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 | ||
602adf40 YS |
64 | #define RBD_MAX_SNAP_NAME_LEN 32 |
65 | #define RBD_MAX_OPT_LEN 1024 | |
66 | ||
67 | #define RBD_SNAP_HEAD_NAME "-" | |
68 | ||
81a89793 AE |
69 | /* |
70 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
71 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
72 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
73 | * enough to hold all possible device names. | |
74 | */ | |
602adf40 | 75 | #define DEV_NAME_LEN 32 |
81a89793 | 76 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 | 77 | |
cc0538b6 | 78 | #define RBD_READ_ONLY_DEFAULT false |
59c2be1e | 79 | |
602adf40 YS |
80 | /* |
81 | * block device image metadata (in-memory version) | |
82 | */ | |
83 | struct rbd_image_header { | |
f84344f3 | 84 | /* These four fields never change for a given rbd image */ |
849b4260 | 85 | char *object_prefix; |
602adf40 YS |
86 | __u8 obj_order; |
87 | __u8 crypt_type; | |
88 | __u8 comp_type; | |
602adf40 | 89 | |
f84344f3 AE |
90 | /* The remaining fields need to be updated occasionally */ |
91 | u64 image_size; | |
92 | struct ceph_snap_context *snapc; | |
602adf40 YS |
93 | char *snap_names; |
94 | u64 *snap_sizes; | |
59c2be1e YS |
95 | |
96 | u64 obj_version; | |
97 | }; | |
98 | ||
99 | struct rbd_options { | |
cc0538b6 | 100 | bool read_only; |
602adf40 YS |
101 | }; |
102 | ||
103 | /* | |
f0f8cef5 | 104 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
105 | */ |
106 | struct rbd_client { | |
107 | struct ceph_client *client; | |
108 | struct kref kref; | |
109 | struct list_head node; | |
110 | }; | |
111 | ||
112 | /* | |
f0f8cef5 | 113 | * a request completion status |
602adf40 | 114 | */ |
1fec7093 YS |
115 | struct rbd_req_status { |
116 | int done; | |
117 | int rc; | |
118 | u64 bytes; | |
119 | }; | |
120 | ||
121 | /* | |
122 | * a collection of requests | |
123 | */ | |
124 | struct rbd_req_coll { | |
125 | int total; | |
126 | int num_done; | |
127 | struct kref kref; | |
128 | struct rbd_req_status status[0]; | |
602adf40 YS |
129 | }; |
130 | ||
f0f8cef5 AE |
131 | /* |
132 | * a single io request | |
133 | */ | |
134 | struct rbd_request { | |
135 | struct request *rq; /* blk layer request */ | |
136 | struct bio *bio; /* cloned bio */ | |
137 | struct page **pages; /* list of used pages */ | |
138 | u64 len; | |
139 | int coll_index; | |
140 | struct rbd_req_coll *coll; | |
141 | }; | |
142 | ||
dfc5606d YS |
143 | struct rbd_snap { |
144 | struct device dev; | |
145 | const char *name; | |
3591538f | 146 | u64 size; |
dfc5606d YS |
147 | struct list_head node; |
148 | u64 id; | |
149 | }; | |
150 | ||
f84344f3 AE |
151 | struct rbd_mapping { |
152 | char *snap_name; | |
153 | u64 snap_id; | |
99c1f08f | 154 | u64 size; |
f84344f3 AE |
155 | bool snap_exists; |
156 | bool read_only; | |
157 | }; | |
158 | ||
602adf40 YS |
159 | /* |
160 | * a single device | |
161 | */ | |
162 | struct rbd_device { | |
de71a297 | 163 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
164 | |
165 | int major; /* blkdev assigned major */ | |
166 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 167 | |
f8c38929 | 168 | struct rbd_options rbd_opts; |
602adf40 YS |
169 | struct rbd_client *rbd_client; |
170 | ||
171 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
172 | ||
173 | spinlock_t lock; /* queue lock */ | |
174 | ||
175 | struct rbd_image_header header; | |
0bed54dc AE |
176 | char *image_name; |
177 | size_t image_name_len; | |
178 | char *header_name; | |
d22f76e7 | 179 | char *pool_name; |
9bb2f334 | 180 | int pool_id; |
602adf40 | 181 | |
59c2be1e YS |
182 | struct ceph_osd_event *watch_event; |
183 | struct ceph_osd_request *watch_request; | |
184 | ||
c666601a JD |
185 | /* protects updating the header */ |
186 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
187 | |
188 | struct rbd_mapping mapping; | |
602adf40 YS |
189 | |
190 | struct list_head node; | |
dfc5606d YS |
191 | |
192 | /* list of snapshots */ | |
193 | struct list_head snaps; | |
194 | ||
195 | /* sysfs related */ | |
196 | struct device dev; | |
197 | }; | |
198 | ||
602adf40 | 199 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 200 | |
602adf40 | 201 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
202 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
203 | ||
432b8587 AE |
204 | static LIST_HEAD(rbd_client_list); /* clients */ |
205 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 206 | |
9fcbb800 | 207 | static int rbd_dev_snap_devs_update(struct rbd_device *rbd_dev); |
dfc5606d | 208 | static void rbd_dev_release(struct device *dev); |
dfc5606d YS |
209 | static ssize_t rbd_snap_add(struct device *dev, |
210 | struct device_attribute *attr, | |
211 | const char *buf, | |
212 | size_t count); | |
14e7085d | 213 | static void __rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 214 | |
f0f8cef5 AE |
215 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
216 | size_t count); | |
217 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
218 | size_t count); | |
219 | ||
220 | static struct bus_attribute rbd_bus_attrs[] = { | |
221 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
222 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
223 | __ATTR_NULL | |
224 | }; | |
225 | ||
226 | static struct bus_type rbd_bus_type = { | |
227 | .name = "rbd", | |
228 | .bus_attrs = rbd_bus_attrs, | |
229 | }; | |
230 | ||
231 | static void rbd_root_dev_release(struct device *dev) | |
232 | { | |
233 | } | |
234 | ||
235 | static struct device rbd_root_dev = { | |
236 | .init_name = "rbd", | |
237 | .release = rbd_root_dev_release, | |
238 | }; | |
239 | ||
aafb230e AE |
240 | #ifdef RBD_DEBUG |
241 | #define rbd_assert(expr) \ | |
242 | if (unlikely(!(expr))) { \ | |
243 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
244 | "at line %d:\n\n" \ | |
245 | "\trbd_assert(%s);\n\n", \ | |
246 | __func__, __LINE__, #expr); \ | |
247 | BUG(); \ | |
248 | } | |
249 | #else /* !RBD_DEBUG */ | |
250 | # define rbd_assert(expr) ((void) 0) | |
251 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 252 | |
dfc5606d YS |
253 | static struct device *rbd_get_dev(struct rbd_device *rbd_dev) |
254 | { | |
255 | return get_device(&rbd_dev->dev); | |
256 | } | |
257 | ||
258 | static void rbd_put_dev(struct rbd_device *rbd_dev) | |
259 | { | |
260 | put_device(&rbd_dev->dev); | |
261 | } | |
602adf40 | 262 | |
1fe5e993 | 263 | static int rbd_refresh_header(struct rbd_device *rbd_dev, u64 *hver); |
59c2be1e | 264 | |
602adf40 YS |
265 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
266 | { | |
f0f8cef5 | 267 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
602adf40 | 268 | |
f84344f3 | 269 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
270 | return -EROFS; |
271 | ||
340c7a2b | 272 | rbd_get_dev(rbd_dev); |
f84344f3 | 273 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
340c7a2b | 274 | |
602adf40 YS |
275 | return 0; |
276 | } | |
277 | ||
dfc5606d YS |
278 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
279 | { | |
280 | struct rbd_device *rbd_dev = disk->private_data; | |
281 | ||
282 | rbd_put_dev(rbd_dev); | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
602adf40 YS |
287 | static const struct block_device_operations rbd_bd_ops = { |
288 | .owner = THIS_MODULE, | |
289 | .open = rbd_open, | |
dfc5606d | 290 | .release = rbd_release, |
602adf40 YS |
291 | }; |
292 | ||
293 | /* | |
294 | * Initialize an rbd client instance. | |
43ae4701 | 295 | * We own *ceph_opts. |
602adf40 | 296 | */ |
f8c38929 | 297 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
298 | { |
299 | struct rbd_client *rbdc; | |
300 | int ret = -ENOMEM; | |
301 | ||
302 | dout("rbd_client_create\n"); | |
303 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); | |
304 | if (!rbdc) | |
305 | goto out_opt; | |
306 | ||
307 | kref_init(&rbdc->kref); | |
308 | INIT_LIST_HEAD(&rbdc->node); | |
309 | ||
bc534d86 AE |
310 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
311 | ||
43ae4701 | 312 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 313 | if (IS_ERR(rbdc->client)) |
bc534d86 | 314 | goto out_mutex; |
43ae4701 | 315 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
316 | |
317 | ret = ceph_open_session(rbdc->client); | |
318 | if (ret < 0) | |
319 | goto out_err; | |
320 | ||
432b8587 | 321 | spin_lock(&rbd_client_list_lock); |
602adf40 | 322 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 323 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 324 | |
bc534d86 AE |
325 | mutex_unlock(&ctl_mutex); |
326 | ||
602adf40 YS |
327 | dout("rbd_client_create created %p\n", rbdc); |
328 | return rbdc; | |
329 | ||
330 | out_err: | |
331 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
332 | out_mutex: |
333 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
334 | kfree(rbdc); |
335 | out_opt: | |
43ae4701 AE |
336 | if (ceph_opts) |
337 | ceph_destroy_options(ceph_opts); | |
28f259b7 | 338 | return ERR_PTR(ret); |
602adf40 YS |
339 | } |
340 | ||
341 | /* | |
1f7ba331 AE |
342 | * Find a ceph client with specific addr and configuration. If |
343 | * found, bump its reference count. | |
602adf40 | 344 | */ |
1f7ba331 | 345 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
346 | { |
347 | struct rbd_client *client_node; | |
1f7ba331 | 348 | bool found = false; |
602adf40 | 349 | |
43ae4701 | 350 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
351 | return NULL; |
352 | ||
1f7ba331 AE |
353 | spin_lock(&rbd_client_list_lock); |
354 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
355 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
356 | kref_get(&client_node->kref); | |
357 | found = true; | |
358 | break; | |
359 | } | |
360 | } | |
361 | spin_unlock(&rbd_client_list_lock); | |
362 | ||
363 | return found ? client_node : NULL; | |
602adf40 YS |
364 | } |
365 | ||
59c2be1e YS |
366 | /* |
367 | * mount options | |
368 | */ | |
369 | enum { | |
59c2be1e YS |
370 | Opt_last_int, |
371 | /* int args above */ | |
372 | Opt_last_string, | |
373 | /* string args above */ | |
cc0538b6 AE |
374 | Opt_read_only, |
375 | Opt_read_write, | |
376 | /* Boolean args above */ | |
377 | Opt_last_bool, | |
59c2be1e YS |
378 | }; |
379 | ||
43ae4701 | 380 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
381 | /* int args above */ |
382 | /* string args above */ | |
f84344f3 | 383 | {Opt_read_only, "mapping.read_only"}, |
cc0538b6 AE |
384 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
385 | {Opt_read_write, "read_write"}, | |
386 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
387 | /* Boolean args above */ | |
59c2be1e YS |
388 | {-1, NULL} |
389 | }; | |
390 | ||
391 | static int parse_rbd_opts_token(char *c, void *private) | |
392 | { | |
43ae4701 | 393 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
394 | substring_t argstr[MAX_OPT_ARGS]; |
395 | int token, intval, ret; | |
396 | ||
43ae4701 | 397 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
398 | if (token < 0) |
399 | return -EINVAL; | |
400 | ||
401 | if (token < Opt_last_int) { | |
402 | ret = match_int(&argstr[0], &intval); | |
403 | if (ret < 0) { | |
404 | pr_err("bad mount option arg (not int) " | |
405 | "at '%s'\n", c); | |
406 | return ret; | |
407 | } | |
408 | dout("got int token %d val %d\n", token, intval); | |
409 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
410 | dout("got string token %d val %s\n", token, | |
411 | argstr[0].from); | |
cc0538b6 AE |
412 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
413 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
414 | } else { |
415 | dout("got token %d\n", token); | |
416 | } | |
417 | ||
418 | switch (token) { | |
cc0538b6 AE |
419 | case Opt_read_only: |
420 | rbd_opts->read_only = true; | |
421 | break; | |
422 | case Opt_read_write: | |
423 | rbd_opts->read_only = false; | |
424 | break; | |
59c2be1e | 425 | default: |
aafb230e AE |
426 | rbd_assert(false); |
427 | break; | |
59c2be1e YS |
428 | } |
429 | return 0; | |
430 | } | |
431 | ||
602adf40 YS |
432 | /* |
433 | * Get a ceph client with specific addr and configuration, if one does | |
434 | * not exist create it. | |
435 | */ | |
f8c38929 AE |
436 | static int rbd_get_client(struct rbd_device *rbd_dev, const char *mon_addr, |
437 | size_t mon_addr_len, char *options) | |
602adf40 | 438 | { |
f8c38929 | 439 | struct rbd_options *rbd_opts = &rbd_dev->rbd_opts; |
43ae4701 | 440 | struct ceph_options *ceph_opts; |
f8c38929 | 441 | struct rbd_client *rbdc; |
59c2be1e | 442 | |
cc0538b6 | 443 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; |
602adf40 | 444 | |
43ae4701 AE |
445 | ceph_opts = ceph_parse_options(options, mon_addr, |
446 | mon_addr + mon_addr_len, | |
447 | parse_rbd_opts_token, rbd_opts); | |
f8c38929 AE |
448 | if (IS_ERR(ceph_opts)) |
449 | return PTR_ERR(ceph_opts); | |
602adf40 | 450 | |
1f7ba331 | 451 | rbdc = rbd_client_find(ceph_opts); |
602adf40 | 452 | if (rbdc) { |
602adf40 | 453 | /* using an existing client */ |
43ae4701 | 454 | ceph_destroy_options(ceph_opts); |
f8c38929 AE |
455 | } else { |
456 | rbdc = rbd_client_create(ceph_opts); | |
457 | if (IS_ERR(rbdc)) | |
458 | return PTR_ERR(rbdc); | |
602adf40 | 459 | } |
f8c38929 | 460 | rbd_dev->rbd_client = rbdc; |
602adf40 | 461 | |
f8c38929 | 462 | return 0; |
602adf40 YS |
463 | } |
464 | ||
465 | /* | |
466 | * Destroy ceph client | |
d23a4b3f | 467 | * |
432b8587 | 468 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
469 | */ |
470 | static void rbd_client_release(struct kref *kref) | |
471 | { | |
472 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
473 | ||
474 | dout("rbd_release_client %p\n", rbdc); | |
cd9d9f5d | 475 | spin_lock(&rbd_client_list_lock); |
602adf40 | 476 | list_del(&rbdc->node); |
cd9d9f5d | 477 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
478 | |
479 | ceph_destroy_client(rbdc->client); | |
480 | kfree(rbdc); | |
481 | } | |
482 | ||
483 | /* | |
484 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
485 | * it. | |
486 | */ | |
487 | static void rbd_put_client(struct rbd_device *rbd_dev) | |
488 | { | |
489 | kref_put(&rbd_dev->rbd_client->kref, rbd_client_release); | |
490 | rbd_dev->rbd_client = NULL; | |
602adf40 YS |
491 | } |
492 | ||
1fec7093 YS |
493 | /* |
494 | * Destroy requests collection | |
495 | */ | |
496 | static void rbd_coll_release(struct kref *kref) | |
497 | { | |
498 | struct rbd_req_coll *coll = | |
499 | container_of(kref, struct rbd_req_coll, kref); | |
500 | ||
501 | dout("rbd_coll_release %p\n", coll); | |
502 | kfree(coll); | |
503 | } | |
602adf40 | 504 | |
8e94af8e AE |
505 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
506 | { | |
103a150f AE |
507 | size_t size; |
508 | u32 snap_count; | |
509 | ||
510 | /* The header has to start with the magic rbd header text */ | |
511 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
512 | return false; | |
513 | ||
514 | /* | |
515 | * The size of a snapshot header has to fit in a size_t, and | |
516 | * that limits the number of snapshots. | |
517 | */ | |
518 | snap_count = le32_to_cpu(ondisk->snap_count); | |
519 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
520 | if (snap_count > size / sizeof (__le64)) | |
521 | return false; | |
522 | ||
523 | /* | |
524 | * Not only that, but the size of the entire the snapshot | |
525 | * header must also be representable in a size_t. | |
526 | */ | |
527 | size -= snap_count * sizeof (__le64); | |
528 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
529 | return false; | |
530 | ||
531 | return true; | |
8e94af8e AE |
532 | } |
533 | ||
602adf40 YS |
534 | /* |
535 | * Create a new header structure, translate header format from the on-disk | |
536 | * header. | |
537 | */ | |
538 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 539 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 540 | { |
ccece235 | 541 | u32 snap_count; |
58c17b0e | 542 | size_t len; |
d2bb24e5 | 543 | size_t size; |
621901d6 | 544 | u32 i; |
602adf40 | 545 | |
6a52325f AE |
546 | memset(header, 0, sizeof (*header)); |
547 | ||
103a150f AE |
548 | snap_count = le32_to_cpu(ondisk->snap_count); |
549 | ||
58c17b0e AE |
550 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
551 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 552 | if (!header->object_prefix) |
602adf40 | 553 | return -ENOMEM; |
58c17b0e AE |
554 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
555 | header->object_prefix[len] = '\0'; | |
00f1f36f | 556 | |
602adf40 | 557 | if (snap_count) { |
f785cc1d AE |
558 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
559 | ||
621901d6 AE |
560 | /* Save a copy of the snapshot names */ |
561 | ||
f785cc1d AE |
562 | if (snap_names_len > (u64) SIZE_MAX) |
563 | return -EIO; | |
564 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 565 | if (!header->snap_names) |
6a52325f | 566 | goto out_err; |
f785cc1d AE |
567 | /* |
568 | * Note that rbd_dev_v1_header_read() guarantees | |
569 | * the ondisk buffer we're working with has | |
570 | * snap_names_len bytes beyond the end of the | |
571 | * snapshot id array, this memcpy() is safe. | |
572 | */ | |
573 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
574 | snap_names_len); | |
6a52325f | 575 | |
621901d6 AE |
576 | /* Record each snapshot's size */ |
577 | ||
d2bb24e5 AE |
578 | size = snap_count * sizeof (*header->snap_sizes); |
579 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 580 | if (!header->snap_sizes) |
6a52325f | 581 | goto out_err; |
621901d6 AE |
582 | for (i = 0; i < snap_count; i++) |
583 | header->snap_sizes[i] = | |
584 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 585 | } else { |
ccece235 | 586 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
587 | header->snap_names = NULL; |
588 | header->snap_sizes = NULL; | |
589 | } | |
849b4260 | 590 | |
602adf40 YS |
591 | header->obj_order = ondisk->options.order; |
592 | header->crypt_type = ondisk->options.crypt_type; | |
593 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 594 | |
621901d6 AE |
595 | /* Allocate and fill in the snapshot context */ |
596 | ||
f84344f3 | 597 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
598 | size = sizeof (struct ceph_snap_context); |
599 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
600 | header->snapc = kzalloc(size, GFP_KERNEL); | |
601 | if (!header->snapc) | |
602 | goto out_err; | |
602adf40 YS |
603 | |
604 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 605 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 606 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
607 | for (i = 0; i < snap_count; i++) |
608 | header->snapc->snaps[i] = | |
609 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
610 | |
611 | return 0; | |
612 | ||
6a52325f | 613 | out_err: |
849b4260 | 614 | kfree(header->snap_sizes); |
ccece235 | 615 | header->snap_sizes = NULL; |
602adf40 | 616 | kfree(header->snap_names); |
ccece235 | 617 | header->snap_names = NULL; |
6a52325f AE |
618 | kfree(header->object_prefix); |
619 | header->object_prefix = NULL; | |
ccece235 | 620 | |
00f1f36f | 621 | return -ENOMEM; |
602adf40 YS |
622 | } |
623 | ||
8836b995 | 624 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 YS |
625 | { |
626 | int i; | |
8836b995 | 627 | struct rbd_image_header *header = &rbd_dev->header; |
602adf40 YS |
628 | char *p = header->snap_names; |
629 | ||
c9aadfe7 AE |
630 | rbd_assert(header->snapc != NULL); |
631 | for (i = 0; i < header->snapc->num_snaps; i++) { | |
00f1f36f | 632 | if (!strcmp(snap_name, p)) { |
602adf40 | 633 | |
00f1f36f | 634 | /* Found it. Pass back its id and/or size */ |
602adf40 | 635 | |
8836b995 AE |
636 | rbd_dev->mapping.snap_id = header->snapc->snaps[i]; |
637 | rbd_dev->mapping.size = header->snap_sizes[i]; | |
638 | ||
00f1f36f AE |
639 | return i; |
640 | } | |
641 | p += strlen(p) + 1; /* Skip ahead to the next name */ | |
642 | } | |
643 | return -ENOENT; | |
602adf40 YS |
644 | } |
645 | ||
4e1105a2 | 646 | static int rbd_header_set_snap(struct rbd_device *rbd_dev, char *snap_name) |
602adf40 | 647 | { |
78dc447d | 648 | int ret; |
602adf40 | 649 | |
0ce1a794 | 650 | down_write(&rbd_dev->header_rwsem); |
602adf40 | 651 | |
4e1105a2 | 652 | if (!memcmp(snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 653 | sizeof (RBD_SNAP_HEAD_NAME))) { |
f84344f3 | 654 | rbd_dev->mapping.snap_id = CEPH_NOSNAP; |
99c1f08f | 655 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
f84344f3 AE |
656 | rbd_dev->mapping.snap_exists = false; |
657 | rbd_dev->mapping.read_only = rbd_dev->rbd_opts.read_only; | |
602adf40 | 658 | } else { |
8836b995 | 659 | ret = snap_by_name(rbd_dev, snap_name); |
602adf40 YS |
660 | if (ret < 0) |
661 | goto done; | |
f84344f3 AE |
662 | rbd_dev->mapping.snap_exists = true; |
663 | rbd_dev->mapping.read_only = true; | |
602adf40 | 664 | } |
4e1105a2 | 665 | rbd_dev->mapping.snap_name = snap_name; |
602adf40 YS |
666 | |
667 | ret = 0; | |
668 | done: | |
0ce1a794 | 669 | up_write(&rbd_dev->header_rwsem); |
602adf40 YS |
670 | return ret; |
671 | } | |
672 | ||
673 | static void rbd_header_free(struct rbd_image_header *header) | |
674 | { | |
849b4260 | 675 | kfree(header->object_prefix); |
d78fd7ae | 676 | header->object_prefix = NULL; |
602adf40 | 677 | kfree(header->snap_sizes); |
d78fd7ae | 678 | header->snap_sizes = NULL; |
849b4260 | 679 | kfree(header->snap_names); |
d78fd7ae | 680 | header->snap_names = NULL; |
d1d25646 | 681 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 682 | header->snapc = NULL; |
602adf40 YS |
683 | } |
684 | ||
65ccfe21 | 685 | static char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 686 | { |
65ccfe21 AE |
687 | char *name; |
688 | u64 segment; | |
689 | int ret; | |
602adf40 | 690 | |
65ccfe21 AE |
691 | name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO); |
692 | if (!name) | |
693 | return NULL; | |
694 | segment = offset >> rbd_dev->header.obj_order; | |
695 | ret = snprintf(name, RBD_MAX_SEG_NAME_LEN, "%s.%012llx", | |
696 | rbd_dev->header.object_prefix, segment); | |
697 | if (ret < 0 || ret >= RBD_MAX_SEG_NAME_LEN) { | |
698 | pr_err("error formatting segment name for #%llu (%d)\n", | |
699 | segment, ret); | |
700 | kfree(name); | |
701 | name = NULL; | |
702 | } | |
602adf40 | 703 | |
65ccfe21 AE |
704 | return name; |
705 | } | |
602adf40 | 706 | |
65ccfe21 AE |
707 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
708 | { | |
709 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 710 | |
65ccfe21 AE |
711 | return offset & (segment_size - 1); |
712 | } | |
713 | ||
714 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
715 | u64 offset, u64 length) | |
716 | { | |
717 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
718 | ||
719 | offset &= segment_size - 1; | |
720 | ||
aafb230e | 721 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
722 | if (offset + length > segment_size) |
723 | length = segment_size - offset; | |
724 | ||
725 | return length; | |
602adf40 YS |
726 | } |
727 | ||
1fec7093 YS |
728 | static int rbd_get_num_segments(struct rbd_image_header *header, |
729 | u64 ofs, u64 len) | |
730 | { | |
df111be6 AE |
731 | u64 start_seg; |
732 | u64 end_seg; | |
733 | ||
734 | if (!len) | |
735 | return 0; | |
736 | if (len - 1 > U64_MAX - ofs) | |
737 | return -ERANGE; | |
738 | ||
739 | start_seg = ofs >> header->obj_order; | |
740 | end_seg = (ofs + len - 1) >> header->obj_order; | |
741 | ||
1fec7093 YS |
742 | return end_seg - start_seg + 1; |
743 | } | |
744 | ||
029bcbd8 JD |
745 | /* |
746 | * returns the size of an object in the image | |
747 | */ | |
748 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
749 | { | |
750 | return 1 << header->obj_order; | |
751 | } | |
752 | ||
602adf40 YS |
753 | /* |
754 | * bio helpers | |
755 | */ | |
756 | ||
757 | static void bio_chain_put(struct bio *chain) | |
758 | { | |
759 | struct bio *tmp; | |
760 | ||
761 | while (chain) { | |
762 | tmp = chain; | |
763 | chain = chain->bi_next; | |
764 | bio_put(tmp); | |
765 | } | |
766 | } | |
767 | ||
768 | /* | |
769 | * zeros a bio chain, starting at specific offset | |
770 | */ | |
771 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
772 | { | |
773 | struct bio_vec *bv; | |
774 | unsigned long flags; | |
775 | void *buf; | |
776 | int i; | |
777 | int pos = 0; | |
778 | ||
779 | while (chain) { | |
780 | bio_for_each_segment(bv, chain, i) { | |
781 | if (pos + bv->bv_len > start_ofs) { | |
782 | int remainder = max(start_ofs - pos, 0); | |
783 | buf = bvec_kmap_irq(bv, &flags); | |
784 | memset(buf + remainder, 0, | |
785 | bv->bv_len - remainder); | |
85b5aaa6 | 786 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
787 | } |
788 | pos += bv->bv_len; | |
789 | } | |
790 | ||
791 | chain = chain->bi_next; | |
792 | } | |
793 | } | |
794 | ||
795 | /* | |
796 | * bio_chain_clone - clone a chain of bios up to a certain length. | |
797 | * might return a bio_pair that will need to be released. | |
798 | */ | |
799 | static struct bio *bio_chain_clone(struct bio **old, struct bio **next, | |
800 | struct bio_pair **bp, | |
801 | int len, gfp_t gfpmask) | |
802 | { | |
542582fc AE |
803 | struct bio *old_chain = *old; |
804 | struct bio *new_chain = NULL; | |
805 | struct bio *tail; | |
602adf40 YS |
806 | int total = 0; |
807 | ||
808 | if (*bp) { | |
809 | bio_pair_release(*bp); | |
810 | *bp = NULL; | |
811 | } | |
812 | ||
813 | while (old_chain && (total < len)) { | |
542582fc AE |
814 | struct bio *tmp; |
815 | ||
602adf40 YS |
816 | tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs); |
817 | if (!tmp) | |
818 | goto err_out; | |
542582fc | 819 | gfpmask &= ~__GFP_WAIT; /* can't wait after the first */ |
602adf40 YS |
820 | |
821 | if (total + old_chain->bi_size > len) { | |
822 | struct bio_pair *bp; | |
823 | ||
824 | /* | |
825 | * this split can only happen with a single paged bio, | |
826 | * split_bio will BUG_ON if this is not the case | |
827 | */ | |
828 | dout("bio_chain_clone split! total=%d remaining=%d" | |
bd919d45 AE |
829 | "bi_size=%u\n", |
830 | total, len - total, old_chain->bi_size); | |
602adf40 YS |
831 | |
832 | /* split the bio. We'll release it either in the next | |
833 | call, or it will have to be released outside */ | |
593a9e7b | 834 | bp = bio_split(old_chain, (len - total) / SECTOR_SIZE); |
602adf40 YS |
835 | if (!bp) |
836 | goto err_out; | |
837 | ||
838 | __bio_clone(tmp, &bp->bio1); | |
839 | ||
840 | *next = &bp->bio2; | |
841 | } else { | |
842 | __bio_clone(tmp, old_chain); | |
843 | *next = old_chain->bi_next; | |
844 | } | |
845 | ||
846 | tmp->bi_bdev = NULL; | |
602adf40 | 847 | tmp->bi_next = NULL; |
542582fc | 848 | if (new_chain) |
602adf40 | 849 | tail->bi_next = tmp; |
542582fc AE |
850 | else |
851 | new_chain = tmp; | |
852 | tail = tmp; | |
602adf40 YS |
853 | old_chain = old_chain->bi_next; |
854 | ||
855 | total += tmp->bi_size; | |
856 | } | |
857 | ||
aafb230e | 858 | rbd_assert(total == len); |
602adf40 | 859 | |
602adf40 YS |
860 | *old = old_chain; |
861 | ||
862 | return new_chain; | |
863 | ||
864 | err_out: | |
865 | dout("bio_chain_clone with err\n"); | |
866 | bio_chain_put(new_chain); | |
867 | return NULL; | |
868 | } | |
869 | ||
870 | /* | |
871 | * helpers for osd request op vectors. | |
872 | */ | |
57cfc106 AE |
873 | static struct ceph_osd_req_op *rbd_create_rw_ops(int num_ops, |
874 | int opcode, u32 payload_len) | |
602adf40 | 875 | { |
57cfc106 AE |
876 | struct ceph_osd_req_op *ops; |
877 | ||
878 | ops = kzalloc(sizeof (*ops) * (num_ops + 1), GFP_NOIO); | |
879 | if (!ops) | |
880 | return NULL; | |
881 | ||
882 | ops[0].op = opcode; | |
883 | ||
602adf40 YS |
884 | /* |
885 | * op extent offset and length will be set later on | |
886 | * in calc_raw_layout() | |
887 | */ | |
57cfc106 AE |
888 | ops[0].payload_len = payload_len; |
889 | ||
890 | return ops; | |
602adf40 YS |
891 | } |
892 | ||
893 | static void rbd_destroy_ops(struct ceph_osd_req_op *ops) | |
894 | { | |
895 | kfree(ops); | |
896 | } | |
897 | ||
1fec7093 YS |
898 | static void rbd_coll_end_req_index(struct request *rq, |
899 | struct rbd_req_coll *coll, | |
900 | int index, | |
901 | int ret, u64 len) | |
902 | { | |
903 | struct request_queue *q; | |
904 | int min, max, i; | |
905 | ||
bd919d45 AE |
906 | dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n", |
907 | coll, index, ret, (unsigned long long) len); | |
1fec7093 YS |
908 | |
909 | if (!rq) | |
910 | return; | |
911 | ||
912 | if (!coll) { | |
913 | blk_end_request(rq, ret, len); | |
914 | return; | |
915 | } | |
916 | ||
917 | q = rq->q; | |
918 | ||
919 | spin_lock_irq(q->queue_lock); | |
920 | coll->status[index].done = 1; | |
921 | coll->status[index].rc = ret; | |
922 | coll->status[index].bytes = len; | |
923 | max = min = coll->num_done; | |
924 | while (max < coll->total && coll->status[max].done) | |
925 | max++; | |
926 | ||
927 | for (i = min; i<max; i++) { | |
928 | __blk_end_request(rq, coll->status[i].rc, | |
929 | coll->status[i].bytes); | |
930 | coll->num_done++; | |
931 | kref_put(&coll->kref, rbd_coll_release); | |
932 | } | |
933 | spin_unlock_irq(q->queue_lock); | |
934 | } | |
935 | ||
936 | static void rbd_coll_end_req(struct rbd_request *req, | |
937 | int ret, u64 len) | |
938 | { | |
939 | rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len); | |
940 | } | |
941 | ||
602adf40 YS |
942 | /* |
943 | * Send ceph osd request | |
944 | */ | |
945 | static int rbd_do_request(struct request *rq, | |
0ce1a794 | 946 | struct rbd_device *rbd_dev, |
602adf40 YS |
947 | struct ceph_snap_context *snapc, |
948 | u64 snapid, | |
aded07ea | 949 | const char *object_name, u64 ofs, u64 len, |
602adf40 YS |
950 | struct bio *bio, |
951 | struct page **pages, | |
952 | int num_pages, | |
953 | int flags, | |
954 | struct ceph_osd_req_op *ops, | |
1fec7093 YS |
955 | struct rbd_req_coll *coll, |
956 | int coll_index, | |
602adf40 | 957 | void (*rbd_cb)(struct ceph_osd_request *req, |
59c2be1e YS |
958 | struct ceph_msg *msg), |
959 | struct ceph_osd_request **linger_req, | |
960 | u64 *ver) | |
602adf40 YS |
961 | { |
962 | struct ceph_osd_request *req; | |
963 | struct ceph_file_layout *layout; | |
964 | int ret; | |
965 | u64 bno; | |
966 | struct timespec mtime = CURRENT_TIME; | |
967 | struct rbd_request *req_data; | |
968 | struct ceph_osd_request_head *reqhead; | |
1dbb4399 | 969 | struct ceph_osd_client *osdc; |
602adf40 | 970 | |
602adf40 | 971 | req_data = kzalloc(sizeof(*req_data), GFP_NOIO); |
1fec7093 YS |
972 | if (!req_data) { |
973 | if (coll) | |
974 | rbd_coll_end_req_index(rq, coll, coll_index, | |
975 | -ENOMEM, len); | |
976 | return -ENOMEM; | |
977 | } | |
978 | ||
979 | if (coll) { | |
980 | req_data->coll = coll; | |
981 | req_data->coll_index = coll_index; | |
982 | } | |
602adf40 | 983 | |
bd919d45 AE |
984 | dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name, |
985 | (unsigned long long) ofs, (unsigned long long) len); | |
602adf40 | 986 | |
0ce1a794 | 987 | osdc = &rbd_dev->rbd_client->client->osdc; |
1dbb4399 AE |
988 | req = ceph_osdc_alloc_request(osdc, flags, snapc, ops, |
989 | false, GFP_NOIO, pages, bio); | |
4ad12621 | 990 | if (!req) { |
4ad12621 | 991 | ret = -ENOMEM; |
602adf40 YS |
992 | goto done_pages; |
993 | } | |
994 | ||
995 | req->r_callback = rbd_cb; | |
996 | ||
997 | req_data->rq = rq; | |
998 | req_data->bio = bio; | |
999 | req_data->pages = pages; | |
1000 | req_data->len = len; | |
1001 | ||
1002 | req->r_priv = req_data; | |
1003 | ||
1004 | reqhead = req->r_request->front.iov_base; | |
1005 | reqhead->snapid = cpu_to_le64(CEPH_NOSNAP); | |
1006 | ||
aded07ea | 1007 | strncpy(req->r_oid, object_name, sizeof(req->r_oid)); |
602adf40 YS |
1008 | req->r_oid_len = strlen(req->r_oid); |
1009 | ||
1010 | layout = &req->r_file_layout; | |
1011 | memset(layout, 0, sizeof(*layout)); | |
1012 | layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
1013 | layout->fl_stripe_count = cpu_to_le32(1); | |
1014 | layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
0ce1a794 | 1015 | layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id); |
1dbb4399 AE |
1016 | ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno, |
1017 | req, ops); | |
602adf40 YS |
1018 | |
1019 | ceph_osdc_build_request(req, ofs, &len, | |
1020 | ops, | |
1021 | snapc, | |
1022 | &mtime, | |
1023 | req->r_oid, req->r_oid_len); | |
602adf40 | 1024 | |
59c2be1e | 1025 | if (linger_req) { |
1dbb4399 | 1026 | ceph_osdc_set_request_linger(osdc, req); |
59c2be1e YS |
1027 | *linger_req = req; |
1028 | } | |
1029 | ||
1dbb4399 | 1030 | ret = ceph_osdc_start_request(osdc, req, false); |
602adf40 YS |
1031 | if (ret < 0) |
1032 | goto done_err; | |
1033 | ||
1034 | if (!rbd_cb) { | |
1dbb4399 | 1035 | ret = ceph_osdc_wait_request(osdc, req); |
59c2be1e YS |
1036 | if (ver) |
1037 | *ver = le64_to_cpu(req->r_reassert_version.version); | |
bd919d45 AE |
1038 | dout("reassert_ver=%llu\n", |
1039 | (unsigned long long) | |
1040 | le64_to_cpu(req->r_reassert_version.version)); | |
602adf40 YS |
1041 | ceph_osdc_put_request(req); |
1042 | } | |
1043 | return ret; | |
1044 | ||
1045 | done_err: | |
1046 | bio_chain_put(req_data->bio); | |
1047 | ceph_osdc_put_request(req); | |
1048 | done_pages: | |
1fec7093 | 1049 | rbd_coll_end_req(req_data, ret, len); |
602adf40 | 1050 | kfree(req_data); |
602adf40 YS |
1051 | return ret; |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * Ceph osd op callback | |
1056 | */ | |
1057 | static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) | |
1058 | { | |
1059 | struct rbd_request *req_data = req->r_priv; | |
1060 | struct ceph_osd_reply_head *replyhead; | |
1061 | struct ceph_osd_op *op; | |
1062 | __s32 rc; | |
1063 | u64 bytes; | |
1064 | int read_op; | |
1065 | ||
1066 | /* parse reply */ | |
1067 | replyhead = msg->front.iov_base; | |
1068 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
1069 | op = (void *)(replyhead + 1); | |
1070 | rc = le32_to_cpu(replyhead->result); | |
1071 | bytes = le64_to_cpu(op->extent.length); | |
895cfcc8 | 1072 | read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ); |
602adf40 | 1073 | |
bd919d45 AE |
1074 | dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n", |
1075 | (unsigned long long) bytes, read_op, (int) rc); | |
602adf40 YS |
1076 | |
1077 | if (rc == -ENOENT && read_op) { | |
1078 | zero_bio_chain(req_data->bio, 0); | |
1079 | rc = 0; | |
1080 | } else if (rc == 0 && read_op && bytes < req_data->len) { | |
1081 | zero_bio_chain(req_data->bio, bytes); | |
1082 | bytes = req_data->len; | |
1083 | } | |
1084 | ||
1fec7093 | 1085 | rbd_coll_end_req(req_data, rc, bytes); |
602adf40 YS |
1086 | |
1087 | if (req_data->bio) | |
1088 | bio_chain_put(req_data->bio); | |
1089 | ||
1090 | ceph_osdc_put_request(req); | |
1091 | kfree(req_data); | |
1092 | } | |
1093 | ||
59c2be1e YS |
1094 | static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) |
1095 | { | |
1096 | ceph_osdc_put_request(req); | |
1097 | } | |
1098 | ||
602adf40 YS |
1099 | /* |
1100 | * Do a synchronous ceph osd operation | |
1101 | */ | |
0ce1a794 | 1102 | static int rbd_req_sync_op(struct rbd_device *rbd_dev, |
602adf40 YS |
1103 | struct ceph_snap_context *snapc, |
1104 | u64 snapid, | |
602adf40 | 1105 | int flags, |
913d2fdc | 1106 | struct ceph_osd_req_op *ops, |
aded07ea | 1107 | const char *object_name, |
602adf40 | 1108 | u64 ofs, u64 len, |
59c2be1e YS |
1109 | char *buf, |
1110 | struct ceph_osd_request **linger_req, | |
1111 | u64 *ver) | |
602adf40 YS |
1112 | { |
1113 | int ret; | |
1114 | struct page **pages; | |
1115 | int num_pages; | |
913d2fdc | 1116 | |
aafb230e | 1117 | rbd_assert(ops != NULL); |
602adf40 YS |
1118 | |
1119 | num_pages = calc_pages_for(ofs , len); | |
1120 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
b8d0638a DC |
1121 | if (IS_ERR(pages)) |
1122 | return PTR_ERR(pages); | |
602adf40 | 1123 | |
0ce1a794 | 1124 | ret = rbd_do_request(NULL, rbd_dev, snapc, snapid, |
aded07ea | 1125 | object_name, ofs, len, NULL, |
602adf40 YS |
1126 | pages, num_pages, |
1127 | flags, | |
1128 | ops, | |
1fec7093 | 1129 | NULL, 0, |
59c2be1e YS |
1130 | NULL, |
1131 | linger_req, ver); | |
602adf40 | 1132 | if (ret < 0) |
913d2fdc | 1133 | goto done; |
602adf40 YS |
1134 | |
1135 | if ((flags & CEPH_OSD_FLAG_READ) && buf) | |
1136 | ret = ceph_copy_from_page_vector(pages, buf, ofs, ret); | |
1137 | ||
602adf40 YS |
1138 | done: |
1139 | ceph_release_page_vector(pages, num_pages); | |
1140 | return ret; | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Do an asynchronous ceph osd operation | |
1145 | */ | |
1146 | static int rbd_do_op(struct request *rq, | |
0ce1a794 | 1147 | struct rbd_device *rbd_dev, |
602adf40 YS |
1148 | struct ceph_snap_context *snapc, |
1149 | u64 snapid, | |
d1f57ea6 | 1150 | int opcode, int flags, |
602adf40 | 1151 | u64 ofs, u64 len, |
1fec7093 YS |
1152 | struct bio *bio, |
1153 | struct rbd_req_coll *coll, | |
1154 | int coll_index) | |
602adf40 YS |
1155 | { |
1156 | char *seg_name; | |
1157 | u64 seg_ofs; | |
1158 | u64 seg_len; | |
1159 | int ret; | |
1160 | struct ceph_osd_req_op *ops; | |
1161 | u32 payload_len; | |
1162 | ||
65ccfe21 | 1163 | seg_name = rbd_segment_name(rbd_dev, ofs); |
602adf40 YS |
1164 | if (!seg_name) |
1165 | return -ENOMEM; | |
65ccfe21 AE |
1166 | seg_len = rbd_segment_length(rbd_dev, ofs, len); |
1167 | seg_ofs = rbd_segment_offset(rbd_dev, ofs); | |
602adf40 YS |
1168 | |
1169 | payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0); | |
1170 | ||
57cfc106 AE |
1171 | ret = -ENOMEM; |
1172 | ops = rbd_create_rw_ops(1, opcode, payload_len); | |
1173 | if (!ops) | |
602adf40 YS |
1174 | goto done; |
1175 | ||
1176 | /* we've taken care of segment sizes earlier when we | |
1177 | cloned the bios. We should never have a segment | |
1178 | truncated at this point */ | |
aafb230e | 1179 | rbd_assert(seg_len == len); |
602adf40 YS |
1180 | |
1181 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1182 | seg_name, seg_ofs, seg_len, | |
1183 | bio, | |
1184 | NULL, 0, | |
1185 | flags, | |
1186 | ops, | |
1fec7093 | 1187 | coll, coll_index, |
59c2be1e | 1188 | rbd_req_cb, 0, NULL); |
11f77002 SW |
1189 | |
1190 | rbd_destroy_ops(ops); | |
602adf40 YS |
1191 | done: |
1192 | kfree(seg_name); | |
1193 | return ret; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Request async osd write | |
1198 | */ | |
1199 | static int rbd_req_write(struct request *rq, | |
1200 | struct rbd_device *rbd_dev, | |
1201 | struct ceph_snap_context *snapc, | |
1202 | u64 ofs, u64 len, | |
1fec7093 YS |
1203 | struct bio *bio, |
1204 | struct rbd_req_coll *coll, | |
1205 | int coll_index) | |
602adf40 YS |
1206 | { |
1207 | return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP, | |
1208 | CEPH_OSD_OP_WRITE, | |
1209 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1fec7093 | 1210 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1211 | } |
1212 | ||
1213 | /* | |
1214 | * Request async osd read | |
1215 | */ | |
1216 | static int rbd_req_read(struct request *rq, | |
1217 | struct rbd_device *rbd_dev, | |
1218 | u64 snapid, | |
1219 | u64 ofs, u64 len, | |
1fec7093 YS |
1220 | struct bio *bio, |
1221 | struct rbd_req_coll *coll, | |
1222 | int coll_index) | |
602adf40 YS |
1223 | { |
1224 | return rbd_do_op(rq, rbd_dev, NULL, | |
b06e6a6b | 1225 | snapid, |
602adf40 YS |
1226 | CEPH_OSD_OP_READ, |
1227 | CEPH_OSD_FLAG_READ, | |
1fec7093 | 1228 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * Request sync osd read | |
1233 | */ | |
0ce1a794 | 1234 | static int rbd_req_sync_read(struct rbd_device *rbd_dev, |
602adf40 | 1235 | u64 snapid, |
aded07ea | 1236 | const char *object_name, |
602adf40 | 1237 | u64 ofs, u64 len, |
59c2be1e YS |
1238 | char *buf, |
1239 | u64 *ver) | |
602adf40 | 1240 | { |
913d2fdc AE |
1241 | struct ceph_osd_req_op *ops; |
1242 | int ret; | |
1243 | ||
1244 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_READ, 0); | |
1245 | if (!ops) | |
1246 | return -ENOMEM; | |
1247 | ||
1248 | ret = rbd_req_sync_op(rbd_dev, NULL, | |
b06e6a6b | 1249 | snapid, |
602adf40 | 1250 | CEPH_OSD_FLAG_READ, |
913d2fdc AE |
1251 | ops, object_name, ofs, len, buf, NULL, ver); |
1252 | rbd_destroy_ops(ops); | |
1253 | ||
1254 | return ret; | |
602adf40 YS |
1255 | } |
1256 | ||
1257 | /* | |
59c2be1e YS |
1258 | * Request sync osd watch |
1259 | */ | |
0ce1a794 | 1260 | static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev, |
59c2be1e | 1261 | u64 ver, |
7f0a24d8 | 1262 | u64 notify_id) |
59c2be1e YS |
1263 | { |
1264 | struct ceph_osd_req_op *ops; | |
11f77002 SW |
1265 | int ret; |
1266 | ||
57cfc106 AE |
1267 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY_ACK, 0); |
1268 | if (!ops) | |
1269 | return -ENOMEM; | |
59c2be1e | 1270 | |
a71b891b | 1271 | ops[0].watch.ver = cpu_to_le64(ver); |
59c2be1e YS |
1272 | ops[0].watch.cookie = notify_id; |
1273 | ops[0].watch.flag = 0; | |
1274 | ||
0ce1a794 | 1275 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
7f0a24d8 | 1276 | rbd_dev->header_name, 0, 0, NULL, |
ad4f232f | 1277 | NULL, 0, |
59c2be1e YS |
1278 | CEPH_OSD_FLAG_READ, |
1279 | ops, | |
1fec7093 | 1280 | NULL, 0, |
59c2be1e YS |
1281 | rbd_simple_req_cb, 0, NULL); |
1282 | ||
1283 | rbd_destroy_ops(ops); | |
1284 | return ret; | |
1285 | } | |
1286 | ||
1287 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1288 | { | |
0ce1a794 | 1289 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
a71b891b | 1290 | u64 hver; |
13143d2d SW |
1291 | int rc; |
1292 | ||
0ce1a794 | 1293 | if (!rbd_dev) |
59c2be1e YS |
1294 | return; |
1295 | ||
bd919d45 AE |
1296 | dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n", |
1297 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1298 | (unsigned int) opcode); | |
1fe5e993 | 1299 | rc = rbd_refresh_header(rbd_dev, &hver); |
13143d2d | 1300 | if (rc) |
f0f8cef5 | 1301 | pr_warning(RBD_DRV_NAME "%d got notification but failed to " |
0ce1a794 | 1302 | " update snaps: %d\n", rbd_dev->major, rc); |
59c2be1e | 1303 | |
7f0a24d8 | 1304 | rbd_req_sync_notify_ack(rbd_dev, hver, notify_id); |
59c2be1e YS |
1305 | } |
1306 | ||
1307 | /* | |
1308 | * Request sync osd watch | |
1309 | */ | |
0e6f322d | 1310 | static int rbd_req_sync_watch(struct rbd_device *rbd_dev) |
59c2be1e YS |
1311 | { |
1312 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1313 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
57cfc106 | 1314 | int ret; |
59c2be1e | 1315 | |
57cfc106 AE |
1316 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1317 | if (!ops) | |
1318 | return -ENOMEM; | |
59c2be1e YS |
1319 | |
1320 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, | |
0ce1a794 | 1321 | (void *)rbd_dev, &rbd_dev->watch_event); |
59c2be1e YS |
1322 | if (ret < 0) |
1323 | goto fail; | |
1324 | ||
0e6f322d | 1325 | ops[0].watch.ver = cpu_to_le64(rbd_dev->header.obj_version); |
0ce1a794 | 1326 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
59c2be1e YS |
1327 | ops[0].watch.flag = 1; |
1328 | ||
0ce1a794 | 1329 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e | 1330 | CEPH_NOSNAP, |
59c2be1e YS |
1331 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1332 | ops, | |
0e6f322d AE |
1333 | rbd_dev->header_name, |
1334 | 0, 0, NULL, | |
0ce1a794 | 1335 | &rbd_dev->watch_request, NULL); |
59c2be1e YS |
1336 | |
1337 | if (ret < 0) | |
1338 | goto fail_event; | |
1339 | ||
1340 | rbd_destroy_ops(ops); | |
1341 | return 0; | |
1342 | ||
1343 | fail_event: | |
0ce1a794 AE |
1344 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1345 | rbd_dev->watch_event = NULL; | |
59c2be1e YS |
1346 | fail: |
1347 | rbd_destroy_ops(ops); | |
1348 | return ret; | |
1349 | } | |
1350 | ||
79e3057c YS |
1351 | /* |
1352 | * Request sync osd unwatch | |
1353 | */ | |
070c633f | 1354 | static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev) |
79e3057c YS |
1355 | { |
1356 | struct ceph_osd_req_op *ops; | |
57cfc106 | 1357 | int ret; |
79e3057c | 1358 | |
57cfc106 AE |
1359 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1360 | if (!ops) | |
1361 | return -ENOMEM; | |
79e3057c YS |
1362 | |
1363 | ops[0].watch.ver = 0; | |
0ce1a794 | 1364 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
79e3057c YS |
1365 | ops[0].watch.flag = 0; |
1366 | ||
0ce1a794 | 1367 | ret = rbd_req_sync_op(rbd_dev, NULL, |
79e3057c | 1368 | CEPH_NOSNAP, |
79e3057c YS |
1369 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1370 | ops, | |
070c633f AE |
1371 | rbd_dev->header_name, |
1372 | 0, 0, NULL, NULL, NULL); | |
1373 | ||
79e3057c YS |
1374 | |
1375 | rbd_destroy_ops(ops); | |
0ce1a794 AE |
1376 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1377 | rbd_dev->watch_event = NULL; | |
79e3057c YS |
1378 | return ret; |
1379 | } | |
1380 | ||
59c2be1e | 1381 | struct rbd_notify_info { |
0ce1a794 | 1382 | struct rbd_device *rbd_dev; |
59c2be1e YS |
1383 | }; |
1384 | ||
1385 | static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1386 | { | |
0ce1a794 AE |
1387 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
1388 | if (!rbd_dev) | |
59c2be1e YS |
1389 | return; |
1390 | ||
bd919d45 AE |
1391 | dout("rbd_notify_cb %s notify_id=%llu opcode=%u\n", |
1392 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1393 | (unsigned int) opcode); | |
59c2be1e YS |
1394 | } |
1395 | ||
1396 | /* | |
1397 | * Request sync osd notify | |
1398 | */ | |
4cb16250 | 1399 | static int rbd_req_sync_notify(struct rbd_device *rbd_dev) |
59c2be1e YS |
1400 | { |
1401 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1402 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
59c2be1e YS |
1403 | struct ceph_osd_event *event; |
1404 | struct rbd_notify_info info; | |
1405 | int payload_len = sizeof(u32) + sizeof(u32); | |
1406 | int ret; | |
1407 | ||
57cfc106 AE |
1408 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY, payload_len); |
1409 | if (!ops) | |
1410 | return -ENOMEM; | |
59c2be1e | 1411 | |
0ce1a794 | 1412 | info.rbd_dev = rbd_dev; |
59c2be1e YS |
1413 | |
1414 | ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1, | |
1415 | (void *)&info, &event); | |
1416 | if (ret < 0) | |
1417 | goto fail; | |
1418 | ||
1419 | ops[0].watch.ver = 1; | |
1420 | ops[0].watch.flag = 1; | |
1421 | ops[0].watch.cookie = event->cookie; | |
1422 | ops[0].watch.prot_ver = RADOS_NOTIFY_VER; | |
1423 | ops[0].watch.timeout = 12; | |
1424 | ||
0ce1a794 | 1425 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e | 1426 | CEPH_NOSNAP, |
59c2be1e YS |
1427 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1428 | ops, | |
4cb16250 AE |
1429 | rbd_dev->header_name, |
1430 | 0, 0, NULL, NULL, NULL); | |
59c2be1e YS |
1431 | if (ret < 0) |
1432 | goto fail_event; | |
1433 | ||
1434 | ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT); | |
1435 | dout("ceph_osdc_wait_event returned %d\n", ret); | |
1436 | rbd_destroy_ops(ops); | |
1437 | return 0; | |
1438 | ||
1439 | fail_event: | |
1440 | ceph_osdc_cancel_event(event); | |
1441 | fail: | |
1442 | rbd_destroy_ops(ops); | |
1443 | return ret; | |
1444 | } | |
1445 | ||
602adf40 YS |
1446 | /* |
1447 | * Request sync osd read | |
1448 | */ | |
0ce1a794 | 1449 | static int rbd_req_sync_exec(struct rbd_device *rbd_dev, |
aded07ea AE |
1450 | const char *object_name, |
1451 | const char *class_name, | |
1452 | const char *method_name, | |
602adf40 | 1453 | const char *data, |
59c2be1e YS |
1454 | int len, |
1455 | u64 *ver) | |
602adf40 YS |
1456 | { |
1457 | struct ceph_osd_req_op *ops; | |
aded07ea AE |
1458 | int class_name_len = strlen(class_name); |
1459 | int method_name_len = strlen(method_name); | |
57cfc106 AE |
1460 | int ret; |
1461 | ||
1462 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_CALL, | |
aded07ea | 1463 | class_name_len + method_name_len + len); |
57cfc106 AE |
1464 | if (!ops) |
1465 | return -ENOMEM; | |
602adf40 | 1466 | |
aded07ea AE |
1467 | ops[0].cls.class_name = class_name; |
1468 | ops[0].cls.class_len = (__u8) class_name_len; | |
1469 | ops[0].cls.method_name = method_name; | |
1470 | ops[0].cls.method_len = (__u8) method_name_len; | |
602adf40 YS |
1471 | ops[0].cls.argc = 0; |
1472 | ops[0].cls.indata = data; | |
1473 | ops[0].cls.indata_len = len; | |
1474 | ||
0ce1a794 | 1475 | ret = rbd_req_sync_op(rbd_dev, NULL, |
602adf40 | 1476 | CEPH_NOSNAP, |
602adf40 YS |
1477 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1478 | ops, | |
d1f57ea6 | 1479 | object_name, 0, 0, NULL, NULL, ver); |
602adf40 YS |
1480 | |
1481 | rbd_destroy_ops(ops); | |
1482 | ||
1483 | dout("cls_exec returned %d\n", ret); | |
1484 | return ret; | |
1485 | } | |
1486 | ||
1fec7093 YS |
1487 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1488 | { | |
1489 | struct rbd_req_coll *coll = | |
1490 | kzalloc(sizeof(struct rbd_req_coll) + | |
1491 | sizeof(struct rbd_req_status) * num_reqs, | |
1492 | GFP_ATOMIC); | |
1493 | ||
1494 | if (!coll) | |
1495 | return NULL; | |
1496 | coll->total = num_reqs; | |
1497 | kref_init(&coll->kref); | |
1498 | return coll; | |
1499 | } | |
1500 | ||
602adf40 YS |
1501 | /* |
1502 | * block device queue callback | |
1503 | */ | |
1504 | static void rbd_rq_fn(struct request_queue *q) | |
1505 | { | |
1506 | struct rbd_device *rbd_dev = q->queuedata; | |
1507 | struct request *rq; | |
1508 | struct bio_pair *bp = NULL; | |
1509 | ||
00f1f36f | 1510 | while ((rq = blk_fetch_request(q))) { |
602adf40 YS |
1511 | struct bio *bio; |
1512 | struct bio *rq_bio, *next_bio = NULL; | |
1513 | bool do_write; | |
bd919d45 AE |
1514 | unsigned int size; |
1515 | u64 op_size = 0; | |
602adf40 | 1516 | u64 ofs; |
1fec7093 YS |
1517 | int num_segs, cur_seg = 0; |
1518 | struct rbd_req_coll *coll; | |
d1d25646 | 1519 | struct ceph_snap_context *snapc; |
602adf40 | 1520 | |
602adf40 YS |
1521 | dout("fetched request\n"); |
1522 | ||
1523 | /* filter out block requests we don't understand */ | |
1524 | if ((rq->cmd_type != REQ_TYPE_FS)) { | |
1525 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1526 | continue; |
602adf40 YS |
1527 | } |
1528 | ||
1529 | /* deduce our operation (read, write) */ | |
1530 | do_write = (rq_data_dir(rq) == WRITE); | |
1531 | ||
1532 | size = blk_rq_bytes(rq); | |
593a9e7b | 1533 | ofs = blk_rq_pos(rq) * SECTOR_SIZE; |
602adf40 | 1534 | rq_bio = rq->bio; |
f84344f3 | 1535 | if (do_write && rbd_dev->mapping.read_only) { |
602adf40 | 1536 | __blk_end_request_all(rq, -EROFS); |
00f1f36f | 1537 | continue; |
602adf40 YS |
1538 | } |
1539 | ||
1540 | spin_unlock_irq(q->queue_lock); | |
1541 | ||
d1d25646 | 1542 | down_read(&rbd_dev->header_rwsem); |
e88a36ec | 1543 | |
f84344f3 AE |
1544 | if (rbd_dev->mapping.snap_id != CEPH_NOSNAP && |
1545 | !rbd_dev->mapping.snap_exists) { | |
e88a36ec | 1546 | up_read(&rbd_dev->header_rwsem); |
d1d25646 JD |
1547 | dout("request for non-existent snapshot"); |
1548 | spin_lock_irq(q->queue_lock); | |
1549 | __blk_end_request_all(rq, -ENXIO); | |
1550 | continue; | |
e88a36ec JD |
1551 | } |
1552 | ||
d1d25646 JD |
1553 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
1554 | ||
1555 | up_read(&rbd_dev->header_rwsem); | |
1556 | ||
602adf40 YS |
1557 | dout("%s 0x%x bytes at 0x%llx\n", |
1558 | do_write ? "write" : "read", | |
bd919d45 | 1559 | size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE); |
602adf40 | 1560 | |
1fec7093 | 1561 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); |
df111be6 AE |
1562 | if (num_segs <= 0) { |
1563 | spin_lock_irq(q->queue_lock); | |
1564 | __blk_end_request_all(rq, num_segs); | |
1565 | ceph_put_snap_context(snapc); | |
1566 | continue; | |
1567 | } | |
1fec7093 YS |
1568 | coll = rbd_alloc_coll(num_segs); |
1569 | if (!coll) { | |
1570 | spin_lock_irq(q->queue_lock); | |
1571 | __blk_end_request_all(rq, -ENOMEM); | |
d1d25646 | 1572 | ceph_put_snap_context(snapc); |
00f1f36f | 1573 | continue; |
1fec7093 YS |
1574 | } |
1575 | ||
602adf40 YS |
1576 | do { |
1577 | /* a bio clone to be passed down to OSD req */ | |
bd919d45 | 1578 | dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt); |
65ccfe21 | 1579 | op_size = rbd_segment_length(rbd_dev, ofs, size); |
1fec7093 | 1580 | kref_get(&coll->kref); |
602adf40 YS |
1581 | bio = bio_chain_clone(&rq_bio, &next_bio, &bp, |
1582 | op_size, GFP_ATOMIC); | |
1583 | if (!bio) { | |
1fec7093 YS |
1584 | rbd_coll_end_req_index(rq, coll, cur_seg, |
1585 | -ENOMEM, op_size); | |
1586 | goto next_seg; | |
602adf40 YS |
1587 | } |
1588 | ||
1fec7093 | 1589 | |
602adf40 YS |
1590 | /* init OSD command: write or read */ |
1591 | if (do_write) | |
1592 | rbd_req_write(rq, rbd_dev, | |
d1d25646 | 1593 | snapc, |
602adf40 | 1594 | ofs, |
1fec7093 YS |
1595 | op_size, bio, |
1596 | coll, cur_seg); | |
602adf40 YS |
1597 | else |
1598 | rbd_req_read(rq, rbd_dev, | |
f84344f3 | 1599 | rbd_dev->mapping.snap_id, |
602adf40 | 1600 | ofs, |
1fec7093 YS |
1601 | op_size, bio, |
1602 | coll, cur_seg); | |
602adf40 | 1603 | |
1fec7093 | 1604 | next_seg: |
602adf40 YS |
1605 | size -= op_size; |
1606 | ofs += op_size; | |
1607 | ||
1fec7093 | 1608 | cur_seg++; |
602adf40 YS |
1609 | rq_bio = next_bio; |
1610 | } while (size > 0); | |
1fec7093 | 1611 | kref_put(&coll->kref, rbd_coll_release); |
602adf40 YS |
1612 | |
1613 | if (bp) | |
1614 | bio_pair_release(bp); | |
602adf40 | 1615 | spin_lock_irq(q->queue_lock); |
d1d25646 JD |
1616 | |
1617 | ceph_put_snap_context(snapc); | |
602adf40 YS |
1618 | } |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1623 | * multiple osd objects. One exception would be with a single page bios, | |
1624 | * which we handle later at bio_chain_clone | |
1625 | */ | |
1626 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1627 | struct bio_vec *bvec) | |
1628 | { | |
1629 | struct rbd_device *rbd_dev = q->queuedata; | |
593a9e7b AE |
1630 | unsigned int chunk_sectors; |
1631 | sector_t sector; | |
1632 | unsigned int bio_sectors; | |
602adf40 YS |
1633 | int max; |
1634 | ||
593a9e7b AE |
1635 | chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); |
1636 | sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev); | |
1637 | bio_sectors = bmd->bi_size >> SECTOR_SHIFT; | |
1638 | ||
602adf40 | 1639 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) |
593a9e7b | 1640 | + bio_sectors)) << SECTOR_SHIFT; |
602adf40 YS |
1641 | if (max < 0) |
1642 | max = 0; /* bio_add cannot handle a negative return */ | |
1643 | if (max <= bvec->bv_len && bio_sectors == 0) | |
1644 | return bvec->bv_len; | |
1645 | return max; | |
1646 | } | |
1647 | ||
1648 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1649 | { | |
1650 | struct gendisk *disk = rbd_dev->disk; | |
1651 | ||
1652 | if (!disk) | |
1653 | return; | |
1654 | ||
1655 | rbd_header_free(&rbd_dev->header); | |
1656 | ||
1657 | if (disk->flags & GENHD_FL_UP) | |
1658 | del_gendisk(disk); | |
1659 | if (disk->queue) | |
1660 | blk_cleanup_queue(disk->queue); | |
1661 | put_disk(disk); | |
1662 | } | |
1663 | ||
1664 | /* | |
4156d998 AE |
1665 | * Read the complete header for the given rbd device. |
1666 | * | |
1667 | * Returns a pointer to a dynamically-allocated buffer containing | |
1668 | * the complete and validated header. Caller can pass the address | |
1669 | * of a variable that will be filled in with the version of the | |
1670 | * header object at the time it was read. | |
1671 | * | |
1672 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 1673 | */ |
4156d998 AE |
1674 | static struct rbd_image_header_ondisk * |
1675 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 1676 | { |
4156d998 | 1677 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 1678 | u32 snap_count = 0; |
4156d998 AE |
1679 | u64 names_size = 0; |
1680 | u32 want_count; | |
1681 | int ret; | |
602adf40 | 1682 | |
00f1f36f | 1683 | /* |
4156d998 AE |
1684 | * The complete header will include an array of its 64-bit |
1685 | * snapshot ids, followed by the names of those snapshots as | |
1686 | * a contiguous block of NUL-terminated strings. Note that | |
1687 | * the number of snapshots could change by the time we read | |
1688 | * it in, in which case we re-read it. | |
00f1f36f | 1689 | */ |
4156d998 AE |
1690 | do { |
1691 | size_t size; | |
1692 | ||
1693 | kfree(ondisk); | |
1694 | ||
1695 | size = sizeof (*ondisk); | |
1696 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
1697 | size += names_size; | |
1698 | ondisk = kmalloc(size, GFP_KERNEL); | |
1699 | if (!ondisk) | |
1700 | return ERR_PTR(-ENOMEM); | |
1701 | ||
1702 | ret = rbd_req_sync_read(rbd_dev, CEPH_NOSNAP, | |
0bed54dc | 1703 | rbd_dev->header_name, |
4156d998 AE |
1704 | 0, size, |
1705 | (char *) ondisk, version); | |
1706 | ||
1707 | if (ret < 0) | |
1708 | goto out_err; | |
1709 | if (WARN_ON((size_t) ret < size)) { | |
1710 | ret = -ENXIO; | |
1711 | pr_warning("short header read for image %s" | |
1712 | " (want %zd got %d)\n", | |
1713 | rbd_dev->image_name, size, ret); | |
1714 | goto out_err; | |
1715 | } | |
1716 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
1717 | ret = -ENXIO; | |
1718 | pr_warning("invalid header for image %s\n", | |
1719 | rbd_dev->image_name); | |
1720 | goto out_err; | |
81e759fb | 1721 | } |
602adf40 | 1722 | |
4156d998 AE |
1723 | names_size = le64_to_cpu(ondisk->snap_names_len); |
1724 | want_count = snap_count; | |
1725 | snap_count = le32_to_cpu(ondisk->snap_count); | |
1726 | } while (snap_count != want_count); | |
00f1f36f | 1727 | |
4156d998 | 1728 | return ondisk; |
00f1f36f | 1729 | |
4156d998 AE |
1730 | out_err: |
1731 | kfree(ondisk); | |
1732 | ||
1733 | return ERR_PTR(ret); | |
1734 | } | |
1735 | ||
1736 | /* | |
1737 | * reload the ondisk the header | |
1738 | */ | |
1739 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1740 | struct rbd_image_header *header) | |
1741 | { | |
1742 | struct rbd_image_header_ondisk *ondisk; | |
1743 | u64 ver = 0; | |
1744 | int ret; | |
602adf40 | 1745 | |
4156d998 AE |
1746 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
1747 | if (IS_ERR(ondisk)) | |
1748 | return PTR_ERR(ondisk); | |
1749 | ret = rbd_header_from_disk(header, ondisk); | |
1750 | if (ret >= 0) | |
1751 | header->obj_version = ver; | |
1752 | kfree(ondisk); | |
1753 | ||
1754 | return ret; | |
602adf40 YS |
1755 | } |
1756 | ||
1757 | /* | |
1758 | * create a snapshot | |
1759 | */ | |
0ce1a794 | 1760 | static int rbd_header_add_snap(struct rbd_device *rbd_dev, |
602adf40 YS |
1761 | const char *snap_name, |
1762 | gfp_t gfp_flags) | |
1763 | { | |
1764 | int name_len = strlen(snap_name); | |
1765 | u64 new_snapid; | |
1766 | int ret; | |
916d4d67 | 1767 | void *data, *p, *e; |
1dbb4399 | 1768 | struct ceph_mon_client *monc; |
602adf40 YS |
1769 | |
1770 | /* we should create a snapshot only if we're pointing at the head */ | |
f84344f3 | 1771 | if (rbd_dev->mapping.snap_id != CEPH_NOSNAP) |
602adf40 YS |
1772 | return -EINVAL; |
1773 | ||
0ce1a794 AE |
1774 | monc = &rbd_dev->rbd_client->client->monc; |
1775 | ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid); | |
bd919d45 | 1776 | dout("created snapid=%llu\n", (unsigned long long) new_snapid); |
602adf40 YS |
1777 | if (ret < 0) |
1778 | return ret; | |
1779 | ||
1780 | data = kmalloc(name_len + 16, gfp_flags); | |
1781 | if (!data) | |
1782 | return -ENOMEM; | |
1783 | ||
916d4d67 SW |
1784 | p = data; |
1785 | e = data + name_len + 16; | |
602adf40 | 1786 | |
916d4d67 SW |
1787 | ceph_encode_string_safe(&p, e, snap_name, name_len, bad); |
1788 | ceph_encode_64_safe(&p, e, new_snapid, bad); | |
602adf40 | 1789 | |
0bed54dc | 1790 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, |
0ce1a794 | 1791 | "rbd", "snap_add", |
d67d4be5 | 1792 | data, p - data, NULL); |
602adf40 | 1793 | |
916d4d67 | 1794 | kfree(data); |
602adf40 | 1795 | |
505cbb9b | 1796 | return ret < 0 ? ret : 0; |
602adf40 YS |
1797 | bad: |
1798 | return -ERANGE; | |
1799 | } | |
1800 | ||
dfc5606d YS |
1801 | static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
1802 | { | |
1803 | struct rbd_snap *snap; | |
a0593290 | 1804 | struct rbd_snap *next; |
dfc5606d | 1805 | |
a0593290 | 1806 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
14e7085d | 1807 | __rbd_remove_snap_dev(snap); |
dfc5606d YS |
1808 | } |
1809 | ||
602adf40 YS |
1810 | /* |
1811 | * only read the first part of the ondisk header, without the snaps info | |
1812 | */ | |
b813623a | 1813 | static int __rbd_refresh_header(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
1814 | { |
1815 | int ret; | |
1816 | struct rbd_image_header h; | |
602adf40 YS |
1817 | |
1818 | ret = rbd_read_header(rbd_dev, &h); | |
1819 | if (ret < 0) | |
1820 | return ret; | |
1821 | ||
a51aa0c0 JD |
1822 | down_write(&rbd_dev->header_rwsem); |
1823 | ||
9db4b3e3 | 1824 | /* resized? */ |
f84344f3 | 1825 | if (rbd_dev->mapping.snap_id == CEPH_NOSNAP) { |
474ef7ce JD |
1826 | sector_t size = (sector_t) h.image_size / SECTOR_SIZE; |
1827 | ||
99c1f08f AE |
1828 | if (size != (sector_t) rbd_dev->mapping.size) { |
1829 | dout("setting size to %llu sectors", | |
1830 | (unsigned long long) size); | |
1831 | rbd_dev->mapping.size = (u64) size; | |
1832 | set_capacity(rbd_dev->disk, size); | |
1833 | } | |
474ef7ce | 1834 | } |
9db4b3e3 | 1835 | |
849b4260 | 1836 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 1837 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 1838 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
1839 | /* osd requests may still refer to snapc */ |
1840 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 1841 | |
b813623a AE |
1842 | if (hver) |
1843 | *hver = h.obj_version; | |
a71b891b | 1844 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 1845 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
1846 | rbd_dev->header.snapc = h.snapc; |
1847 | rbd_dev->header.snap_names = h.snap_names; | |
1848 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
1849 | /* Free the extra copy of the object prefix */ |
1850 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
1851 | kfree(h.object_prefix); | |
1852 | ||
9fcbb800 | 1853 | ret = rbd_dev_snap_devs_update(rbd_dev); |
dfc5606d | 1854 | |
c666601a | 1855 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1856 | |
dfc5606d | 1857 | return ret; |
602adf40 YS |
1858 | } |
1859 | ||
1fe5e993 AE |
1860 | static int rbd_refresh_header(struct rbd_device *rbd_dev, u64 *hver) |
1861 | { | |
1862 | int ret; | |
1863 | ||
1864 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
1865 | ret = __rbd_refresh_header(rbd_dev, hver); | |
1866 | mutex_unlock(&ctl_mutex); | |
1867 | ||
1868 | return ret; | |
1869 | } | |
1870 | ||
602adf40 YS |
1871 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
1872 | { | |
1873 | struct gendisk *disk; | |
1874 | struct request_queue *q; | |
1875 | int rc; | |
593a9e7b | 1876 | u64 segment_size; |
602adf40 YS |
1877 | |
1878 | /* contact OSD, request size info about the object being mapped */ | |
1879 | rc = rbd_read_header(rbd_dev, &rbd_dev->header); | |
1880 | if (rc) | |
1881 | return rc; | |
1882 | ||
dfc5606d | 1883 | /* no need to lock here, as rbd_dev is not registered yet */ |
9fcbb800 | 1884 | rc = rbd_dev_snap_devs_update(rbd_dev); |
dfc5606d YS |
1885 | if (rc) |
1886 | return rc; | |
1887 | ||
4e1105a2 | 1888 | rc = rbd_header_set_snap(rbd_dev, snap_name); |
602adf40 YS |
1889 | if (rc) |
1890 | return rc; | |
1891 | ||
1892 | /* create gendisk info */ | |
1893 | rc = -ENOMEM; | |
1894 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); | |
1895 | if (!disk) | |
1896 | goto out; | |
1897 | ||
f0f8cef5 | 1898 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 1899 | rbd_dev->dev_id); |
602adf40 YS |
1900 | disk->major = rbd_dev->major; |
1901 | disk->first_minor = 0; | |
1902 | disk->fops = &rbd_bd_ops; | |
1903 | disk->private_data = rbd_dev; | |
1904 | ||
1905 | /* init rq */ | |
1906 | rc = -ENOMEM; | |
1907 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); | |
1908 | if (!q) | |
1909 | goto out_disk; | |
029bcbd8 | 1910 | |
593a9e7b AE |
1911 | /* We use the default size, but let's be explicit about it. */ |
1912 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1913 | ||
029bcbd8 | 1914 | /* set io sizes to object size */ |
593a9e7b AE |
1915 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1916 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1917 | blk_queue_max_segment_size(q, segment_size); | |
1918 | blk_queue_io_min(q, segment_size); | |
1919 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1920 | |
602adf40 YS |
1921 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1922 | disk->queue = q; | |
1923 | ||
1924 | q->queuedata = rbd_dev; | |
1925 | ||
1926 | rbd_dev->disk = disk; | |
602adf40 YS |
1927 | |
1928 | /* finally, announce the disk to the world */ | |
99c1f08f | 1929 | set_capacity(disk, (sector_t) rbd_dev->mapping.size / SECTOR_SIZE); |
602adf40 YS |
1930 | add_disk(disk); |
1931 | ||
1932 | pr_info("%s: added with size 0x%llx\n", | |
99c1f08f | 1933 | disk->disk_name, (unsigned long long) rbd_dev->mapping.size); |
602adf40 YS |
1934 | return 0; |
1935 | ||
1936 | out_disk: | |
1937 | put_disk(disk); | |
1938 | out: | |
1939 | return rc; | |
1940 | } | |
1941 | ||
dfc5606d YS |
1942 | /* |
1943 | sysfs | |
1944 | */ | |
1945 | ||
593a9e7b AE |
1946 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1947 | { | |
1948 | return container_of(dev, struct rbd_device, dev); | |
1949 | } | |
1950 | ||
dfc5606d YS |
1951 | static ssize_t rbd_size_show(struct device *dev, |
1952 | struct device_attribute *attr, char *buf) | |
1953 | { | |
593a9e7b | 1954 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
1955 | sector_t size; |
1956 | ||
1957 | down_read(&rbd_dev->header_rwsem); | |
1958 | size = get_capacity(rbd_dev->disk); | |
1959 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 1960 | |
a51aa0c0 | 1961 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
1962 | } |
1963 | ||
1964 | static ssize_t rbd_major_show(struct device *dev, | |
1965 | struct device_attribute *attr, char *buf) | |
1966 | { | |
593a9e7b | 1967 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 1968 | |
dfc5606d YS |
1969 | return sprintf(buf, "%d\n", rbd_dev->major); |
1970 | } | |
1971 | ||
1972 | static ssize_t rbd_client_id_show(struct device *dev, | |
1973 | struct device_attribute *attr, char *buf) | |
602adf40 | 1974 | { |
593a9e7b | 1975 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1976 | |
1dbb4399 AE |
1977 | return sprintf(buf, "client%lld\n", |
1978 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
1979 | } |
1980 | ||
dfc5606d YS |
1981 | static ssize_t rbd_pool_show(struct device *dev, |
1982 | struct device_attribute *attr, char *buf) | |
602adf40 | 1983 | { |
593a9e7b | 1984 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1985 | |
1986 | return sprintf(buf, "%s\n", rbd_dev->pool_name); | |
1987 | } | |
1988 | ||
9bb2f334 AE |
1989 | static ssize_t rbd_pool_id_show(struct device *dev, |
1990 | struct device_attribute *attr, char *buf) | |
1991 | { | |
1992 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1993 | ||
1994 | return sprintf(buf, "%d\n", rbd_dev->pool_id); | |
1995 | } | |
1996 | ||
dfc5606d YS |
1997 | static ssize_t rbd_name_show(struct device *dev, |
1998 | struct device_attribute *attr, char *buf) | |
1999 | { | |
593a9e7b | 2000 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2001 | |
0bed54dc | 2002 | return sprintf(buf, "%s\n", rbd_dev->image_name); |
dfc5606d YS |
2003 | } |
2004 | ||
2005 | static ssize_t rbd_snap_show(struct device *dev, | |
2006 | struct device_attribute *attr, | |
2007 | char *buf) | |
2008 | { | |
593a9e7b | 2009 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2010 | |
f84344f3 | 2011 | return sprintf(buf, "%s\n", rbd_dev->mapping.snap_name); |
dfc5606d YS |
2012 | } |
2013 | ||
2014 | static ssize_t rbd_image_refresh(struct device *dev, | |
2015 | struct device_attribute *attr, | |
2016 | const char *buf, | |
2017 | size_t size) | |
2018 | { | |
593a9e7b | 2019 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2020 | int ret; |
602adf40 | 2021 | |
1fe5e993 | 2022 | ret = rbd_refresh_header(rbd_dev, NULL); |
b813623a AE |
2023 | |
2024 | return ret < 0 ? ret : size; | |
dfc5606d | 2025 | } |
602adf40 | 2026 | |
dfc5606d YS |
2027 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
2028 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); | |
2029 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2030 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2031 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d YS |
2032 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
2033 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); | |
2034 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
2035 | static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add); | |
dfc5606d YS |
2036 | |
2037 | static struct attribute *rbd_attrs[] = { | |
2038 | &dev_attr_size.attr, | |
2039 | &dev_attr_major.attr, | |
2040 | &dev_attr_client_id.attr, | |
2041 | &dev_attr_pool.attr, | |
9bb2f334 | 2042 | &dev_attr_pool_id.attr, |
dfc5606d YS |
2043 | &dev_attr_name.attr, |
2044 | &dev_attr_current_snap.attr, | |
2045 | &dev_attr_refresh.attr, | |
2046 | &dev_attr_create_snap.attr, | |
dfc5606d YS |
2047 | NULL |
2048 | }; | |
2049 | ||
2050 | static struct attribute_group rbd_attr_group = { | |
2051 | .attrs = rbd_attrs, | |
2052 | }; | |
2053 | ||
2054 | static const struct attribute_group *rbd_attr_groups[] = { | |
2055 | &rbd_attr_group, | |
2056 | NULL | |
2057 | }; | |
2058 | ||
2059 | static void rbd_sysfs_dev_release(struct device *dev) | |
2060 | { | |
2061 | } | |
2062 | ||
2063 | static struct device_type rbd_device_type = { | |
2064 | .name = "rbd", | |
2065 | .groups = rbd_attr_groups, | |
2066 | .release = rbd_sysfs_dev_release, | |
2067 | }; | |
2068 | ||
2069 | ||
2070 | /* | |
2071 | sysfs - snapshots | |
2072 | */ | |
2073 | ||
2074 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2075 | struct device_attribute *attr, | |
2076 | char *buf) | |
2077 | { | |
2078 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2079 | ||
3591538f | 2080 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2081 | } |
2082 | ||
2083 | static ssize_t rbd_snap_id_show(struct device *dev, | |
2084 | struct device_attribute *attr, | |
2085 | char *buf) | |
2086 | { | |
2087 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2088 | ||
3591538f | 2089 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
2090 | } |
2091 | ||
2092 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); | |
2093 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
2094 | ||
2095 | static struct attribute *rbd_snap_attrs[] = { | |
2096 | &dev_attr_snap_size.attr, | |
2097 | &dev_attr_snap_id.attr, | |
2098 | NULL, | |
2099 | }; | |
2100 | ||
2101 | static struct attribute_group rbd_snap_attr_group = { | |
2102 | .attrs = rbd_snap_attrs, | |
2103 | }; | |
2104 | ||
2105 | static void rbd_snap_dev_release(struct device *dev) | |
2106 | { | |
2107 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2108 | kfree(snap->name); | |
2109 | kfree(snap); | |
2110 | } | |
2111 | ||
2112 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2113 | &rbd_snap_attr_group, | |
2114 | NULL | |
2115 | }; | |
2116 | ||
2117 | static struct device_type rbd_snap_device_type = { | |
2118 | .groups = rbd_snap_attr_groups, | |
2119 | .release = rbd_snap_dev_release, | |
2120 | }; | |
2121 | ||
14e7085d | 2122 | static void __rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2123 | { |
2124 | list_del(&snap->node); | |
2125 | device_unregister(&snap->dev); | |
2126 | } | |
2127 | ||
14e7085d | 2128 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2129 | struct device *parent) |
2130 | { | |
2131 | struct device *dev = &snap->dev; | |
2132 | int ret; | |
2133 | ||
2134 | dev->type = &rbd_snap_device_type; | |
2135 | dev->parent = parent; | |
2136 | dev->release = rbd_snap_dev_release; | |
2137 | dev_set_name(dev, "snap_%s", snap->name); | |
2138 | ret = device_register(dev); | |
2139 | ||
2140 | return ret; | |
2141 | } | |
2142 | ||
4e891e0a AE |
2143 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
2144 | int i, const char *name) | |
dfc5606d | 2145 | { |
4e891e0a | 2146 | struct rbd_snap *snap; |
dfc5606d | 2147 | int ret; |
4e891e0a AE |
2148 | |
2149 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2150 | if (!snap) |
4e891e0a AE |
2151 | return ERR_PTR(-ENOMEM); |
2152 | ||
2153 | ret = -ENOMEM; | |
dfc5606d | 2154 | snap->name = kstrdup(name, GFP_KERNEL); |
4e891e0a AE |
2155 | if (!snap->name) |
2156 | goto err; | |
2157 | ||
dfc5606d YS |
2158 | snap->size = rbd_dev->header.snap_sizes[i]; |
2159 | snap->id = rbd_dev->header.snapc->snaps[i]; | |
2160 | if (device_is_registered(&rbd_dev->dev)) { | |
14e7085d | 2161 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); |
dfc5606d YS |
2162 | if (ret < 0) |
2163 | goto err; | |
2164 | } | |
4e891e0a AE |
2165 | |
2166 | return snap; | |
2167 | ||
dfc5606d YS |
2168 | err: |
2169 | kfree(snap->name); | |
2170 | kfree(snap); | |
4e891e0a AE |
2171 | |
2172 | return ERR_PTR(ret); | |
dfc5606d YS |
2173 | } |
2174 | ||
2175 | /* | |
35938150 AE |
2176 | * Scan the rbd device's current snapshot list and compare it to the |
2177 | * newly-received snapshot context. Remove any existing snapshots | |
2178 | * not present in the new snapshot context. Add a new snapshot for | |
2179 | * any snaphots in the snapshot context not in the current list. | |
2180 | * And verify there are no changes to snapshots we already know | |
2181 | * about. | |
2182 | * | |
2183 | * Assumes the snapshots in the snapshot context are sorted by | |
2184 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
2185 | * are also maintained in that order.) | |
dfc5606d | 2186 | */ |
9fcbb800 | 2187 | static int rbd_dev_snap_devs_update(struct rbd_device *rbd_dev) |
dfc5606d | 2188 | { |
35938150 AE |
2189 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
2190 | const u32 snap_count = snapc->num_snaps; | |
2191 | char *snap_name = rbd_dev->header.snap_names; | |
2192 | struct list_head *head = &rbd_dev->snaps; | |
2193 | struct list_head *links = head->next; | |
2194 | u32 index = 0; | |
dfc5606d | 2195 | |
9fcbb800 | 2196 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
2197 | while (index < snap_count || links != head) { |
2198 | u64 snap_id; | |
2199 | struct rbd_snap *snap; | |
dfc5606d | 2200 | |
35938150 AE |
2201 | snap_id = index < snap_count ? snapc->snaps[index] |
2202 | : CEPH_NOSNAP; | |
2203 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
2204 | : NULL; | |
aafb230e | 2205 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 2206 | |
35938150 AE |
2207 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
2208 | struct list_head *next = links->next; | |
dfc5606d | 2209 | |
35938150 | 2210 | /* Existing snapshot not in the new snap context */ |
dfc5606d | 2211 | |
f84344f3 AE |
2212 | if (rbd_dev->mapping.snap_id == snap->id) |
2213 | rbd_dev->mapping.snap_exists = false; | |
35938150 | 2214 | __rbd_remove_snap_dev(snap); |
9fcbb800 | 2215 | dout("%ssnap id %llu has been removed\n", |
f84344f3 AE |
2216 | rbd_dev->mapping.snap_id == snap->id ? |
2217 | "mapped " : "", | |
9fcbb800 | 2218 | (unsigned long long) snap->id); |
35938150 AE |
2219 | |
2220 | /* Done with this list entry; advance */ | |
2221 | ||
2222 | links = next; | |
dfc5606d YS |
2223 | continue; |
2224 | } | |
35938150 | 2225 | |
9fcbb800 AE |
2226 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
2227 | (unsigned long long) snap_id); | |
35938150 AE |
2228 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
2229 | struct rbd_snap *new_snap; | |
2230 | ||
2231 | /* We haven't seen this snapshot before */ | |
2232 | ||
2233 | new_snap = __rbd_add_snap_dev(rbd_dev, index, | |
2234 | snap_name); | |
9fcbb800 AE |
2235 | if (IS_ERR(new_snap)) { |
2236 | int err = PTR_ERR(new_snap); | |
2237 | ||
2238 | dout(" failed to add dev, error %d\n", err); | |
2239 | ||
2240 | return err; | |
2241 | } | |
35938150 AE |
2242 | |
2243 | /* New goes before existing, or at end of list */ | |
2244 | ||
9fcbb800 | 2245 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
2246 | if (snap) |
2247 | list_add_tail(&new_snap->node, &snap->node); | |
2248 | else | |
523f3258 | 2249 | list_add_tail(&new_snap->node, head); |
35938150 AE |
2250 | } else { |
2251 | /* Already have this one */ | |
2252 | ||
9fcbb800 AE |
2253 | dout(" already present\n"); |
2254 | ||
aafb230e AE |
2255 | rbd_assert(snap->size == |
2256 | rbd_dev->header.snap_sizes[index]); | |
2257 | rbd_assert(!strcmp(snap->name, snap_name)); | |
35938150 AE |
2258 | |
2259 | /* Done with this list entry; advance */ | |
2260 | ||
2261 | links = links->next; | |
dfc5606d | 2262 | } |
35938150 AE |
2263 | |
2264 | /* Advance to the next entry in the snapshot context */ | |
2265 | ||
2266 | index++; | |
2267 | snap_name += strlen(snap_name) + 1; | |
dfc5606d | 2268 | } |
9fcbb800 | 2269 | dout("%s: done\n", __func__); |
dfc5606d YS |
2270 | |
2271 | return 0; | |
2272 | } | |
2273 | ||
dfc5606d YS |
2274 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
2275 | { | |
f0f8cef5 | 2276 | int ret; |
dfc5606d YS |
2277 | struct device *dev; |
2278 | struct rbd_snap *snap; | |
2279 | ||
2280 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2281 | dev = &rbd_dev->dev; | |
2282 | ||
2283 | dev->bus = &rbd_bus_type; | |
2284 | dev->type = &rbd_device_type; | |
2285 | dev->parent = &rbd_root_dev; | |
2286 | dev->release = rbd_dev_release; | |
de71a297 | 2287 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d YS |
2288 | ret = device_register(dev); |
2289 | if (ret < 0) | |
f0f8cef5 | 2290 | goto out; |
dfc5606d YS |
2291 | |
2292 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
14e7085d | 2293 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); |
dfc5606d | 2294 | if (ret < 0) |
602adf40 YS |
2295 | break; |
2296 | } | |
f0f8cef5 | 2297 | out: |
dfc5606d YS |
2298 | mutex_unlock(&ctl_mutex); |
2299 | return ret; | |
602adf40 YS |
2300 | } |
2301 | ||
dfc5606d YS |
2302 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
2303 | { | |
2304 | device_unregister(&rbd_dev->dev); | |
2305 | } | |
2306 | ||
59c2be1e YS |
2307 | static int rbd_init_watch_dev(struct rbd_device *rbd_dev) |
2308 | { | |
2309 | int ret, rc; | |
2310 | ||
2311 | do { | |
0e6f322d | 2312 | ret = rbd_req_sync_watch(rbd_dev); |
59c2be1e | 2313 | if (ret == -ERANGE) { |
1fe5e993 | 2314 | rc = rbd_refresh_header(rbd_dev, NULL); |
59c2be1e YS |
2315 | if (rc < 0) |
2316 | return rc; | |
2317 | } | |
2318 | } while (ret == -ERANGE); | |
2319 | ||
2320 | return ret; | |
2321 | } | |
2322 | ||
e2839308 | 2323 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
2324 | |
2325 | /* | |
499afd5b AE |
2326 | * Get a unique rbd identifier for the given new rbd_dev, and add |
2327 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 2328 | */ |
e2839308 | 2329 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 2330 | { |
e2839308 | 2331 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
2332 | |
2333 | spin_lock(&rbd_dev_list_lock); | |
2334 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
2335 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
2336 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
2337 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 2338 | } |
b7f23c36 | 2339 | |
1ddbe94e | 2340 | /* |
499afd5b AE |
2341 | * Remove an rbd_dev from the global list, and record that its |
2342 | * identifier is no longer in use. | |
1ddbe94e | 2343 | */ |
e2839308 | 2344 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 2345 | { |
d184f6bf | 2346 | struct list_head *tmp; |
de71a297 | 2347 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
2348 | int max_id; |
2349 | ||
aafb230e | 2350 | rbd_assert(rbd_id > 0); |
499afd5b | 2351 | |
e2839308 AE |
2352 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
2353 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
2354 | spin_lock(&rbd_dev_list_lock); |
2355 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
2356 | |
2357 | /* | |
2358 | * If the id being "put" is not the current maximum, there | |
2359 | * is nothing special we need to do. | |
2360 | */ | |
e2839308 | 2361 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
2362 | spin_unlock(&rbd_dev_list_lock); |
2363 | return; | |
2364 | } | |
2365 | ||
2366 | /* | |
2367 | * We need to update the current maximum id. Search the | |
2368 | * list to find out what it is. We're more likely to find | |
2369 | * the maximum at the end, so search the list backward. | |
2370 | */ | |
2371 | max_id = 0; | |
2372 | list_for_each_prev(tmp, &rbd_dev_list) { | |
2373 | struct rbd_device *rbd_dev; | |
2374 | ||
2375 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
2376 | if (rbd_id > max_id) | |
2377 | max_id = rbd_id; | |
2378 | } | |
499afd5b | 2379 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 2380 | |
1ddbe94e | 2381 | /* |
e2839308 | 2382 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
2383 | * which case it now accurately reflects the new maximum. |
2384 | * Be careful not to overwrite the maximum value in that | |
2385 | * case. | |
1ddbe94e | 2386 | */ |
e2839308 AE |
2387 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
2388 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
2389 | } |
2390 | ||
e28fff26 AE |
2391 | /* |
2392 | * Skips over white space at *buf, and updates *buf to point to the | |
2393 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
2394 | * the token (string of non-white space characters) found. Note |
2395 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
2396 | */ |
2397 | static inline size_t next_token(const char **buf) | |
2398 | { | |
2399 | /* | |
2400 | * These are the characters that produce nonzero for | |
2401 | * isspace() in the "C" and "POSIX" locales. | |
2402 | */ | |
2403 | const char *spaces = " \f\n\r\t\v"; | |
2404 | ||
2405 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
2406 | ||
2407 | return strcspn(*buf, spaces); /* Return token length */ | |
2408 | } | |
2409 | ||
2410 | /* | |
2411 | * Finds the next token in *buf, and if the provided token buffer is | |
2412 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
2413 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
2414 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
2415 | * |
2416 | * Returns the length of the token found (not including the '\0'). | |
2417 | * Return value will be 0 if no token is found, and it will be >= | |
2418 | * token_size if the token would not fit. | |
2419 | * | |
593a9e7b | 2420 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
2421 | * found token. Note that this occurs even if the token buffer is |
2422 | * too small to hold it. | |
2423 | */ | |
2424 | static inline size_t copy_token(const char **buf, | |
2425 | char *token, | |
2426 | size_t token_size) | |
2427 | { | |
2428 | size_t len; | |
2429 | ||
2430 | len = next_token(buf); | |
2431 | if (len < token_size) { | |
2432 | memcpy(token, *buf, len); | |
2433 | *(token + len) = '\0'; | |
2434 | } | |
2435 | *buf += len; | |
2436 | ||
2437 | return len; | |
2438 | } | |
2439 | ||
ea3352f4 AE |
2440 | /* |
2441 | * Finds the next token in *buf, dynamically allocates a buffer big | |
2442 | * enough to hold a copy of it, and copies the token into the new | |
2443 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
2444 | * that a duplicate buffer is created even for a zero-length token. | |
2445 | * | |
2446 | * Returns a pointer to the newly-allocated duplicate, or a null | |
2447 | * pointer if memory for the duplicate was not available. If | |
2448 | * the lenp argument is a non-null pointer, the length of the token | |
2449 | * (not including the '\0') is returned in *lenp. | |
2450 | * | |
2451 | * If successful, the *buf pointer will be updated to point beyond | |
2452 | * the end of the found token. | |
2453 | * | |
2454 | * Note: uses GFP_KERNEL for allocation. | |
2455 | */ | |
2456 | static inline char *dup_token(const char **buf, size_t *lenp) | |
2457 | { | |
2458 | char *dup; | |
2459 | size_t len; | |
2460 | ||
2461 | len = next_token(buf); | |
2462 | dup = kmalloc(len + 1, GFP_KERNEL); | |
2463 | if (!dup) | |
2464 | return NULL; | |
2465 | ||
2466 | memcpy(dup, *buf, len); | |
2467 | *(dup + len) = '\0'; | |
2468 | *buf += len; | |
2469 | ||
2470 | if (lenp) | |
2471 | *lenp = len; | |
2472 | ||
2473 | return dup; | |
2474 | } | |
2475 | ||
a725f65e | 2476 | /* |
3feeb894 AE |
2477 | * This fills in the pool_name, image_name, image_name_len, rbd_dev, |
2478 | * rbd_md_name, and name fields of the given rbd_dev, based on the | |
2479 | * list of monitor addresses and other options provided via | |
2480 | * /sys/bus/rbd/add. Returns a pointer to a dynamically-allocated | |
2481 | * copy of the snapshot name to map if successful, or a | |
2482 | * pointer-coded error otherwise. | |
d22f76e7 AE |
2483 | * |
2484 | * Note: rbd_dev is assumed to have been initially zero-filled. | |
a725f65e | 2485 | */ |
3feeb894 AE |
2486 | static char *rbd_add_parse_args(struct rbd_device *rbd_dev, |
2487 | const char *buf, | |
2488 | const char **mon_addrs, | |
2489 | size_t *mon_addrs_size, | |
2490 | char *options, | |
2491 | size_t options_size) | |
e28fff26 | 2492 | { |
d22f76e7 | 2493 | size_t len; |
3feeb894 AE |
2494 | char *err_ptr = ERR_PTR(-EINVAL); |
2495 | char *snap_name; | |
e28fff26 AE |
2496 | |
2497 | /* The first four tokens are required */ | |
2498 | ||
7ef3214a AE |
2499 | len = next_token(&buf); |
2500 | if (!len) | |
3feeb894 | 2501 | return err_ptr; |
5214ecc4 | 2502 | *mon_addrs_size = len + 1; |
7ef3214a AE |
2503 | *mon_addrs = buf; |
2504 | ||
2505 | buf += len; | |
a725f65e | 2506 | |
e28fff26 AE |
2507 | len = copy_token(&buf, options, options_size); |
2508 | if (!len || len >= options_size) | |
3feeb894 | 2509 | return err_ptr; |
e28fff26 | 2510 | |
3feeb894 | 2511 | err_ptr = ERR_PTR(-ENOMEM); |
d22f76e7 AE |
2512 | rbd_dev->pool_name = dup_token(&buf, NULL); |
2513 | if (!rbd_dev->pool_name) | |
d22f76e7 | 2514 | goto out_err; |
e28fff26 | 2515 | |
0bed54dc AE |
2516 | rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len); |
2517 | if (!rbd_dev->image_name) | |
bf3e5ae1 | 2518 | goto out_err; |
a725f65e | 2519 | |
cb8627c7 AE |
2520 | /* Create the name of the header object */ |
2521 | ||
0bed54dc | 2522 | rbd_dev->header_name = kmalloc(rbd_dev->image_name_len |
bf3e5ae1 AE |
2523 | + sizeof (RBD_SUFFIX), |
2524 | GFP_KERNEL); | |
0bed54dc | 2525 | if (!rbd_dev->header_name) |
cb8627c7 | 2526 | goto out_err; |
0bed54dc | 2527 | sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX); |
a725f65e | 2528 | |
3feeb894 AE |
2529 | /* Snapshot name is optional */ |
2530 | len = next_token(&buf); | |
820a5f3e | 2531 | if (!len) { |
3feeb894 AE |
2532 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
2533 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
849b4260 | 2534 | } |
3feeb894 AE |
2535 | snap_name = kmalloc(len + 1, GFP_KERNEL); |
2536 | if (!snap_name) | |
2537 | goto out_err; | |
2538 | memcpy(snap_name, buf, len); | |
2539 | *(snap_name + len) = '\0'; | |
e28fff26 | 2540 | |
3feeb894 AE |
2541 | dout(" SNAP_NAME is <%s>, len is %zd\n", snap_name, len); |
2542 | ||
2543 | return snap_name; | |
d22f76e7 AE |
2544 | |
2545 | out_err: | |
0bed54dc | 2546 | kfree(rbd_dev->header_name); |
d78fd7ae | 2547 | rbd_dev->header_name = NULL; |
0bed54dc | 2548 | kfree(rbd_dev->image_name); |
d78fd7ae AE |
2549 | rbd_dev->image_name = NULL; |
2550 | rbd_dev->image_name_len = 0; | |
d22f76e7 AE |
2551 | kfree(rbd_dev->pool_name); |
2552 | rbd_dev->pool_name = NULL; | |
2553 | ||
3feeb894 | 2554 | return err_ptr; |
a725f65e AE |
2555 | } |
2556 | ||
59c2be1e YS |
2557 | static ssize_t rbd_add(struct bus_type *bus, |
2558 | const char *buf, | |
2559 | size_t count) | |
602adf40 | 2560 | { |
cb8627c7 AE |
2561 | char *options; |
2562 | struct rbd_device *rbd_dev = NULL; | |
7ef3214a AE |
2563 | const char *mon_addrs = NULL; |
2564 | size_t mon_addrs_size = 0; | |
27cc2594 AE |
2565 | struct ceph_osd_client *osdc; |
2566 | int rc = -ENOMEM; | |
3feeb894 | 2567 | char *snap_name; |
602adf40 YS |
2568 | |
2569 | if (!try_module_get(THIS_MODULE)) | |
2570 | return -ENODEV; | |
2571 | ||
60571c7d | 2572 | options = kmalloc(count, GFP_KERNEL); |
602adf40 | 2573 | if (!options) |
27cc2594 | 2574 | goto err_nomem; |
cb8627c7 AE |
2575 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
2576 | if (!rbd_dev) | |
2577 | goto err_nomem; | |
602adf40 YS |
2578 | |
2579 | /* static rbd_device initialization */ | |
2580 | spin_lock_init(&rbd_dev->lock); | |
2581 | INIT_LIST_HEAD(&rbd_dev->node); | |
dfc5606d | 2582 | INIT_LIST_HEAD(&rbd_dev->snaps); |
c666601a | 2583 | init_rwsem(&rbd_dev->header_rwsem); |
602adf40 | 2584 | |
d184f6bf | 2585 | /* generate unique id: find highest unique id, add one */ |
e2839308 | 2586 | rbd_dev_id_get(rbd_dev); |
602adf40 | 2587 | |
a725f65e | 2588 | /* Fill in the device name, now that we have its id. */ |
81a89793 AE |
2589 | BUILD_BUG_ON(DEV_NAME_LEN |
2590 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
de71a297 | 2591 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); |
a725f65e | 2592 | |
602adf40 | 2593 | /* parse add command */ |
3feeb894 AE |
2594 | snap_name = rbd_add_parse_args(rbd_dev, buf, |
2595 | &mon_addrs, &mon_addrs_size, options, count); | |
2596 | if (IS_ERR(snap_name)) { | |
2597 | rc = PTR_ERR(snap_name); | |
f0f8cef5 | 2598 | goto err_put_id; |
3feeb894 | 2599 | } |
e124a82f | 2600 | |
f8c38929 AE |
2601 | rc = rbd_get_client(rbd_dev, mon_addrs, mon_addrs_size - 1, options); |
2602 | if (rc < 0) | |
f0f8cef5 | 2603 | goto err_put_id; |
602adf40 | 2604 | |
602adf40 | 2605 | /* pick the pool */ |
1dbb4399 | 2606 | osdc = &rbd_dev->rbd_client->client->osdc; |
602adf40 YS |
2607 | rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name); |
2608 | if (rc < 0) | |
2609 | goto err_out_client; | |
9bb2f334 | 2610 | rbd_dev->pool_id = rc; |
602adf40 YS |
2611 | |
2612 | /* register our block device */ | |
27cc2594 AE |
2613 | rc = register_blkdev(0, rbd_dev->name); |
2614 | if (rc < 0) | |
602adf40 | 2615 | goto err_out_client; |
27cc2594 | 2616 | rbd_dev->major = rc; |
602adf40 | 2617 | |
dfc5606d YS |
2618 | rc = rbd_bus_add_dev(rbd_dev); |
2619 | if (rc) | |
766fc439 YS |
2620 | goto err_out_blkdev; |
2621 | ||
32eec68d AE |
2622 | /* |
2623 | * At this point cleanup in the event of an error is the job | |
2624 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
2625 | * | |
2626 | * Set up and announce blkdev mapping. | |
2627 | */ | |
602adf40 YS |
2628 | rc = rbd_init_disk(rbd_dev); |
2629 | if (rc) | |
766fc439 | 2630 | goto err_out_bus; |
602adf40 | 2631 | |
59c2be1e YS |
2632 | rc = rbd_init_watch_dev(rbd_dev); |
2633 | if (rc) | |
2634 | goto err_out_bus; | |
2635 | ||
602adf40 YS |
2636 | return count; |
2637 | ||
766fc439 | 2638 | err_out_bus: |
766fc439 YS |
2639 | /* this will also clean up rest of rbd_dev stuff */ |
2640 | ||
2641 | rbd_bus_del_dev(rbd_dev); | |
2642 | kfree(options); | |
766fc439 YS |
2643 | return rc; |
2644 | ||
602adf40 YS |
2645 | err_out_blkdev: |
2646 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
2647 | err_out_client: | |
2648 | rbd_put_client(rbd_dev); | |
f0f8cef5 | 2649 | err_put_id: |
cb8627c7 | 2650 | if (rbd_dev->pool_name) { |
f84344f3 | 2651 | kfree(rbd_dev->mapping.snap_name); |
0bed54dc AE |
2652 | kfree(rbd_dev->header_name); |
2653 | kfree(rbd_dev->image_name); | |
cb8627c7 AE |
2654 | kfree(rbd_dev->pool_name); |
2655 | } | |
e2839308 | 2656 | rbd_dev_id_put(rbd_dev); |
27cc2594 | 2657 | err_nomem: |
27cc2594 | 2658 | kfree(rbd_dev); |
cb8627c7 | 2659 | kfree(options); |
27cc2594 | 2660 | |
602adf40 YS |
2661 | dout("Error adding device %s\n", buf); |
2662 | module_put(THIS_MODULE); | |
27cc2594 AE |
2663 | |
2664 | return (ssize_t) rc; | |
602adf40 YS |
2665 | } |
2666 | ||
de71a297 | 2667 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
2668 | { |
2669 | struct list_head *tmp; | |
2670 | struct rbd_device *rbd_dev; | |
2671 | ||
e124a82f | 2672 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
2673 | list_for_each(tmp, &rbd_dev_list) { |
2674 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 2675 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 2676 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 2677 | return rbd_dev; |
e124a82f | 2678 | } |
602adf40 | 2679 | } |
e124a82f | 2680 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
2681 | return NULL; |
2682 | } | |
2683 | ||
dfc5606d | 2684 | static void rbd_dev_release(struct device *dev) |
602adf40 | 2685 | { |
593a9e7b | 2686 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2687 | |
1dbb4399 AE |
2688 | if (rbd_dev->watch_request) { |
2689 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
2690 | ||
2691 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 2692 | rbd_dev->watch_request); |
1dbb4399 | 2693 | } |
59c2be1e | 2694 | if (rbd_dev->watch_event) |
070c633f | 2695 | rbd_req_sync_unwatch(rbd_dev); |
59c2be1e | 2696 | |
602adf40 YS |
2697 | rbd_put_client(rbd_dev); |
2698 | ||
2699 | /* clean up and free blkdev */ | |
2700 | rbd_free_disk(rbd_dev); | |
2701 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d AE |
2702 | |
2703 | /* done with the id, and with the rbd_dev */ | |
f84344f3 | 2704 | kfree(rbd_dev->mapping.snap_name); |
0bed54dc | 2705 | kfree(rbd_dev->header_name); |
d22f76e7 | 2706 | kfree(rbd_dev->pool_name); |
0bed54dc | 2707 | kfree(rbd_dev->image_name); |
e2839308 | 2708 | rbd_dev_id_put(rbd_dev); |
602adf40 YS |
2709 | kfree(rbd_dev); |
2710 | ||
2711 | /* release module ref */ | |
2712 | module_put(THIS_MODULE); | |
602adf40 YS |
2713 | } |
2714 | ||
dfc5606d YS |
2715 | static ssize_t rbd_remove(struct bus_type *bus, |
2716 | const char *buf, | |
2717 | size_t count) | |
602adf40 YS |
2718 | { |
2719 | struct rbd_device *rbd_dev = NULL; | |
2720 | int target_id, rc; | |
2721 | unsigned long ul; | |
2722 | int ret = count; | |
2723 | ||
2724 | rc = strict_strtoul(buf, 10, &ul); | |
2725 | if (rc) | |
2726 | return rc; | |
2727 | ||
2728 | /* convert to int; abort if we lost anything in the conversion */ | |
2729 | target_id = (int) ul; | |
2730 | if (target_id != ul) | |
2731 | return -EINVAL; | |
2732 | ||
2733 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2734 | ||
2735 | rbd_dev = __rbd_get_dev(target_id); | |
2736 | if (!rbd_dev) { | |
2737 | ret = -ENOENT; | |
2738 | goto done; | |
2739 | } | |
2740 | ||
dfc5606d YS |
2741 | __rbd_remove_all_snaps(rbd_dev); |
2742 | rbd_bus_del_dev(rbd_dev); | |
602adf40 YS |
2743 | |
2744 | done: | |
2745 | mutex_unlock(&ctl_mutex); | |
aafb230e | 2746 | |
602adf40 YS |
2747 | return ret; |
2748 | } | |
2749 | ||
dfc5606d YS |
2750 | static ssize_t rbd_snap_add(struct device *dev, |
2751 | struct device_attribute *attr, | |
2752 | const char *buf, | |
2753 | size_t count) | |
602adf40 | 2754 | { |
593a9e7b | 2755 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
2756 | int ret; |
2757 | char *name = kmalloc(count + 1, GFP_KERNEL); | |
602adf40 YS |
2758 | if (!name) |
2759 | return -ENOMEM; | |
2760 | ||
dfc5606d | 2761 | snprintf(name, count, "%s", buf); |
602adf40 YS |
2762 | |
2763 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2764 | ||
602adf40 YS |
2765 | ret = rbd_header_add_snap(rbd_dev, |
2766 | name, GFP_KERNEL); | |
2767 | if (ret < 0) | |
59c2be1e | 2768 | goto err_unlock; |
602adf40 | 2769 | |
b813623a | 2770 | ret = __rbd_refresh_header(rbd_dev, NULL); |
602adf40 | 2771 | if (ret < 0) |
59c2be1e YS |
2772 | goto err_unlock; |
2773 | ||
2774 | /* shouldn't hold ctl_mutex when notifying.. notify might | |
2775 | trigger a watch callback that would need to get that mutex */ | |
2776 | mutex_unlock(&ctl_mutex); | |
2777 | ||
2778 | /* make a best effort, don't error if failed */ | |
4cb16250 | 2779 | rbd_req_sync_notify(rbd_dev); |
602adf40 YS |
2780 | |
2781 | ret = count; | |
59c2be1e YS |
2782 | kfree(name); |
2783 | return ret; | |
2784 | ||
2785 | err_unlock: | |
602adf40 | 2786 | mutex_unlock(&ctl_mutex); |
602adf40 YS |
2787 | kfree(name); |
2788 | return ret; | |
2789 | } | |
2790 | ||
602adf40 YS |
2791 | /* |
2792 | * create control files in sysfs | |
dfc5606d | 2793 | * /sys/bus/rbd/... |
602adf40 YS |
2794 | */ |
2795 | static int rbd_sysfs_init(void) | |
2796 | { | |
dfc5606d | 2797 | int ret; |
602adf40 | 2798 | |
fed4c143 | 2799 | ret = device_register(&rbd_root_dev); |
21079786 | 2800 | if (ret < 0) |
dfc5606d | 2801 | return ret; |
602adf40 | 2802 | |
fed4c143 AE |
2803 | ret = bus_register(&rbd_bus_type); |
2804 | if (ret < 0) | |
2805 | device_unregister(&rbd_root_dev); | |
602adf40 | 2806 | |
602adf40 YS |
2807 | return ret; |
2808 | } | |
2809 | ||
2810 | static void rbd_sysfs_cleanup(void) | |
2811 | { | |
dfc5606d | 2812 | bus_unregister(&rbd_bus_type); |
fed4c143 | 2813 | device_unregister(&rbd_root_dev); |
602adf40 YS |
2814 | } |
2815 | ||
2816 | int __init rbd_init(void) | |
2817 | { | |
2818 | int rc; | |
2819 | ||
2820 | rc = rbd_sysfs_init(); | |
2821 | if (rc) | |
2822 | return rc; | |
f0f8cef5 | 2823 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
2824 | return 0; |
2825 | } | |
2826 | ||
2827 | void __exit rbd_exit(void) | |
2828 | { | |
2829 | rbd_sysfs_cleanup(); | |
2830 | } | |
2831 | ||
2832 | module_init(rbd_init); | |
2833 | module_exit(rbd_exit); | |
2834 | ||
2835 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
2836 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
2837 | MODULE_DESCRIPTION("rados block device"); | |
2838 | ||
2839 | /* following authorship retained from original osdblk.c */ | |
2840 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
2841 | ||
2842 | MODULE_LICENSE("GPL"); |