Commit | Line | Data |
---|---|---|
602adf40 YS |
1 | /* |
2 | rbd.c -- Export ceph rados objects as a Linux block device | |
3 | ||
4 | ||
5 | based on drivers/block/osdblk.c: | |
6 | ||
7 | Copyright 2009 Red Hat, Inc. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; see the file COPYING. If not, write to | |
20 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | ||
22 | ||
23 | ||
dfc5606d | 24 | For usage instructions, please refer to: |
602adf40 | 25 | |
dfc5606d | 26 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
27 | |
28 | */ | |
29 | ||
30 | #include <linux/ceph/libceph.h> | |
31 | #include <linux/ceph/osd_client.h> | |
32 | #include <linux/ceph/mon_client.h> | |
33 | #include <linux/ceph/decode.h> | |
59c2be1e | 34 | #include <linux/parser.h> |
602adf40 YS |
35 | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/device.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/blkdev.h> | |
41 | ||
42 | #include "rbd_types.h" | |
43 | ||
aafb230e AE |
44 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
45 | ||
593a9e7b AE |
46 | /* |
47 | * The basic unit of block I/O is a sector. It is interpreted in a | |
48 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
49 | * universally 512 bytes. These symbols are just slightly more | |
50 | * meaningful than the bare numbers they represent. | |
51 | */ | |
52 | #define SECTOR_SHIFT 9 | |
53 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
54 | ||
2647ba38 | 55 | /* It might be useful to have these defined elsewhere */ |
df111be6 | 56 | |
2647ba38 AE |
57 | #define U8_MAX ((u8) (~0U)) |
58 | #define U16_MAX ((u16) (~0U)) | |
59 | #define U32_MAX ((u32) (~0U)) | |
60 | #define U64_MAX ((u64) (~0ULL)) | |
df111be6 | 61 | |
f0f8cef5 AE |
62 | #define RBD_DRV_NAME "rbd" |
63 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
64 | |
65 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
66 | ||
d4b125e9 AE |
67 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
68 | #define RBD_MAX_SNAP_NAME_LEN \ | |
69 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
70 | ||
35d489f9 | 71 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
72 | |
73 | #define RBD_SNAP_HEAD_NAME "-" | |
74 | ||
9e15b77d AE |
75 | /* This allows a single page to hold an image name sent by OSD */ |
76 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 77 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 78 | |
1e130199 | 79 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 80 | |
d889140c AE |
81 | /* Feature bits */ |
82 | ||
83 | #define RBD_FEATURE_LAYERING 1 | |
84 | ||
85 | /* Features supported by this (client software) implementation. */ | |
86 | ||
87 | #define RBD_FEATURES_ALL (0) | |
88 | ||
81a89793 AE |
89 | /* |
90 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
91 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
92 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
93 | * enough to hold all possible device names. | |
94 | */ | |
602adf40 | 95 | #define DEV_NAME_LEN 32 |
81a89793 | 96 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
97 | |
98 | /* | |
99 | * block device image metadata (in-memory version) | |
100 | */ | |
101 | struct rbd_image_header { | |
f84344f3 | 102 | /* These four fields never change for a given rbd image */ |
849b4260 | 103 | char *object_prefix; |
34b13184 | 104 | u64 features; |
602adf40 YS |
105 | __u8 obj_order; |
106 | __u8 crypt_type; | |
107 | __u8 comp_type; | |
602adf40 | 108 | |
f84344f3 AE |
109 | /* The remaining fields need to be updated occasionally */ |
110 | u64 image_size; | |
111 | struct ceph_snap_context *snapc; | |
602adf40 YS |
112 | char *snap_names; |
113 | u64 *snap_sizes; | |
59c2be1e YS |
114 | |
115 | u64 obj_version; | |
116 | }; | |
117 | ||
0d7dbfce AE |
118 | /* |
119 | * An rbd image specification. | |
120 | * | |
121 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
122 | * identify an image. Each rbd_dev structure includes a pointer to |
123 | * an rbd_spec structure that encapsulates this identity. | |
124 | * | |
125 | * Each of the id's in an rbd_spec has an associated name. For a | |
126 | * user-mapped image, the names are supplied and the id's associated | |
127 | * with them are looked up. For a layered image, a parent image is | |
128 | * defined by the tuple, and the names are looked up. | |
129 | * | |
130 | * An rbd_dev structure contains a parent_spec pointer which is | |
131 | * non-null if the image it represents is a child in a layered | |
132 | * image. This pointer will refer to the rbd_spec structure used | |
133 | * by the parent rbd_dev for its own identity (i.e., the structure | |
134 | * is shared between the parent and child). | |
135 | * | |
136 | * Since these structures are populated once, during the discovery | |
137 | * phase of image construction, they are effectively immutable so | |
138 | * we make no effort to synchronize access to them. | |
139 | * | |
140 | * Note that code herein does not assume the image name is known (it | |
141 | * could be a null pointer). | |
0d7dbfce AE |
142 | */ |
143 | struct rbd_spec { | |
144 | u64 pool_id; | |
145 | char *pool_name; | |
146 | ||
147 | char *image_id; | |
0d7dbfce | 148 | char *image_name; |
0d7dbfce AE |
149 | |
150 | u64 snap_id; | |
151 | char *snap_name; | |
152 | ||
153 | struct kref kref; | |
154 | }; | |
155 | ||
602adf40 | 156 | /* |
f0f8cef5 | 157 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
158 | */ |
159 | struct rbd_client { | |
160 | struct ceph_client *client; | |
161 | struct kref kref; | |
162 | struct list_head node; | |
163 | }; | |
164 | ||
165 | /* | |
f0f8cef5 | 166 | * a request completion status |
602adf40 | 167 | */ |
1fec7093 YS |
168 | struct rbd_req_status { |
169 | int done; | |
8986cb37 | 170 | s32 rc; |
1fec7093 YS |
171 | u64 bytes; |
172 | }; | |
173 | ||
174 | /* | |
175 | * a collection of requests | |
176 | */ | |
177 | struct rbd_req_coll { | |
178 | int total; | |
179 | int num_done; | |
180 | struct kref kref; | |
181 | struct rbd_req_status status[0]; | |
602adf40 YS |
182 | }; |
183 | ||
f0f8cef5 AE |
184 | /* |
185 | * a single io request | |
186 | */ | |
187 | struct rbd_request { | |
188 | struct request *rq; /* blk layer request */ | |
189 | struct bio *bio; /* cloned bio */ | |
190 | struct page **pages; /* list of used pages */ | |
191 | u64 len; | |
192 | int coll_index; | |
193 | struct rbd_req_coll *coll; | |
194 | }; | |
195 | ||
dfc5606d YS |
196 | struct rbd_snap { |
197 | struct device dev; | |
198 | const char *name; | |
3591538f | 199 | u64 size; |
dfc5606d YS |
200 | struct list_head node; |
201 | u64 id; | |
34b13184 | 202 | u64 features; |
dfc5606d YS |
203 | }; |
204 | ||
f84344f3 | 205 | struct rbd_mapping { |
99c1f08f | 206 | u64 size; |
34b13184 | 207 | u64 features; |
f84344f3 AE |
208 | bool read_only; |
209 | }; | |
210 | ||
602adf40 YS |
211 | /* |
212 | * a single device | |
213 | */ | |
214 | struct rbd_device { | |
de71a297 | 215 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
216 | |
217 | int major; /* blkdev assigned major */ | |
218 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 219 | |
a30b71b9 | 220 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
221 | struct rbd_client *rbd_client; |
222 | ||
223 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
224 | ||
225 | spinlock_t lock; /* queue lock */ | |
226 | ||
227 | struct rbd_image_header header; | |
d78b650a | 228 | atomic_t exists; |
0d7dbfce | 229 | struct rbd_spec *spec; |
602adf40 | 230 | |
0d7dbfce | 231 | char *header_name; |
971f839a | 232 | |
0903e875 AE |
233 | struct ceph_file_layout layout; |
234 | ||
59c2be1e YS |
235 | struct ceph_osd_event *watch_event; |
236 | struct ceph_osd_request *watch_request; | |
237 | ||
86b00e0d AE |
238 | struct rbd_spec *parent_spec; |
239 | u64 parent_overlap; | |
240 | ||
c666601a JD |
241 | /* protects updating the header */ |
242 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
243 | |
244 | struct rbd_mapping mapping; | |
602adf40 YS |
245 | |
246 | struct list_head node; | |
dfc5606d YS |
247 | |
248 | /* list of snapshots */ | |
249 | struct list_head snaps; | |
250 | ||
251 | /* sysfs related */ | |
252 | struct device dev; | |
42382b70 | 253 | unsigned long open_count; |
dfc5606d YS |
254 | }; |
255 | ||
602adf40 | 256 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 257 | |
602adf40 | 258 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
259 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
260 | ||
432b8587 AE |
261 | static LIST_HEAD(rbd_client_list); /* clients */ |
262 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 263 | |
304f6808 AE |
264 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
265 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev); | |
266 | ||
dfc5606d | 267 | static void rbd_dev_release(struct device *dev); |
41f38c2b | 268 | static void rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 269 | |
f0f8cef5 AE |
270 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
271 | size_t count); | |
272 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
273 | size_t count); | |
274 | ||
275 | static struct bus_attribute rbd_bus_attrs[] = { | |
276 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
277 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
278 | __ATTR_NULL | |
279 | }; | |
280 | ||
281 | static struct bus_type rbd_bus_type = { | |
282 | .name = "rbd", | |
283 | .bus_attrs = rbd_bus_attrs, | |
284 | }; | |
285 | ||
286 | static void rbd_root_dev_release(struct device *dev) | |
287 | { | |
288 | } | |
289 | ||
290 | static struct device rbd_root_dev = { | |
291 | .init_name = "rbd", | |
292 | .release = rbd_root_dev_release, | |
293 | }; | |
294 | ||
06ecc6cb AE |
295 | static __printf(2, 3) |
296 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
297 | { | |
298 | struct va_format vaf; | |
299 | va_list args; | |
300 | ||
301 | va_start(args, fmt); | |
302 | vaf.fmt = fmt; | |
303 | vaf.va = &args; | |
304 | ||
305 | if (!rbd_dev) | |
306 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
307 | else if (rbd_dev->disk) | |
308 | printk(KERN_WARNING "%s: %s: %pV\n", | |
309 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
310 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
311 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
312 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
313 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
314 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
315 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
316 | else /* punt */ | |
317 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
318 | RBD_DRV_NAME, rbd_dev, &vaf); | |
319 | va_end(args); | |
320 | } | |
321 | ||
aafb230e AE |
322 | #ifdef RBD_DEBUG |
323 | #define rbd_assert(expr) \ | |
324 | if (unlikely(!(expr))) { \ | |
325 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
326 | "at line %d:\n\n" \ | |
327 | "\trbd_assert(%s);\n\n", \ | |
328 | __func__, __LINE__, #expr); \ | |
329 | BUG(); \ | |
330 | } | |
331 | #else /* !RBD_DEBUG */ | |
332 | # define rbd_assert(expr) ((void) 0) | |
333 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 334 | |
117973fb AE |
335 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
336 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 337 | |
602adf40 YS |
338 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
339 | { | |
f0f8cef5 | 340 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
602adf40 | 341 | |
f84344f3 | 342 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
343 | return -EROFS; |
344 | ||
42382b70 | 345 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 346 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 347 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 AE |
348 | rbd_dev->open_count++; |
349 | mutex_unlock(&ctl_mutex); | |
340c7a2b | 350 | |
602adf40 YS |
351 | return 0; |
352 | } | |
353 | ||
dfc5606d YS |
354 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
355 | { | |
356 | struct rbd_device *rbd_dev = disk->private_data; | |
357 | ||
42382b70 AE |
358 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
359 | rbd_assert(rbd_dev->open_count > 0); | |
360 | rbd_dev->open_count--; | |
c3e946ce | 361 | put_device(&rbd_dev->dev); |
42382b70 | 362 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
363 | |
364 | return 0; | |
365 | } | |
366 | ||
602adf40 YS |
367 | static const struct block_device_operations rbd_bd_ops = { |
368 | .owner = THIS_MODULE, | |
369 | .open = rbd_open, | |
dfc5606d | 370 | .release = rbd_release, |
602adf40 YS |
371 | }; |
372 | ||
373 | /* | |
374 | * Initialize an rbd client instance. | |
43ae4701 | 375 | * We own *ceph_opts. |
602adf40 | 376 | */ |
f8c38929 | 377 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
378 | { |
379 | struct rbd_client *rbdc; | |
380 | int ret = -ENOMEM; | |
381 | ||
382 | dout("rbd_client_create\n"); | |
383 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); | |
384 | if (!rbdc) | |
385 | goto out_opt; | |
386 | ||
387 | kref_init(&rbdc->kref); | |
388 | INIT_LIST_HEAD(&rbdc->node); | |
389 | ||
bc534d86 AE |
390 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
391 | ||
43ae4701 | 392 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 393 | if (IS_ERR(rbdc->client)) |
bc534d86 | 394 | goto out_mutex; |
43ae4701 | 395 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
396 | |
397 | ret = ceph_open_session(rbdc->client); | |
398 | if (ret < 0) | |
399 | goto out_err; | |
400 | ||
432b8587 | 401 | spin_lock(&rbd_client_list_lock); |
602adf40 | 402 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 403 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 404 | |
bc534d86 AE |
405 | mutex_unlock(&ctl_mutex); |
406 | ||
602adf40 YS |
407 | dout("rbd_client_create created %p\n", rbdc); |
408 | return rbdc; | |
409 | ||
410 | out_err: | |
411 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
412 | out_mutex: |
413 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
414 | kfree(rbdc); |
415 | out_opt: | |
43ae4701 AE |
416 | if (ceph_opts) |
417 | ceph_destroy_options(ceph_opts); | |
28f259b7 | 418 | return ERR_PTR(ret); |
602adf40 YS |
419 | } |
420 | ||
421 | /* | |
1f7ba331 AE |
422 | * Find a ceph client with specific addr and configuration. If |
423 | * found, bump its reference count. | |
602adf40 | 424 | */ |
1f7ba331 | 425 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
426 | { |
427 | struct rbd_client *client_node; | |
1f7ba331 | 428 | bool found = false; |
602adf40 | 429 | |
43ae4701 | 430 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
431 | return NULL; |
432 | ||
1f7ba331 AE |
433 | spin_lock(&rbd_client_list_lock); |
434 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
435 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
436 | kref_get(&client_node->kref); | |
437 | found = true; | |
438 | break; | |
439 | } | |
440 | } | |
441 | spin_unlock(&rbd_client_list_lock); | |
442 | ||
443 | return found ? client_node : NULL; | |
602adf40 YS |
444 | } |
445 | ||
59c2be1e YS |
446 | /* |
447 | * mount options | |
448 | */ | |
449 | enum { | |
59c2be1e YS |
450 | Opt_last_int, |
451 | /* int args above */ | |
452 | Opt_last_string, | |
453 | /* string args above */ | |
cc0538b6 AE |
454 | Opt_read_only, |
455 | Opt_read_write, | |
456 | /* Boolean args above */ | |
457 | Opt_last_bool, | |
59c2be1e YS |
458 | }; |
459 | ||
43ae4701 | 460 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
461 | /* int args above */ |
462 | /* string args above */ | |
be466c1c | 463 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
464 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
465 | {Opt_read_write, "read_write"}, | |
466 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
467 | /* Boolean args above */ | |
59c2be1e YS |
468 | {-1, NULL} |
469 | }; | |
470 | ||
98571b5a AE |
471 | struct rbd_options { |
472 | bool read_only; | |
473 | }; | |
474 | ||
475 | #define RBD_READ_ONLY_DEFAULT false | |
476 | ||
59c2be1e YS |
477 | static int parse_rbd_opts_token(char *c, void *private) |
478 | { | |
43ae4701 | 479 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
480 | substring_t argstr[MAX_OPT_ARGS]; |
481 | int token, intval, ret; | |
482 | ||
43ae4701 | 483 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
484 | if (token < 0) |
485 | return -EINVAL; | |
486 | ||
487 | if (token < Opt_last_int) { | |
488 | ret = match_int(&argstr[0], &intval); | |
489 | if (ret < 0) { | |
490 | pr_err("bad mount option arg (not int) " | |
491 | "at '%s'\n", c); | |
492 | return ret; | |
493 | } | |
494 | dout("got int token %d val %d\n", token, intval); | |
495 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
496 | dout("got string token %d val %s\n", token, | |
497 | argstr[0].from); | |
cc0538b6 AE |
498 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
499 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
500 | } else { |
501 | dout("got token %d\n", token); | |
502 | } | |
503 | ||
504 | switch (token) { | |
cc0538b6 AE |
505 | case Opt_read_only: |
506 | rbd_opts->read_only = true; | |
507 | break; | |
508 | case Opt_read_write: | |
509 | rbd_opts->read_only = false; | |
510 | break; | |
59c2be1e | 511 | default: |
aafb230e AE |
512 | rbd_assert(false); |
513 | break; | |
59c2be1e YS |
514 | } |
515 | return 0; | |
516 | } | |
517 | ||
602adf40 YS |
518 | /* |
519 | * Get a ceph client with specific addr and configuration, if one does | |
520 | * not exist create it. | |
521 | */ | |
9d3997fd | 522 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 523 | { |
f8c38929 | 524 | struct rbd_client *rbdc; |
59c2be1e | 525 | |
1f7ba331 | 526 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 527 | if (rbdc) /* using an existing client */ |
43ae4701 | 528 | ceph_destroy_options(ceph_opts); |
9d3997fd | 529 | else |
f8c38929 | 530 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 531 | |
9d3997fd | 532 | return rbdc; |
602adf40 YS |
533 | } |
534 | ||
535 | /* | |
536 | * Destroy ceph client | |
d23a4b3f | 537 | * |
432b8587 | 538 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
539 | */ |
540 | static void rbd_client_release(struct kref *kref) | |
541 | { | |
542 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
543 | ||
544 | dout("rbd_release_client %p\n", rbdc); | |
cd9d9f5d | 545 | spin_lock(&rbd_client_list_lock); |
602adf40 | 546 | list_del(&rbdc->node); |
cd9d9f5d | 547 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
548 | |
549 | ceph_destroy_client(rbdc->client); | |
550 | kfree(rbdc); | |
551 | } | |
552 | ||
553 | /* | |
554 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
555 | * it. | |
556 | */ | |
9d3997fd | 557 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 558 | { |
c53d5893 AE |
559 | if (rbdc) |
560 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
561 | } |
562 | ||
1fec7093 YS |
563 | /* |
564 | * Destroy requests collection | |
565 | */ | |
566 | static void rbd_coll_release(struct kref *kref) | |
567 | { | |
568 | struct rbd_req_coll *coll = | |
569 | container_of(kref, struct rbd_req_coll, kref); | |
570 | ||
571 | dout("rbd_coll_release %p\n", coll); | |
572 | kfree(coll); | |
573 | } | |
602adf40 | 574 | |
a30b71b9 AE |
575 | static bool rbd_image_format_valid(u32 image_format) |
576 | { | |
577 | return image_format == 1 || image_format == 2; | |
578 | } | |
579 | ||
8e94af8e AE |
580 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
581 | { | |
103a150f AE |
582 | size_t size; |
583 | u32 snap_count; | |
584 | ||
585 | /* The header has to start with the magic rbd header text */ | |
586 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
587 | return false; | |
588 | ||
db2388b6 AE |
589 | /* The bio layer requires at least sector-sized I/O */ |
590 | ||
591 | if (ondisk->options.order < SECTOR_SHIFT) | |
592 | return false; | |
593 | ||
594 | /* If we use u64 in a few spots we may be able to loosen this */ | |
595 | ||
596 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
597 | return false; | |
598 | ||
103a150f AE |
599 | /* |
600 | * The size of a snapshot header has to fit in a size_t, and | |
601 | * that limits the number of snapshots. | |
602 | */ | |
603 | snap_count = le32_to_cpu(ondisk->snap_count); | |
604 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
605 | if (snap_count > size / sizeof (__le64)) | |
606 | return false; | |
607 | ||
608 | /* | |
609 | * Not only that, but the size of the entire the snapshot | |
610 | * header must also be representable in a size_t. | |
611 | */ | |
612 | size -= snap_count * sizeof (__le64); | |
613 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
614 | return false; | |
615 | ||
616 | return true; | |
8e94af8e AE |
617 | } |
618 | ||
602adf40 YS |
619 | /* |
620 | * Create a new header structure, translate header format from the on-disk | |
621 | * header. | |
622 | */ | |
623 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 624 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 625 | { |
ccece235 | 626 | u32 snap_count; |
58c17b0e | 627 | size_t len; |
d2bb24e5 | 628 | size_t size; |
621901d6 | 629 | u32 i; |
602adf40 | 630 | |
6a52325f AE |
631 | memset(header, 0, sizeof (*header)); |
632 | ||
103a150f AE |
633 | snap_count = le32_to_cpu(ondisk->snap_count); |
634 | ||
58c17b0e AE |
635 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
636 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 637 | if (!header->object_prefix) |
602adf40 | 638 | return -ENOMEM; |
58c17b0e AE |
639 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
640 | header->object_prefix[len] = '\0'; | |
00f1f36f | 641 | |
602adf40 | 642 | if (snap_count) { |
f785cc1d AE |
643 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
644 | ||
621901d6 AE |
645 | /* Save a copy of the snapshot names */ |
646 | ||
f785cc1d AE |
647 | if (snap_names_len > (u64) SIZE_MAX) |
648 | return -EIO; | |
649 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 650 | if (!header->snap_names) |
6a52325f | 651 | goto out_err; |
f785cc1d AE |
652 | /* |
653 | * Note that rbd_dev_v1_header_read() guarantees | |
654 | * the ondisk buffer we're working with has | |
655 | * snap_names_len bytes beyond the end of the | |
656 | * snapshot id array, this memcpy() is safe. | |
657 | */ | |
658 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
659 | snap_names_len); | |
6a52325f | 660 | |
621901d6 AE |
661 | /* Record each snapshot's size */ |
662 | ||
d2bb24e5 AE |
663 | size = snap_count * sizeof (*header->snap_sizes); |
664 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 665 | if (!header->snap_sizes) |
6a52325f | 666 | goto out_err; |
621901d6 AE |
667 | for (i = 0; i < snap_count; i++) |
668 | header->snap_sizes[i] = | |
669 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 670 | } else { |
ccece235 | 671 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
672 | header->snap_names = NULL; |
673 | header->snap_sizes = NULL; | |
674 | } | |
849b4260 | 675 | |
34b13184 | 676 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
677 | header->obj_order = ondisk->options.order; |
678 | header->crypt_type = ondisk->options.crypt_type; | |
679 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 680 | |
621901d6 AE |
681 | /* Allocate and fill in the snapshot context */ |
682 | ||
f84344f3 | 683 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
684 | size = sizeof (struct ceph_snap_context); |
685 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
686 | header->snapc = kzalloc(size, GFP_KERNEL); | |
687 | if (!header->snapc) | |
688 | goto out_err; | |
602adf40 YS |
689 | |
690 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 691 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 692 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
693 | for (i = 0; i < snap_count; i++) |
694 | header->snapc->snaps[i] = | |
695 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
696 | |
697 | return 0; | |
698 | ||
6a52325f | 699 | out_err: |
849b4260 | 700 | kfree(header->snap_sizes); |
ccece235 | 701 | header->snap_sizes = NULL; |
602adf40 | 702 | kfree(header->snap_names); |
ccece235 | 703 | header->snap_names = NULL; |
6a52325f AE |
704 | kfree(header->object_prefix); |
705 | header->object_prefix = NULL; | |
ccece235 | 706 | |
00f1f36f | 707 | return -ENOMEM; |
602adf40 YS |
708 | } |
709 | ||
9e15b77d AE |
710 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
711 | { | |
712 | struct rbd_snap *snap; | |
713 | ||
714 | if (snap_id == CEPH_NOSNAP) | |
715 | return RBD_SNAP_HEAD_NAME; | |
716 | ||
717 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
718 | if (snap_id == snap->id) | |
719 | return snap->name; | |
720 | ||
721 | return NULL; | |
722 | } | |
723 | ||
8836b995 | 724 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 | 725 | { |
602adf40 | 726 | |
e86924a8 | 727 | struct rbd_snap *snap; |
602adf40 | 728 | |
e86924a8 AE |
729 | list_for_each_entry(snap, &rbd_dev->snaps, node) { |
730 | if (!strcmp(snap_name, snap->name)) { | |
0d7dbfce | 731 | rbd_dev->spec->snap_id = snap->id; |
e86924a8 | 732 | rbd_dev->mapping.size = snap->size; |
34b13184 | 733 | rbd_dev->mapping.features = snap->features; |
602adf40 | 734 | |
e86924a8 | 735 | return 0; |
00f1f36f | 736 | } |
00f1f36f | 737 | } |
e86924a8 | 738 | |
00f1f36f | 739 | return -ENOENT; |
602adf40 YS |
740 | } |
741 | ||
819d52bf | 742 | static int rbd_dev_set_mapping(struct rbd_device *rbd_dev) |
602adf40 | 743 | { |
78dc447d | 744 | int ret; |
602adf40 | 745 | |
0d7dbfce | 746 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 747 | sizeof (RBD_SNAP_HEAD_NAME))) { |
0d7dbfce | 748 | rbd_dev->spec->snap_id = CEPH_NOSNAP; |
99c1f08f | 749 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 750 | rbd_dev->mapping.features = rbd_dev->header.features; |
e86924a8 | 751 | ret = 0; |
602adf40 | 752 | } else { |
0d7dbfce | 753 | ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); |
602adf40 YS |
754 | if (ret < 0) |
755 | goto done; | |
f84344f3 | 756 | rbd_dev->mapping.read_only = true; |
602adf40 | 757 | } |
d78b650a | 758 | atomic_set(&rbd_dev->exists, 1); |
602adf40 | 759 | done: |
602adf40 YS |
760 | return ret; |
761 | } | |
762 | ||
763 | static void rbd_header_free(struct rbd_image_header *header) | |
764 | { | |
849b4260 | 765 | kfree(header->object_prefix); |
d78fd7ae | 766 | header->object_prefix = NULL; |
602adf40 | 767 | kfree(header->snap_sizes); |
d78fd7ae | 768 | header->snap_sizes = NULL; |
849b4260 | 769 | kfree(header->snap_names); |
d78fd7ae | 770 | header->snap_names = NULL; |
d1d25646 | 771 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 772 | header->snapc = NULL; |
602adf40 YS |
773 | } |
774 | ||
98571b5a | 775 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 776 | { |
65ccfe21 AE |
777 | char *name; |
778 | u64 segment; | |
779 | int ret; | |
602adf40 | 780 | |
2fd82b9e | 781 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
782 | if (!name) |
783 | return NULL; | |
784 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 785 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 786 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 787 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
788 | pr_err("error formatting segment name for #%llu (%d)\n", |
789 | segment, ret); | |
790 | kfree(name); | |
791 | name = NULL; | |
792 | } | |
602adf40 | 793 | |
65ccfe21 AE |
794 | return name; |
795 | } | |
602adf40 | 796 | |
65ccfe21 AE |
797 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
798 | { | |
799 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 800 | |
65ccfe21 AE |
801 | return offset & (segment_size - 1); |
802 | } | |
803 | ||
804 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
805 | u64 offset, u64 length) | |
806 | { | |
807 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
808 | ||
809 | offset &= segment_size - 1; | |
810 | ||
aafb230e | 811 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
812 | if (offset + length > segment_size) |
813 | length = segment_size - offset; | |
814 | ||
815 | return length; | |
602adf40 YS |
816 | } |
817 | ||
1fec7093 YS |
818 | static int rbd_get_num_segments(struct rbd_image_header *header, |
819 | u64 ofs, u64 len) | |
820 | { | |
df111be6 AE |
821 | u64 start_seg; |
822 | u64 end_seg; | |
38901e0f | 823 | u64 result; |
df111be6 AE |
824 | |
825 | if (!len) | |
826 | return 0; | |
827 | if (len - 1 > U64_MAX - ofs) | |
828 | return -ERANGE; | |
829 | ||
830 | start_seg = ofs >> header->obj_order; | |
831 | end_seg = (ofs + len - 1) >> header->obj_order; | |
832 | ||
38901e0f AE |
833 | result = end_seg - start_seg + 1; |
834 | if (result > (u64) INT_MAX) | |
835 | return -ERANGE; | |
836 | ||
837 | return (int) result; | |
1fec7093 YS |
838 | } |
839 | ||
029bcbd8 JD |
840 | /* |
841 | * returns the size of an object in the image | |
842 | */ | |
843 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
844 | { | |
845 | return 1 << header->obj_order; | |
846 | } | |
847 | ||
602adf40 YS |
848 | /* |
849 | * bio helpers | |
850 | */ | |
851 | ||
852 | static void bio_chain_put(struct bio *chain) | |
853 | { | |
854 | struct bio *tmp; | |
855 | ||
856 | while (chain) { | |
857 | tmp = chain; | |
858 | chain = chain->bi_next; | |
859 | bio_put(tmp); | |
860 | } | |
861 | } | |
862 | ||
863 | /* | |
864 | * zeros a bio chain, starting at specific offset | |
865 | */ | |
866 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
867 | { | |
868 | struct bio_vec *bv; | |
869 | unsigned long flags; | |
870 | void *buf; | |
871 | int i; | |
872 | int pos = 0; | |
873 | ||
874 | while (chain) { | |
875 | bio_for_each_segment(bv, chain, i) { | |
876 | if (pos + bv->bv_len > start_ofs) { | |
877 | int remainder = max(start_ofs - pos, 0); | |
878 | buf = bvec_kmap_irq(bv, &flags); | |
879 | memset(buf + remainder, 0, | |
880 | bv->bv_len - remainder); | |
85b5aaa6 | 881 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
882 | } |
883 | pos += bv->bv_len; | |
884 | } | |
885 | ||
886 | chain = chain->bi_next; | |
887 | } | |
888 | } | |
889 | ||
890 | /* | |
f7760dad AE |
891 | * Clone a portion of a bio, starting at the given byte offset |
892 | * and continuing for the number of bytes indicated. | |
602adf40 | 893 | */ |
f7760dad AE |
894 | static struct bio *bio_clone_range(struct bio *bio_src, |
895 | unsigned int offset, | |
896 | unsigned int len, | |
897 | gfp_t gfpmask) | |
602adf40 | 898 | { |
f7760dad AE |
899 | struct bio_vec *bv; |
900 | unsigned int resid; | |
901 | unsigned short idx; | |
902 | unsigned int voff; | |
903 | unsigned short end_idx; | |
904 | unsigned short vcnt; | |
905 | struct bio *bio; | |
906 | ||
907 | /* Handle the easy case for the caller */ | |
908 | ||
909 | if (!offset && len == bio_src->bi_size) | |
910 | return bio_clone(bio_src, gfpmask); | |
911 | ||
912 | if (WARN_ON_ONCE(!len)) | |
913 | return NULL; | |
914 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
915 | return NULL; | |
916 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
917 | return NULL; | |
918 | ||
919 | /* Find first affected segment... */ | |
920 | ||
921 | resid = offset; | |
922 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
923 | if (resid < bv->bv_len) | |
924 | break; | |
925 | resid -= bv->bv_len; | |
602adf40 | 926 | } |
f7760dad | 927 | voff = resid; |
602adf40 | 928 | |
f7760dad | 929 | /* ...and the last affected segment */ |
602adf40 | 930 | |
f7760dad AE |
931 | resid += len; |
932 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
933 | if (resid <= bv->bv_len) | |
934 | break; | |
935 | resid -= bv->bv_len; | |
936 | } | |
937 | vcnt = end_idx - idx + 1; | |
938 | ||
939 | /* Build the clone */ | |
940 | ||
941 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
942 | if (!bio) | |
943 | return NULL; /* ENOMEM */ | |
602adf40 | 944 | |
f7760dad AE |
945 | bio->bi_bdev = bio_src->bi_bdev; |
946 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
947 | bio->bi_rw = bio_src->bi_rw; | |
948 | bio->bi_flags |= 1 << BIO_CLONED; | |
949 | ||
950 | /* | |
951 | * Copy over our part of the bio_vec, then update the first | |
952 | * and last (or only) entries. | |
953 | */ | |
954 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
955 | vcnt * sizeof (struct bio_vec)); | |
956 | bio->bi_io_vec[0].bv_offset += voff; | |
957 | if (vcnt > 1) { | |
958 | bio->bi_io_vec[0].bv_len -= voff; | |
959 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
960 | } else { | |
961 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
962 | } |
963 | ||
f7760dad AE |
964 | bio->bi_vcnt = vcnt; |
965 | bio->bi_size = len; | |
966 | bio->bi_idx = 0; | |
967 | ||
968 | return bio; | |
969 | } | |
970 | ||
971 | /* | |
972 | * Clone a portion of a bio chain, starting at the given byte offset | |
973 | * into the first bio in the source chain and continuing for the | |
974 | * number of bytes indicated. The result is another bio chain of | |
975 | * exactly the given length, or a null pointer on error. | |
976 | * | |
977 | * The bio_src and offset parameters are both in-out. On entry they | |
978 | * refer to the first source bio and the offset into that bio where | |
979 | * the start of data to be cloned is located. | |
980 | * | |
981 | * On return, bio_src is updated to refer to the bio in the source | |
982 | * chain that contains first un-cloned byte, and *offset will | |
983 | * contain the offset of that byte within that bio. | |
984 | */ | |
985 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
986 | unsigned int *offset, | |
987 | unsigned int len, | |
988 | gfp_t gfpmask) | |
989 | { | |
990 | struct bio *bi = *bio_src; | |
991 | unsigned int off = *offset; | |
992 | struct bio *chain = NULL; | |
993 | struct bio **end; | |
994 | ||
995 | /* Build up a chain of clone bios up to the limit */ | |
996 | ||
997 | if (!bi || off >= bi->bi_size || !len) | |
998 | return NULL; /* Nothing to clone */ | |
602adf40 | 999 | |
f7760dad AE |
1000 | end = &chain; |
1001 | while (len) { | |
1002 | unsigned int bi_size; | |
1003 | struct bio *bio; | |
1004 | ||
f5400b7a AE |
1005 | if (!bi) { |
1006 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1007 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1008 | } |
f7760dad AE |
1009 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1010 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1011 | if (!bio) | |
1012 | goto out_err; /* ENOMEM */ | |
1013 | ||
1014 | *end = bio; | |
1015 | end = &bio->bi_next; | |
602adf40 | 1016 | |
f7760dad AE |
1017 | off += bi_size; |
1018 | if (off == bi->bi_size) { | |
1019 | bi = bi->bi_next; | |
1020 | off = 0; | |
1021 | } | |
1022 | len -= bi_size; | |
1023 | } | |
1024 | *bio_src = bi; | |
1025 | *offset = off; | |
1026 | ||
1027 | return chain; | |
1028 | out_err: | |
1029 | bio_chain_put(chain); | |
602adf40 | 1030 | |
602adf40 YS |
1031 | return NULL; |
1032 | } | |
1033 | ||
8d23bf29 AE |
1034 | struct ceph_osd_req_op *rbd_osd_req_op_create(u16 opcode, ...) |
1035 | { | |
1036 | struct ceph_osd_req_op *op; | |
1037 | va_list args; | |
2647ba38 | 1038 | size_t size; |
8d23bf29 AE |
1039 | |
1040 | op = kzalloc(sizeof (*op), GFP_NOIO); | |
1041 | if (!op) | |
1042 | return NULL; | |
1043 | op->op = opcode; | |
1044 | va_start(args, opcode); | |
1045 | switch (opcode) { | |
1046 | case CEPH_OSD_OP_READ: | |
1047 | case CEPH_OSD_OP_WRITE: | |
1048 | /* rbd_osd_req_op_create(READ, offset, length) */ | |
1049 | /* rbd_osd_req_op_create(WRITE, offset, length) */ | |
1050 | op->extent.offset = va_arg(args, u64); | |
1051 | op->extent.length = va_arg(args, u64); | |
1052 | if (opcode == CEPH_OSD_OP_WRITE) | |
1053 | op->payload_len = op->extent.length; | |
1054 | break; | |
2647ba38 AE |
1055 | case CEPH_OSD_OP_CALL: |
1056 | /* rbd_osd_req_op_create(CALL, class, method, data, datalen) */ | |
1057 | op->cls.class_name = va_arg(args, char *); | |
1058 | size = strlen(op->cls.class_name); | |
1059 | rbd_assert(size <= (size_t) U8_MAX); | |
1060 | op->cls.class_len = size; | |
1061 | op->payload_len = size; | |
1062 | ||
1063 | op->cls.method_name = va_arg(args, char *); | |
1064 | size = strlen(op->cls.method_name); | |
1065 | rbd_assert(size <= (size_t) U8_MAX); | |
1066 | op->cls.method_len = size; | |
1067 | op->payload_len += size; | |
1068 | ||
1069 | op->cls.argc = 0; | |
1070 | op->cls.indata = va_arg(args, void *); | |
1071 | size = va_arg(args, size_t); | |
1072 | rbd_assert(size <= (size_t) U32_MAX); | |
1073 | op->cls.indata_len = (u32) size; | |
1074 | op->payload_len += size; | |
1075 | break; | |
5efea49a AE |
1076 | case CEPH_OSD_OP_NOTIFY_ACK: |
1077 | case CEPH_OSD_OP_WATCH: | |
1078 | /* rbd_osd_req_op_create(NOTIFY_ACK, cookie, version) */ | |
1079 | /* rbd_osd_req_op_create(WATCH, cookie, version, flag) */ | |
1080 | op->watch.cookie = va_arg(args, u64); | |
1081 | op->watch.ver = va_arg(args, u64); | |
1082 | op->watch.ver = cpu_to_le64(op->watch.ver); | |
1083 | if (opcode == CEPH_OSD_OP_WATCH && va_arg(args, int)) | |
1084 | op->watch.flag = (u8) 1; | |
1085 | break; | |
8d23bf29 AE |
1086 | default: |
1087 | rbd_warn(NULL, "unsupported opcode %hu\n", opcode); | |
1088 | kfree(op); | |
1089 | op = NULL; | |
1090 | break; | |
1091 | } | |
1092 | va_end(args); | |
1093 | ||
1094 | return op; | |
1095 | } | |
1096 | ||
1097 | static void rbd_osd_req_op_destroy(struct ceph_osd_req_op *op) | |
1098 | { | |
1099 | kfree(op); | |
1100 | } | |
1101 | ||
1fec7093 YS |
1102 | static void rbd_coll_end_req_index(struct request *rq, |
1103 | struct rbd_req_coll *coll, | |
1104 | int index, | |
8986cb37 | 1105 | s32 ret, u64 len) |
1fec7093 YS |
1106 | { |
1107 | struct request_queue *q; | |
1108 | int min, max, i; | |
1109 | ||
bd919d45 | 1110 | dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n", |
8986cb37 | 1111 | coll, index, (int)ret, (unsigned long long)len); |
1fec7093 YS |
1112 | |
1113 | if (!rq) | |
1114 | return; | |
1115 | ||
1116 | if (!coll) { | |
1117 | blk_end_request(rq, ret, len); | |
1118 | return; | |
1119 | } | |
1120 | ||
1121 | q = rq->q; | |
1122 | ||
1123 | spin_lock_irq(q->queue_lock); | |
1124 | coll->status[index].done = 1; | |
1125 | coll->status[index].rc = ret; | |
1126 | coll->status[index].bytes = len; | |
1127 | max = min = coll->num_done; | |
1128 | while (max < coll->total && coll->status[max].done) | |
1129 | max++; | |
1130 | ||
1131 | for (i = min; i<max; i++) { | |
8986cb37 | 1132 | __blk_end_request(rq, (int)coll->status[i].rc, |
1fec7093 YS |
1133 | coll->status[i].bytes); |
1134 | coll->num_done++; | |
1135 | kref_put(&coll->kref, rbd_coll_release); | |
1136 | } | |
1137 | spin_unlock_irq(q->queue_lock); | |
1138 | } | |
1139 | ||
725afc97 | 1140 | static void rbd_coll_end_req(struct rbd_request *rbd_req, |
8986cb37 | 1141 | s32 ret, u64 len) |
1fec7093 | 1142 | { |
725afc97 AE |
1143 | rbd_coll_end_req_index(rbd_req->rq, |
1144 | rbd_req->coll, rbd_req->coll_index, | |
1145 | ret, len); | |
1fec7093 YS |
1146 | } |
1147 | ||
602adf40 YS |
1148 | /* |
1149 | * Send ceph osd request | |
1150 | */ | |
1151 | static int rbd_do_request(struct request *rq, | |
0ce1a794 | 1152 | struct rbd_device *rbd_dev, |
602adf40 YS |
1153 | struct ceph_snap_context *snapc, |
1154 | u64 snapid, | |
aded07ea | 1155 | const char *object_name, u64 ofs, u64 len, |
602adf40 YS |
1156 | struct bio *bio, |
1157 | struct page **pages, | |
1158 | int num_pages, | |
1159 | int flags, | |
30573d68 | 1160 | struct ceph_osd_req_op *op, |
1fec7093 YS |
1161 | struct rbd_req_coll *coll, |
1162 | int coll_index, | |
5f29ddd4 AE |
1163 | void (*rbd_cb)(struct ceph_osd_request *, |
1164 | struct ceph_msg *), | |
59c2be1e | 1165 | u64 *ver) |
602adf40 | 1166 | { |
2e53c6c3 | 1167 | struct ceph_osd_client *osdc; |
5f29ddd4 | 1168 | struct ceph_osd_request *osd_req; |
2e53c6c3 | 1169 | struct rbd_request *rbd_req = NULL; |
602adf40 | 1170 | struct timespec mtime = CURRENT_TIME; |
2e53c6c3 | 1171 | int ret; |
602adf40 | 1172 | |
f7760dad AE |
1173 | dout("rbd_do_request object_name=%s ofs=%llu len=%llu coll=%p[%d]\n", |
1174 | object_name, (unsigned long long) ofs, | |
1175 | (unsigned long long) len, coll, coll_index); | |
602adf40 | 1176 | |
0ce1a794 | 1177 | osdc = &rbd_dev->rbd_client->client->osdc; |
30573d68 | 1178 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_NOIO); |
2e53c6c3 AE |
1179 | if (!osd_req) |
1180 | return -ENOMEM; | |
602adf40 | 1181 | |
d178a9e7 | 1182 | osd_req->r_flags = flags; |
54a54007 AE |
1183 | osd_req->r_pages = pages; |
1184 | if (bio) { | |
1185 | osd_req->r_bio = bio; | |
1186 | bio_get(osd_req->r_bio); | |
1187 | } | |
602adf40 | 1188 | |
18216657 | 1189 | if (coll) { |
2e53c6c3 AE |
1190 | ret = -ENOMEM; |
1191 | rbd_req = kmalloc(sizeof(*rbd_req), GFP_NOIO); | |
1192 | if (!rbd_req) | |
1193 | goto done_osd_req; | |
1194 | ||
1195 | rbd_req->rq = rq; | |
1196 | rbd_req->bio = bio; | |
1197 | rbd_req->pages = pages; | |
1198 | rbd_req->len = len; | |
1199 | rbd_req->coll = coll; | |
18216657 | 1200 | rbd_req->coll_index = coll_index; |
2e53c6c3 | 1201 | } |
602adf40 | 1202 | |
2e53c6c3 | 1203 | osd_req->r_callback = rbd_cb; |
5f29ddd4 | 1204 | osd_req->r_priv = rbd_req; |
602adf40 | 1205 | |
5f29ddd4 AE |
1206 | strncpy(osd_req->r_oid, object_name, sizeof(osd_req->r_oid)); |
1207 | osd_req->r_oid_len = strlen(osd_req->r_oid); | |
602adf40 | 1208 | |
0903e875 | 1209 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ |
e01e7927 AE |
1210 | osd_req->r_num_pages = calc_pages_for(ofs, len); |
1211 | osd_req->r_page_alignment = ofs & ~PAGE_MASK; | |
602adf40 | 1212 | |
30573d68 | 1213 | ceph_osdc_build_request(osd_req, ofs, len, 1, op, |
ae7ca4a3 | 1214 | snapc, snapid, &mtime); |
602adf40 | 1215 | |
8b84de79 | 1216 | if (op->op == CEPH_OSD_OP_WATCH && op->watch.flag) { |
5f29ddd4 | 1217 | ceph_osdc_set_request_linger(osdc, osd_req); |
8b84de79 | 1218 | rbd_dev->watch_request = osd_req; |
59c2be1e YS |
1219 | } |
1220 | ||
5f29ddd4 | 1221 | ret = ceph_osdc_start_request(osdc, osd_req, false); |
602adf40 YS |
1222 | if (ret < 0) |
1223 | goto done_err; | |
1224 | ||
1225 | if (!rbd_cb) { | |
5f29ddd4 AE |
1226 | u64 version; |
1227 | ||
1228 | ret = ceph_osdc_wait_request(osdc, osd_req); | |
1229 | version = le64_to_cpu(osd_req->r_reassert_version.version); | |
59c2be1e | 1230 | if (ver) |
5f29ddd4 AE |
1231 | *ver = version; |
1232 | dout("reassert_ver=%llu\n", (unsigned long long) version); | |
1233 | ceph_osdc_put_request(osd_req); | |
602adf40 YS |
1234 | } |
1235 | return ret; | |
1236 | ||
1237 | done_err: | |
2e53c6c3 AE |
1238 | if (bio) |
1239 | bio_chain_put(osd_req->r_bio); | |
725afc97 | 1240 | kfree(rbd_req); |
2e53c6c3 AE |
1241 | done_osd_req: |
1242 | ceph_osdc_put_request(osd_req); | |
1243 | ||
602adf40 YS |
1244 | return ret; |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * Ceph osd op callback | |
1249 | */ | |
5f29ddd4 | 1250 | static void rbd_req_cb(struct ceph_osd_request *osd_req, struct ceph_msg *msg) |
602adf40 | 1251 | { |
5f29ddd4 | 1252 | struct rbd_request *rbd_req = osd_req->r_priv; |
602adf40 YS |
1253 | struct ceph_osd_reply_head *replyhead; |
1254 | struct ceph_osd_op *op; | |
8986cb37 | 1255 | s32 rc; |
602adf40 YS |
1256 | u64 bytes; |
1257 | int read_op; | |
1258 | ||
1259 | /* parse reply */ | |
1260 | replyhead = msg->front.iov_base; | |
1261 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
1262 | op = (void *)(replyhead + 1); | |
8986cb37 | 1263 | rc = (s32)le32_to_cpu(replyhead->result); |
602adf40 | 1264 | bytes = le64_to_cpu(op->extent.length); |
895cfcc8 | 1265 | read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ); |
602adf40 | 1266 | |
bd919d45 AE |
1267 | dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n", |
1268 | (unsigned long long) bytes, read_op, (int) rc); | |
602adf40 | 1269 | |
8986cb37 | 1270 | if (rc == (s32)-ENOENT && read_op) { |
725afc97 | 1271 | zero_bio_chain(rbd_req->bio, 0); |
602adf40 | 1272 | rc = 0; |
725afc97 AE |
1273 | } else if (rc == 0 && read_op && bytes < rbd_req->len) { |
1274 | zero_bio_chain(rbd_req->bio, bytes); | |
1275 | bytes = rbd_req->len; | |
602adf40 YS |
1276 | } |
1277 | ||
725afc97 | 1278 | rbd_coll_end_req(rbd_req, rc, bytes); |
602adf40 | 1279 | |
725afc97 AE |
1280 | if (rbd_req->bio) |
1281 | bio_chain_put(rbd_req->bio); | |
602adf40 | 1282 | |
5f29ddd4 | 1283 | ceph_osdc_put_request(osd_req); |
725afc97 | 1284 | kfree(rbd_req); |
602adf40 YS |
1285 | } |
1286 | ||
5f29ddd4 AE |
1287 | static void rbd_simple_req_cb(struct ceph_osd_request *osd_req, |
1288 | struct ceph_msg *msg) | |
59c2be1e | 1289 | { |
5f29ddd4 | 1290 | ceph_osdc_put_request(osd_req); |
59c2be1e YS |
1291 | } |
1292 | ||
602adf40 YS |
1293 | /* |
1294 | * Do a synchronous ceph osd operation | |
1295 | */ | |
0ce1a794 | 1296 | static int rbd_req_sync_op(struct rbd_device *rbd_dev, |
602adf40 | 1297 | int flags, |
30573d68 | 1298 | struct ceph_osd_req_op *op, |
aded07ea | 1299 | const char *object_name, |
f8d4de6e AE |
1300 | u64 ofs, u64 inbound_size, |
1301 | char *inbound, | |
59c2be1e | 1302 | u64 *ver) |
602adf40 YS |
1303 | { |
1304 | int ret; | |
1305 | struct page **pages; | |
1306 | int num_pages; | |
913d2fdc | 1307 | |
30573d68 | 1308 | rbd_assert(op != NULL); |
602adf40 | 1309 | |
f8d4de6e | 1310 | num_pages = calc_pages_for(ofs, inbound_size); |
602adf40 | 1311 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
b8d0638a DC |
1312 | if (IS_ERR(pages)) |
1313 | return PTR_ERR(pages); | |
602adf40 | 1314 | |
25704ac9 | 1315 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
f8d4de6e | 1316 | object_name, ofs, inbound_size, NULL, |
602adf40 YS |
1317 | pages, num_pages, |
1318 | flags, | |
30573d68 | 1319 | op, |
1fec7093 | 1320 | NULL, 0, |
59c2be1e | 1321 | NULL, |
8b84de79 | 1322 | ver); |
602adf40 | 1323 | if (ret < 0) |
913d2fdc | 1324 | goto done; |
602adf40 | 1325 | |
f8d4de6e AE |
1326 | if ((flags & CEPH_OSD_FLAG_READ) && inbound) |
1327 | ret = ceph_copy_from_page_vector(pages, inbound, ofs, ret); | |
602adf40 | 1328 | |
602adf40 YS |
1329 | done: |
1330 | ceph_release_page_vector(pages, num_pages); | |
1331 | return ret; | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Do an asynchronous ceph osd operation | |
1336 | */ | |
1337 | static int rbd_do_op(struct request *rq, | |
0ce1a794 | 1338 | struct rbd_device *rbd_dev, |
602adf40 | 1339 | struct ceph_snap_context *snapc, |
602adf40 | 1340 | u64 ofs, u64 len, |
1fec7093 YS |
1341 | struct bio *bio, |
1342 | struct rbd_req_coll *coll, | |
1343 | int coll_index) | |
602adf40 | 1344 | { |
98571b5a | 1345 | const char *seg_name; |
602adf40 YS |
1346 | u64 seg_ofs; |
1347 | u64 seg_len; | |
1348 | int ret; | |
139b4318 | 1349 | struct ceph_osd_req_op *op; |
ff2e4bb5 AE |
1350 | int opcode; |
1351 | int flags; | |
4634246d | 1352 | u64 snapid; |
602adf40 | 1353 | |
65ccfe21 | 1354 | seg_name = rbd_segment_name(rbd_dev, ofs); |
602adf40 YS |
1355 | if (!seg_name) |
1356 | return -ENOMEM; | |
65ccfe21 AE |
1357 | seg_len = rbd_segment_length(rbd_dev, ofs, len); |
1358 | seg_ofs = rbd_segment_offset(rbd_dev, ofs); | |
602adf40 | 1359 | |
ff2e4bb5 AE |
1360 | if (rq_data_dir(rq) == WRITE) { |
1361 | opcode = CEPH_OSD_OP_WRITE; | |
1362 | flags = CEPH_OSD_FLAG_WRITE|CEPH_OSD_FLAG_ONDISK; | |
4634246d | 1363 | snapid = CEPH_NOSNAP; |
ff2e4bb5 AE |
1364 | } else { |
1365 | opcode = CEPH_OSD_OP_READ; | |
1366 | flags = CEPH_OSD_FLAG_READ; | |
a7b4c65f | 1367 | rbd_assert(!snapc); |
0d7dbfce | 1368 | snapid = rbd_dev->spec->snap_id; |
ff2e4bb5 | 1369 | } |
602adf40 | 1370 | |
57cfc106 | 1371 | ret = -ENOMEM; |
8d23bf29 | 1372 | op = rbd_osd_req_op_create(opcode, seg_ofs, seg_len); |
139b4318 | 1373 | if (!op) |
602adf40 YS |
1374 | goto done; |
1375 | ||
1376 | /* we've taken care of segment sizes earlier when we | |
1377 | cloned the bios. We should never have a segment | |
1378 | truncated at this point */ | |
aafb230e | 1379 | rbd_assert(seg_len == len); |
602adf40 YS |
1380 | |
1381 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1382 | seg_name, seg_ofs, seg_len, | |
1383 | bio, | |
1384 | NULL, 0, | |
1385 | flags, | |
30573d68 | 1386 | op, |
1fec7093 | 1387 | coll, coll_index, |
8b84de79 | 1388 | rbd_req_cb, NULL); |
cd323ac0 AE |
1389 | if (ret < 0) |
1390 | rbd_coll_end_req_index(rq, coll, coll_index, | |
1391 | (s32)ret, seg_len); | |
8d23bf29 | 1392 | rbd_osd_req_op_destroy(op); |
602adf40 YS |
1393 | done: |
1394 | kfree(seg_name); | |
1395 | return ret; | |
1396 | } | |
1397 | ||
602adf40 YS |
1398 | /* |
1399 | * Request sync osd read | |
1400 | */ | |
0ce1a794 | 1401 | static int rbd_req_sync_read(struct rbd_device *rbd_dev, |
aded07ea | 1402 | const char *object_name, |
602adf40 | 1403 | u64 ofs, u64 len, |
59c2be1e YS |
1404 | char *buf, |
1405 | u64 *ver) | |
602adf40 | 1406 | { |
139b4318 | 1407 | struct ceph_osd_req_op *op; |
913d2fdc AE |
1408 | int ret; |
1409 | ||
8d23bf29 | 1410 | op = rbd_osd_req_op_create(CEPH_OSD_OP_READ, ofs, len); |
139b4318 | 1411 | if (!op) |
913d2fdc AE |
1412 | return -ENOMEM; |
1413 | ||
25704ac9 | 1414 | ret = rbd_req_sync_op(rbd_dev, CEPH_OSD_FLAG_READ, |
8b84de79 | 1415 | op, object_name, ofs, len, buf, ver); |
8d23bf29 | 1416 | rbd_osd_req_op_destroy(op); |
913d2fdc AE |
1417 | |
1418 | return ret; | |
602adf40 YS |
1419 | } |
1420 | ||
1421 | /* | |
59c2be1e YS |
1422 | * Request sync osd watch |
1423 | */ | |
0ce1a794 | 1424 | static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev, |
59c2be1e | 1425 | u64 ver, |
7f0a24d8 | 1426 | u64 notify_id) |
59c2be1e | 1427 | { |
139b4318 | 1428 | struct ceph_osd_req_op *op; |
11f77002 SW |
1429 | int ret; |
1430 | ||
5efea49a | 1431 | op = rbd_osd_req_op_create(CEPH_OSD_OP_NOTIFY_ACK, notify_id, ver); |
139b4318 | 1432 | if (!op) |
57cfc106 | 1433 | return -ENOMEM; |
59c2be1e | 1434 | |
0ce1a794 | 1435 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
7f0a24d8 | 1436 | rbd_dev->header_name, 0, 0, NULL, |
ad4f232f | 1437 | NULL, 0, |
59c2be1e | 1438 | CEPH_OSD_FLAG_READ, |
30573d68 | 1439 | op, |
1fec7093 | 1440 | NULL, 0, |
8b84de79 | 1441 | rbd_simple_req_cb, NULL); |
59c2be1e | 1442 | |
5efea49a AE |
1443 | rbd_osd_req_op_destroy(op); |
1444 | ||
59c2be1e YS |
1445 | return ret; |
1446 | } | |
1447 | ||
1448 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1449 | { | |
0ce1a794 | 1450 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
a71b891b | 1451 | u64 hver; |
13143d2d SW |
1452 | int rc; |
1453 | ||
0ce1a794 | 1454 | if (!rbd_dev) |
59c2be1e YS |
1455 | return; |
1456 | ||
bd919d45 AE |
1457 | dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n", |
1458 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1459 | (unsigned int) opcode); | |
117973fb | 1460 | rc = rbd_dev_refresh(rbd_dev, &hver); |
13143d2d | 1461 | if (rc) |
06ecc6cb AE |
1462 | rbd_warn(rbd_dev, "got notification but failed to " |
1463 | " update snaps: %d\n", rc); | |
59c2be1e | 1464 | |
7f0a24d8 | 1465 | rbd_req_sync_notify_ack(rbd_dev, hver, notify_id); |
59c2be1e YS |
1466 | } |
1467 | ||
1468 | /* | |
907703d0 AE |
1469 | * Request sync osd watch/unwatch. The value of "start" determines |
1470 | * whether a watch request is being initiated or torn down. | |
59c2be1e | 1471 | */ |
907703d0 | 1472 | static int rbd_req_sync_watch(struct rbd_device *rbd_dev, int start) |
59c2be1e | 1473 | { |
5efea49a AE |
1474 | struct ceph_osd_req_op *op; |
1475 | int ret = 0; | |
59c2be1e | 1476 | |
c0430647 AE |
1477 | rbd_assert(start ^ !!rbd_dev->watch_event); |
1478 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
1479 | ||
907703d0 AE |
1480 | if (start) { |
1481 | struct ceph_osd_client *osdc; | |
79e3057c | 1482 | |
907703d0 AE |
1483 | osdc = &rbd_dev->rbd_client->client->osdc; |
1484 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, rbd_dev, | |
1485 | &rbd_dev->watch_event); | |
1486 | if (ret < 0) | |
5efea49a | 1487 | return ret; |
907703d0 | 1488 | } |
79e3057c | 1489 | |
5efea49a AE |
1490 | op = rbd_osd_req_op_create(CEPH_OSD_OP_WATCH, |
1491 | rbd_dev->watch_event->cookie, | |
1492 | rbd_dev->header.obj_version, start); | |
1493 | if (op) | |
1494 | ret = rbd_req_sync_op(rbd_dev, | |
79e3057c | 1495 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
907703d0 | 1496 | op, rbd_dev->header_name, |
8b84de79 | 1497 | 0, 0, NULL, NULL); |
79e3057c | 1498 | |
5efea49a AE |
1499 | /* Cancel the event if we're tearing down, or on error */ |
1500 | ||
1501 | if (!start || !op || ret < 0) { | |
907703d0 AE |
1502 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1503 | rbd_dev->watch_event = NULL; | |
1504 | } | |
5efea49a | 1505 | rbd_osd_req_op_destroy(op); |
907703d0 | 1506 | |
79e3057c YS |
1507 | return ret; |
1508 | } | |
1509 | ||
602adf40 | 1510 | /* |
3cb4a687 | 1511 | * Synchronous osd object method call |
602adf40 | 1512 | */ |
0ce1a794 | 1513 | static int rbd_req_sync_exec(struct rbd_device *rbd_dev, |
aded07ea AE |
1514 | const char *object_name, |
1515 | const char *class_name, | |
1516 | const char *method_name, | |
3cb4a687 AE |
1517 | const char *outbound, |
1518 | size_t outbound_size, | |
f8d4de6e AE |
1519 | char *inbound, |
1520 | size_t inbound_size, | |
59c2be1e | 1521 | u64 *ver) |
602adf40 | 1522 | { |
139b4318 | 1523 | struct ceph_osd_req_op *op; |
57cfc106 AE |
1524 | int ret; |
1525 | ||
3cb4a687 AE |
1526 | /* |
1527 | * Any input parameters required by the method we're calling | |
1528 | * will be sent along with the class and method names as | |
1529 | * part of the message payload. That data and its size are | |
1530 | * supplied via the indata and indata_len fields (named from | |
1531 | * the perspective of the server side) in the OSD request | |
1532 | * operation. | |
1533 | */ | |
2647ba38 AE |
1534 | op = rbd_osd_req_op_create(CEPH_OSD_OP_CALL, class_name, |
1535 | method_name, outbound, outbound_size); | |
139b4318 | 1536 | if (!op) |
57cfc106 | 1537 | return -ENOMEM; |
602adf40 | 1538 | |
30573d68 | 1539 | ret = rbd_req_sync_op(rbd_dev, CEPH_OSD_FLAG_READ, op, |
f8d4de6e | 1540 | object_name, 0, inbound_size, inbound, |
8b84de79 | 1541 | ver); |
602adf40 | 1542 | |
2647ba38 | 1543 | rbd_osd_req_op_destroy(op); |
602adf40 YS |
1544 | |
1545 | dout("cls_exec returned %d\n", ret); | |
1546 | return ret; | |
1547 | } | |
1548 | ||
1fec7093 YS |
1549 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1550 | { | |
1551 | struct rbd_req_coll *coll = | |
1552 | kzalloc(sizeof(struct rbd_req_coll) + | |
1553 | sizeof(struct rbd_req_status) * num_reqs, | |
1554 | GFP_ATOMIC); | |
1555 | ||
1556 | if (!coll) | |
1557 | return NULL; | |
1558 | coll->total = num_reqs; | |
1559 | kref_init(&coll->kref); | |
1560 | return coll; | |
1561 | } | |
1562 | ||
8295cda7 AE |
1563 | static int rbd_dev_do_request(struct request *rq, |
1564 | struct rbd_device *rbd_dev, | |
1565 | struct ceph_snap_context *snapc, | |
1566 | u64 ofs, unsigned int size, | |
1567 | struct bio *bio_chain) | |
1568 | { | |
1569 | int num_segs; | |
1570 | struct rbd_req_coll *coll; | |
1571 | unsigned int bio_offset; | |
1572 | int cur_seg = 0; | |
1573 | ||
1574 | dout("%s 0x%x bytes at 0x%llx\n", | |
1575 | rq_data_dir(rq) == WRITE ? "write" : "read", | |
1576 | size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE); | |
1577 | ||
1578 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); | |
1579 | if (num_segs <= 0) | |
1580 | return num_segs; | |
1581 | ||
1582 | coll = rbd_alloc_coll(num_segs); | |
1583 | if (!coll) | |
1584 | return -ENOMEM; | |
1585 | ||
1586 | bio_offset = 0; | |
1587 | do { | |
1588 | u64 limit = rbd_segment_length(rbd_dev, ofs, size); | |
1589 | unsigned int clone_size; | |
1590 | struct bio *bio_clone; | |
1591 | ||
1592 | BUG_ON(limit > (u64)UINT_MAX); | |
1593 | clone_size = (unsigned int)limit; | |
1594 | dout("bio_chain->bi_vcnt=%hu\n", bio_chain->bi_vcnt); | |
1595 | ||
1596 | kref_get(&coll->kref); | |
1597 | ||
1598 | /* Pass a cloned bio chain via an osd request */ | |
1599 | ||
1600 | bio_clone = bio_chain_clone_range(&bio_chain, | |
1601 | &bio_offset, clone_size, | |
1602 | GFP_ATOMIC); | |
1603 | if (bio_clone) | |
1604 | (void)rbd_do_op(rq, rbd_dev, snapc, | |
1605 | ofs, clone_size, | |
1606 | bio_clone, coll, cur_seg); | |
1607 | else | |
1608 | rbd_coll_end_req_index(rq, coll, cur_seg, | |
1609 | (s32)-ENOMEM, | |
1610 | clone_size); | |
1611 | size -= clone_size; | |
1612 | ofs += clone_size; | |
1613 | ||
1614 | cur_seg++; | |
1615 | } while (size > 0); | |
1616 | kref_put(&coll->kref, rbd_coll_release); | |
1617 | ||
1618 | return 0; | |
1619 | } | |
1620 | ||
602adf40 YS |
1621 | /* |
1622 | * block device queue callback | |
1623 | */ | |
1624 | static void rbd_rq_fn(struct request_queue *q) | |
1625 | { | |
1626 | struct rbd_device *rbd_dev = q->queuedata; | |
b395e8b5 | 1627 | bool read_only = rbd_dev->mapping.read_only; |
602adf40 | 1628 | struct request *rq; |
602adf40 | 1629 | |
00f1f36f | 1630 | while ((rq = blk_fetch_request(q))) { |
b395e8b5 AE |
1631 | struct ceph_snap_context *snapc = NULL; |
1632 | unsigned int size = 0; | |
8295cda7 | 1633 | int result; |
602adf40 | 1634 | |
602adf40 YS |
1635 | dout("fetched request\n"); |
1636 | ||
b395e8b5 AE |
1637 | /* Filter out block requests we don't understand */ |
1638 | ||
602adf40 YS |
1639 | if ((rq->cmd_type != REQ_TYPE_FS)) { |
1640 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1641 | continue; |
602adf40 | 1642 | } |
b395e8b5 | 1643 | spin_unlock_irq(q->queue_lock); |
602adf40 | 1644 | |
a7b4c65f AE |
1645 | /* Write requests need a reference to the snapshot context */ |
1646 | ||
1647 | if (rq_data_dir(rq) == WRITE) { | |
1648 | result = -EROFS; | |
1649 | if (read_only) /* Can't write to a read-only device */ | |
1650 | goto out_end_request; | |
1651 | ||
1652 | /* | |
1653 | * Note that each osd request will take its | |
1654 | * own reference to the snapshot context | |
1655 | * supplied. The reference we take here | |
1656 | * just guarantees the one we provide stays | |
1657 | * valid. | |
1658 | */ | |
1659 | down_read(&rbd_dev->header_rwsem); | |
b395e8b5 | 1660 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
a7b4c65f | 1661 | up_read(&rbd_dev->header_rwsem); |
b395e8b5 | 1662 | rbd_assert(snapc != NULL); |
a7b4c65f | 1663 | } else if (!atomic_read(&rbd_dev->exists)) { |
0d7dbfce | 1664 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); |
d1d25646 | 1665 | dout("request for non-existent snapshot"); |
b395e8b5 AE |
1666 | result = -ENXIO; |
1667 | goto out_end_request; | |
e88a36ec JD |
1668 | } |
1669 | ||
f7760dad | 1670 | size = blk_rq_bytes(rq); |
b395e8b5 AE |
1671 | result = rbd_dev_do_request(rq, rbd_dev, snapc, |
1672 | blk_rq_pos(rq) * SECTOR_SIZE, | |
1673 | size, rq->bio); | |
1674 | out_end_request: | |
a7b4c65f AE |
1675 | if (snapc) |
1676 | ceph_put_snap_context(snapc); | |
8295cda7 AE |
1677 | spin_lock_irq(q->queue_lock); |
1678 | if (!size || result < 0) | |
1679 | __blk_end_request_all(rq, result); | |
602adf40 YS |
1680 | } |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1685 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 1686 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
1687 | */ |
1688 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1689 | struct bio_vec *bvec) | |
1690 | { | |
1691 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
1692 | sector_t sector_offset; |
1693 | sector_t sectors_per_obj; | |
1694 | sector_t obj_sector_offset; | |
1695 | int ret; | |
1696 | ||
1697 | /* | |
1698 | * Find how far into its rbd object the partition-relative | |
1699 | * bio start sector is to offset relative to the enclosing | |
1700 | * device. | |
1701 | */ | |
1702 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
1703 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
1704 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
1705 | ||
1706 | /* | |
1707 | * Compute the number of bytes from that offset to the end | |
1708 | * of the object. Account for what's already used by the bio. | |
1709 | */ | |
1710 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
1711 | if (ret > bmd->bi_size) | |
1712 | ret -= bmd->bi_size; | |
1713 | else | |
1714 | ret = 0; | |
1715 | ||
1716 | /* | |
1717 | * Don't send back more than was asked for. And if the bio | |
1718 | * was empty, let the whole thing through because: "Note | |
1719 | * that a block device *must* allow a single page to be | |
1720 | * added to an empty bio." | |
1721 | */ | |
1722 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
1723 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
1724 | ret = (int) bvec->bv_len; | |
1725 | ||
1726 | return ret; | |
602adf40 YS |
1727 | } |
1728 | ||
1729 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1730 | { | |
1731 | struct gendisk *disk = rbd_dev->disk; | |
1732 | ||
1733 | if (!disk) | |
1734 | return; | |
1735 | ||
602adf40 YS |
1736 | if (disk->flags & GENHD_FL_UP) |
1737 | del_gendisk(disk); | |
1738 | if (disk->queue) | |
1739 | blk_cleanup_queue(disk->queue); | |
1740 | put_disk(disk); | |
1741 | } | |
1742 | ||
1743 | /* | |
4156d998 AE |
1744 | * Read the complete header for the given rbd device. |
1745 | * | |
1746 | * Returns a pointer to a dynamically-allocated buffer containing | |
1747 | * the complete and validated header. Caller can pass the address | |
1748 | * of a variable that will be filled in with the version of the | |
1749 | * header object at the time it was read. | |
1750 | * | |
1751 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 1752 | */ |
4156d998 AE |
1753 | static struct rbd_image_header_ondisk * |
1754 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 1755 | { |
4156d998 | 1756 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 1757 | u32 snap_count = 0; |
4156d998 AE |
1758 | u64 names_size = 0; |
1759 | u32 want_count; | |
1760 | int ret; | |
602adf40 | 1761 | |
00f1f36f | 1762 | /* |
4156d998 AE |
1763 | * The complete header will include an array of its 64-bit |
1764 | * snapshot ids, followed by the names of those snapshots as | |
1765 | * a contiguous block of NUL-terminated strings. Note that | |
1766 | * the number of snapshots could change by the time we read | |
1767 | * it in, in which case we re-read it. | |
00f1f36f | 1768 | */ |
4156d998 AE |
1769 | do { |
1770 | size_t size; | |
1771 | ||
1772 | kfree(ondisk); | |
1773 | ||
1774 | size = sizeof (*ondisk); | |
1775 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
1776 | size += names_size; | |
1777 | ondisk = kmalloc(size, GFP_KERNEL); | |
1778 | if (!ondisk) | |
1779 | return ERR_PTR(-ENOMEM); | |
1780 | ||
4775618d | 1781 | ret = rbd_req_sync_read(rbd_dev, rbd_dev->header_name, |
4156d998 AE |
1782 | 0, size, |
1783 | (char *) ondisk, version); | |
1784 | ||
1785 | if (ret < 0) | |
1786 | goto out_err; | |
1787 | if (WARN_ON((size_t) ret < size)) { | |
1788 | ret = -ENXIO; | |
06ecc6cb AE |
1789 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
1790 | size, ret); | |
4156d998 AE |
1791 | goto out_err; |
1792 | } | |
1793 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
1794 | ret = -ENXIO; | |
06ecc6cb | 1795 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 1796 | goto out_err; |
81e759fb | 1797 | } |
602adf40 | 1798 | |
4156d998 AE |
1799 | names_size = le64_to_cpu(ondisk->snap_names_len); |
1800 | want_count = snap_count; | |
1801 | snap_count = le32_to_cpu(ondisk->snap_count); | |
1802 | } while (snap_count != want_count); | |
00f1f36f | 1803 | |
4156d998 | 1804 | return ondisk; |
00f1f36f | 1805 | |
4156d998 AE |
1806 | out_err: |
1807 | kfree(ondisk); | |
1808 | ||
1809 | return ERR_PTR(ret); | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * reload the ondisk the header | |
1814 | */ | |
1815 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1816 | struct rbd_image_header *header) | |
1817 | { | |
1818 | struct rbd_image_header_ondisk *ondisk; | |
1819 | u64 ver = 0; | |
1820 | int ret; | |
602adf40 | 1821 | |
4156d998 AE |
1822 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
1823 | if (IS_ERR(ondisk)) | |
1824 | return PTR_ERR(ondisk); | |
1825 | ret = rbd_header_from_disk(header, ondisk); | |
1826 | if (ret >= 0) | |
1827 | header->obj_version = ver; | |
1828 | kfree(ondisk); | |
1829 | ||
1830 | return ret; | |
602adf40 YS |
1831 | } |
1832 | ||
41f38c2b | 1833 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
1834 | { |
1835 | struct rbd_snap *snap; | |
a0593290 | 1836 | struct rbd_snap *next; |
dfc5606d | 1837 | |
a0593290 | 1838 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
41f38c2b | 1839 | rbd_remove_snap_dev(snap); |
dfc5606d YS |
1840 | } |
1841 | ||
9478554a AE |
1842 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
1843 | { | |
1844 | sector_t size; | |
1845 | ||
0d7dbfce | 1846 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
1847 | return; |
1848 | ||
1849 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
1850 | dout("setting size to %llu sectors", (unsigned long long) size); | |
1851 | rbd_dev->mapping.size = (u64) size; | |
1852 | set_capacity(rbd_dev->disk, size); | |
1853 | } | |
1854 | ||
602adf40 YS |
1855 | /* |
1856 | * only read the first part of the ondisk header, without the snaps info | |
1857 | */ | |
117973fb | 1858 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
1859 | { |
1860 | int ret; | |
1861 | struct rbd_image_header h; | |
602adf40 YS |
1862 | |
1863 | ret = rbd_read_header(rbd_dev, &h); | |
1864 | if (ret < 0) | |
1865 | return ret; | |
1866 | ||
a51aa0c0 JD |
1867 | down_write(&rbd_dev->header_rwsem); |
1868 | ||
9478554a AE |
1869 | /* Update image size, and check for resize of mapped image */ |
1870 | rbd_dev->header.image_size = h.image_size; | |
1871 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 1872 | |
849b4260 | 1873 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 1874 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 1875 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
1876 | /* osd requests may still refer to snapc */ |
1877 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 1878 | |
b813623a AE |
1879 | if (hver) |
1880 | *hver = h.obj_version; | |
a71b891b | 1881 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 1882 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
1883 | rbd_dev->header.snapc = h.snapc; |
1884 | rbd_dev->header.snap_names = h.snap_names; | |
1885 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
1886 | /* Free the extra copy of the object prefix */ |
1887 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
1888 | kfree(h.object_prefix); | |
1889 | ||
304f6808 AE |
1890 | ret = rbd_dev_snaps_update(rbd_dev); |
1891 | if (!ret) | |
1892 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 1893 | |
c666601a | 1894 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1895 | |
dfc5606d | 1896 | return ret; |
602adf40 YS |
1897 | } |
1898 | ||
117973fb | 1899 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
1900 | { |
1901 | int ret; | |
1902 | ||
117973fb | 1903 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 1904 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
1905 | if (rbd_dev->image_format == 1) |
1906 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
1907 | else | |
1908 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
1909 | mutex_unlock(&ctl_mutex); |
1910 | ||
1911 | return ret; | |
1912 | } | |
1913 | ||
602adf40 YS |
1914 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
1915 | { | |
1916 | struct gendisk *disk; | |
1917 | struct request_queue *q; | |
593a9e7b | 1918 | u64 segment_size; |
602adf40 | 1919 | |
602adf40 | 1920 | /* create gendisk info */ |
602adf40 YS |
1921 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
1922 | if (!disk) | |
1fcdb8aa | 1923 | return -ENOMEM; |
602adf40 | 1924 | |
f0f8cef5 | 1925 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 1926 | rbd_dev->dev_id); |
602adf40 YS |
1927 | disk->major = rbd_dev->major; |
1928 | disk->first_minor = 0; | |
1929 | disk->fops = &rbd_bd_ops; | |
1930 | disk->private_data = rbd_dev; | |
1931 | ||
1932 | /* init rq */ | |
602adf40 YS |
1933 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); |
1934 | if (!q) | |
1935 | goto out_disk; | |
029bcbd8 | 1936 | |
593a9e7b AE |
1937 | /* We use the default size, but let's be explicit about it. */ |
1938 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1939 | ||
029bcbd8 | 1940 | /* set io sizes to object size */ |
593a9e7b AE |
1941 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1942 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1943 | blk_queue_max_segment_size(q, segment_size); | |
1944 | blk_queue_io_min(q, segment_size); | |
1945 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1946 | |
602adf40 YS |
1947 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1948 | disk->queue = q; | |
1949 | ||
1950 | q->queuedata = rbd_dev; | |
1951 | ||
1952 | rbd_dev->disk = disk; | |
602adf40 | 1953 | |
12f02944 AE |
1954 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
1955 | ||
602adf40 | 1956 | return 0; |
602adf40 YS |
1957 | out_disk: |
1958 | put_disk(disk); | |
1fcdb8aa AE |
1959 | |
1960 | return -ENOMEM; | |
602adf40 YS |
1961 | } |
1962 | ||
dfc5606d YS |
1963 | /* |
1964 | sysfs | |
1965 | */ | |
1966 | ||
593a9e7b AE |
1967 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1968 | { | |
1969 | return container_of(dev, struct rbd_device, dev); | |
1970 | } | |
1971 | ||
dfc5606d YS |
1972 | static ssize_t rbd_size_show(struct device *dev, |
1973 | struct device_attribute *attr, char *buf) | |
1974 | { | |
593a9e7b | 1975 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
1976 | sector_t size; |
1977 | ||
1978 | down_read(&rbd_dev->header_rwsem); | |
1979 | size = get_capacity(rbd_dev->disk); | |
1980 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 1981 | |
a51aa0c0 | 1982 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
1983 | } |
1984 | ||
34b13184 AE |
1985 | /* |
1986 | * Note this shows the features for whatever's mapped, which is not | |
1987 | * necessarily the base image. | |
1988 | */ | |
1989 | static ssize_t rbd_features_show(struct device *dev, | |
1990 | struct device_attribute *attr, char *buf) | |
1991 | { | |
1992 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
1993 | ||
1994 | return sprintf(buf, "0x%016llx\n", | |
1995 | (unsigned long long) rbd_dev->mapping.features); | |
1996 | } | |
1997 | ||
dfc5606d YS |
1998 | static ssize_t rbd_major_show(struct device *dev, |
1999 | struct device_attribute *attr, char *buf) | |
2000 | { | |
593a9e7b | 2001 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2002 | |
dfc5606d YS |
2003 | return sprintf(buf, "%d\n", rbd_dev->major); |
2004 | } | |
2005 | ||
2006 | static ssize_t rbd_client_id_show(struct device *dev, | |
2007 | struct device_attribute *attr, char *buf) | |
602adf40 | 2008 | { |
593a9e7b | 2009 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2010 | |
1dbb4399 AE |
2011 | return sprintf(buf, "client%lld\n", |
2012 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
2013 | } |
2014 | ||
dfc5606d YS |
2015 | static ssize_t rbd_pool_show(struct device *dev, |
2016 | struct device_attribute *attr, char *buf) | |
602adf40 | 2017 | { |
593a9e7b | 2018 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2019 | |
0d7dbfce | 2020 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
2021 | } |
2022 | ||
9bb2f334 AE |
2023 | static ssize_t rbd_pool_id_show(struct device *dev, |
2024 | struct device_attribute *attr, char *buf) | |
2025 | { | |
2026 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2027 | ||
0d7dbfce AE |
2028 | return sprintf(buf, "%llu\n", |
2029 | (unsigned long long) rbd_dev->spec->pool_id); | |
9bb2f334 AE |
2030 | } |
2031 | ||
dfc5606d YS |
2032 | static ssize_t rbd_name_show(struct device *dev, |
2033 | struct device_attribute *attr, char *buf) | |
2034 | { | |
593a9e7b | 2035 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2036 | |
a92ffdf8 AE |
2037 | if (rbd_dev->spec->image_name) |
2038 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
2039 | ||
2040 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
2041 | } |
2042 | ||
589d30e0 AE |
2043 | static ssize_t rbd_image_id_show(struct device *dev, |
2044 | struct device_attribute *attr, char *buf) | |
2045 | { | |
2046 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2047 | ||
0d7dbfce | 2048 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
2049 | } |
2050 | ||
34b13184 AE |
2051 | /* |
2052 | * Shows the name of the currently-mapped snapshot (or | |
2053 | * RBD_SNAP_HEAD_NAME for the base image). | |
2054 | */ | |
dfc5606d YS |
2055 | static ssize_t rbd_snap_show(struct device *dev, |
2056 | struct device_attribute *attr, | |
2057 | char *buf) | |
2058 | { | |
593a9e7b | 2059 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2060 | |
0d7dbfce | 2061 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
2062 | } |
2063 | ||
86b00e0d AE |
2064 | /* |
2065 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
2066 | * for the parent image. If there is no parent, simply shows | |
2067 | * "(no parent image)". | |
2068 | */ | |
2069 | static ssize_t rbd_parent_show(struct device *dev, | |
2070 | struct device_attribute *attr, | |
2071 | char *buf) | |
2072 | { | |
2073 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2074 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
2075 | int count; | |
2076 | char *bufp = buf; | |
2077 | ||
2078 | if (!spec) | |
2079 | return sprintf(buf, "(no parent image)\n"); | |
2080 | ||
2081 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
2082 | (unsigned long long) spec->pool_id, spec->pool_name); | |
2083 | if (count < 0) | |
2084 | return count; | |
2085 | bufp += count; | |
2086 | ||
2087 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
2088 | spec->image_name ? spec->image_name : "(unknown)"); | |
2089 | if (count < 0) | |
2090 | return count; | |
2091 | bufp += count; | |
2092 | ||
2093 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
2094 | (unsigned long long) spec->snap_id, spec->snap_name); | |
2095 | if (count < 0) | |
2096 | return count; | |
2097 | bufp += count; | |
2098 | ||
2099 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
2100 | if (count < 0) | |
2101 | return count; | |
2102 | bufp += count; | |
2103 | ||
2104 | return (ssize_t) (bufp - buf); | |
2105 | } | |
2106 | ||
dfc5606d YS |
2107 | static ssize_t rbd_image_refresh(struct device *dev, |
2108 | struct device_attribute *attr, | |
2109 | const char *buf, | |
2110 | size_t size) | |
2111 | { | |
593a9e7b | 2112 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2113 | int ret; |
602adf40 | 2114 | |
117973fb | 2115 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
2116 | |
2117 | return ret < 0 ? ret : size; | |
dfc5606d | 2118 | } |
602adf40 | 2119 | |
dfc5606d | 2120 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 2121 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
2122 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
2123 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2124 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2125 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 2126 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 2127 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
2128 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
2129 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 2130 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
2131 | |
2132 | static struct attribute *rbd_attrs[] = { | |
2133 | &dev_attr_size.attr, | |
34b13184 | 2134 | &dev_attr_features.attr, |
dfc5606d YS |
2135 | &dev_attr_major.attr, |
2136 | &dev_attr_client_id.attr, | |
2137 | &dev_attr_pool.attr, | |
9bb2f334 | 2138 | &dev_attr_pool_id.attr, |
dfc5606d | 2139 | &dev_attr_name.attr, |
589d30e0 | 2140 | &dev_attr_image_id.attr, |
dfc5606d | 2141 | &dev_attr_current_snap.attr, |
86b00e0d | 2142 | &dev_attr_parent.attr, |
dfc5606d | 2143 | &dev_attr_refresh.attr, |
dfc5606d YS |
2144 | NULL |
2145 | }; | |
2146 | ||
2147 | static struct attribute_group rbd_attr_group = { | |
2148 | .attrs = rbd_attrs, | |
2149 | }; | |
2150 | ||
2151 | static const struct attribute_group *rbd_attr_groups[] = { | |
2152 | &rbd_attr_group, | |
2153 | NULL | |
2154 | }; | |
2155 | ||
2156 | static void rbd_sysfs_dev_release(struct device *dev) | |
2157 | { | |
2158 | } | |
2159 | ||
2160 | static struct device_type rbd_device_type = { | |
2161 | .name = "rbd", | |
2162 | .groups = rbd_attr_groups, | |
2163 | .release = rbd_sysfs_dev_release, | |
2164 | }; | |
2165 | ||
2166 | ||
2167 | /* | |
2168 | sysfs - snapshots | |
2169 | */ | |
2170 | ||
2171 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2172 | struct device_attribute *attr, | |
2173 | char *buf) | |
2174 | { | |
2175 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2176 | ||
3591538f | 2177 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2178 | } |
2179 | ||
2180 | static ssize_t rbd_snap_id_show(struct device *dev, | |
2181 | struct device_attribute *attr, | |
2182 | char *buf) | |
2183 | { | |
2184 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2185 | ||
3591538f | 2186 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
2187 | } |
2188 | ||
34b13184 AE |
2189 | static ssize_t rbd_snap_features_show(struct device *dev, |
2190 | struct device_attribute *attr, | |
2191 | char *buf) | |
2192 | { | |
2193 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2194 | ||
2195 | return sprintf(buf, "0x%016llx\n", | |
2196 | (unsigned long long) snap->features); | |
2197 | } | |
2198 | ||
dfc5606d YS |
2199 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2200 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2201 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2202 | |
2203 | static struct attribute *rbd_snap_attrs[] = { | |
2204 | &dev_attr_snap_size.attr, | |
2205 | &dev_attr_snap_id.attr, | |
34b13184 | 2206 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2207 | NULL, |
2208 | }; | |
2209 | ||
2210 | static struct attribute_group rbd_snap_attr_group = { | |
2211 | .attrs = rbd_snap_attrs, | |
2212 | }; | |
2213 | ||
2214 | static void rbd_snap_dev_release(struct device *dev) | |
2215 | { | |
2216 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2217 | kfree(snap->name); | |
2218 | kfree(snap); | |
2219 | } | |
2220 | ||
2221 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2222 | &rbd_snap_attr_group, | |
2223 | NULL | |
2224 | }; | |
2225 | ||
2226 | static struct device_type rbd_snap_device_type = { | |
2227 | .groups = rbd_snap_attr_groups, | |
2228 | .release = rbd_snap_dev_release, | |
2229 | }; | |
2230 | ||
8b8fb99c AE |
2231 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
2232 | { | |
2233 | kref_get(&spec->kref); | |
2234 | ||
2235 | return spec; | |
2236 | } | |
2237 | ||
2238 | static void rbd_spec_free(struct kref *kref); | |
2239 | static void rbd_spec_put(struct rbd_spec *spec) | |
2240 | { | |
2241 | if (spec) | |
2242 | kref_put(&spec->kref, rbd_spec_free); | |
2243 | } | |
2244 | ||
2245 | static struct rbd_spec *rbd_spec_alloc(void) | |
2246 | { | |
2247 | struct rbd_spec *spec; | |
2248 | ||
2249 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
2250 | if (!spec) | |
2251 | return NULL; | |
2252 | kref_init(&spec->kref); | |
2253 | ||
2254 | rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */ | |
2255 | ||
2256 | return spec; | |
2257 | } | |
2258 | ||
2259 | static void rbd_spec_free(struct kref *kref) | |
2260 | { | |
2261 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
2262 | ||
2263 | kfree(spec->pool_name); | |
2264 | kfree(spec->image_id); | |
2265 | kfree(spec->image_name); | |
2266 | kfree(spec->snap_name); | |
2267 | kfree(spec); | |
2268 | } | |
2269 | ||
c53d5893 AE |
2270 | struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
2271 | struct rbd_spec *spec) | |
2272 | { | |
2273 | struct rbd_device *rbd_dev; | |
2274 | ||
2275 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
2276 | if (!rbd_dev) | |
2277 | return NULL; | |
2278 | ||
2279 | spin_lock_init(&rbd_dev->lock); | |
d78b650a | 2280 | atomic_set(&rbd_dev->exists, 0); |
c53d5893 AE |
2281 | INIT_LIST_HEAD(&rbd_dev->node); |
2282 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
2283 | init_rwsem(&rbd_dev->header_rwsem); | |
2284 | ||
2285 | rbd_dev->spec = spec; | |
2286 | rbd_dev->rbd_client = rbdc; | |
2287 | ||
0903e875 AE |
2288 | /* Initialize the layout used for all rbd requests */ |
2289 | ||
2290 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2291 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
2292 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
2293 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
2294 | ||
c53d5893 AE |
2295 | return rbd_dev; |
2296 | } | |
2297 | ||
2298 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
2299 | { | |
86b00e0d | 2300 | rbd_spec_put(rbd_dev->parent_spec); |
c53d5893 AE |
2301 | kfree(rbd_dev->header_name); |
2302 | rbd_put_client(rbd_dev->rbd_client); | |
2303 | rbd_spec_put(rbd_dev->spec); | |
2304 | kfree(rbd_dev); | |
2305 | } | |
2306 | ||
304f6808 AE |
2307 | static bool rbd_snap_registered(struct rbd_snap *snap) |
2308 | { | |
2309 | bool ret = snap->dev.type == &rbd_snap_device_type; | |
2310 | bool reg = device_is_registered(&snap->dev); | |
2311 | ||
2312 | rbd_assert(!ret ^ reg); | |
2313 | ||
2314 | return ret; | |
2315 | } | |
2316 | ||
41f38c2b | 2317 | static void rbd_remove_snap_dev(struct rbd_snap *snap) |
dfc5606d YS |
2318 | { |
2319 | list_del(&snap->node); | |
304f6808 AE |
2320 | if (device_is_registered(&snap->dev)) |
2321 | device_unregister(&snap->dev); | |
dfc5606d YS |
2322 | } |
2323 | ||
14e7085d | 2324 | static int rbd_register_snap_dev(struct rbd_snap *snap, |
dfc5606d YS |
2325 | struct device *parent) |
2326 | { | |
2327 | struct device *dev = &snap->dev; | |
2328 | int ret; | |
2329 | ||
2330 | dev->type = &rbd_snap_device_type; | |
2331 | dev->parent = parent; | |
2332 | dev->release = rbd_snap_dev_release; | |
d4b125e9 | 2333 | dev_set_name(dev, "%s%s", RBD_SNAP_DEV_NAME_PREFIX, snap->name); |
304f6808 AE |
2334 | dout("%s: registering device for snapshot %s\n", __func__, snap->name); |
2335 | ||
dfc5606d YS |
2336 | ret = device_register(dev); |
2337 | ||
2338 | return ret; | |
2339 | } | |
2340 | ||
4e891e0a | 2341 | static struct rbd_snap *__rbd_add_snap_dev(struct rbd_device *rbd_dev, |
c8d18425 | 2342 | const char *snap_name, |
34b13184 AE |
2343 | u64 snap_id, u64 snap_size, |
2344 | u64 snap_features) | |
dfc5606d | 2345 | { |
4e891e0a | 2346 | struct rbd_snap *snap; |
dfc5606d | 2347 | int ret; |
4e891e0a AE |
2348 | |
2349 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 2350 | if (!snap) |
4e891e0a AE |
2351 | return ERR_PTR(-ENOMEM); |
2352 | ||
2353 | ret = -ENOMEM; | |
c8d18425 | 2354 | snap->name = kstrdup(snap_name, GFP_KERNEL); |
4e891e0a AE |
2355 | if (!snap->name) |
2356 | goto err; | |
2357 | ||
c8d18425 AE |
2358 | snap->id = snap_id; |
2359 | snap->size = snap_size; | |
34b13184 | 2360 | snap->features = snap_features; |
4e891e0a AE |
2361 | |
2362 | return snap; | |
2363 | ||
dfc5606d YS |
2364 | err: |
2365 | kfree(snap->name); | |
2366 | kfree(snap); | |
4e891e0a AE |
2367 | |
2368 | return ERR_PTR(ret); | |
dfc5606d YS |
2369 | } |
2370 | ||
cd892126 AE |
2371 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
2372 | u64 *snap_size, u64 *snap_features) | |
2373 | { | |
2374 | char *snap_name; | |
2375 | ||
2376 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
2377 | ||
2378 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
2379 | *snap_features = 0; /* No features for v1 */ | |
2380 | ||
2381 | /* Skip over names until we find the one we are looking for */ | |
2382 | ||
2383 | snap_name = rbd_dev->header.snap_names; | |
2384 | while (which--) | |
2385 | snap_name += strlen(snap_name) + 1; | |
2386 | ||
2387 | return snap_name; | |
2388 | } | |
2389 | ||
9d475de5 AE |
2390 | /* |
2391 | * Get the size and object order for an image snapshot, or if | |
2392 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
2393 | * image. | |
2394 | */ | |
2395 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
2396 | u8 *order, u64 *snap_size) | |
2397 | { | |
2398 | __le64 snapid = cpu_to_le64(snap_id); | |
2399 | int ret; | |
2400 | struct { | |
2401 | u8 order; | |
2402 | __le64 size; | |
2403 | } __attribute__ ((packed)) size_buf = { 0 }; | |
2404 | ||
2405 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2406 | "rbd", "get_size", | |
2407 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2408 | (char *) &size_buf, sizeof (size_buf), NULL); |
9d475de5 AE |
2409 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2410 | if (ret < 0) | |
2411 | return ret; | |
2412 | ||
2413 | *order = size_buf.order; | |
2414 | *snap_size = le64_to_cpu(size_buf.size); | |
2415 | ||
2416 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2417 | (unsigned long long) snap_id, (unsigned int) *order, | |
2418 | (unsigned long long) *snap_size); | |
2419 | ||
2420 | return 0; | |
2421 | } | |
2422 | ||
2423 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2424 | { | |
2425 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2426 | &rbd_dev->header.obj_order, | |
2427 | &rbd_dev->header.image_size); | |
2428 | } | |
2429 | ||
1e130199 AE |
2430 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2431 | { | |
2432 | void *reply_buf; | |
2433 | int ret; | |
2434 | void *p; | |
2435 | ||
2436 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2437 | if (!reply_buf) | |
2438 | return -ENOMEM; | |
2439 | ||
2440 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2441 | "rbd", "get_object_prefix", | |
2442 | NULL, 0, | |
07b2391f | 2443 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL); |
1e130199 AE |
2444 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2445 | if (ret < 0) | |
2446 | goto out; | |
a0ea3a40 | 2447 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
1e130199 AE |
2448 | |
2449 | p = reply_buf; | |
2450 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2451 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2452 | NULL, GFP_NOIO); | |
2453 | ||
2454 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2455 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2456 | rbd_dev->header.object_prefix = NULL; | |
2457 | } else { | |
2458 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2459 | } | |
2460 | ||
2461 | out: | |
2462 | kfree(reply_buf); | |
2463 | ||
2464 | return ret; | |
2465 | } | |
2466 | ||
b1b5402a AE |
2467 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2468 | u64 *snap_features) | |
2469 | { | |
2470 | __le64 snapid = cpu_to_le64(snap_id); | |
2471 | struct { | |
2472 | __le64 features; | |
2473 | __le64 incompat; | |
2474 | } features_buf = { 0 }; | |
d889140c | 2475 | u64 incompat; |
b1b5402a AE |
2476 | int ret; |
2477 | ||
2478 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2479 | "rbd", "get_features", | |
2480 | (char *) &snapid, sizeof (snapid), | |
2481 | (char *) &features_buf, sizeof (features_buf), | |
07b2391f | 2482 | NULL); |
b1b5402a AE |
2483 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2484 | if (ret < 0) | |
2485 | return ret; | |
d889140c AE |
2486 | |
2487 | incompat = le64_to_cpu(features_buf.incompat); | |
2488 | if (incompat & ~RBD_FEATURES_ALL) | |
b8f5c6ed | 2489 | return -ENXIO; |
d889140c | 2490 | |
b1b5402a AE |
2491 | *snap_features = le64_to_cpu(features_buf.features); |
2492 | ||
2493 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2494 | (unsigned long long) snap_id, | |
2495 | (unsigned long long) *snap_features, | |
2496 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2497 | ||
2498 | return 0; | |
2499 | } | |
2500 | ||
2501 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2502 | { | |
2503 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2504 | &rbd_dev->header.features); | |
2505 | } | |
2506 | ||
86b00e0d AE |
2507 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
2508 | { | |
2509 | struct rbd_spec *parent_spec; | |
2510 | size_t size; | |
2511 | void *reply_buf = NULL; | |
2512 | __le64 snapid; | |
2513 | void *p; | |
2514 | void *end; | |
2515 | char *image_id; | |
2516 | u64 overlap; | |
86b00e0d AE |
2517 | int ret; |
2518 | ||
2519 | parent_spec = rbd_spec_alloc(); | |
2520 | if (!parent_spec) | |
2521 | return -ENOMEM; | |
2522 | ||
2523 | size = sizeof (__le64) + /* pool_id */ | |
2524 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
2525 | sizeof (__le64) + /* snap_id */ | |
2526 | sizeof (__le64); /* overlap */ | |
2527 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2528 | if (!reply_buf) { | |
2529 | ret = -ENOMEM; | |
2530 | goto out_err; | |
2531 | } | |
2532 | ||
2533 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
2534 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2535 | "rbd", "get_parent", | |
2536 | (char *) &snapid, sizeof (snapid), | |
07b2391f | 2537 | (char *) reply_buf, size, NULL); |
86b00e0d AE |
2538 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2539 | if (ret < 0) | |
2540 | goto out_err; | |
2541 | ||
2542 | ret = -ERANGE; | |
2543 | p = reply_buf; | |
2544 | end = (char *) reply_buf + size; | |
2545 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); | |
2546 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
2547 | goto out; /* No parent? No problem. */ | |
2548 | ||
0903e875 AE |
2549 | /* The ceph file layout needs to fit pool id in 32 bits */ |
2550 | ||
2551 | ret = -EIO; | |
2552 | if (WARN_ON(parent_spec->pool_id > (u64) U32_MAX)) | |
2553 | goto out; | |
2554 | ||
979ed480 | 2555 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
2556 | if (IS_ERR(image_id)) { |
2557 | ret = PTR_ERR(image_id); | |
2558 | goto out_err; | |
2559 | } | |
2560 | parent_spec->image_id = image_id; | |
2561 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
2562 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
2563 | ||
2564 | rbd_dev->parent_overlap = overlap; | |
2565 | rbd_dev->parent_spec = parent_spec; | |
2566 | parent_spec = NULL; /* rbd_dev now owns this */ | |
2567 | out: | |
2568 | ret = 0; | |
2569 | out_err: | |
2570 | kfree(reply_buf); | |
2571 | rbd_spec_put(parent_spec); | |
2572 | ||
2573 | return ret; | |
2574 | } | |
2575 | ||
9e15b77d AE |
2576 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
2577 | { | |
2578 | size_t image_id_size; | |
2579 | char *image_id; | |
2580 | void *p; | |
2581 | void *end; | |
2582 | size_t size; | |
2583 | void *reply_buf = NULL; | |
2584 | size_t len = 0; | |
2585 | char *image_name = NULL; | |
2586 | int ret; | |
2587 | ||
2588 | rbd_assert(!rbd_dev->spec->image_name); | |
2589 | ||
69e7a02f AE |
2590 | len = strlen(rbd_dev->spec->image_id); |
2591 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
2592 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
2593 | if (!image_id) | |
2594 | return NULL; | |
2595 | ||
2596 | p = image_id; | |
2597 | end = (char *) image_id + image_id_size; | |
69e7a02f | 2598 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len); |
9e15b77d AE |
2599 | |
2600 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
2601 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2602 | if (!reply_buf) | |
2603 | goto out; | |
2604 | ||
2605 | ret = rbd_req_sync_exec(rbd_dev, RBD_DIRECTORY, | |
2606 | "rbd", "dir_get_name", | |
2607 | image_id, image_id_size, | |
07b2391f | 2608 | (char *) reply_buf, size, NULL); |
9e15b77d AE |
2609 | if (ret < 0) |
2610 | goto out; | |
2611 | p = reply_buf; | |
2612 | end = (char *) reply_buf + size; | |
2613 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); | |
2614 | if (IS_ERR(image_name)) | |
2615 | image_name = NULL; | |
2616 | else | |
2617 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
2618 | out: | |
2619 | kfree(reply_buf); | |
2620 | kfree(image_id); | |
2621 | ||
2622 | return image_name; | |
2623 | } | |
2624 | ||
2625 | /* | |
2626 | * When a parent image gets probed, we only have the pool, image, | |
2627 | * and snapshot ids but not the names of any of them. This call | |
2628 | * is made later to fill in those names. It has to be done after | |
2629 | * rbd_dev_snaps_update() has completed because some of the | |
2630 | * information (in particular, snapshot name) is not available | |
2631 | * until then. | |
2632 | */ | |
2633 | static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev) | |
2634 | { | |
2635 | struct ceph_osd_client *osdc; | |
2636 | const char *name; | |
2637 | void *reply_buf = NULL; | |
2638 | int ret; | |
2639 | ||
2640 | if (rbd_dev->spec->pool_name) | |
2641 | return 0; /* Already have the names */ | |
2642 | ||
2643 | /* Look up the pool name */ | |
2644 | ||
2645 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2646 | name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id); | |
935dc89f AE |
2647 | if (!name) { |
2648 | rbd_warn(rbd_dev, "there is no pool with id %llu", | |
2649 | rbd_dev->spec->pool_id); /* Really a BUG() */ | |
2650 | return -EIO; | |
2651 | } | |
9e15b77d AE |
2652 | |
2653 | rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL); | |
2654 | if (!rbd_dev->spec->pool_name) | |
2655 | return -ENOMEM; | |
2656 | ||
2657 | /* Fetch the image name; tolerate failure here */ | |
2658 | ||
2659 | name = rbd_dev_image_name(rbd_dev); | |
69e7a02f | 2660 | if (name) |
9e15b77d | 2661 | rbd_dev->spec->image_name = (char *) name; |
69e7a02f | 2662 | else |
06ecc6cb | 2663 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d AE |
2664 | |
2665 | /* Look up the snapshot name. */ | |
2666 | ||
2667 | name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id); | |
2668 | if (!name) { | |
935dc89f AE |
2669 | rbd_warn(rbd_dev, "no snapshot with id %llu", |
2670 | rbd_dev->spec->snap_id); /* Really a BUG() */ | |
9e15b77d AE |
2671 | ret = -EIO; |
2672 | goto out_err; | |
2673 | } | |
2674 | rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL); | |
2675 | if(!rbd_dev->spec->snap_name) | |
2676 | goto out_err; | |
2677 | ||
2678 | return 0; | |
2679 | out_err: | |
2680 | kfree(reply_buf); | |
2681 | kfree(rbd_dev->spec->pool_name); | |
2682 | rbd_dev->spec->pool_name = NULL; | |
2683 | ||
2684 | return ret; | |
2685 | } | |
2686 | ||
6e14b1a6 | 2687 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
2688 | { |
2689 | size_t size; | |
2690 | int ret; | |
2691 | void *reply_buf; | |
2692 | void *p; | |
2693 | void *end; | |
2694 | u64 seq; | |
2695 | u32 snap_count; | |
2696 | struct ceph_snap_context *snapc; | |
2697 | u32 i; | |
2698 | ||
2699 | /* | |
2700 | * We'll need room for the seq value (maximum snapshot id), | |
2701 | * snapshot count, and array of that many snapshot ids. | |
2702 | * For now we have a fixed upper limit on the number we're | |
2703 | * prepared to receive. | |
2704 | */ | |
2705 | size = sizeof (__le64) + sizeof (__le32) + | |
2706 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
2707 | reply_buf = kzalloc(size, GFP_KERNEL); | |
2708 | if (!reply_buf) | |
2709 | return -ENOMEM; | |
2710 | ||
2711 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2712 | "rbd", "get_snapcontext", | |
2713 | NULL, 0, | |
07b2391f | 2714 | reply_buf, size, ver); |
35d489f9 AE |
2715 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2716 | if (ret < 0) | |
2717 | goto out; | |
2718 | ||
2719 | ret = -ERANGE; | |
2720 | p = reply_buf; | |
2721 | end = (char *) reply_buf + size; | |
2722 | ceph_decode_64_safe(&p, end, seq, out); | |
2723 | ceph_decode_32_safe(&p, end, snap_count, out); | |
2724 | ||
2725 | /* | |
2726 | * Make sure the reported number of snapshot ids wouldn't go | |
2727 | * beyond the end of our buffer. But before checking that, | |
2728 | * make sure the computed size of the snapshot context we | |
2729 | * allocate is representable in a size_t. | |
2730 | */ | |
2731 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
2732 | / sizeof (u64)) { | |
2733 | ret = -EINVAL; | |
2734 | goto out; | |
2735 | } | |
2736 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
2737 | goto out; | |
2738 | ||
2739 | size = sizeof (struct ceph_snap_context) + | |
2740 | snap_count * sizeof (snapc->snaps[0]); | |
2741 | snapc = kmalloc(size, GFP_KERNEL); | |
2742 | if (!snapc) { | |
2743 | ret = -ENOMEM; | |
2744 | goto out; | |
2745 | } | |
2746 | ||
2747 | atomic_set(&snapc->nref, 1); | |
2748 | snapc->seq = seq; | |
2749 | snapc->num_snaps = snap_count; | |
2750 | for (i = 0; i < snap_count; i++) | |
2751 | snapc->snaps[i] = ceph_decode_64(&p); | |
2752 | ||
2753 | rbd_dev->header.snapc = snapc; | |
2754 | ||
2755 | dout(" snap context seq = %llu, snap_count = %u\n", | |
2756 | (unsigned long long) seq, (unsigned int) snap_count); | |
2757 | ||
2758 | out: | |
2759 | kfree(reply_buf); | |
2760 | ||
2761 | return 0; | |
2762 | } | |
2763 | ||
b8b1e2db AE |
2764 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
2765 | { | |
2766 | size_t size; | |
2767 | void *reply_buf; | |
2768 | __le64 snap_id; | |
2769 | int ret; | |
2770 | void *p; | |
2771 | void *end; | |
b8b1e2db AE |
2772 | char *snap_name; |
2773 | ||
2774 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
2775 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2776 | if (!reply_buf) | |
2777 | return ERR_PTR(-ENOMEM); | |
2778 | ||
2779 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
2780 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2781 | "rbd", "get_snapshot_name", | |
2782 | (char *) &snap_id, sizeof (snap_id), | |
07b2391f | 2783 | reply_buf, size, NULL); |
b8b1e2db AE |
2784 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2785 | if (ret < 0) | |
2786 | goto out; | |
2787 | ||
2788 | p = reply_buf; | |
2789 | end = (char *) reply_buf + size; | |
e5c35534 | 2790 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
b8b1e2db AE |
2791 | if (IS_ERR(snap_name)) { |
2792 | ret = PTR_ERR(snap_name); | |
2793 | goto out; | |
2794 | } else { | |
2795 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
2796 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
2797 | } | |
2798 | kfree(reply_buf); | |
2799 | ||
2800 | return snap_name; | |
2801 | out: | |
2802 | kfree(reply_buf); | |
2803 | ||
2804 | return ERR_PTR(ret); | |
2805 | } | |
2806 | ||
2807 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2808 | u64 *snap_size, u64 *snap_features) | |
2809 | { | |
e0b49868 | 2810 | u64 snap_id; |
b8b1e2db AE |
2811 | u8 order; |
2812 | int ret; | |
2813 | ||
2814 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
2815 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
2816 | if (ret) | |
2817 | return ERR_PTR(ret); | |
2818 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
2819 | if (ret) | |
2820 | return ERR_PTR(ret); | |
2821 | ||
2822 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
2823 | } | |
2824 | ||
2825 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2826 | u64 *snap_size, u64 *snap_features) | |
2827 | { | |
2828 | if (rbd_dev->image_format == 1) | |
2829 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
2830 | snap_size, snap_features); | |
2831 | if (rbd_dev->image_format == 2) | |
2832 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
2833 | snap_size, snap_features); | |
2834 | return ERR_PTR(-EINVAL); | |
2835 | } | |
2836 | ||
117973fb AE |
2837 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
2838 | { | |
2839 | int ret; | |
2840 | __u8 obj_order; | |
2841 | ||
2842 | down_write(&rbd_dev->header_rwsem); | |
2843 | ||
2844 | /* Grab old order first, to see if it changes */ | |
2845 | ||
2846 | obj_order = rbd_dev->header.obj_order, | |
2847 | ret = rbd_dev_v2_image_size(rbd_dev); | |
2848 | if (ret) | |
2849 | goto out; | |
2850 | if (rbd_dev->header.obj_order != obj_order) { | |
2851 | ret = -EIO; | |
2852 | goto out; | |
2853 | } | |
2854 | rbd_update_mapping_size(rbd_dev); | |
2855 | ||
2856 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
2857 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
2858 | if (ret) | |
2859 | goto out; | |
2860 | ret = rbd_dev_snaps_update(rbd_dev); | |
2861 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
2862 | if (ret) | |
2863 | goto out; | |
2864 | ret = rbd_dev_snaps_register(rbd_dev); | |
2865 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
2866 | out: | |
2867 | up_write(&rbd_dev->header_rwsem); | |
2868 | ||
2869 | return ret; | |
2870 | } | |
2871 | ||
dfc5606d | 2872 | /* |
35938150 AE |
2873 | * Scan the rbd device's current snapshot list and compare it to the |
2874 | * newly-received snapshot context. Remove any existing snapshots | |
2875 | * not present in the new snapshot context. Add a new snapshot for | |
2876 | * any snaphots in the snapshot context not in the current list. | |
2877 | * And verify there are no changes to snapshots we already know | |
2878 | * about. | |
2879 | * | |
2880 | * Assumes the snapshots in the snapshot context are sorted by | |
2881 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
2882 | * are also maintained in that order.) | |
dfc5606d | 2883 | */ |
304f6808 | 2884 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 2885 | { |
35938150 AE |
2886 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
2887 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
2888 | struct list_head *head = &rbd_dev->snaps; |
2889 | struct list_head *links = head->next; | |
2890 | u32 index = 0; | |
dfc5606d | 2891 | |
9fcbb800 | 2892 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
2893 | while (index < snap_count || links != head) { |
2894 | u64 snap_id; | |
2895 | struct rbd_snap *snap; | |
cd892126 AE |
2896 | char *snap_name; |
2897 | u64 snap_size = 0; | |
2898 | u64 snap_features = 0; | |
dfc5606d | 2899 | |
35938150 AE |
2900 | snap_id = index < snap_count ? snapc->snaps[index] |
2901 | : CEPH_NOSNAP; | |
2902 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
2903 | : NULL; | |
aafb230e | 2904 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 2905 | |
35938150 AE |
2906 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
2907 | struct list_head *next = links->next; | |
dfc5606d | 2908 | |
35938150 | 2909 | /* Existing snapshot not in the new snap context */ |
dfc5606d | 2910 | |
0d7dbfce | 2911 | if (rbd_dev->spec->snap_id == snap->id) |
d78b650a | 2912 | atomic_set(&rbd_dev->exists, 0); |
41f38c2b | 2913 | rbd_remove_snap_dev(snap); |
9fcbb800 | 2914 | dout("%ssnap id %llu has been removed\n", |
0d7dbfce AE |
2915 | rbd_dev->spec->snap_id == snap->id ? |
2916 | "mapped " : "", | |
9fcbb800 | 2917 | (unsigned long long) snap->id); |
35938150 AE |
2918 | |
2919 | /* Done with this list entry; advance */ | |
2920 | ||
2921 | links = next; | |
dfc5606d YS |
2922 | continue; |
2923 | } | |
35938150 | 2924 | |
b8b1e2db AE |
2925 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
2926 | &snap_size, &snap_features); | |
cd892126 AE |
2927 | if (IS_ERR(snap_name)) |
2928 | return PTR_ERR(snap_name); | |
2929 | ||
9fcbb800 AE |
2930 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
2931 | (unsigned long long) snap_id); | |
35938150 AE |
2932 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
2933 | struct rbd_snap *new_snap; | |
2934 | ||
2935 | /* We haven't seen this snapshot before */ | |
2936 | ||
c8d18425 | 2937 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 2938 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
2939 | if (IS_ERR(new_snap)) { |
2940 | int err = PTR_ERR(new_snap); | |
2941 | ||
2942 | dout(" failed to add dev, error %d\n", err); | |
2943 | ||
2944 | return err; | |
2945 | } | |
35938150 AE |
2946 | |
2947 | /* New goes before existing, or at end of list */ | |
2948 | ||
9fcbb800 | 2949 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
2950 | if (snap) |
2951 | list_add_tail(&new_snap->node, &snap->node); | |
2952 | else | |
523f3258 | 2953 | list_add_tail(&new_snap->node, head); |
35938150 AE |
2954 | } else { |
2955 | /* Already have this one */ | |
2956 | ||
9fcbb800 AE |
2957 | dout(" already present\n"); |
2958 | ||
cd892126 | 2959 | rbd_assert(snap->size == snap_size); |
aafb230e | 2960 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 2961 | rbd_assert(snap->features == snap_features); |
35938150 AE |
2962 | |
2963 | /* Done with this list entry; advance */ | |
2964 | ||
2965 | links = links->next; | |
dfc5606d | 2966 | } |
35938150 AE |
2967 | |
2968 | /* Advance to the next entry in the snapshot context */ | |
2969 | ||
2970 | index++; | |
dfc5606d | 2971 | } |
9fcbb800 | 2972 | dout("%s: done\n", __func__); |
dfc5606d YS |
2973 | |
2974 | return 0; | |
2975 | } | |
2976 | ||
304f6808 AE |
2977 | /* |
2978 | * Scan the list of snapshots and register the devices for any that | |
2979 | * have not already been registered. | |
2980 | */ | |
2981 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
2982 | { | |
2983 | struct rbd_snap *snap; | |
2984 | int ret = 0; | |
2985 | ||
2986 | dout("%s called\n", __func__); | |
86ff77bb AE |
2987 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
2988 | return -EIO; | |
304f6808 AE |
2989 | |
2990 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
2991 | if (!rbd_snap_registered(snap)) { | |
2992 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
2993 | if (ret < 0) | |
2994 | break; | |
2995 | } | |
2996 | } | |
2997 | dout("%s: returning %d\n", __func__, ret); | |
2998 | ||
2999 | return ret; | |
3000 | } | |
3001 | ||
dfc5606d YS |
3002 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
3003 | { | |
dfc5606d | 3004 | struct device *dev; |
cd789ab9 | 3005 | int ret; |
dfc5606d YS |
3006 | |
3007 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 3008 | |
cd789ab9 | 3009 | dev = &rbd_dev->dev; |
dfc5606d YS |
3010 | dev->bus = &rbd_bus_type; |
3011 | dev->type = &rbd_device_type; | |
3012 | dev->parent = &rbd_root_dev; | |
3013 | dev->release = rbd_dev_release; | |
de71a297 | 3014 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 3015 | ret = device_register(dev); |
dfc5606d | 3016 | |
dfc5606d | 3017 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 3018 | |
dfc5606d | 3019 | return ret; |
602adf40 YS |
3020 | } |
3021 | ||
dfc5606d YS |
3022 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
3023 | { | |
3024 | device_unregister(&rbd_dev->dev); | |
3025 | } | |
3026 | ||
e2839308 | 3027 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
3028 | |
3029 | /* | |
499afd5b AE |
3030 | * Get a unique rbd identifier for the given new rbd_dev, and add |
3031 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 3032 | */ |
e2839308 | 3033 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 3034 | { |
e2839308 | 3035 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
3036 | |
3037 | spin_lock(&rbd_dev_list_lock); | |
3038 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
3039 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
3040 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
3041 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 3042 | } |
b7f23c36 | 3043 | |
1ddbe94e | 3044 | /* |
499afd5b AE |
3045 | * Remove an rbd_dev from the global list, and record that its |
3046 | * identifier is no longer in use. | |
1ddbe94e | 3047 | */ |
e2839308 | 3048 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 3049 | { |
d184f6bf | 3050 | struct list_head *tmp; |
de71a297 | 3051 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
3052 | int max_id; |
3053 | ||
aafb230e | 3054 | rbd_assert(rbd_id > 0); |
499afd5b | 3055 | |
e2839308 AE |
3056 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
3057 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
3058 | spin_lock(&rbd_dev_list_lock); |
3059 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
3060 | |
3061 | /* | |
3062 | * If the id being "put" is not the current maximum, there | |
3063 | * is nothing special we need to do. | |
3064 | */ | |
e2839308 | 3065 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
3066 | spin_unlock(&rbd_dev_list_lock); |
3067 | return; | |
3068 | } | |
3069 | ||
3070 | /* | |
3071 | * We need to update the current maximum id. Search the | |
3072 | * list to find out what it is. We're more likely to find | |
3073 | * the maximum at the end, so search the list backward. | |
3074 | */ | |
3075 | max_id = 0; | |
3076 | list_for_each_prev(tmp, &rbd_dev_list) { | |
3077 | struct rbd_device *rbd_dev; | |
3078 | ||
3079 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
3080 | if (rbd_dev->dev_id > max_id) |
3081 | max_id = rbd_dev->dev_id; | |
d184f6bf | 3082 | } |
499afd5b | 3083 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 3084 | |
1ddbe94e | 3085 | /* |
e2839308 | 3086 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
3087 | * which case it now accurately reflects the new maximum. |
3088 | * Be careful not to overwrite the maximum value in that | |
3089 | * case. | |
1ddbe94e | 3090 | */ |
e2839308 AE |
3091 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
3092 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
3093 | } |
3094 | ||
e28fff26 AE |
3095 | /* |
3096 | * Skips over white space at *buf, and updates *buf to point to the | |
3097 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
3098 | * the token (string of non-white space characters) found. Note |
3099 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
3100 | */ |
3101 | static inline size_t next_token(const char **buf) | |
3102 | { | |
3103 | /* | |
3104 | * These are the characters that produce nonzero for | |
3105 | * isspace() in the "C" and "POSIX" locales. | |
3106 | */ | |
3107 | const char *spaces = " \f\n\r\t\v"; | |
3108 | ||
3109 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
3110 | ||
3111 | return strcspn(*buf, spaces); /* Return token length */ | |
3112 | } | |
3113 | ||
3114 | /* | |
3115 | * Finds the next token in *buf, and if the provided token buffer is | |
3116 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
3117 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
3118 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
3119 | * |
3120 | * Returns the length of the token found (not including the '\0'). | |
3121 | * Return value will be 0 if no token is found, and it will be >= | |
3122 | * token_size if the token would not fit. | |
3123 | * | |
593a9e7b | 3124 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
3125 | * found token. Note that this occurs even if the token buffer is |
3126 | * too small to hold it. | |
3127 | */ | |
3128 | static inline size_t copy_token(const char **buf, | |
3129 | char *token, | |
3130 | size_t token_size) | |
3131 | { | |
3132 | size_t len; | |
3133 | ||
3134 | len = next_token(buf); | |
3135 | if (len < token_size) { | |
3136 | memcpy(token, *buf, len); | |
3137 | *(token + len) = '\0'; | |
3138 | } | |
3139 | *buf += len; | |
3140 | ||
3141 | return len; | |
3142 | } | |
3143 | ||
ea3352f4 AE |
3144 | /* |
3145 | * Finds the next token in *buf, dynamically allocates a buffer big | |
3146 | * enough to hold a copy of it, and copies the token into the new | |
3147 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
3148 | * that a duplicate buffer is created even for a zero-length token. | |
3149 | * | |
3150 | * Returns a pointer to the newly-allocated duplicate, or a null | |
3151 | * pointer if memory for the duplicate was not available. If | |
3152 | * the lenp argument is a non-null pointer, the length of the token | |
3153 | * (not including the '\0') is returned in *lenp. | |
3154 | * | |
3155 | * If successful, the *buf pointer will be updated to point beyond | |
3156 | * the end of the found token. | |
3157 | * | |
3158 | * Note: uses GFP_KERNEL for allocation. | |
3159 | */ | |
3160 | static inline char *dup_token(const char **buf, size_t *lenp) | |
3161 | { | |
3162 | char *dup; | |
3163 | size_t len; | |
3164 | ||
3165 | len = next_token(buf); | |
4caf35f9 | 3166 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
3167 | if (!dup) |
3168 | return NULL; | |
ea3352f4 AE |
3169 | *(dup + len) = '\0'; |
3170 | *buf += len; | |
3171 | ||
3172 | if (lenp) | |
3173 | *lenp = len; | |
3174 | ||
3175 | return dup; | |
3176 | } | |
3177 | ||
a725f65e | 3178 | /* |
859c31df AE |
3179 | * Parse the options provided for an "rbd add" (i.e., rbd image |
3180 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
3181 | * and the data written is passed here via a NUL-terminated buffer. | |
3182 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 3183 | * |
859c31df AE |
3184 | * The information extracted from these options is recorded in |
3185 | * the other parameters which return dynamically-allocated | |
3186 | * structures: | |
3187 | * ceph_opts | |
3188 | * The address of a pointer that will refer to a ceph options | |
3189 | * structure. Caller must release the returned pointer using | |
3190 | * ceph_destroy_options() when it is no longer needed. | |
3191 | * rbd_opts | |
3192 | * Address of an rbd options pointer. Fully initialized by | |
3193 | * this function; caller must release with kfree(). | |
3194 | * spec | |
3195 | * Address of an rbd image specification pointer. Fully | |
3196 | * initialized by this function based on parsed options. | |
3197 | * Caller must release with rbd_spec_put(). | |
3198 | * | |
3199 | * The options passed take this form: | |
3200 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
3201 | * where: | |
3202 | * <mon_addrs> | |
3203 | * A comma-separated list of one or more monitor addresses. | |
3204 | * A monitor address is an ip address, optionally followed | |
3205 | * by a port number (separated by a colon). | |
3206 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
3207 | * <options> | |
3208 | * A comma-separated list of ceph and/or rbd options. | |
3209 | * <pool_name> | |
3210 | * The name of the rados pool containing the rbd image. | |
3211 | * <image_name> | |
3212 | * The name of the image in that pool to map. | |
3213 | * <snap_id> | |
3214 | * An optional snapshot id. If provided, the mapping will | |
3215 | * present data from the image at the time that snapshot was | |
3216 | * created. The image head is used if no snapshot id is | |
3217 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 3218 | */ |
859c31df | 3219 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 3220 | struct ceph_options **ceph_opts, |
859c31df AE |
3221 | struct rbd_options **opts, |
3222 | struct rbd_spec **rbd_spec) | |
e28fff26 | 3223 | { |
d22f76e7 | 3224 | size_t len; |
859c31df | 3225 | char *options; |
0ddebc0c AE |
3226 | const char *mon_addrs; |
3227 | size_t mon_addrs_size; | |
859c31df | 3228 | struct rbd_spec *spec = NULL; |
4e9afeba | 3229 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3230 | struct ceph_options *copts; |
dc79b113 | 3231 | int ret; |
e28fff26 AE |
3232 | |
3233 | /* The first four tokens are required */ | |
3234 | ||
7ef3214a | 3235 | len = next_token(&buf); |
4fb5d671 AE |
3236 | if (!len) { |
3237 | rbd_warn(NULL, "no monitor address(es) provided"); | |
3238 | return -EINVAL; | |
3239 | } | |
0ddebc0c | 3240 | mon_addrs = buf; |
f28e565a | 3241 | mon_addrs_size = len + 1; |
7ef3214a | 3242 | buf += len; |
a725f65e | 3243 | |
dc79b113 | 3244 | ret = -EINVAL; |
f28e565a AE |
3245 | options = dup_token(&buf, NULL); |
3246 | if (!options) | |
dc79b113 | 3247 | return -ENOMEM; |
4fb5d671 AE |
3248 | if (!*options) { |
3249 | rbd_warn(NULL, "no options provided"); | |
3250 | goto out_err; | |
3251 | } | |
e28fff26 | 3252 | |
859c31df AE |
3253 | spec = rbd_spec_alloc(); |
3254 | if (!spec) | |
f28e565a | 3255 | goto out_mem; |
859c31df AE |
3256 | |
3257 | spec->pool_name = dup_token(&buf, NULL); | |
3258 | if (!spec->pool_name) | |
3259 | goto out_mem; | |
4fb5d671 AE |
3260 | if (!*spec->pool_name) { |
3261 | rbd_warn(NULL, "no pool name provided"); | |
3262 | goto out_err; | |
3263 | } | |
e28fff26 | 3264 | |
69e7a02f | 3265 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 3266 | if (!spec->image_name) |
f28e565a | 3267 | goto out_mem; |
4fb5d671 AE |
3268 | if (!*spec->image_name) { |
3269 | rbd_warn(NULL, "no image name provided"); | |
3270 | goto out_err; | |
3271 | } | |
d4b125e9 | 3272 | |
f28e565a AE |
3273 | /* |
3274 | * Snapshot name is optional; default is to use "-" | |
3275 | * (indicating the head/no snapshot). | |
3276 | */ | |
3feeb894 | 3277 | len = next_token(&buf); |
820a5f3e | 3278 | if (!len) { |
3feeb894 AE |
3279 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
3280 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 3281 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 3282 | ret = -ENAMETOOLONG; |
f28e565a | 3283 | goto out_err; |
849b4260 | 3284 | } |
4caf35f9 | 3285 | spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
859c31df | 3286 | if (!spec->snap_name) |
f28e565a | 3287 | goto out_mem; |
859c31df | 3288 | *(spec->snap_name + len) = '\0'; |
e5c35534 | 3289 | |
0ddebc0c | 3290 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 3291 | |
4e9afeba AE |
3292 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
3293 | if (!rbd_opts) | |
3294 | goto out_mem; | |
3295 | ||
3296 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 3297 | |
859c31df | 3298 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 3299 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 3300 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
3301 | if (IS_ERR(copts)) { |
3302 | ret = PTR_ERR(copts); | |
dc79b113 AE |
3303 | goto out_err; |
3304 | } | |
859c31df AE |
3305 | kfree(options); |
3306 | ||
3307 | *ceph_opts = copts; | |
4e9afeba | 3308 | *opts = rbd_opts; |
859c31df | 3309 | *rbd_spec = spec; |
0ddebc0c | 3310 | |
dc79b113 | 3311 | return 0; |
f28e565a | 3312 | out_mem: |
dc79b113 | 3313 | ret = -ENOMEM; |
d22f76e7 | 3314 | out_err: |
859c31df AE |
3315 | kfree(rbd_opts); |
3316 | rbd_spec_put(spec); | |
f28e565a | 3317 | kfree(options); |
d22f76e7 | 3318 | |
dc79b113 | 3319 | return ret; |
a725f65e AE |
3320 | } |
3321 | ||
589d30e0 AE |
3322 | /* |
3323 | * An rbd format 2 image has a unique identifier, distinct from the | |
3324 | * name given to it by the user. Internally, that identifier is | |
3325 | * what's used to specify the names of objects related to the image. | |
3326 | * | |
3327 | * A special "rbd id" object is used to map an rbd image name to its | |
3328 | * id. If that object doesn't exist, then there is no v2 rbd image | |
3329 | * with the supplied name. | |
3330 | * | |
3331 | * This function will record the given rbd_dev's image_id field if | |
3332 | * it can be determined, and in that case will return 0. If any | |
3333 | * errors occur a negative errno will be returned and the rbd_dev's | |
3334 | * image_id field will be unchanged (and should be NULL). | |
3335 | */ | |
3336 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
3337 | { | |
3338 | int ret; | |
3339 | size_t size; | |
3340 | char *object_name; | |
3341 | void *response; | |
3342 | void *p; | |
3343 | ||
2c0d0a10 AE |
3344 | /* |
3345 | * When probing a parent image, the image id is already | |
3346 | * known (and the image name likely is not). There's no | |
3347 | * need to fetch the image id again in this case. | |
3348 | */ | |
3349 | if (rbd_dev->spec->image_id) | |
3350 | return 0; | |
3351 | ||
589d30e0 AE |
3352 | /* |
3353 | * First, see if the format 2 image id file exists, and if | |
3354 | * so, get the image's persistent id from it. | |
3355 | */ | |
69e7a02f | 3356 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
3357 | object_name = kmalloc(size, GFP_NOIO); |
3358 | if (!object_name) | |
3359 | return -ENOMEM; | |
0d7dbfce | 3360 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
3361 | dout("rbd id object name is %s\n", object_name); |
3362 | ||
3363 | /* Response will be an encoded string, which includes a length */ | |
3364 | ||
3365 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
3366 | response = kzalloc(size, GFP_NOIO); | |
3367 | if (!response) { | |
3368 | ret = -ENOMEM; | |
3369 | goto out; | |
3370 | } | |
3371 | ||
3372 | ret = rbd_req_sync_exec(rbd_dev, object_name, | |
3373 | "rbd", "get_id", | |
3374 | NULL, 0, | |
07b2391f | 3375 | response, RBD_IMAGE_ID_LEN_MAX, NULL); |
589d30e0 AE |
3376 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
3377 | if (ret < 0) | |
3378 | goto out; | |
a0ea3a40 | 3379 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
589d30e0 AE |
3380 | |
3381 | p = response; | |
0d7dbfce | 3382 | rbd_dev->spec->image_id = ceph_extract_encoded_string(&p, |
589d30e0 | 3383 | p + RBD_IMAGE_ID_LEN_MAX, |
979ed480 | 3384 | NULL, GFP_NOIO); |
0d7dbfce AE |
3385 | if (IS_ERR(rbd_dev->spec->image_id)) { |
3386 | ret = PTR_ERR(rbd_dev->spec->image_id); | |
3387 | rbd_dev->spec->image_id = NULL; | |
589d30e0 | 3388 | } else { |
0d7dbfce | 3389 | dout("image_id is %s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3390 | } |
3391 | out: | |
3392 | kfree(response); | |
3393 | kfree(object_name); | |
3394 | ||
3395 | return ret; | |
3396 | } | |
3397 | ||
a30b71b9 AE |
3398 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
3399 | { | |
3400 | int ret; | |
3401 | size_t size; | |
3402 | ||
3403 | /* Version 1 images have no id; empty string is used */ | |
3404 | ||
0d7dbfce AE |
3405 | rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL); |
3406 | if (!rbd_dev->spec->image_id) | |
a30b71b9 | 3407 | return -ENOMEM; |
a30b71b9 AE |
3408 | |
3409 | /* Record the header object name for this rbd image. */ | |
3410 | ||
69e7a02f | 3411 | size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX); |
a30b71b9 AE |
3412 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3413 | if (!rbd_dev->header_name) { | |
3414 | ret = -ENOMEM; | |
3415 | goto out_err; | |
3416 | } | |
0d7dbfce AE |
3417 | sprintf(rbd_dev->header_name, "%s%s", |
3418 | rbd_dev->spec->image_name, RBD_SUFFIX); | |
a30b71b9 AE |
3419 | |
3420 | /* Populate rbd image metadata */ | |
3421 | ||
3422 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
3423 | if (ret < 0) | |
3424 | goto out_err; | |
86b00e0d AE |
3425 | |
3426 | /* Version 1 images have no parent (no layering) */ | |
3427 | ||
3428 | rbd_dev->parent_spec = NULL; | |
3429 | rbd_dev->parent_overlap = 0; | |
3430 | ||
a30b71b9 AE |
3431 | rbd_dev->image_format = 1; |
3432 | ||
3433 | dout("discovered version 1 image, header name is %s\n", | |
3434 | rbd_dev->header_name); | |
3435 | ||
3436 | return 0; | |
3437 | ||
3438 | out_err: | |
3439 | kfree(rbd_dev->header_name); | |
3440 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
3441 | kfree(rbd_dev->spec->image_id); |
3442 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
3443 | |
3444 | return ret; | |
3445 | } | |
3446 | ||
3447 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
3448 | { | |
3449 | size_t size; | |
9d475de5 | 3450 | int ret; |
6e14b1a6 | 3451 | u64 ver = 0; |
a30b71b9 AE |
3452 | |
3453 | /* | |
3454 | * Image id was filled in by the caller. Record the header | |
3455 | * object name for this rbd image. | |
3456 | */ | |
979ed480 | 3457 | size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id); |
a30b71b9 AE |
3458 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3459 | if (!rbd_dev->header_name) | |
3460 | return -ENOMEM; | |
3461 | sprintf(rbd_dev->header_name, "%s%s", | |
0d7dbfce | 3462 | RBD_HEADER_PREFIX, rbd_dev->spec->image_id); |
9d475de5 AE |
3463 | |
3464 | /* Get the size and object order for the image */ | |
3465 | ||
3466 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
3467 | if (ret < 0) |
3468 | goto out_err; | |
3469 | ||
3470 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
3471 | ||
3472 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
3473 | if (ret < 0) |
3474 | goto out_err; | |
3475 | ||
d889140c | 3476 | /* Get the and check features for the image */ |
b1b5402a AE |
3477 | |
3478 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
3479 | if (ret < 0) |
3480 | goto out_err; | |
35d489f9 | 3481 | |
86b00e0d AE |
3482 | /* If the image supports layering, get the parent info */ |
3483 | ||
3484 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
3485 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3486 | if (ret < 0) | |
3487 | goto out_err; | |
3488 | } | |
3489 | ||
6e14b1a6 AE |
3490 | /* crypto and compression type aren't (yet) supported for v2 images */ |
3491 | ||
3492 | rbd_dev->header.crypt_type = 0; | |
3493 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 3494 | |
6e14b1a6 AE |
3495 | /* Get the snapshot context, plus the header version */ |
3496 | ||
3497 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
3498 | if (ret) |
3499 | goto out_err; | |
6e14b1a6 AE |
3500 | rbd_dev->header.obj_version = ver; |
3501 | ||
a30b71b9 AE |
3502 | rbd_dev->image_format = 2; |
3503 | ||
3504 | dout("discovered version 2 image, header name is %s\n", | |
3505 | rbd_dev->header_name); | |
3506 | ||
35152979 | 3507 | return 0; |
9d475de5 | 3508 | out_err: |
86b00e0d AE |
3509 | rbd_dev->parent_overlap = 0; |
3510 | rbd_spec_put(rbd_dev->parent_spec); | |
3511 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
3512 | kfree(rbd_dev->header_name); |
3513 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3514 | kfree(rbd_dev->header.object_prefix); |
3515 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3516 | |
3517 | return ret; | |
a30b71b9 AE |
3518 | } |
3519 | ||
83a06263 AE |
3520 | static int rbd_dev_probe_finish(struct rbd_device *rbd_dev) |
3521 | { | |
3522 | int ret; | |
3523 | ||
3524 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3525 | ret = rbd_dev_snaps_update(rbd_dev); | |
3526 | if (ret) | |
3527 | return ret; | |
3528 | ||
9e15b77d AE |
3529 | ret = rbd_dev_probe_update_spec(rbd_dev); |
3530 | if (ret) | |
3531 | goto err_out_snaps; | |
3532 | ||
83a06263 AE |
3533 | ret = rbd_dev_set_mapping(rbd_dev); |
3534 | if (ret) | |
3535 | goto err_out_snaps; | |
3536 | ||
3537 | /* generate unique id: find highest unique id, add one */ | |
3538 | rbd_dev_id_get(rbd_dev); | |
3539 | ||
3540 | /* Fill in the device name, now that we have its id. */ | |
3541 | BUILD_BUG_ON(DEV_NAME_LEN | |
3542 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3543 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3544 | ||
3545 | /* Get our block major device number. */ | |
3546 | ||
3547 | ret = register_blkdev(0, rbd_dev->name); | |
3548 | if (ret < 0) | |
3549 | goto err_out_id; | |
3550 | rbd_dev->major = ret; | |
3551 | ||
3552 | /* Set up the blkdev mapping. */ | |
3553 | ||
3554 | ret = rbd_init_disk(rbd_dev); | |
3555 | if (ret) | |
3556 | goto err_out_blkdev; | |
3557 | ||
3558 | ret = rbd_bus_add_dev(rbd_dev); | |
3559 | if (ret) | |
3560 | goto err_out_disk; | |
3561 | ||
3562 | /* | |
3563 | * At this point cleanup in the event of an error is the job | |
3564 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
3565 | */ | |
3566 | down_write(&rbd_dev->header_rwsem); | |
3567 | ret = rbd_dev_snaps_register(rbd_dev); | |
3568 | up_write(&rbd_dev->header_rwsem); | |
3569 | if (ret) | |
3570 | goto err_out_bus; | |
3571 | ||
c0430647 | 3572 | ret = rbd_req_sync_watch(rbd_dev, 1); |
83a06263 AE |
3573 | if (ret) |
3574 | goto err_out_bus; | |
3575 | ||
3576 | /* Everything's ready. Announce the disk to the world. */ | |
3577 | ||
3578 | add_disk(rbd_dev->disk); | |
3579 | ||
3580 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
3581 | (unsigned long long) rbd_dev->mapping.size); | |
3582 | ||
3583 | return ret; | |
3584 | err_out_bus: | |
3585 | /* this will also clean up rest of rbd_dev stuff */ | |
3586 | ||
3587 | rbd_bus_del_dev(rbd_dev); | |
3588 | ||
3589 | return ret; | |
3590 | err_out_disk: | |
3591 | rbd_free_disk(rbd_dev); | |
3592 | err_out_blkdev: | |
3593 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
3594 | err_out_id: | |
3595 | rbd_dev_id_put(rbd_dev); | |
3596 | err_out_snaps: | |
3597 | rbd_remove_all_snaps(rbd_dev); | |
3598 | ||
3599 | return ret; | |
3600 | } | |
3601 | ||
a30b71b9 AE |
3602 | /* |
3603 | * Probe for the existence of the header object for the given rbd | |
3604 | * device. For format 2 images this includes determining the image | |
3605 | * id. | |
3606 | */ | |
3607 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
3608 | { | |
3609 | int ret; | |
3610 | ||
3611 | /* | |
3612 | * Get the id from the image id object. If it's not a | |
3613 | * format 2 image, we'll get ENOENT back, and we'll assume | |
3614 | * it's a format 1 image. | |
3615 | */ | |
3616 | ret = rbd_dev_image_id(rbd_dev); | |
3617 | if (ret) | |
3618 | ret = rbd_dev_v1_probe(rbd_dev); | |
3619 | else | |
3620 | ret = rbd_dev_v2_probe(rbd_dev); | |
83a06263 | 3621 | if (ret) { |
a30b71b9 AE |
3622 | dout("probe failed, returning %d\n", ret); |
3623 | ||
83a06263 AE |
3624 | return ret; |
3625 | } | |
3626 | ||
3627 | ret = rbd_dev_probe_finish(rbd_dev); | |
3628 | if (ret) | |
3629 | rbd_header_free(&rbd_dev->header); | |
3630 | ||
a30b71b9 AE |
3631 | return ret; |
3632 | } | |
3633 | ||
59c2be1e YS |
3634 | static ssize_t rbd_add(struct bus_type *bus, |
3635 | const char *buf, | |
3636 | size_t count) | |
602adf40 | 3637 | { |
cb8627c7 | 3638 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 3639 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 3640 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3641 | struct rbd_spec *spec = NULL; |
9d3997fd | 3642 | struct rbd_client *rbdc; |
27cc2594 AE |
3643 | struct ceph_osd_client *osdc; |
3644 | int rc = -ENOMEM; | |
602adf40 YS |
3645 | |
3646 | if (!try_module_get(THIS_MODULE)) | |
3647 | return -ENODEV; | |
3648 | ||
602adf40 | 3649 | /* parse add command */ |
859c31df | 3650 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 3651 | if (rc < 0) |
bd4ba655 | 3652 | goto err_out_module; |
78cea76e | 3653 | |
9d3997fd AE |
3654 | rbdc = rbd_get_client(ceph_opts); |
3655 | if (IS_ERR(rbdc)) { | |
3656 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 3657 | goto err_out_args; |
9d3997fd | 3658 | } |
c53d5893 | 3659 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 3660 | |
602adf40 | 3661 | /* pick the pool */ |
9d3997fd | 3662 | osdc = &rbdc->client->osdc; |
859c31df | 3663 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
3664 | if (rc < 0) |
3665 | goto err_out_client; | |
859c31df AE |
3666 | spec->pool_id = (u64) rc; |
3667 | ||
0903e875 AE |
3668 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3669 | ||
3670 | if (WARN_ON(spec->pool_id > (u64) U32_MAX)) { | |
3671 | rc = -EIO; | |
3672 | goto err_out_client; | |
3673 | } | |
3674 | ||
c53d5893 | 3675 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
3676 | if (!rbd_dev) |
3677 | goto err_out_client; | |
c53d5893 AE |
3678 | rbdc = NULL; /* rbd_dev now owns this */ |
3679 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 3680 | |
bd4ba655 | 3681 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
3682 | kfree(rbd_opts); |
3683 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 3684 | |
a30b71b9 AE |
3685 | rc = rbd_dev_probe(rbd_dev); |
3686 | if (rc < 0) | |
c53d5893 | 3687 | goto err_out_rbd_dev; |
05fd6f6f | 3688 | |
602adf40 | 3689 | return count; |
c53d5893 AE |
3690 | err_out_rbd_dev: |
3691 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 3692 | err_out_client: |
9d3997fd | 3693 | rbd_put_client(rbdc); |
0ddebc0c | 3694 | err_out_args: |
78cea76e AE |
3695 | if (ceph_opts) |
3696 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 3697 | kfree(rbd_opts); |
859c31df | 3698 | rbd_spec_put(spec); |
bd4ba655 AE |
3699 | err_out_module: |
3700 | module_put(THIS_MODULE); | |
27cc2594 | 3701 | |
602adf40 | 3702 | dout("Error adding device %s\n", buf); |
27cc2594 AE |
3703 | |
3704 | return (ssize_t) rc; | |
602adf40 YS |
3705 | } |
3706 | ||
de71a297 | 3707 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
3708 | { |
3709 | struct list_head *tmp; | |
3710 | struct rbd_device *rbd_dev; | |
3711 | ||
e124a82f | 3712 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
3713 | list_for_each(tmp, &rbd_dev_list) { |
3714 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 3715 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 3716 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 3717 | return rbd_dev; |
e124a82f | 3718 | } |
602adf40 | 3719 | } |
e124a82f | 3720 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
3721 | return NULL; |
3722 | } | |
3723 | ||
dfc5606d | 3724 | static void rbd_dev_release(struct device *dev) |
602adf40 | 3725 | { |
593a9e7b | 3726 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3727 | |
1dbb4399 AE |
3728 | if (rbd_dev->watch_request) { |
3729 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
3730 | ||
3731 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 3732 | rbd_dev->watch_request); |
1dbb4399 | 3733 | } |
59c2be1e | 3734 | if (rbd_dev->watch_event) |
907703d0 | 3735 | rbd_req_sync_watch(rbd_dev, 0); |
602adf40 YS |
3736 | |
3737 | /* clean up and free blkdev */ | |
3738 | rbd_free_disk(rbd_dev); | |
3739 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 3740 | |
2ac4e75d AE |
3741 | /* release allocated disk header fields */ |
3742 | rbd_header_free(&rbd_dev->header); | |
3743 | ||
32eec68d | 3744 | /* done with the id, and with the rbd_dev */ |
e2839308 | 3745 | rbd_dev_id_put(rbd_dev); |
c53d5893 AE |
3746 | rbd_assert(rbd_dev->rbd_client != NULL); |
3747 | rbd_dev_destroy(rbd_dev); | |
602adf40 YS |
3748 | |
3749 | /* release module ref */ | |
3750 | module_put(THIS_MODULE); | |
602adf40 YS |
3751 | } |
3752 | ||
dfc5606d YS |
3753 | static ssize_t rbd_remove(struct bus_type *bus, |
3754 | const char *buf, | |
3755 | size_t count) | |
602adf40 YS |
3756 | { |
3757 | struct rbd_device *rbd_dev = NULL; | |
3758 | int target_id, rc; | |
3759 | unsigned long ul; | |
3760 | int ret = count; | |
3761 | ||
3762 | rc = strict_strtoul(buf, 10, &ul); | |
3763 | if (rc) | |
3764 | return rc; | |
3765 | ||
3766 | /* convert to int; abort if we lost anything in the conversion */ | |
3767 | target_id = (int) ul; | |
3768 | if (target_id != ul) | |
3769 | return -EINVAL; | |
3770 | ||
3771 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
3772 | ||
3773 | rbd_dev = __rbd_get_dev(target_id); | |
3774 | if (!rbd_dev) { | |
3775 | ret = -ENOENT; | |
3776 | goto done; | |
42382b70 AE |
3777 | } |
3778 | ||
3779 | if (rbd_dev->open_count) { | |
3780 | ret = -EBUSY; | |
3781 | goto done; | |
602adf40 YS |
3782 | } |
3783 | ||
41f38c2b | 3784 | rbd_remove_all_snaps(rbd_dev); |
dfc5606d | 3785 | rbd_bus_del_dev(rbd_dev); |
602adf40 YS |
3786 | |
3787 | done: | |
3788 | mutex_unlock(&ctl_mutex); | |
aafb230e | 3789 | |
602adf40 YS |
3790 | return ret; |
3791 | } | |
3792 | ||
602adf40 YS |
3793 | /* |
3794 | * create control files in sysfs | |
dfc5606d | 3795 | * /sys/bus/rbd/... |
602adf40 YS |
3796 | */ |
3797 | static int rbd_sysfs_init(void) | |
3798 | { | |
dfc5606d | 3799 | int ret; |
602adf40 | 3800 | |
fed4c143 | 3801 | ret = device_register(&rbd_root_dev); |
21079786 | 3802 | if (ret < 0) |
dfc5606d | 3803 | return ret; |
602adf40 | 3804 | |
fed4c143 AE |
3805 | ret = bus_register(&rbd_bus_type); |
3806 | if (ret < 0) | |
3807 | device_unregister(&rbd_root_dev); | |
602adf40 | 3808 | |
602adf40 YS |
3809 | return ret; |
3810 | } | |
3811 | ||
3812 | static void rbd_sysfs_cleanup(void) | |
3813 | { | |
dfc5606d | 3814 | bus_unregister(&rbd_bus_type); |
fed4c143 | 3815 | device_unregister(&rbd_root_dev); |
602adf40 YS |
3816 | } |
3817 | ||
3818 | int __init rbd_init(void) | |
3819 | { | |
3820 | int rc; | |
3821 | ||
3822 | rc = rbd_sysfs_init(); | |
3823 | if (rc) | |
3824 | return rc; | |
f0f8cef5 | 3825 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
3826 | return 0; |
3827 | } | |
3828 | ||
3829 | void __exit rbd_exit(void) | |
3830 | { | |
3831 | rbd_sysfs_cleanup(); | |
3832 | } | |
3833 | ||
3834 | module_init(rbd_init); | |
3835 | module_exit(rbd_exit); | |
3836 | ||
3837 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
3838 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
3839 | MODULE_DESCRIPTION("rados block device"); | |
3840 | ||
3841 | /* following authorship retained from original osdblk.c */ | |
3842 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
3843 | ||
3844 | MODULE_LICENSE("GPL"); |