroy_ops(ops); | |
1436 | ||
1437 | dout("cls_exec returned %d\n", ret); | |
1438 | return ret; | |
1439 | } | |
1440 | ||
1fec7093 YS |
1441 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1442 | { | |
1443 | struct rbd_req_coll *coll = | |
1444 | kzalloc(sizeof(struct rbd_req_coll) + | |
1445 | sizeof(struct rbd_req_status) * num_reqs, | |
1446 | GFP_ATOMIC); | |
1447 | ||
1448 | if (!coll) | |
1449 | return NULL; | |
1450 | coll->total = num_reqs; | |
1451 | kref_init(&coll->kref); | |
1452 | return coll; | |
1453 | } | |
1454 | ||
602adf40 YS |
1455 | /* |
1456 | * block device queue callback | |
1457 | */ | |
1458 | static void rbd_rq_fn(struct request_queue *q) | |
1459 | { | |
1460 | struct rbd_device *rbd_dev = q->queuedata; | |
1461 | struct request *rq; | |
1462 | struct bio_pair *bp = NULL; | |
1463 | ||
00f1f36f | 1464 | while ((rq = blk_fetch_request(q))) { |
602adf40 YS |
1465 | struct bio *bio; |
1466 | struct bio *rq_bio, *next_bio = NULL; | |
1467 | bool do_write; | |
bd919d45 AE |
1468 | unsigned int size; |
1469 | u64 op_size = 0; | |
602adf40 | 1470 | u64 ofs; |
1fec7093 YS |
1471 | int num_segs, cur_seg = 0; |
1472 | struct rbd_req_coll *coll; | |
d1d25646 | 1473 | struct ceph_snap_context *snapc; |
602adf40 | 1474 | |
602adf40 YS |
1475 | dout("fetched request\n"); |
1476 | ||
1477 | /* filter out block requests we don't understand */ | |
1478 | if ((rq->cmd_type != REQ_TYPE_FS)) { | |
1479 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1480 | continue; |
602adf40 YS |
1481 | } |
1482 | ||
1483 | /* deduce our operation (read, write) */ | |
1484 | do_write = (rq_data_dir(rq) == WRITE); | |
1485 | ||
1486 | size = blk_rq_bytes(rq); | |
593a9e7b | 1487 | ofs = blk_rq_pos(rq) * SECTOR_SIZE; |
602adf40 | 1488 | rq_bio = rq->bio; |
f84344f3 | 1489 | if (do_write && rbd_dev->mapping.read_only) { |
602adf40 | 1490 | __blk_end_request_all(rq, -EROFS); |
00f1f36f | 1491 | continue; |
602adf40 YS |
1492 | } |
1493 | ||
1494 | spin_unlock_irq(q->queue_lock); | |
1495 | ||
d1d25646 | 1496 | down_read(&rbd_dev->header_rwsem); |
e88a36ec | 1497 | |
f84344f3 AE |
1498 | if (rbd_dev->mapping.snap_id != CEPH_NOSNAP && |
1499 | !rbd_dev->mapping.snap_exists) { | |
e88a36ec | 1500 | up_read(&rbd_dev->header_rwsem); |
d1d25646 JD |
1501 | dout("request for non-existent snapshot"); |
1502 | spin_lock_irq(q->queue_lock); | |
1503 | __blk_end_request_all(rq, -ENXIO); | |
1504 | continue; | |
e88a36ec JD |
1505 | } |
1506 | ||
d1d25646 JD |
1507 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
1508 | ||
1509 | up_read(&rbd_dev->header_rwsem); | |
1510 | ||
602adf40 YS |
1511 | dout("%s 0x%x bytes at 0x%llx\n", |
1512 | do_write ? "write" : "read", | |
bd919d45 | 1513 | size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE); |
602adf40 | 1514 | |
1fec7093 | 1515 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); |
df111be6 AE |
1516 | if (num_segs <= 0) { |
1517 | spin_lock_irq(q->queue_lock); | |
1518 | __blk_end_request_all(rq, num_segs); | |
1519 | ceph_put_snap_context(snapc); | |
1520 | continue; | |
1521 | } | |
1fec7093 YS |
1522 | coll = rbd_alloc_coll(num_segs); |
1523 | if (!coll) { | |
1524 | spin_lock_irq(q->queue_lock); | |
1525 | __blk_end_request_all(rq, -ENOMEM); | |
d1d25646 | 1526 | ceph_put_snap_context(snapc); |
00f1f36f | 1527 | continue; |
1fec7093 YS |
1528 | } |
1529 | ||
602adf40 YS |
1530 | do { |
1531 | /* a bio clone to be passed down to OSD req */ | |
bd919d45 | 1532 | dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt); |
65ccfe21 | 1533 | op_size = rbd_segment_length(rbd_dev, ofs, size); |
1fec7093 | 1534 | kref_get(&coll->kref); |
602adf40 YS |
1535 | bio = bio_chain_clone(&rq_bio, &next_bio, &bp, |
1536 | op_size, GFP_ATOMIC); | |
4634246d AE |
1537 | if (bio) |
1538 | (void) rbd_do_op(rq, rbd_dev, snapc, | |
1539 | ofs, op_size, | |
1540 | bio, coll, cur_seg); | |
1541 | else | |
1fec7093 YS |
1542 | rbd_coll_end_req_index(rq, coll, cur_seg, |
1543 | -ENOMEM, op_size); | |
602adf40 YS |
1544 | size -= op_size; |
1545 | ofs += op_size; | |
1546 | ||
1fec7093 | 1547 | cur_seg++; |
602adf40 YS |
1548 | rq_bio = next_bio; |
1549 | } while (size > 0); | |
1fec7093 | 1550 | kref_put(&coll->kref, rbd_coll_release); |
602adf40 YS |
1551 | |
1552 | if (bp) | |
1553 | bio_pair_release(bp); | |
602adf40 | 1554 | spin_lock_irq(q->queue_lock); |
d1d25646 JD |
1555 | |
1556 | ceph_put_snap_context(snapc); | |
602adf40 YS |
1557 | } |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1562 | * multiple osd objects. One exception would be with a single page bios, | |
1563 | * which we handle later at bio_chain_clone | |
1564 | */ | |
1565 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1566 | struct bio_vec *bvec) | |
1567 | { | |
1568 | struct rbd_device *rbd_dev = q->queuedata; | |
593a9e7b AE |
1569 | unsigned int chunk_sectors; |
1570 | sector_t sector; | |
1571 | unsigned int bio_sectors; | |
602adf40 YS |
1572 | int max; |
1573 | ||
593a9e7b AE |
1574 | chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); |
1575 | sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev); | |
1576 | bio_sectors = bmd->bi_size >> SECTOR_SHIFT; | |
1577 | ||
602adf40 | 1578 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) |
593a9e7b | 1579 | + bio_sectors)) << SECTOR_SHIFT; |
602adf40 YS |
1580 | if (max < 0) |
1581 | max = 0; /* bio_add cannot handle a negative return */ | |
1582 | if (max <= bvec->bv_len && bio_sectors == 0) | |
1583 | return bvec->bv_len; | |
1584 | return max; | |
1585 | } | |
1586 | ||
1587 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1588 | { | |
1589 | struct gendisk *disk = rbd_dev->disk; | |
1590 | ||
1591 | if (!disk) | |
1592 | return; | |
1593 | ||
602adf40 YS |
1594 | if (disk->flags & GENHD_FL_UP) |
1595 | del_gendisk(disk); | |
1596 | if (disk->queue) | |
1597 | blk_cleanup_queue(disk->queue); | |
1598 | put_disk(disk); | |
1599 | } | |
1600 | ||
1601 | /* | |
4156d998 AE |
1602 | * Read the complete header for the given rbd device. |
1603 | * | |
1604 | * Returns a pointer to a dynamically-allocated buffer containing | |
1605 | * the complete and validated header. Caller can pass the address | |
1606 | * of a variable that will be filled in with the version of the | |
1607 | * header object at the time it was read. | |
1608 | * | |
1609 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 1610 | */ |
4156d998 AE |
1611 | static struct rbd_image_header_ondisk * |
1612 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 1613 | { |
4156d998 | 1614 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 1615 | u32 snap_count = 0; |
4156d998 AE |
1616 | u64 names_size = 0; |
1617 | u32 want_count; | |
1618 | int ret; | |
602adf40 | 1619 | |
00f1f36f | 1620 | /* |
4156d998 AE |
1621 | * The complete header will include an array of its 64-bit |
1622 | * snapshot ids, followed by the names of those snapshots as | |
1623 | * a contiguous block of NUL-terminated strings. Note that | |
1624 | * the number of snapshots could change by the time we read | |
1625 | * it in, in which case we re-read it. | |
00f1f36f | 1626 | */ |
4156d998 AE |
1627 | do { |
1628 | size_t size; | |
1629 | ||
1630 | kfree(ondisk); | |
1631 | ||
1632 | size = sizeof (*ondisk); | |
1633 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
1634 | size += names_size; | |
1635 | ondisk = kmalloc(size, GFP_KERNEL); | |
1636 | if (!ondisk) | |
1637 | return ERR_PTR(-ENOMEM); | |
1638 | ||
1639 | ret = rbd_req_sync_read(rbd_dev, CEPH_NOSNAP, | |
0bed54dc | 1640 | rbd_dev->header_name, |
4156d998 AE |
1641 | 0, size, |
1642 | (char *) ondisk, version); | |
1643 | ||
1644 | if (ret < 0) | |
1645 | goto out_err; | |
1646 | if (WARN_ON((size_t) ret < size)) { | |
1647 | ret = -ENXIO; | |
1648 | pr_warning("short header read for image %s" | |
1649 | " (want %zd got %d)\n", | |
1650 | rbd_dev->image_name, size, ret); | |
1651 | goto out_err; | |
1652 | } | |
1653 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
1654 | ret = -ENXIO; | |
1655 | pr_warning("invalid header for image %s\n", | |
1656 | rbd_dev->image_name); | |
1657 | goto out_err; | |
81e759fb | 1658 | } |
602adf40 | 1659 | |
4156d998 AE |
1660 | names_size = le64_to_cpu(ondisk->snap_names_len); |
1661 | want_count = snap_count; | |
1662 | snap_count = le32_to_cpu(ondisk->snap_count); | |
1663 | } while (snap_count != want_count); | |
00f1f36f | 1664 | |
4156d998 | 1665 | return ondisk; |
00f1f36f | 1666 | |
4156d998 AE |
1667 | out_err: |
1668 | kfree(ondisk); | |
1669 | ||
1670 | return ERR_PTR(ret); | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * reload the ondisk the header | |
1675 | */ | |
1676 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1677 | struct rbd_image_header *header) | |
1678 | { | |
1679 | struct rbd_image_header_ondisk *ondisk; | |
1680 | u64 ver = 0; | |
1681 | int ret; | |
602adf40 | 1682 | |
4156d998 AE |
1683 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
1684 | if (IS_ERR(ondisk)) | |
1685 | return PTR_ERR(ondisk); | |
1686 | ret = rbd_header_from_disk(header, ondisk); | |
1687 | if (ret >= 0) | |
1688 | header->obj_version = ver; | |
1689 | kfree(ondisk); | |
1690 | ||
1691 | return ret; | |
602adf40 YS |
1692 | } |
1693 | ||
dfc5606d YS |
1694 | static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
1695 | { | |
1696 | struct rbd_snap *snap; | |
a0593290 | 1697 | struct rbd_snap *next; |
dfc5606d | 1698 | |
a0593290 | 1699 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
14e7085d | 1700 | __rbd_remove_snap_dev(snap); |
dfc5606d YS |
1701 | } |
1702 | ||
9478554a AE |
1703 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
1704 | { | |
1705 | sector_t size; | |
1706 | ||
1707 | if (rbd_dev->mapping.snap_id != CEPH_NOSNAP) | |
1708 | return; | |
1709 | ||
1710 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
1711 | dout("setting size to %llu sectors", (unsigned long long) size); | |
1712 | rbd_dev->mapping.size = (u64) size; | |
1713 | set_capacity(rbd_dev->disk, size); | |
1714 | } | |
1715 | ||
602adf40 YS |
1716 | /* |
1717 | * only read the first part of the ondisk header, without the snaps info | |
1718 | */ | |
117973fb | 1719 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
1720 | { |
1721 | int ret; | |
1722 | struct rbd_image_header h; | |
602adf40 YS |
1723 | |
1724 | ret = rbd_read_header(rbd_dev, &h); | |
1725 | if (ret < 0) | |
1726 | return ret; | |
1727 | ||
a51aa0c0 JD |
1728 | down_write(&rbd_dev->header_rwsem); |
1729 | ||
9478554a AE |
1730 | /* Update image size, and check for resize of mapped image */ |
1731 | rbd_dev->header.image_size = h.image_size; | |
1732 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 1733 | |
849b4260 | 1734 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 1735 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 1736 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
1737 | /* osd requests may still refer to snapc */ |
1738 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 1739 | |
b813623a AE |
1740 | if (hver) |
1741 | *hver = h.obj_version; | |
a71b891b | 1742 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 1743 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
1744 | rbd_dev->header.snapc = h.snapc; |
1745 | rbd_dev->header.snap_names = h.snap_names; | |
1746 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
1747 | /* Free the extra copy of the object prefix */ |
1748 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
1749 | kfree(h.object_prefix); | |
1750 | ||
304f6808 AE |
1751 | ret = rbd_dev_snaps_update(rbd_dev); |
1752 | if (!ret) | |
1753 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 1754 | |
c666601a | 1755 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1756 | |
dfc5606d | 1757 | return ret; |
602adf40 YS |
1758 | } |
1759 | ||
117973fb | 1760 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
1761 | { |
1762 | int ret; | |
1763 | ||
117973fb | 1764 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 1765 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
1766 | if (rbd_dev->image_format == 1) |
1767 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
1768 | else | |
1769 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
1770 | mutex_unlock(&ctl_mutex); |
1771 | ||
1772 | return ret; | |
1773 | } | |
1774 | ||
602adf40 YS |
1775 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
1776 | { | |
1777 | struct gendisk *disk; | |
1778 | struct request_queue *q; | |
593a9e7b | 1779 | u64 segment_size; |
602adf40 | 1780 | |
602adf40 | 1781 | /* create gendisk info */ |
602adf40 YS |
1782 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
1783 | if (!disk) | |
1fcdb8aa | 1784 | return -ENOMEM; |
602adf40 | 1785 | |
f0f8cef5 | 1786 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 1787 | rbd_dev->dev_id); |
602adf40 YS |
1788 | disk->major = rbd_dev->major; |
1789 | disk->first_minor = 0; | |
1790 | disk->fops = &rbd_bd_ops; | |
1791 | disk->private_data = rbd_dev; | |
1792 | ||
1793 | /* init rq */ | |
602adf40 YS |
1794 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); |
1795 | if (!q) | |
1796 | goto out_disk; | |
029bcbd8 | 1797 | |
593a9e7b AE |
1798 | /* We use the default size, but let's be explicit about it. */ |
1799 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1800 | ||
029bcbd8 | 1801 | /* set io sizes to object size */ |
593a9e7b AE |
1802 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1803 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1804 | blk_queue_max_segment_size(q, segment_size); | |
1805 | blk_queue_io_min(q, segment_size); | |
1806 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1807 | |
602adf40 YS |
1808 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1809 | disk->queue = q; | |
1810 | ||
1811 | q->queuedata = rbd_dev; | |
1812 | ||
1813 | rbd_dev->disk = disk; | |
602adf40 | 1814 | |
12f02944 AE |
1815 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
1816 | ||
602adf40 | 1817 | return 0; |
602adf40 YS |
1818 | out_disk: |
1819 | put_disk(disk); | |
1fcdb8aa AE |
1820 | |
1821 | return -ENOMEM; | |
602adf40 YS |
1822 | } |
1823 | ||
dfc5606d YS |
1824 | /* |
1825 | sysfs | |
1826 | */ | |
1827 | ||
593a9e7b AE |
1828 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1829 | { | |
1830 | return container_of(dev, struct rbd_device, dev); | |
1831 | } | |
1832 | ||
dfc5606d YS |
1833 | static ssize_t rbd_size_show(struct device *dev, |
1834 | struct device_attribute *attr, char *buf) | |
1835 | { | |
593a9e7b | 1836 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
1837 | sector_t size; |
1838 | ||
1839 | down_read(&rbd_dev->header_rwsem); | |
1840 | size = get_capacity(rbd_dev->disk); | |
1841 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 1842 | |
a51aa0c0 | 1843 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
1844 | } |
1845 | ||
34b13184 AE |
1846 | /* |
1847 | * Note this shows the features for whatever's mapped, which is not | |
1848 | * necessarily the base image. | |
1849 | */ | |
1850 | static ssize_t rbd_features_show(struct device *dev, | |
1851 | struct device_attribute *attr, char *buf) | |
1852 | { | |
1853 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1854 | ||
1855 | return sprintf(buf, "0x%016llx\n", | |
1856 | (unsigned long long) rbd_dev->mapping.features); | |
1857 | } | |
1858 | ||
dfc5606d YS |
1859 | static ssize_t rbd_major_show(struct device *dev, |
1860 | struct device_attribute *attr, char *buf) | |
1861 | { | |
593a9e7b | 1862 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 1863 | |
dfc5606d YS |
1864 | return sprintf(buf, "%d\n", rbd_dev->major); |
1865 | } | |
1866 | ||
1867 | static ssize_t rbd_client_id_show(struct device *dev, | |
1868 | struct device_attribute *attr, char *buf) | |
602adf40 | 1869 | { |
593a9e7b | 1870 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1871 | |
1dbb4399 AE |
1872 | return sprintf(buf, "client%lld\n", |
1873 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
1874 | } |
1875 | ||
dfc5606d YS |
1876 | static ssize_t rbd_pool_show(struct device *dev, |
1877 | struct device_attribute *attr, char *buf) | |
602adf40 | 1878 | { |
593a9e7b | 1879 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1880 | |
1881 | return sprintf(buf, "%s\n", rbd_dev->pool_name); | |
1882 | } | |
1883 | ||
9bb2f334 AE |
1884 | static ssize_t rbd_pool_id_show(struct device *dev, |
1885 | struct device_attribute *attr, char *buf) | |
1886 | { | |
1887 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1888 | ||
1889 | return sprintf(buf, "%d\n", rbd_dev->pool_id); | |
1890 | } | |
1891 | ||
dfc5606d YS |
1892 | static ssize_t rbd_name_show(struct device *dev, |
1893 | struct device_attribute *attr, char *buf) | |
1894 | { | |
593a9e7b | 1895 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1896 | |
0bed54dc | 1897 | return sprintf(buf, "%s\n", rbd_dev->image_name); |
dfc5606d YS |
1898 | } |
1899 | ||
589d30e0 AE |
1900 | static ssize_t rbd_image_id_show(struct device *dev, |
1901 | struct device_attribute *attr, char *buf) | |
1902 | { | |
1903 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1904 | ||
1905 | return sprintf(buf, "%s\n", rbd_dev->image_id); | |
1906 | } | |
1907 | ||
34b13184 AE |
1908 | /* |
1909 | * Shows the name of the currently-mapped snapshot (or | |
1910 | * RBD_SNAP_HEAD_NAME for the base image). | |
1911 | */ | |
dfc5606d YS |
1912 | static ssize_t rbd_snap_show(struct device *dev, |
1913 | struct device_attribute *attr, | |
1914 | char *buf) | |
1915 | { | |
593a9e7b | 1916 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1917 | |
f84344f3 | 1918 | return sprintf(buf, "%s\n", rbd_dev->mapping.snap_name); |
dfc5606d YS |
1919 | } |
1920 | ||
1921 | static ssize_t rbd_image_refresh(struct device *dev, | |
1922 | struct device_attribute *attr, | |
1923 | const char *buf, | |
1924 | size_t size) | |
1925 | { | |
593a9e7b | 1926 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 1927 | int ret; |
602adf40 | 1928 | |
117973fb | 1929 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
1930 | |
1931 | return ret < 0 ? ret : size; | |
dfc5606d | 1932 | } |
602adf40 | 1933 | |
dfc5606d | 1934 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 1935 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
1936 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
1937 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
1938 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 1939 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 1940 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 1941 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
1942 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
1943 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
dfc5606d YS |
1944 | |
1945 | static struct attribute *rbd_attrs[] = { | |
1946 | &dev_attr_size.attr, | |
34b13184 | 1947 | &dev_attr_features.attr, |
dfc5606d YS |
1948 | &dev_attr_major.attr, |
1949 | &dev_attr_client_id.attr, | |
1950 | &dev_attr_pool.attr, | |
9bb2f334 | 1951 | &dev_attr_pool_id.attr, |
dfc5606d | 1952 | &dev_attr_name.attr, |
589d30e0 | 1953 | &dev_attr_image_id.attr, |
dfc5606d YS |
1954 | &dev_attr_current_snap.attr, |
1955 | &dev_attr_refresh.attr, | |
dfc5606d YS |
1956 | NULL |
1957 | }; | |
1958 | ||
1959 | static struct attribute_group rbd_attr_group = { | |
1960 | .attrs = rbd_attrs, | |
1961 | }; | |
1962 | ||
1963 | static const struct attribute_group *rbd_attr_groups[] = { | |
1964 | &rbd_attr_group, | |
1965 | NULL | |
1966 | }; | |
1967 | ||
1968 | static void rbd_sysfs_dev_release(struct device *dev) | |
1969 | { | |
1970 | } | |
1971 | ||
1972 | static struct device_type rbd_device_type = { | |
1973 | .name = "rbd", | |
1974 | .groups = rbd_attr_groups, | |
1975 | .release = rbd_sysfs_dev_release, | |
1976 | }; | |
1977 | ||
1978 | ||
1979 | /* | |
1980 | sysfs - snapshots | |
1981 | */ | |
1982 | ||
1983 | static ssize_t rbd_snap_size_show(struct device *dev, | |
1984 | struct device_attribute *attr, | |
1985 | char *buf) | |
1986 | { | |
1987 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1988 | ||
3591538f | 1989 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
1990 | } |
1991 | ||
1992 | static ssize_t rbd_snap_id_show(struct device *dev, | |
1993 | struct device_attribute *attr, | |
1994 | char *buf) | |
1995 | { | |
1996 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1997 | ||
3591538f | 1998 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
1999 | } |
2000 | ||
34b13184 AE |
2001 | static ssize_t rbd_snap_features_show(struct device *dev, |
2002 | struct device_attribute *attr, | |
2003 | char *buf) | |
2004 | { | |
2005 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2006 | ||
2007 | return sprintf(buf, "0x%016llx\n", | |
2008 | (unsigned long long) snap->features); | |
2009 | } | |
2010 | ||
dfc5606d YS |
2011 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2012 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2013 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2014 | |
2015 | static struct attribute *rbd_snap_attrs[] = { | |
2016 | &dev_attr_snap_size.attr, | |
2017 | &dev_attr_snap_id.attr, | |
34b13184 | 2018 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2019 | NULL, |
2020 | }; | |
2021 | ||
2022 | static struct attribute_group rbd_snap_attr_group = { | |
2023 | .attrs = rbd_snap_attrs, | |
2024 | }; | |
2025 | ||
2026 | static void rbd_snap_dev_release(struct device *dev) | |
2027 | { | |
2028 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2029 | kfree(snap->name); | |
2030 | kfree(snap); | |
2031 | } | |
2032 | ||
2033 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2034 | &rbd_snap_attr_group, | |
2035 | NULL | |
2036 | }; | |
2037 | ||
2038 | static struct device_type rbd_snap_device_type = { | |
2039 | .groups = rbd_snap_attr_groups, | |
2040 | .release = rbd_snap_dev_release, | |
2041 | }; | |
2042 | ||
304f6808 AE |
2043 | static bool rbd_snap_registered(struct rbd_snap *snap) |
2044 | { | |
2045 | bool ret = snap->dev.type == &rbd_snap_device_type; | |
2046 | bool reg = device_is_registered(&snap->dev); | |
2047 | ||
2048 | rbd_assert(!ret ^ reg); | |
2049 | ||
2050 | return ret; | |
2051 | } | |
2052 | ||
14e7085d | 2053 | static void __rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2054 | { |
2055 | list_del(&snap->node); | |
304f6808 AE |
2056 | if (device_is_registered(&snap->dev)) |
2057 | device_unregister(&snap->dev); | |
dfc5606d YS |
2058 | } |
2059 | ||
14e7085d | 2060 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2061 | struct device *parent) |
2062 | { | |
2063 | struct device *dev = &snap->dev; | |
2064 | int ret; | |
2065 | ||
2066 | dev->type = &rbd_snap_device_type; | |
2067 | dev->parent = parent; | |
2068 | dev->release = rbd_snap_dev_release; | |
d4b125e9 | 2069 | dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name); |
304f6808 AE |
2070 | dout("%s: registering device for snapshot %s\n", __func__, snap->name); |
2071 | ||
dfc5606d YS |
2072 | ret = device_register(dev); |
2073 | ||
2074 | return ret; | |
2075 | } | |
2076 | ||
4e891e0a | 2077 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
c8d18425 | 2078 | const char *snap_name, |
34b13184 AE |
2079 | u64 snap_id, u64 snap_size, |
2080 | u64 snap_features) | |
dfc5606d | 2081 | { |
4e891e0a | 2082 | struct rbd_snap *snap; |
dfc5606d | 2083 | int ret; |
4e891e0a AE |
2084 | |
2085 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2086 | if (!snap) |
4e891e0a AE |
2087 | return ERR_PTR(-ENOMEM); |
2088 | ||
2089 | ret = -ENOMEM; | |
c8d18425 | 2090 | snap->name = kstrdup(snap_name, GFP_KERNEL); |
4e891e0a AE |
2091 | if (!snap->name) |
2092 | goto err; | |
2093 | ||
c8d18425 AE |
2094 | snap->id = snap_id; |
2095 | snap->size = snap_size; | |
34b13184 | 2096 | snap->features = snap_features; |
4e891e0a AE |
2097 | |
2098 | return snap; | |
2099 | ||
dfc5606d YS |
2100 | err: |
2101 | kfree(snap->name); | |
2102 | kfree(snap); | |
4e891e0a AE |
2103 | |
2104 | return ERR_PTR(ret); | |
dfc5606d YS |
2105 | } |
2106 | ||
cd892126 AE |
2107 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
2108 | u64 *snap_size, u64 *snap_features) | |
2109 | { | |
2110 | char *snap_name; | |
2111 | ||
2112 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
2113 | ||
2114 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
2115 | *snap_features = 0; /* No features for v1 */ | |
2116 | ||
2117 | /* Skip over names until we find the one we are looking for */ | |
2118 | ||
2119 | snap_name = rbd_dev->header.snap_names; | |
2120 | while (which--) | |
2121 | snap_name += strlen(snap_name) + 1; | |
2122 | ||
2123 | return snap_name; | |
2124 | } | |
2125 | ||
9d475de5 AE |
2126 | /* |
2127 | * Get the size and object order for an image snapshot, or if | |
2128 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
2129 | * image. | |
2130 | */ | |
2131 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
2132 | u8 *order, u64 *snap_size) | |
2133 | { | |
2134 | __le64 snapid = cpu_to_le64(snap_id); | |
2135 | int ret; | |
2136 | struct { | |
2137 | u8 order; | |
2138 | __le64 size; | |
2139 | } __attribute__ ((packed)) size_buf = { 0 }; | |
2140 | ||
2141 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2142 | "rbd", "get_size", | |
2143 | (char *) &snapid, sizeof (snapid), | |
2144 | (char *) &size_buf, sizeof (size_buf), | |
2145 | CEPH_OSD_FLAG_READ, NULL); | |
2146 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2147 | if (ret < 0) | |
2148 | return ret; | |
2149 | ||
2150 | *order = size_buf.order; | |
2151 | *snap_size = le64_to_cpu(size_buf.size); | |
2152 | ||
2153 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2154 | (unsigned long long) snap_id, (unsigned int) *order, | |
2155 | (unsigned long long) *snap_size); | |
2156 | ||
2157 | return 0; | |
2158 | } | |
2159 | ||
2160 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2161 | { | |
2162 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2163 | &rbd_dev->header.obj_order, | |
2164 | &rbd_dev->header.image_size); | |
2165 | } | |
2166 | ||
1e130199 AE |
2167 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2168 | { | |
2169 | void *reply_buf; | |
2170 | int ret; | |
2171 | void *p; | |
2172 | ||
2173 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2174 | if (!reply_buf) | |
2175 | return -ENOMEM; | |
2176 | ||
2177 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2178 | "rbd", "get_object_prefix", | |
2179 | NULL, 0, | |
2180 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, | |
2181 | CEPH_OSD_FLAG_READ, NULL); | |
2182 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2183 | if (ret < 0) | |
2184 | goto out; | |
a0ea3a40 | 2185 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
1e130199 AE |
2186 | |
2187 | p = reply_buf; | |
2188 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2189 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2190 | NULL, GFP_NOIO); | |
2191 | ||
2192 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2193 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2194 | rbd_dev->header.object_prefix = NULL; | |
2195 | } else { | |
2196 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2197 | } | |
2198 | ||
2199 | out: | |
2200 | kfree(reply_buf); | |
2201 | ||
2202 | return ret; | |
2203 | } | |
2204 | ||
b1b5402a AE |
2205 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2206 | u64 *snap_features) | |
2207 | { | |
2208 | __le64 snapid = cpu_to_le64(snap_id); | |
2209 | struct { | |
2210 | __le64 features; | |
2211 | __le64 incompat; | |
2212 | } features_buf = { 0 }; | |
d889140c | 2213 | u64 incompat; |
b1b5402a AE |
2214 | int ret; |
2215 | ||
2216 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2217 | "rbd", "get_features", | |
2218 | (char *) &snapid, sizeof (snapid), | |
2219 | (char *) &features_buf, sizeof (features_buf), | |
2220 | CEPH_OSD_FLAG_READ, NULL); | |
2221 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2222 | if (ret < 0) | |
2223 | return ret; | |
d889140c AE |
2224 | |
2225 | incompat = le64_to_cpu(features_buf.incompat); | |
2226 | if (incompat & ~RBD_FEATURES_ALL) | |
2227 | return -ENOTSUPP; | |
2228 | ||
b1b5402a AE |
2229 | *snap_features = le64_to_cpu(features_buf.features); |
2230 | ||
2231 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2232 | (unsigned long long) snap_id, | |
2233 | (unsigned long long) *snap_features, | |
2234 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2235 | ||
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2240 | { | |
2241 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2242 | &rbd_dev->header.features); | |
2243 | } | |
2244 | ||
6e14b1a6 | 2245 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
2246 | { |
2247 | size_t size; | |
2248 | int ret; | |
2249 | void *reply_buf; | |
2250 | void *p; | |
2251 | void *end; | |
2252 | u64 seq; | |
2253 | u32 snap_count; | |
2254 | struct ceph_snap_context *snapc; | |
2255 | u32 i; | |
2256 | ||
2257 | /* | |
2258 | * We'll need room for the seq value (maximum snapshot id), | |
2259 | * snapshot count, and array of that many snapshot ids. | |
2260 | * For now we have a fixed upper limit on the number we're | |
2261 | * prepared to receive. | |
2262 | */ | |
2263 | size = sizeof (__le64) + sizeof (__le32) + | |
2264 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
2265 | reply_buf = kzalloc(size, GFP_KERNEL); | |
2266 | if (!reply_buf) | |
2267 | return -ENOMEM; | |
2268 | ||
2269 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2270 | "rbd", "get_snapcontext", | |
2271 | NULL, 0, | |
2272 | reply_buf, size, | |
6e14b1a6 | 2273 | CEPH_OSD_FLAG_READ, ver); |
35d489f9 AE |
2274 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2275 | if (ret < 0) | |
2276 | goto out; | |
2277 | ||
2278 | ret = -ERANGE; | |
2279 | p = reply_buf; | |
2280 | end = (char *) reply_buf + size; | |
2281 | ceph_decode_64_safe(&p, end, seq, out); | |
2282 | ceph_decode_32_safe(&p, end, snap_count, out); | |
2283 | ||
2284 | /* | |
2285 | * Make sure the reported number of snapshot ids wouldn't go | |
2286 | * beyond the end of our buffer. But before checking that, | |
2287 | * make sure the computed size of the snapshot context we | |
2288 | * allocate is representable in a size_t. | |
2289 | */ | |
2290 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
2291 | / sizeof (u64)) { | |
2292 | ret = -EINVAL; | |
2293 | goto out; | |
2294 | } | |
2295 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
2296 | goto out; | |
2297 | ||
2298 | size = sizeof (struct ceph_snap_context) + | |
2299 | snap_count * sizeof (snapc->snaps[0]); | |
2300 | snapc = kmalloc(size, GFP_KERNEL); | |
2301 | if (!snapc) { | |
2302 | ret = -ENOMEM; | |
2303 | goto out; | |
2304 | } | |
2305 | ||
2306 | atomic_set(&snapc->nref, 1); | |
2307 | snapc->seq = seq; | |
2308 | snapc->num_snaps = snap_count; | |
2309 | for (i = 0; i < snap_count; i++) | |
2310 | snapc->snaps[i] = ceph_decode_64(&p); | |
2311 | ||
2312 | rbd_dev->header.snapc = snapc; | |
2313 | ||
2314 | dout(" snap context seq = %llu, snap_count = %u\n", | |
2315 | (unsigned long long) seq, (unsigned int) snap_count); | |
2316 | ||
2317 | out: | |
2318 | kfree(reply_buf); | |
2319 | ||
2320 | return 0; | |
2321 | } | |
2322 | ||
b8b1e2db AE |
2323 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
2324 | { | |
2325 | size_t size; | |
2326 | void *reply_buf; | |
2327 | __le64 snap_id; | |
2328 | int ret; | |
2329 | void *p; | |
2330 | void *end; | |
2331 | size_t snap_name_len; | |
2332 | char *snap_name; | |
2333 | ||
2334 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
2335 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2336 | if (!reply_buf) | |
2337 | return ERR_PTR(-ENOMEM); | |
2338 | ||
2339 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
2340 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2341 | "rbd", "get_snapshot_name", | |
2342 | (char *) &snap_id, sizeof (snap_id), | |
2343 | reply_buf, size, | |
2344 | CEPH_OSD_FLAG_READ, NULL); | |
2345 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2346 | if (ret < 0) | |
2347 | goto out; | |
2348 | ||
2349 | p = reply_buf; | |
2350 | end = (char *) reply_buf + size; | |
2351 | snap_name_len = 0; | |
2352 | snap_name = ceph_extract_encoded_string(&p, end, &snap_name_len, | |
2353 | GFP_KERNEL); | |
2354 | if (IS_ERR(snap_name)) { | |
2355 | ret = PTR_ERR(snap_name); | |
2356 | goto out; | |
2357 | } else { | |
2358 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
2359 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
2360 | } | |
2361 | kfree(reply_buf); | |
2362 | ||
2363 | return snap_name; | |
2364 | out: | |
2365 | kfree(reply_buf); | |
2366 | ||
2367 | return ERR_PTR(ret); | |
2368 | } | |
2369 | ||
2370 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2371 | u64 *snap_size, u64 *snap_features) | |
2372 | { | |
2373 | __le64 snap_id; | |
2374 | u8 order; | |
2375 | int ret; | |
2376 | ||
2377 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
2378 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
2379 | if (ret) | |
2380 | return ERR_PTR(ret); | |
2381 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
2382 | if (ret) | |
2383 | return ERR_PTR(ret); | |
2384 | ||
2385 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
2386 | } | |
2387 | ||
2388 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2389 | u64 *snap_size, u64 *snap_features) | |
2390 | { | |
2391 | if (rbd_dev->image_format == 1) | |
2392 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
2393 | snap_size, snap_features); | |
2394 | if (rbd_dev->image_format == 2) | |
2395 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
2396 | snap_size, snap_features); | |
2397 | return ERR_PTR(-EINVAL); | |
2398 | } | |
2399 | ||
117973fb AE |
2400 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
2401 | { | |
2402 | int ret; | |
2403 | __u8 obj_order; | |
2404 | ||
2405 | down_write(&rbd_dev->header_rwsem); | |
2406 | ||
2407 | /* Grab old order first, to see if it changes */ | |
2408 | ||
2409 | obj_order = rbd_dev->header.obj_order, | |
2410 | ret = rbd_dev_v2_image_size(rbd_dev); | |
2411 | if (ret) | |
2412 | goto out; | |
2413 | if (rbd_dev->header.obj_order != obj_order) { | |
2414 | ret = -EIO; | |
2415 | goto out; | |
2416 | } | |
2417 | rbd_update_mapping_size(rbd_dev); | |
2418 | ||
2419 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
2420 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
2421 | if (ret) | |
2422 | goto out; | |
2423 | ret = rbd_dev_snaps_update(rbd_dev); | |
2424 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
2425 | if (ret) | |
2426 | goto out; | |
2427 | ret = rbd_dev_snaps_register(rbd_dev); | |
2428 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
2429 | out: | |
2430 | up_write(&rbd_dev->header_rwsem); | |
2431 | ||
2432 | return ret; | |
2433 | } | |
2434 | ||
dfc5606d | 2435 | /* |
35938150 AE |
2436 | * Scan the rbd device's current snapshot list and compare it to the |
2437 | * newly-received snapshot context. Remove any existing snapshots | |
2438 | * not present in the new snapshot context. Add a new snapshot for | |
2439 | * any snaphots in the snapshot context not in the current list. | |
2440 | * And verify there are no changes to snapshots we already know | |
2441 | * about. | |
2442 | * | |
2443 | * Assumes the snapshots in the snapshot context are sorted by | |
2444 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
2445 | * are also maintained in that order.) | |
dfc5606d | 2446 | */ |
304f6808 | 2447 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 2448 | { |
35938150 AE |
2449 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
2450 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
2451 | struct list_head *head = &rbd_dev->snaps; |
2452 | struct list_head *links = head->next; | |
2453 | u32 index = 0; | |
dfc5606d | 2454 | |
9fcbb800 | 2455 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
2456 | while (index < snap_count || links != head) { |
2457 | u64 snap_id; | |
2458 | struct rbd_snap *snap; | |
cd892126 AE |
2459 | char *snap_name; |
2460 | u64 snap_size = 0; | |
2461 | u64 snap_features = 0; | |
dfc5606d | 2462 | |
35938150 AE |
2463 | snap_id = index < snap_count ? snapc->snaps[index] |
2464 | : CEPH_NOSNAP; | |
2465 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
2466 | : NULL; | |
aafb230e | 2467 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 2468 | |
35938150 AE |
2469 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
2470 | struct list_head *next = links->next; | |
dfc5606d | 2471 | |
35938150 | 2472 | /* Existing snapshot not in the new snap context */ |
dfc5606d | 2473 | |
f84344f3 AE |
2474 | if (rbd_dev->mapping.snap_id == snap->id) |
2475 | rbd_dev->mapping.snap_exists = false; | |
35938150 | 2476 | __rbd_remove_snap_dev(snap); |
9fcbb800 | 2477 | dout("%ssnap id %llu has been removed\n", |
f84344f3 AE |
2478 | rbd_dev->mapping.snap_id == snap->id ? |
2479 | "mapped " : "", | |
9fcbb800 | 2480 | (unsigned long long) snap->id); |
35938150 AE |
2481 | |
2482 | /* Done with this list entry; advance */ | |
2483 | ||
2484 | links = next; | |
dfc5606d YS |
2485 | continue; |
2486 | } | |
35938150 | 2487 | |
b8b1e2db AE |
2488 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
2489 | &snap_size, &snap_features); | |
cd892126 AE |
2490 | if (IS_ERR(snap_name)) |
2491 | return PTR_ERR(snap_name); | |
2492 | ||
9fcbb800 AE |
2493 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
2494 | (unsigned long long) snap_id); | |
35938150 AE |
2495 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
2496 | struct rbd_snap *new_snap; | |
2497 | ||
2498 | /* We haven't seen this snapshot before */ | |
2499 | ||
c8d18425 | 2500 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 2501 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
2502 | if (IS_ERR(new_snap)) { |
2503 | int err = PTR_ERR(new_snap); | |
2504 | ||
2505 | dout(" failed to add dev, error %d\n", err); | |
2506 | ||
2507 | return err; | |
2508 | } | |
35938150 AE |
2509 | |
2510 | /* New goes before existing, or at end of list */ | |
2511 | ||
9fcbb800 | 2512 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
2513 | if (snap) |
2514 | list_add_tail(&new_snap->node, &snap->node); | |
2515 | else | |
523f3258 | 2516 | list_add_tail(&new_snap->node, head); |
35938150 AE |
2517 | } else { |
2518 | /* Already have this one */ | |
2519 | ||
9fcbb800 AE |
2520 | dout(" already present\n"); |
2521 | ||
cd892126 | 2522 | rbd_assert(snap->size == snap_size); |
aafb230e | 2523 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 2524 | rbd_assert(snap->features == snap_features); |
35938150 AE |
2525 | |
2526 | /* Done with this list entry; advance */ | |
2527 | ||
2528 | links = links->next; | |
dfc5606d | 2529 | } |
35938150 AE |
2530 | |
2531 | /* Advance to the next entry in the snapshot context */ | |
2532 | ||
2533 | index++; | |
dfc5606d | 2534 | } |
9fcbb800 | 2535 | dout("%s: done\n", __func__); |
dfc5606d YS |
2536 | |
2537 | return 0; | |
2538 | } | |
2539 | ||
304f6808 AE |
2540 | /* |
2541 | * Scan the list of snapshots and register the devices for any that | |
2542 | * have not already been registered. | |
2543 | */ | |
2544 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
2545 | { | |
2546 | struct rbd_snap *snap; | |
2547 | int ret = 0; | |
2548 | ||
2549 | dout("%s called\n", __func__); | |
86ff77bb AE |
2550 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
2551 | return -EIO; | |
304f6808 AE |
2552 | |
2553 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
2554 | if (!rbd_snap_registered(snap)) { | |
2555 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
2556 | if (ret < 0) | |
2557 | break; | |
2558 | } | |
2559 | } | |
2560 | dout("%s: returning %d\n", __func__, ret); | |
2561 | ||
2562 | return ret; | |
2563 | } | |
2564 | ||
dfc5606d YS |
2565 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
2566 | { | |
dfc5606d | 2567 | struct device *dev; |
cd789ab9 | 2568 | int ret; |
dfc5606d YS |
2569 | |
2570 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 2571 | |
cd789ab9 | 2572 | dev = &rbd_dev->dev; |
dfc5606d YS |
2573 | dev->bus = &rbd_bus_type; |
2574 | dev->type = &rbd_device_type; | |
2575 | dev->parent = &rbd_root_dev; | |
2576 | dev->release = rbd_dev_release; | |
de71a297 | 2577 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 2578 | ret = device_register(dev); |
dfc5606d | 2579 | |
dfc5606d | 2580 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 2581 | |
dfc5606d | 2582 | return ret; |
602adf40 YS |
2583 | } |
2584 | ||
dfc5606d YS |
2585 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
2586 | { | |
2587 | device_unregister(&rbd_dev->dev); | |
2588 | } | |
2589 | ||
59c2be1e YS |
2590 | static int rbd_init_watch_dev(struct rbd_device *rbd_dev) |
2591 | { | |
2592 | int ret, rc; | |
2593 | ||
2594 | do { | |
0e6f322d | 2595 | ret = rbd_req_sync_watch(rbd_dev); |
59c2be1e | 2596 | if (ret == -ERANGE) { |
117973fb | 2597 | rc = rbd_dev_refresh(rbd_dev, NULL); |
59c2be1e YS |
2598 | if (rc < 0) |
2599 | return rc; | |
2600 | } | |
2601 | } while (ret == -ERANGE); | |
2602 | ||
2603 | return ret; | |
2604 | } | |
2605 | ||
e2839308 | 2606 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
2607 | |
2608 | /* | |
499afd5b AE |
2609 | * Get a unique rbd identifier for the given new rbd_dev, and add |
2610 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 2611 | */ |
e2839308 | 2612 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 2613 | { |
e2839308 | 2614 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
2615 | |
2616 | spin_lock(&rbd_dev_list_lock); | |
2617 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
2618 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
2619 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
2620 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 2621 | } |
b7f23c36 | 2622 | |
1ddbe94e | 2623 | /* |
499afd5b AE |
2624 | * Remove an rbd_dev from the global list, and record that its |
2625 | * identifier is no longer in use. | |
1ddbe94e | 2626 | */ |
e2839308 | 2627 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 2628 | { |
d184f6bf | 2629 | struct list_head *tmp; |
de71a297 | 2630 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
2631 | int max_id; |
2632 | ||
aafb230e | 2633 | rbd_assert(rbd_id > 0); |
499afd5b | 2634 | |
e2839308 AE |
2635 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
2636 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
2637 | spin_lock(&rbd_dev_list_lock); |
2638 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
2639 | |
2640 | /* | |
2641 | * If the id being "put" is not the current maximum, there | |
2642 | * is nothing special we need to do. | |
2643 | */ | |
e2839308 | 2644 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
2645 | spin_unlock(&rbd_dev_list_lock); |
2646 | return; | |
2647 | } | |
2648 | ||
2649 | /* | |
2650 | * We need to update the current maximum id. Search the | |
2651 | * list to find out what it is. We're more likely to find | |
2652 | * the maximum at the end, so search the list backward. | |
2653 | */ | |
2654 | max_id = 0; | |
2655 | list_for_each_prev(tmp, &rbd_dev_list) { | |
2656 | struct rbd_device *rbd_dev; | |
2657 | ||
2658 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
2659 | if (rbd_dev->dev_id > max_id) |
2660 | max_id = rbd_dev->dev_id; | |
d184f6bf | 2661 | } |
499afd5b | 2662 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 2663 | |
1ddbe94e | 2664 | /* |
e2839308 | 2665 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
2666 | * which case it now accurately reflects the new maximum. |
2667 | * Be careful not to overwrite the maximum value in that | |
2668 | * case. | |
1ddbe94e | 2669 | */ |
e2839308 AE |
2670 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
2671 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
2672 | } |
2673 | ||
e28fff26 AE |
2674 | /* |
2675 | * Skips over white space at *buf, and updates *buf to point to the | |
2676 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
2677 | * the token (string of non-white space characters) found. Note |
2678 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
2679 | */ |
2680 | static inline size_t next_token(const char **buf) | |
2681 | { | |
2682 | /* | |
2683 | * These are the characters that produce nonzero for | |
2684 | * isspace() in the "C" and "POSIX" locales. | |
2685 | */ | |
2686 | const char *spaces = " \f\n\r\t\v"; | |
2687 | ||
2688 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
2689 | ||
2690 | return strcspn(*buf, spaces); /* Return token length */ | |
2691 | } | |
2692 | ||
2693 | /* | |
2694 | * Finds the next token in *buf, and if the provided token buffer is | |
2695 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
2696 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
2697 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
2698 | * |
2699 | * Returns the length of the token found (not including the '\0'). | |
2700 | * Return value will be 0 if no token is found, and it will be >= | |
2701 | * token_size if the token would not fit. | |
2702 | * | |
593a9e7b | 2703 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
2704 | * found token. Note that this occurs even if the token buffer is |
2705 | * too small to hold it. | |
2706 | */ | |
2707 | static inline size_t copy_token(const char **buf, | |
2708 | char *token, | |
2709 | size_t token_size) | |
2710 | { | |
2711 | size_t len; | |
2712 | ||
2713 | len = next_token(buf); | |
2714 | if (len < token_size) { | |
2715 | memcpy(token, *buf, len); | |
2716 | *(token + len) = '\0'; | |
2717 | } | |
2718 | *buf += len; | |
2719 | ||
2720 | return len; | |
2721 | } | |
2722 | ||
ea3352f4 AE |
2723 | /* |
2724 | * Finds the next token in *buf, dynamically allocates a buffer big | |
2725 | * enough to hold a copy of it, and copies the token into the new | |
2726 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
2727 | * that a duplicate buffer is created even for a zero-length token. | |
2728 | * | |
2729 | * Returns a pointer to the newly-allocated duplicate, or a null | |
2730 | * pointer if memory for the duplicate was not available. If | |
2731 | * the lenp argument is a non-null pointer, the length of the token | |
2732 | * (not including the '\0') is returned in *lenp. | |
2733 | * | |
2734 | * If successful, the *buf pointer will be updated to point beyond | |
2735 | * the end of the found token. | |
2736 | * | |
2737 | * Note: uses GFP_KERNEL for allocation. | |
2738 | */ | |
2739 | static inline char *dup_token(const char **buf, size_t *lenp) | |
2740 | { | |
2741 | char *dup; | |
2742 | size_t len; | |
2743 | ||
2744 | len = next_token(buf); | |
2745 | dup = kmalloc(len + 1, GFP_KERNEL); | |
2746 | if (!dup) | |
2747 | return NULL; | |
2748 | ||
2749 | memcpy(dup, *buf, len); | |
2750 | *(dup + len) = '\0'; | |
2751 | *buf += len; | |
2752 | ||
2753 | if (lenp) | |
2754 | *lenp = len; | |
2755 | ||
2756 | return dup; | |
2757 | } | |
2758 | ||
a725f65e | 2759 | /* |
3feeb894 AE |
2760 | * This fills in the pool_name, image_name, image_name_len, rbd_dev, |
2761 | * rbd_md_name, and name fields of the given rbd_dev, based on the | |
2762 | * list of monitor addresses and other options provided via | |
2763 | * /sys/bus/rbd/add. Returns a pointer to a dynamically-allocated | |
2764 | * copy of the snapshot name to map if successful, or a | |
2765 | * pointer-coded error otherwise. | |
d22f76e7 AE |
2766 | * |
2767 | * Note: rbd_dev is assumed to have been initially zero-filled. | |
a725f65e | 2768 | */ |
3feeb894 AE |
2769 | static char *rbd_add_parse_args(struct rbd_device *rbd_dev, |
2770 | const char *buf, | |
2771 | const char **mon_addrs, | |
2772 | size_t *mon_addrs_size, | |
2773 | char *options, | |
2774 | size_t options_size) | |
e28fff26 | 2775 | { |
d22f76e7 | 2776 | size_t len; |
3feeb894 AE |
2777 | char *err_ptr = ERR_PTR(-EINVAL); |
2778 | char *snap_name; | |
e28fff26 AE |
2779 | |
2780 | /* The first four tokens are required */ | |
2781 | ||
7ef3214a AE |
2782 | len = next_token(&buf); |
2783 | if (!len) | |
3feeb894 | 2784 | return err_ptr; |
5214ecc4 | 2785 | *mon_addrs_size = len + 1; |
7ef3214a AE |
2786 | *mon_addrs = buf; |
2787 | ||
2788 | buf += len; | |
a725f65e | 2789 | |
e28fff26 AE |
2790 | len = copy_token(&buf, options, options_size); |
2791 | if (!len || len >= options_size) | |
3feeb894 | 2792 | return err_ptr; |
e28fff26 | 2793 | |
3feeb894 | 2794 | err_ptr = ERR_PTR(-ENOMEM); |
d22f76e7 AE |
2795 | rbd_dev->pool_name = dup_token(&buf, NULL); |
2796 | if (!rbd_dev->pool_name) | |
d22f76e7 | 2797 | goto out_err; |
e28fff26 | 2798 | |
0bed54dc AE |
2799 | rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len); |
2800 | if (!rbd_dev->image_name) | |
bf3e5ae1 | 2801 | goto out_err; |
a725f65e | 2802 | |
d4b125e9 AE |
2803 | /* Snapshot name is optional; default is to use "head" */ |
2804 | ||
3feeb894 | 2805 | len = next_token(&buf); |
d4b125e9 AE |
2806 | if (len > RBD_MAX_SNAP_NAME_LEN) { |
2807 | err_ptr = ERR_PTR(-ENAMETOOLONG); | |
2808 | goto out_err; | |
2809 | } | |
820a5f3e | 2810 | if (!len) { |
3feeb894 AE |
2811 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
2812 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
849b4260 | 2813 | } |
3feeb894 AE |
2814 | snap_name = kmalloc(len + 1, GFP_KERNEL); |
2815 | if (!snap_name) | |
2816 | goto out_err; | |
2817 | memcpy(snap_name, buf, len); | |
2818 | *(snap_name + len) = '\0'; | |
e28fff26 | 2819 | |
3feeb894 | 2820 | return snap_name; |
d22f76e7 AE |
2821 | |
2822 | out_err: | |
0bed54dc | 2823 | kfree(rbd_dev->image_name); |
d78fd7ae AE |
2824 | rbd_dev->image_name = NULL; |
2825 | rbd_dev->image_name_len = 0; | |
d22f76e7 AE |
2826 | kfree(rbd_dev->pool_name); |
2827 | rbd_dev->pool_name = NULL; | |
2828 | ||
3feeb894 | 2829 | return err_ptr; |
a725f65e AE |
2830 | } |
2831 | ||
589d30e0 AE |
2832 | /* |
2833 | * An rbd format 2 image has a unique identifier, distinct from the | |
2834 | * name given to it by the user. Internally, that identifier is | |
2835 | * what's used to specify the names of objects related to the image. | |
2836 | * | |
2837 | * A special "rbd id" object is used to map an rbd image name to its | |
2838 | * id. If that object doesn't exist, then there is no v2 rbd image | |
2839 | * with the supplied name. | |
2840 | * | |
2841 | * This function will record the given rbd_dev's image_id field if | |
2842 | * it can be determined, and in that case will return 0. If any | |
2843 | * errors occur a negative errno will be returned and the rbd_dev's | |
2844 | * image_id field will be unchanged (and should be NULL). | |
2845 | */ | |
2846 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
2847 | { | |
2848 | int ret; | |
2849 | size_t size; | |
2850 | char *object_name; | |
2851 | void *response; | |
2852 | void *p; | |
2853 | ||
2854 | /* | |
2855 | * First, see if the format 2 image id file exists, and if | |
2856 | * so, get the image's persistent id from it. | |
2857 | */ | |
2858 | size = sizeof (RBD_ID_PREFIX) + rbd_dev->image_name_len; | |
2859 | object_name = kmalloc(size, GFP_NOIO); | |
2860 | if (!object_name) | |
2861 | return -ENOMEM; | |
2862 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->image_name); | |
2863 | dout("rbd id object name is %s\n", object_name); | |
2864 | ||
2865 | /* Response will be an encoded string, which includes a length */ | |
2866 | ||
2867 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
2868 | response = kzalloc(size, GFP_NOIO); | |
2869 | if (!response) { | |
2870 | ret = -ENOMEM; | |
2871 | goto out; | |
2872 | } | |
2873 | ||
2874 | ret = rbd_req_sync_exec(rbd_dev, object_name, | |
2875 | "rbd", "get_id", | |
2876 | NULL, 0, | |
2877 | response, RBD_IMAGE_ID_LEN_MAX, | |
2878 | CEPH_OSD_FLAG_READ, NULL); | |
2879 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2880 | if (ret < 0) | |
2881 | goto out; | |
a0ea3a40 | 2882 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
589d30e0 AE |
2883 | |
2884 | p = response; | |
2885 | rbd_dev->image_id = ceph_extract_encoded_string(&p, | |
2886 | p + RBD_IMAGE_ID_LEN_MAX, | |
2887 | &rbd_dev->image_id_len, | |
2888 | GFP_NOIO); | |
2889 | if (IS_ERR(rbd_dev->image_id)) { | |
2890 | ret = PTR_ERR(rbd_dev->image_id); | |
2891 | rbd_dev->image_id = NULL; | |
2892 | } else { | |
2893 | dout("image_id is %s\n", rbd_dev->image_id); | |
2894 | } | |
2895 | out: | |
2896 | kfree(response); | |
2897 | kfree(object_name); | |
2898 | ||
2899 | return ret; | |
2900 | } | |
2901 | ||
a30b71b9 AE |
2902 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
2903 | { | |
2904 | int ret; | |
2905 | size_t size; | |
2906 | ||
2907 | /* Version 1 images have no id; empty string is used */ | |
2908 | ||
2909 | rbd_dev->image_id = kstrdup("", GFP_KERNEL); | |
2910 | if (!rbd_dev->image_id) | |
2911 | return -ENOMEM; | |
2912 | rbd_dev->image_id_len = 0; | |
2913 | ||
2914 | /* Record the header object name for this rbd image. */ | |
2915 | ||
2916 | size = rbd_dev->image_name_len + sizeof (RBD_SUFFIX); | |
2917 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
2918 | if (!rbd_dev->header_name) { | |
2919 | ret = -ENOMEM; | |
2920 | goto out_err; | |
2921 | } | |
2922 | sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX); | |
2923 | ||
2924 | /* Populate rbd image metadata */ | |
2925 | ||
2926 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
2927 | if (ret < 0) | |
2928 | goto out_err; | |
2929 | rbd_dev->image_format = 1; | |
2930 | ||
2931 | dout("discovered version 1 image, header name is %s\n", | |
2932 | rbd_dev->header_name); | |
2933 | ||
2934 | return 0; | |
2935 | ||
2936 | out_err: | |
2937 | kfree(rbd_dev->header_name); | |
2938 | rbd_dev->header_name = NULL; | |
2939 | kfree(rbd_dev->image_id); | |
2940 | rbd_dev->image_id = NULL; | |
2941 | ||
2942 | return ret; | |
2943 | } | |
2944 | ||
2945 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
2946 | { | |
2947 | size_t size; | |
9d475de5 | 2948 | int ret; |
6e14b1a6 | 2949 | u64 ver = 0; |
a30b71b9 AE |
2950 | |
2951 | /* | |
2952 | * Image id was filled in by the caller. Record the header | |
2953 | * object name for this rbd image. | |
2954 | */ | |
2955 | size = sizeof (RBD_HEADER_PREFIX) + rbd_dev->image_id_len; | |
2956 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
2957 | if (!rbd_dev->header_name) | |
2958 | return -ENOMEM; | |
2959 | sprintf(rbd_dev->header_name, "%s%s", | |
2960 | RBD_HEADER_PREFIX, rbd_dev->image_id); | |
9d475de5 AE |
2961 | |
2962 | /* Get the size and object order for the image */ | |
2963 | ||
2964 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
2965 | if (ret < 0) |
2966 | goto out_err; | |
2967 | ||
2968 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
2969 | ||
2970 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
2971 | if (ret < 0) |
2972 | goto out_err; | |
2973 | ||
d889140c | 2974 | /* Get the and check features for the image */ |
b1b5402a AE |
2975 | |
2976 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
2977 | if (ret < 0) |
2978 | goto out_err; | |
35d489f9 | 2979 | |
6e14b1a6 AE |
2980 | /* crypto and compression type aren't (yet) supported for v2 images */ |
2981 | ||
2982 | rbd_dev->header.crypt_type = 0; | |
2983 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 2984 | |
6e14b1a6 AE |
2985 | /* Get the snapshot context, plus the header version */ |
2986 | ||
2987 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
2988 | if (ret) |
2989 | goto out_err; | |
6e14b1a6 AE |
2990 | rbd_dev->header.obj_version = ver; |
2991 | ||
a30b71b9 AE |
2992 | rbd_dev->image_format = 2; |
2993 | ||
2994 | dout("discovered version 2 image, header name is %s\n", | |
2995 | rbd_dev->header_name); | |
2996 | ||
35152979 | 2997 | return 0; |
9d475de5 AE |
2998 | out_err: |
2999 | kfree(rbd_dev->header_name); | |
3000 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3001 | kfree(rbd_dev->header.object_prefix); |
3002 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3003 | |
3004 | return ret; | |
a30b71b9 AE |
3005 | } |
3006 | ||
3007 | /* | |
3008 | * Probe for the existence of the header object for the given rbd | |
3009 | * device. For format 2 images this includes determining the image | |
3010 | * id. | |
3011 | */ | |
3012 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
3013 | { | |
3014 | int ret; | |
3015 | ||
3016 | /* | |
3017 | * Get the id from the image id object. If it's not a | |
3018 | * format 2 image, we'll get ENOENT back, and we'll assume | |
3019 | * it's a format 1 image. | |
3020 | */ | |
3021 | ret = rbd_dev_image_id(rbd_dev); | |
3022 | if (ret) | |
3023 | ret = rbd_dev_v1_probe(rbd_dev); | |
3024 | else | |
3025 | ret = rbd_dev_v2_probe(rbd_dev); | |
3026 | if (ret) | |
3027 | dout("probe failed, returning %d\n", ret); | |
3028 | ||
3029 | return ret; | |
3030 | } | |
3031 | ||
59c2be1e YS |
3032 | static ssize_t rbd_add(struct bus_type *bus, |
3033 | const char *buf, | |
3034 | size_t count) | |
602adf40 | 3035 | { |
cb8627c7 AE |
3036 | char *options; |
3037 | struct rbd_device *rbd_dev = NULL; | |
7ef3214a AE |
3038 | const char *mon_addrs = NULL; |
3039 | size_t mon_addrs_size = 0; | |
27cc2594 AE |
3040 | struct ceph_osd_client *osdc; |
3041 | int rc = -ENOMEM; | |
3feeb894 | 3042 | char *snap_name; |
602adf40 YS |
3043 | |
3044 | if (!try_module_get(THIS_MODULE)) | |
3045 | return -ENODEV; | |
3046 | ||
60571c7d | 3047 | options = kmalloc(count, GFP_KERNEL); |
602adf40 | 3048 | if (!options) |
85ae8926 | 3049 | goto err_out_mem; |
cb8627c7 AE |
3050 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
3051 | if (!rbd_dev) | |
85ae8926 | 3052 | goto err_out_mem; |
602adf40 YS |
3053 | |
3054 | /* static rbd_device initialization */ | |
3055 | spin_lock_init(&rbd_dev->lock); | |
3056 | INIT_LIST_HEAD(&rbd_dev->node); | |
dfc5606d | 3057 | INIT_LIST_HEAD(&rbd_dev->snaps); |
c666601a | 3058 | init_rwsem(&rbd_dev->header_rwsem); |
602adf40 | 3059 | |
602adf40 | 3060 | /* parse add command */ |
3feeb894 AE |
3061 | snap_name = rbd_add_parse_args(rbd_dev, buf, |
3062 | &mon_addrs, &mon_addrs_size, options, count); | |
3063 | if (IS_ERR(snap_name)) { | |
3064 | rc = PTR_ERR(snap_name); | |
85ae8926 | 3065 | goto err_out_mem; |
3feeb894 | 3066 | } |
e124a82f | 3067 | |
f8c38929 AE |
3068 | rc = rbd_get_client(rbd_dev, mon_addrs, mon_addrs_size - 1, options); |
3069 | if (rc < 0) | |
85ae8926 | 3070 | goto err_out_args; |
602adf40 | 3071 | |
602adf40 | 3072 | /* pick the pool */ |
1dbb4399 | 3073 | osdc = &rbd_dev->rbd_client->client->osdc; |
602adf40 YS |
3074 | rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name); |
3075 | if (rc < 0) | |
3076 | goto err_out_client; | |
9bb2f334 | 3077 | rbd_dev->pool_id = rc; |
602adf40 | 3078 | |
a30b71b9 AE |
3079 | rc = rbd_dev_probe(rbd_dev); |
3080 | if (rc < 0) | |
05fd6f6f AE |
3081 | goto err_out_client; |
3082 | ||
3083 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3084 | rc = rbd_dev_snaps_update(rbd_dev); | |
3085 | if (rc) | |
3086 | goto err_out_header; | |
3087 | ||
3088 | rc = rbd_dev_set_mapping(rbd_dev, snap_name); | |
3089 | if (rc) | |
3090 | goto err_out_header; | |
3091 | ||
85ae8926 AE |
3092 | /* generate unique id: find highest unique id, add one */ |
3093 | rbd_dev_id_get(rbd_dev); | |
3094 | ||
3095 | /* Fill in the device name, now that we have its id. */ | |
3096 | BUILD_BUG_ON(DEV_NAME_LEN | |
3097 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3098 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3099 | ||
3100 | /* Get our block major device number. */ | |
3101 | ||
27cc2594 AE |
3102 | rc = register_blkdev(0, rbd_dev->name); |
3103 | if (rc < 0) | |
85ae8926 | 3104 | goto err_out_id; |
27cc2594 | 3105 | rbd_dev->major = rc; |
602adf40 | 3106 | |
0f308a31 AE |
3107 | /* Set up the blkdev mapping. */ |
3108 | ||
3109 | rc = rbd_init_disk(rbd_dev); | |
dfc5606d | 3110 | if (rc) |
766fc439 YS |
3111 | goto err_out_blkdev; |
3112 | ||
0f308a31 AE |
3113 | rc = rbd_bus_add_dev(rbd_dev); |
3114 | if (rc) | |
3115 | goto err_out_disk; | |
3116 | ||
32eec68d AE |
3117 | /* |
3118 | * At this point cleanup in the event of an error is the job | |
3119 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
32eec68d | 3120 | */ |
2ac4e75d | 3121 | |
4bb1f1ed | 3122 | down_write(&rbd_dev->header_rwsem); |
5ed16177 | 3123 | rc = rbd_dev_snaps_register(rbd_dev); |
4bb1f1ed | 3124 | up_write(&rbd_dev->header_rwsem); |
2ac4e75d AE |
3125 | if (rc) |
3126 | goto err_out_bus; | |
3127 | ||
3ee4001e AE |
3128 | rc = rbd_init_watch_dev(rbd_dev); |
3129 | if (rc) | |
3130 | goto err_out_bus; | |
3131 | ||
2ac4e75d AE |
3132 | /* Everything's ready. Announce the disk to the world. */ |
3133 | ||
2ac4e75d | 3134 | add_disk(rbd_dev->disk); |
3ee4001e | 3135 | |
2ac4e75d AE |
3136 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, |
3137 | (unsigned long long) rbd_dev->mapping.size); | |
3138 | ||
602adf40 YS |
3139 | return count; |
3140 | ||
766fc439 | 3141 | err_out_bus: |
766fc439 YS |
3142 | /* this will also clean up rest of rbd_dev stuff */ |
3143 | ||
3144 | rbd_bus_del_dev(rbd_dev); | |
3145 | kfree(options); | |
766fc439 YS |
3146 | return rc; |
3147 | ||
0f308a31 AE |
3148 | err_out_disk: |
3149 | rbd_free_disk(rbd_dev); | |
602adf40 YS |
3150 | err_out_blkdev: |
3151 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
85ae8926 AE |
3152 | err_out_id: |
3153 | rbd_dev_id_put(rbd_dev); | |
05fd6f6f AE |
3154 | err_out_header: |
3155 | rbd_header_free(&rbd_dev->header); | |
602adf40 | 3156 | err_out_client: |
3fcf2581 | 3157 | kfree(rbd_dev->header_name); |
602adf40 | 3158 | rbd_put_client(rbd_dev); |
589d30e0 | 3159 | kfree(rbd_dev->image_id); |
85ae8926 AE |
3160 | err_out_args: |
3161 | kfree(rbd_dev->mapping.snap_name); | |
3162 | kfree(rbd_dev->image_name); | |
3163 | kfree(rbd_dev->pool_name); | |
3164 | err_out_mem: | |
27cc2594 | 3165 | kfree(rbd_dev); |
cb8627c7 | 3166 | kfree(options); |
27cc2594 | 3167 | |
602adf40 YS |
3168 | dout("Error adding device %s\n", buf); |
3169 | module_put(THIS_MODULE); | |
27cc2594 AE |
3170 | |
3171 | return (ssize_t) rc; | |
602adf40 YS |
3172 | } |
3173 | ||
de71a297 | 3174 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
3175 | { |
3176 | struct list_head *tmp; | |
3177 | struct rbd_device *rbd_dev; | |
3178 | ||
e124a82f | 3179 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
3180 | list_for_each(tmp, &rbd_dev_list) { |
3181 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 3182 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 3183 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 3184 | return rbd_dev; |
e124a82f | 3185 | } |
602adf40 | 3186 | } |
e124a82f | 3187 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
3188 | return NULL; |
3189 | } | |
3190 | ||
dfc5606d | 3191 | static void rbd_dev_release(struct device *dev) |
602adf40 | 3192 | { |
593a9e7b | 3193 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3194 | |
1dbb4399 AE |
3195 | if (rbd_dev->watch_request) { |
3196 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
3197 | ||
3198 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 3199 | rbd_dev->watch_request); |
1dbb4399 | 3200 | } |
59c2be1e | 3201 | if (rbd_dev->watch_event) |
070c633f | 3202 | rbd_req_sync_unwatch(rbd_dev); |
59c2be1e | 3203 | |
602adf40 YS |
3204 | rbd_put_client(rbd_dev); |
3205 | ||
3206 | /* clean up and free blkdev */ | |
3207 | rbd_free_disk(rbd_dev); | |
3208 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 3209 | |
2ac4e75d AE |
3210 | /* release allocated disk header fields */ |
3211 | rbd_header_free(&rbd_dev->header); | |
3212 | ||
32eec68d | 3213 | /* done with the id, and with the rbd_dev */ |
f84344f3 | 3214 | kfree(rbd_dev->mapping.snap_name); |
589d30e0 | 3215 | kfree(rbd_dev->image_id); |
0bed54dc | 3216 | kfree(rbd_dev->header_name); |
d22f76e7 | 3217 | kfree(rbd_dev->pool_name); |
0bed54dc | 3218 | kfree(rbd_dev->image_name); |
e2839308 | 3219 | rbd_dev_id_put(rbd_dev); |
602adf40 YS |
3220 | kfree(rbd_dev); |
3221 | ||
3222 | /* release module ref */ | |
3223 | module_put(THIS_MODULE); | |
602adf40 YS |
3224 | } |
3225 | ||
dfc5606d YS |
3226 | static ssize_t rbd_remove(struct bus_type *bus, |
3227 | const char *buf, | |
3228 | size_t count) | |
602adf40 YS |
3229 | { |
3230 | struct rbd_device *rbd_dev = NULL; | |
3231 | int target_id, rc; | |
3232 | unsigned long ul; | |
3233 | int ret = count; | |
3234 | ||
3235 | rc = strict_strtoul(buf, 10, &ul); | |
3236 | if (rc) | |
3237 | return rc; | |
3238 | ||
3239 | /* convert to int; abort if we lost anything in the conversion */ | |
3240 | target_id = (int) ul; | |
3241 | if (target_id != ul) | |
3242 | return -EINVAL; | |
3243 | ||
3244 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
3245 | ||
3246 | rbd_dev = __rbd_get_dev(target_id); | |
3247 | if (!rbd_dev) { | |
3248 | ret = -ENOENT; | |
3249 | goto done; | |
3250 | } | |
3251 | ||
dfc5606d YS |
3252 | __rbd_remove_all_snaps(rbd_dev); |
3253 | rbd_bus_del_dev(rbd_dev); | |
602adf40 YS |
3254 | |
3255 | done: | |
3256 | mutex_unlock(&ctl_mutex); | |
aafb230e | 3257 | |
602adf40 YS |
3258 | return ret; |
3259 | } | |
3260 | ||
602adf40 YS |
3261 | /* |
3262 | * create control files in sysfs | |
dfc5606d | 3263 | * /sys/bus/rbd/... |
602adf40 YS |
3264 | */ |
3265 | static int rbd_sysfs_init(void) | |
3266 | { | |
dfc5606d | 3267 | int ret; |
602adf40 | 3268 | |
fed4c143 | 3269 | ret = device_register(&rbd_root_dev); |
21079786 | 3270 | if (ret < 0) |
dfc5606d | 3271 | return ret; |
602adf40 | 3272 | |
fed4c143 AE |
3273 | ret = bus_register(&rbd_bus_type); |
3274 | if (ret < 0) | |
3275 | device_unregister(&rbd_root_dev); | |
602adf40 | 3276 | |
602adf40 YS |
3277 | return ret; |
3278 | } | |
3279 | ||
3280 | static void rbd_sysfs_cleanup(void) | |
3281 | { | |
dfc5606d | 3282 | bus_unregister(&rbd_bus_type); |
fed4c143 | 3283 | device_unregister(&rbd_root_dev); |
602adf40 YS |
3284 | } |
3285 | ||
3286 | int __init rbd_init(void) | |
3287 | { | |
3288 | int rc; | |
3289 | ||
3290 | rc = rbd_sysfs_init(); | |
3291 | if (rc) | |
3292 | return rc; | |
f0f8cef5 | 3293 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
3294 | return 0; |
3295 | } | |
3296 | ||
3297 | void __exit rbd_exit(void) | |
3298 | { | |
3299 | rbd_sysfs_cleanup(); | |
3300 | } | |
3301 | ||
3302 | module_init(rbd_init); | |
3303 | module_exit(rbd_exit); | |
3304 | ||
3305 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
3306 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
3307 | MODULE_DESCRIPTION("rados block device"); | |
3308 | ||
3309 | /* following authorship retained from original osdblk.c */ | |
3310 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
3311 | ||
3312 | MODULE_LICENSE("GPL"); |