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 | ||
593a9e7b AE |
44 | /* |
45 | * The basic unit of block I/O is a sector. It is interpreted in a | |
46 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
47 | * universally 512 bytes. These symbols are just slightly more | |
48 | * meaningful than the bare numbers they represent. | |
49 | */ | |
50 | #define SECTOR_SHIFT 9 | |
51 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
52 | ||
f0f8cef5 AE |
53 | #define RBD_DRV_NAME "rbd" |
54 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
55 | |
56 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
57 | ||
602adf40 YS |
58 | #define RBD_MAX_SNAP_NAME_LEN 32 |
59 | #define RBD_MAX_OPT_LEN 1024 | |
60 | ||
61 | #define RBD_SNAP_HEAD_NAME "-" | |
62 | ||
81a89793 AE |
63 | /* |
64 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
65 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
66 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
67 | * enough to hold all possible device names. | |
68 | */ | |
602adf40 | 69 | #define DEV_NAME_LEN 32 |
81a89793 | 70 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 | 71 | |
59c2be1e YS |
72 | #define RBD_NOTIFY_TIMEOUT_DEFAULT 10 |
73 | ||
602adf40 YS |
74 | /* |
75 | * block device image metadata (in-memory version) | |
76 | */ | |
77 | struct rbd_image_header { | |
78 | u64 image_size; | |
849b4260 | 79 | char *object_prefix; |
602adf40 YS |
80 | __u8 obj_order; |
81 | __u8 crypt_type; | |
82 | __u8 comp_type; | |
602adf40 YS |
83 | struct ceph_snap_context *snapc; |
84 | size_t snap_names_len; | |
602adf40 YS |
85 | u32 total_snaps; |
86 | ||
87 | char *snap_names; | |
88 | u64 *snap_sizes; | |
59c2be1e YS |
89 | |
90 | u64 obj_version; | |
91 | }; | |
92 | ||
93 | struct rbd_options { | |
94 | int notify_timeout; | |
602adf40 YS |
95 | }; |
96 | ||
97 | /* | |
f0f8cef5 | 98 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
99 | */ |
100 | struct rbd_client { | |
101 | struct ceph_client *client; | |
59c2be1e | 102 | struct rbd_options *rbd_opts; |
602adf40 YS |
103 | struct kref kref; |
104 | struct list_head node; | |
105 | }; | |
106 | ||
107 | /* | |
f0f8cef5 | 108 | * a request completion status |
602adf40 | 109 | */ |
1fec7093 YS |
110 | struct rbd_req_status { |
111 | int done; | |
112 | int rc; | |
113 | u64 bytes; | |
114 | }; | |
115 | ||
116 | /* | |
117 | * a collection of requests | |
118 | */ | |
119 | struct rbd_req_coll { | |
120 | int total; | |
121 | int num_done; | |
122 | struct kref kref; | |
123 | struct rbd_req_status status[0]; | |
602adf40 YS |
124 | }; |
125 | ||
f0f8cef5 AE |
126 | /* |
127 | * a single io request | |
128 | */ | |
129 | struct rbd_request { | |
130 | struct request *rq; /* blk layer request */ | |
131 | struct bio *bio; /* cloned bio */ | |
132 | struct page **pages; /* list of used pages */ | |
133 | u64 len; | |
134 | int coll_index; | |
135 | struct rbd_req_coll *coll; | |
136 | }; | |
137 | ||
dfc5606d YS |
138 | struct rbd_snap { |
139 | struct device dev; | |
140 | const char *name; | |
3591538f | 141 | u64 size; |
dfc5606d YS |
142 | struct list_head node; |
143 | u64 id; | |
144 | }; | |
145 | ||
602adf40 YS |
146 | /* |
147 | * a single device | |
148 | */ | |
149 | struct rbd_device { | |
150 | int id; /* blkdev unique id */ | |
151 | ||
152 | int major; /* blkdev assigned major */ | |
153 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
154 | struct request_queue *q; | |
155 | ||
602adf40 YS |
156 | struct rbd_client *rbd_client; |
157 | ||
158 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
159 | ||
160 | spinlock_t lock; /* queue lock */ | |
161 | ||
162 | struct rbd_image_header header; | |
0bed54dc AE |
163 | char *image_name; |
164 | size_t image_name_len; | |
165 | char *header_name; | |
d22f76e7 | 166 | char *pool_name; |
9bb2f334 | 167 | int pool_id; |
602adf40 | 168 | |
59c2be1e YS |
169 | struct ceph_osd_event *watch_event; |
170 | struct ceph_osd_request *watch_request; | |
171 | ||
c666601a JD |
172 | /* protects updating the header */ |
173 | struct rw_semaphore header_rwsem; | |
e88a36ec | 174 | /* name of the snapshot this device reads from */ |
820a5f3e | 175 | char *snap_name; |
e88a36ec | 176 | /* id of the snapshot this device reads from */ |
77dfe99f | 177 | u64 snap_id; /* current snapshot id */ |
e88a36ec JD |
178 | /* whether the snap_id this device reads from still exists */ |
179 | bool snap_exists; | |
180 | int read_only; | |
602adf40 YS |
181 | |
182 | struct list_head node; | |
dfc5606d YS |
183 | |
184 | /* list of snapshots */ | |
185 | struct list_head snaps; | |
186 | ||
187 | /* sysfs related */ | |
188 | struct device dev; | |
189 | }; | |
190 | ||
602adf40 | 191 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 192 | |
602adf40 | 193 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
194 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
195 | ||
432b8587 AE |
196 | static LIST_HEAD(rbd_client_list); /* clients */ |
197 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 198 | |
dfc5606d YS |
199 | static int __rbd_init_snaps_header(struct rbd_device *rbd_dev); |
200 | static void rbd_dev_release(struct device *dev); | |
dfc5606d YS |
201 | static ssize_t rbd_snap_add(struct device *dev, |
202 | struct device_attribute *attr, | |
203 | const char *buf, | |
204 | size_t count); | |
205 | static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev, | |
69932487 | 206 | struct rbd_snap *snap); |
dfc5606d | 207 | |
f0f8cef5 AE |
208 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
209 | size_t count); | |
210 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
211 | size_t count); | |
212 | ||
213 | static struct bus_attribute rbd_bus_attrs[] = { | |
214 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
215 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
216 | __ATTR_NULL | |
217 | }; | |
218 | ||
219 | static struct bus_type rbd_bus_type = { | |
220 | .name = "rbd", | |
221 | .bus_attrs = rbd_bus_attrs, | |
222 | }; | |
223 | ||
224 | static void rbd_root_dev_release(struct device *dev) | |
225 | { | |
226 | } | |
227 | ||
228 | static struct device rbd_root_dev = { | |
229 | .init_name = "rbd", | |
230 | .release = rbd_root_dev_release, | |
231 | }; | |
232 | ||
dfc5606d | 233 | |
dfc5606d YS |
234 | static struct device *rbd_get_dev(struct rbd_device *rbd_dev) |
235 | { | |
236 | return get_device(&rbd_dev->dev); | |
237 | } | |
238 | ||
239 | static void rbd_put_dev(struct rbd_device *rbd_dev) | |
240 | { | |
241 | put_device(&rbd_dev->dev); | |
242 | } | |
602adf40 | 243 | |
263c6ca0 | 244 | static int __rbd_refresh_header(struct rbd_device *rbd_dev); |
59c2be1e | 245 | |
602adf40 YS |
246 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
247 | { | |
f0f8cef5 | 248 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
602adf40 | 249 | |
dfc5606d YS |
250 | rbd_get_dev(rbd_dev); |
251 | ||
602adf40 YS |
252 | set_device_ro(bdev, rbd_dev->read_only); |
253 | ||
254 | if ((mode & FMODE_WRITE) && rbd_dev->read_only) | |
255 | return -EROFS; | |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
dfc5606d YS |
260 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
261 | { | |
262 | struct rbd_device *rbd_dev = disk->private_data; | |
263 | ||
264 | rbd_put_dev(rbd_dev); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
602adf40 YS |
269 | static const struct block_device_operations rbd_bd_ops = { |
270 | .owner = THIS_MODULE, | |
271 | .open = rbd_open, | |
dfc5606d | 272 | .release = rbd_release, |
602adf40 YS |
273 | }; |
274 | ||
275 | /* | |
276 | * Initialize an rbd client instance. | |
43ae4701 | 277 | * We own *ceph_opts. |
602adf40 | 278 | */ |
43ae4701 | 279 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts, |
59c2be1e | 280 | struct rbd_options *rbd_opts) |
602adf40 YS |
281 | { |
282 | struct rbd_client *rbdc; | |
283 | int ret = -ENOMEM; | |
284 | ||
285 | dout("rbd_client_create\n"); | |
286 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); | |
287 | if (!rbdc) | |
288 | goto out_opt; | |
289 | ||
290 | kref_init(&rbdc->kref); | |
291 | INIT_LIST_HEAD(&rbdc->node); | |
292 | ||
bc534d86 AE |
293 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
294 | ||
43ae4701 | 295 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 296 | if (IS_ERR(rbdc->client)) |
bc534d86 | 297 | goto out_mutex; |
43ae4701 | 298 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
299 | |
300 | ret = ceph_open_session(rbdc->client); | |
301 | if (ret < 0) | |
302 | goto out_err; | |
303 | ||
59c2be1e YS |
304 | rbdc->rbd_opts = rbd_opts; |
305 | ||
432b8587 | 306 | spin_lock(&rbd_client_list_lock); |
602adf40 | 307 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 308 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 309 | |
bc534d86 AE |
310 | mutex_unlock(&ctl_mutex); |
311 | ||
602adf40 YS |
312 | dout("rbd_client_create created %p\n", rbdc); |
313 | return rbdc; | |
314 | ||
315 | out_err: | |
316 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
317 | out_mutex: |
318 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
319 | kfree(rbdc); |
320 | out_opt: | |
43ae4701 AE |
321 | if (ceph_opts) |
322 | ceph_destroy_options(ceph_opts); | |
28f259b7 | 323 | return ERR_PTR(ret); |
602adf40 YS |
324 | } |
325 | ||
326 | /* | |
327 | * Find a ceph client with specific addr and configuration. | |
328 | */ | |
43ae4701 | 329 | static struct rbd_client *__rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
330 | { |
331 | struct rbd_client *client_node; | |
332 | ||
43ae4701 | 333 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
334 | return NULL; |
335 | ||
336 | list_for_each_entry(client_node, &rbd_client_list, node) | |
43ae4701 | 337 | if (!ceph_compare_options(ceph_opts, client_node->client)) |
602adf40 YS |
338 | return client_node; |
339 | return NULL; | |
340 | } | |
341 | ||
59c2be1e YS |
342 | /* |
343 | * mount options | |
344 | */ | |
345 | enum { | |
346 | Opt_notify_timeout, | |
347 | Opt_last_int, | |
348 | /* int args above */ | |
349 | Opt_last_string, | |
350 | /* string args above */ | |
351 | }; | |
352 | ||
43ae4701 | 353 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
354 | {Opt_notify_timeout, "notify_timeout=%d"}, |
355 | /* int args above */ | |
356 | /* string args above */ | |
357 | {-1, NULL} | |
358 | }; | |
359 | ||
360 | static int parse_rbd_opts_token(char *c, void *private) | |
361 | { | |
43ae4701 | 362 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
363 | substring_t argstr[MAX_OPT_ARGS]; |
364 | int token, intval, ret; | |
365 | ||
43ae4701 | 366 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
367 | if (token < 0) |
368 | return -EINVAL; | |
369 | ||
370 | if (token < Opt_last_int) { | |
371 | ret = match_int(&argstr[0], &intval); | |
372 | if (ret < 0) { | |
373 | pr_err("bad mount option arg (not int) " | |
374 | "at '%s'\n", c); | |
375 | return ret; | |
376 | } | |
377 | dout("got int token %d val %d\n", token, intval); | |
378 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
379 | dout("got string token %d val %s\n", token, | |
380 | argstr[0].from); | |
381 | } else { | |
382 | dout("got token %d\n", token); | |
383 | } | |
384 | ||
385 | switch (token) { | |
386 | case Opt_notify_timeout: | |
43ae4701 | 387 | rbd_opts->notify_timeout = intval; |
59c2be1e YS |
388 | break; |
389 | default: | |
390 | BUG_ON(token); | |
391 | } | |
392 | return 0; | |
393 | } | |
394 | ||
602adf40 YS |
395 | /* |
396 | * Get a ceph client with specific addr and configuration, if one does | |
397 | * not exist create it. | |
398 | */ | |
5214ecc4 AE |
399 | static struct rbd_client *rbd_get_client(const char *mon_addr, |
400 | size_t mon_addr_len, | |
401 | char *options) | |
602adf40 YS |
402 | { |
403 | struct rbd_client *rbdc; | |
43ae4701 | 404 | struct ceph_options *ceph_opts; |
59c2be1e YS |
405 | struct rbd_options *rbd_opts; |
406 | ||
407 | rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL); | |
408 | if (!rbd_opts) | |
d720bcb0 | 409 | return ERR_PTR(-ENOMEM); |
59c2be1e YS |
410 | |
411 | rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT; | |
602adf40 | 412 | |
43ae4701 AE |
413 | ceph_opts = ceph_parse_options(options, mon_addr, |
414 | mon_addr + mon_addr_len, | |
415 | parse_rbd_opts_token, rbd_opts); | |
416 | if (IS_ERR(ceph_opts)) { | |
d720bcb0 | 417 | kfree(rbd_opts); |
43ae4701 | 418 | return ERR_CAST(ceph_opts); |
ee57741c | 419 | } |
602adf40 | 420 | |
432b8587 | 421 | spin_lock(&rbd_client_list_lock); |
43ae4701 | 422 | rbdc = __rbd_client_find(ceph_opts); |
602adf40 | 423 | if (rbdc) { |
602adf40 YS |
424 | /* using an existing client */ |
425 | kref_get(&rbdc->kref); | |
432b8587 | 426 | spin_unlock(&rbd_client_list_lock); |
e6994d3d | 427 | |
43ae4701 | 428 | ceph_destroy_options(ceph_opts); |
e6994d3d AE |
429 | kfree(rbd_opts); |
430 | ||
d720bcb0 | 431 | return rbdc; |
602adf40 | 432 | } |
432b8587 | 433 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 434 | |
43ae4701 | 435 | rbdc = rbd_client_create(ceph_opts, rbd_opts); |
d97081b0 | 436 | |
d720bcb0 AE |
437 | if (IS_ERR(rbdc)) |
438 | kfree(rbd_opts); | |
602adf40 | 439 | |
d720bcb0 | 440 | return rbdc; |
602adf40 YS |
441 | } |
442 | ||
443 | /* | |
444 | * Destroy ceph client | |
d23a4b3f | 445 | * |
432b8587 | 446 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
447 | */ |
448 | static void rbd_client_release(struct kref *kref) | |
449 | { | |
450 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
451 | ||
452 | dout("rbd_release_client %p\n", rbdc); | |
cd9d9f5d | 453 | spin_lock(&rbd_client_list_lock); |
602adf40 | 454 | list_del(&rbdc->node); |
cd9d9f5d | 455 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
456 | |
457 | ceph_destroy_client(rbdc->client); | |
59c2be1e | 458 | kfree(rbdc->rbd_opts); |
602adf40 YS |
459 | kfree(rbdc); |
460 | } | |
461 | ||
462 | /* | |
463 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
464 | * it. | |
465 | */ | |
466 | static void rbd_put_client(struct rbd_device *rbd_dev) | |
467 | { | |
468 | kref_put(&rbd_dev->rbd_client->kref, rbd_client_release); | |
469 | rbd_dev->rbd_client = NULL; | |
602adf40 YS |
470 | } |
471 | ||
1fec7093 YS |
472 | /* |
473 | * Destroy requests collection | |
474 | */ | |
475 | static void rbd_coll_release(struct kref *kref) | |
476 | { | |
477 | struct rbd_req_coll *coll = | |
478 | container_of(kref, struct rbd_req_coll, kref); | |
479 | ||
480 | dout("rbd_coll_release %p\n", coll); | |
481 | kfree(coll); | |
482 | } | |
602adf40 | 483 | |
8e94af8e AE |
484 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
485 | { | |
486 | return !memcmp(&ondisk->text, | |
487 | RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT)); | |
488 | } | |
489 | ||
602adf40 YS |
490 | /* |
491 | * Create a new header structure, translate header format from the on-disk | |
492 | * header. | |
493 | */ | |
494 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
495 | struct rbd_image_header_ondisk *ondisk, | |
50f7c4c9 | 496 | u32 allocated_snaps, |
602adf40 YS |
497 | gfp_t gfp_flags) |
498 | { | |
50f7c4c9 | 499 | u32 i, snap_count; |
602adf40 | 500 | |
8e94af8e | 501 | if (!rbd_dev_ondisk_valid(ondisk)) |
81e759fb | 502 | return -ENXIO; |
81e759fb | 503 | |
00f1f36f | 504 | snap_count = le32_to_cpu(ondisk->snap_count); |
50f7c4c9 XW |
505 | if (snap_count > (UINT_MAX - sizeof(struct ceph_snap_context)) |
506 | / sizeof (*ondisk)) | |
507 | return -EINVAL; | |
602adf40 | 508 | header->snapc = kmalloc(sizeof(struct ceph_snap_context) + |
f9f9a190 | 509 | snap_count * sizeof(u64), |
602adf40 YS |
510 | gfp_flags); |
511 | if (!header->snapc) | |
512 | return -ENOMEM; | |
00f1f36f | 513 | |
00f1f36f | 514 | header->snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
602adf40 YS |
515 | if (snap_count) { |
516 | header->snap_names = kmalloc(header->snap_names_len, | |
f8ad495a | 517 | gfp_flags); |
602adf40 YS |
518 | if (!header->snap_names) |
519 | goto err_snapc; | |
520 | header->snap_sizes = kmalloc(snap_count * sizeof(u64), | |
f8ad495a | 521 | gfp_flags); |
602adf40 YS |
522 | if (!header->snap_sizes) |
523 | goto err_names; | |
524 | } else { | |
525 | header->snap_names = NULL; | |
526 | header->snap_sizes = NULL; | |
527 | } | |
849b4260 AE |
528 | |
529 | header->object_prefix = kmalloc(sizeof (ondisk->block_name) + 1, | |
530 | gfp_flags); | |
531 | if (!header->object_prefix) | |
532 | goto err_sizes; | |
533 | ||
ca1e49a6 | 534 | memcpy(header->object_prefix, ondisk->block_name, |
602adf40 | 535 | sizeof(ondisk->block_name)); |
849b4260 | 536 | header->object_prefix[sizeof (ondisk->block_name)] = '\0'; |
602adf40 YS |
537 | |
538 | header->image_size = le64_to_cpu(ondisk->image_size); | |
539 | header->obj_order = ondisk->options.order; | |
540 | header->crypt_type = ondisk->options.crypt_type; | |
541 | header->comp_type = ondisk->options.comp_type; | |
542 | ||
543 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 544 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 YS |
545 | header->snapc->num_snaps = snap_count; |
546 | header->total_snaps = snap_count; | |
547 | ||
21079786 | 548 | if (snap_count && allocated_snaps == snap_count) { |
602adf40 YS |
549 | for (i = 0; i < snap_count; i++) { |
550 | header->snapc->snaps[i] = | |
551 | le64_to_cpu(ondisk->snaps[i].id); | |
552 | header->snap_sizes[i] = | |
553 | le64_to_cpu(ondisk->snaps[i].image_size); | |
554 | } | |
555 | ||
556 | /* copy snapshot names */ | |
557 | memcpy(header->snap_names, &ondisk->snaps[i], | |
558 | header->snap_names_len); | |
559 | } | |
560 | ||
561 | return 0; | |
562 | ||
849b4260 AE |
563 | err_sizes: |
564 | kfree(header->snap_sizes); | |
602adf40 YS |
565 | err_names: |
566 | kfree(header->snap_names); | |
567 | err_snapc: | |
568 | kfree(header->snapc); | |
00f1f36f | 569 | return -ENOMEM; |
602adf40 YS |
570 | } |
571 | ||
602adf40 YS |
572 | static int snap_by_name(struct rbd_image_header *header, const char *snap_name, |
573 | u64 *seq, u64 *size) | |
574 | { | |
575 | int i; | |
576 | char *p = header->snap_names; | |
577 | ||
00f1f36f AE |
578 | for (i = 0; i < header->total_snaps; i++) { |
579 | if (!strcmp(snap_name, p)) { | |
602adf40 | 580 | |
00f1f36f | 581 | /* Found it. Pass back its id and/or size */ |
602adf40 | 582 | |
00f1f36f AE |
583 | if (seq) |
584 | *seq = header->snapc->snaps[i]; | |
585 | if (size) | |
586 | *size = header->snap_sizes[i]; | |
587 | return i; | |
588 | } | |
589 | p += strlen(p) + 1; /* Skip ahead to the next name */ | |
590 | } | |
591 | return -ENOENT; | |
602adf40 YS |
592 | } |
593 | ||
0ce1a794 | 594 | static int rbd_header_set_snap(struct rbd_device *rbd_dev, u64 *size) |
602adf40 | 595 | { |
78dc447d | 596 | int ret; |
602adf40 | 597 | |
0ce1a794 | 598 | down_write(&rbd_dev->header_rwsem); |
602adf40 | 599 | |
0ce1a794 | 600 | if (!memcmp(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 601 | sizeof (RBD_SNAP_HEAD_NAME))) { |
0ce1a794 | 602 | rbd_dev->snap_id = CEPH_NOSNAP; |
e88a36ec | 603 | rbd_dev->snap_exists = false; |
0ce1a794 | 604 | rbd_dev->read_only = 0; |
602adf40 | 605 | if (size) |
78dc447d | 606 | *size = rbd_dev->header.image_size; |
602adf40 | 607 | } else { |
78dc447d AE |
608 | u64 snap_id = 0; |
609 | ||
610 | ret = snap_by_name(&rbd_dev->header, rbd_dev->snap_name, | |
611 | &snap_id, size); | |
602adf40 YS |
612 | if (ret < 0) |
613 | goto done; | |
78dc447d | 614 | rbd_dev->snap_id = snap_id; |
e88a36ec | 615 | rbd_dev->snap_exists = true; |
0ce1a794 | 616 | rbd_dev->read_only = 1; |
602adf40 YS |
617 | } |
618 | ||
619 | ret = 0; | |
620 | done: | |
0ce1a794 | 621 | up_write(&rbd_dev->header_rwsem); |
602adf40 YS |
622 | return ret; |
623 | } | |
624 | ||
625 | static void rbd_header_free(struct rbd_image_header *header) | |
626 | { | |
849b4260 | 627 | kfree(header->object_prefix); |
602adf40 | 628 | kfree(header->snap_sizes); |
849b4260 | 629 | kfree(header->snap_names); |
d1d25646 | 630 | ceph_put_snap_context(header->snapc); |
602adf40 YS |
631 | } |
632 | ||
633 | /* | |
634 | * get the actual striped segment name, offset and length | |
635 | */ | |
636 | static u64 rbd_get_segment(struct rbd_image_header *header, | |
ca1e49a6 | 637 | const char *object_prefix, |
602adf40 YS |
638 | u64 ofs, u64 len, |
639 | char *seg_name, u64 *segofs) | |
640 | { | |
641 | u64 seg = ofs >> header->obj_order; | |
642 | ||
643 | if (seg_name) | |
644 | snprintf(seg_name, RBD_MAX_SEG_NAME_LEN, | |
ca1e49a6 | 645 | "%s.%012llx", object_prefix, seg); |
602adf40 YS |
646 | |
647 | ofs = ofs & ((1 << header->obj_order) - 1); | |
648 | len = min_t(u64, len, (1 << header->obj_order) - ofs); | |
649 | ||
650 | if (segofs) | |
651 | *segofs = ofs; | |
652 | ||
653 | return len; | |
654 | } | |
655 | ||
1fec7093 YS |
656 | static int rbd_get_num_segments(struct rbd_image_header *header, |
657 | u64 ofs, u64 len) | |
658 | { | |
659 | u64 start_seg = ofs >> header->obj_order; | |
660 | u64 end_seg = (ofs + len - 1) >> header->obj_order; | |
661 | return end_seg - start_seg + 1; | |
662 | } | |
663 | ||
029bcbd8 JD |
664 | /* |
665 | * returns the size of an object in the image | |
666 | */ | |
667 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
668 | { | |
669 | return 1 << header->obj_order; | |
670 | } | |
671 | ||
602adf40 YS |
672 | /* |
673 | * bio helpers | |
674 | */ | |
675 | ||
676 | static void bio_chain_put(struct bio *chain) | |
677 | { | |
678 | struct bio *tmp; | |
679 | ||
680 | while (chain) { | |
681 | tmp = chain; | |
682 | chain = chain->bi_next; | |
683 | bio_put(tmp); | |
684 | } | |
685 | } | |
686 | ||
687 | /* | |
688 | * zeros a bio chain, starting at specific offset | |
689 | */ | |
690 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
691 | { | |
692 | struct bio_vec *bv; | |
693 | unsigned long flags; | |
694 | void *buf; | |
695 | int i; | |
696 | int pos = 0; | |
697 | ||
698 | while (chain) { | |
699 | bio_for_each_segment(bv, chain, i) { | |
700 | if (pos + bv->bv_len > start_ofs) { | |
701 | int remainder = max(start_ofs - pos, 0); | |
702 | buf = bvec_kmap_irq(bv, &flags); | |
703 | memset(buf + remainder, 0, | |
704 | bv->bv_len - remainder); | |
85b5aaa6 | 705 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
706 | } |
707 | pos += bv->bv_len; | |
708 | } | |
709 | ||
710 | chain = chain->bi_next; | |
711 | } | |
712 | } | |
713 | ||
714 | /* | |
715 | * bio_chain_clone - clone a chain of bios up to a certain length. | |
716 | * might return a bio_pair that will need to be released. | |
717 | */ | |
718 | static struct bio *bio_chain_clone(struct bio **old, struct bio **next, | |
719 | struct bio_pair **bp, | |
720 | int len, gfp_t gfpmask) | |
721 | { | |
722 | struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL; | |
723 | int total = 0; | |
724 | ||
725 | if (*bp) { | |
726 | bio_pair_release(*bp); | |
727 | *bp = NULL; | |
728 | } | |
729 | ||
730 | while (old_chain && (total < len)) { | |
731 | tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs); | |
732 | if (!tmp) | |
733 | goto err_out; | |
734 | ||
735 | if (total + old_chain->bi_size > len) { | |
736 | struct bio_pair *bp; | |
737 | ||
738 | /* | |
739 | * this split can only happen with a single paged bio, | |
740 | * split_bio will BUG_ON if this is not the case | |
741 | */ | |
742 | dout("bio_chain_clone split! total=%d remaining=%d" | |
bd919d45 AE |
743 | "bi_size=%u\n", |
744 | total, len - total, old_chain->bi_size); | |
602adf40 YS |
745 | |
746 | /* split the bio. We'll release it either in the next | |
747 | call, or it will have to be released outside */ | |
593a9e7b | 748 | bp = bio_split(old_chain, (len - total) / SECTOR_SIZE); |
602adf40 YS |
749 | if (!bp) |
750 | goto err_out; | |
751 | ||
752 | __bio_clone(tmp, &bp->bio1); | |
753 | ||
754 | *next = &bp->bio2; | |
755 | } else { | |
756 | __bio_clone(tmp, old_chain); | |
757 | *next = old_chain->bi_next; | |
758 | } | |
759 | ||
760 | tmp->bi_bdev = NULL; | |
761 | gfpmask &= ~__GFP_WAIT; | |
762 | tmp->bi_next = NULL; | |
763 | ||
764 | if (!new_chain) { | |
765 | new_chain = tail = tmp; | |
766 | } else { | |
767 | tail->bi_next = tmp; | |
768 | tail = tmp; | |
769 | } | |
770 | old_chain = old_chain->bi_next; | |
771 | ||
772 | total += tmp->bi_size; | |
773 | } | |
774 | ||
775 | BUG_ON(total < len); | |
776 | ||
777 | if (tail) | |
778 | tail->bi_next = NULL; | |
779 | ||
780 | *old = old_chain; | |
781 | ||
782 | return new_chain; | |
783 | ||
784 | err_out: | |
785 | dout("bio_chain_clone with err\n"); | |
786 | bio_chain_put(new_chain); | |
787 | return NULL; | |
788 | } | |
789 | ||
790 | /* | |
791 | * helpers for osd request op vectors. | |
792 | */ | |
793 | static int rbd_create_rw_ops(struct ceph_osd_req_op **ops, | |
794 | int num_ops, | |
795 | int opcode, | |
796 | u32 payload_len) | |
797 | { | |
798 | *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1), | |
799 | GFP_NOIO); | |
800 | if (!*ops) | |
801 | return -ENOMEM; | |
802 | (*ops)[0].op = opcode; | |
803 | /* | |
804 | * op extent offset and length will be set later on | |
805 | * in calc_raw_layout() | |
806 | */ | |
807 | (*ops)[0].payload_len = payload_len; | |
808 | return 0; | |
809 | } | |
810 | ||
811 | static void rbd_destroy_ops(struct ceph_osd_req_op *ops) | |
812 | { | |
813 | kfree(ops); | |
814 | } | |
815 | ||
1fec7093 YS |
816 | static void rbd_coll_end_req_index(struct request *rq, |
817 | struct rbd_req_coll *coll, | |
818 | int index, | |
819 | int ret, u64 len) | |
820 | { | |
821 | struct request_queue *q; | |
822 | int min, max, i; | |
823 | ||
bd919d45 AE |
824 | dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n", |
825 | coll, index, ret, (unsigned long long) len); | |
1fec7093 YS |
826 | |
827 | if (!rq) | |
828 | return; | |
829 | ||
830 | if (!coll) { | |
831 | blk_end_request(rq, ret, len); | |
832 | return; | |
833 | } | |
834 | ||
835 | q = rq->q; | |
836 | ||
837 | spin_lock_irq(q->queue_lock); | |
838 | coll->status[index].done = 1; | |
839 | coll->status[index].rc = ret; | |
840 | coll->status[index].bytes = len; | |
841 | max = min = coll->num_done; | |
842 | while (max < coll->total && coll->status[max].done) | |
843 | max++; | |
844 | ||
845 | for (i = min; i<max; i++) { | |
846 | __blk_end_request(rq, coll->status[i].rc, | |
847 | coll->status[i].bytes); | |
848 | coll->num_done++; | |
849 | kref_put(&coll->kref, rbd_coll_release); | |
850 | } | |
851 | spin_unlock_irq(q->queue_lock); | |
852 | } | |
853 | ||
854 | static void rbd_coll_end_req(struct rbd_request *req, | |
855 | int ret, u64 len) | |
856 | { | |
857 | rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len); | |
858 | } | |
859 | ||
602adf40 YS |
860 | /* |
861 | * Send ceph osd request | |
862 | */ | |
863 | static int rbd_do_request(struct request *rq, | |
0ce1a794 | 864 | struct rbd_device *rbd_dev, |
602adf40 YS |
865 | struct ceph_snap_context *snapc, |
866 | u64 snapid, | |
aded07ea | 867 | const char *object_name, u64 ofs, u64 len, |
602adf40 YS |
868 | struct bio *bio, |
869 | struct page **pages, | |
870 | int num_pages, | |
871 | int flags, | |
872 | struct ceph_osd_req_op *ops, | |
1fec7093 YS |
873 | struct rbd_req_coll *coll, |
874 | int coll_index, | |
602adf40 | 875 | void (*rbd_cb)(struct ceph_osd_request *req, |
59c2be1e YS |
876 | struct ceph_msg *msg), |
877 | struct ceph_osd_request **linger_req, | |
878 | u64 *ver) | |
602adf40 YS |
879 | { |
880 | struct ceph_osd_request *req; | |
881 | struct ceph_file_layout *layout; | |
882 | int ret; | |
883 | u64 bno; | |
884 | struct timespec mtime = CURRENT_TIME; | |
885 | struct rbd_request *req_data; | |
886 | struct ceph_osd_request_head *reqhead; | |
1dbb4399 | 887 | struct ceph_osd_client *osdc; |
602adf40 | 888 | |
602adf40 | 889 | req_data = kzalloc(sizeof(*req_data), GFP_NOIO); |
1fec7093 YS |
890 | if (!req_data) { |
891 | if (coll) | |
892 | rbd_coll_end_req_index(rq, coll, coll_index, | |
893 | -ENOMEM, len); | |
894 | return -ENOMEM; | |
895 | } | |
896 | ||
897 | if (coll) { | |
898 | req_data->coll = coll; | |
899 | req_data->coll_index = coll_index; | |
900 | } | |
602adf40 | 901 | |
bd919d45 AE |
902 | dout("rbd_do_request object_name=%s ofs=%llu len=%llu\n", object_name, |
903 | (unsigned long long) ofs, (unsigned long long) len); | |
602adf40 | 904 | |
0ce1a794 | 905 | osdc = &rbd_dev->rbd_client->client->osdc; |
1dbb4399 AE |
906 | req = ceph_osdc_alloc_request(osdc, flags, snapc, ops, |
907 | false, GFP_NOIO, pages, bio); | |
4ad12621 | 908 | if (!req) { |
4ad12621 | 909 | ret = -ENOMEM; |
602adf40 YS |
910 | goto done_pages; |
911 | } | |
912 | ||
913 | req->r_callback = rbd_cb; | |
914 | ||
915 | req_data->rq = rq; | |
916 | req_data->bio = bio; | |
917 | req_data->pages = pages; | |
918 | req_data->len = len; | |
919 | ||
920 | req->r_priv = req_data; | |
921 | ||
922 | reqhead = req->r_request->front.iov_base; | |
923 | reqhead->snapid = cpu_to_le64(CEPH_NOSNAP); | |
924 | ||
aded07ea | 925 | strncpy(req->r_oid, object_name, sizeof(req->r_oid)); |
602adf40 YS |
926 | req->r_oid_len = strlen(req->r_oid); |
927 | ||
928 | layout = &req->r_file_layout; | |
929 | memset(layout, 0, sizeof(*layout)); | |
930 | layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
931 | layout->fl_stripe_count = cpu_to_le32(1); | |
932 | layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
0ce1a794 | 933 | layout->fl_pg_pool = cpu_to_le32(rbd_dev->pool_id); |
1dbb4399 AE |
934 | ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno, |
935 | req, ops); | |
602adf40 YS |
936 | |
937 | ceph_osdc_build_request(req, ofs, &len, | |
938 | ops, | |
939 | snapc, | |
940 | &mtime, | |
941 | req->r_oid, req->r_oid_len); | |
602adf40 | 942 | |
59c2be1e | 943 | if (linger_req) { |
1dbb4399 | 944 | ceph_osdc_set_request_linger(osdc, req); |
59c2be1e YS |
945 | *linger_req = req; |
946 | } | |
947 | ||
1dbb4399 | 948 | ret = ceph_osdc_start_request(osdc, req, false); |
602adf40 YS |
949 | if (ret < 0) |
950 | goto done_err; | |
951 | ||
952 | if (!rbd_cb) { | |
1dbb4399 | 953 | ret = ceph_osdc_wait_request(osdc, req); |
59c2be1e YS |
954 | if (ver) |
955 | *ver = le64_to_cpu(req->r_reassert_version.version); | |
bd919d45 AE |
956 | dout("reassert_ver=%llu\n", |
957 | (unsigned long long) | |
958 | le64_to_cpu(req->r_reassert_version.version)); | |
602adf40 YS |
959 | ceph_osdc_put_request(req); |
960 | } | |
961 | return ret; | |
962 | ||
963 | done_err: | |
964 | bio_chain_put(req_data->bio); | |
965 | ceph_osdc_put_request(req); | |
966 | done_pages: | |
1fec7093 | 967 | rbd_coll_end_req(req_data, ret, len); |
602adf40 | 968 | kfree(req_data); |
602adf40 YS |
969 | return ret; |
970 | } | |
971 | ||
972 | /* | |
973 | * Ceph osd op callback | |
974 | */ | |
975 | static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) | |
976 | { | |
977 | struct rbd_request *req_data = req->r_priv; | |
978 | struct ceph_osd_reply_head *replyhead; | |
979 | struct ceph_osd_op *op; | |
980 | __s32 rc; | |
981 | u64 bytes; | |
982 | int read_op; | |
983 | ||
984 | /* parse reply */ | |
985 | replyhead = msg->front.iov_base; | |
986 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
987 | op = (void *)(replyhead + 1); | |
988 | rc = le32_to_cpu(replyhead->result); | |
989 | bytes = le64_to_cpu(op->extent.length); | |
895cfcc8 | 990 | read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ); |
602adf40 | 991 | |
bd919d45 AE |
992 | dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n", |
993 | (unsigned long long) bytes, read_op, (int) rc); | |
602adf40 YS |
994 | |
995 | if (rc == -ENOENT && read_op) { | |
996 | zero_bio_chain(req_data->bio, 0); | |
997 | rc = 0; | |
998 | } else if (rc == 0 && read_op && bytes < req_data->len) { | |
999 | zero_bio_chain(req_data->bio, bytes); | |
1000 | bytes = req_data->len; | |
1001 | } | |
1002 | ||
1fec7093 | 1003 | rbd_coll_end_req(req_data, rc, bytes); |
602adf40 YS |
1004 | |
1005 | if (req_data->bio) | |
1006 | bio_chain_put(req_data->bio); | |
1007 | ||
1008 | ceph_osdc_put_request(req); | |
1009 | kfree(req_data); | |
1010 | } | |
1011 | ||
59c2be1e YS |
1012 | static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) |
1013 | { | |
1014 | ceph_osdc_put_request(req); | |
1015 | } | |
1016 | ||
602adf40 YS |
1017 | /* |
1018 | * Do a synchronous ceph osd operation | |
1019 | */ | |
0ce1a794 | 1020 | static int rbd_req_sync_op(struct rbd_device *rbd_dev, |
602adf40 YS |
1021 | struct ceph_snap_context *snapc, |
1022 | u64 snapid, | |
1023 | int opcode, | |
1024 | int flags, | |
1025 | struct ceph_osd_req_op *orig_ops, | |
aded07ea | 1026 | const char *object_name, |
602adf40 | 1027 | u64 ofs, u64 len, |
59c2be1e YS |
1028 | char *buf, |
1029 | struct ceph_osd_request **linger_req, | |
1030 | u64 *ver) | |
602adf40 YS |
1031 | { |
1032 | int ret; | |
1033 | struct page **pages; | |
1034 | int num_pages; | |
1035 | struct ceph_osd_req_op *ops = orig_ops; | |
1036 | u32 payload_len; | |
1037 | ||
1038 | num_pages = calc_pages_for(ofs , len); | |
1039 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
b8d0638a DC |
1040 | if (IS_ERR(pages)) |
1041 | return PTR_ERR(pages); | |
602adf40 YS |
1042 | |
1043 | if (!orig_ops) { | |
1044 | payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0); | |
1045 | ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len); | |
1046 | if (ret < 0) | |
1047 | goto done; | |
1048 | ||
1049 | if ((flags & CEPH_OSD_FLAG_WRITE) && buf) { | |
1050 | ret = ceph_copy_to_page_vector(pages, buf, ofs, len); | |
1051 | if (ret < 0) | |
1052 | goto done_ops; | |
1053 | } | |
1054 | } | |
1055 | ||
0ce1a794 | 1056 | ret = rbd_do_request(NULL, rbd_dev, snapc, snapid, |
aded07ea | 1057 | object_name, ofs, len, NULL, |
602adf40 YS |
1058 | pages, num_pages, |
1059 | flags, | |
1060 | ops, | |
1fec7093 | 1061 | NULL, 0, |
59c2be1e YS |
1062 | NULL, |
1063 | linger_req, ver); | |
602adf40 YS |
1064 | if (ret < 0) |
1065 | goto done_ops; | |
1066 | ||
1067 | if ((flags & CEPH_OSD_FLAG_READ) && buf) | |
1068 | ret = ceph_copy_from_page_vector(pages, buf, ofs, ret); | |
1069 | ||
1070 | done_ops: | |
1071 | if (!orig_ops) | |
1072 | rbd_destroy_ops(ops); | |
1073 | done: | |
1074 | ceph_release_page_vector(pages, num_pages); | |
1075 | return ret; | |
1076 | } | |
1077 | ||
1078 | /* | |
1079 | * Do an asynchronous ceph osd operation | |
1080 | */ | |
1081 | static int rbd_do_op(struct request *rq, | |
0ce1a794 | 1082 | struct rbd_device *rbd_dev, |
602adf40 YS |
1083 | struct ceph_snap_context *snapc, |
1084 | u64 snapid, | |
d1f57ea6 | 1085 | int opcode, int flags, |
602adf40 | 1086 | u64 ofs, u64 len, |
1fec7093 YS |
1087 | struct bio *bio, |
1088 | struct rbd_req_coll *coll, | |
1089 | int coll_index) | |
602adf40 YS |
1090 | { |
1091 | char *seg_name; | |
1092 | u64 seg_ofs; | |
1093 | u64 seg_len; | |
1094 | int ret; | |
1095 | struct ceph_osd_req_op *ops; | |
1096 | u32 payload_len; | |
1097 | ||
1098 | seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO); | |
1099 | if (!seg_name) | |
1100 | return -ENOMEM; | |
1101 | ||
1102 | seg_len = rbd_get_segment(&rbd_dev->header, | |
ca1e49a6 | 1103 | rbd_dev->header.object_prefix, |
602adf40 YS |
1104 | ofs, len, |
1105 | seg_name, &seg_ofs); | |
602adf40 YS |
1106 | |
1107 | payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0); | |
1108 | ||
1109 | ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len); | |
1110 | if (ret < 0) | |
1111 | goto done; | |
1112 | ||
1113 | /* we've taken care of segment sizes earlier when we | |
1114 | cloned the bios. We should never have a segment | |
1115 | truncated at this point */ | |
1116 | BUG_ON(seg_len < len); | |
1117 | ||
1118 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1119 | seg_name, seg_ofs, seg_len, | |
1120 | bio, | |
1121 | NULL, 0, | |
1122 | flags, | |
1123 | ops, | |
1fec7093 | 1124 | coll, coll_index, |
59c2be1e | 1125 | rbd_req_cb, 0, NULL); |
11f77002 SW |
1126 | |
1127 | rbd_destroy_ops(ops); | |
602adf40 YS |
1128 | done: |
1129 | kfree(seg_name); | |
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * Request async osd write | |
1135 | */ | |
1136 | static int rbd_req_write(struct request *rq, | |
1137 | struct rbd_device *rbd_dev, | |
1138 | struct ceph_snap_context *snapc, | |
1139 | u64 ofs, u64 len, | |
1fec7093 YS |
1140 | struct bio *bio, |
1141 | struct rbd_req_coll *coll, | |
1142 | int coll_index) | |
602adf40 YS |
1143 | { |
1144 | return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP, | |
1145 | CEPH_OSD_OP_WRITE, | |
1146 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1fec7093 | 1147 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1148 | } |
1149 | ||
1150 | /* | |
1151 | * Request async osd read | |
1152 | */ | |
1153 | static int rbd_req_read(struct request *rq, | |
1154 | struct rbd_device *rbd_dev, | |
1155 | u64 snapid, | |
1156 | u64 ofs, u64 len, | |
1fec7093 YS |
1157 | struct bio *bio, |
1158 | struct rbd_req_coll *coll, | |
1159 | int coll_index) | |
602adf40 YS |
1160 | { |
1161 | return rbd_do_op(rq, rbd_dev, NULL, | |
b06e6a6b | 1162 | snapid, |
602adf40 YS |
1163 | CEPH_OSD_OP_READ, |
1164 | CEPH_OSD_FLAG_READ, | |
1fec7093 | 1165 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1166 | } |
1167 | ||
1168 | /* | |
1169 | * Request sync osd read | |
1170 | */ | |
0ce1a794 | 1171 | static int rbd_req_sync_read(struct rbd_device *rbd_dev, |
602adf40 YS |
1172 | struct ceph_snap_context *snapc, |
1173 | u64 snapid, | |
aded07ea | 1174 | const char *object_name, |
602adf40 | 1175 | u64 ofs, u64 len, |
59c2be1e YS |
1176 | char *buf, |
1177 | u64 *ver) | |
602adf40 | 1178 | { |
0ce1a794 | 1179 | return rbd_req_sync_op(rbd_dev, NULL, |
b06e6a6b | 1180 | snapid, |
602adf40 YS |
1181 | CEPH_OSD_OP_READ, |
1182 | CEPH_OSD_FLAG_READ, | |
1183 | NULL, | |
d1f57ea6 | 1184 | object_name, ofs, len, buf, NULL, ver); |
602adf40 YS |
1185 | } |
1186 | ||
1187 | /* | |
59c2be1e YS |
1188 | * Request sync osd watch |
1189 | */ | |
0ce1a794 | 1190 | static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev, |
59c2be1e YS |
1191 | u64 ver, |
1192 | u64 notify_id, | |
aded07ea | 1193 | const char *object_name) |
59c2be1e YS |
1194 | { |
1195 | struct ceph_osd_req_op *ops; | |
11f77002 SW |
1196 | int ret; |
1197 | ||
1198 | ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0); | |
59c2be1e YS |
1199 | if (ret < 0) |
1200 | return ret; | |
1201 | ||
a71b891b | 1202 | ops[0].watch.ver = cpu_to_le64(ver); |
59c2be1e YS |
1203 | ops[0].watch.cookie = notify_id; |
1204 | ops[0].watch.flag = 0; | |
1205 | ||
0ce1a794 | 1206 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
aded07ea | 1207 | object_name, 0, 0, NULL, |
ad4f232f | 1208 | NULL, 0, |
59c2be1e YS |
1209 | CEPH_OSD_FLAG_READ, |
1210 | ops, | |
1fec7093 | 1211 | NULL, 0, |
59c2be1e YS |
1212 | rbd_simple_req_cb, 0, NULL); |
1213 | ||
1214 | rbd_destroy_ops(ops); | |
1215 | return ret; | |
1216 | } | |
1217 | ||
1218 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1219 | { | |
0ce1a794 | 1220 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
a71b891b | 1221 | u64 hver; |
13143d2d SW |
1222 | int rc; |
1223 | ||
0ce1a794 | 1224 | if (!rbd_dev) |
59c2be1e YS |
1225 | return; |
1226 | ||
bd919d45 AE |
1227 | dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n", |
1228 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1229 | (unsigned int) opcode); | |
59c2be1e | 1230 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
0ce1a794 | 1231 | rc = __rbd_refresh_header(rbd_dev); |
a71b891b | 1232 | hver = rbd_dev->header.obj_version; |
59c2be1e | 1233 | mutex_unlock(&ctl_mutex); |
13143d2d | 1234 | if (rc) |
f0f8cef5 | 1235 | pr_warning(RBD_DRV_NAME "%d got notification but failed to " |
0ce1a794 | 1236 | " update snaps: %d\n", rbd_dev->major, rc); |
59c2be1e | 1237 | |
a71b891b | 1238 | rbd_req_sync_notify_ack(rbd_dev, hver, notify_id, rbd_dev->header_name); |
59c2be1e YS |
1239 | } |
1240 | ||
1241 | /* | |
1242 | * Request sync osd watch | |
1243 | */ | |
0ce1a794 | 1244 | static int rbd_req_sync_watch(struct rbd_device *rbd_dev, |
aded07ea | 1245 | const char *object_name, |
59c2be1e YS |
1246 | u64 ver) |
1247 | { | |
1248 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1249 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
59c2be1e YS |
1250 | |
1251 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0); | |
1252 | if (ret < 0) | |
1253 | return ret; | |
1254 | ||
1255 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, | |
0ce1a794 | 1256 | (void *)rbd_dev, &rbd_dev->watch_event); |
59c2be1e YS |
1257 | if (ret < 0) |
1258 | goto fail; | |
1259 | ||
1260 | ops[0].watch.ver = cpu_to_le64(ver); | |
0ce1a794 | 1261 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
59c2be1e YS |
1262 | ops[0].watch.flag = 1; |
1263 | ||
0ce1a794 | 1264 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e YS |
1265 | CEPH_NOSNAP, |
1266 | 0, | |
1267 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1268 | ops, | |
d1f57ea6 | 1269 | object_name, 0, 0, NULL, |
0ce1a794 | 1270 | &rbd_dev->watch_request, NULL); |
59c2be1e YS |
1271 | |
1272 | if (ret < 0) | |
1273 | goto fail_event; | |
1274 | ||
1275 | rbd_destroy_ops(ops); | |
1276 | return 0; | |
1277 | ||
1278 | fail_event: | |
0ce1a794 AE |
1279 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1280 | rbd_dev->watch_event = NULL; | |
59c2be1e YS |
1281 | fail: |
1282 | rbd_destroy_ops(ops); | |
1283 | return ret; | |
1284 | } | |
1285 | ||
79e3057c YS |
1286 | /* |
1287 | * Request sync osd unwatch | |
1288 | */ | |
0ce1a794 | 1289 | static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev, |
aded07ea | 1290 | const char *object_name) |
79e3057c YS |
1291 | { |
1292 | struct ceph_osd_req_op *ops; | |
1293 | ||
1294 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0); | |
1295 | if (ret < 0) | |
1296 | return ret; | |
1297 | ||
1298 | ops[0].watch.ver = 0; | |
0ce1a794 | 1299 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
79e3057c YS |
1300 | ops[0].watch.flag = 0; |
1301 | ||
0ce1a794 | 1302 | ret = rbd_req_sync_op(rbd_dev, NULL, |
79e3057c YS |
1303 | CEPH_NOSNAP, |
1304 | 0, | |
1305 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1306 | ops, | |
d1f57ea6 | 1307 | object_name, 0, 0, NULL, NULL, NULL); |
79e3057c YS |
1308 | |
1309 | rbd_destroy_ops(ops); | |
0ce1a794 AE |
1310 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1311 | rbd_dev->watch_event = NULL; | |
79e3057c YS |
1312 | return ret; |
1313 | } | |
1314 | ||
59c2be1e | 1315 | struct rbd_notify_info { |
0ce1a794 | 1316 | struct rbd_device *rbd_dev; |
59c2be1e YS |
1317 | }; |
1318 | ||
1319 | static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1320 | { | |
0ce1a794 AE |
1321 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
1322 | if (!rbd_dev) | |
59c2be1e YS |
1323 | return; |
1324 | ||
bd919d45 AE |
1325 | dout("rbd_notify_cb %s notify_id=%llu opcode=%u\n", |
1326 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1327 | (unsigned int) opcode); | |
59c2be1e YS |
1328 | } |
1329 | ||
1330 | /* | |
1331 | * Request sync osd notify | |
1332 | */ | |
0ce1a794 | 1333 | static int rbd_req_sync_notify(struct rbd_device *rbd_dev, |
aded07ea | 1334 | const char *object_name) |
59c2be1e YS |
1335 | { |
1336 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1337 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
59c2be1e YS |
1338 | struct ceph_osd_event *event; |
1339 | struct rbd_notify_info info; | |
1340 | int payload_len = sizeof(u32) + sizeof(u32); | |
1341 | int ret; | |
1342 | ||
1343 | ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len); | |
1344 | if (ret < 0) | |
1345 | return ret; | |
1346 | ||
0ce1a794 | 1347 | info.rbd_dev = rbd_dev; |
59c2be1e YS |
1348 | |
1349 | ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1, | |
1350 | (void *)&info, &event); | |
1351 | if (ret < 0) | |
1352 | goto fail; | |
1353 | ||
1354 | ops[0].watch.ver = 1; | |
1355 | ops[0].watch.flag = 1; | |
1356 | ops[0].watch.cookie = event->cookie; | |
1357 | ops[0].watch.prot_ver = RADOS_NOTIFY_VER; | |
1358 | ops[0].watch.timeout = 12; | |
1359 | ||
0ce1a794 | 1360 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e YS |
1361 | CEPH_NOSNAP, |
1362 | 0, | |
1363 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1364 | ops, | |
d1f57ea6 | 1365 | object_name, 0, 0, NULL, NULL, NULL); |
59c2be1e YS |
1366 | if (ret < 0) |
1367 | goto fail_event; | |
1368 | ||
1369 | ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT); | |
1370 | dout("ceph_osdc_wait_event returned %d\n", ret); | |
1371 | rbd_destroy_ops(ops); | |
1372 | return 0; | |
1373 | ||
1374 | fail_event: | |
1375 | ceph_osdc_cancel_event(event); | |
1376 | fail: | |
1377 | rbd_destroy_ops(ops); | |
1378 | return ret; | |
1379 | } | |
1380 | ||
602adf40 YS |
1381 | /* |
1382 | * Request sync osd read | |
1383 | */ | |
0ce1a794 | 1384 | static int rbd_req_sync_exec(struct rbd_device *rbd_dev, |
aded07ea AE |
1385 | const char *object_name, |
1386 | const char *class_name, | |
1387 | const char *method_name, | |
602adf40 | 1388 | const char *data, |
59c2be1e YS |
1389 | int len, |
1390 | u64 *ver) | |
602adf40 YS |
1391 | { |
1392 | struct ceph_osd_req_op *ops; | |
aded07ea AE |
1393 | int class_name_len = strlen(class_name); |
1394 | int method_name_len = strlen(method_name); | |
602adf40 | 1395 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL, |
aded07ea | 1396 | class_name_len + method_name_len + len); |
602adf40 YS |
1397 | if (ret < 0) |
1398 | return ret; | |
1399 | ||
aded07ea AE |
1400 | ops[0].cls.class_name = class_name; |
1401 | ops[0].cls.class_len = (__u8) class_name_len; | |
1402 | ops[0].cls.method_name = method_name; | |
1403 | ops[0].cls.method_len = (__u8) method_name_len; | |
602adf40 YS |
1404 | ops[0].cls.argc = 0; |
1405 | ops[0].cls.indata = data; | |
1406 | ops[0].cls.indata_len = len; | |
1407 | ||
0ce1a794 | 1408 | ret = rbd_req_sync_op(rbd_dev, NULL, |
602adf40 YS |
1409 | CEPH_NOSNAP, |
1410 | 0, | |
1411 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1412 | ops, | |
d1f57ea6 | 1413 | object_name, 0, 0, NULL, NULL, ver); |
602adf40 YS |
1414 | |
1415 | rbd_destroy_ops(ops); | |
1416 | ||
1417 | dout("cls_exec returned %d\n", ret); | |
1418 | return ret; | |
1419 | } | |
1420 | ||
1fec7093 YS |
1421 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1422 | { | |
1423 | struct rbd_req_coll *coll = | |
1424 | kzalloc(sizeof(struct rbd_req_coll) + | |
1425 | sizeof(struct rbd_req_status) * num_reqs, | |
1426 | GFP_ATOMIC); | |
1427 | ||
1428 | if (!coll) | |
1429 | return NULL; | |
1430 | coll->total = num_reqs; | |
1431 | kref_init(&coll->kref); | |
1432 | return coll; | |
1433 | } | |
1434 | ||
602adf40 YS |
1435 | /* |
1436 | * block device queue callback | |
1437 | */ | |
1438 | static void rbd_rq_fn(struct request_queue *q) | |
1439 | { | |
1440 | struct rbd_device *rbd_dev = q->queuedata; | |
1441 | struct request *rq; | |
1442 | struct bio_pair *bp = NULL; | |
1443 | ||
00f1f36f | 1444 | while ((rq = blk_fetch_request(q))) { |
602adf40 YS |
1445 | struct bio *bio; |
1446 | struct bio *rq_bio, *next_bio = NULL; | |
1447 | bool do_write; | |
bd919d45 AE |
1448 | unsigned int size; |
1449 | u64 op_size = 0; | |
602adf40 | 1450 | u64 ofs; |
1fec7093 YS |
1451 | int num_segs, cur_seg = 0; |
1452 | struct rbd_req_coll *coll; | |
d1d25646 | 1453 | struct ceph_snap_context *snapc; |
602adf40 YS |
1454 | |
1455 | /* peek at request from block layer */ | |
1456 | if (!rq) | |
1457 | break; | |
1458 | ||
1459 | dout("fetched request\n"); | |
1460 | ||
1461 | /* filter out block requests we don't understand */ | |
1462 | if ((rq->cmd_type != REQ_TYPE_FS)) { | |
1463 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1464 | continue; |
602adf40 YS |
1465 | } |
1466 | ||
1467 | /* deduce our operation (read, write) */ | |
1468 | do_write = (rq_data_dir(rq) == WRITE); | |
1469 | ||
1470 | size = blk_rq_bytes(rq); | |
593a9e7b | 1471 | ofs = blk_rq_pos(rq) * SECTOR_SIZE; |
602adf40 YS |
1472 | rq_bio = rq->bio; |
1473 | if (do_write && rbd_dev->read_only) { | |
1474 | __blk_end_request_all(rq, -EROFS); | |
00f1f36f | 1475 | continue; |
602adf40 YS |
1476 | } |
1477 | ||
1478 | spin_unlock_irq(q->queue_lock); | |
1479 | ||
d1d25646 | 1480 | down_read(&rbd_dev->header_rwsem); |
e88a36ec | 1481 | |
d1d25646 | 1482 | if (rbd_dev->snap_id != CEPH_NOSNAP && !rbd_dev->snap_exists) { |
e88a36ec | 1483 | up_read(&rbd_dev->header_rwsem); |
d1d25646 JD |
1484 | dout("request for non-existent snapshot"); |
1485 | spin_lock_irq(q->queue_lock); | |
1486 | __blk_end_request_all(rq, -ENXIO); | |
1487 | continue; | |
e88a36ec JD |
1488 | } |
1489 | ||
d1d25646 JD |
1490 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
1491 | ||
1492 | up_read(&rbd_dev->header_rwsem); | |
1493 | ||
602adf40 YS |
1494 | dout("%s 0x%x bytes at 0x%llx\n", |
1495 | do_write ? "write" : "read", | |
bd919d45 | 1496 | size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE); |
602adf40 | 1497 | |
1fec7093 YS |
1498 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); |
1499 | coll = rbd_alloc_coll(num_segs); | |
1500 | if (!coll) { | |
1501 | spin_lock_irq(q->queue_lock); | |
1502 | __blk_end_request_all(rq, -ENOMEM); | |
d1d25646 | 1503 | ceph_put_snap_context(snapc); |
00f1f36f | 1504 | continue; |
1fec7093 YS |
1505 | } |
1506 | ||
602adf40 YS |
1507 | do { |
1508 | /* a bio clone to be passed down to OSD req */ | |
bd919d45 | 1509 | dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt); |
602adf40 | 1510 | op_size = rbd_get_segment(&rbd_dev->header, |
ca1e49a6 | 1511 | rbd_dev->header.object_prefix, |
602adf40 YS |
1512 | ofs, size, |
1513 | NULL, NULL); | |
1fec7093 | 1514 | kref_get(&coll->kref); |
602adf40 YS |
1515 | bio = bio_chain_clone(&rq_bio, &next_bio, &bp, |
1516 | op_size, GFP_ATOMIC); | |
1517 | if (!bio) { | |
1fec7093 YS |
1518 | rbd_coll_end_req_index(rq, coll, cur_seg, |
1519 | -ENOMEM, op_size); | |
1520 | goto next_seg; | |
602adf40 YS |
1521 | } |
1522 | ||
1fec7093 | 1523 | |
602adf40 YS |
1524 | /* init OSD command: write or read */ |
1525 | if (do_write) | |
1526 | rbd_req_write(rq, rbd_dev, | |
d1d25646 | 1527 | snapc, |
602adf40 | 1528 | ofs, |
1fec7093 YS |
1529 | op_size, bio, |
1530 | coll, cur_seg); | |
602adf40 YS |
1531 | else |
1532 | rbd_req_read(rq, rbd_dev, | |
77dfe99f | 1533 | rbd_dev->snap_id, |
602adf40 | 1534 | ofs, |
1fec7093 YS |
1535 | op_size, bio, |
1536 | coll, cur_seg); | |
602adf40 | 1537 | |
1fec7093 | 1538 | next_seg: |
602adf40 YS |
1539 | size -= op_size; |
1540 | ofs += op_size; | |
1541 | ||
1fec7093 | 1542 | cur_seg++; |
602adf40 YS |
1543 | rq_bio = next_bio; |
1544 | } while (size > 0); | |
1fec7093 | 1545 | kref_put(&coll->kref, rbd_coll_release); |
602adf40 YS |
1546 | |
1547 | if (bp) | |
1548 | bio_pair_release(bp); | |
602adf40 | 1549 | spin_lock_irq(q->queue_lock); |
d1d25646 JD |
1550 | |
1551 | ceph_put_snap_context(snapc); | |
602adf40 YS |
1552 | } |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1557 | * multiple osd objects. One exception would be with a single page bios, | |
1558 | * which we handle later at bio_chain_clone | |
1559 | */ | |
1560 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1561 | struct bio_vec *bvec) | |
1562 | { | |
1563 | struct rbd_device *rbd_dev = q->queuedata; | |
593a9e7b AE |
1564 | unsigned int chunk_sectors; |
1565 | sector_t sector; | |
1566 | unsigned int bio_sectors; | |
602adf40 YS |
1567 | int max; |
1568 | ||
593a9e7b AE |
1569 | chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); |
1570 | sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev); | |
1571 | bio_sectors = bmd->bi_size >> SECTOR_SHIFT; | |
1572 | ||
602adf40 | 1573 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) |
593a9e7b | 1574 | + bio_sectors)) << SECTOR_SHIFT; |
602adf40 YS |
1575 | if (max < 0) |
1576 | max = 0; /* bio_add cannot handle a negative return */ | |
1577 | if (max <= bvec->bv_len && bio_sectors == 0) | |
1578 | return bvec->bv_len; | |
1579 | return max; | |
1580 | } | |
1581 | ||
1582 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1583 | { | |
1584 | struct gendisk *disk = rbd_dev->disk; | |
1585 | ||
1586 | if (!disk) | |
1587 | return; | |
1588 | ||
1589 | rbd_header_free(&rbd_dev->header); | |
1590 | ||
1591 | if (disk->flags & GENHD_FL_UP) | |
1592 | del_gendisk(disk); | |
1593 | if (disk->queue) | |
1594 | blk_cleanup_queue(disk->queue); | |
1595 | put_disk(disk); | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * reload the ondisk the header | |
1600 | */ | |
1601 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1602 | struct rbd_image_header *header) | |
1603 | { | |
1604 | ssize_t rc; | |
1605 | struct rbd_image_header_ondisk *dh; | |
50f7c4c9 | 1606 | u32 snap_count = 0; |
59c2be1e | 1607 | u64 ver; |
00f1f36f | 1608 | size_t len; |
602adf40 | 1609 | |
00f1f36f AE |
1610 | /* |
1611 | * First reads the fixed-size header to determine the number | |
1612 | * of snapshots, then re-reads it, along with all snapshot | |
1613 | * records as well as their stored names. | |
1614 | */ | |
1615 | len = sizeof (*dh); | |
602adf40 | 1616 | while (1) { |
602adf40 YS |
1617 | dh = kmalloc(len, GFP_KERNEL); |
1618 | if (!dh) | |
1619 | return -ENOMEM; | |
1620 | ||
1621 | rc = rbd_req_sync_read(rbd_dev, | |
1622 | NULL, CEPH_NOSNAP, | |
0bed54dc | 1623 | rbd_dev->header_name, |
602adf40 | 1624 | 0, len, |
59c2be1e | 1625 | (char *)dh, &ver); |
602adf40 YS |
1626 | if (rc < 0) |
1627 | goto out_dh; | |
1628 | ||
1629 | rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL); | |
81e759fb | 1630 | if (rc < 0) { |
00f1f36f | 1631 | if (rc == -ENXIO) |
81e759fb | 1632 | pr_warning("unrecognized header format" |
0bed54dc AE |
1633 | " for image %s\n", |
1634 | rbd_dev->image_name); | |
602adf40 | 1635 | goto out_dh; |
81e759fb | 1636 | } |
602adf40 | 1637 | |
00f1f36f AE |
1638 | if (snap_count == header->total_snaps) |
1639 | break; | |
1640 | ||
1641 | snap_count = header->total_snaps; | |
1642 | len = sizeof (*dh) + | |
1643 | snap_count * sizeof(struct rbd_image_snap_ondisk) + | |
1644 | header->snap_names_len; | |
1645 | ||
1646 | rbd_header_free(header); | |
1647 | kfree(dh); | |
602adf40 | 1648 | } |
59c2be1e | 1649 | header->obj_version = ver; |
602adf40 YS |
1650 | |
1651 | out_dh: | |
1652 | kfree(dh); | |
1653 | return rc; | |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * create a snapshot | |
1658 | */ | |
0ce1a794 | 1659 | static int rbd_header_add_snap(struct rbd_device *rbd_dev, |
602adf40 YS |
1660 | const char *snap_name, |
1661 | gfp_t gfp_flags) | |
1662 | { | |
1663 | int name_len = strlen(snap_name); | |
1664 | u64 new_snapid; | |
1665 | int ret; | |
916d4d67 | 1666 | void *data, *p, *e; |
59c2be1e | 1667 | u64 ver; |
1dbb4399 | 1668 | struct ceph_mon_client *monc; |
602adf40 YS |
1669 | |
1670 | /* we should create a snapshot only if we're pointing at the head */ | |
0ce1a794 | 1671 | if (rbd_dev->snap_id != CEPH_NOSNAP) |
602adf40 YS |
1672 | return -EINVAL; |
1673 | ||
0ce1a794 AE |
1674 | monc = &rbd_dev->rbd_client->client->monc; |
1675 | ret = ceph_monc_create_snapid(monc, rbd_dev->pool_id, &new_snapid); | |
bd919d45 | 1676 | dout("created snapid=%llu\n", (unsigned long long) new_snapid); |
602adf40 YS |
1677 | if (ret < 0) |
1678 | return ret; | |
1679 | ||
1680 | data = kmalloc(name_len + 16, gfp_flags); | |
1681 | if (!data) | |
1682 | return -ENOMEM; | |
1683 | ||
916d4d67 SW |
1684 | p = data; |
1685 | e = data + name_len + 16; | |
602adf40 | 1686 | |
916d4d67 SW |
1687 | ceph_encode_string_safe(&p, e, snap_name, name_len, bad); |
1688 | ceph_encode_64_safe(&p, e, new_snapid, bad); | |
602adf40 | 1689 | |
0bed54dc | 1690 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, |
0ce1a794 | 1691 | "rbd", "snap_add", |
916d4d67 | 1692 | data, p - data, &ver); |
602adf40 | 1693 | |
916d4d67 | 1694 | kfree(data); |
602adf40 | 1695 | |
505cbb9b | 1696 | return ret < 0 ? ret : 0; |
602adf40 YS |
1697 | bad: |
1698 | return -ERANGE; | |
1699 | } | |
1700 | ||
dfc5606d YS |
1701 | static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
1702 | { | |
1703 | struct rbd_snap *snap; | |
a0593290 | 1704 | struct rbd_snap *next; |
dfc5606d | 1705 | |
a0593290 | 1706 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
dfc5606d | 1707 | __rbd_remove_snap_dev(rbd_dev, snap); |
dfc5606d YS |
1708 | } |
1709 | ||
602adf40 YS |
1710 | /* |
1711 | * only read the first part of the ondisk header, without the snaps info | |
1712 | */ | |
263c6ca0 | 1713 | static int __rbd_refresh_header(struct rbd_device *rbd_dev) |
602adf40 YS |
1714 | { |
1715 | int ret; | |
1716 | struct rbd_image_header h; | |
602adf40 YS |
1717 | |
1718 | ret = rbd_read_header(rbd_dev, &h); | |
1719 | if (ret < 0) | |
1720 | return ret; | |
1721 | ||
a51aa0c0 JD |
1722 | down_write(&rbd_dev->header_rwsem); |
1723 | ||
9db4b3e3 | 1724 | /* resized? */ |
474ef7ce JD |
1725 | if (rbd_dev->snap_id == CEPH_NOSNAP) { |
1726 | sector_t size = (sector_t) h.image_size / SECTOR_SIZE; | |
1727 | ||
1728 | dout("setting size to %llu sectors", (unsigned long long) size); | |
1729 | set_capacity(rbd_dev->disk, size); | |
1730 | } | |
9db4b3e3 | 1731 | |
849b4260 | 1732 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 1733 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 1734 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
1735 | /* osd requests may still refer to snapc */ |
1736 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 1737 | |
a71b891b | 1738 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 1739 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
1740 | rbd_dev->header.total_snaps = h.total_snaps; |
1741 | rbd_dev->header.snapc = h.snapc; | |
1742 | rbd_dev->header.snap_names = h.snap_names; | |
dfc5606d | 1743 | rbd_dev->header.snap_names_len = h.snap_names_len; |
602adf40 | 1744 | rbd_dev->header.snap_sizes = h.snap_sizes; |
849b4260 AE |
1745 | /* Free the extra copy of the object prefix */ |
1746 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
1747 | kfree(h.object_prefix); | |
1748 | ||
dfc5606d YS |
1749 | ret = __rbd_init_snaps_header(rbd_dev); |
1750 | ||
c666601a | 1751 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1752 | |
dfc5606d | 1753 | return ret; |
602adf40 YS |
1754 | } |
1755 | ||
1756 | static int rbd_init_disk(struct rbd_device *rbd_dev) | |
1757 | { | |
1758 | struct gendisk *disk; | |
1759 | struct request_queue *q; | |
1760 | int rc; | |
593a9e7b | 1761 | u64 segment_size; |
602adf40 YS |
1762 | u64 total_size = 0; |
1763 | ||
1764 | /* contact OSD, request size info about the object being mapped */ | |
1765 | rc = rbd_read_header(rbd_dev, &rbd_dev->header); | |
1766 | if (rc) | |
1767 | return rc; | |
1768 | ||
dfc5606d YS |
1769 | /* no need to lock here, as rbd_dev is not registered yet */ |
1770 | rc = __rbd_init_snaps_header(rbd_dev); | |
1771 | if (rc) | |
1772 | return rc; | |
1773 | ||
cc9d734c | 1774 | rc = rbd_header_set_snap(rbd_dev, &total_size); |
602adf40 YS |
1775 | if (rc) |
1776 | return rc; | |
1777 | ||
1778 | /* create gendisk info */ | |
1779 | rc = -ENOMEM; | |
1780 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); | |
1781 | if (!disk) | |
1782 | goto out; | |
1783 | ||
f0f8cef5 | 1784 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
aedfec59 | 1785 | rbd_dev->id); |
602adf40 YS |
1786 | disk->major = rbd_dev->major; |
1787 | disk->first_minor = 0; | |
1788 | disk->fops = &rbd_bd_ops; | |
1789 | disk->private_data = rbd_dev; | |
1790 | ||
1791 | /* init rq */ | |
1792 | rc = -ENOMEM; | |
1793 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); | |
1794 | if (!q) | |
1795 | goto out_disk; | |
029bcbd8 | 1796 | |
593a9e7b AE |
1797 | /* We use the default size, but let's be explicit about it. */ |
1798 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1799 | ||
029bcbd8 | 1800 | /* set io sizes to object size */ |
593a9e7b AE |
1801 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1802 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1803 | blk_queue_max_segment_size(q, segment_size); | |
1804 | blk_queue_io_min(q, segment_size); | |
1805 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1806 | |
602adf40 YS |
1807 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1808 | disk->queue = q; | |
1809 | ||
1810 | q->queuedata = rbd_dev; | |
1811 | ||
1812 | rbd_dev->disk = disk; | |
1813 | rbd_dev->q = q; | |
1814 | ||
1815 | /* finally, announce the disk to the world */ | |
593a9e7b | 1816 | set_capacity(disk, total_size / SECTOR_SIZE); |
602adf40 YS |
1817 | add_disk(disk); |
1818 | ||
1819 | pr_info("%s: added with size 0x%llx\n", | |
1820 | disk->disk_name, (unsigned long long)total_size); | |
1821 | return 0; | |
1822 | ||
1823 | out_disk: | |
1824 | put_disk(disk); | |
1825 | out: | |
1826 | return rc; | |
1827 | } | |
1828 | ||
dfc5606d YS |
1829 | /* |
1830 | sysfs | |
1831 | */ | |
1832 | ||
593a9e7b AE |
1833 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1834 | { | |
1835 | return container_of(dev, struct rbd_device, dev); | |
1836 | } | |
1837 | ||
dfc5606d YS |
1838 | static ssize_t rbd_size_show(struct device *dev, |
1839 | struct device_attribute *attr, char *buf) | |
1840 | { | |
593a9e7b | 1841 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
1842 | sector_t size; |
1843 | ||
1844 | down_read(&rbd_dev->header_rwsem); | |
1845 | size = get_capacity(rbd_dev->disk); | |
1846 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 1847 | |
a51aa0c0 | 1848 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
1849 | } |
1850 | ||
1851 | static ssize_t rbd_major_show(struct device *dev, | |
1852 | struct device_attribute *attr, char *buf) | |
1853 | { | |
593a9e7b | 1854 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 1855 | |
dfc5606d YS |
1856 | return sprintf(buf, "%d\n", rbd_dev->major); |
1857 | } | |
1858 | ||
1859 | static ssize_t rbd_client_id_show(struct device *dev, | |
1860 | struct device_attribute *attr, char *buf) | |
602adf40 | 1861 | { |
593a9e7b | 1862 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1863 | |
1dbb4399 AE |
1864 | return sprintf(buf, "client%lld\n", |
1865 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
1866 | } |
1867 | ||
dfc5606d YS |
1868 | static ssize_t rbd_pool_show(struct device *dev, |
1869 | struct device_attribute *attr, char *buf) | |
602adf40 | 1870 | { |
593a9e7b | 1871 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1872 | |
1873 | return sprintf(buf, "%s\n", rbd_dev->pool_name); | |
1874 | } | |
1875 | ||
9bb2f334 AE |
1876 | static ssize_t rbd_pool_id_show(struct device *dev, |
1877 | struct device_attribute *attr, char *buf) | |
1878 | { | |
1879 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1880 | ||
1881 | return sprintf(buf, "%d\n", rbd_dev->pool_id); | |
1882 | } | |
1883 | ||
dfc5606d YS |
1884 | static ssize_t rbd_name_show(struct device *dev, |
1885 | struct device_attribute *attr, char *buf) | |
1886 | { | |
593a9e7b | 1887 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1888 | |
0bed54dc | 1889 | return sprintf(buf, "%s\n", rbd_dev->image_name); |
dfc5606d YS |
1890 | } |
1891 | ||
1892 | static ssize_t rbd_snap_show(struct device *dev, | |
1893 | struct device_attribute *attr, | |
1894 | char *buf) | |
1895 | { | |
593a9e7b | 1896 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1897 | |
1898 | return sprintf(buf, "%s\n", rbd_dev->snap_name); | |
1899 | } | |
1900 | ||
1901 | static ssize_t rbd_image_refresh(struct device *dev, | |
1902 | struct device_attribute *attr, | |
1903 | const char *buf, | |
1904 | size_t size) | |
1905 | { | |
593a9e7b | 1906 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1907 | int rc; |
1908 | int ret = size; | |
602adf40 YS |
1909 | |
1910 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
1911 | ||
263c6ca0 | 1912 | rc = __rbd_refresh_header(rbd_dev); |
dfc5606d YS |
1913 | if (rc < 0) |
1914 | ret = rc; | |
602adf40 | 1915 | |
dfc5606d YS |
1916 | mutex_unlock(&ctl_mutex); |
1917 | return ret; | |
1918 | } | |
602adf40 | 1919 | |
dfc5606d YS |
1920 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
1921 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); | |
1922 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
1923 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 1924 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d YS |
1925 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
1926 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); | |
1927 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
1928 | static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add); | |
dfc5606d YS |
1929 | |
1930 | static struct attribute *rbd_attrs[] = { | |
1931 | &dev_attr_size.attr, | |
1932 | &dev_attr_major.attr, | |
1933 | &dev_attr_client_id.attr, | |
1934 | &dev_attr_pool.attr, | |
9bb2f334 | 1935 | &dev_attr_pool_id.attr, |
dfc5606d YS |
1936 | &dev_attr_name.attr, |
1937 | &dev_attr_current_snap.attr, | |
1938 | &dev_attr_refresh.attr, | |
1939 | &dev_attr_create_snap.attr, | |
dfc5606d YS |
1940 | NULL |
1941 | }; | |
1942 | ||
1943 | static struct attribute_group rbd_attr_group = { | |
1944 | .attrs = rbd_attrs, | |
1945 | }; | |
1946 | ||
1947 | static const struct attribute_group *rbd_attr_groups[] = { | |
1948 | &rbd_attr_group, | |
1949 | NULL | |
1950 | }; | |
1951 | ||
1952 | static void rbd_sysfs_dev_release(struct device *dev) | |
1953 | { | |
1954 | } | |
1955 | ||
1956 | static struct device_type rbd_device_type = { | |
1957 | .name = "rbd", | |
1958 | .groups = rbd_attr_groups, | |
1959 | .release = rbd_sysfs_dev_release, | |
1960 | }; | |
1961 | ||
1962 | ||
1963 | /* | |
1964 | sysfs - snapshots | |
1965 | */ | |
1966 | ||
1967 | static ssize_t rbd_snap_size_show(struct device *dev, | |
1968 | struct device_attribute *attr, | |
1969 | char *buf) | |
1970 | { | |
1971 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1972 | ||
3591538f | 1973 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
1974 | } |
1975 | ||
1976 | static ssize_t rbd_snap_id_show(struct device *dev, | |
1977 | struct device_attribute *attr, | |
1978 | char *buf) | |
1979 | { | |
1980 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1981 | ||
3591538f | 1982 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
1983 | } |
1984 | ||
1985 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); | |
1986 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
1987 | ||
1988 | static struct attribute *rbd_snap_attrs[] = { | |
1989 | &dev_attr_snap_size.attr, | |
1990 | &dev_attr_snap_id.attr, | |
1991 | NULL, | |
1992 | }; | |
1993 | ||
1994 | static struct attribute_group rbd_snap_attr_group = { | |
1995 | .attrs = rbd_snap_attrs, | |
1996 | }; | |
1997 | ||
1998 | static void rbd_snap_dev_release(struct device *dev) | |
1999 | { | |
2000 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2001 | kfree(snap->name); | |
2002 | kfree(snap); | |
2003 | } | |
2004 | ||
2005 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2006 | &rbd_snap_attr_group, | |
2007 | NULL | |
2008 | }; | |
2009 | ||
2010 | static struct device_type rbd_snap_device_type = { | |
2011 | .groups = rbd_snap_attr_groups, | |
2012 | .release = rbd_snap_dev_release, | |
2013 | }; | |
2014 | ||
2015 | static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev, | |
2016 | struct rbd_snap *snap) | |
2017 | { | |
2018 | list_del(&snap->node); | |
2019 | device_unregister(&snap->dev); | |
2020 | } | |
2021 | ||
2022 | static int rbd_register_snap_dev(struct rbd_device *rbd_dev, | |
2023 | struct rbd_snap *snap, | |
2024 | struct device *parent) | |
2025 | { | |
2026 | struct device *dev = &snap->dev; | |
2027 | int ret; | |
2028 | ||
2029 | dev->type = &rbd_snap_device_type; | |
2030 | dev->parent = parent; | |
2031 | dev->release = rbd_snap_dev_release; | |
2032 | dev_set_name(dev, "snap_%s", snap->name); | |
2033 | ret = device_register(dev); | |
2034 | ||
2035 | return ret; | |
2036 | } | |
2037 | ||
2038 | static int __rbd_add_snap_dev(struct rbd_device *rbd_dev, | |
2039 | int i, const char *name, | |
2040 | struct rbd_snap **snapp) | |
2041 | { | |
2042 | int ret; | |
2043 | struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL); | |
2044 | if (!snap) | |
2045 | return -ENOMEM; | |
2046 | snap->name = kstrdup(name, GFP_KERNEL); | |
2047 | snap->size = rbd_dev->header.snap_sizes[i]; | |
2048 | snap->id = rbd_dev->header.snapc->snaps[i]; | |
2049 | if (device_is_registered(&rbd_dev->dev)) { | |
2050 | ret = rbd_register_snap_dev(rbd_dev, snap, | |
2051 | &rbd_dev->dev); | |
2052 | if (ret < 0) | |
2053 | goto err; | |
2054 | } | |
2055 | *snapp = snap; | |
2056 | return 0; | |
2057 | err: | |
2058 | kfree(snap->name); | |
2059 | kfree(snap); | |
2060 | return ret; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * search for the previous snap in a null delimited string list | |
2065 | */ | |
2066 | const char *rbd_prev_snap_name(const char *name, const char *start) | |
2067 | { | |
2068 | if (name < start + 2) | |
2069 | return NULL; | |
2070 | ||
2071 | name -= 2; | |
2072 | while (*name) { | |
2073 | if (name == start) | |
2074 | return start; | |
2075 | name--; | |
2076 | } | |
2077 | return name + 1; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * compare the old list of snapshots that we have to what's in the header | |
2082 | * and update it accordingly. Note that the header holds the snapshots | |
2083 | * in a reverse order (from newest to oldest) and we need to go from | |
2084 | * older to new so that we don't get a duplicate snap name when | |
2085 | * doing the process (e.g., removed snapshot and recreated a new | |
2086 | * one with the same name. | |
2087 | */ | |
2088 | static int __rbd_init_snaps_header(struct rbd_device *rbd_dev) | |
2089 | { | |
2090 | const char *name, *first_name; | |
2091 | int i = rbd_dev->header.total_snaps; | |
2092 | struct rbd_snap *snap, *old_snap = NULL; | |
2093 | int ret; | |
2094 | struct list_head *p, *n; | |
2095 | ||
2096 | first_name = rbd_dev->header.snap_names; | |
2097 | name = first_name + rbd_dev->header.snap_names_len; | |
2098 | ||
2099 | list_for_each_prev_safe(p, n, &rbd_dev->snaps) { | |
2100 | u64 cur_id; | |
2101 | ||
2102 | old_snap = list_entry(p, struct rbd_snap, node); | |
2103 | ||
2104 | if (i) | |
2105 | cur_id = rbd_dev->header.snapc->snaps[i - 1]; | |
2106 | ||
2107 | if (!i || old_snap->id < cur_id) { | |
e88a36ec JD |
2108 | /* |
2109 | * old_snap->id was skipped, thus was | |
2110 | * removed. If this rbd_dev is mapped to | |
2111 | * the removed snapshot, record that it no | |
2112 | * longer exists, to prevent further I/O. | |
2113 | */ | |
2114 | if (rbd_dev->snap_id == old_snap->id) | |
2115 | rbd_dev->snap_exists = false; | |
dfc5606d YS |
2116 | __rbd_remove_snap_dev(rbd_dev, old_snap); |
2117 | continue; | |
2118 | } | |
2119 | if (old_snap->id == cur_id) { | |
2120 | /* we have this snapshot already */ | |
2121 | i--; | |
2122 | name = rbd_prev_snap_name(name, first_name); | |
2123 | continue; | |
2124 | } | |
2125 | for (; i > 0; | |
2126 | i--, name = rbd_prev_snap_name(name, first_name)) { | |
2127 | if (!name) { | |
2128 | WARN_ON(1); | |
2129 | return -EINVAL; | |
2130 | } | |
2131 | cur_id = rbd_dev->header.snapc->snaps[i]; | |
2132 | /* snapshot removal? handle it above */ | |
2133 | if (cur_id >= old_snap->id) | |
2134 | break; | |
2135 | /* a new snapshot */ | |
2136 | ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap); | |
2137 | if (ret < 0) | |
2138 | return ret; | |
2139 | ||
2140 | /* note that we add it backward so using n and not p */ | |
2141 | list_add(&snap->node, n); | |
2142 | p = &snap->node; | |
2143 | } | |
2144 | } | |
2145 | /* we're done going over the old snap list, just add what's left */ | |
2146 | for (; i > 0; i--) { | |
2147 | name = rbd_prev_snap_name(name, first_name); | |
2148 | if (!name) { | |
2149 | WARN_ON(1); | |
2150 | return -EINVAL; | |
2151 | } | |
2152 | ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap); | |
2153 | if (ret < 0) | |
2154 | return ret; | |
2155 | list_add(&snap->node, &rbd_dev->snaps); | |
2156 | } | |
2157 | ||
2158 | return 0; | |
2159 | } | |
2160 | ||
dfc5606d YS |
2161 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
2162 | { | |
f0f8cef5 | 2163 | int ret; |
dfc5606d YS |
2164 | struct device *dev; |
2165 | struct rbd_snap *snap; | |
2166 | ||
2167 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2168 | dev = &rbd_dev->dev; | |
2169 | ||
2170 | dev->bus = &rbd_bus_type; | |
2171 | dev->type = &rbd_device_type; | |
2172 | dev->parent = &rbd_root_dev; | |
2173 | dev->release = rbd_dev_release; | |
2174 | dev_set_name(dev, "%d", rbd_dev->id); | |
2175 | ret = device_register(dev); | |
2176 | if (ret < 0) | |
f0f8cef5 | 2177 | goto out; |
dfc5606d YS |
2178 | |
2179 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
2180 | ret = rbd_register_snap_dev(rbd_dev, snap, | |
2181 | &rbd_dev->dev); | |
2182 | if (ret < 0) | |
602adf40 YS |
2183 | break; |
2184 | } | |
f0f8cef5 | 2185 | out: |
dfc5606d YS |
2186 | mutex_unlock(&ctl_mutex); |
2187 | return ret; | |
602adf40 YS |
2188 | } |
2189 | ||
dfc5606d YS |
2190 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
2191 | { | |
2192 | device_unregister(&rbd_dev->dev); | |
2193 | } | |
2194 | ||
59c2be1e YS |
2195 | static int rbd_init_watch_dev(struct rbd_device *rbd_dev) |
2196 | { | |
2197 | int ret, rc; | |
2198 | ||
2199 | do { | |
0bed54dc | 2200 | ret = rbd_req_sync_watch(rbd_dev, rbd_dev->header_name, |
59c2be1e YS |
2201 | rbd_dev->header.obj_version); |
2202 | if (ret == -ERANGE) { | |
2203 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
263c6ca0 | 2204 | rc = __rbd_refresh_header(rbd_dev); |
59c2be1e YS |
2205 | mutex_unlock(&ctl_mutex); |
2206 | if (rc < 0) | |
2207 | return rc; | |
2208 | } | |
2209 | } while (ret == -ERANGE); | |
2210 | ||
2211 | return ret; | |
2212 | } | |
2213 | ||
1ddbe94e AE |
2214 | static atomic64_t rbd_id_max = ATOMIC64_INIT(0); |
2215 | ||
2216 | /* | |
499afd5b AE |
2217 | * Get a unique rbd identifier for the given new rbd_dev, and add |
2218 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 2219 | */ |
499afd5b | 2220 | static void rbd_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 2221 | { |
499afd5b AE |
2222 | rbd_dev->id = atomic64_inc_return(&rbd_id_max); |
2223 | ||
2224 | spin_lock(&rbd_dev_list_lock); | |
2225 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
2226 | spin_unlock(&rbd_dev_list_lock); | |
1ddbe94e | 2227 | } |
b7f23c36 | 2228 | |
1ddbe94e | 2229 | /* |
499afd5b AE |
2230 | * Remove an rbd_dev from the global list, and record that its |
2231 | * identifier is no longer in use. | |
1ddbe94e | 2232 | */ |
499afd5b | 2233 | static void rbd_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 2234 | { |
d184f6bf AE |
2235 | struct list_head *tmp; |
2236 | int rbd_id = rbd_dev->id; | |
2237 | int max_id; | |
2238 | ||
2239 | BUG_ON(rbd_id < 1); | |
499afd5b AE |
2240 | |
2241 | spin_lock(&rbd_dev_list_lock); | |
2242 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
2243 | |
2244 | /* | |
2245 | * If the id being "put" is not the current maximum, there | |
2246 | * is nothing special we need to do. | |
2247 | */ | |
2248 | if (rbd_id != atomic64_read(&rbd_id_max)) { | |
2249 | spin_unlock(&rbd_dev_list_lock); | |
2250 | return; | |
2251 | } | |
2252 | ||
2253 | /* | |
2254 | * We need to update the current maximum id. Search the | |
2255 | * list to find out what it is. We're more likely to find | |
2256 | * the maximum at the end, so search the list backward. | |
2257 | */ | |
2258 | max_id = 0; | |
2259 | list_for_each_prev(tmp, &rbd_dev_list) { | |
2260 | struct rbd_device *rbd_dev; | |
2261 | ||
2262 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
2263 | if (rbd_id > max_id) | |
2264 | max_id = rbd_id; | |
2265 | } | |
499afd5b | 2266 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 2267 | |
1ddbe94e | 2268 | /* |
d184f6bf AE |
2269 | * The max id could have been updated by rbd_id_get(), in |
2270 | * which case it now accurately reflects the new maximum. | |
2271 | * Be careful not to overwrite the maximum value in that | |
2272 | * case. | |
1ddbe94e | 2273 | */ |
d184f6bf | 2274 | atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id); |
b7f23c36 AE |
2275 | } |
2276 | ||
e28fff26 AE |
2277 | /* |
2278 | * Skips over white space at *buf, and updates *buf to point to the | |
2279 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
2280 | * the token (string of non-white space characters) found. Note |
2281 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
2282 | */ |
2283 | static inline size_t next_token(const char **buf) | |
2284 | { | |
2285 | /* | |
2286 | * These are the characters that produce nonzero for | |
2287 | * isspace() in the "C" and "POSIX" locales. | |
2288 | */ | |
2289 | const char *spaces = " \f\n\r\t\v"; | |
2290 | ||
2291 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
2292 | ||
2293 | return strcspn(*buf, spaces); /* Return token length */ | |
2294 | } | |
2295 | ||
2296 | /* | |
2297 | * Finds the next token in *buf, and if the provided token buffer is | |
2298 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
2299 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
2300 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
2301 | * |
2302 | * Returns the length of the token found (not including the '\0'). | |
2303 | * Return value will be 0 if no token is found, and it will be >= | |
2304 | * token_size if the token would not fit. | |
2305 | * | |
593a9e7b | 2306 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
2307 | * found token. Note that this occurs even if the token buffer is |
2308 | * too small to hold it. | |
2309 | */ | |
2310 | static inline size_t copy_token(const char **buf, | |
2311 | char *token, | |
2312 | size_t token_size) | |
2313 | { | |
2314 | size_t len; | |
2315 | ||
2316 | len = next_token(buf); | |
2317 | if (len < token_size) { | |
2318 | memcpy(token, *buf, len); | |
2319 | *(token + len) = '\0'; | |
2320 | } | |
2321 | *buf += len; | |
2322 | ||
2323 | return len; | |
2324 | } | |
2325 | ||
ea3352f4 AE |
2326 | /* |
2327 | * Finds the next token in *buf, dynamically allocates a buffer big | |
2328 | * enough to hold a copy of it, and copies the token into the new | |
2329 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
2330 | * that a duplicate buffer is created even for a zero-length token. | |
2331 | * | |
2332 | * Returns a pointer to the newly-allocated duplicate, or a null | |
2333 | * pointer if memory for the duplicate was not available. If | |
2334 | * the lenp argument is a non-null pointer, the length of the token | |
2335 | * (not including the '\0') is returned in *lenp. | |
2336 | * | |
2337 | * If successful, the *buf pointer will be updated to point beyond | |
2338 | * the end of the found token. | |
2339 | * | |
2340 | * Note: uses GFP_KERNEL for allocation. | |
2341 | */ | |
2342 | static inline char *dup_token(const char **buf, size_t *lenp) | |
2343 | { | |
2344 | char *dup; | |
2345 | size_t len; | |
2346 | ||
2347 | len = next_token(buf); | |
2348 | dup = kmalloc(len + 1, GFP_KERNEL); | |
2349 | if (!dup) | |
2350 | return NULL; | |
2351 | ||
2352 | memcpy(dup, *buf, len); | |
2353 | *(dup + len) = '\0'; | |
2354 | *buf += len; | |
2355 | ||
2356 | if (lenp) | |
2357 | *lenp = len; | |
2358 | ||
2359 | return dup; | |
2360 | } | |
2361 | ||
a725f65e | 2362 | /* |
0bed54dc | 2363 | * This fills in the pool_name, image_name, image_name_len, snap_name, |
a725f65e AE |
2364 | * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based |
2365 | * on the list of monitor addresses and other options provided via | |
2366 | * /sys/bus/rbd/add. | |
d22f76e7 AE |
2367 | * |
2368 | * Note: rbd_dev is assumed to have been initially zero-filled. | |
a725f65e AE |
2369 | */ |
2370 | static int rbd_add_parse_args(struct rbd_device *rbd_dev, | |
2371 | const char *buf, | |
7ef3214a | 2372 | const char **mon_addrs, |
5214ecc4 | 2373 | size_t *mon_addrs_size, |
e28fff26 | 2374 | char *options, |
0bed54dc | 2375 | size_t options_size) |
e28fff26 | 2376 | { |
d22f76e7 AE |
2377 | size_t len; |
2378 | int ret; | |
e28fff26 AE |
2379 | |
2380 | /* The first four tokens are required */ | |
2381 | ||
7ef3214a AE |
2382 | len = next_token(&buf); |
2383 | if (!len) | |
a725f65e | 2384 | return -EINVAL; |
5214ecc4 | 2385 | *mon_addrs_size = len + 1; |
7ef3214a AE |
2386 | *mon_addrs = buf; |
2387 | ||
2388 | buf += len; | |
a725f65e | 2389 | |
e28fff26 AE |
2390 | len = copy_token(&buf, options, options_size); |
2391 | if (!len || len >= options_size) | |
2392 | return -EINVAL; | |
2393 | ||
bf3e5ae1 | 2394 | ret = -ENOMEM; |
d22f76e7 AE |
2395 | rbd_dev->pool_name = dup_token(&buf, NULL); |
2396 | if (!rbd_dev->pool_name) | |
d22f76e7 | 2397 | goto out_err; |
e28fff26 | 2398 | |
0bed54dc AE |
2399 | rbd_dev->image_name = dup_token(&buf, &rbd_dev->image_name_len); |
2400 | if (!rbd_dev->image_name) | |
bf3e5ae1 | 2401 | goto out_err; |
a725f65e | 2402 | |
cb8627c7 AE |
2403 | /* Create the name of the header object */ |
2404 | ||
0bed54dc | 2405 | rbd_dev->header_name = kmalloc(rbd_dev->image_name_len |
bf3e5ae1 AE |
2406 | + sizeof (RBD_SUFFIX), |
2407 | GFP_KERNEL); | |
0bed54dc | 2408 | if (!rbd_dev->header_name) |
cb8627c7 | 2409 | goto out_err; |
0bed54dc | 2410 | sprintf(rbd_dev->header_name, "%s%s", rbd_dev->image_name, RBD_SUFFIX); |
a725f65e | 2411 | |
e28fff26 | 2412 | /* |
820a5f3e AE |
2413 | * The snapshot name is optional. If none is is supplied, |
2414 | * we use the default value. | |
e28fff26 | 2415 | */ |
820a5f3e AE |
2416 | rbd_dev->snap_name = dup_token(&buf, &len); |
2417 | if (!rbd_dev->snap_name) | |
2418 | goto out_err; | |
2419 | if (!len) { | |
2420 | /* Replace the empty name with the default */ | |
2421 | kfree(rbd_dev->snap_name); | |
2422 | rbd_dev->snap_name | |
2423 | = kmalloc(sizeof (RBD_SNAP_HEAD_NAME), GFP_KERNEL); | |
2424 | if (!rbd_dev->snap_name) | |
2425 | goto out_err; | |
2426 | ||
e28fff26 AE |
2427 | memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME, |
2428 | sizeof (RBD_SNAP_HEAD_NAME)); | |
849b4260 | 2429 | } |
e28fff26 | 2430 | |
a725f65e | 2431 | return 0; |
d22f76e7 AE |
2432 | |
2433 | out_err: | |
0bed54dc AE |
2434 | kfree(rbd_dev->header_name); |
2435 | kfree(rbd_dev->image_name); | |
d22f76e7 AE |
2436 | kfree(rbd_dev->pool_name); |
2437 | rbd_dev->pool_name = NULL; | |
2438 | ||
2439 | return ret; | |
a725f65e AE |
2440 | } |
2441 | ||
59c2be1e YS |
2442 | static ssize_t rbd_add(struct bus_type *bus, |
2443 | const char *buf, | |
2444 | size_t count) | |
602adf40 | 2445 | { |
cb8627c7 AE |
2446 | char *options; |
2447 | struct rbd_device *rbd_dev = NULL; | |
7ef3214a AE |
2448 | const char *mon_addrs = NULL; |
2449 | size_t mon_addrs_size = 0; | |
27cc2594 AE |
2450 | struct ceph_osd_client *osdc; |
2451 | int rc = -ENOMEM; | |
602adf40 YS |
2452 | |
2453 | if (!try_module_get(THIS_MODULE)) | |
2454 | return -ENODEV; | |
2455 | ||
60571c7d | 2456 | options = kmalloc(count, GFP_KERNEL); |
602adf40 | 2457 | if (!options) |
27cc2594 | 2458 | goto err_nomem; |
cb8627c7 AE |
2459 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
2460 | if (!rbd_dev) | |
2461 | goto err_nomem; | |
602adf40 YS |
2462 | |
2463 | /* static rbd_device initialization */ | |
2464 | spin_lock_init(&rbd_dev->lock); | |
2465 | INIT_LIST_HEAD(&rbd_dev->node); | |
dfc5606d | 2466 | INIT_LIST_HEAD(&rbd_dev->snaps); |
c666601a | 2467 | init_rwsem(&rbd_dev->header_rwsem); |
602adf40 | 2468 | |
d184f6bf | 2469 | /* generate unique id: find highest unique id, add one */ |
499afd5b | 2470 | rbd_id_get(rbd_dev); |
602adf40 | 2471 | |
a725f65e | 2472 | /* Fill in the device name, now that we have its id. */ |
81a89793 AE |
2473 | BUILD_BUG_ON(DEV_NAME_LEN |
2474 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
2475 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id); | |
a725f65e | 2476 | |
602adf40 | 2477 | /* parse add command */ |
7ef3214a | 2478 | rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size, |
e28fff26 | 2479 | options, count); |
a725f65e | 2480 | if (rc) |
f0f8cef5 | 2481 | goto err_put_id; |
e124a82f | 2482 | |
5214ecc4 AE |
2483 | rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1, |
2484 | options); | |
d720bcb0 AE |
2485 | if (IS_ERR(rbd_dev->rbd_client)) { |
2486 | rc = PTR_ERR(rbd_dev->rbd_client); | |
f0f8cef5 | 2487 | goto err_put_id; |
d720bcb0 | 2488 | } |
602adf40 | 2489 | |
602adf40 | 2490 | /* pick the pool */ |
1dbb4399 | 2491 | osdc = &rbd_dev->rbd_client->client->osdc; |
602adf40 YS |
2492 | rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name); |
2493 | if (rc < 0) | |
2494 | goto err_out_client; | |
9bb2f334 | 2495 | rbd_dev->pool_id = rc; |
602adf40 YS |
2496 | |
2497 | /* register our block device */ | |
27cc2594 AE |
2498 | rc = register_blkdev(0, rbd_dev->name); |
2499 | if (rc < 0) | |
602adf40 | 2500 | goto err_out_client; |
27cc2594 | 2501 | rbd_dev->major = rc; |
602adf40 | 2502 | |
dfc5606d YS |
2503 | rc = rbd_bus_add_dev(rbd_dev); |
2504 | if (rc) | |
766fc439 YS |
2505 | goto err_out_blkdev; |
2506 | ||
32eec68d AE |
2507 | /* |
2508 | * At this point cleanup in the event of an error is the job | |
2509 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
2510 | * | |
2511 | * Set up and announce blkdev mapping. | |
2512 | */ | |
602adf40 YS |
2513 | rc = rbd_init_disk(rbd_dev); |
2514 | if (rc) | |
766fc439 | 2515 | goto err_out_bus; |
602adf40 | 2516 | |
59c2be1e YS |
2517 | rc = rbd_init_watch_dev(rbd_dev); |
2518 | if (rc) | |
2519 | goto err_out_bus; | |
2520 | ||
602adf40 YS |
2521 | return count; |
2522 | ||
766fc439 | 2523 | err_out_bus: |
766fc439 YS |
2524 | /* this will also clean up rest of rbd_dev stuff */ |
2525 | ||
2526 | rbd_bus_del_dev(rbd_dev); | |
2527 | kfree(options); | |
766fc439 YS |
2528 | return rc; |
2529 | ||
602adf40 YS |
2530 | err_out_blkdev: |
2531 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
2532 | err_out_client: | |
2533 | rbd_put_client(rbd_dev); | |
f0f8cef5 | 2534 | err_put_id: |
cb8627c7 | 2535 | if (rbd_dev->pool_name) { |
820a5f3e | 2536 | kfree(rbd_dev->snap_name); |
0bed54dc AE |
2537 | kfree(rbd_dev->header_name); |
2538 | kfree(rbd_dev->image_name); | |
cb8627c7 AE |
2539 | kfree(rbd_dev->pool_name); |
2540 | } | |
499afd5b | 2541 | rbd_id_put(rbd_dev); |
27cc2594 | 2542 | err_nomem: |
27cc2594 | 2543 | kfree(rbd_dev); |
cb8627c7 | 2544 | kfree(options); |
27cc2594 | 2545 | |
602adf40 YS |
2546 | dout("Error adding device %s\n", buf); |
2547 | module_put(THIS_MODULE); | |
27cc2594 AE |
2548 | |
2549 | return (ssize_t) rc; | |
602adf40 YS |
2550 | } |
2551 | ||
2552 | static struct rbd_device *__rbd_get_dev(unsigned long id) | |
2553 | { | |
2554 | struct list_head *tmp; | |
2555 | struct rbd_device *rbd_dev; | |
2556 | ||
e124a82f | 2557 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
2558 | list_for_each(tmp, &rbd_dev_list) { |
2559 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
e124a82f AE |
2560 | if (rbd_dev->id == id) { |
2561 | spin_unlock(&rbd_dev_list_lock); | |
602adf40 | 2562 | return rbd_dev; |
e124a82f | 2563 | } |
602adf40 | 2564 | } |
e124a82f | 2565 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
2566 | return NULL; |
2567 | } | |
2568 | ||
dfc5606d | 2569 | static void rbd_dev_release(struct device *dev) |
602adf40 | 2570 | { |
593a9e7b | 2571 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2572 | |
1dbb4399 AE |
2573 | if (rbd_dev->watch_request) { |
2574 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
2575 | ||
2576 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 2577 | rbd_dev->watch_request); |
1dbb4399 | 2578 | } |
59c2be1e | 2579 | if (rbd_dev->watch_event) |
0bed54dc | 2580 | rbd_req_sync_unwatch(rbd_dev, rbd_dev->header_name); |
59c2be1e | 2581 | |
602adf40 YS |
2582 | rbd_put_client(rbd_dev); |
2583 | ||
2584 | /* clean up and free blkdev */ | |
2585 | rbd_free_disk(rbd_dev); | |
2586 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d AE |
2587 | |
2588 | /* done with the id, and with the rbd_dev */ | |
820a5f3e | 2589 | kfree(rbd_dev->snap_name); |
0bed54dc | 2590 | kfree(rbd_dev->header_name); |
d22f76e7 | 2591 | kfree(rbd_dev->pool_name); |
0bed54dc | 2592 | kfree(rbd_dev->image_name); |
32eec68d | 2593 | rbd_id_put(rbd_dev); |
602adf40 YS |
2594 | kfree(rbd_dev); |
2595 | ||
2596 | /* release module ref */ | |
2597 | module_put(THIS_MODULE); | |
602adf40 YS |
2598 | } |
2599 | ||
dfc5606d YS |
2600 | static ssize_t rbd_remove(struct bus_type *bus, |
2601 | const char *buf, | |
2602 | size_t count) | |
602adf40 YS |
2603 | { |
2604 | struct rbd_device *rbd_dev = NULL; | |
2605 | int target_id, rc; | |
2606 | unsigned long ul; | |
2607 | int ret = count; | |
2608 | ||
2609 | rc = strict_strtoul(buf, 10, &ul); | |
2610 | if (rc) | |
2611 | return rc; | |
2612 | ||
2613 | /* convert to int; abort if we lost anything in the conversion */ | |
2614 | target_id = (int) ul; | |
2615 | if (target_id != ul) | |
2616 | return -EINVAL; | |
2617 | ||
2618 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2619 | ||
2620 | rbd_dev = __rbd_get_dev(target_id); | |
2621 | if (!rbd_dev) { | |
2622 | ret = -ENOENT; | |
2623 | goto done; | |
2624 | } | |
2625 | ||
dfc5606d YS |
2626 | __rbd_remove_all_snaps(rbd_dev); |
2627 | rbd_bus_del_dev(rbd_dev); | |
602adf40 YS |
2628 | |
2629 | done: | |
2630 | mutex_unlock(&ctl_mutex); | |
2631 | return ret; | |
2632 | } | |
2633 | ||
dfc5606d YS |
2634 | static ssize_t rbd_snap_add(struct device *dev, |
2635 | struct device_attribute *attr, | |
2636 | const char *buf, | |
2637 | size_t count) | |
602adf40 | 2638 | { |
593a9e7b | 2639 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
2640 | int ret; |
2641 | char *name = kmalloc(count + 1, GFP_KERNEL); | |
602adf40 YS |
2642 | if (!name) |
2643 | return -ENOMEM; | |
2644 | ||
dfc5606d | 2645 | snprintf(name, count, "%s", buf); |
602adf40 YS |
2646 | |
2647 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2648 | ||
602adf40 YS |
2649 | ret = rbd_header_add_snap(rbd_dev, |
2650 | name, GFP_KERNEL); | |
2651 | if (ret < 0) | |
59c2be1e | 2652 | goto err_unlock; |
602adf40 | 2653 | |
263c6ca0 | 2654 | ret = __rbd_refresh_header(rbd_dev); |
602adf40 | 2655 | if (ret < 0) |
59c2be1e YS |
2656 | goto err_unlock; |
2657 | ||
2658 | /* shouldn't hold ctl_mutex when notifying.. notify might | |
2659 | trigger a watch callback that would need to get that mutex */ | |
2660 | mutex_unlock(&ctl_mutex); | |
2661 | ||
2662 | /* make a best effort, don't error if failed */ | |
0bed54dc | 2663 | rbd_req_sync_notify(rbd_dev, rbd_dev->header_name); |
602adf40 YS |
2664 | |
2665 | ret = count; | |
59c2be1e YS |
2666 | kfree(name); |
2667 | return ret; | |
2668 | ||
2669 | err_unlock: | |
602adf40 | 2670 | mutex_unlock(&ctl_mutex); |
602adf40 YS |
2671 | kfree(name); |
2672 | return ret; | |
2673 | } | |
2674 | ||
602adf40 YS |
2675 | /* |
2676 | * create control files in sysfs | |
dfc5606d | 2677 | * /sys/bus/rbd/... |
602adf40 YS |
2678 | */ |
2679 | static int rbd_sysfs_init(void) | |
2680 | { | |
dfc5606d | 2681 | int ret; |
602adf40 | 2682 | |
fed4c143 | 2683 | ret = device_register(&rbd_root_dev); |
21079786 | 2684 | if (ret < 0) |
dfc5606d | 2685 | return ret; |
602adf40 | 2686 | |
fed4c143 AE |
2687 | ret = bus_register(&rbd_bus_type); |
2688 | if (ret < 0) | |
2689 | device_unregister(&rbd_root_dev); | |
602adf40 | 2690 | |
602adf40 YS |
2691 | return ret; |
2692 | } | |
2693 | ||
2694 | static void rbd_sysfs_cleanup(void) | |
2695 | { | |
dfc5606d | 2696 | bus_unregister(&rbd_bus_type); |
fed4c143 | 2697 | device_unregister(&rbd_root_dev); |
602adf40 YS |
2698 | } |
2699 | ||
2700 | int __init rbd_init(void) | |
2701 | { | |
2702 | int rc; | |
2703 | ||
2704 | rc = rbd_sysfs_init(); | |
2705 | if (rc) | |
2706 | return rc; | |
f0f8cef5 | 2707 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
2708 | return 0; |
2709 | } | |
2710 | ||
2711 | void __exit rbd_exit(void) | |
2712 | { | |
2713 | rbd_sysfs_cleanup(); | |
2714 | } | |
2715 | ||
2716 | module_init(rbd_init); | |
2717 | module_exit(rbd_exit); | |
2718 | ||
2719 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
2720 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
2721 | MODULE_DESCRIPTION("rados block device"); | |
2722 | ||
2723 | /* following authorship retained from original osdblk.c */ | |
2724 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
2725 | ||
2726 | MODULE_LICENSE("GPL"); |