]>
Commit | Line | Data |
---|---|---|
602adf40 YS |
1 | /* |
2 | rbd.c -- Export ceph rados objects as a Linux block device | |
3 | ||
4 | ||
5 | based on drivers/block/osdblk.c: | |
6 | ||
7 | Copyright 2009 Red Hat, Inc. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; see the file COPYING. If not, write to | |
20 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | ||
22 | ||
23 | ||
dfc5606d | 24 | For usage instructions, please refer to: |
602adf40 | 25 | |
dfc5606d | 26 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
27 | |
28 | */ | |
29 | ||
30 | #include <linux/ceph/libceph.h> | |
31 | #include <linux/ceph/osd_client.h> | |
32 | #include <linux/ceph/mon_client.h> | |
33 | #include <linux/ceph/decode.h> | |
59c2be1e | 34 | #include <linux/parser.h> |
602adf40 YS |
35 | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/device.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/blkdev.h> | |
41 | ||
42 | #include "rbd_types.h" | |
43 | ||
aafb230e AE |
44 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
45 | ||
593a9e7b AE |
46 | /* |
47 | * The basic unit of block I/O is a sector. It is interpreted in a | |
48 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
49 | * universally 512 bytes. These symbols are just slightly more | |
50 | * meaningful than the bare numbers they represent. | |
51 | */ | |
52 | #define SECTOR_SHIFT 9 | |
53 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
54 | ||
df111be6 AE |
55 | /* It might be useful to have this defined elsewhere too */ |
56 | ||
57 | #define U64_MAX ((u64) (~0ULL)) | |
58 | ||
f0f8cef5 AE |
59 | #define RBD_DRV_NAME "rbd" |
60 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
61 | |
62 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
63 | ||
d4b125e9 AE |
64 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
65 | #define RBD_MAX_SNAP_NAME_LEN \ | |
66 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
67 | ||
35d489f9 | 68 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
69 | #define RBD_MAX_OPT_LEN 1024 |
70 | ||
71 | #define RBD_SNAP_HEAD_NAME "-" | |
72 | ||
9e15b77d AE |
73 | /* This allows a single page to hold an image name sent by OSD */ |
74 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 75 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 76 | |
1e130199 | 77 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 78 | |
d889140c AE |
79 | /* Feature bits */ |
80 | ||
81 | #define RBD_FEATURE_LAYERING 1 | |
82 | ||
83 | /* Features supported by this (client software) implementation. */ | |
84 | ||
85 | #define RBD_FEATURES_ALL (0) | |
86 | ||
81a89793 AE |
87 | /* |
88 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
89 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
90 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
91 | * enough to hold all possible device names. | |
92 | */ | |
602adf40 | 93 | #define DEV_NAME_LEN 32 |
81a89793 | 94 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 | 95 | |
cc0538b6 | 96 | #define RBD_READ_ONLY_DEFAULT false |
59c2be1e | 97 | |
602adf40 YS |
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 | ||
59c2be1e | 156 | struct rbd_options { |
cc0538b6 | 157 | bool read_only; |
602adf40 YS |
158 | }; |
159 | ||
160 | /* | |
f0f8cef5 | 161 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
162 | */ |
163 | struct rbd_client { | |
164 | struct ceph_client *client; | |
165 | struct kref kref; | |
166 | struct list_head node; | |
167 | }; | |
168 | ||
169 | /* | |
f0f8cef5 | 170 | * a request completion status |
602adf40 | 171 | */ |
1fec7093 YS |
172 | struct rbd_req_status { |
173 | int done; | |
174 | int rc; | |
175 | u64 bytes; | |
176 | }; | |
177 | ||
178 | /* | |
179 | * a collection of requests | |
180 | */ | |
181 | struct rbd_req_coll { | |
182 | int total; | |
183 | int num_done; | |
184 | struct kref kref; | |
185 | struct rbd_req_status status[0]; | |
602adf40 YS |
186 | }; |
187 | ||
f0f8cef5 AE |
188 | /* |
189 | * a single io request | |
190 | */ | |
191 | struct rbd_request { | |
192 | struct request *rq; /* blk layer request */ | |
193 | struct bio *bio; /* cloned bio */ | |
194 | struct page **pages; /* list of used pages */ | |
195 | u64 len; | |
196 | int coll_index; | |
197 | struct rbd_req_coll *coll; | |
198 | }; | |
199 | ||
dfc5606d YS |
200 | struct rbd_snap { |
201 | struct device dev; | |
202 | const char *name; | |
3591538f | 203 | u64 size; |
dfc5606d YS |
204 | struct list_head node; |
205 | u64 id; | |
34b13184 | 206 | u64 features; |
dfc5606d YS |
207 | }; |
208 | ||
f84344f3 | 209 | struct rbd_mapping { |
99c1f08f | 210 | u64 size; |
34b13184 | 211 | u64 features; |
f84344f3 AE |
212 | bool read_only; |
213 | }; | |
214 | ||
602adf40 YS |
215 | /* |
216 | * a single device | |
217 | */ | |
218 | struct rbd_device { | |
de71a297 | 219 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
220 | |
221 | int major; /* blkdev assigned major */ | |
222 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 223 | |
a30b71b9 | 224 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
225 | struct rbd_client *rbd_client; |
226 | ||
227 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
228 | ||
229 | spinlock_t lock; /* queue lock */ | |
230 | ||
231 | struct rbd_image_header header; | |
daba5fdb | 232 | bool exists; |
0d7dbfce | 233 | struct rbd_spec *spec; |
602adf40 | 234 | |
0d7dbfce | 235 | char *header_name; |
971f839a | 236 | |
59c2be1e YS |
237 | struct ceph_osd_event *watch_event; |
238 | struct ceph_osd_request *watch_request; | |
239 | ||
86b00e0d AE |
240 | struct rbd_spec *parent_spec; |
241 | u64 parent_overlap; | |
242 | ||
c666601a JD |
243 | /* protects updating the header */ |
244 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
245 | |
246 | struct rbd_mapping mapping; | |
602adf40 YS |
247 | |
248 | struct list_head node; | |
dfc5606d YS |
249 | |
250 | /* list of snapshots */ | |
251 | struct list_head snaps; | |
252 | ||
253 | /* sysfs related */ | |
254 | struct device dev; | |
42382b70 | 255 | unsigned long open_count; |
dfc5606d YS |
256 | }; |
257 | ||
602adf40 | 258 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 259 | |
602adf40 | 260 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
261 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
262 | ||
432b8587 AE |
263 | static LIST_HEAD(rbd_client_list); /* clients */ |
264 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 265 | |
304f6808 AE |
266 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
267 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev); | |
268 | ||
dfc5606d | 269 | static void rbd_dev_release(struct device *dev); |
41f38c2b | 270 | static void rbd_remove_snap_dev(struct rbd_snap *snap); |
dfc5606d | 271 | |
f0f8cef5 AE |
272 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
273 | size_t count); | |
274 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
275 | size_t count); | |
276 | ||
277 | static struct bus_attribute rbd_bus_attrs[] = { | |
278 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
279 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
280 | __ATTR_NULL | |
281 | }; | |
282 | ||
283 | static struct bus_type rbd_bus_type = { | |
284 | .name = "rbd", | |
285 | .bus_attrs = rbd_bus_attrs, | |
286 | }; | |
287 | ||
288 | static void rbd_root_dev_release(struct device *dev) | |
289 | { | |
290 | } | |
291 | ||
292 | static struct device rbd_root_dev = { | |
293 | .init_name = "rbd", | |
294 | .release = rbd_root_dev_release, | |
295 | }; | |
296 | ||
06ecc6cb AE |
297 | static __printf(2, 3) |
298 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
299 | { | |
300 | struct va_format vaf; | |
301 | va_list args; | |
302 | ||
303 | va_start(args, fmt); | |
304 | vaf.fmt = fmt; | |
305 | vaf.va = &args; | |
306 | ||
307 | if (!rbd_dev) | |
308 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
309 | else if (rbd_dev->disk) | |
310 | printk(KERN_WARNING "%s: %s: %pV\n", | |
311 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
312 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
313 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
314 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
315 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
316 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
317 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
318 | else /* punt */ | |
319 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
320 | RBD_DRV_NAME, rbd_dev, &vaf); | |
321 | va_end(args); | |
322 | } | |
323 | ||
aafb230e AE |
324 | #ifdef RBD_DEBUG |
325 | #define rbd_assert(expr) \ | |
326 | if (unlikely(!(expr))) { \ | |
327 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
328 | "at line %d:\n\n" \ | |
329 | "\trbd_assert(%s);\n\n", \ | |
330 | __func__, __LINE__, #expr); \ | |
331 | BUG(); \ | |
332 | } | |
333 | #else /* !RBD_DEBUG */ | |
334 | # define rbd_assert(expr) ((void) 0) | |
335 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 336 | |
117973fb AE |
337 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
338 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 339 | |
602adf40 YS |
340 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
341 | { | |
f0f8cef5 | 342 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
602adf40 | 343 | |
f84344f3 | 344 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
345 | return -EROFS; |
346 | ||
42382b70 | 347 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 348 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 349 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 AE |
350 | rbd_dev->open_count++; |
351 | mutex_unlock(&ctl_mutex); | |
340c7a2b | 352 | |
602adf40 YS |
353 | return 0; |
354 | } | |
355 | ||
dfc5606d YS |
356 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
357 | { | |
358 | struct rbd_device *rbd_dev = disk->private_data; | |
359 | ||
42382b70 AE |
360 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
361 | rbd_assert(rbd_dev->open_count > 0); | |
362 | rbd_dev->open_count--; | |
c3e946ce | 363 | put_device(&rbd_dev->dev); |
42382b70 | 364 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
365 | |
366 | return 0; | |
367 | } | |
368 | ||
602adf40 YS |
369 | static const struct block_device_operations rbd_bd_ops = { |
370 | .owner = THIS_MODULE, | |
371 | .open = rbd_open, | |
dfc5606d | 372 | .release = rbd_release, |
602adf40 YS |
373 | }; |
374 | ||
375 | /* | |
376 | * Initialize an rbd client instance. | |
43ae4701 | 377 | * We own *ceph_opts. |
602adf40 | 378 | */ |
f8c38929 | 379 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
380 | { |
381 | struct rbd_client *rbdc; | |
382 | int ret = -ENOMEM; | |
383 | ||
384 | dout("rbd_client_create\n"); | |
385 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); | |
386 | if (!rbdc) | |
387 | goto out_opt; | |
388 | ||
389 | kref_init(&rbdc->kref); | |
390 | INIT_LIST_HEAD(&rbdc->node); | |
391 | ||
bc534d86 AE |
392 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
393 | ||
43ae4701 | 394 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 395 | if (IS_ERR(rbdc->client)) |
bc534d86 | 396 | goto out_mutex; |
43ae4701 | 397 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
398 | |
399 | ret = ceph_open_session(rbdc->client); | |
400 | if (ret < 0) | |
401 | goto out_err; | |
402 | ||
432b8587 | 403 | spin_lock(&rbd_client_list_lock); |
602adf40 | 404 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 405 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 406 | |
bc534d86 AE |
407 | mutex_unlock(&ctl_mutex); |
408 | ||
602adf40 YS |
409 | dout("rbd_client_create created %p\n", rbdc); |
410 | return rbdc; | |
411 | ||
412 | out_err: | |
413 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
414 | out_mutex: |
415 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
416 | kfree(rbdc); |
417 | out_opt: | |
43ae4701 AE |
418 | if (ceph_opts) |
419 | ceph_destroy_options(ceph_opts); | |
28f259b7 | 420 | return ERR_PTR(ret); |
602adf40 YS |
421 | } |
422 | ||
423 | /* | |
1f7ba331 AE |
424 | * Find a ceph client with specific addr and configuration. If |
425 | * found, bump its reference count. | |
602adf40 | 426 | */ |
1f7ba331 | 427 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
428 | { |
429 | struct rbd_client *client_node; | |
1f7ba331 | 430 | bool found = false; |
602adf40 | 431 | |
43ae4701 | 432 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
433 | return NULL; |
434 | ||
1f7ba331 AE |
435 | spin_lock(&rbd_client_list_lock); |
436 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
437 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
438 | kref_get(&client_node->kref); | |
439 | found = true; | |
440 | break; | |
441 | } | |
442 | } | |
443 | spin_unlock(&rbd_client_list_lock); | |
444 | ||
445 | return found ? client_node : NULL; | |
602adf40 YS |
446 | } |
447 | ||
59c2be1e YS |
448 | /* |
449 | * mount options | |
450 | */ | |
451 | enum { | |
59c2be1e YS |
452 | Opt_last_int, |
453 | /* int args above */ | |
454 | Opt_last_string, | |
455 | /* string args above */ | |
cc0538b6 AE |
456 | Opt_read_only, |
457 | Opt_read_write, | |
458 | /* Boolean args above */ | |
459 | Opt_last_bool, | |
59c2be1e YS |
460 | }; |
461 | ||
43ae4701 | 462 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
463 | /* int args above */ |
464 | /* string args above */ | |
be466c1c | 465 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
466 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
467 | {Opt_read_write, "read_write"}, | |
468 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
469 | /* Boolean args above */ | |
59c2be1e YS |
470 | {-1, NULL} |
471 | }; | |
472 | ||
473 | static int parse_rbd_opts_token(char *c, void *private) | |
474 | { | |
43ae4701 | 475 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
476 | substring_t argstr[MAX_OPT_ARGS]; |
477 | int token, intval, ret; | |
478 | ||
43ae4701 | 479 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
480 | if (token < 0) |
481 | return -EINVAL; | |
482 | ||
483 | if (token < Opt_last_int) { | |
484 | ret = match_int(&argstr[0], &intval); | |
485 | if (ret < 0) { | |
486 | pr_err("bad mount option arg (not int) " | |
487 | "at '%s'\n", c); | |
488 | return ret; | |
489 | } | |
490 | dout("got int token %d val %d\n", token, intval); | |
491 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
492 | dout("got string token %d val %s\n", token, | |
493 | argstr[0].from); | |
cc0538b6 AE |
494 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
495 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
496 | } else { |
497 | dout("got token %d\n", token); | |
498 | } | |
499 | ||
500 | switch (token) { | |
cc0538b6 AE |
501 | case Opt_read_only: |
502 | rbd_opts->read_only = true; | |
503 | break; | |
504 | case Opt_read_write: | |
505 | rbd_opts->read_only = false; | |
506 | break; | |
59c2be1e | 507 | default: |
aafb230e AE |
508 | rbd_assert(false); |
509 | break; | |
59c2be1e YS |
510 | } |
511 | return 0; | |
512 | } | |
513 | ||
602adf40 YS |
514 | /* |
515 | * Get a ceph client with specific addr and configuration, if one does | |
516 | * not exist create it. | |
517 | */ | |
9d3997fd | 518 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 519 | { |
f8c38929 | 520 | struct rbd_client *rbdc; |
59c2be1e | 521 | |
1f7ba331 | 522 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 523 | if (rbdc) /* using an existing client */ |
43ae4701 | 524 | ceph_destroy_options(ceph_opts); |
9d3997fd | 525 | else |
f8c38929 | 526 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 527 | |
9d3997fd | 528 | return rbdc; |
602adf40 YS |
529 | } |
530 | ||
531 | /* | |
532 | * Destroy ceph client | |
d23a4b3f | 533 | * |
432b8587 | 534 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
535 | */ |
536 | static void rbd_client_release(struct kref *kref) | |
537 | { | |
538 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
539 | ||
540 | dout("rbd_release_client %p\n", rbdc); | |
cd9d9f5d | 541 | spin_lock(&rbd_client_list_lock); |
602adf40 | 542 | list_del(&rbdc->node); |
cd9d9f5d | 543 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
544 | |
545 | ceph_destroy_client(rbdc->client); | |
546 | kfree(rbdc); | |
547 | } | |
548 | ||
549 | /* | |
550 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
551 | * it. | |
552 | */ | |
9d3997fd | 553 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 554 | { |
c53d5893 AE |
555 | if (rbdc) |
556 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
557 | } |
558 | ||
1fec7093 YS |
559 | /* |
560 | * Destroy requests collection | |
561 | */ | |
562 | static void rbd_coll_release(struct kref *kref) | |
563 | { | |
564 | struct rbd_req_coll *coll = | |
565 | container_of(kref, struct rbd_req_coll, kref); | |
566 | ||
567 | dout("rbd_coll_release %p\n", coll); | |
568 | kfree(coll); | |
569 | } | |
602adf40 | 570 | |
a30b71b9 AE |
571 | static bool rbd_image_format_valid(u32 image_format) |
572 | { | |
573 | return image_format == 1 || image_format == 2; | |
574 | } | |
575 | ||
8e94af8e AE |
576 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
577 | { | |
103a150f AE |
578 | size_t size; |
579 | u32 snap_count; | |
580 | ||
581 | /* The header has to start with the magic rbd header text */ | |
582 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
583 | return false; | |
584 | ||
db2388b6 AE |
585 | /* The bio layer requires at least sector-sized I/O */ |
586 | ||
587 | if (ondisk->options.order < SECTOR_SHIFT) | |
588 | return false; | |
589 | ||
590 | /* If we use u64 in a few spots we may be able to loosen this */ | |
591 | ||
592 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
593 | return false; | |
594 | ||
103a150f AE |
595 | /* |
596 | * The size of a snapshot header has to fit in a size_t, and | |
597 | * that limits the number of snapshots. | |
598 | */ | |
599 | snap_count = le32_to_cpu(ondisk->snap_count); | |
600 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
601 | if (snap_count > size / sizeof (__le64)) | |
602 | return false; | |
603 | ||
604 | /* | |
605 | * Not only that, but the size of the entire the snapshot | |
606 | * header must also be representable in a size_t. | |
607 | */ | |
608 | size -= snap_count * sizeof (__le64); | |
609 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
610 | return false; | |
611 | ||
612 | return true; | |
8e94af8e AE |
613 | } |
614 | ||
602adf40 YS |
615 | /* |
616 | * Create a new header structure, translate header format from the on-disk | |
617 | * header. | |
618 | */ | |
619 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 620 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 621 | { |
ccece235 | 622 | u32 snap_count; |
58c17b0e | 623 | size_t len; |
d2bb24e5 | 624 | size_t size; |
621901d6 | 625 | u32 i; |
602adf40 | 626 | |
6a52325f AE |
627 | memset(header, 0, sizeof (*header)); |
628 | ||
103a150f AE |
629 | snap_count = le32_to_cpu(ondisk->snap_count); |
630 | ||
58c17b0e AE |
631 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
632 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 633 | if (!header->object_prefix) |
602adf40 | 634 | return -ENOMEM; |
58c17b0e AE |
635 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
636 | header->object_prefix[len] = '\0'; | |
00f1f36f | 637 | |
602adf40 | 638 | if (snap_count) { |
f785cc1d AE |
639 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
640 | ||
621901d6 AE |
641 | /* Save a copy of the snapshot names */ |
642 | ||
f785cc1d AE |
643 | if (snap_names_len > (u64) SIZE_MAX) |
644 | return -EIO; | |
645 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 646 | if (!header->snap_names) |
6a52325f | 647 | goto out_err; |
f785cc1d AE |
648 | /* |
649 | * Note that rbd_dev_v1_header_read() guarantees | |
650 | * the ondisk buffer we're working with has | |
651 | * snap_names_len bytes beyond the end of the | |
652 | * snapshot id array, this memcpy() is safe. | |
653 | */ | |
654 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
655 | snap_names_len); | |
6a52325f | 656 | |
621901d6 AE |
657 | /* Record each snapshot's size */ |
658 | ||
d2bb24e5 AE |
659 | size = snap_count * sizeof (*header->snap_sizes); |
660 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 661 | if (!header->snap_sizes) |
6a52325f | 662 | goto out_err; |
621901d6 AE |
663 | for (i = 0; i < snap_count; i++) |
664 | header->snap_sizes[i] = | |
665 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 | 666 | } else { |
ccece235 | 667 | WARN_ON(ondisk->snap_names_len); |
602adf40 YS |
668 | header->snap_names = NULL; |
669 | header->snap_sizes = NULL; | |
670 | } | |
849b4260 | 671 | |
34b13184 | 672 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
673 | header->obj_order = ondisk->options.order; |
674 | header->crypt_type = ondisk->options.crypt_type; | |
675 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 676 | |
621901d6 AE |
677 | /* Allocate and fill in the snapshot context */ |
678 | ||
f84344f3 | 679 | header->image_size = le64_to_cpu(ondisk->image_size); |
6a52325f AE |
680 | size = sizeof (struct ceph_snap_context); |
681 | size += snap_count * sizeof (header->snapc->snaps[0]); | |
682 | header->snapc = kzalloc(size, GFP_KERNEL); | |
683 | if (!header->snapc) | |
684 | goto out_err; | |
602adf40 YS |
685 | |
686 | atomic_set(&header->snapc->nref, 1); | |
505cbb9b | 687 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
602adf40 | 688 | header->snapc->num_snaps = snap_count; |
621901d6 AE |
689 | for (i = 0; i < snap_count; i++) |
690 | header->snapc->snaps[i] = | |
691 | le64_to_cpu(ondisk->snaps[i].id); | |
602adf40 YS |
692 | |
693 | return 0; | |
694 | ||
6a52325f | 695 | out_err: |
849b4260 | 696 | kfree(header->snap_sizes); |
ccece235 | 697 | header->snap_sizes = NULL; |
602adf40 | 698 | kfree(header->snap_names); |
ccece235 | 699 | header->snap_names = NULL; |
6a52325f AE |
700 | kfree(header->object_prefix); |
701 | header->object_prefix = NULL; | |
ccece235 | 702 | |
00f1f36f | 703 | return -ENOMEM; |
602adf40 YS |
704 | } |
705 | ||
9e15b77d AE |
706 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
707 | { | |
708 | struct rbd_snap *snap; | |
709 | ||
710 | if (snap_id == CEPH_NOSNAP) | |
711 | return RBD_SNAP_HEAD_NAME; | |
712 | ||
713 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
714 | if (snap_id == snap->id) | |
715 | return snap->name; | |
716 | ||
717 | return NULL; | |
718 | } | |
719 | ||
8836b995 | 720 | static int snap_by_name(struct rbd_device *rbd_dev, const char *snap_name) |
602adf40 | 721 | { |
602adf40 | 722 | |
e86924a8 | 723 | struct rbd_snap *snap; |
602adf40 | 724 | |
e86924a8 AE |
725 | list_for_each_entry(snap, &rbd_dev->snaps, node) { |
726 | if (!strcmp(snap_name, snap->name)) { | |
0d7dbfce | 727 | rbd_dev->spec->snap_id = snap->id; |
e86924a8 | 728 | rbd_dev->mapping.size = snap->size; |
34b13184 | 729 | rbd_dev->mapping.features = snap->features; |
602adf40 | 730 | |
e86924a8 | 731 | return 0; |
00f1f36f | 732 | } |
00f1f36f | 733 | } |
e86924a8 | 734 | |
00f1f36f | 735 | return -ENOENT; |
602adf40 YS |
736 | } |
737 | ||
819d52bf | 738 | static int rbd_dev_set_mapping(struct rbd_device *rbd_dev) |
602adf40 | 739 | { |
78dc447d | 740 | int ret; |
602adf40 | 741 | |
0d7dbfce | 742 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 743 | sizeof (RBD_SNAP_HEAD_NAME))) { |
0d7dbfce | 744 | rbd_dev->spec->snap_id = CEPH_NOSNAP; |
99c1f08f | 745 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 746 | rbd_dev->mapping.features = rbd_dev->header.features; |
e86924a8 | 747 | ret = 0; |
602adf40 | 748 | } else { |
0d7dbfce | 749 | ret = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); |
602adf40 YS |
750 | if (ret < 0) |
751 | goto done; | |
f84344f3 | 752 | rbd_dev->mapping.read_only = true; |
602adf40 | 753 | } |
daba5fdb | 754 | rbd_dev->exists = true; |
602adf40 | 755 | done: |
602adf40 YS |
756 | return ret; |
757 | } | |
758 | ||
759 | static void rbd_header_free(struct rbd_image_header *header) | |
760 | { | |
849b4260 | 761 | kfree(header->object_prefix); |
d78fd7ae | 762 | header->object_prefix = NULL; |
602adf40 | 763 | kfree(header->snap_sizes); |
d78fd7ae | 764 | header->snap_sizes = NULL; |
849b4260 | 765 | kfree(header->snap_names); |
d78fd7ae | 766 | header->snap_names = NULL; |
d1d25646 | 767 | ceph_put_snap_context(header->snapc); |
d78fd7ae | 768 | header->snapc = NULL; |
602adf40 YS |
769 | } |
770 | ||
65ccfe21 | 771 | static char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 772 | { |
65ccfe21 AE |
773 | char *name; |
774 | u64 segment; | |
775 | int ret; | |
602adf40 | 776 | |
2fd82b9e | 777 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
778 | if (!name) |
779 | return NULL; | |
780 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 781 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 782 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 783 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
784 | pr_err("error formatting segment name for #%llu (%d)\n", |
785 | segment, ret); | |
786 | kfree(name); | |
787 | name = NULL; | |
788 | } | |
602adf40 | 789 | |
65ccfe21 AE |
790 | return name; |
791 | } | |
602adf40 | 792 | |
65ccfe21 AE |
793 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
794 | { | |
795 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 796 | |
65ccfe21 AE |
797 | return offset & (segment_size - 1); |
798 | } | |
799 | ||
800 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
801 | u64 offset, u64 length) | |
802 | { | |
803 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
804 | ||
805 | offset &= segment_size - 1; | |
806 | ||
aafb230e | 807 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
808 | if (offset + length > segment_size) |
809 | length = segment_size - offset; | |
810 | ||
811 | return length; | |
602adf40 YS |
812 | } |
813 | ||
1fec7093 YS |
814 | static int rbd_get_num_segments(struct rbd_image_header *header, |
815 | u64 ofs, u64 len) | |
816 | { | |
df111be6 AE |
817 | u64 start_seg; |
818 | u64 end_seg; | |
819 | ||
820 | if (!len) | |
821 | return 0; | |
822 | if (len - 1 > U64_MAX - ofs) | |
823 | return -ERANGE; | |
824 | ||
825 | start_seg = ofs >> header->obj_order; | |
826 | end_seg = (ofs + len - 1) >> header->obj_order; | |
827 | ||
1fec7093 YS |
828 | return end_seg - start_seg + 1; |
829 | } | |
830 | ||
029bcbd8 JD |
831 | /* |
832 | * returns the size of an object in the image | |
833 | */ | |
834 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
835 | { | |
836 | return 1 << header->obj_order; | |
837 | } | |
838 | ||
602adf40 YS |
839 | /* |
840 | * bio helpers | |
841 | */ | |
842 | ||
843 | static void bio_chain_put(struct bio *chain) | |
844 | { | |
845 | struct bio *tmp; | |
846 | ||
847 | while (chain) { | |
848 | tmp = chain; | |
849 | chain = chain->bi_next; | |
850 | bio_put(tmp); | |
851 | } | |
852 | } | |
853 | ||
854 | /* | |
855 | * zeros a bio chain, starting at specific offset | |
856 | */ | |
857 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
858 | { | |
859 | struct bio_vec *bv; | |
860 | unsigned long flags; | |
861 | void *buf; | |
862 | int i; | |
863 | int pos = 0; | |
864 | ||
865 | while (chain) { | |
866 | bio_for_each_segment(bv, chain, i) { | |
867 | if (pos + bv->bv_len > start_ofs) { | |
868 | int remainder = max(start_ofs - pos, 0); | |
869 | buf = bvec_kmap_irq(bv, &flags); | |
870 | memset(buf + remainder, 0, | |
871 | bv->bv_len - remainder); | |
85b5aaa6 | 872 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
873 | } |
874 | pos += bv->bv_len; | |
875 | } | |
876 | ||
877 | chain = chain->bi_next; | |
878 | } | |
879 | } | |
880 | ||
881 | /* | |
f7760dad AE |
882 | * Clone a portion of a bio, starting at the given byte offset |
883 | * and continuing for the number of bytes indicated. | |
602adf40 | 884 | */ |
f7760dad AE |
885 | static struct bio *bio_clone_range(struct bio *bio_src, |
886 | unsigned int offset, | |
887 | unsigned int len, | |
888 | gfp_t gfpmask) | |
602adf40 | 889 | { |
f7760dad AE |
890 | struct bio_vec *bv; |
891 | unsigned int resid; | |
892 | unsigned short idx; | |
893 | unsigned int voff; | |
894 | unsigned short end_idx; | |
895 | unsigned short vcnt; | |
896 | struct bio *bio; | |
897 | ||
898 | /* Handle the easy case for the caller */ | |
899 | ||
900 | if (!offset && len == bio_src->bi_size) | |
901 | return bio_clone(bio_src, gfpmask); | |
902 | ||
903 | if (WARN_ON_ONCE(!len)) | |
904 | return NULL; | |
905 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
906 | return NULL; | |
907 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
908 | return NULL; | |
909 | ||
910 | /* Find first affected segment... */ | |
911 | ||
912 | resid = offset; | |
913 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
914 | if (resid < bv->bv_len) | |
915 | break; | |
916 | resid -= bv->bv_len; | |
602adf40 | 917 | } |
f7760dad | 918 | voff = resid; |
602adf40 | 919 | |
f7760dad | 920 | /* ...and the last affected segment */ |
602adf40 | 921 | |
f7760dad AE |
922 | resid += len; |
923 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
924 | if (resid <= bv->bv_len) | |
925 | break; | |
926 | resid -= bv->bv_len; | |
927 | } | |
928 | vcnt = end_idx - idx + 1; | |
929 | ||
930 | /* Build the clone */ | |
931 | ||
932 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
933 | if (!bio) | |
934 | return NULL; /* ENOMEM */ | |
602adf40 | 935 | |
f7760dad AE |
936 | bio->bi_bdev = bio_src->bi_bdev; |
937 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
938 | bio->bi_rw = bio_src->bi_rw; | |
939 | bio->bi_flags |= 1 << BIO_CLONED; | |
940 | ||
941 | /* | |
942 | * Copy over our part of the bio_vec, then update the first | |
943 | * and last (or only) entries. | |
944 | */ | |
945 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
946 | vcnt * sizeof (struct bio_vec)); | |
947 | bio->bi_io_vec[0].bv_offset += voff; | |
948 | if (vcnt > 1) { | |
949 | bio->bi_io_vec[0].bv_len -= voff; | |
950 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
951 | } else { | |
952 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
953 | } |
954 | ||
f7760dad AE |
955 | bio->bi_vcnt = vcnt; |
956 | bio->bi_size = len; | |
957 | bio->bi_idx = 0; | |
958 | ||
959 | return bio; | |
960 | } | |
961 | ||
962 | /* | |
963 | * Clone a portion of a bio chain, starting at the given byte offset | |
964 | * into the first bio in the source chain and continuing for the | |
965 | * number of bytes indicated. The result is another bio chain of | |
966 | * exactly the given length, or a null pointer on error. | |
967 | * | |
968 | * The bio_src and offset parameters are both in-out. On entry they | |
969 | * refer to the first source bio and the offset into that bio where | |
970 | * the start of data to be cloned is located. | |
971 | * | |
972 | * On return, bio_src is updated to refer to the bio in the source | |
973 | * chain that contains first un-cloned byte, and *offset will | |
974 | * contain the offset of that byte within that bio. | |
975 | */ | |
976 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
977 | unsigned int *offset, | |
978 | unsigned int len, | |
979 | gfp_t gfpmask) | |
980 | { | |
981 | struct bio *bi = *bio_src; | |
982 | unsigned int off = *offset; | |
983 | struct bio *chain = NULL; | |
984 | struct bio **end; | |
985 | ||
986 | /* Build up a chain of clone bios up to the limit */ | |
987 | ||
988 | if (!bi || off >= bi->bi_size || !len) | |
989 | return NULL; /* Nothing to clone */ | |
602adf40 | 990 | |
f7760dad AE |
991 | end = &chain; |
992 | while (len) { | |
993 | unsigned int bi_size; | |
994 | struct bio *bio; | |
995 | ||
996 | if (!bi) | |
997 | goto out_err; /* EINVAL; ran out of bio's */ | |
998 | bi_size = min_t(unsigned int, bi->bi_size - off, len); | |
999 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1000 | if (!bio) | |
1001 | goto out_err; /* ENOMEM */ | |
1002 | ||
1003 | *end = bio; | |
1004 | end = &bio->bi_next; | |
602adf40 | 1005 | |
f7760dad AE |
1006 | off += bi_size; |
1007 | if (off == bi->bi_size) { | |
1008 | bi = bi->bi_next; | |
1009 | off = 0; | |
1010 | } | |
1011 | len -= bi_size; | |
1012 | } | |
1013 | *bio_src = bi; | |
1014 | *offset = off; | |
1015 | ||
1016 | return chain; | |
1017 | out_err: | |
1018 | bio_chain_put(chain); | |
602adf40 | 1019 | |
602adf40 YS |
1020 | return NULL; |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * helpers for osd request op vectors. | |
1025 | */ | |
57cfc106 AE |
1026 | static struct ceph_osd_req_op *rbd_create_rw_ops(int num_ops, |
1027 | int opcode, u32 payload_len) | |
602adf40 | 1028 | { |
57cfc106 AE |
1029 | struct ceph_osd_req_op *ops; |
1030 | ||
1031 | ops = kzalloc(sizeof (*ops) * (num_ops + 1), GFP_NOIO); | |
1032 | if (!ops) | |
1033 | return NULL; | |
1034 | ||
1035 | ops[0].op = opcode; | |
1036 | ||
602adf40 YS |
1037 | /* |
1038 | * op extent offset and length will be set later on | |
1039 | * in calc_raw_layout() | |
1040 | */ | |
57cfc106 AE |
1041 | ops[0].payload_len = payload_len; |
1042 | ||
1043 | return ops; | |
602adf40 YS |
1044 | } |
1045 | ||
1046 | static void rbd_destroy_ops(struct ceph_osd_req_op *ops) | |
1047 | { | |
1048 | kfree(ops); | |
1049 | } | |
1050 | ||
1fec7093 YS |
1051 | static void rbd_coll_end_req_index(struct request *rq, |
1052 | struct rbd_req_coll *coll, | |
1053 | int index, | |
1054 | int ret, u64 len) | |
1055 | { | |
1056 | struct request_queue *q; | |
1057 | int min, max, i; | |
1058 | ||
bd919d45 AE |
1059 | dout("rbd_coll_end_req_index %p index %d ret %d len %llu\n", |
1060 | coll, index, ret, (unsigned long long) len); | |
1fec7093 YS |
1061 | |
1062 | if (!rq) | |
1063 | return; | |
1064 | ||
1065 | if (!coll) { | |
1066 | blk_end_request(rq, ret, len); | |
1067 | return; | |
1068 | } | |
1069 | ||
1070 | q = rq->q; | |
1071 | ||
1072 | spin_lock_irq(q->queue_lock); | |
1073 | coll->status[index].done = 1; | |
1074 | coll->status[index].rc = ret; | |
1075 | coll->status[index].bytes = len; | |
1076 | max = min = coll->num_done; | |
1077 | while (max < coll->total && coll->status[max].done) | |
1078 | max++; | |
1079 | ||
1080 | for (i = min; i<max; i++) { | |
1081 | __blk_end_request(rq, coll->status[i].rc, | |
1082 | coll->status[i].bytes); | |
1083 | coll->num_done++; | |
1084 | kref_put(&coll->kref, rbd_coll_release); | |
1085 | } | |
1086 | spin_unlock_irq(q->queue_lock); | |
1087 | } | |
1088 | ||
1089 | static void rbd_coll_end_req(struct rbd_request *req, | |
1090 | int ret, u64 len) | |
1091 | { | |
1092 | rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len); | |
1093 | } | |
1094 | ||
602adf40 YS |
1095 | /* |
1096 | * Send ceph osd request | |
1097 | */ | |
1098 | static int rbd_do_request(struct request *rq, | |
0ce1a794 | 1099 | struct rbd_device *rbd_dev, |
602adf40 YS |
1100 | struct ceph_snap_context *snapc, |
1101 | u64 snapid, | |
aded07ea | 1102 | const char *object_name, u64 ofs, u64 len, |
602adf40 YS |
1103 | struct bio *bio, |
1104 | struct page **pages, | |
1105 | int num_pages, | |
1106 | int flags, | |
1107 | struct ceph_osd_req_op *ops, | |
1fec7093 YS |
1108 | struct rbd_req_coll *coll, |
1109 | int coll_index, | |
602adf40 | 1110 | void (*rbd_cb)(struct ceph_osd_request *req, |
59c2be1e YS |
1111 | struct ceph_msg *msg), |
1112 | struct ceph_osd_request **linger_req, | |
1113 | u64 *ver) | |
602adf40 YS |
1114 | { |
1115 | struct ceph_osd_request *req; | |
1116 | struct ceph_file_layout *layout; | |
1117 | int ret; | |
1118 | u64 bno; | |
1119 | struct timespec mtime = CURRENT_TIME; | |
1120 | struct rbd_request *req_data; | |
1121 | struct ceph_osd_request_head *reqhead; | |
1dbb4399 | 1122 | struct ceph_osd_client *osdc; |
602adf40 | 1123 | |
602adf40 | 1124 | req_data = kzalloc(sizeof(*req_data), GFP_NOIO); |
1fec7093 YS |
1125 | if (!req_data) { |
1126 | if (coll) | |
1127 | rbd_coll_end_req_index(rq, coll, coll_index, | |
1128 | -ENOMEM, len); | |
1129 | return -ENOMEM; | |
1130 | } | |
1131 | ||
1132 | if (coll) { | |
1133 | req_data->coll = coll; | |
1134 | req_data->coll_index = coll_index; | |
1135 | } | |
602adf40 | 1136 | |
f7760dad AE |
1137 | dout("rbd_do_request object_name=%s ofs=%llu len=%llu coll=%p[%d]\n", |
1138 | object_name, (unsigned long long) ofs, | |
1139 | (unsigned long long) len, coll, coll_index); | |
602adf40 | 1140 | |
0ce1a794 | 1141 | osdc = &rbd_dev->rbd_client->client->osdc; |
1dbb4399 AE |
1142 | req = ceph_osdc_alloc_request(osdc, flags, snapc, ops, |
1143 | false, GFP_NOIO, pages, bio); | |
4ad12621 | 1144 | if (!req) { |
4ad12621 | 1145 | ret = -ENOMEM; |
602adf40 YS |
1146 | goto done_pages; |
1147 | } | |
1148 | ||
1149 | req->r_callback = rbd_cb; | |
1150 | ||
1151 | req_data->rq = rq; | |
1152 | req_data->bio = bio; | |
1153 | req_data->pages = pages; | |
1154 | req_data->len = len; | |
1155 | ||
1156 | req->r_priv = req_data; | |
1157 | ||
1158 | reqhead = req->r_request->front.iov_base; | |
1159 | reqhead->snapid = cpu_to_le64(CEPH_NOSNAP); | |
1160 | ||
aded07ea | 1161 | strncpy(req->r_oid, object_name, sizeof(req->r_oid)); |
602adf40 YS |
1162 | req->r_oid_len = strlen(req->r_oid); |
1163 | ||
1164 | layout = &req->r_file_layout; | |
1165 | memset(layout, 0, sizeof(*layout)); | |
1166 | layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
1167 | layout->fl_stripe_count = cpu_to_le32(1); | |
1168 | layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
0d7dbfce | 1169 | layout->fl_pg_pool = cpu_to_le32((int) rbd_dev->spec->pool_id); |
6cae3717 SW |
1170 | ret = ceph_calc_raw_layout(osdc, layout, snapid, ofs, &len, &bno, |
1171 | req, ops); | |
1172 | rbd_assert(ret == 0); | |
602adf40 YS |
1173 | |
1174 | ceph_osdc_build_request(req, ofs, &len, | |
1175 | ops, | |
1176 | snapc, | |
1177 | &mtime, | |
1178 | req->r_oid, req->r_oid_len); | |
602adf40 | 1179 | |
59c2be1e | 1180 | if (linger_req) { |
1dbb4399 | 1181 | ceph_osdc_set_request_linger(osdc, req); |
59c2be1e YS |
1182 | *linger_req = req; |
1183 | } | |
1184 | ||
1dbb4399 | 1185 | ret = ceph_osdc_start_request(osdc, req, false); |
602adf40 YS |
1186 | if (ret < 0) |
1187 | goto done_err; | |
1188 | ||
1189 | if (!rbd_cb) { | |
1dbb4399 | 1190 | ret = ceph_osdc_wait_request(osdc, req); |
59c2be1e YS |
1191 | if (ver) |
1192 | *ver = le64_to_cpu(req->r_reassert_version.version); | |
bd919d45 AE |
1193 | dout("reassert_ver=%llu\n", |
1194 | (unsigned long long) | |
1195 | le64_to_cpu(req->r_reassert_version.version)); | |
602adf40 YS |
1196 | ceph_osdc_put_request(req); |
1197 | } | |
1198 | return ret; | |
1199 | ||
1200 | done_err: | |
1201 | bio_chain_put(req_data->bio); | |
1202 | ceph_osdc_put_request(req); | |
1203 | done_pages: | |
1fec7093 | 1204 | rbd_coll_end_req(req_data, ret, len); |
602adf40 | 1205 | kfree(req_data); |
602adf40 YS |
1206 | return ret; |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Ceph osd op callback | |
1211 | */ | |
1212 | static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) | |
1213 | { | |
1214 | struct rbd_request *req_data = req->r_priv; | |
1215 | struct ceph_osd_reply_head *replyhead; | |
1216 | struct ceph_osd_op *op; | |
1217 | __s32 rc; | |
1218 | u64 bytes; | |
1219 | int read_op; | |
1220 | ||
1221 | /* parse reply */ | |
1222 | replyhead = msg->front.iov_base; | |
1223 | WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); | |
1224 | op = (void *)(replyhead + 1); | |
1225 | rc = le32_to_cpu(replyhead->result); | |
1226 | bytes = le64_to_cpu(op->extent.length); | |
895cfcc8 | 1227 | read_op = (le16_to_cpu(op->op) == CEPH_OSD_OP_READ); |
602adf40 | 1228 | |
bd919d45 AE |
1229 | dout("rbd_req_cb bytes=%llu readop=%d rc=%d\n", |
1230 | (unsigned long long) bytes, read_op, (int) rc); | |
602adf40 YS |
1231 | |
1232 | if (rc == -ENOENT && read_op) { | |
1233 | zero_bio_chain(req_data->bio, 0); | |
1234 | rc = 0; | |
1235 | } else if (rc == 0 && read_op && bytes < req_data->len) { | |
1236 | zero_bio_chain(req_data->bio, bytes); | |
1237 | bytes = req_data->len; | |
1238 | } | |
1239 | ||
1fec7093 | 1240 | rbd_coll_end_req(req_data, rc, bytes); |
602adf40 YS |
1241 | |
1242 | if (req_data->bio) | |
1243 | bio_chain_put(req_data->bio); | |
1244 | ||
1245 | ceph_osdc_put_request(req); | |
1246 | kfree(req_data); | |
1247 | } | |
1248 | ||
59c2be1e YS |
1249 | static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg) |
1250 | { | |
1251 | ceph_osdc_put_request(req); | |
1252 | } | |
1253 | ||
602adf40 YS |
1254 | /* |
1255 | * Do a synchronous ceph osd operation | |
1256 | */ | |
0ce1a794 | 1257 | static int rbd_req_sync_op(struct rbd_device *rbd_dev, |
602adf40 YS |
1258 | struct ceph_snap_context *snapc, |
1259 | u64 snapid, | |
602adf40 | 1260 | int flags, |
913d2fdc | 1261 | struct ceph_osd_req_op *ops, |
aded07ea | 1262 | const char *object_name, |
f8d4de6e AE |
1263 | u64 ofs, u64 inbound_size, |
1264 | char *inbound, | |
59c2be1e YS |
1265 | struct ceph_osd_request **linger_req, |
1266 | u64 *ver) | |
602adf40 YS |
1267 | { |
1268 | int ret; | |
1269 | struct page **pages; | |
1270 | int num_pages; | |
913d2fdc | 1271 | |
aafb230e | 1272 | rbd_assert(ops != NULL); |
602adf40 | 1273 | |
f8d4de6e | 1274 | num_pages = calc_pages_for(ofs, inbound_size); |
602adf40 | 1275 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
b8d0638a DC |
1276 | if (IS_ERR(pages)) |
1277 | return PTR_ERR(pages); | |
602adf40 | 1278 | |
0ce1a794 | 1279 | ret = rbd_do_request(NULL, rbd_dev, snapc, snapid, |
f8d4de6e | 1280 | object_name, ofs, inbound_size, NULL, |
602adf40 YS |
1281 | pages, num_pages, |
1282 | flags, | |
1283 | ops, | |
1fec7093 | 1284 | NULL, 0, |
59c2be1e YS |
1285 | NULL, |
1286 | linger_req, ver); | |
602adf40 | 1287 | if (ret < 0) |
913d2fdc | 1288 | goto done; |
602adf40 | 1289 | |
f8d4de6e AE |
1290 | if ((flags & CEPH_OSD_FLAG_READ) && inbound) |
1291 | ret = ceph_copy_from_page_vector(pages, inbound, ofs, ret); | |
602adf40 | 1292 | |
602adf40 YS |
1293 | done: |
1294 | ceph_release_page_vector(pages, num_pages); | |
1295 | return ret; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Do an asynchronous ceph osd operation | |
1300 | */ | |
1301 | static int rbd_do_op(struct request *rq, | |
0ce1a794 | 1302 | struct rbd_device *rbd_dev, |
602adf40 | 1303 | struct ceph_snap_context *snapc, |
602adf40 | 1304 | u64 ofs, u64 len, |
1fec7093 YS |
1305 | struct bio *bio, |
1306 | struct rbd_req_coll *coll, | |
1307 | int coll_index) | |
602adf40 YS |
1308 | { |
1309 | char *seg_name; | |
1310 | u64 seg_ofs; | |
1311 | u64 seg_len; | |
1312 | int ret; | |
1313 | struct ceph_osd_req_op *ops; | |
1314 | u32 payload_len; | |
ff2e4bb5 AE |
1315 | int opcode; |
1316 | int flags; | |
4634246d | 1317 | u64 snapid; |
602adf40 | 1318 | |
65ccfe21 | 1319 | seg_name = rbd_segment_name(rbd_dev, ofs); |
602adf40 YS |
1320 | if (!seg_name) |
1321 | return -ENOMEM; | |
65ccfe21 AE |
1322 | seg_len = rbd_segment_length(rbd_dev, ofs, len); |
1323 | seg_ofs = rbd_segment_offset(rbd_dev, ofs); | |
602adf40 | 1324 | |
ff2e4bb5 AE |
1325 | if (rq_data_dir(rq) == WRITE) { |
1326 | opcode = CEPH_OSD_OP_WRITE; | |
1327 | flags = CEPH_OSD_FLAG_WRITE|CEPH_OSD_FLAG_ONDISK; | |
4634246d | 1328 | snapid = CEPH_NOSNAP; |
ff2e4bb5 AE |
1329 | payload_len = seg_len; |
1330 | } else { | |
1331 | opcode = CEPH_OSD_OP_READ; | |
1332 | flags = CEPH_OSD_FLAG_READ; | |
4634246d | 1333 | snapc = NULL; |
0d7dbfce | 1334 | snapid = rbd_dev->spec->snap_id; |
ff2e4bb5 AE |
1335 | payload_len = 0; |
1336 | } | |
602adf40 | 1337 | |
57cfc106 AE |
1338 | ret = -ENOMEM; |
1339 | ops = rbd_create_rw_ops(1, opcode, payload_len); | |
1340 | if (!ops) | |
602adf40 YS |
1341 | goto done; |
1342 | ||
1343 | /* we've taken care of segment sizes earlier when we | |
1344 | cloned the bios. We should never have a segment | |
1345 | truncated at this point */ | |
aafb230e | 1346 | rbd_assert(seg_len == len); |
602adf40 YS |
1347 | |
1348 | ret = rbd_do_request(rq, rbd_dev, snapc, snapid, | |
1349 | seg_name, seg_ofs, seg_len, | |
1350 | bio, | |
1351 | NULL, 0, | |
1352 | flags, | |
1353 | ops, | |
1fec7093 | 1354 | coll, coll_index, |
59c2be1e | 1355 | rbd_req_cb, 0, NULL); |
11f77002 SW |
1356 | |
1357 | rbd_destroy_ops(ops); | |
602adf40 YS |
1358 | done: |
1359 | kfree(seg_name); | |
1360 | return ret; | |
1361 | } | |
1362 | ||
602adf40 YS |
1363 | /* |
1364 | * Request sync osd read | |
1365 | */ | |
0ce1a794 | 1366 | static int rbd_req_sync_read(struct rbd_device *rbd_dev, |
602adf40 | 1367 | u64 snapid, |
aded07ea | 1368 | const char *object_name, |
602adf40 | 1369 | u64 ofs, u64 len, |
59c2be1e YS |
1370 | char *buf, |
1371 | u64 *ver) | |
602adf40 | 1372 | { |
913d2fdc AE |
1373 | struct ceph_osd_req_op *ops; |
1374 | int ret; | |
1375 | ||
1376 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_READ, 0); | |
1377 | if (!ops) | |
1378 | return -ENOMEM; | |
1379 | ||
1380 | ret = rbd_req_sync_op(rbd_dev, NULL, | |
b06e6a6b | 1381 | snapid, |
602adf40 | 1382 | CEPH_OSD_FLAG_READ, |
913d2fdc AE |
1383 | ops, object_name, ofs, len, buf, NULL, ver); |
1384 | rbd_destroy_ops(ops); | |
1385 | ||
1386 | return ret; | |
602adf40 YS |
1387 | } |
1388 | ||
1389 | /* | |
59c2be1e YS |
1390 | * Request sync osd watch |
1391 | */ | |
0ce1a794 | 1392 | static int rbd_req_sync_notify_ack(struct rbd_device *rbd_dev, |
59c2be1e | 1393 | u64 ver, |
7f0a24d8 | 1394 | u64 notify_id) |
59c2be1e YS |
1395 | { |
1396 | struct ceph_osd_req_op *ops; | |
11f77002 SW |
1397 | int ret; |
1398 | ||
57cfc106 AE |
1399 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_NOTIFY_ACK, 0); |
1400 | if (!ops) | |
1401 | return -ENOMEM; | |
59c2be1e | 1402 | |
a71b891b | 1403 | ops[0].watch.ver = cpu_to_le64(ver); |
59c2be1e YS |
1404 | ops[0].watch.cookie = notify_id; |
1405 | ops[0].watch.flag = 0; | |
1406 | ||
0ce1a794 | 1407 | ret = rbd_do_request(NULL, rbd_dev, NULL, CEPH_NOSNAP, |
7f0a24d8 | 1408 | rbd_dev->header_name, 0, 0, NULL, |
ad4f232f | 1409 | NULL, 0, |
59c2be1e YS |
1410 | CEPH_OSD_FLAG_READ, |
1411 | ops, | |
1fec7093 | 1412 | NULL, 0, |
59c2be1e YS |
1413 | rbd_simple_req_cb, 0, NULL); |
1414 | ||
1415 | rbd_destroy_ops(ops); | |
1416 | return ret; | |
1417 | } | |
1418 | ||
1419 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
1420 | { | |
0ce1a794 | 1421 | struct rbd_device *rbd_dev = (struct rbd_device *)data; |
a71b891b | 1422 | u64 hver; |
13143d2d SW |
1423 | int rc; |
1424 | ||
0ce1a794 | 1425 | if (!rbd_dev) |
59c2be1e YS |
1426 | return; |
1427 | ||
bd919d45 AE |
1428 | dout("rbd_watch_cb %s notify_id=%llu opcode=%u\n", |
1429 | rbd_dev->header_name, (unsigned long long) notify_id, | |
1430 | (unsigned int) opcode); | |
117973fb | 1431 | rc = rbd_dev_refresh(rbd_dev, &hver); |
13143d2d | 1432 | if (rc) |
06ecc6cb AE |
1433 | rbd_warn(rbd_dev, "got notification but failed to " |
1434 | " update snaps: %d\n", rc); | |
59c2be1e | 1435 | |
7f0a24d8 | 1436 | rbd_req_sync_notify_ack(rbd_dev, hver, notify_id); |
59c2be1e YS |
1437 | } |
1438 | ||
1439 | /* | |
1440 | * Request sync osd watch | |
1441 | */ | |
0e6f322d | 1442 | static int rbd_req_sync_watch(struct rbd_device *rbd_dev) |
59c2be1e YS |
1443 | { |
1444 | struct ceph_osd_req_op *ops; | |
0ce1a794 | 1445 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
57cfc106 | 1446 | int ret; |
59c2be1e | 1447 | |
57cfc106 AE |
1448 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1449 | if (!ops) | |
1450 | return -ENOMEM; | |
59c2be1e YS |
1451 | |
1452 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0, | |
0ce1a794 | 1453 | (void *)rbd_dev, &rbd_dev->watch_event); |
59c2be1e YS |
1454 | if (ret < 0) |
1455 | goto fail; | |
1456 | ||
0e6f322d | 1457 | ops[0].watch.ver = cpu_to_le64(rbd_dev->header.obj_version); |
0ce1a794 | 1458 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
59c2be1e YS |
1459 | ops[0].watch.flag = 1; |
1460 | ||
0ce1a794 | 1461 | ret = rbd_req_sync_op(rbd_dev, NULL, |
59c2be1e | 1462 | CEPH_NOSNAP, |
59c2be1e YS |
1463 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1464 | ops, | |
0e6f322d AE |
1465 | rbd_dev->header_name, |
1466 | 0, 0, NULL, | |
0ce1a794 | 1467 | &rbd_dev->watch_request, NULL); |
59c2be1e YS |
1468 | |
1469 | if (ret < 0) | |
1470 | goto fail_event; | |
1471 | ||
1472 | rbd_destroy_ops(ops); | |
1473 | return 0; | |
1474 | ||
1475 | fail_event: | |
0ce1a794 AE |
1476 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1477 | rbd_dev->watch_event = NULL; | |
59c2be1e YS |
1478 | fail: |
1479 | rbd_destroy_ops(ops); | |
1480 | return ret; | |
1481 | } | |
1482 | ||
79e3057c YS |
1483 | /* |
1484 | * Request sync osd unwatch | |
1485 | */ | |
070c633f | 1486 | static int rbd_req_sync_unwatch(struct rbd_device *rbd_dev) |
79e3057c YS |
1487 | { |
1488 | struct ceph_osd_req_op *ops; | |
57cfc106 | 1489 | int ret; |
79e3057c | 1490 | |
57cfc106 AE |
1491 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_WATCH, 0); |
1492 | if (!ops) | |
1493 | return -ENOMEM; | |
79e3057c YS |
1494 | |
1495 | ops[0].watch.ver = 0; | |
0ce1a794 | 1496 | ops[0].watch.cookie = cpu_to_le64(rbd_dev->watch_event->cookie); |
79e3057c YS |
1497 | ops[0].watch.flag = 0; |
1498 | ||
0ce1a794 | 1499 | ret = rbd_req_sync_op(rbd_dev, NULL, |
79e3057c | 1500 | CEPH_NOSNAP, |
79e3057c YS |
1501 | CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, |
1502 | ops, | |
070c633f AE |
1503 | rbd_dev->header_name, |
1504 | 0, 0, NULL, NULL, NULL); | |
1505 | ||
79e3057c YS |
1506 | |
1507 | rbd_destroy_ops(ops); | |
0ce1a794 AE |
1508 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
1509 | rbd_dev->watch_event = NULL; | |
79e3057c YS |
1510 | return ret; |
1511 | } | |
1512 | ||
602adf40 | 1513 | /* |
3cb4a687 | 1514 | * Synchronous osd object method call |
602adf40 | 1515 | */ |
0ce1a794 | 1516 | static int rbd_req_sync_exec(struct rbd_device *rbd_dev, |
aded07ea AE |
1517 | const char *object_name, |
1518 | const char *class_name, | |
1519 | const char *method_name, | |
3cb4a687 AE |
1520 | const char *outbound, |
1521 | size_t outbound_size, | |
f8d4de6e AE |
1522 | char *inbound, |
1523 | size_t inbound_size, | |
3cb4a687 | 1524 | int flags, |
59c2be1e | 1525 | u64 *ver) |
602adf40 YS |
1526 | { |
1527 | struct ceph_osd_req_op *ops; | |
aded07ea AE |
1528 | int class_name_len = strlen(class_name); |
1529 | int method_name_len = strlen(method_name); | |
3cb4a687 | 1530 | int payload_size; |
57cfc106 AE |
1531 | int ret; |
1532 | ||
3cb4a687 AE |
1533 | /* |
1534 | * Any input parameters required by the method we're calling | |
1535 | * will be sent along with the class and method names as | |
1536 | * part of the message payload. That data and its size are | |
1537 | * supplied via the indata and indata_len fields (named from | |
1538 | * the perspective of the server side) in the OSD request | |
1539 | * operation. | |
1540 | */ | |
1541 | payload_size = class_name_len + method_name_len + outbound_size; | |
1542 | ops = rbd_create_rw_ops(1, CEPH_OSD_OP_CALL, payload_size); | |
57cfc106 AE |
1543 | if (!ops) |
1544 | return -ENOMEM; | |
602adf40 | 1545 | |
aded07ea AE |
1546 | ops[0].cls.class_name = class_name; |
1547 | ops[0].cls.class_len = (__u8) class_name_len; | |
1548 | ops[0].cls.method_name = method_name; | |
1549 | ops[0].cls.method_len = (__u8) method_name_len; | |
602adf40 | 1550 | ops[0].cls.argc = 0; |
3cb4a687 AE |
1551 | ops[0].cls.indata = outbound; |
1552 | ops[0].cls.indata_len = outbound_size; | |
602adf40 | 1553 | |
0ce1a794 | 1554 | ret = rbd_req_sync_op(rbd_dev, NULL, |
602adf40 | 1555 | CEPH_NOSNAP, |
3cb4a687 | 1556 | flags, ops, |
f8d4de6e AE |
1557 | object_name, 0, inbound_size, inbound, |
1558 | NULL, ver); | |
602adf40 YS |
1559 | |
1560 | rbd_destroy_ops(ops); | |
1561 | ||
1562 | dout("cls_exec returned %d\n", ret); | |
1563 | return ret; | |
1564 | } | |
1565 | ||
1fec7093 YS |
1566 | static struct rbd_req_coll *rbd_alloc_coll(int num_reqs) |
1567 | { | |
1568 | struct rbd_req_coll *coll = | |
1569 | kzalloc(sizeof(struct rbd_req_coll) + | |
1570 | sizeof(struct rbd_req_status) * num_reqs, | |
1571 | GFP_ATOMIC); | |
1572 | ||
1573 | if (!coll) | |
1574 | return NULL; | |
1575 | coll->total = num_reqs; | |
1576 | kref_init(&coll->kref); | |
1577 | return coll; | |
1578 | } | |
1579 | ||
602adf40 YS |
1580 | /* |
1581 | * block device queue callback | |
1582 | */ | |
1583 | static void rbd_rq_fn(struct request_queue *q) | |
1584 | { | |
1585 | struct rbd_device *rbd_dev = q->queuedata; | |
1586 | struct request *rq; | |
602adf40 | 1587 | |
00f1f36f | 1588 | while ((rq = blk_fetch_request(q))) { |
602adf40 | 1589 | struct bio *bio; |
602adf40 | 1590 | bool do_write; |
bd919d45 | 1591 | unsigned int size; |
602adf40 | 1592 | u64 ofs; |
1fec7093 YS |
1593 | int num_segs, cur_seg = 0; |
1594 | struct rbd_req_coll *coll; | |
d1d25646 | 1595 | struct ceph_snap_context *snapc; |
f7760dad | 1596 | unsigned int bio_offset; |
602adf40 | 1597 | |
602adf40 YS |
1598 | dout("fetched request\n"); |
1599 | ||
1600 | /* filter out block requests we don't understand */ | |
1601 | if ((rq->cmd_type != REQ_TYPE_FS)) { | |
1602 | __blk_end_request_all(rq, 0); | |
00f1f36f | 1603 | continue; |
602adf40 YS |
1604 | } |
1605 | ||
1606 | /* deduce our operation (read, write) */ | |
1607 | do_write = (rq_data_dir(rq) == WRITE); | |
f84344f3 | 1608 | if (do_write && rbd_dev->mapping.read_only) { |
602adf40 | 1609 | __blk_end_request_all(rq, -EROFS); |
00f1f36f | 1610 | continue; |
602adf40 YS |
1611 | } |
1612 | ||
1613 | spin_unlock_irq(q->queue_lock); | |
1614 | ||
d1d25646 | 1615 | down_read(&rbd_dev->header_rwsem); |
e88a36ec | 1616 | |
daba5fdb | 1617 | if (!rbd_dev->exists) { |
0d7dbfce | 1618 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); |
e88a36ec | 1619 | up_read(&rbd_dev->header_rwsem); |
d1d25646 JD |
1620 | dout("request for non-existent snapshot"); |
1621 | spin_lock_irq(q->queue_lock); | |
1622 | __blk_end_request_all(rq, -ENXIO); | |
1623 | continue; | |
e88a36ec JD |
1624 | } |
1625 | ||
d1d25646 JD |
1626 | snapc = ceph_get_snap_context(rbd_dev->header.snapc); |
1627 | ||
1628 | up_read(&rbd_dev->header_rwsem); | |
1629 | ||
f7760dad AE |
1630 | size = blk_rq_bytes(rq); |
1631 | ofs = blk_rq_pos(rq) * SECTOR_SIZE; | |
1632 | bio = rq->bio; | |
1633 | ||
602adf40 YS |
1634 | dout("%s 0x%x bytes at 0x%llx\n", |
1635 | do_write ? "write" : "read", | |
bd919d45 | 1636 | size, (unsigned long long) blk_rq_pos(rq) * SECTOR_SIZE); |
602adf40 | 1637 | |
1fec7093 | 1638 | num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size); |
df111be6 AE |
1639 | if (num_segs <= 0) { |
1640 | spin_lock_irq(q->queue_lock); | |
1641 | __blk_end_request_all(rq, num_segs); | |
1642 | ceph_put_snap_context(snapc); | |
1643 | continue; | |
1644 | } | |
1fec7093 YS |
1645 | coll = rbd_alloc_coll(num_segs); |
1646 | if (!coll) { | |
1647 | spin_lock_irq(q->queue_lock); | |
1648 | __blk_end_request_all(rq, -ENOMEM); | |
d1d25646 | 1649 | ceph_put_snap_context(snapc); |
00f1f36f | 1650 | continue; |
1fec7093 YS |
1651 | } |
1652 | ||
f7760dad | 1653 | bio_offset = 0; |
602adf40 | 1654 | do { |
f7760dad AE |
1655 | u64 limit = rbd_segment_length(rbd_dev, ofs, size); |
1656 | unsigned int chain_size; | |
1657 | struct bio *bio_chain; | |
1658 | ||
1659 | BUG_ON(limit > (u64) UINT_MAX); | |
1660 | chain_size = (unsigned int) limit; | |
bd919d45 | 1661 | dout("rq->bio->bi_vcnt=%hu\n", rq->bio->bi_vcnt); |
f7760dad | 1662 | |
1fec7093 | 1663 | kref_get(&coll->kref); |
f7760dad AE |
1664 | |
1665 | /* Pass a cloned bio chain via an osd request */ | |
1666 | ||
1667 | bio_chain = bio_chain_clone_range(&bio, | |
1668 | &bio_offset, chain_size, | |
1669 | GFP_ATOMIC); | |
1670 | if (bio_chain) | |
4634246d | 1671 | (void) rbd_do_op(rq, rbd_dev, snapc, |
f7760dad AE |
1672 | ofs, chain_size, |
1673 | bio_chain, coll, cur_seg); | |
4634246d | 1674 | else |
1fec7093 | 1675 | rbd_coll_end_req_index(rq, coll, cur_seg, |
f7760dad AE |
1676 | -ENOMEM, chain_size); |
1677 | size -= chain_size; | |
1678 | ofs += chain_size; | |
602adf40 | 1679 | |
1fec7093 | 1680 | cur_seg++; |
602adf40 | 1681 | } while (size > 0); |
1fec7093 | 1682 | kref_put(&coll->kref, rbd_coll_release); |
602adf40 | 1683 | |
602adf40 | 1684 | spin_lock_irq(q->queue_lock); |
d1d25646 JD |
1685 | |
1686 | ceph_put_snap_context(snapc); | |
602adf40 YS |
1687 | } |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * a queue callback. Makes sure that we don't create a bio that spans across | |
1692 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 1693 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
1694 | */ |
1695 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
1696 | struct bio_vec *bvec) | |
1697 | { | |
1698 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
1699 | sector_t sector_offset; |
1700 | sector_t sectors_per_obj; | |
1701 | sector_t obj_sector_offset; | |
1702 | int ret; | |
1703 | ||
1704 | /* | |
1705 | * Find how far into its rbd object the partition-relative | |
1706 | * bio start sector is to offset relative to the enclosing | |
1707 | * device. | |
1708 | */ | |
1709 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
1710 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
1711 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
1712 | ||
1713 | /* | |
1714 | * Compute the number of bytes from that offset to the end | |
1715 | * of the object. Account for what's already used by the bio. | |
1716 | */ | |
1717 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
1718 | if (ret > bmd->bi_size) | |
1719 | ret -= bmd->bi_size; | |
1720 | else | |
1721 | ret = 0; | |
1722 | ||
1723 | /* | |
1724 | * Don't send back more than was asked for. And if the bio | |
1725 | * was empty, let the whole thing through because: "Note | |
1726 | * that a block device *must* allow a single page to be | |
1727 | * added to an empty bio." | |
1728 | */ | |
1729 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
1730 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
1731 | ret = (int) bvec->bv_len; | |
1732 | ||
1733 | return ret; | |
602adf40 YS |
1734 | } |
1735 | ||
1736 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
1737 | { | |
1738 | struct gendisk *disk = rbd_dev->disk; | |
1739 | ||
1740 | if (!disk) | |
1741 | return; | |
1742 | ||
602adf40 YS |
1743 | if (disk->flags & GENHD_FL_UP) |
1744 | del_gendisk(disk); | |
1745 | if (disk->queue) | |
1746 | blk_cleanup_queue(disk->queue); | |
1747 | put_disk(disk); | |
1748 | } | |
1749 | ||
1750 | /* | |
4156d998 AE |
1751 | * Read the complete header for the given rbd device. |
1752 | * | |
1753 | * Returns a pointer to a dynamically-allocated buffer containing | |
1754 | * the complete and validated header. Caller can pass the address | |
1755 | * of a variable that will be filled in with the version of the | |
1756 | * header object at the time it was read. | |
1757 | * | |
1758 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 1759 | */ |
4156d998 AE |
1760 | static struct rbd_image_header_ondisk * |
1761 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 1762 | { |
4156d998 | 1763 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 1764 | u32 snap_count = 0; |
4156d998 AE |
1765 | u64 names_size = 0; |
1766 | u32 want_count; | |
1767 | int ret; | |
602adf40 | 1768 | |
00f1f36f | 1769 | /* |
4156d998 AE |
1770 | * The complete header will include an array of its 64-bit |
1771 | * snapshot ids, followed by the names of those snapshots as | |
1772 | * a contiguous block of NUL-terminated strings. Note that | |
1773 | * the number of snapshots could change by the time we read | |
1774 | * it in, in which case we re-read it. | |
00f1f36f | 1775 | */ |
4156d998 AE |
1776 | do { |
1777 | size_t size; | |
1778 | ||
1779 | kfree(ondisk); | |
1780 | ||
1781 | size = sizeof (*ondisk); | |
1782 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
1783 | size += names_size; | |
1784 | ondisk = kmalloc(size, GFP_KERNEL); | |
1785 | if (!ondisk) | |
1786 | return ERR_PTR(-ENOMEM); | |
1787 | ||
1788 | ret = rbd_req_sync_read(rbd_dev, CEPH_NOSNAP, | |
0bed54dc | 1789 | rbd_dev->header_name, |
4156d998 AE |
1790 | 0, size, |
1791 | (char *) ondisk, version); | |
1792 | ||
1793 | if (ret < 0) | |
1794 | goto out_err; | |
1795 | if (WARN_ON((size_t) ret < size)) { | |
1796 | ret = -ENXIO; | |
06ecc6cb AE |
1797 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
1798 | size, ret); | |
4156d998 AE |
1799 | goto out_err; |
1800 | } | |
1801 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
1802 | ret = -ENXIO; | |
06ecc6cb | 1803 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 1804 | goto out_err; |
81e759fb | 1805 | } |
602adf40 | 1806 | |
4156d998 AE |
1807 | names_size = le64_to_cpu(ondisk->snap_names_len); |
1808 | want_count = snap_count; | |
1809 | snap_count = le32_to_cpu(ondisk->snap_count); | |
1810 | } while (snap_count != want_count); | |
00f1f36f | 1811 | |
4156d998 | 1812 | return ondisk; |
00f1f36f | 1813 | |
4156d998 AE |
1814 | out_err: |
1815 | kfree(ondisk); | |
1816 | ||
1817 | return ERR_PTR(ret); | |
1818 | } | |
1819 | ||
1820 | /* | |
1821 | * reload the ondisk the header | |
1822 | */ | |
1823 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
1824 | struct rbd_image_header *header) | |
1825 | { | |
1826 | struct rbd_image_header_ondisk *ondisk; | |
1827 | u64 ver = 0; | |
1828 | int ret; | |
602adf40 | 1829 | |
4156d998 AE |
1830 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
1831 | if (IS_ERR(ondisk)) | |
1832 | return PTR_ERR(ondisk); | |
1833 | ret = rbd_header_from_disk(header, ondisk); | |
1834 | if (ret >= 0) | |
1835 | header->obj_version = ver; | |
1836 | kfree(ondisk); | |
1837 | ||
1838 | return ret; | |
602adf40 YS |
1839 | } |
1840 | ||
41f38c2b | 1841 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
1842 | { |
1843 | struct rbd_snap *snap; | |
a0593290 | 1844 | struct rbd_snap *next; |
dfc5606d | 1845 | |
a0593290 | 1846 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) |
41f38c2b | 1847 | rbd_remove_snap_dev(snap); |
dfc5606d YS |
1848 | } |
1849 | ||
9478554a AE |
1850 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
1851 | { | |
1852 | sector_t size; | |
1853 | ||
0d7dbfce | 1854 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
1855 | return; |
1856 | ||
1857 | size = (sector_t) rbd_dev->header.image_size / SECTOR_SIZE; | |
1858 | dout("setting size to %llu sectors", (unsigned long long) size); | |
1859 | rbd_dev->mapping.size = (u64) size; | |
1860 | set_capacity(rbd_dev->disk, size); | |
1861 | } | |
1862 | ||
602adf40 YS |
1863 | /* |
1864 | * only read the first part of the ondisk header, without the snaps info | |
1865 | */ | |
117973fb | 1866 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
1867 | { |
1868 | int ret; | |
1869 | struct rbd_image_header h; | |
602adf40 YS |
1870 | |
1871 | ret = rbd_read_header(rbd_dev, &h); | |
1872 | if (ret < 0) | |
1873 | return ret; | |
1874 | ||
a51aa0c0 JD |
1875 | down_write(&rbd_dev->header_rwsem); |
1876 | ||
9478554a AE |
1877 | /* Update image size, and check for resize of mapped image */ |
1878 | rbd_dev->header.image_size = h.image_size; | |
1879 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 1880 | |
849b4260 | 1881 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 1882 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 1883 | kfree(rbd_dev->header.snap_names); |
d1d25646 JD |
1884 | /* osd requests may still refer to snapc */ |
1885 | ceph_put_snap_context(rbd_dev->header.snapc); | |
602adf40 | 1886 | |
b813623a AE |
1887 | if (hver) |
1888 | *hver = h.obj_version; | |
a71b891b | 1889 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 1890 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
1891 | rbd_dev->header.snapc = h.snapc; |
1892 | rbd_dev->header.snap_names = h.snap_names; | |
1893 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 AE |
1894 | /* Free the extra copy of the object prefix */ |
1895 | WARN_ON(strcmp(rbd_dev->header.object_prefix, h.object_prefix)); | |
1896 | kfree(h.object_prefix); | |
1897 | ||
304f6808 AE |
1898 | ret = rbd_dev_snaps_update(rbd_dev); |
1899 | if (!ret) | |
1900 | ret = rbd_dev_snaps_register(rbd_dev); | |
dfc5606d | 1901 | |
c666601a | 1902 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 1903 | |
dfc5606d | 1904 | return ret; |
602adf40 YS |
1905 | } |
1906 | ||
117973fb | 1907 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
1908 | { |
1909 | int ret; | |
1910 | ||
117973fb | 1911 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 1912 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
1913 | if (rbd_dev->image_format == 1) |
1914 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
1915 | else | |
1916 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 AE |
1917 | mutex_unlock(&ctl_mutex); |
1918 | ||
1919 | return ret; | |
1920 | } | |
1921 | ||
602adf40 YS |
1922 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
1923 | { | |
1924 | struct gendisk *disk; | |
1925 | struct request_queue *q; | |
593a9e7b | 1926 | u64 segment_size; |
602adf40 | 1927 | |
602adf40 | 1928 | /* create gendisk info */ |
602adf40 YS |
1929 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
1930 | if (!disk) | |
1fcdb8aa | 1931 | return -ENOMEM; |
602adf40 | 1932 | |
f0f8cef5 | 1933 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 1934 | rbd_dev->dev_id); |
602adf40 YS |
1935 | disk->major = rbd_dev->major; |
1936 | disk->first_minor = 0; | |
1937 | disk->fops = &rbd_bd_ops; | |
1938 | disk->private_data = rbd_dev; | |
1939 | ||
1940 | /* init rq */ | |
602adf40 YS |
1941 | q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock); |
1942 | if (!q) | |
1943 | goto out_disk; | |
029bcbd8 | 1944 | |
593a9e7b AE |
1945 | /* We use the default size, but let's be explicit about it. */ |
1946 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
1947 | ||
029bcbd8 | 1948 | /* set io sizes to object size */ |
593a9e7b AE |
1949 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
1950 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
1951 | blk_queue_max_segment_size(q, segment_size); | |
1952 | blk_queue_io_min(q, segment_size); | |
1953 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 1954 | |
602adf40 YS |
1955 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
1956 | disk->queue = q; | |
1957 | ||
1958 | q->queuedata = rbd_dev; | |
1959 | ||
1960 | rbd_dev->disk = disk; | |
602adf40 | 1961 | |
12f02944 AE |
1962 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
1963 | ||
602adf40 | 1964 | return 0; |
602adf40 YS |
1965 | out_disk: |
1966 | put_disk(disk); | |
1fcdb8aa AE |
1967 | |
1968 | return -ENOMEM; | |
602adf40 YS |
1969 | } |
1970 | ||
dfc5606d YS |
1971 | /* |
1972 | sysfs | |
1973 | */ | |
1974 | ||
593a9e7b AE |
1975 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
1976 | { | |
1977 | return container_of(dev, struct rbd_device, dev); | |
1978 | } | |
1979 | ||
dfc5606d YS |
1980 | static ssize_t rbd_size_show(struct device *dev, |
1981 | struct device_attribute *attr, char *buf) | |
1982 | { | |
593a9e7b | 1983 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 JD |
1984 | sector_t size; |
1985 | ||
1986 | down_read(&rbd_dev->header_rwsem); | |
1987 | size = get_capacity(rbd_dev->disk); | |
1988 | up_read(&rbd_dev->header_rwsem); | |
dfc5606d | 1989 | |
a51aa0c0 | 1990 | return sprintf(buf, "%llu\n", (unsigned long long) size * SECTOR_SIZE); |
dfc5606d YS |
1991 | } |
1992 | ||
34b13184 AE |
1993 | /* |
1994 | * Note this shows the features for whatever's mapped, which is not | |
1995 | * necessarily the base image. | |
1996 | */ | |
1997 | static ssize_t rbd_features_show(struct device *dev, | |
1998 | struct device_attribute *attr, char *buf) | |
1999 | { | |
2000 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2001 | ||
2002 | return sprintf(buf, "0x%016llx\n", | |
2003 | (unsigned long long) rbd_dev->mapping.features); | |
2004 | } | |
2005 | ||
dfc5606d YS |
2006 | static ssize_t rbd_major_show(struct device *dev, |
2007 | struct device_attribute *attr, char *buf) | |
2008 | { | |
593a9e7b | 2009 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 2010 | |
dfc5606d YS |
2011 | return sprintf(buf, "%d\n", rbd_dev->major); |
2012 | } | |
2013 | ||
2014 | static ssize_t rbd_client_id_show(struct device *dev, | |
2015 | struct device_attribute *attr, char *buf) | |
602adf40 | 2016 | { |
593a9e7b | 2017 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2018 | |
1dbb4399 AE |
2019 | return sprintf(buf, "client%lld\n", |
2020 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
2021 | } |
2022 | ||
dfc5606d YS |
2023 | static ssize_t rbd_pool_show(struct device *dev, |
2024 | struct device_attribute *attr, char *buf) | |
602adf40 | 2025 | { |
593a9e7b | 2026 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2027 | |
0d7dbfce | 2028 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
2029 | } |
2030 | ||
9bb2f334 AE |
2031 | static ssize_t rbd_pool_id_show(struct device *dev, |
2032 | struct device_attribute *attr, char *buf) | |
2033 | { | |
2034 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2035 | ||
0d7dbfce AE |
2036 | return sprintf(buf, "%llu\n", |
2037 | (unsigned long long) rbd_dev->spec->pool_id); | |
9bb2f334 AE |
2038 | } |
2039 | ||
dfc5606d YS |
2040 | static ssize_t rbd_name_show(struct device *dev, |
2041 | struct device_attribute *attr, char *buf) | |
2042 | { | |
593a9e7b | 2043 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2044 | |
a92ffdf8 AE |
2045 | if (rbd_dev->spec->image_name) |
2046 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
2047 | ||
2048 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
2049 | } |
2050 | ||
589d30e0 AE |
2051 | static ssize_t rbd_image_id_show(struct device *dev, |
2052 | struct device_attribute *attr, char *buf) | |
2053 | { | |
2054 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2055 | ||
0d7dbfce | 2056 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
2057 | } |
2058 | ||
34b13184 AE |
2059 | /* |
2060 | * Shows the name of the currently-mapped snapshot (or | |
2061 | * RBD_SNAP_HEAD_NAME for the base image). | |
2062 | */ | |
dfc5606d YS |
2063 | static ssize_t rbd_snap_show(struct device *dev, |
2064 | struct device_attribute *attr, | |
2065 | char *buf) | |
2066 | { | |
593a9e7b | 2067 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 2068 | |
0d7dbfce | 2069 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
2070 | } |
2071 | ||
86b00e0d AE |
2072 | /* |
2073 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
2074 | * for the parent image. If there is no parent, simply shows | |
2075 | * "(no parent image)". | |
2076 | */ | |
2077 | static ssize_t rbd_parent_show(struct device *dev, | |
2078 | struct device_attribute *attr, | |
2079 | char *buf) | |
2080 | { | |
2081 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
2082 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
2083 | int count; | |
2084 | char *bufp = buf; | |
2085 | ||
2086 | if (!spec) | |
2087 | return sprintf(buf, "(no parent image)\n"); | |
2088 | ||
2089 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
2090 | (unsigned long long) spec->pool_id, spec->pool_name); | |
2091 | if (count < 0) | |
2092 | return count; | |
2093 | bufp += count; | |
2094 | ||
2095 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
2096 | spec->image_name ? spec->image_name : "(unknown)"); | |
2097 | if (count < 0) | |
2098 | return count; | |
2099 | bufp += count; | |
2100 | ||
2101 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
2102 | (unsigned long long) spec->snap_id, spec->snap_name); | |
2103 | if (count < 0) | |
2104 | return count; | |
2105 | bufp += count; | |
2106 | ||
2107 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
2108 | if (count < 0) | |
2109 | return count; | |
2110 | bufp += count; | |
2111 | ||
2112 | return (ssize_t) (bufp - buf); | |
2113 | } | |
2114 | ||
dfc5606d YS |
2115 | static ssize_t rbd_image_refresh(struct device *dev, |
2116 | struct device_attribute *attr, | |
2117 | const char *buf, | |
2118 | size_t size) | |
2119 | { | |
593a9e7b | 2120 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 2121 | int ret; |
602adf40 | 2122 | |
117973fb | 2123 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
2124 | |
2125 | return ret < 0 ? ret : size; | |
dfc5606d | 2126 | } |
602adf40 | 2127 | |
dfc5606d | 2128 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 2129 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
2130 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
2131 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
2132 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 2133 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 2134 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 2135 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
2136 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
2137 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 2138 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
2139 | |
2140 | static struct attribute *rbd_attrs[] = { | |
2141 | &dev_attr_size.attr, | |
34b13184 | 2142 | &dev_attr_features.attr, |
dfc5606d YS |
2143 | &dev_attr_major.attr, |
2144 | &dev_attr_client_id.attr, | |
2145 | &dev_attr_pool.attr, | |
9bb2f334 | 2146 | &dev_attr_pool_id.attr, |
dfc5606d | 2147 | &dev_attr_name.attr, |
589d30e0 | 2148 | &dev_attr_image_id.attr, |
dfc5606d | 2149 | &dev_attr_current_snap.attr, |
86b00e0d | 2150 | &dev_attr_parent.attr, |
dfc5606d | 2151 | &dev_attr_refresh.attr, |
dfc5606d YS |
2152 | NULL |
2153 | }; | |
2154 | ||
2155 | static struct attribute_group rbd_attr_group = { | |
2156 | .attrs = rbd_attrs, | |
2157 | }; | |
2158 | ||
2159 | static const struct attribute_group *rbd_attr_groups[] = { | |
2160 | &rbd_attr_group, | |
2161 | NULL | |
2162 | }; | |
2163 | ||
2164 | static void rbd_sysfs_dev_release(struct device *dev) | |
2165 | { | |
2166 | } | |
2167 | ||
2168 | static struct device_type rbd_device_type = { | |
2169 | .name = "rbd", | |
2170 | .groups = rbd_attr_groups, | |
2171 | .release = rbd_sysfs_dev_release, | |
2172 | }; | |
2173 | ||
2174 | ||
2175 | /* | |
2176 | sysfs - snapshots | |
2177 | */ | |
2178 | ||
2179 | static ssize_t rbd_snap_size_show(struct device *dev, | |
2180 | struct device_attribute *attr, | |
2181 | char *buf) | |
2182 | { | |
2183 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2184 | ||
3591538f | 2185 | return sprintf(buf, "%llu\n", (unsigned long long)snap->size); |
dfc5606d YS |
2186 | } |
2187 | ||
2188 | static ssize_t rbd_snap_id_show(struct device *dev, | |
2189 | struct device_attribute *attr, | |
2190 | char *buf) | |
2191 | { | |
2192 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2193 | ||
3591538f | 2194 | return sprintf(buf, "%llu\n", (unsigned long long)snap->id); |
dfc5606d YS |
2195 | } |
2196 | ||
34b13184 AE |
2197 | static ssize_t rbd_snap_features_show(struct device *dev, |
2198 | struct device_attribute *attr, | |
2199 | char *buf) | |
2200 | { | |
2201 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2202 | ||
2203 | return sprintf(buf, "0x%016llx\n", | |
2204 | (unsigned long long) snap->features); | |
2205 | } | |
2206 | ||
dfc5606d YS |
2207 | static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL); |
2208 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); | |
34b13184 | 2209 | static DEVICE_ATTR(snap_features, S_IRUGO, rbd_snap_features_show, NULL); |
dfc5606d YS |
2210 | |
2211 | static struct attribute *rbd_snap_attrs[] = { | |
2212 | &dev_attr_snap_size.attr, | |
2213 | &dev_attr_snap_id.attr, | |
34b13184 | 2214 | &dev_attr_snap_features.attr, |
dfc5606d YS |
2215 | NULL, |
2216 | }; | |
2217 | ||
2218 | static struct attribute_group rbd_snap_attr_group = { | |
2219 | .attrs = rbd_snap_attrs, | |
2220 | }; | |
2221 | ||
2222 | static void rbd_snap_dev_release(struct device *dev) | |
2223 | { | |
2224 | struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev); | |
2225 | kfree(snap->name); | |
2226 | kfree(snap); | |
2227 | } | |
2228 | ||
2229 | static const struct attribute_group *rbd_snap_attr_groups[] = { | |
2230 | &rbd_snap_attr_group, | |
2231 | NULL | |
2232 | }; | |
2233 | ||
2234 | static struct device_type rbd_snap_device_type = { | |
2235 | .groups = rbd_snap_attr_groups, | |
2236 | .release = rbd_snap_dev_release, | |
2237 | }; | |
2238 | ||
8b8fb99c AE |
2239 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
2240 | { | |
2241 | kref_get(&spec->kref); | |
2242 | ||
2243 | return spec; | |
2244 | } | |
2245 | ||
2246 | static void rbd_spec_free(struct kref *kref); | |
2247 | static void rbd_spec_put(struct rbd_spec *spec) | |
2248 | { | |
2249 | if (spec) | |
2250 | kref_put(&spec->kref, rbd_spec_free); | |
2251 | } | |
2252 | ||
2253 | static struct rbd_spec *rbd_spec_alloc(void) | |
2254 | { | |
2255 | struct rbd_spec *spec; | |
2256 | ||
2257 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
2258 | if (!spec) | |
2259 | return NULL; | |
2260 | kref_init(&spec->kref); | |
2261 | ||
2262 | rbd_spec_put(rbd_spec_get(spec)); /* TEMPORARY */ | |
2263 | ||
2264 | return spec; | |
2265 | } | |
2266 | ||
2267 | static void rbd_spec_free(struct kref *kref) | |
2268 | { | |
2269 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
2270 | ||
2271 | kfree(spec->pool_name); | |
2272 | kfree(spec->image_id); | |
2273 | kfree(spec->image_name); | |
2274 | kfree(spec->snap_name); | |
2275 | kfree(spec); | |
2276 | } | |
2277 | ||
c53d5893 AE |
2278 | struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
2279 | struct rbd_spec *spec) | |
2280 | { | |
2281 | struct rbd_device *rbd_dev; | |
2282 | ||
2283 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
2284 | if (!rbd_dev) | |
2285 | return NULL; | |
2286 | ||
2287 | spin_lock_init(&rbd_dev->lock); | |
2288 | INIT_LIST_HEAD(&rbd_dev->node); | |
2289 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
2290 | init_rwsem(&rbd_dev->header_rwsem); | |
2291 | ||
2292 | rbd_dev->spec = spec; | |
2293 | rbd_dev->rbd_client = rbdc; | |
2294 | ||
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), | |
2408 | (char *) &size_buf, sizeof (size_buf), | |
2409 | CEPH_OSD_FLAG_READ, NULL); | |
2410 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2411 | if (ret < 0) | |
2412 | return ret; | |
2413 | ||
2414 | *order = size_buf.order; | |
2415 | *snap_size = le64_to_cpu(size_buf.size); | |
2416 | ||
2417 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
2418 | (unsigned long long) snap_id, (unsigned int) *order, | |
2419 | (unsigned long long) *snap_size); | |
2420 | ||
2421 | return 0; | |
2422 | } | |
2423 | ||
2424 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
2425 | { | |
2426 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
2427 | &rbd_dev->header.obj_order, | |
2428 | &rbd_dev->header.image_size); | |
2429 | } | |
2430 | ||
1e130199 AE |
2431 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
2432 | { | |
2433 | void *reply_buf; | |
2434 | int ret; | |
2435 | void *p; | |
2436 | ||
2437 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
2438 | if (!reply_buf) | |
2439 | return -ENOMEM; | |
2440 | ||
2441 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2442 | "rbd", "get_object_prefix", | |
2443 | NULL, 0, | |
2444 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, | |
2445 | CEPH_OSD_FLAG_READ, NULL); | |
2446 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2447 | if (ret < 0) | |
2448 | goto out; | |
a0ea3a40 | 2449 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
1e130199 AE |
2450 | |
2451 | p = reply_buf; | |
2452 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
2453 | p + RBD_OBJ_PREFIX_LEN_MAX, | |
2454 | NULL, GFP_NOIO); | |
2455 | ||
2456 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
2457 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
2458 | rbd_dev->header.object_prefix = NULL; | |
2459 | } else { | |
2460 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
2461 | } | |
2462 | ||
2463 | out: | |
2464 | kfree(reply_buf); | |
2465 | ||
2466 | return ret; | |
2467 | } | |
2468 | ||
b1b5402a AE |
2469 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
2470 | u64 *snap_features) | |
2471 | { | |
2472 | __le64 snapid = cpu_to_le64(snap_id); | |
2473 | struct { | |
2474 | __le64 features; | |
2475 | __le64 incompat; | |
2476 | } features_buf = { 0 }; | |
d889140c | 2477 | u64 incompat; |
b1b5402a AE |
2478 | int ret; |
2479 | ||
2480 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2481 | "rbd", "get_features", | |
2482 | (char *) &snapid, sizeof (snapid), | |
2483 | (char *) &features_buf, sizeof (features_buf), | |
2484 | CEPH_OSD_FLAG_READ, NULL); | |
2485 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2486 | if (ret < 0) | |
2487 | return ret; | |
d889140c AE |
2488 | |
2489 | incompat = le64_to_cpu(features_buf.incompat); | |
2490 | if (incompat & ~RBD_FEATURES_ALL) | |
b8f5c6ed | 2491 | return -ENXIO; |
d889140c | 2492 | |
b1b5402a AE |
2493 | *snap_features = le64_to_cpu(features_buf.features); |
2494 | ||
2495 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
2496 | (unsigned long long) snap_id, | |
2497 | (unsigned long long) *snap_features, | |
2498 | (unsigned long long) le64_to_cpu(features_buf.incompat)); | |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
2504 | { | |
2505 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
2506 | &rbd_dev->header.features); | |
2507 | } | |
2508 | ||
86b00e0d AE |
2509 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
2510 | { | |
2511 | struct rbd_spec *parent_spec; | |
2512 | size_t size; | |
2513 | void *reply_buf = NULL; | |
2514 | __le64 snapid; | |
2515 | void *p; | |
2516 | void *end; | |
2517 | char *image_id; | |
2518 | u64 overlap; | |
86b00e0d AE |
2519 | int ret; |
2520 | ||
2521 | parent_spec = rbd_spec_alloc(); | |
2522 | if (!parent_spec) | |
2523 | return -ENOMEM; | |
2524 | ||
2525 | size = sizeof (__le64) + /* pool_id */ | |
2526 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
2527 | sizeof (__le64) + /* snap_id */ | |
2528 | sizeof (__le64); /* overlap */ | |
2529 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2530 | if (!reply_buf) { | |
2531 | ret = -ENOMEM; | |
2532 | goto out_err; | |
2533 | } | |
2534 | ||
2535 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
2536 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2537 | "rbd", "get_parent", | |
2538 | (char *) &snapid, sizeof (snapid), | |
2539 | (char *) reply_buf, size, | |
2540 | CEPH_OSD_FLAG_READ, NULL); | |
2541 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2542 | if (ret < 0) | |
2543 | goto out_err; | |
2544 | ||
2545 | ret = -ERANGE; | |
2546 | p = reply_buf; | |
2547 | end = (char *) reply_buf + size; | |
2548 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); | |
2549 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
2550 | goto out; /* No parent? No problem. */ | |
2551 | ||
979ed480 | 2552 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
2553 | if (IS_ERR(image_id)) { |
2554 | ret = PTR_ERR(image_id); | |
2555 | goto out_err; | |
2556 | } | |
2557 | parent_spec->image_id = image_id; | |
2558 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
2559 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
2560 | ||
2561 | rbd_dev->parent_overlap = overlap; | |
2562 | rbd_dev->parent_spec = parent_spec; | |
2563 | parent_spec = NULL; /* rbd_dev now owns this */ | |
2564 | out: | |
2565 | ret = 0; | |
2566 | out_err: | |
2567 | kfree(reply_buf); | |
2568 | rbd_spec_put(parent_spec); | |
2569 | ||
2570 | return ret; | |
2571 | } | |
2572 | ||
9e15b77d AE |
2573 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
2574 | { | |
2575 | size_t image_id_size; | |
2576 | char *image_id; | |
2577 | void *p; | |
2578 | void *end; | |
2579 | size_t size; | |
2580 | void *reply_buf = NULL; | |
2581 | size_t len = 0; | |
2582 | char *image_name = NULL; | |
2583 | int ret; | |
2584 | ||
2585 | rbd_assert(!rbd_dev->spec->image_name); | |
2586 | ||
69e7a02f AE |
2587 | len = strlen(rbd_dev->spec->image_id); |
2588 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
2589 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
2590 | if (!image_id) | |
2591 | return NULL; | |
2592 | ||
2593 | p = image_id; | |
2594 | end = (char *) image_id + image_id_size; | |
69e7a02f | 2595 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32) len); |
9e15b77d AE |
2596 | |
2597 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
2598 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2599 | if (!reply_buf) | |
2600 | goto out; | |
2601 | ||
2602 | ret = rbd_req_sync_exec(rbd_dev, RBD_DIRECTORY, | |
2603 | "rbd", "dir_get_name", | |
2604 | image_id, image_id_size, | |
2605 | (char *) reply_buf, size, | |
2606 | CEPH_OSD_FLAG_READ, NULL); | |
2607 | if (ret < 0) | |
2608 | goto out; | |
2609 | p = reply_buf; | |
2610 | end = (char *) reply_buf + size; | |
2611 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); | |
2612 | if (IS_ERR(image_name)) | |
2613 | image_name = NULL; | |
2614 | else | |
2615 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
2616 | out: | |
2617 | kfree(reply_buf); | |
2618 | kfree(image_id); | |
2619 | ||
2620 | return image_name; | |
2621 | } | |
2622 | ||
2623 | /* | |
2624 | * When a parent image gets probed, we only have the pool, image, | |
2625 | * and snapshot ids but not the names of any of them. This call | |
2626 | * is made later to fill in those names. It has to be done after | |
2627 | * rbd_dev_snaps_update() has completed because some of the | |
2628 | * information (in particular, snapshot name) is not available | |
2629 | * until then. | |
2630 | */ | |
2631 | static int rbd_dev_probe_update_spec(struct rbd_device *rbd_dev) | |
2632 | { | |
2633 | struct ceph_osd_client *osdc; | |
2634 | const char *name; | |
2635 | void *reply_buf = NULL; | |
2636 | int ret; | |
2637 | ||
2638 | if (rbd_dev->spec->pool_name) | |
2639 | return 0; /* Already have the names */ | |
2640 | ||
2641 | /* Look up the pool name */ | |
2642 | ||
2643 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2644 | name = ceph_pg_pool_name_by_id(osdc->osdmap, rbd_dev->spec->pool_id); | |
2645 | if (!name) | |
2646 | return -EIO; /* pool id too large (>= 2^31) */ | |
2647 | ||
2648 | rbd_dev->spec->pool_name = kstrdup(name, GFP_KERNEL); | |
2649 | if (!rbd_dev->spec->pool_name) | |
2650 | return -ENOMEM; | |
2651 | ||
2652 | /* Fetch the image name; tolerate failure here */ | |
2653 | ||
2654 | name = rbd_dev_image_name(rbd_dev); | |
69e7a02f | 2655 | if (name) |
9e15b77d | 2656 | rbd_dev->spec->image_name = (char *) name; |
69e7a02f | 2657 | else |
06ecc6cb | 2658 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d AE |
2659 | |
2660 | /* Look up the snapshot name. */ | |
2661 | ||
2662 | name = rbd_snap_name(rbd_dev, rbd_dev->spec->snap_id); | |
2663 | if (!name) { | |
2664 | ret = -EIO; | |
2665 | goto out_err; | |
2666 | } | |
2667 | rbd_dev->spec->snap_name = kstrdup(name, GFP_KERNEL); | |
2668 | if(!rbd_dev->spec->snap_name) | |
2669 | goto out_err; | |
2670 | ||
2671 | return 0; | |
2672 | out_err: | |
2673 | kfree(reply_buf); | |
2674 | kfree(rbd_dev->spec->pool_name); | |
2675 | rbd_dev->spec->pool_name = NULL; | |
2676 | ||
2677 | return ret; | |
2678 | } | |
2679 | ||
6e14b1a6 | 2680 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
2681 | { |
2682 | size_t size; | |
2683 | int ret; | |
2684 | void *reply_buf; | |
2685 | void *p; | |
2686 | void *end; | |
2687 | u64 seq; | |
2688 | u32 snap_count; | |
2689 | struct ceph_snap_context *snapc; | |
2690 | u32 i; | |
2691 | ||
2692 | /* | |
2693 | * We'll need room for the seq value (maximum snapshot id), | |
2694 | * snapshot count, and array of that many snapshot ids. | |
2695 | * For now we have a fixed upper limit on the number we're | |
2696 | * prepared to receive. | |
2697 | */ | |
2698 | size = sizeof (__le64) + sizeof (__le32) + | |
2699 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
2700 | reply_buf = kzalloc(size, GFP_KERNEL); | |
2701 | if (!reply_buf) | |
2702 | return -ENOMEM; | |
2703 | ||
2704 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2705 | "rbd", "get_snapcontext", | |
2706 | NULL, 0, | |
2707 | reply_buf, size, | |
6e14b1a6 | 2708 | CEPH_OSD_FLAG_READ, ver); |
35d489f9 AE |
2709 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); |
2710 | if (ret < 0) | |
2711 | goto out; | |
2712 | ||
2713 | ret = -ERANGE; | |
2714 | p = reply_buf; | |
2715 | end = (char *) reply_buf + size; | |
2716 | ceph_decode_64_safe(&p, end, seq, out); | |
2717 | ceph_decode_32_safe(&p, end, snap_count, out); | |
2718 | ||
2719 | /* | |
2720 | * Make sure the reported number of snapshot ids wouldn't go | |
2721 | * beyond the end of our buffer. But before checking that, | |
2722 | * make sure the computed size of the snapshot context we | |
2723 | * allocate is representable in a size_t. | |
2724 | */ | |
2725 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
2726 | / sizeof (u64)) { | |
2727 | ret = -EINVAL; | |
2728 | goto out; | |
2729 | } | |
2730 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
2731 | goto out; | |
2732 | ||
2733 | size = sizeof (struct ceph_snap_context) + | |
2734 | snap_count * sizeof (snapc->snaps[0]); | |
2735 | snapc = kmalloc(size, GFP_KERNEL); | |
2736 | if (!snapc) { | |
2737 | ret = -ENOMEM; | |
2738 | goto out; | |
2739 | } | |
2740 | ||
2741 | atomic_set(&snapc->nref, 1); | |
2742 | snapc->seq = seq; | |
2743 | snapc->num_snaps = snap_count; | |
2744 | for (i = 0; i < snap_count; i++) | |
2745 | snapc->snaps[i] = ceph_decode_64(&p); | |
2746 | ||
2747 | rbd_dev->header.snapc = snapc; | |
2748 | ||
2749 | dout(" snap context seq = %llu, snap_count = %u\n", | |
2750 | (unsigned long long) seq, (unsigned int) snap_count); | |
2751 | ||
2752 | out: | |
2753 | kfree(reply_buf); | |
2754 | ||
2755 | return 0; | |
2756 | } | |
2757 | ||
b8b1e2db AE |
2758 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
2759 | { | |
2760 | size_t size; | |
2761 | void *reply_buf; | |
2762 | __le64 snap_id; | |
2763 | int ret; | |
2764 | void *p; | |
2765 | void *end; | |
b8b1e2db AE |
2766 | char *snap_name; |
2767 | ||
2768 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
2769 | reply_buf = kmalloc(size, GFP_KERNEL); | |
2770 | if (!reply_buf) | |
2771 | return ERR_PTR(-ENOMEM); | |
2772 | ||
2773 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); | |
2774 | ret = rbd_req_sync_exec(rbd_dev, rbd_dev->header_name, | |
2775 | "rbd", "get_snapshot_name", | |
2776 | (char *) &snap_id, sizeof (snap_id), | |
2777 | reply_buf, size, | |
2778 | CEPH_OSD_FLAG_READ, NULL); | |
2779 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
2780 | if (ret < 0) | |
2781 | goto out; | |
2782 | ||
2783 | p = reply_buf; | |
2784 | end = (char *) reply_buf + size; | |
e5c35534 | 2785 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
b8b1e2db AE |
2786 | if (IS_ERR(snap_name)) { |
2787 | ret = PTR_ERR(snap_name); | |
2788 | goto out; | |
2789 | } else { | |
2790 | dout(" snap_id 0x%016llx snap_name = %s\n", | |
2791 | (unsigned long long) le64_to_cpu(snap_id), snap_name); | |
2792 | } | |
2793 | kfree(reply_buf); | |
2794 | ||
2795 | return snap_name; | |
2796 | out: | |
2797 | kfree(reply_buf); | |
2798 | ||
2799 | return ERR_PTR(ret); | |
2800 | } | |
2801 | ||
2802 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2803 | u64 *snap_size, u64 *snap_features) | |
2804 | { | |
2805 | __le64 snap_id; | |
2806 | u8 order; | |
2807 | int ret; | |
2808 | ||
2809 | snap_id = rbd_dev->header.snapc->snaps[which]; | |
2810 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, &order, snap_size); | |
2811 | if (ret) | |
2812 | return ERR_PTR(ret); | |
2813 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, snap_features); | |
2814 | if (ret) | |
2815 | return ERR_PTR(ret); | |
2816 | ||
2817 | return rbd_dev_v2_snap_name(rbd_dev, which); | |
2818 | } | |
2819 | ||
2820 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
2821 | u64 *snap_size, u64 *snap_features) | |
2822 | { | |
2823 | if (rbd_dev->image_format == 1) | |
2824 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
2825 | snap_size, snap_features); | |
2826 | if (rbd_dev->image_format == 2) | |
2827 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
2828 | snap_size, snap_features); | |
2829 | return ERR_PTR(-EINVAL); | |
2830 | } | |
2831 | ||
117973fb AE |
2832 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
2833 | { | |
2834 | int ret; | |
2835 | __u8 obj_order; | |
2836 | ||
2837 | down_write(&rbd_dev->header_rwsem); | |
2838 | ||
2839 | /* Grab old order first, to see if it changes */ | |
2840 | ||
2841 | obj_order = rbd_dev->header.obj_order, | |
2842 | ret = rbd_dev_v2_image_size(rbd_dev); | |
2843 | if (ret) | |
2844 | goto out; | |
2845 | if (rbd_dev->header.obj_order != obj_order) { | |
2846 | ret = -EIO; | |
2847 | goto out; | |
2848 | } | |
2849 | rbd_update_mapping_size(rbd_dev); | |
2850 | ||
2851 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
2852 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
2853 | if (ret) | |
2854 | goto out; | |
2855 | ret = rbd_dev_snaps_update(rbd_dev); | |
2856 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
2857 | if (ret) | |
2858 | goto out; | |
2859 | ret = rbd_dev_snaps_register(rbd_dev); | |
2860 | dout("rbd_dev_snaps_register returned %d\n", ret); | |
2861 | out: | |
2862 | up_write(&rbd_dev->header_rwsem); | |
2863 | ||
2864 | return ret; | |
2865 | } | |
2866 | ||
dfc5606d | 2867 | /* |
35938150 AE |
2868 | * Scan the rbd device's current snapshot list and compare it to the |
2869 | * newly-received snapshot context. Remove any existing snapshots | |
2870 | * not present in the new snapshot context. Add a new snapshot for | |
2871 | * any snaphots in the snapshot context not in the current list. | |
2872 | * And verify there are no changes to snapshots we already know | |
2873 | * about. | |
2874 | * | |
2875 | * Assumes the snapshots in the snapshot context are sorted by | |
2876 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
2877 | * are also maintained in that order.) | |
dfc5606d | 2878 | */ |
304f6808 | 2879 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 2880 | { |
35938150 AE |
2881 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
2882 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
2883 | struct list_head *head = &rbd_dev->snaps; |
2884 | struct list_head *links = head->next; | |
2885 | u32 index = 0; | |
dfc5606d | 2886 | |
9fcbb800 | 2887 | dout("%s: snap count is %u\n", __func__, (unsigned int) snap_count); |
35938150 AE |
2888 | while (index < snap_count || links != head) { |
2889 | u64 snap_id; | |
2890 | struct rbd_snap *snap; | |
cd892126 AE |
2891 | char *snap_name; |
2892 | u64 snap_size = 0; | |
2893 | u64 snap_features = 0; | |
dfc5606d | 2894 | |
35938150 AE |
2895 | snap_id = index < snap_count ? snapc->snaps[index] |
2896 | : CEPH_NOSNAP; | |
2897 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
2898 | : NULL; | |
aafb230e | 2899 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 2900 | |
35938150 AE |
2901 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
2902 | struct list_head *next = links->next; | |
dfc5606d | 2903 | |
35938150 | 2904 | /* Existing snapshot not in the new snap context */ |
dfc5606d | 2905 | |
0d7dbfce | 2906 | if (rbd_dev->spec->snap_id == snap->id) |
daba5fdb | 2907 | rbd_dev->exists = false; |
41f38c2b | 2908 | rbd_remove_snap_dev(snap); |
9fcbb800 | 2909 | dout("%ssnap id %llu has been removed\n", |
0d7dbfce AE |
2910 | rbd_dev->spec->snap_id == snap->id ? |
2911 | "mapped " : "", | |
9fcbb800 | 2912 | (unsigned long long) snap->id); |
35938150 AE |
2913 | |
2914 | /* Done with this list entry; advance */ | |
2915 | ||
2916 | links = next; | |
dfc5606d YS |
2917 | continue; |
2918 | } | |
35938150 | 2919 | |
b8b1e2db AE |
2920 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
2921 | &snap_size, &snap_features); | |
cd892126 AE |
2922 | if (IS_ERR(snap_name)) |
2923 | return PTR_ERR(snap_name); | |
2924 | ||
9fcbb800 AE |
2925 | dout("entry %u: snap_id = %llu\n", (unsigned int) snap_count, |
2926 | (unsigned long long) snap_id); | |
35938150 AE |
2927 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
2928 | struct rbd_snap *new_snap; | |
2929 | ||
2930 | /* We haven't seen this snapshot before */ | |
2931 | ||
c8d18425 | 2932 | new_snap = __rbd_add_snap_dev(rbd_dev, snap_name, |
cd892126 | 2933 | snap_id, snap_size, snap_features); |
9fcbb800 AE |
2934 | if (IS_ERR(new_snap)) { |
2935 | int err = PTR_ERR(new_snap); | |
2936 | ||
2937 | dout(" failed to add dev, error %d\n", err); | |
2938 | ||
2939 | return err; | |
2940 | } | |
35938150 AE |
2941 | |
2942 | /* New goes before existing, or at end of list */ | |
2943 | ||
9fcbb800 | 2944 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
2945 | if (snap) |
2946 | list_add_tail(&new_snap->node, &snap->node); | |
2947 | else | |
523f3258 | 2948 | list_add_tail(&new_snap->node, head); |
35938150 AE |
2949 | } else { |
2950 | /* Already have this one */ | |
2951 | ||
9fcbb800 AE |
2952 | dout(" already present\n"); |
2953 | ||
cd892126 | 2954 | rbd_assert(snap->size == snap_size); |
aafb230e | 2955 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 2956 | rbd_assert(snap->features == snap_features); |
35938150 AE |
2957 | |
2958 | /* Done with this list entry; advance */ | |
2959 | ||
2960 | links = links->next; | |
dfc5606d | 2961 | } |
35938150 AE |
2962 | |
2963 | /* Advance to the next entry in the snapshot context */ | |
2964 | ||
2965 | index++; | |
dfc5606d | 2966 | } |
9fcbb800 | 2967 | dout("%s: done\n", __func__); |
dfc5606d YS |
2968 | |
2969 | return 0; | |
2970 | } | |
2971 | ||
304f6808 AE |
2972 | /* |
2973 | * Scan the list of snapshots and register the devices for any that | |
2974 | * have not already been registered. | |
2975 | */ | |
2976 | static int rbd_dev_snaps_register(struct rbd_device *rbd_dev) | |
2977 | { | |
2978 | struct rbd_snap *snap; | |
2979 | int ret = 0; | |
2980 | ||
2981 | dout("%s called\n", __func__); | |
86ff77bb AE |
2982 | if (WARN_ON(!device_is_registered(&rbd_dev->dev))) |
2983 | return -EIO; | |
304f6808 AE |
2984 | |
2985 | list_for_each_entry(snap, &rbd_dev->snaps, node) { | |
2986 | if (!rbd_snap_registered(snap)) { | |
2987 | ret = rbd_register_snap_dev(snap, &rbd_dev->dev); | |
2988 | if (ret < 0) | |
2989 | break; | |
2990 | } | |
2991 | } | |
2992 | dout("%s: returning %d\n", __func__, ret); | |
2993 | ||
2994 | return ret; | |
2995 | } | |
2996 | ||
dfc5606d YS |
2997 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
2998 | { | |
dfc5606d | 2999 | struct device *dev; |
cd789ab9 | 3000 | int ret; |
dfc5606d YS |
3001 | |
3002 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 3003 | |
cd789ab9 | 3004 | dev = &rbd_dev->dev; |
dfc5606d YS |
3005 | dev->bus = &rbd_bus_type; |
3006 | dev->type = &rbd_device_type; | |
3007 | dev->parent = &rbd_root_dev; | |
3008 | dev->release = rbd_dev_release; | |
de71a297 | 3009 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 3010 | ret = device_register(dev); |
dfc5606d | 3011 | |
dfc5606d | 3012 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 3013 | |
dfc5606d | 3014 | return ret; |
602adf40 YS |
3015 | } |
3016 | ||
dfc5606d YS |
3017 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
3018 | { | |
3019 | device_unregister(&rbd_dev->dev); | |
3020 | } | |
3021 | ||
59c2be1e YS |
3022 | static int rbd_init_watch_dev(struct rbd_device *rbd_dev) |
3023 | { | |
3024 | int ret, rc; | |
3025 | ||
3026 | do { | |
0e6f322d | 3027 | ret = rbd_req_sync_watch(rbd_dev); |
59c2be1e | 3028 | if (ret == -ERANGE) { |
117973fb | 3029 | rc = rbd_dev_refresh(rbd_dev, NULL); |
59c2be1e YS |
3030 | if (rc < 0) |
3031 | return rc; | |
3032 | } | |
3033 | } while (ret == -ERANGE); | |
3034 | ||
3035 | return ret; | |
3036 | } | |
3037 | ||
e2839308 | 3038 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
3039 | |
3040 | /* | |
499afd5b AE |
3041 | * Get a unique rbd identifier for the given new rbd_dev, and add |
3042 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 3043 | */ |
e2839308 | 3044 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 3045 | { |
e2839308 | 3046 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
3047 | |
3048 | spin_lock(&rbd_dev_list_lock); | |
3049 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
3050 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
3051 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
3052 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 3053 | } |
b7f23c36 | 3054 | |
1ddbe94e | 3055 | /* |
499afd5b AE |
3056 | * Remove an rbd_dev from the global list, and record that its |
3057 | * identifier is no longer in use. | |
1ddbe94e | 3058 | */ |
e2839308 | 3059 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 3060 | { |
d184f6bf | 3061 | struct list_head *tmp; |
de71a297 | 3062 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
3063 | int max_id; |
3064 | ||
aafb230e | 3065 | rbd_assert(rbd_id > 0); |
499afd5b | 3066 | |
e2839308 AE |
3067 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
3068 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
3069 | spin_lock(&rbd_dev_list_lock); |
3070 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
3071 | |
3072 | /* | |
3073 | * If the id being "put" is not the current maximum, there | |
3074 | * is nothing special we need to do. | |
3075 | */ | |
e2839308 | 3076 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
3077 | spin_unlock(&rbd_dev_list_lock); |
3078 | return; | |
3079 | } | |
3080 | ||
3081 | /* | |
3082 | * We need to update the current maximum id. Search the | |
3083 | * list to find out what it is. We're more likely to find | |
3084 | * the maximum at the end, so search the list backward. | |
3085 | */ | |
3086 | max_id = 0; | |
3087 | list_for_each_prev(tmp, &rbd_dev_list) { | |
3088 | struct rbd_device *rbd_dev; | |
3089 | ||
3090 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
3091 | if (rbd_dev->dev_id > max_id) |
3092 | max_id = rbd_dev->dev_id; | |
d184f6bf | 3093 | } |
499afd5b | 3094 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 3095 | |
1ddbe94e | 3096 | /* |
e2839308 | 3097 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
3098 | * which case it now accurately reflects the new maximum. |
3099 | * Be careful not to overwrite the maximum value in that | |
3100 | * case. | |
1ddbe94e | 3101 | */ |
e2839308 AE |
3102 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
3103 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
3104 | } |
3105 | ||
e28fff26 AE |
3106 | /* |
3107 | * Skips over white space at *buf, and updates *buf to point to the | |
3108 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
3109 | * the token (string of non-white space characters) found. Note |
3110 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
3111 | */ |
3112 | static inline size_t next_token(const char **buf) | |
3113 | { | |
3114 | /* | |
3115 | * These are the characters that produce nonzero for | |
3116 | * isspace() in the "C" and "POSIX" locales. | |
3117 | */ | |
3118 | const char *spaces = " \f\n\r\t\v"; | |
3119 | ||
3120 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
3121 | ||
3122 | return strcspn(*buf, spaces); /* Return token length */ | |
3123 | } | |
3124 | ||
3125 | /* | |
3126 | * Finds the next token in *buf, and if the provided token buffer is | |
3127 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
3128 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
3129 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
3130 | * |
3131 | * Returns the length of the token found (not including the '\0'). | |
3132 | * Return value will be 0 if no token is found, and it will be >= | |
3133 | * token_size if the token would not fit. | |
3134 | * | |
593a9e7b | 3135 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
3136 | * found token. Note that this occurs even if the token buffer is |
3137 | * too small to hold it. | |
3138 | */ | |
3139 | static inline size_t copy_token(const char **buf, | |
3140 | char *token, | |
3141 | size_t token_size) | |
3142 | { | |
3143 | size_t len; | |
3144 | ||
3145 | len = next_token(buf); | |
3146 | if (len < token_size) { | |
3147 | memcpy(token, *buf, len); | |
3148 | *(token + len) = '\0'; | |
3149 | } | |
3150 | *buf += len; | |
3151 | ||
3152 | return len; | |
3153 | } | |
3154 | ||
ea3352f4 AE |
3155 | /* |
3156 | * Finds the next token in *buf, dynamically allocates a buffer big | |
3157 | * enough to hold a copy of it, and copies the token into the new | |
3158 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
3159 | * that a duplicate buffer is created even for a zero-length token. | |
3160 | * | |
3161 | * Returns a pointer to the newly-allocated duplicate, or a null | |
3162 | * pointer if memory for the duplicate was not available. If | |
3163 | * the lenp argument is a non-null pointer, the length of the token | |
3164 | * (not including the '\0') is returned in *lenp. | |
3165 | * | |
3166 | * If successful, the *buf pointer will be updated to point beyond | |
3167 | * the end of the found token. | |
3168 | * | |
3169 | * Note: uses GFP_KERNEL for allocation. | |
3170 | */ | |
3171 | static inline char *dup_token(const char **buf, size_t *lenp) | |
3172 | { | |
3173 | char *dup; | |
3174 | size_t len; | |
3175 | ||
3176 | len = next_token(buf); | |
4caf35f9 | 3177 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
3178 | if (!dup) |
3179 | return NULL; | |
ea3352f4 AE |
3180 | *(dup + len) = '\0'; |
3181 | *buf += len; | |
3182 | ||
3183 | if (lenp) | |
3184 | *lenp = len; | |
3185 | ||
3186 | return dup; | |
3187 | } | |
3188 | ||
a725f65e | 3189 | /* |
859c31df AE |
3190 | * Parse the options provided for an "rbd add" (i.e., rbd image |
3191 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
3192 | * and the data written is passed here via a NUL-terminated buffer. | |
3193 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 3194 | * |
859c31df AE |
3195 | * The information extracted from these options is recorded in |
3196 | * the other parameters which return dynamically-allocated | |
3197 | * structures: | |
3198 | * ceph_opts | |
3199 | * The address of a pointer that will refer to a ceph options | |
3200 | * structure. Caller must release the returned pointer using | |
3201 | * ceph_destroy_options() when it is no longer needed. | |
3202 | * rbd_opts | |
3203 | * Address of an rbd options pointer. Fully initialized by | |
3204 | * this function; caller must release with kfree(). | |
3205 | * spec | |
3206 | * Address of an rbd image specification pointer. Fully | |
3207 | * initialized by this function based on parsed options. | |
3208 | * Caller must release with rbd_spec_put(). | |
3209 | * | |
3210 | * The options passed take this form: | |
3211 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
3212 | * where: | |
3213 | * <mon_addrs> | |
3214 | * A comma-separated list of one or more monitor addresses. | |
3215 | * A monitor address is an ip address, optionally followed | |
3216 | * by a port number (separated by a colon). | |
3217 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
3218 | * <options> | |
3219 | * A comma-separated list of ceph and/or rbd options. | |
3220 | * <pool_name> | |
3221 | * The name of the rados pool containing the rbd image. | |
3222 | * <image_name> | |
3223 | * The name of the image in that pool to map. | |
3224 | * <snap_id> | |
3225 | * An optional snapshot id. If provided, the mapping will | |
3226 | * present data from the image at the time that snapshot was | |
3227 | * created. The image head is used if no snapshot id is | |
3228 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 3229 | */ |
859c31df | 3230 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 3231 | struct ceph_options **ceph_opts, |
859c31df AE |
3232 | struct rbd_options **opts, |
3233 | struct rbd_spec **rbd_spec) | |
e28fff26 | 3234 | { |
d22f76e7 | 3235 | size_t len; |
859c31df | 3236 | char *options; |
0ddebc0c AE |
3237 | const char *mon_addrs; |
3238 | size_t mon_addrs_size; | |
859c31df | 3239 | struct rbd_spec *spec = NULL; |
4e9afeba | 3240 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3241 | struct ceph_options *copts; |
dc79b113 | 3242 | int ret; |
e28fff26 AE |
3243 | |
3244 | /* The first four tokens are required */ | |
3245 | ||
7ef3214a AE |
3246 | len = next_token(&buf); |
3247 | if (!len) | |
dc79b113 | 3248 | return -EINVAL; /* Missing monitor address(es) */ |
0ddebc0c | 3249 | mon_addrs = buf; |
f28e565a | 3250 | mon_addrs_size = len + 1; |
7ef3214a | 3251 | buf += len; |
a725f65e | 3252 | |
dc79b113 | 3253 | ret = -EINVAL; |
f28e565a AE |
3254 | options = dup_token(&buf, NULL); |
3255 | if (!options) | |
dc79b113 | 3256 | return -ENOMEM; |
f28e565a AE |
3257 | if (!*options) |
3258 | goto out_err; /* Missing options */ | |
e28fff26 | 3259 | |
859c31df AE |
3260 | spec = rbd_spec_alloc(); |
3261 | if (!spec) | |
f28e565a | 3262 | goto out_mem; |
859c31df AE |
3263 | |
3264 | spec->pool_name = dup_token(&buf, NULL); | |
3265 | if (!spec->pool_name) | |
3266 | goto out_mem; | |
3267 | if (!*spec->pool_name) | |
f28e565a | 3268 | goto out_err; /* Missing pool name */ |
e28fff26 | 3269 | |
69e7a02f | 3270 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 3271 | if (!spec->image_name) |
f28e565a | 3272 | goto out_mem; |
859c31df | 3273 | if (!*spec->image_name) |
f28e565a | 3274 | goto out_err; /* Missing image name */ |
d4b125e9 | 3275 | |
f28e565a AE |
3276 | /* |
3277 | * Snapshot name is optional; default is to use "-" | |
3278 | * (indicating the head/no snapshot). | |
3279 | */ | |
3feeb894 | 3280 | len = next_token(&buf); |
820a5f3e | 3281 | if (!len) { |
3feeb894 AE |
3282 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
3283 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 3284 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 3285 | ret = -ENAMETOOLONG; |
f28e565a | 3286 | goto out_err; |
849b4260 | 3287 | } |
4caf35f9 | 3288 | spec->snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
859c31df | 3289 | if (!spec->snap_name) |
f28e565a | 3290 | goto out_mem; |
859c31df | 3291 | *(spec->snap_name + len) = '\0'; |
e5c35534 | 3292 | |
0ddebc0c | 3293 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 3294 | |
4e9afeba AE |
3295 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
3296 | if (!rbd_opts) | |
3297 | goto out_mem; | |
3298 | ||
3299 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 3300 | |
859c31df | 3301 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 3302 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 3303 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
3304 | if (IS_ERR(copts)) { |
3305 | ret = PTR_ERR(copts); | |
dc79b113 AE |
3306 | goto out_err; |
3307 | } | |
859c31df AE |
3308 | kfree(options); |
3309 | ||
3310 | *ceph_opts = copts; | |
4e9afeba | 3311 | *opts = rbd_opts; |
859c31df | 3312 | *rbd_spec = spec; |
0ddebc0c | 3313 | |
dc79b113 | 3314 | return 0; |
f28e565a | 3315 | out_mem: |
dc79b113 | 3316 | ret = -ENOMEM; |
d22f76e7 | 3317 | out_err: |
859c31df AE |
3318 | kfree(rbd_opts); |
3319 | rbd_spec_put(spec); | |
f28e565a | 3320 | kfree(options); |
d22f76e7 | 3321 | |
dc79b113 | 3322 | return ret; |
a725f65e AE |
3323 | } |
3324 | ||
589d30e0 AE |
3325 | /* |
3326 | * An rbd format 2 image has a unique identifier, distinct from the | |
3327 | * name given to it by the user. Internally, that identifier is | |
3328 | * what's used to specify the names of objects related to the image. | |
3329 | * | |
3330 | * A special "rbd id" object is used to map an rbd image name to its | |
3331 | * id. If that object doesn't exist, then there is no v2 rbd image | |
3332 | * with the supplied name. | |
3333 | * | |
3334 | * This function will record the given rbd_dev's image_id field if | |
3335 | * it can be determined, and in that case will return 0. If any | |
3336 | * errors occur a negative errno will be returned and the rbd_dev's | |
3337 | * image_id field will be unchanged (and should be NULL). | |
3338 | */ | |
3339 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
3340 | { | |
3341 | int ret; | |
3342 | size_t size; | |
3343 | char *object_name; | |
3344 | void *response; | |
3345 | void *p; | |
3346 | ||
2c0d0a10 AE |
3347 | /* |
3348 | * When probing a parent image, the image id is already | |
3349 | * known (and the image name likely is not). There's no | |
3350 | * need to fetch the image id again in this case. | |
3351 | */ | |
3352 | if (rbd_dev->spec->image_id) | |
3353 | return 0; | |
3354 | ||
589d30e0 AE |
3355 | /* |
3356 | * First, see if the format 2 image id file exists, and if | |
3357 | * so, get the image's persistent id from it. | |
3358 | */ | |
69e7a02f | 3359 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
3360 | object_name = kmalloc(size, GFP_NOIO); |
3361 | if (!object_name) | |
3362 | return -ENOMEM; | |
0d7dbfce | 3363 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
3364 | dout("rbd id object name is %s\n", object_name); |
3365 | ||
3366 | /* Response will be an encoded string, which includes a length */ | |
3367 | ||
3368 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
3369 | response = kzalloc(size, GFP_NOIO); | |
3370 | if (!response) { | |
3371 | ret = -ENOMEM; | |
3372 | goto out; | |
3373 | } | |
3374 | ||
3375 | ret = rbd_req_sync_exec(rbd_dev, object_name, | |
3376 | "rbd", "get_id", | |
3377 | NULL, 0, | |
3378 | response, RBD_IMAGE_ID_LEN_MAX, | |
3379 | CEPH_OSD_FLAG_READ, NULL); | |
3380 | dout("%s: rbd_req_sync_exec returned %d\n", __func__, ret); | |
3381 | if (ret < 0) | |
3382 | goto out; | |
a0ea3a40 | 3383 | ret = 0; /* rbd_req_sync_exec() can return positive */ |
589d30e0 AE |
3384 | |
3385 | p = response; | |
0d7dbfce | 3386 | rbd_dev->spec->image_id = ceph_extract_encoded_string(&p, |
589d30e0 | 3387 | p + RBD_IMAGE_ID_LEN_MAX, |
979ed480 | 3388 | NULL, GFP_NOIO); |
0d7dbfce AE |
3389 | if (IS_ERR(rbd_dev->spec->image_id)) { |
3390 | ret = PTR_ERR(rbd_dev->spec->image_id); | |
3391 | rbd_dev->spec->image_id = NULL; | |
589d30e0 | 3392 | } else { |
0d7dbfce | 3393 | dout("image_id is %s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3394 | } |
3395 | out: | |
3396 | kfree(response); | |
3397 | kfree(object_name); | |
3398 | ||
3399 | return ret; | |
3400 | } | |
3401 | ||
a30b71b9 AE |
3402 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
3403 | { | |
3404 | int ret; | |
3405 | size_t size; | |
3406 | ||
3407 | /* Version 1 images have no id; empty string is used */ | |
3408 | ||
0d7dbfce AE |
3409 | rbd_dev->spec->image_id = kstrdup("", GFP_KERNEL); |
3410 | if (!rbd_dev->spec->image_id) | |
a30b71b9 | 3411 | return -ENOMEM; |
a30b71b9 AE |
3412 | |
3413 | /* Record the header object name for this rbd image. */ | |
3414 | ||
69e7a02f | 3415 | size = strlen(rbd_dev->spec->image_name) + sizeof (RBD_SUFFIX); |
a30b71b9 AE |
3416 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3417 | if (!rbd_dev->header_name) { | |
3418 | ret = -ENOMEM; | |
3419 | goto out_err; | |
3420 | } | |
0d7dbfce AE |
3421 | sprintf(rbd_dev->header_name, "%s%s", |
3422 | rbd_dev->spec->image_name, RBD_SUFFIX); | |
a30b71b9 AE |
3423 | |
3424 | /* Populate rbd image metadata */ | |
3425 | ||
3426 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
3427 | if (ret < 0) | |
3428 | goto out_err; | |
86b00e0d AE |
3429 | |
3430 | /* Version 1 images have no parent (no layering) */ | |
3431 | ||
3432 | rbd_dev->parent_spec = NULL; | |
3433 | rbd_dev->parent_overlap = 0; | |
3434 | ||
a30b71b9 AE |
3435 | rbd_dev->image_format = 1; |
3436 | ||
3437 | dout("discovered version 1 image, header name is %s\n", | |
3438 | rbd_dev->header_name); | |
3439 | ||
3440 | return 0; | |
3441 | ||
3442 | out_err: | |
3443 | kfree(rbd_dev->header_name); | |
3444 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
3445 | kfree(rbd_dev->spec->image_id); |
3446 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
3447 | |
3448 | return ret; | |
3449 | } | |
3450 | ||
3451 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
3452 | { | |
3453 | size_t size; | |
9d475de5 | 3454 | int ret; |
6e14b1a6 | 3455 | u64 ver = 0; |
a30b71b9 AE |
3456 | |
3457 | /* | |
3458 | * Image id was filled in by the caller. Record the header | |
3459 | * object name for this rbd image. | |
3460 | */ | |
979ed480 | 3461 | size = sizeof (RBD_HEADER_PREFIX) + strlen(rbd_dev->spec->image_id); |
a30b71b9 AE |
3462 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); |
3463 | if (!rbd_dev->header_name) | |
3464 | return -ENOMEM; | |
3465 | sprintf(rbd_dev->header_name, "%s%s", | |
0d7dbfce | 3466 | RBD_HEADER_PREFIX, rbd_dev->spec->image_id); |
9d475de5 AE |
3467 | |
3468 | /* Get the size and object order for the image */ | |
3469 | ||
3470 | ret = rbd_dev_v2_image_size(rbd_dev); | |
1e130199 AE |
3471 | if (ret < 0) |
3472 | goto out_err; | |
3473 | ||
3474 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
3475 | ||
3476 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
b1b5402a AE |
3477 | if (ret < 0) |
3478 | goto out_err; | |
3479 | ||
d889140c | 3480 | /* Get the and check features for the image */ |
b1b5402a AE |
3481 | |
3482 | ret = rbd_dev_v2_features(rbd_dev); | |
9d475de5 AE |
3483 | if (ret < 0) |
3484 | goto out_err; | |
35d489f9 | 3485 | |
86b00e0d AE |
3486 | /* If the image supports layering, get the parent info */ |
3487 | ||
3488 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
3489 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3490 | if (ret < 0) | |
3491 | goto out_err; | |
3492 | } | |
3493 | ||
6e14b1a6 AE |
3494 | /* crypto and compression type aren't (yet) supported for v2 images */ |
3495 | ||
3496 | rbd_dev->header.crypt_type = 0; | |
3497 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 3498 | |
6e14b1a6 AE |
3499 | /* Get the snapshot context, plus the header version */ |
3500 | ||
3501 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
3502 | if (ret) |
3503 | goto out_err; | |
6e14b1a6 AE |
3504 | rbd_dev->header.obj_version = ver; |
3505 | ||
a30b71b9 AE |
3506 | rbd_dev->image_format = 2; |
3507 | ||
3508 | dout("discovered version 2 image, header name is %s\n", | |
3509 | rbd_dev->header_name); | |
3510 | ||
35152979 | 3511 | return 0; |
9d475de5 | 3512 | out_err: |
86b00e0d AE |
3513 | rbd_dev->parent_overlap = 0; |
3514 | rbd_spec_put(rbd_dev->parent_spec); | |
3515 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
3516 | kfree(rbd_dev->header_name); |
3517 | rbd_dev->header_name = NULL; | |
1e130199 AE |
3518 | kfree(rbd_dev->header.object_prefix); |
3519 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
3520 | |
3521 | return ret; | |
a30b71b9 AE |
3522 | } |
3523 | ||
83a06263 AE |
3524 | static int rbd_dev_probe_finish(struct rbd_device *rbd_dev) |
3525 | { | |
3526 | int ret; | |
3527 | ||
3528 | /* no need to lock here, as rbd_dev is not registered yet */ | |
3529 | ret = rbd_dev_snaps_update(rbd_dev); | |
3530 | if (ret) | |
3531 | return ret; | |
3532 | ||
9e15b77d AE |
3533 | ret = rbd_dev_probe_update_spec(rbd_dev); |
3534 | if (ret) | |
3535 | goto err_out_snaps; | |
3536 | ||
83a06263 AE |
3537 | ret = rbd_dev_set_mapping(rbd_dev); |
3538 | if (ret) | |
3539 | goto err_out_snaps; | |
3540 | ||
3541 | /* generate unique id: find highest unique id, add one */ | |
3542 | rbd_dev_id_get(rbd_dev); | |
3543 | ||
3544 | /* Fill in the device name, now that we have its id. */ | |
3545 | BUILD_BUG_ON(DEV_NAME_LEN | |
3546 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
3547 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
3548 | ||
3549 | /* Get our block major device number. */ | |
3550 | ||
3551 | ret = register_blkdev(0, rbd_dev->name); | |
3552 | if (ret < 0) | |
3553 | goto err_out_id; | |
3554 | rbd_dev->major = ret; | |
3555 | ||
3556 | /* Set up the blkdev mapping. */ | |
3557 | ||
3558 | ret = rbd_init_disk(rbd_dev); | |
3559 | if (ret) | |
3560 | goto err_out_blkdev; | |
3561 | ||
3562 | ret = rbd_bus_add_dev(rbd_dev); | |
3563 | if (ret) | |
3564 | goto err_out_disk; | |
3565 | ||
3566 | /* | |
3567 | * At this point cleanup in the event of an error is the job | |
3568 | * of the sysfs code (initiated by rbd_bus_del_dev()). | |
3569 | */ | |
3570 | down_write(&rbd_dev->header_rwsem); | |
3571 | ret = rbd_dev_snaps_register(rbd_dev); | |
3572 | up_write(&rbd_dev->header_rwsem); | |
3573 | if (ret) | |
3574 | goto err_out_bus; | |
3575 | ||
3576 | ret = rbd_init_watch_dev(rbd_dev); | |
3577 | if (ret) | |
3578 | goto err_out_bus; | |
3579 | ||
3580 | /* Everything's ready. Announce the disk to the world. */ | |
3581 | ||
3582 | add_disk(rbd_dev->disk); | |
3583 | ||
3584 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
3585 | (unsigned long long) rbd_dev->mapping.size); | |
3586 | ||
3587 | return ret; | |
3588 | err_out_bus: | |
3589 | /* this will also clean up rest of rbd_dev stuff */ | |
3590 | ||
3591 | rbd_bus_del_dev(rbd_dev); | |
3592 | ||
3593 | return ret; | |
3594 | err_out_disk: | |
3595 | rbd_free_disk(rbd_dev); | |
3596 | err_out_blkdev: | |
3597 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
3598 | err_out_id: | |
3599 | rbd_dev_id_put(rbd_dev); | |
3600 | err_out_snaps: | |
3601 | rbd_remove_all_snaps(rbd_dev); | |
3602 | ||
3603 | return ret; | |
3604 | } | |
3605 | ||
a30b71b9 AE |
3606 | /* |
3607 | * Probe for the existence of the header object for the given rbd | |
3608 | * device. For format 2 images this includes determining the image | |
3609 | * id. | |
3610 | */ | |
3611 | static int rbd_dev_probe(struct rbd_device *rbd_dev) | |
3612 | { | |
3613 | int ret; | |
3614 | ||
3615 | /* | |
3616 | * Get the id from the image id object. If it's not a | |
3617 | * format 2 image, we'll get ENOENT back, and we'll assume | |
3618 | * it's a format 1 image. | |
3619 | */ | |
3620 | ret = rbd_dev_image_id(rbd_dev); | |
3621 | if (ret) | |
3622 | ret = rbd_dev_v1_probe(rbd_dev); | |
3623 | else | |
3624 | ret = rbd_dev_v2_probe(rbd_dev); | |
83a06263 | 3625 | if (ret) { |
a30b71b9 AE |
3626 | dout("probe failed, returning %d\n", ret); |
3627 | ||
83a06263 AE |
3628 | return ret; |
3629 | } | |
3630 | ||
3631 | ret = rbd_dev_probe_finish(rbd_dev); | |
3632 | if (ret) | |
3633 | rbd_header_free(&rbd_dev->header); | |
3634 | ||
a30b71b9 AE |
3635 | return ret; |
3636 | } | |
3637 | ||
59c2be1e YS |
3638 | static ssize_t rbd_add(struct bus_type *bus, |
3639 | const char *buf, | |
3640 | size_t count) | |
602adf40 | 3641 | { |
cb8627c7 | 3642 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 3643 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 3644 | struct rbd_options *rbd_opts = NULL; |
859c31df | 3645 | struct rbd_spec *spec = NULL; |
9d3997fd | 3646 | struct rbd_client *rbdc; |
27cc2594 AE |
3647 | struct ceph_osd_client *osdc; |
3648 | int rc = -ENOMEM; | |
602adf40 YS |
3649 | |
3650 | if (!try_module_get(THIS_MODULE)) | |
3651 | return -ENODEV; | |
3652 | ||
602adf40 | 3653 | /* parse add command */ |
859c31df | 3654 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 3655 | if (rc < 0) |
bd4ba655 | 3656 | goto err_out_module; |
78cea76e | 3657 | |
9d3997fd AE |
3658 | rbdc = rbd_get_client(ceph_opts); |
3659 | if (IS_ERR(rbdc)) { | |
3660 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 3661 | goto err_out_args; |
9d3997fd | 3662 | } |
c53d5893 | 3663 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 3664 | |
602adf40 | 3665 | /* pick the pool */ |
9d3997fd | 3666 | osdc = &rbdc->client->osdc; |
859c31df | 3667 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
3668 | if (rc < 0) |
3669 | goto err_out_client; | |
859c31df AE |
3670 | spec->pool_id = (u64) rc; |
3671 | ||
c53d5893 | 3672 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
3673 | if (!rbd_dev) |
3674 | goto err_out_client; | |
c53d5893 AE |
3675 | rbdc = NULL; /* rbd_dev now owns this */ |
3676 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 3677 | |
bd4ba655 | 3678 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
3679 | kfree(rbd_opts); |
3680 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 3681 | |
a30b71b9 AE |
3682 | rc = rbd_dev_probe(rbd_dev); |
3683 | if (rc < 0) | |
c53d5893 | 3684 | goto err_out_rbd_dev; |
05fd6f6f | 3685 | |
602adf40 | 3686 | return count; |
c53d5893 AE |
3687 | err_out_rbd_dev: |
3688 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 3689 | err_out_client: |
9d3997fd | 3690 | rbd_put_client(rbdc); |
0ddebc0c | 3691 | err_out_args: |
78cea76e AE |
3692 | if (ceph_opts) |
3693 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 3694 | kfree(rbd_opts); |
859c31df | 3695 | rbd_spec_put(spec); |
bd4ba655 AE |
3696 | err_out_module: |
3697 | module_put(THIS_MODULE); | |
27cc2594 | 3698 | |
602adf40 | 3699 | dout("Error adding device %s\n", buf); |
27cc2594 AE |
3700 | |
3701 | return (ssize_t) rc; | |
602adf40 YS |
3702 | } |
3703 | ||
de71a297 | 3704 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
3705 | { |
3706 | struct list_head *tmp; | |
3707 | struct rbd_device *rbd_dev; | |
3708 | ||
e124a82f | 3709 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
3710 | list_for_each(tmp, &rbd_dev_list) { |
3711 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 3712 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 3713 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 3714 | return rbd_dev; |
e124a82f | 3715 | } |
602adf40 | 3716 | } |
e124a82f | 3717 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
3718 | return NULL; |
3719 | } | |
3720 | ||
dfc5606d | 3721 | static void rbd_dev_release(struct device *dev) |
602adf40 | 3722 | { |
593a9e7b | 3723 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3724 | |
1dbb4399 AE |
3725 | if (rbd_dev->watch_request) { |
3726 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
3727 | ||
3728 | ceph_osdc_unregister_linger_request(&client->osdc, | |
59c2be1e | 3729 | rbd_dev->watch_request); |
1dbb4399 | 3730 | } |
59c2be1e | 3731 | if (rbd_dev->watch_event) |
070c633f | 3732 | rbd_req_sync_unwatch(rbd_dev); |
59c2be1e | 3733 | |
602adf40 YS |
3734 | |
3735 | /* clean up and free blkdev */ | |
3736 | rbd_free_disk(rbd_dev); | |
3737 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
32eec68d | 3738 | |
2ac4e75d AE |
3739 | /* release allocated disk header fields */ |
3740 | rbd_header_free(&rbd_dev->header); | |
3741 | ||
32eec68d | 3742 | /* done with the id, and with the rbd_dev */ |
e2839308 | 3743 | rbd_dev_id_put(rbd_dev); |
c53d5893 AE |
3744 | rbd_assert(rbd_dev->rbd_client != NULL); |
3745 | rbd_dev_destroy(rbd_dev); | |
602adf40 YS |
3746 | |
3747 | /* release module ref */ | |
3748 | module_put(THIS_MODULE); | |
602adf40 YS |
3749 | } |
3750 | ||
dfc5606d YS |
3751 | static ssize_t rbd_remove(struct bus_type *bus, |
3752 | const char *buf, | |
3753 | size_t count) | |
602adf40 YS |
3754 | { |
3755 | struct rbd_device *rbd_dev = NULL; | |
3756 | int target_id, rc; | |
3757 | unsigned long ul; | |
3758 | int ret = count; | |
3759 | ||
3760 | rc = strict_strtoul(buf, 10, &ul); | |
3761 | if (rc) | |
3762 | return rc; | |
3763 | ||
3764 | /* convert to int; abort if we lost anything in the conversion */ | |
3765 | target_id = (int) ul; | |
3766 | if (target_id != ul) | |
3767 | return -EINVAL; | |
3768 | ||
3769 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
3770 | ||
3771 | rbd_dev = __rbd_get_dev(target_id); | |
3772 | if (!rbd_dev) { | |
3773 | ret = -ENOENT; | |
3774 | goto done; | |
42382b70 AE |
3775 | } |
3776 | ||
3777 | if (rbd_dev->open_count) { | |
3778 | ret = -EBUSY; | |
3779 | goto done; | |
602adf40 YS |
3780 | } |
3781 | ||
41f38c2b | 3782 | rbd_remove_all_snaps(rbd_dev); |
dfc5606d | 3783 | rbd_bus_del_dev(rbd_dev); |
602adf40 YS |
3784 | |
3785 | done: | |
3786 | mutex_unlock(&ctl_mutex); | |
aafb230e | 3787 | |
602adf40 YS |
3788 | return ret; |
3789 | } | |
3790 | ||
602adf40 YS |
3791 | /* |
3792 | * create control files in sysfs | |
dfc5606d | 3793 | * /sys/bus/rbd/... |
602adf40 YS |
3794 | */ |
3795 | static int rbd_sysfs_init(void) | |
3796 | { | |
dfc5606d | 3797 | int ret; |
602adf40 | 3798 | |
fed4c143 | 3799 | ret = device_register(&rbd_root_dev); |
21079786 | 3800 | if (ret < 0) |
dfc5606d | 3801 | return ret; |
602adf40 | 3802 | |
fed4c143 AE |
3803 | ret = bus_register(&rbd_bus_type); |
3804 | if (ret < 0) | |
3805 | device_unregister(&rbd_root_dev); | |
602adf40 | 3806 | |
602adf40 YS |
3807 | return ret; |
3808 | } | |
3809 | ||
3810 | static void rbd_sysfs_cleanup(void) | |
3811 | { | |
dfc5606d | 3812 | bus_unregister(&rbd_bus_type); |
fed4c143 | 3813 | device_unregister(&rbd_root_dev); |
602adf40 YS |
3814 | } |
3815 | ||
3816 | int __init rbd_init(void) | |
3817 | { | |
3818 | int rc; | |
3819 | ||
3820 | rc = rbd_sysfs_init(); | |
3821 | if (rc) | |
3822 | return rc; | |
f0f8cef5 | 3823 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
3824 | return 0; |
3825 | } | |
3826 | ||
3827 | void __exit rbd_exit(void) | |
3828 | { | |
3829 | rbd_sysfs_cleanup(); | |
3830 | } | |
3831 | ||
3832 | module_init(rbd_init); | |
3833 | module_exit(rbd_exit); | |
3834 | ||
3835 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
3836 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
3837 | MODULE_DESCRIPTION("rados block device"); | |
3838 | ||
3839 | /* following authorship retained from original osdblk.c */ | |
3840 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
3841 | ||
3842 | MODULE_LICENSE("GPL"); |