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 | ||
21079786 | 58 | #define RBD_MAX_MD_NAME_LEN (RBD_MAX_OBJ_NAME_LEN + sizeof(RBD_SUFFIX)) |
602adf40 YS |
59 | #define RBD_MAX_POOL_NAME_LEN 64 |
60 | #define RBD_MAX_SNAP_NAME_LEN 32 | |
61 | #define RBD_MAX_OPT_LEN 1024 | |
62 | ||
63 | #define RBD_SNAP_HEAD_NAME "-" | |
64 | ||
81a89793 AE |
65 | /* |
66 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
67 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
68 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
69 | * enough to hold all possible device names. | |
70 | */ | |
602adf40 | 71 | #define DEV_NAME_LEN 32 |
81a89793 | 72 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 | 73 | |
59c2be1e YS |
74 | #define RBD_NOTIFY_TIMEOUT_DEFAULT 10 |
75 | ||
602adf40 YS |
76 | /* |
77 | * block device image metadata (in-memory version) | |
78 | */ | |
79 | struct rbd_image_header { | |
80 | u64 image_size; | |
81 | char block_name[32]; | |
82 | __u8 obj_order; | |
83 | __u8 crypt_type; | |
84 | __u8 comp_type; | |
602adf40 YS |
85 | struct ceph_snap_context *snapc; |
86 | size_t snap_names_len; | |
87 | u64 snap_seq; | |
88 | u32 total_snaps; | |
89 | ||
90 | char *snap_names; | |
91 | u64 *snap_sizes; | |
59c2be1e YS |
92 | |
93 | u64 obj_version; | |
94 | }; | |
95 | ||
96 | struct rbd_options { | |
97 | int notify_timeout; | |
602adf40 YS |
98 | }; |
99 | ||
100 | /* | |
f0f8cef5 | 101 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
102 | */ |
103 | struct rbd_client { | |
104 | struct ceph_client *client; | |
59c2be1e | 105 | struct rbd_options *rbd_opts; |
602adf40 YS |
106 | struct kref kref; |
107 | struct list_head node; | |
108 | }; | |
109 | ||
110 | /* | |
f0f8cef5 | 111 | * a request completion status |
602adf40 | 112 | */ |
1fec7093 YS |
113 | struct rbd_req_status { |
114 | int done; | |
115 | int rc; | |
116 | u64 bytes; | |
117 | }; | |
118 | ||
119 | /* | |
120 | * a collection of requests | |
121 | */ | |
122 | struct rbd_req_coll { | |
123 | int total; | |
124 | int num_done; | |
125 | struct kref kref; | |
126 | struct rbd_req_status status[0]; | |
602adf40 YS |
127 | }; |
128 | ||
f0f8cef5 AE |
129 | /* |
130 | * a single io request | |
131 | */ | |
132 | struct rbd_request { | |
133 | struct request *rq; /* blk layer request */ | |
134 | struct bio *bio; /* cloned bio */ | |
135 | struct page **pages; /* list of used pages */ | |
136 | u64 len; | |
137 | int coll_index; | |
138 | struct rbd_req_coll *coll; | |
139 | }; | |
140 | ||
dfc5606d YS |
141 | struct rbd_snap { |
142 | struct device dev; | |
143 | const char *name; | |
144 | size_t size; | |
145 | struct list_head node; | |
146 | u64 id; | |
147 | }; | |
148 | ||
602adf40 YS |
149 | /* |
150 | * a single device | |
151 | */ | |
152 | struct rbd_device { | |
153 | int id; /* blkdev unique id */ | |
154 | ||
155 | int major; /* blkdev assigned major */ | |
156 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
157 | struct request_queue *q; | |
158 | ||
602adf40 YS |
159 | struct rbd_client *rbd_client; |
160 | ||
161 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
162 | ||
163 | spinlock_t lock; /* queue lock */ | |
164 | ||
165 | struct rbd_image_header header; | |
166 | char obj[RBD_MAX_OBJ_NAME_LEN]; /* rbd image name */ | |
167 | int obj_len; | |
168 | char obj_md_name[RBD_MAX_MD_NAME_LEN]; /* hdr nm. */ | |
169 | char pool_name[RBD_MAX_POOL_NAME_LEN]; | |
170 | int poolid; | |
171 | ||
59c2be1e YS |
172 | struct ceph_osd_event *watch_event; |
173 | struct ceph_osd_request *watch_request; | |
174 | ||
c666601a JD |
175 | /* protects updating the header */ |
176 | struct rw_semaphore header_rwsem; | |
602adf40 YS |
177 | char snap_name[RBD_MAX_SNAP_NAME_LEN]; |
178 | u32 cur_snap; /* index+1 of current snapshot within snap context | |
179 | 0 - for the head */ | |
180 | int read_only; | |
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 | |
59c2be1e YS |
244 | static int __rbd_update_snaps(struct rbd_device *rbd_dev); |
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. | |
277 | * We own *opt. | |
278 | */ | |
59c2be1e YS |
279 | static struct rbd_client *rbd_client_create(struct ceph_options *opt, |
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 | ||
6ab00d46 | 295 | rbdc->client = ceph_create_client(opt, rbdc, 0, 0); |
602adf40 | 296 | if (IS_ERR(rbdc->client)) |
bc534d86 | 297 | goto out_mutex; |
28f259b7 | 298 | opt = NULL; /* Now rbdc->client is responsible for opt */ |
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: | |
28f259b7 VK |
321 | if (opt) |
322 | ceph_destroy_options(opt); | |
323 | return ERR_PTR(ret); | |
602adf40 YS |
324 | } |
325 | ||
326 | /* | |
327 | * Find a ceph client with specific addr and configuration. | |
328 | */ | |
329 | static struct rbd_client *__rbd_client_find(struct ceph_options *opt) | |
330 | { | |
331 | struct rbd_client *client_node; | |
332 | ||
333 | if (opt->flags & CEPH_OPT_NOSHARE) | |
334 | return NULL; | |
335 | ||
336 | list_for_each_entry(client_node, &rbd_client_list, node) | |
337 | if (ceph_compare_options(opt, client_node->client) == 0) | |
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 | ||
353 | static match_table_t rbdopt_tokens = { | |
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 | { | |
362 | struct rbd_options *rbdopt = private; | |
363 | substring_t argstr[MAX_OPT_ARGS]; | |
364 | int token, intval, ret; | |
365 | ||
21079786 | 366 | token = match_token(c, rbdopt_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: | |
387 | rbdopt->notify_timeout = intval; | |
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; | |
404 | struct ceph_options *opt; | |
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 | |
ee57741c | 413 | opt = ceph_parse_options(options, mon_addr, |
5214ecc4 | 414 | mon_addr + mon_addr_len, |
21079786 | 415 | parse_rbd_opts_token, rbd_opts); |
ee57741c | 416 | if (IS_ERR(opt)) { |
d720bcb0 AE |
417 | kfree(rbd_opts); |
418 | return ERR_CAST(opt); | |
ee57741c | 419 | } |
602adf40 | 420 | |
432b8587 | 421 | spin_lock(&rbd_client_list_lock); |
602adf40 YS |
422 | rbdc = __rbd_client_find(opt); |
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 | |
e6994d3d AE |
428 | ceph_destroy_options(opt); |
429 | kfree(rbd_opts); | |
430 | ||
d720bcb0 | 431 | return rbdc; |
602adf40 | 432 | } |
432b8587 | 433 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 434 | |
59c2be1e | 435 | rbdc = rbd_client_create(opt, 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 YS |
483 | |
484 | /* | |
485 | * Create a new header structure, translate header format from the on-disk | |
486 | * header. | |
487 | */ | |
488 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
489 | struct rbd_image_header_ondisk *ondisk, | |
490 | int allocated_snaps, | |
491 | gfp_t gfp_flags) | |
492 | { | |
493 | int i; | |
00f1f36f | 494 | u32 snap_count; |
602adf40 | 495 | |
21079786 | 496 | if (memcmp(ondisk, RBD_HEADER_TEXT, sizeof(RBD_HEADER_TEXT))) |
81e759fb | 497 | return -ENXIO; |
81e759fb | 498 | |
00f1f36f | 499 | snap_count = le32_to_cpu(ondisk->snap_count); |
602adf40 | 500 | header->snapc = kmalloc(sizeof(struct ceph_snap_context) + |
21079786 | 501 | snap_count * sizeof (*ondisk), |
602adf40 YS |
502 | gfp_flags); |
503 | if (!header->snapc) | |
504 | return -ENOMEM; | |
00f1f36f | 505 | |
00f1f36f | 506 | header->snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
602adf40 YS |
507 | if (snap_count) { |
508 | header->snap_names = kmalloc(header->snap_names_len, | |
509 | GFP_KERNEL); | |
510 | if (!header->snap_names) | |
511 | goto err_snapc; | |
512 | header->snap_sizes = kmalloc(snap_count * sizeof(u64), | |
513 | GFP_KERNEL); | |
514 | if (!header->snap_sizes) | |
515 | goto err_names; | |
516 | } else { | |
517 | header->snap_names = NULL; | |
518 | header->snap_sizes = NULL; | |
519 | } | |
520 | memcpy(header->block_name, ondisk->block_name, | |
521 | sizeof(ondisk->block_name)); | |
522 | ||
523 | header->image_size = le64_to_cpu(ondisk->image_size); | |
524 | header->obj_order = ondisk->options.order; | |
525 | header->crypt_type = ondisk->options.crypt_type; | |
526 | header->comp_type = ondisk->options.comp_type; | |
527 | ||
528 | atomic_set(&header->snapc->nref, 1); | |
529 | header->snap_seq = le64_to_cpu(ondisk->snap_seq); | |
530 | header->snapc->num_snaps = snap_count; | |
531 | header->total_snaps = snap_count; | |
532 | ||
21079786 | 533 | if (snap_count && allocated_snaps == snap_count) { |
602adf40 YS |
534 | for (i = 0; i < snap_count; i++) { |
535 | header->snapc->snaps[i] = | |
536 | le64_to_cpu(ondisk->snaps[i].id); | |
537 | header->snap_sizes[i] = | |
538 | le64_to_cpu(ondisk->snaps[i].image_size); | |
539 | } | |
540 | ||
541 | /* copy snapshot names */ | |
542 | memcpy(header->snap_names, &ondisk->snaps[i], | |
543 | header->snap_names_len); | |
544 | } | |
545 | ||
546 | return 0; | |
547 | ||
548 | err_names: | |
549 | kfree(header->snap_names); | |
550 | err_snapc: | |
551 | kfree(header->snapc); | |
00f1f36f | 552 | return -ENOMEM; |
602adf40 YS |
553 | } |
554 | ||
555 | static int snap_index(struct rbd_image_header *header, int snap_num) | |
556 | { | |
557 | return header->total_snaps - snap_num; | |
558 | } | |
559 | ||
560 | static u64 cur_snap_id(struct rbd_device *rbd_dev) | |
561 | { | |
562 | struct rbd_image_header *header = &rbd_dev->header; | |
563 | ||
564 | if (!rbd_dev->cur_snap) | |
565 | return 0; | |
566 | ||
567 | return header->snapc->snaps[snap_index(header, rbd_dev->cur_snap)]; | |
568 | } | |
569 | ||
570 | static int snap_by_name(struct rbd_image_header *header, const char *snap_name, | |
571 | u64 *seq, u64 *size) | |
572 | { | |
573 | int i; | |
574 | char *p = header->snap_names; | |
575 | ||
00f1f36f AE |
576 | for (i = 0; i < header->total_snaps; i++) { |
577 | if (!strcmp(snap_name, p)) { | |
602adf40 | 578 | |
00f1f36f | 579 | /* Found it. Pass back its id and/or size */ |
602adf40 | 580 | |
00f1f36f AE |
581 | if (seq) |
582 | *seq = header->snapc->snaps[i]; | |
583 | if (size) | |
584 | *size = header->snap_sizes[i]; | |
585 | return i; | |
586 | } | |
587 | p += strlen(p) + 1; /* Skip ahead to the next name */ | |
588 | } | |
589 | return -ENOENT; | |
602adf40 YS |
590 | } |
591 | ||
cc9d734c | 592 | static int rbd_header_set_snap(struct rbd_device *dev, u64 *size) |
602adf40 YS |
593 | { |
594 | struct rbd_image_header *header = &dev->header; | |
595 | struct ceph_snap_context *snapc = header->snapc; | |
596 | int ret = -ENOENT; | |
597 | ||
cc9d734c JD |
598 | BUILD_BUG_ON(sizeof (dev->snap_name) < sizeof (RBD_SNAP_HEAD_NAME)); |
599 | ||
c666601a | 600 | down_write(&dev->header_rwsem); |
602adf40 | 601 | |
cc9d734c JD |
602 | if (!memcmp(dev->snap_name, RBD_SNAP_HEAD_NAME, |
603 | sizeof (RBD_SNAP_HEAD_NAME))) { | |
602adf40 YS |
604 | if (header->total_snaps) |
605 | snapc->seq = header->snap_seq; | |
606 | else | |
607 | snapc->seq = 0; | |
608 | dev->cur_snap = 0; | |
609 | dev->read_only = 0; | |
610 | if (size) | |
611 | *size = header->image_size; | |
612 | } else { | |
cc9d734c | 613 | ret = snap_by_name(header, dev->snap_name, &snapc->seq, size); |
602adf40 YS |
614 | if (ret < 0) |
615 | goto done; | |
616 | ||
617 | dev->cur_snap = header->total_snaps - ret; | |
618 | dev->read_only = 1; | |
619 | } | |
620 | ||
621 | ret = 0; | |
622 | done: | |
c666601a | 623 | up_write(&dev->header_rwsem); |
602adf40 YS |
624 | return ret; |
625 | } | |
626 | ||
627 | static void rbd_header_free(struct rbd_image_header *header) | |
628 | { | |
629 | kfree(header->snapc); | |
630 | kfree(header->snap_names); | |
631 | kfree(header->snap_sizes); | |
632 | } | |
633 | ||
634 | /* | |
635 | * get the actual striped segment name, offset and length | |
636 | */ | |
637 | static u64 rbd_get_segment(struct rbd_image_header *header, | |
638 | const char *block_name, | |
639 | u64 ofs, u64 len, | |
640 | char *seg_name, u64 *segofs) | |
641 | { | |
642 | u64 seg = ofs >> header->obj_order; | |
643 | ||
644 | if (seg_name) | |
645 | snprintf(seg_name, RBD_MAX_SEG_NAME_LEN, | |
646 | "%s.%012llx", block_name, seg); | |
647 | ||
648 | ofs = ofs & ((1 << header->obj_order) - 1); | |
649 | len = min_t(u64, len, (1 << header->obj_order) - ofs); | |
650 | ||
651 | if (segofs) | |
652 | *segofs = ofs; | |
653 | ||
654 | return len; | |
655 | } | |
656 | ||
1fec7093 YS |
657 | static int rbd_get_num_segments(struct rbd_image_header *header, |
658 | u64 ofs, u64 len) | |
659 | { | |
660 | u64 start_seg = ofs >> header->obj_order; | |
661 | u64 end_seg = (ofs + len - 1) >> header->obj_order; | |
662 | return end_seg - start_seg + 1; | |
663 | } | |
664 | ||
029bcbd8 JD |
665 | /* |
666 | * returns the size of an object in the image | |
667 | */ | |
668 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
669 | { | |
670 | return 1 << header->obj_order; | |
671 | } | |
672 | ||
602adf40 YS |
673 | /* |
674 | * bio helpers | |
675 | */ | |
676 | ||
677 | static void bio_chain_put(struct bio *chain) | |
678 | { | |
679 | struct bio *tmp; | |
680 | ||
681 | while (chain) { | |
682 | tmp = chain; | |
683 | chain = chain->bi_next; | |
684 | bio_put(tmp); | |
685 | } | |
686 | } | |
687 | ||
688 | /* | |
689 | * zeros a bio chain, starting at specific offset | |
690 | */ | |
691 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
692 | { | |
693 | struct bio_vec *bv; | |
694 | unsigned long flags; | |
695 | void *buf; | |
696 | int i; | |
697 | int pos = 0; | |
698 | ||
699 | while (chain) { | |
700 | bio_for_each_segment(bv, chain, i) { | |
701 | if (pos + bv->bv_len > start_ofs) { | |
702 | int remainder = max(start_ofs - pos, 0); | |
703 | buf = bvec_kmap_irq(bv, &flags); | |
704 | memset(buf + remainder, 0, | |
705 | bv->bv_len - remainder); | |
85b5aaa6 | 706 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
707 | } |
708 | pos += bv->bv_len; | |
709 | } | |
710 | ||
711 | chain = chain->bi_next; | |
712 | } | |
713 | } | |
714 | ||
715 | /* | |
716 | * bio_chain_clone - clone a chain of bios up to a certain length. | |
717 | * might return a bio_pair that will need to be released. | |
718 | */ | |
719 | static struct bio *bio_chain_clone(struct bio **old, struct bio **next, | |
720 | struct bio_pair **bp, | |
721 | int len, gfp_t gfpmask) | |
722 | { | |
723 | struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL; | |
724 | int total = 0; | |
725 | ||
726 | if (*bp) { | |
727 | bio_pair_release(*bp); | |
728 | *bp = NULL; | |
729 | } | |
730 | ||
731 | while (old_chain && (total < len)) { | |
732 | tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs); | |
733 | if (!tmp) | |
734 | goto err_out; | |
735 | ||
736 | if (total + old_chain->bi_size > len) { | |
737 | struct bio_pair *bp; | |
738 | ||
739 | /* | |
740 | * this split can only happen with a single paged bio, | |
741 | * split_bio will BUG_ON if this is not the case | |
742 | */ | |
743 | dout("bio_chain_clone split! total=%d remaining=%d" | |
744 | "bi_size=%d\n", | |
745 | (int)total, (int)len-total, | |
746 | (int)old_chain->bi_size); | |
747 | ||
748 | /* split the bio. We'll release it either in the next | |
749 | call, or it will have to be released outside */ | |
593a9e7b | 750 | bp = bio_split(old_chain, (len - total) / SECTOR_SIZE); |
602adf40 YS |
751 | if (!bp) |
752 | goto err_out; | |
753 | ||
754 | __bio_clone(tmp, &bp->bio1); | |
755 | ||
756 | *next = &bp->bio2; | |
757 | } else { | |
758 | __bio_clone(tmp, old_chain); | |
759 | *next = old_chain->bi_next; | |
760 | } | |
761 | ||
762 | tmp->bi_bdev = NULL; | |
763 | gfpmask &= ~__GFP_WAIT; | |
764 | tmp->bi_next = NULL; | |
765 | ||
766 | if (!new_chain) { | |
767 | new_chain = tail = tmp; | |
768 | } else { | |
769 | tail->bi_next = tmp; | |
770 | tail = tmp; | |
771 | } | |
772 | old_chain = old_chain->bi_next; | |
773 | ||
774 | total += tmp->bi_size; | |
775 | } | |
776 | ||
777 | BUG_ON(total < len); | |
778 | ||
779 | if (tail) | |
780 | tail->bi_next = NULL; | |
781 | ||
782 | *old = old_chain; | |
783 | ||
784 | return new_chain; | |
785 | ||
786 | err_out: | |
787 | dout("bio_chain_clone with err\n"); | |
788 | bio_chain_put(new_chain); | |
789 | return NULL; | |
790 | } | |
791 | ||
792 | /* | |
793 | * helpers for osd request op vectors. | |
794 | */ | |
795 | static int rbd_create_rw_ops(struct ceph_osd_req_op **ops, | |
796 | int num_ops, | |
797 | int opcode, | |
798 | u32 payload_len) | |
799 | { | |
800 | *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1), | |
801 | GFP_NOIO); | |
802 | if (!*ops) | |
803 | return -ENOMEM; | |
804 | (*ops)[0].op = opcode; | |
805 | /* | |
806 | * op extent offset and length will be set later on | |
807 | * in calc_raw_layout() | |
808 | */ | |
809 | (*ops)[0].payload_len = payload_len; | |
810 | return 0; | |
811 | } | |
812 | ||
813 | static void rbd_destroy_ops(struct ceph_osd_req_op *ops) | |
814 | { | |
815 | kfree(ops); | |
816 | } | |
817 | ||
1fec7093 YS |
818 | static void rbd_coll_end_req_index(struct request *rq, |
819 | struct rbd_req_coll *coll, | |
820 | int index, | |
821 | int ret, u64 len) | |
822 | { | |
823 | struct request_queue *q; | |
824 | int min, max, i; | |
825 | ||
826 | dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n", | |
827 | coll, index, ret, len); | |
828 | ||
829 | if (!rq) | |
830 | return; | |
831 | ||
832 | if (!coll) { | |
833 | blk_end_request(rq, ret, len); | |
834 | return; | |
835 | } | |
836 | ||
837 | q = rq->q; | |
838 | ||
839 | spin_lock_irq(q->queue_lock); | |
840 | coll->status[index].done = 1; | |
841 | coll->status[index].rc = ret; | |
842 | coll->status[index].bytes = len; | |
843 | max = min = coll->num_done; | |
844 | while (max < coll->total && coll->status[max].done) | |
845 | max++; | |
846 | ||
847 | for (i = min; i<max; i++) { | |
848 | __blk_end_request(rq, coll->status[i].rc, | |
849 | coll->status[i].bytes); | |
850 | coll->num_done++; | |
851 | kref_put(&coll->kref, rbd_coll_release); | |
852 | } | |
853 | spin_unlock_irq(q->queue_lock); | |
854 | } | |
855 | ||
856 | static void rbd_coll_end_req(struct rbd_request *req, | |
857 | int ret, u64 len) | |
858 | { | |
859 | rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len); | |
860 | } | |
861 | ||
602adf40 YS |
862 | /* |
863 | * Send ceph osd request | |
864 | */ | |
865 | static int rbd_do_request(struct request *rq, | |
866 | struct rbd_device *dev, | |
867 | struct ceph_snap_context *snapc, | |
868 | u64 snapid, | |
869 | const char *obj, u64 ofs, u64 len, | |
870 | struct bio *bio, | |
871 | struct page **pages, | |
872 | int num_pages, | |
873 | int flags, | |
874 | struct ceph_osd_req_op *ops, | |
875 | int num_reply, | |
1fec7093 YS |
876 | struct rbd_req_coll *coll, |
877 | int coll_index, | |
602adf40 | 878 | void (*rbd_cb)(struct ceph_osd_request *req, |
59c2be1e YS |
879 | struct ceph_msg *msg), |
880 | struct ceph_osd_request **linger_req, | |
881 | u64 *ver) | |
602adf40 YS |
882 | { |
883 | struct ceph_osd_request *req; | |
884 | struct ceph_file_layout *layout; | |
885 | int ret; | |
886 | u64 bno; | |
887 | struct timespec mtime = CURRENT_TIME; | |
888 | struct rbd_request *req_data; | |
889 | struct ceph_osd_request_head *reqhead; | |
1dbb4399 | 890 | struct ceph_osd_client *osdc; |
602adf40 | 891 | |
602adf40 | 892 | req_data = kzalloc(sizeof(*req_data), GFP_NOIO); |
1fec7093 YS |
893 | if (!req_data) { |
894 | if (coll) | |
895 | rbd_coll_end_req_index(rq, coll, coll_index, | |
896 | -ENOMEM, len); | |
897 | return -ENOMEM; | |
898 | } | |
899 | ||
900 | if (coll) { | |
901 | req_data->coll = coll; | |
902 | req_data->coll_index = coll_index; | |
903 | } | |
602adf40 | 904 | |
1fec7093 | 905 | dout("rbd_do_request obj=%s ofs=%lld len=%lld\n", obj, len, ofs); |
602adf40 | 906 | |
c666601a | 907 | down_read(&dev->header_rwsem); |
602adf40 | 908 | |
1dbb4399 AE |
909 | osdc = &dev->rbd_client->client->osdc; |
910 | req = ceph_osdc_alloc_request(osdc, flags, snapc, ops, | |
911 | false, GFP_NOIO, pages, bio); | |
4ad12621 | 912 | if (!req) { |
c666601a | 913 | up_read(&dev->header_rwsem); |
4ad12621 | 914 | ret = -ENOMEM; |
602adf40 YS |
915 | goto done_pages; |
916 | } | |
917 | ||
918 | req->r_callback = rbd_cb; | |
919 | ||
920 | req_data->rq = rq; | |
921 | req_data->bio = bio; | |
922 | req_data->pages = pages; | |
923 | req_data->len = len; | |
924 | ||
925 | req->r_priv = req_data; | |
926 | ||
927 | reqhead = req->r_request->front.iov_base; | |
928 | reqhead->snapid = cpu_to_le64(CEPH_NOSNAP); | |
929 | ||
930 | strncpy(req->r_oid, obj, sizeof(req->r_oid)); | |
931 | req->r_oid_len = strlen(req->r_oid); | |
932 | ||
933 | layout = &req->r_file_layout; | |
934 | memset(layout, 0, sizeof(*layout)); | |
935 | layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
936 | layout->fl_stripe_count = cpu_to_le32(1); | |
937 | layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
602adf40 | 938 | layout->fl_pg_pool = cpu_to_le32(dev->poolid); |
1dbb4399 AE |
939 | ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno, |
940 | req, ops); | |
602adf40 YS |
941 | |
942 | ceph_osdc_build_request(req, ofs, &len, | |
943 | ops, | |
944 | snapc, | |
945 | &mtime, | |
946 | req->r_oid, req->r_oid_len); | |
c666601a | 947 | up_read(&dev->header_rwsem); |
602adf40 | 948 | |
59c2be1e | 949 | if (linger_req) { |
1dbb4399 | 950 | ceph_osdc_set_request_linger(osdc, req); |
59c2be1e YS |
951 | *linger_req = req; |
952 | } | |
953 | ||
1dbb4399 | 954 | ret = ceph_osdc_start_request(osdc, req, false); |
602adf40 YS |
955 | if (ret < 0) |
956 | goto done_err; | |
957 | ||
958 | if (!rbd_cb) { | |
1dbb4399 | 959 | ret = ceph_osdc_wait_request(osdc, req); |
59c2be1e YS |
960 | if (ver) |
961 | *ver = le64_to_cpu(req->r_reassert_version.version); | |
1fec7093 YS |
962 | dout("reassert_ver=%lld\n", |
963 | le64_to_cpu(req->r_reassert_version.version)); | |
602adf40 YS |
964 | ceph_osdc_put_request(req); |
965 | } | |
966 | return ret; | |
967 | ||
968 | done_err: | |
969 | bio_chain_put(req_data->bio); | |
970 | ceph_osdc_put_request(req); | |
971 | done_pages: | |
1fec7093 | 972 | rbd_coll_end_req(req_data, ret, len); |
602adf40 | 973 | kfree(req_data); |
602adf40 YS |
974 | return ret; |
975 | } | |
976 | ||
977 | /* | |
978 | * Ceph osd op callback | |
979 | */ | |
980 | static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) | |
981 | { | |
982 | struct rbd_request *req_data = req->r_priv; | |
983 | struct ceph_osd_reply_head *replyhead; | |
984 | struct ceph_osd_op *op; | |
985 | __s32 rc; | |
986 | u64 bytes; | |
987 | int read_op; | |
988 | ||
989 | /* parse reply */ | |
990 | replyhead = msg->front.iov_base; | |
991 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
992 | op = (void *)(replyhead + 1); | |
993 | rc = le32_to_cpu(replyhead->result); | |
994 | bytes = le64_to_cpu(op->extent.length); | |
995 | read_op = (le32_to_cpu(op->op) == CEPH_OSD_OP_READ); | |
996 | ||
997 | dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc); | |
998 | ||
999 | if (rc == -ENOENT && read_op) { | |
1000 | zero_bio_chain(req_data->bio, 0); | |
1001 | rc = 0; | |
1002 | } else if (rc == 0 && read_op && bytes < req_data->len) { | |
1003 | zero_bio_chain(req_data->bio, bytes); | |
1004 | bytes = req_data->len; | |
1005 | } | |
1006 | ||
1fec7093 | 1007 | rbd_coll_end_req(req_data, rc, bytes); |
602adf40 YS |
1008 | |
1009 | if (req_data->bio) | |
1010 | bio_chain_put(req_data->bio); | |
1011 | ||
1012 | ceph_osdc_put_request(req); | |
1013 | kfree(req_data); | |
1014 | } | |
1015 | ||
59c2be1e YS |
1016 | static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) |
1017 | { | |
1018 | ceph_osdc_put_request(req); | |
1019 | } | |
1020 | ||
602adf40 YS |
1021 | /* |
1022 | * Do a synchronous ceph osd operation | |
1023 | */ | |
1024 | static int rbd_req_sync_op(struct rbd_device *dev, | |
1025 | struct ceph_snap_context *snapc, | |
1026 | u64 snapid, | |
1027 | int opcode, | |
1028 | int flags, | |
1029 | struct ceph_osd_req_op *orig_ops, | |
1030 | int num_reply, | |
1031 | const char *obj, | |
1032 | u64 ofs, u64 len, | |
59c2be1e YS |
1033 | char *buf, |
1034 | struct ceph_osd_request **linger_req, | |
1035 | u64 *ver) | |
602adf40 YS |
1036 | { |
1037 | int ret; | |
1038 | struct page **pages; | |
1039 | int num_pages; | |
1040 | struct ceph_osd_req_op *ops = orig_ops; | |
1041 | u32 payload_len; | |
1042 | ||
1043 | num_pages = calc_pages_for(ofs , len); | |
1044 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
b8d0638a DC |
1045 | if (IS_ERR(pages)) |
1046 | return PTR_ERR(pages); | |
602adf40 YS |
1047 | |
1048 | if (!orig_ops) { | |
1049 | payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0); | |
1050 | ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len); | |
1051 | if (ret < 0) | |
1052 | goto done; | |
1053 | ||
1054 | if ((flags & CEPH_OSD_FLAG_WRITE) && buf) { | |
1055 | ret = ceph_copy_to_page_vector(pages, buf, ofs, len); | |
1056 | if (ret < 0) | |
1057 | goto done_ops; | |
1058 | } | |
1059 | } | |
1060 | ||
1061 | ret = rbd_do_request(NULL, dev, snapc, snapid, | |
1062 | obj, ofs, len, NULL, | |
1063 | pages, num_pages, | |
1064 | flags, | |
1065 | ops, | |
1066 | 2, | |
1fec7093 | 1067 | NULL, 0, |
59c2be1e YS |
1068 | NULL, |
1069 | linger_req, ver); | |
602adf40 YS |
1070 | if (ret < 0) |
1071 | goto done_ops; | |
1072 | ||
1073 | if ((flags & CEPH_OSD_FLAG_READ) && buf) | |
1074 | ret = ceph_copy_from_page_vector(pages, buf, ofs, ret); | |
1075 | ||
1076 | done_ops: | |
1077 | if (!orig_ops) | |
1078 | rbd_destroy_ops(ops); | |
1079 | done: | |
1080 | ceph_release_page_vector(pages, num_pages); | |
1081 | return ret; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Do an asynchronous ceph osd operation | |
1086 | */ | |
1087 | static int rbd_do_op(struct request *rq, | |
1088 | struct rbd_device *rbd_dev , | |
1089 | struct ceph_snap_context *snapc, | |
1090 | u64 snapid, | |
1091 | int opcode, int flags, int num_reply, | |
1092 | u64 ofs, u64 len, | |
1fec7093 YS |
1093 | struct bio *bio, |
1094 | struct rbd_req_coll *coll, | |
1095 | int coll_index) | |
602adf40 YS |
1096 | { |
1097 | char *seg_name; | |
1098 | u64 seg_ofs; | |
1099 | u64 seg_len; | |
1100 | int ret; | |
1101 | struct ceph_osd_req_op *ops; | |
1102 | u32 payload_len; | |
1103 | ||
1104 | seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO); | |
1105 | if (!seg_name) | |
1106 | return -ENOMEM; | |
1107 | ||
1108 | seg_len = rbd_get_segment(&rbd_dev->header, | |
1109 | rbd_dev->header.block_name, | |
1110 | ofs, len, | |
1111 | seg_name, &seg_ofs); | |
602adf40 YS |
1112 | |
1113 | payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0); | |
1114 | ||
1115 | ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len); | |
1116 | if (ret < 0) | |
1117 | goto done; | |
1118 | ||
1119 | /* we've taken care of segment sizes earlier when we | |
1120 | cloned the bios. We should never have a segment | |
1121 | truncated at this point */ | |
1122 | BUG_ON(seg_len < len); | |
1123 | ||
1124 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1125 | seg_name, seg_ofs, seg_len, | |
1126 | bio, | |
1127 | NULL, 0, | |
1128 | flags, | |
1129 | ops, | |
1130 | num_reply, | |
1fec7093 | 1131 | coll, coll_index, |
59c2be1e | 1132 | rbd_req_cb, 0, NULL); |
11f77002 SW |
1133 | |
1134 | rbd_destroy_ops(ops); | |
602adf40 YS |
1135 | done: |
1136 | kfree(seg_name); | |
1137 | return ret; | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * Request async osd write | |
1142 | */ | |
1143 | static int rbd_req_write(struct request *rq, | |
1144 | struct rbd_device *rbd_dev, | |
1145 | struct ceph_snap_context *snapc, | |
1146 | u64 ofs, u64 len, | |
1fec7093 YS |
1147 | struct bio *bio, |
1148 | struct rbd_req_coll *coll, | |
1149 | int coll_index) | |
602adf40 YS |
1150 | { |
1151 | return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP, | |
1152 | CEPH_OSD_OP_WRITE, | |
1153 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1154 | 2, | |
1fec7093 | 1155 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1156 | } |
1157 | ||
1158 | /* | |
1159 | * Request async osd read | |
1160 | */ | |
1161 | static int rbd_req_read(struct request *rq, | |
1162 | struct rbd_device *rbd_dev, | |
1163 | u64 snapid, | |
1164 | u64 ofs, u64 len, | |
1fec7093 YS |
1165 | struct bio *bio, |
1166 | struct rbd_req_coll *coll, | |
1167 | int coll_index) | |
602adf40 YS |
1168 | { |
1169 | return rbd_do_op(rq, rbd_dev, NULL, | |
1170 | (snapid ? snapid : CEPH_NOSNAP), | |
1171 | CEPH_OSD_OP_READ, | |
1172 | CEPH_OSD_FLAG_READ, | |
1173 | 2, | |
1fec7093 | 1174 | ofs, len, bio, coll, coll_index); |
602adf40 YS |
1175 | } |
1176 | ||
1177 | /* | |
1178 | * Request sync osd read | |
1179 | */ | |
1180 | static int rbd_req_sync_read(struct rbd_device *dev, | |
1181 | struct ceph_snap_context *snapc, | |
1182 | u64 snapid, | |
1183 | const char *obj, | |
1184 | u64 ofs, u64 len, | |
59c2be1e YS |
1185 | char *buf, |
1186 | u64 *ver) | |
602adf40 YS |
1187 | { |
1188 | return rbd_req_sync_op(dev, NULL, | |
1189 | (snapid ? snapid : CEPH_NOSNAP), | |
1190 | CEPH_OSD_OP_READ, | |
1191 | CEPH_OSD_FLAG_READ, | |
1192 | NULL, | |
59c2be1e | 1193 | 1, obj, ofs, len, buf, NULL, ver); |
602adf40 YS |
1194 | } |
1195 | ||
1196 | /* | |
59c2be1e YS |
1197 | * Request sync osd watch |
1198 | */ | |
1199 | static int rbd_req_sync_notify_ack(struct rbd_device *dev, | |
1200 | u64 ver, | |
1201 | u64 notify_id, | |
1202 | const char *obj) | |
1203 | { | |
1204 | struct ceph_osd_req_op *ops; | |
1205 | struct page **pages = NULL; | |
11f77002 SW |
1206 | int ret; |
1207 | ||
1208 | ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0); | |
59c2be1e YS |
1209 | if (ret < 0) |
1210 | return ret; | |
1211 | ||
1212 | ops[0].watch.ver = cpu_to_le64(dev->header.obj_version); | |
1213 | ops[0].watch.cookie = notify_id; | |
1214 | ops[0].watch.flag = 0; | |
1215 | ||
1216 | ret = rbd_do_request(NULL, dev, NULL, CEPH_NOSNAP, | |
1217 | obj, 0, 0, NULL, | |
1218 | pages, 0, | |
1219 | CEPH_OSD_FLAG_READ, | |
1220 | ops, | |
1221 | 1, | |
1fec7093 | 1222 | NULL, 0, |
59c2be1e YS |
1223 | rbd_simple_req_cb, 0, NULL); |
1224 | ||
1225 | rbd_destroy_ops(ops); | |
1226 | return ret; | |
1227 | } | |
1228 | ||
1229 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1230 | { | |
1231 | struct rbd_device *dev = (struct rbd_device *)data; | |
13143d2d SW |
1232 | int rc; |
1233 | ||
59c2be1e YS |
1234 | if (!dev) |
1235 | return; | |
1236 | ||
1237 | dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name, | |
1238 | notify_id, (int)opcode); | |
1239 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
13143d2d | 1240 | rc = __rbd_update_snaps(dev); |
59c2be1e | 1241 | mutex_unlock(&ctl_mutex); |
13143d2d | 1242 | if (rc) |
f0f8cef5 AE |
1243 | pr_warning(RBD_DRV_NAME "%d got notification but failed to " |
1244 | " update snaps: %d\n", dev->major, rc); | |
59c2be1e YS |
1245 | |
1246 | rbd_req_sync_notify_ack(dev, ver, notify_id, dev->obj_md_name); | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * Request sync osd watch | |
1251 | */ | |
1252 | static int rbd_req_sync_watch(struct rbd_device *dev, | |
1253 | const char *obj, | |
1254 | u64 ver) | |
1255 | { | |
1256 | struct ceph_osd_req_op *ops; | |
1dbb4399 | 1257 | struct ceph_osd_client *osdc = &dev->rbd_client->client->osdc; |
59c2be1e YS |
1258 | |
1259 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0); | |
1260 | if (ret < 0) | |
1261 | return ret; | |
1262 | ||
1263 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, | |
1264 | (void *)dev, &dev->watch_event); | |
1265 | if (ret < 0) | |
1266 | goto fail; | |
1267 | ||
1268 | ops[0].watch.ver = cpu_to_le64(ver); | |
1269 | ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie); | |
1270 | ops[0].watch.flag = 1; | |
1271 | ||
1272 | ret = rbd_req_sync_op(dev, NULL, | |
1273 | CEPH_NOSNAP, | |
1274 | 0, | |
1275 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1276 | ops, | |
1277 | 1, obj, 0, 0, NULL, | |
1278 | &dev->watch_request, NULL); | |
1279 | ||
1280 | if (ret < 0) | |
1281 | goto fail_event; | |
1282 | ||
1283 | rbd_destroy_ops(ops); | |
1284 | return 0; | |
1285 | ||
1286 | fail_event: | |
1287 | ceph_osdc_cancel_event(dev->watch_event); | |
1288 | dev->watch_event = NULL; | |
1289 | fail: | |
1290 | rbd_destroy_ops(ops); | |
1291 | return ret; | |
1292 | } | |
1293 | ||
79e3057c YS |
1294 | /* |
1295 | * Request sync osd unwatch | |
1296 | */ | |
1297 | static int rbd_req_sync_unwatch(struct rbd_device *dev, | |
1298 | const char *obj) | |
1299 | { | |
1300 | struct ceph_osd_req_op *ops; | |
1301 | ||
1302 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0); | |
1303 | if (ret < 0) | |
1304 | return ret; | |
1305 | ||
1306 | ops[0].watch.ver = 0; | |
1307 | ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie); | |
1308 | ops[0].watch.flag = 0; | |
1309 | ||
1310 | ret = rbd_req_sync_op(dev, NULL, | |
1311 | CEPH_NOSNAP, | |
1312 | 0, | |
1313 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1314 | ops, | |
1315 | 1, obj, 0, 0, NULL, NULL, NULL); | |
1316 | ||
1317 | rbd_destroy_ops(ops); | |
1318 | ceph_osdc_cancel_event(dev->watch_event); | |
1319 | dev->watch_event = NULL; | |
1320 | return ret; | |
1321 | } | |
1322 | ||
59c2be1e YS |
1323 | struct rbd_notify_info { |
1324 | struct rbd_device *dev; | |
1325 | }; | |
1326 | ||
1327 | static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1328 | { | |
1329 | struct rbd_device *dev = (struct rbd_device *)data; | |
1330 | if (!dev) | |
1331 | return; | |
1332 | ||
1333 | dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name, | |
1334 | notify_id, (int)opcode); | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * Request sync osd notify | |
1339 | */ | |
1340 | static int rbd_req_sync_notify(struct rbd_device *dev, | |
1341 | const char *obj) | |
1342 | { | |
1343 | struct ceph_osd_req_op *ops; | |
1dbb4399 | 1344 | struct ceph_osd_client *osdc = &dev->rbd_client->client->osdc; |
59c2be1e YS |
1345 | struct ceph_osd_event *event; |
1346 | struct rbd_notify_info info; | |
1347 | int payload_len = sizeof(u32) + sizeof(u32); | |
1348 | int ret; | |
1349 | ||
1350 | ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len); | |
1351 | if (ret < 0) | |
1352 | return ret; | |
1353 | ||
1354 | info.dev = dev; | |
1355 | ||
1356 | ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1, | |
1357 | (void *)&info, &event); | |
1358 | if (ret < 0) | |
1359 | goto fail; | |
1360 | ||
1361 | ops[0].watch.ver = 1; | |
1362 | ops[0].watch.flag = 1; | |
1363 | ops[0].watch.cookie = event->cookie; | |
1364 | ops[0].watch.prot_ver = RADOS_NOTIFY_VER; | |
1365 | ops[0].watch.timeout = 12; | |
1366 | ||
1367 | ret = rbd_req_sync_op(dev, NULL, | |
1368 | CEPH_NOSNAP, | |
1369 | 0, | |
1370 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1371 | ops, | |
1372 | 1, obj, 0, 0, NULL, NULL, NULL); | |
1373 | if (ret < 0) | |
1374 | goto fail_event; | |
1375 | ||
1376 | ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT); | |
1377 | dout("ceph_osdc_wait_event returned %d\n", ret); | |
1378 | rbd_destroy_ops(ops); | |
1379 | return 0; | |
1380 | ||
1381 | fail_event: | |
1382 | ceph_osdc_cancel_event(event); | |
1383 | fail: | |
1384 | rbd_destroy_ops(ops); | |
1385 | return ret; | |
1386 | } | |
1387 | ||
602adf40 YS |
1388 | /* |
1389 | * Request sync osd read | |
1390 | */ | |
1391 | static int rbd_req_sync_exec(struct rbd_device *dev, | |
1392 | const char *obj, | |
1393 | const char *cls, | |
1394 | const char *method, | |
1395 | const char *data, | |
59c2be1e YS |
1396 | int len, |
1397 | u64 *ver) | |
602adf40 YS |
1398 | { |
1399 | struct ceph_osd_req_op *ops; | |
1400 | int cls_len = strlen(cls); | |
1401 | int method_len = strlen(method); | |
1402 | int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL, | |
1403 | cls_len + method_len + len); | |
1404 | if (ret < 0) | |
1405 | return ret; | |
1406 | ||
1407 | ops[0].cls.class_name = cls; | |
1408 | ops[0].cls.class_len = (__u8)cls_len; | |
1409 | ops[0].cls.method_name = method; | |
1410 | ops[0].cls.method_len = (__u8)method_len; | |
1411 | ops[0].cls.argc = 0; | |
1412 | ops[0].cls.indata = data; | |
1413 | ops[0].cls.indata_len = len; | |
1414 | ||
1415 | ret = rbd_req_sync_op(dev, NULL, | |
1416 | CEPH_NOSNAP, | |
1417 | 0, | |
1418 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, | |
1419 | ops, | |
59c2be1e | 1420 | 1, obj, 0, 0, NULL, NULL, ver); |
602adf40 YS |
1421 | |
1422 | rbd_destroy_ops(ops); | |
1423 | ||
1424 | dout("cls_exec returned %d\n", ret); | |
1425 | return ret; | |
1426 | } | |
1427 | ||
1fec7093 YS |
1428 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1429 | { | |
1430 | struct rbd_req_coll *coll = | |
1431 | kzalloc(sizeof(struct rbd_req_coll) + | |
1432 | sizeof(struct rbd_req_status) * num_reqs, | |
1433 | GFP_ATOMIC); | |
1434 | ||
1435 | if (!coll) | |
1436 | return NULL; | |
1437 | coll->total = num_reqs; | |
1438 | kref_init(&coll->kref); | |
1439 | return coll; | |
1440 | } | |
1441 | ||
602adf40 YS |
1442 | /* |
1443 | * block device queue callback | |
1444 | */ | |
1445 | static void rbd_rq_fn(struct request_queue *q) | |
1446 | { | |
1447 | struct rbd_device *rbd_dev = q->queuedata; | |
1448 | struct request *rq; | |
1449 | struct bio_pair *bp = NULL; | |
1450 | ||
00f1f36f | 1451 | while ((rq = blk_fetch_request(q))) { |
602adf40 YS |
1452 | struct bio *bio; |
1453 | struct bio *rq_bio, *next_bio = NULL; | |
1454 | bool do_write; | |
1455 | int size, op_size = 0; | |
1456 | u64 ofs; | |
1fec7093 YS |
1457 | int num_segs, cur_seg = 0; |
1458 | struct rbd_req_coll *coll; | |
602adf40 YS |
1459 | |
1460 | /* peek at request from block layer */ | |
1461 | if (!rq) | |
1462 | break; | |
1463 | ||
1464 | dout("fetched request\n"); | |
1465 | ||
1466 | /* filter out block requests we don't understand */ | |
1467 | if ((rq->cmd_type != REQ_TYPE_FS)) { | |
1468 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1469 | continue; |
602adf40 YS |
1470 | } |
1471 | ||
1472 | /* deduce our operation (read, write) */ | |
1473 | do_write = (rq_data_dir(rq) == WRITE); | |
1474 | ||
1475 | size = blk_rq_bytes(rq); | |
593a9e7b | 1476 | ofs = blk_rq_pos(rq) * SECTOR_SIZE; |
602adf40 YS |
1477 | rq_bio = rq->bio; |
1478 | if (do_write && rbd_dev->read_only) { | |
1479 | __blk_end_request_all(rq, -EROFS); | |
00f1f36f | 1480 | continue; |
602adf40 YS |
1481 | } |
1482 | ||
1483 | spin_unlock_irq(q->queue_lock); | |
1484 | ||
1485 | dout("%s 0x%x bytes at 0x%llx\n", | |
1486 | do_write ? "write" : "read", | |
593a9e7b | 1487 | size, blk_rq_pos(rq) * SECTOR_SIZE); |
602adf40 | 1488 | |
1fec7093 YS |
1489 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); |
1490 | coll = rbd_alloc_coll(num_segs); | |
1491 | if (!coll) { | |
1492 | spin_lock_irq(q->queue_lock); | |
1493 | __blk_end_request_all(rq, -ENOMEM); | |
00f1f36f | 1494 | continue; |
1fec7093 YS |
1495 | } |
1496 | ||
602adf40 YS |
1497 | do { |
1498 | /* a bio clone to be passed down to OSD req */ | |
1499 | dout("rq->bio->bi_vcnt=%d\n", rq->bio->bi_vcnt); | |
1500 | op_size = rbd_get_segment(&rbd_dev->header, | |
1501 | rbd_dev->header.block_name, | |
1502 | ofs, size, | |
1503 | NULL, NULL); | |
1fec7093 | 1504 | kref_get(&coll->kref); |
602adf40 YS |
1505 | bio = bio_chain_clone(&rq_bio, &next_bio, &bp, |
1506 | op_size, GFP_ATOMIC); | |
1507 | if (!bio) { | |
1fec7093 YS |
1508 | rbd_coll_end_req_index(rq, coll, cur_seg, |
1509 | -ENOMEM, op_size); | |
1510 | goto next_seg; | |
602adf40 YS |
1511 | } |
1512 | ||
1fec7093 | 1513 | |
602adf40 YS |
1514 | /* init OSD command: write or read */ |
1515 | if (do_write) | |
1516 | rbd_req_write(rq, rbd_dev, | |
1517 | rbd_dev->header.snapc, | |
1518 | ofs, | |
1fec7093 YS |
1519 | op_size, bio, |
1520 | coll, cur_seg); | |
602adf40 YS |
1521 | else |
1522 | rbd_req_read(rq, rbd_dev, | |
1523 | cur_snap_id(rbd_dev), | |
1524 | ofs, | |
1fec7093 YS |
1525 | op_size, bio, |
1526 | coll, cur_seg); | |
602adf40 | 1527 | |
1fec7093 | 1528 | next_seg: |
602adf40 YS |
1529 | size -= op_size; |
1530 | ofs += op_size; | |
1531 | ||
1fec7093 | 1532 | cur_seg++; |
602adf40 YS |
1533 | rq_bio = next_bio; |
1534 | } while (size > 0); | |
1fec7093 | 1535 | kref_put(&coll->kref, rbd_coll_release); |
602adf40 YS |
1536 | |
1537 | if (bp) | |
1538 | bio_pair_release(bp); | |
602adf40 | 1539 | spin_lock_irq(q->queue_lock); |
602adf40 YS |
1540 | } |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1545 | * multiple osd objects. One exception would be with a single page bios, | |
1546 | * which we handle later at bio_chain_clone | |
1547 | */ | |
1548 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1549 | struct bio_vec *bvec) | |
1550 | { | |
1551 | struct rbd_device *rbd_dev = q->queuedata; | |
593a9e7b AE |
1552 | unsigned int chunk_sectors; |
1553 | sector_t sector; | |
1554 | unsigned int bio_sectors; | |
602adf40 YS |
1555 | int max; |
1556 | ||
593a9e7b AE |
1557 | chunk_sectors = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); |
1558 | sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev); | |
1559 | bio_sectors = bmd->bi_size >> SECTOR_SHIFT; | |
1560 | ||
602adf40 | 1561 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) |
593a9e7b | 1562 | + bio_sectors)) << SECTOR_SHIFT; |
602adf40 YS |
1563 | if (max < 0) |
1564 | max = 0; /* bio_add cannot handle a negative return */ | |
1565 | if (max <= bvec->bv_len && bio_sectors == 0) | |
1566 | return bvec->bv_len; | |
1567 | return max; | |
1568 | } | |
1569 | ||
1570 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1571 | { | |
1572 | struct gendisk *disk = rbd_dev->disk; | |
1573 | ||
1574 | if (!disk) | |
1575 | return; | |
1576 | ||
1577 | rbd_header_free(&rbd_dev->header); | |
1578 | ||
1579 | if (disk->flags & GENHD_FL_UP) | |
1580 | del_gendisk(disk); | |
1581 | if (disk->queue) | |
1582 | blk_cleanup_queue(disk->queue); | |
1583 | put_disk(disk); | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * reload the ondisk the header | |
1588 | */ | |
1589 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1590 | struct rbd_image_header *header) | |
1591 | { | |
1592 | ssize_t rc; | |
1593 | struct rbd_image_header_ondisk *dh; | |
1594 | int snap_count = 0; | |
59c2be1e | 1595 | u64 ver; |
00f1f36f | 1596 | size_t len; |
602adf40 | 1597 | |
00f1f36f AE |
1598 | /* |
1599 | * First reads the fixed-size header to determine the number | |
1600 | * of snapshots, then re-reads it, along with all snapshot | |
1601 | * records as well as their stored names. | |
1602 | */ | |
1603 | len = sizeof (*dh); | |
602adf40 | 1604 | while (1) { |
602adf40 YS |
1605 | dh = kmalloc(len, GFP_KERNEL); |
1606 | if (!dh) | |
1607 | return -ENOMEM; | |
1608 | ||
1609 | rc = rbd_req_sync_read(rbd_dev, | |
1610 | NULL, CEPH_NOSNAP, | |
1611 | rbd_dev->obj_md_name, | |
1612 | 0, len, | |
59c2be1e | 1613 | (char *)dh, &ver); |
602adf40 YS |
1614 | if (rc < 0) |
1615 | goto out_dh; | |
1616 | ||
1617 | rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL); | |
81e759fb | 1618 | if (rc < 0) { |
00f1f36f | 1619 | if (rc == -ENXIO) |
81e759fb JD |
1620 | pr_warning("unrecognized header format" |
1621 | " for image %s", rbd_dev->obj); | |
602adf40 | 1622 | goto out_dh; |
81e759fb | 1623 | } |
602adf40 | 1624 | |
00f1f36f AE |
1625 | if (snap_count == header->total_snaps) |
1626 | break; | |
1627 | ||
1628 | snap_count = header->total_snaps; | |
1629 | len = sizeof (*dh) + | |
1630 | snap_count * sizeof(struct rbd_image_snap_ondisk) + | |
1631 | header->snap_names_len; | |
1632 | ||
1633 | rbd_header_free(header); | |
1634 | kfree(dh); | |
602adf40 | 1635 | } |
59c2be1e | 1636 | header->obj_version = ver; |
602adf40 YS |
1637 | |
1638 | out_dh: | |
1639 | kfree(dh); | |
1640 | return rc; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * create a snapshot | |
1645 | */ | |
1646 | static int rbd_header_add_snap(struct rbd_device *dev, | |
1647 | const char *snap_name, | |
1648 | gfp_t gfp_flags) | |
1649 | { | |
1650 | int name_len = strlen(snap_name); | |
1651 | u64 new_snapid; | |
1652 | int ret; | |
916d4d67 | 1653 | void *data, *p, *e; |
59c2be1e | 1654 | u64 ver; |
1dbb4399 | 1655 | struct ceph_mon_client *monc; |
602adf40 YS |
1656 | |
1657 | /* we should create a snapshot only if we're pointing at the head */ | |
1658 | if (dev->cur_snap) | |
1659 | return -EINVAL; | |
1660 | ||
1dbb4399 AE |
1661 | monc = &dev->rbd_client->client->monc; |
1662 | ret = ceph_monc_create_snapid(monc, dev->poolid, &new_snapid); | |
602adf40 YS |
1663 | dout("created snapid=%lld\n", new_snapid); |
1664 | if (ret < 0) | |
1665 | return ret; | |
1666 | ||
1667 | data = kmalloc(name_len + 16, gfp_flags); | |
1668 | if (!data) | |
1669 | return -ENOMEM; | |
1670 | ||
916d4d67 SW |
1671 | p = data; |
1672 | e = data + name_len + 16; | |
602adf40 | 1673 | |
916d4d67 SW |
1674 | ceph_encode_string_safe(&p, e, snap_name, name_len, bad); |
1675 | ceph_encode_64_safe(&p, e, new_snapid, bad); | |
602adf40 YS |
1676 | |
1677 | ret = rbd_req_sync_exec(dev, dev->obj_md_name, "rbd", "snap_add", | |
916d4d67 | 1678 | data, p - data, &ver); |
602adf40 | 1679 | |
916d4d67 | 1680 | kfree(data); |
602adf40 YS |
1681 | |
1682 | if (ret < 0) | |
1683 | return ret; | |
1684 | ||
1685 | dev->header.snapc->seq = new_snapid; | |
1686 | ||
1687 | return 0; | |
1688 | bad: | |
1689 | return -ERANGE; | |
1690 | } | |
1691 | ||
dfc5606d YS |
1692 | static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
1693 | { | |
1694 | struct rbd_snap *snap; | |
1695 | ||
1696 | while (!list_empty(&rbd_dev->snaps)) { | |
1697 | snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node); | |
1698 | __rbd_remove_snap_dev(rbd_dev, snap); | |
1699 | } | |
1700 | } | |
1701 | ||
602adf40 YS |
1702 | /* |
1703 | * only read the first part of the ondisk header, without the snaps info | |
1704 | */ | |
dfc5606d | 1705 | static int __rbd_update_snaps(struct rbd_device *rbd_dev) |
602adf40 YS |
1706 | { |
1707 | int ret; | |
1708 | struct rbd_image_header h; | |
1709 | u64 snap_seq; | |
59c2be1e | 1710 | int follow_seq = 0; |
602adf40 YS |
1711 | |
1712 | ret = rbd_read_header(rbd_dev, &h); | |
1713 | if (ret < 0) | |
1714 | return ret; | |
1715 | ||
9db4b3e3 | 1716 | /* resized? */ |
593a9e7b | 1717 | set_capacity(rbd_dev->disk, h.image_size / SECTOR_SIZE); |
9db4b3e3 | 1718 | |
c666601a | 1719 | down_write(&rbd_dev->header_rwsem); |
602adf40 YS |
1720 | |
1721 | snap_seq = rbd_dev->header.snapc->seq; | |
59c2be1e YS |
1722 | if (rbd_dev->header.total_snaps && |
1723 | rbd_dev->header.snapc->snaps[0] == snap_seq) | |
1724 | /* pointing at the head, will need to follow that | |
1725 | if head moves */ | |
1726 | follow_seq = 1; | |
602adf40 YS |
1727 | |
1728 | kfree(rbd_dev->header.snapc); | |
1729 | kfree(rbd_dev->header.snap_names); | |
1730 | kfree(rbd_dev->header.snap_sizes); | |
1731 | ||
1732 | rbd_dev->header.total_snaps = h.total_snaps; | |
1733 | rbd_dev->header.snapc = h.snapc; | |
1734 | rbd_dev->header.snap_names = h.snap_names; | |
dfc5606d | 1735 | rbd_dev->header.snap_names_len = h.snap_names_len; |
602adf40 | 1736 | rbd_dev->header.snap_sizes = h.snap_sizes; |
59c2be1e YS |
1737 | if (follow_seq) |
1738 | rbd_dev->header.snapc->seq = rbd_dev->header.snapc->snaps[0]; | |
1739 | else | |
1740 | rbd_dev->header.snapc->seq = snap_seq; | |
602adf40 | 1741 | |
dfc5606d YS |
1742 | ret = __rbd_init_snaps_header(rbd_dev); |
1743 | ||
c666601a | 1744 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1745 | |
dfc5606d | 1746 | return ret; |
602adf40 YS |
1747 | } |
1748 | ||
1749 | static int rbd_init_disk(struct rbd_device *rbd_dev) | |
1750 | { | |
1751 | struct gendisk *disk; | |
1752 | struct request_queue *q; | |
1753 | int rc; | |
593a9e7b | 1754 | u64 segment_size; |
602adf40 YS |
1755 | u64 total_size = 0; |
1756 | ||
1757 | /* contact OSD, request size info about the object being mapped */ | |
1758 | rc = rbd_read_header(rbd_dev, &rbd_dev->header); | |
1759 | if (rc) | |
1760 | return rc; | |
1761 | ||
dfc5606d YS |
1762 | /* no need to lock here, as rbd_dev is not registered yet */ |
1763 | rc = __rbd_init_snaps_header(rbd_dev); | |
1764 | if (rc) | |
1765 | return rc; | |
1766 | ||
cc9d734c | 1767 | rc = rbd_header_set_snap(rbd_dev, &total_size); |
602adf40 YS |
1768 | if (rc) |
1769 | return rc; | |
1770 | ||
1771 | /* create gendisk info */ | |
1772 | rc = -ENOMEM; | |
1773 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); | |
1774 | if (!disk) | |
1775 | goto out; | |
1776 | ||
f0f8cef5 | 1777 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
aedfec59 | 1778 | rbd_dev->id); |
602adf40 YS |
1779 | disk->major = rbd_dev->major; |
1780 | disk->first_minor = 0; | |
1781 | disk->fops = &rbd_bd_ops; | |
1782 | disk->private_data = rbd_dev; | |
1783 | ||
1784 | /* init rq */ | |
1785 | rc = -ENOMEM; | |
1786 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); | |
1787 | if (!q) | |
1788 | goto out_disk; | |
029bcbd8 | 1789 | |
593a9e7b AE |
1790 | /* We use the default size, but let's be explicit about it. */ |
1791 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1792 | ||
029bcbd8 | 1793 | /* set io sizes to object size */ |
593a9e7b AE |
1794 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1795 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1796 | blk_queue_max_segment_size(q, segment_size); | |
1797 | blk_queue_io_min(q, segment_size); | |
1798 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1799 | |
602adf40 YS |
1800 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1801 | disk->queue = q; | |
1802 | ||
1803 | q->queuedata = rbd_dev; | |
1804 | ||
1805 | rbd_dev->disk = disk; | |
1806 | rbd_dev->q = q; | |
1807 | ||
1808 | /* finally, announce the disk to the world */ | |
593a9e7b | 1809 | set_capacity(disk, total_size / SECTOR_SIZE); |
602adf40 YS |
1810 | add_disk(disk); |
1811 | ||
1812 | pr_info("%s: added with size 0x%llx\n", | |
1813 | disk->disk_name, (unsigned long long)total_size); | |
1814 | return 0; | |
1815 | ||
1816 | out_disk: | |
1817 | put_disk(disk); | |
1818 | out: | |
1819 | return rc; | |
1820 | } | |
1821 | ||
dfc5606d YS |
1822 | /* |
1823 | sysfs | |
1824 | */ | |
1825 | ||
593a9e7b AE |
1826 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1827 | { | |
1828 | return container_of(dev, struct rbd_device, dev); | |
1829 | } | |
1830 | ||
dfc5606d YS |
1831 | static ssize_t rbd_size_show(struct device *dev, |
1832 | struct device_attribute *attr, char *buf) | |
1833 | { | |
593a9e7b | 1834 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1835 | |
1836 | return sprintf(buf, "%llu\n", (unsigned long long)rbd_dev->header.image_size); | |
1837 | } | |
1838 | ||
1839 | static ssize_t rbd_major_show(struct device *dev, | |
1840 | struct device_attribute *attr, char *buf) | |
1841 | { | |
593a9e7b | 1842 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 1843 | |
dfc5606d YS |
1844 | return sprintf(buf, "%d\n", rbd_dev->major); |
1845 | } | |
1846 | ||
1847 | static ssize_t rbd_client_id_show(struct device *dev, | |
1848 | struct device_attribute *attr, char *buf) | |
602adf40 | 1849 | { |
593a9e7b | 1850 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 1851 | |
1dbb4399 AE |
1852 | return sprintf(buf, "client%lld\n", |
1853 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
1854 | } |
1855 | ||
dfc5606d YS |
1856 | static ssize_t rbd_pool_show(struct device *dev, |
1857 | struct device_attribute *attr, char *buf) | |
602adf40 | 1858 | { |
593a9e7b | 1859 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1860 | |
1861 | return sprintf(buf, "%s\n", rbd_dev->pool_name); | |
1862 | } | |
1863 | ||
1864 | static ssize_t rbd_name_show(struct device *dev, | |
1865 | struct device_attribute *attr, char *buf) | |
1866 | { | |
593a9e7b | 1867 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1868 | |
1869 | return sprintf(buf, "%s\n", rbd_dev->obj); | |
1870 | } | |
1871 | ||
1872 | static ssize_t rbd_snap_show(struct device *dev, | |
1873 | struct device_attribute *attr, | |
1874 | char *buf) | |
1875 | { | |
593a9e7b | 1876 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1877 | |
1878 | return sprintf(buf, "%s\n", rbd_dev->snap_name); | |
1879 | } | |
1880 | ||
1881 | static ssize_t rbd_image_refresh(struct device *dev, | |
1882 | struct device_attribute *attr, | |
1883 | const char *buf, | |
1884 | size_t size) | |
1885 | { | |
593a9e7b | 1886 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
1887 | int rc; |
1888 | int ret = size; | |
602adf40 YS |
1889 | |
1890 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
1891 | ||
dfc5606d YS |
1892 | rc = __rbd_update_snaps(rbd_dev); |
1893 | if (rc < 0) | |
1894 | ret = rc; | |
602adf40 | 1895 | |
dfc5606d YS |
1896 | mutex_unlock(&ctl_mutex); |
1897 | return ret; | |
1898 | } | |
602adf40 | 1899 | |
dfc5606d YS |
1900 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
1901 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); | |
1902 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
1903 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
1904 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); | |
1905 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); | |
1906 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
1907 | static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add); | |
dfc5606d YS |
1908 | |
1909 | static struct attribute *rbd_attrs[] = { | |
1910 | &dev_attr_size.attr, | |
1911 | &dev_attr_major.attr, | |
1912 | &dev_attr_client_id.attr, | |
1913 | &dev_attr_pool.attr, | |
1914 | &dev_attr_name.attr, | |
1915 | &dev_attr_current_snap.attr, | |
1916 | &dev_attr_refresh.attr, | |
1917 | &dev_attr_create_snap.attr, | |
dfc5606d YS |
1918 | NULL |
1919 | }; | |
1920 | ||
1921 | static struct attribute_group rbd_attr_group = { | |
1922 | .attrs = rbd_attrs, | |
1923 | }; | |
1924 | ||
1925 | static const struct attribute_group *rbd_attr_groups[] = { | |
1926 | &rbd_attr_group, | |
1927 | NULL | |
1928 | }; | |
1929 | ||
1930 | static void rbd_sysfs_dev_release(struct device *dev) | |
1931 | { | |
1932 | } | |
1933 | ||
1934 | static struct device_type rbd_device_type = { | |
1935 | .name = "rbd", | |
1936 | .groups = rbd_attr_groups, | |
1937 | .release = rbd_sysfs_dev_release, | |
1938 | }; | |
1939 | ||
1940 | ||
1941 | /* | |
1942 | sysfs - snapshots | |
1943 | */ | |
1944 | ||
1945 | static ssize_t rbd_snap_size_show(struct device *dev, | |
1946 | struct device_attribute *attr, | |
1947 | char *buf) | |
1948 | { | |
1949 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1950 | ||
593a9e7b | 1951 | return sprintf(buf, "%zd\n", snap->size); |
dfc5606d YS |
1952 | } |
1953 | ||
1954 | static ssize_t rbd_snap_id_show(struct device *dev, | |
1955 | struct device_attribute *attr, | |
1956 | char *buf) | |
1957 | { | |
1958 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1959 | ||
593a9e7b | 1960 | return sprintf(buf, "%llu\n", (unsigned long long) snap->id); |
dfc5606d YS |
1961 | } |
1962 | ||
1963 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); | |
1964 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
1965 | ||
1966 | static struct attribute *rbd_snap_attrs[] = { | |
1967 | &dev_attr_snap_size.attr, | |
1968 | &dev_attr_snap_id.attr, | |
1969 | NULL, | |
1970 | }; | |
1971 | ||
1972 | static struct attribute_group rbd_snap_attr_group = { | |
1973 | .attrs = rbd_snap_attrs, | |
1974 | }; | |
1975 | ||
1976 | static void rbd_snap_dev_release(struct device *dev) | |
1977 | { | |
1978 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
1979 | kfree(snap->name); | |
1980 | kfree(snap); | |
1981 | } | |
1982 | ||
1983 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
1984 | &rbd_snap_attr_group, | |
1985 | NULL | |
1986 | }; | |
1987 | ||
1988 | static struct device_type rbd_snap_device_type = { | |
1989 | .groups = rbd_snap_attr_groups, | |
1990 | .release = rbd_snap_dev_release, | |
1991 | }; | |
1992 | ||
1993 | static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev, | |
1994 | struct rbd_snap *snap) | |
1995 | { | |
1996 | list_del(&snap->node); | |
1997 | device_unregister(&snap->dev); | |
1998 | } | |
1999 | ||
2000 | static int rbd_register_snap_dev(struct rbd_device *rbd_dev, | |
2001 | struct rbd_snap *snap, | |
2002 | struct device *parent) | |
2003 | { | |
2004 | struct device *dev = &snap->dev; | |
2005 | int ret; | |
2006 | ||
2007 | dev->type = &rbd_snap_device_type; | |
2008 | dev->parent = parent; | |
2009 | dev->release = rbd_snap_dev_release; | |
2010 | dev_set_name(dev, "snap_%s", snap->name); | |
2011 | ret = device_register(dev); | |
2012 | ||
2013 | return ret; | |
2014 | } | |
2015 | ||
2016 | static int __rbd_add_snap_dev(struct rbd_device *rbd_dev, | |
2017 | int i, const char *name, | |
2018 | struct rbd_snap **snapp) | |
2019 | { | |
2020 | int ret; | |
2021 | struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL); | |
2022 | if (!snap) | |
2023 | return -ENOMEM; | |
2024 | snap->name = kstrdup(name, GFP_KERNEL); | |
2025 | snap->size = rbd_dev->header.snap_sizes[i]; | |
2026 | snap->id = rbd_dev->header.snapc->snaps[i]; | |
2027 | if (device_is_registered(&rbd_dev->dev)) { | |
2028 | ret = rbd_register_snap_dev(rbd_dev, snap, | |
2029 | &rbd_dev->dev); | |
2030 | if (ret < 0) | |
2031 | goto err; | |
2032 | } | |
2033 | *snapp = snap; | |
2034 | return 0; | |
2035 | err: | |
2036 | kfree(snap->name); | |
2037 | kfree(snap); | |
2038 | return ret; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * search for the previous snap in a null delimited string list | |
2043 | */ | |
2044 | const char *rbd_prev_snap_name(const char *name, const char *start) | |
2045 | { | |
2046 | if (name < start + 2) | |
2047 | return NULL; | |
2048 | ||
2049 | name -= 2; | |
2050 | while (*name) { | |
2051 | if (name == start) | |
2052 | return start; | |
2053 | name--; | |
2054 | } | |
2055 | return name + 1; | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * compare the old list of snapshots that we have to what's in the header | |
2060 | * and update it accordingly. Note that the header holds the snapshots | |
2061 | * in a reverse order (from newest to oldest) and we need to go from | |
2062 | * older to new so that we don't get a duplicate snap name when | |
2063 | * doing the process (e.g., removed snapshot and recreated a new | |
2064 | * one with the same name. | |
2065 | */ | |
2066 | static int __rbd_init_snaps_header(struct rbd_device *rbd_dev) | |
2067 | { | |
2068 | const char *name, *first_name; | |
2069 | int i = rbd_dev->header.total_snaps; | |
2070 | struct rbd_snap *snap, *old_snap = NULL; | |
2071 | int ret; | |
2072 | struct list_head *p, *n; | |
2073 | ||
2074 | first_name = rbd_dev->header.snap_names; | |
2075 | name = first_name + rbd_dev->header.snap_names_len; | |
2076 | ||
2077 | list_for_each_prev_safe(p, n, &rbd_dev->snaps) { | |
2078 | u64 cur_id; | |
2079 | ||
2080 | old_snap = list_entry(p, struct rbd_snap, node); | |
2081 | ||
2082 | if (i) | |
2083 | cur_id = rbd_dev->header.snapc->snaps[i - 1]; | |
2084 | ||
2085 | if (!i || old_snap->id < cur_id) { | |
2086 | /* old_snap->id was skipped, thus was removed */ | |
2087 | __rbd_remove_snap_dev(rbd_dev, old_snap); | |
2088 | continue; | |
2089 | } | |
2090 | if (old_snap->id == cur_id) { | |
2091 | /* we have this snapshot already */ | |
2092 | i--; | |
2093 | name = rbd_prev_snap_name(name, first_name); | |
2094 | continue; | |
2095 | } | |
2096 | for (; i > 0; | |
2097 | i--, name = rbd_prev_snap_name(name, first_name)) { | |
2098 | if (!name) { | |
2099 | WARN_ON(1); | |
2100 | return -EINVAL; | |
2101 | } | |
2102 | cur_id = rbd_dev->header.snapc->snaps[i]; | |
2103 | /* snapshot removal? handle it above */ | |
2104 | if (cur_id >= old_snap->id) | |
2105 | break; | |
2106 | /* a new snapshot */ | |
2107 | ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap); | |
2108 | if (ret < 0) | |
2109 | return ret; | |
2110 | ||
2111 | /* note that we add it backward so using n and not p */ | |
2112 | list_add(&snap->node, n); | |
2113 | p = &snap->node; | |
2114 | } | |
2115 | } | |
2116 | /* we're done going over the old snap list, just add what's left */ | |
2117 | for (; i > 0; i--) { | |
2118 | name = rbd_prev_snap_name(name, first_name); | |
2119 | if (!name) { | |
2120 | WARN_ON(1); | |
2121 | return -EINVAL; | |
2122 | } | |
2123 | ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap); | |
2124 | if (ret < 0) | |
2125 | return ret; | |
2126 | list_add(&snap->node, &rbd_dev->snaps); | |
2127 | } | |
2128 | ||
2129 | return 0; | |
2130 | } | |
2131 | ||
dfc5606d YS |
2132 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
2133 | { | |
f0f8cef5 | 2134 | int ret; |
dfc5606d YS |
2135 | struct device *dev; |
2136 | struct rbd_snap *snap; | |
2137 | ||
2138 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2139 | dev = &rbd_dev->dev; | |
2140 | ||
2141 | dev->bus = &rbd_bus_type; | |
2142 | dev->type = &rbd_device_type; | |
2143 | dev->parent = &rbd_root_dev; | |
2144 | dev->release = rbd_dev_release; | |
2145 | dev_set_name(dev, "%d", rbd_dev->id); | |
2146 | ret = device_register(dev); | |
2147 | if (ret < 0) | |
f0f8cef5 | 2148 | goto out; |
dfc5606d YS |
2149 | |
2150 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
2151 | ret = rbd_register_snap_dev(rbd_dev, snap, | |
2152 | &rbd_dev->dev); | |
2153 | if (ret < 0) | |
602adf40 YS |
2154 | break; |
2155 | } | |
f0f8cef5 | 2156 | out: |
dfc5606d YS |
2157 | mutex_unlock(&ctl_mutex); |
2158 | return ret; | |
602adf40 YS |
2159 | } |
2160 | ||
dfc5606d YS |
2161 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
2162 | { | |
2163 | device_unregister(&rbd_dev->dev); | |
2164 | } | |
2165 | ||
59c2be1e YS |
2166 | static int rbd_init_watch_dev(struct rbd_device *rbd_dev) |
2167 | { | |
2168 | int ret, rc; | |
2169 | ||
2170 | do { | |
2171 | ret = rbd_req_sync_watch(rbd_dev, rbd_dev->obj_md_name, | |
2172 | rbd_dev->header.obj_version); | |
2173 | if (ret == -ERANGE) { | |
2174 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2175 | rc = __rbd_update_snaps(rbd_dev); | |
2176 | mutex_unlock(&ctl_mutex); | |
2177 | if (rc < 0) | |
2178 | return rc; | |
2179 | } | |
2180 | } while (ret == -ERANGE); | |
2181 | ||
2182 | return ret; | |
2183 | } | |
2184 | ||
1ddbe94e AE |
2185 | static atomic64_t rbd_id_max = ATOMIC64_INIT(0); |
2186 | ||
2187 | /* | |
499afd5b AE |
2188 | * Get a unique rbd identifier for the given new rbd_dev, and add |
2189 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 2190 | */ |
499afd5b | 2191 | static void rbd_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 2192 | { |
499afd5b AE |
2193 | rbd_dev->id = atomic64_inc_return(&rbd_id_max); |
2194 | ||
2195 | spin_lock(&rbd_dev_list_lock); | |
2196 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
2197 | spin_unlock(&rbd_dev_list_lock); | |
1ddbe94e | 2198 | } |
b7f23c36 | 2199 | |
1ddbe94e | 2200 | /* |
499afd5b AE |
2201 | * Remove an rbd_dev from the global list, and record that its |
2202 | * identifier is no longer in use. | |
1ddbe94e | 2203 | */ |
499afd5b | 2204 | static void rbd_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 2205 | { |
d184f6bf AE |
2206 | struct list_head *tmp; |
2207 | int rbd_id = rbd_dev->id; | |
2208 | int max_id; | |
2209 | ||
2210 | BUG_ON(rbd_id < 1); | |
499afd5b AE |
2211 | |
2212 | spin_lock(&rbd_dev_list_lock); | |
2213 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
2214 | |
2215 | /* | |
2216 | * If the id being "put" is not the current maximum, there | |
2217 | * is nothing special we need to do. | |
2218 | */ | |
2219 | if (rbd_id != atomic64_read(&rbd_id_max)) { | |
2220 | spin_unlock(&rbd_dev_list_lock); | |
2221 | return; | |
2222 | } | |
2223 | ||
2224 | /* | |
2225 | * We need to update the current maximum id. Search the | |
2226 | * list to find out what it is. We're more likely to find | |
2227 | * the maximum at the end, so search the list backward. | |
2228 | */ | |
2229 | max_id = 0; | |
2230 | list_for_each_prev(tmp, &rbd_dev_list) { | |
2231 | struct rbd_device *rbd_dev; | |
2232 | ||
2233 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
2234 | if (rbd_id > max_id) | |
2235 | max_id = rbd_id; | |
2236 | } | |
499afd5b | 2237 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 2238 | |
1ddbe94e | 2239 | /* |
d184f6bf AE |
2240 | * The max id could have been updated by rbd_id_get(), in |
2241 | * which case it now accurately reflects the new maximum. | |
2242 | * Be careful not to overwrite the maximum value in that | |
2243 | * case. | |
1ddbe94e | 2244 | */ |
d184f6bf | 2245 | atomic64_cmpxchg(&rbd_id_max, rbd_id, max_id); |
b7f23c36 AE |
2246 | } |
2247 | ||
e28fff26 AE |
2248 | /* |
2249 | * Skips over white space at *buf, and updates *buf to point to the | |
2250 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
2251 | * the token (string of non-white space characters) found. Note |
2252 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
2253 | */ |
2254 | static inline size_t next_token(const char **buf) | |
2255 | { | |
2256 | /* | |
2257 | * These are the characters that produce nonzero for | |
2258 | * isspace() in the "C" and "POSIX" locales. | |
2259 | */ | |
2260 | const char *spaces = " \f\n\r\t\v"; | |
2261 | ||
2262 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
2263 | ||
2264 | return strcspn(*buf, spaces); /* Return token length */ | |
2265 | } | |
2266 | ||
2267 | /* | |
2268 | * Finds the next token in *buf, and if the provided token buffer is | |
2269 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
2270 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
2271 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
2272 | * |
2273 | * Returns the length of the token found (not including the '\0'). | |
2274 | * Return value will be 0 if no token is found, and it will be >= | |
2275 | * token_size if the token would not fit. | |
2276 | * | |
593a9e7b | 2277 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
2278 | * found token. Note that this occurs even if the token buffer is |
2279 | * too small to hold it. | |
2280 | */ | |
2281 | static inline size_t copy_token(const char **buf, | |
2282 | char *token, | |
2283 | size_t token_size) | |
2284 | { | |
2285 | size_t len; | |
2286 | ||
2287 | len = next_token(buf); | |
2288 | if (len < token_size) { | |
2289 | memcpy(token, *buf, len); | |
2290 | *(token + len) = '\0'; | |
2291 | } | |
2292 | *buf += len; | |
2293 | ||
2294 | return len; | |
2295 | } | |
2296 | ||
a725f65e AE |
2297 | /* |
2298 | * This fills in the pool_name, obj, obj_len, snap_name, obj_len, | |
2299 | * rbd_dev, rbd_md_name, and name fields of the given rbd_dev, based | |
2300 | * on the list of monitor addresses and other options provided via | |
2301 | * /sys/bus/rbd/add. | |
2302 | */ | |
2303 | static int rbd_add_parse_args(struct rbd_device *rbd_dev, | |
2304 | const char *buf, | |
7ef3214a | 2305 | const char **mon_addrs, |
5214ecc4 | 2306 | size_t *mon_addrs_size, |
e28fff26 AE |
2307 | char *options, |
2308 | size_t options_size) | |
2309 | { | |
2310 | size_t len; | |
2311 | ||
2312 | /* The first four tokens are required */ | |
2313 | ||
7ef3214a AE |
2314 | len = next_token(&buf); |
2315 | if (!len) | |
a725f65e | 2316 | return -EINVAL; |
5214ecc4 | 2317 | *mon_addrs_size = len + 1; |
7ef3214a AE |
2318 | *mon_addrs = buf; |
2319 | ||
2320 | buf += len; | |
a725f65e | 2321 | |
e28fff26 AE |
2322 | len = copy_token(&buf, options, options_size); |
2323 | if (!len || len >= options_size) | |
2324 | return -EINVAL; | |
2325 | ||
2326 | len = copy_token(&buf, rbd_dev->pool_name, sizeof (rbd_dev->pool_name)); | |
2327 | if (!len || len >= sizeof (rbd_dev->pool_name)) | |
2328 | return -EINVAL; | |
2329 | ||
2330 | len = copy_token(&buf, rbd_dev->obj, sizeof (rbd_dev->obj)); | |
2331 | if (!len || len >= sizeof (rbd_dev->obj)) | |
2332 | return -EINVAL; | |
2333 | ||
2334 | /* We have the object length in hand, save it. */ | |
2335 | ||
2336 | rbd_dev->obj_len = len; | |
a725f65e | 2337 | |
81a89793 AE |
2338 | BUILD_BUG_ON(RBD_MAX_MD_NAME_LEN |
2339 | < RBD_MAX_OBJ_NAME_LEN + sizeof (RBD_SUFFIX)); | |
2340 | sprintf(rbd_dev->obj_md_name, "%s%s", rbd_dev->obj, RBD_SUFFIX); | |
a725f65e | 2341 | |
e28fff26 AE |
2342 | /* |
2343 | * The snapshot name is optional, but it's an error if it's | |
2344 | * too long. If no snapshot is supplied, fill in the default. | |
2345 | */ | |
2346 | len = copy_token(&buf, rbd_dev->snap_name, sizeof (rbd_dev->snap_name)); | |
2347 | if (!len) | |
2348 | memcpy(rbd_dev->snap_name, RBD_SNAP_HEAD_NAME, | |
2349 | sizeof (RBD_SNAP_HEAD_NAME)); | |
2350 | else if (len >= sizeof (rbd_dev->snap_name)) | |
2351 | return -EINVAL; | |
2352 | ||
a725f65e AE |
2353 | return 0; |
2354 | } | |
2355 | ||
59c2be1e YS |
2356 | static ssize_t rbd_add(struct bus_type *bus, |
2357 | const char *buf, | |
2358 | size_t count) | |
602adf40 | 2359 | { |
602adf40 | 2360 | struct rbd_device *rbd_dev; |
7ef3214a AE |
2361 | const char *mon_addrs = NULL; |
2362 | size_t mon_addrs_size = 0; | |
27cc2594 AE |
2363 | char *options = NULL; |
2364 | struct ceph_osd_client *osdc; | |
2365 | int rc = -ENOMEM; | |
602adf40 YS |
2366 | |
2367 | if (!try_module_get(THIS_MODULE)) | |
2368 | return -ENODEV; | |
2369 | ||
27cc2594 AE |
2370 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
2371 | if (!rbd_dev) | |
2372 | goto err_nomem; | |
60571c7d | 2373 | options = kmalloc(count, GFP_KERNEL); |
602adf40 | 2374 | if (!options) |
27cc2594 | 2375 | goto err_nomem; |
602adf40 YS |
2376 | |
2377 | /* static rbd_device initialization */ | |
2378 | spin_lock_init(&rbd_dev->lock); | |
2379 | INIT_LIST_HEAD(&rbd_dev->node); | |
dfc5606d | 2380 | INIT_LIST_HEAD(&rbd_dev->snaps); |
c666601a | 2381 | init_rwsem(&rbd_dev->header_rwsem); |
602adf40 | 2382 | |
c666601a | 2383 | init_rwsem(&rbd_dev->header_rwsem); |
0e805a1d | 2384 | |
d184f6bf | 2385 | /* generate unique id: find highest unique id, add one */ |
499afd5b | 2386 | rbd_id_get(rbd_dev); |
602adf40 | 2387 | |
a725f65e | 2388 | /* Fill in the device name, now that we have its id. */ |
81a89793 AE |
2389 | BUILD_BUG_ON(DEV_NAME_LEN |
2390 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
2391 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->id); | |
a725f65e | 2392 | |
602adf40 | 2393 | /* parse add command */ |
7ef3214a | 2394 | rc = rbd_add_parse_args(rbd_dev, buf, &mon_addrs, &mon_addrs_size, |
e28fff26 | 2395 | options, count); |
a725f65e | 2396 | if (rc) |
f0f8cef5 | 2397 | goto err_put_id; |
e124a82f | 2398 | |
5214ecc4 AE |
2399 | rbd_dev->rbd_client = rbd_get_client(mon_addrs, mon_addrs_size - 1, |
2400 | options); | |
d720bcb0 AE |
2401 | if (IS_ERR(rbd_dev->rbd_client)) { |
2402 | rc = PTR_ERR(rbd_dev->rbd_client); | |
f0f8cef5 | 2403 | goto err_put_id; |
d720bcb0 | 2404 | } |
602adf40 | 2405 | |
602adf40 | 2406 | /* pick the pool */ |
1dbb4399 | 2407 | osdc = &rbd_dev->rbd_client->client->osdc; |
602adf40 YS |
2408 | rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name); |
2409 | if (rc < 0) | |
2410 | goto err_out_client; | |
2411 | rbd_dev->poolid = rc; | |
2412 | ||
2413 | /* register our block device */ | |
27cc2594 AE |
2414 | rc = register_blkdev(0, rbd_dev->name); |
2415 | if (rc < 0) | |
602adf40 | 2416 | goto err_out_client; |
27cc2594 | 2417 | rbd_dev->major = rc; |
602adf40 | 2418 | |
dfc5606d YS |
2419 | rc = rbd_bus_add_dev(rbd_dev); |
2420 | if (rc) | |
766fc439 YS |
2421 | goto err_out_blkdev; |
2422 | ||
32eec68d AE |
2423 | /* |
2424 | * At this point cleanup in the event of an error is the job | |
2425 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
2426 | * | |
2427 | * Set up and announce blkdev mapping. | |
2428 | */ | |
602adf40 YS |
2429 | rc = rbd_init_disk(rbd_dev); |
2430 | if (rc) | |
766fc439 | 2431 | goto err_out_bus; |
602adf40 | 2432 | |
59c2be1e YS |
2433 | rc = rbd_init_watch_dev(rbd_dev); |
2434 | if (rc) | |
2435 | goto err_out_bus; | |
2436 | ||
602adf40 YS |
2437 | return count; |
2438 | ||
766fc439 | 2439 | err_out_bus: |
766fc439 YS |
2440 | /* this will also clean up rest of rbd_dev stuff */ |
2441 | ||
2442 | rbd_bus_del_dev(rbd_dev); | |
2443 | kfree(options); | |
766fc439 YS |
2444 | return rc; |
2445 | ||
602adf40 YS |
2446 | err_out_blkdev: |
2447 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
2448 | err_out_client: | |
2449 | rbd_put_client(rbd_dev); | |
f0f8cef5 | 2450 | err_put_id: |
499afd5b | 2451 | rbd_id_put(rbd_dev); |
27cc2594 | 2452 | err_nomem: |
602adf40 | 2453 | kfree(options); |
27cc2594 AE |
2454 | kfree(rbd_dev); |
2455 | ||
602adf40 YS |
2456 | dout("Error adding device %s\n", buf); |
2457 | module_put(THIS_MODULE); | |
27cc2594 AE |
2458 | |
2459 | return (ssize_t) rc; | |
602adf40 YS |
2460 | } |
2461 | ||
2462 | static struct rbd_device *__rbd_get_dev(unsigned long id) | |
2463 | { | |
2464 | struct list_head *tmp; | |
2465 | struct rbd_device *rbd_dev; | |
2466 | ||
e124a82f | 2467 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
2468 | list_for_each(tmp, &rbd_dev_list) { |
2469 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
e124a82f AE |
2470 | if (rbd_dev->id == id) { |
2471 | spin_unlock(&rbd_dev_list_lock); | |
602adf40 | 2472 | return rbd_dev; |
e124a82f | 2473 | } |
602adf40 | 2474 | } |
e124a82f | 2475 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
2476 | return NULL; |
2477 | } | |
2478 | ||
dfc5606d | 2479 | static void rbd_dev_release(struct device *dev) |
602adf40 | 2480 | { |
593a9e7b | 2481 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2482 | |
1dbb4399 AE |
2483 | if (rbd_dev->watch_request) { |
2484 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
2485 | ||
2486 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 2487 | rbd_dev->watch_request); |
1dbb4399 | 2488 | } |
59c2be1e | 2489 | if (rbd_dev->watch_event) |
79e3057c | 2490 | rbd_req_sync_unwatch(rbd_dev, rbd_dev->obj_md_name); |
59c2be1e | 2491 | |
602adf40 YS |
2492 | rbd_put_client(rbd_dev); |
2493 | ||
2494 | /* clean up and free blkdev */ | |
2495 | rbd_free_disk(rbd_dev); | |
2496 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d AE |
2497 | |
2498 | /* done with the id, and with the rbd_dev */ | |
2499 | rbd_id_put(rbd_dev); | |
602adf40 YS |
2500 | kfree(rbd_dev); |
2501 | ||
2502 | /* release module ref */ | |
2503 | module_put(THIS_MODULE); | |
602adf40 YS |
2504 | } |
2505 | ||
dfc5606d YS |
2506 | static ssize_t rbd_remove(struct bus_type *bus, |
2507 | const char *buf, | |
2508 | size_t count) | |
602adf40 YS |
2509 | { |
2510 | struct rbd_device *rbd_dev = NULL; | |
2511 | int target_id, rc; | |
2512 | unsigned long ul; | |
2513 | int ret = count; | |
2514 | ||
2515 | rc = strict_strtoul(buf, 10, &ul); | |
2516 | if (rc) | |
2517 | return rc; | |
2518 | ||
2519 | /* convert to int; abort if we lost anything in the conversion */ | |
2520 | target_id = (int) ul; | |
2521 | if (target_id != ul) | |
2522 | return -EINVAL; | |
2523 | ||
2524 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2525 | ||
2526 | rbd_dev = __rbd_get_dev(target_id); | |
2527 | if (!rbd_dev) { | |
2528 | ret = -ENOENT; | |
2529 | goto done; | |
2530 | } | |
2531 | ||
dfc5606d YS |
2532 | __rbd_remove_all_snaps(rbd_dev); |
2533 | rbd_bus_del_dev(rbd_dev); | |
602adf40 YS |
2534 | |
2535 | done: | |
2536 | mutex_unlock(&ctl_mutex); | |
2537 | return ret; | |
2538 | } | |
2539 | ||
dfc5606d YS |
2540 | static ssize_t rbd_snap_add(struct device *dev, |
2541 | struct device_attribute *attr, | |
2542 | const char *buf, | |
2543 | size_t count) | |
602adf40 | 2544 | { |
593a9e7b | 2545 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d YS |
2546 | int ret; |
2547 | char *name = kmalloc(count + 1, GFP_KERNEL); | |
602adf40 YS |
2548 | if (!name) |
2549 | return -ENOMEM; | |
2550 | ||
dfc5606d | 2551 | snprintf(name, count, "%s", buf); |
602adf40 YS |
2552 | |
2553 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
2554 | ||
602adf40 YS |
2555 | ret = rbd_header_add_snap(rbd_dev, |
2556 | name, GFP_KERNEL); | |
2557 | if (ret < 0) | |
59c2be1e | 2558 | goto err_unlock; |
602adf40 | 2559 | |
dfc5606d | 2560 | ret = __rbd_update_snaps(rbd_dev); |
602adf40 | 2561 | if (ret < 0) |
59c2be1e YS |
2562 | goto err_unlock; |
2563 | ||
2564 | /* shouldn't hold ctl_mutex when notifying.. notify might | |
2565 | trigger a watch callback that would need to get that mutex */ | |
2566 | mutex_unlock(&ctl_mutex); | |
2567 | ||
2568 | /* make a best effort, don't error if failed */ | |
2569 | rbd_req_sync_notify(rbd_dev, rbd_dev->obj_md_name); | |
602adf40 YS |
2570 | |
2571 | ret = count; | |
59c2be1e YS |
2572 | kfree(name); |
2573 | return ret; | |
2574 | ||
2575 | err_unlock: | |
602adf40 | 2576 | mutex_unlock(&ctl_mutex); |
602adf40 YS |
2577 | kfree(name); |
2578 | return ret; | |
2579 | } | |
2580 | ||
602adf40 YS |
2581 | /* |
2582 | * create control files in sysfs | |
dfc5606d | 2583 | * /sys/bus/rbd/... |
602adf40 YS |
2584 | */ |
2585 | static int rbd_sysfs_init(void) | |
2586 | { | |
dfc5606d | 2587 | int ret; |
602adf40 | 2588 | |
fed4c143 | 2589 | ret = device_register(&rbd_root_dev); |
21079786 | 2590 | if (ret < 0) |
dfc5606d | 2591 | return ret; |
602adf40 | 2592 | |
fed4c143 AE |
2593 | ret = bus_register(&rbd_bus_type); |
2594 | if (ret < 0) | |
2595 | device_unregister(&rbd_root_dev); | |
602adf40 | 2596 | |
602adf40 YS |
2597 | return ret; |
2598 | } | |
2599 | ||
2600 | static void rbd_sysfs_cleanup(void) | |
2601 | { | |
dfc5606d | 2602 | bus_unregister(&rbd_bus_type); |
fed4c143 | 2603 | device_unregister(&rbd_root_dev); |
602adf40 YS |
2604 | } |
2605 | ||
2606 | int __init rbd_init(void) | |
2607 | { | |
2608 | int rc; | |
2609 | ||
2610 | rc = rbd_sysfs_init(); | |
2611 | if (rc) | |
2612 | return rc; | |
f0f8cef5 | 2613 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
2614 | return 0; |
2615 | } | |
2616 | ||
2617 | void __exit rbd_exit(void) | |
2618 | { | |
2619 | rbd_sysfs_cleanup(); | |
2620 | } | |
2621 | ||
2622 | module_init(rbd_init); | |
2623 | module_exit(rbd_exit); | |
2624 | ||
2625 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
2626 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
2627 | MODULE_DESCRIPTION("rados block device"); | |
2628 | ||
2629 | /* following authorship retained from original osdblk.c */ | |
2630 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
2631 | ||
2632 | MODULE_LICENSE("GPL"); |