]>
Commit | Line | Data |
---|---|---|
e2a58ee5 | 1 | |
602adf40 YS |
2 | /* |
3 | rbd.c -- Export ceph rados objects as a Linux block device | |
4 | ||
5 | ||
6 | based on drivers/block/osdblk.c: | |
7 | ||
8 | Copyright 2009 Red Hat, Inc. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; see the file COPYING. If not, write to | |
21 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
22 | ||
23 | ||
24 | ||
dfc5606d | 25 | For usage instructions, please refer to: |
602adf40 | 26 | |
dfc5606d | 27 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
28 | |
29 | */ | |
30 | ||
31 | #include <linux/ceph/libceph.h> | |
32 | #include <linux/ceph/osd_client.h> | |
33 | #include <linux/ceph/mon_client.h> | |
ed95b21a | 34 | #include <linux/ceph/cls_lock_client.h> |
602adf40 | 35 | #include <linux/ceph/decode.h> |
59c2be1e | 36 | #include <linux/parser.h> |
30d1cff8 | 37 | #include <linux/bsearch.h> |
602adf40 YS |
38 | |
39 | #include <linux/kernel.h> | |
40 | #include <linux/device.h> | |
41 | #include <linux/module.h> | |
7ad18afa | 42 | #include <linux/blk-mq.h> |
602adf40 YS |
43 | #include <linux/fs.h> |
44 | #include <linux/blkdev.h> | |
1c2a9dfe | 45 | #include <linux/slab.h> |
f8a22fc2 | 46 | #include <linux/idr.h> |
bc1ecc65 | 47 | #include <linux/workqueue.h> |
602adf40 YS |
48 | |
49 | #include "rbd_types.h" | |
50 | ||
aafb230e AE |
51 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
52 | ||
593a9e7b AE |
53 | /* |
54 | * The basic unit of block I/O is a sector. It is interpreted in a | |
55 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
56 | * universally 512 bytes. These symbols are just slightly more | |
57 | * meaningful than the bare numbers they represent. | |
58 | */ | |
59 | #define SECTOR_SHIFT 9 | |
60 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
61 | ||
a2acd00e AE |
62 | /* |
63 | * Increment the given counter and return its updated value. | |
64 | * If the counter is already 0 it will not be incremented. | |
65 | * If the counter is already at its maximum value returns | |
66 | * -EINVAL without updating it. | |
67 | */ | |
68 | static int atomic_inc_return_safe(atomic_t *v) | |
69 | { | |
70 | unsigned int counter; | |
71 | ||
72 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
73 | if (counter <= (unsigned int)INT_MAX) | |
74 | return (int)counter; | |
75 | ||
76 | atomic_dec(v); | |
77 | ||
78 | return -EINVAL; | |
79 | } | |
80 | ||
81 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
82 | static int atomic_dec_return_safe(atomic_t *v) | |
83 | { | |
84 | int counter; | |
85 | ||
86 | counter = atomic_dec_return(v); | |
87 | if (counter >= 0) | |
88 | return counter; | |
89 | ||
90 | atomic_inc(v); | |
91 | ||
92 | return -EINVAL; | |
93 | } | |
94 | ||
f0f8cef5 | 95 | #define RBD_DRV_NAME "rbd" |
602adf40 | 96 | |
7e513d43 ID |
97 | #define RBD_MINORS_PER_MAJOR 256 |
98 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 99 | |
6d69bb53 ID |
100 | #define RBD_MAX_PARENT_CHAIN_LEN 16 |
101 | ||
d4b125e9 AE |
102 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
103 | #define RBD_MAX_SNAP_NAME_LEN \ | |
104 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
105 | ||
35d489f9 | 106 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
107 | |
108 | #define RBD_SNAP_HEAD_NAME "-" | |
109 | ||
9682fc6d AE |
110 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
111 | ||
9e15b77d AE |
112 | /* This allows a single page to hold an image name sent by OSD */ |
113 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 114 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 115 | |
1e130199 | 116 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 117 | |
ed95b21a | 118 | #define RBD_NOTIFY_TIMEOUT 5 /* seconds */ |
99d16943 ID |
119 | #define RBD_RETRY_DELAY msecs_to_jiffies(1000) |
120 | ||
d889140c AE |
121 | /* Feature bits */ |
122 | ||
8767b293 ID |
123 | #define RBD_FEATURE_LAYERING (1ULL<<0) |
124 | #define RBD_FEATURE_STRIPINGV2 (1ULL<<1) | |
125 | #define RBD_FEATURE_EXCLUSIVE_LOCK (1ULL<<2) | |
126 | #define RBD_FEATURE_DATA_POOL (1ULL<<7) | |
127 | ||
ed95b21a ID |
128 | #define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \ |
129 | RBD_FEATURE_STRIPINGV2 | \ | |
7e97332e ID |
130 | RBD_FEATURE_EXCLUSIVE_LOCK | \ |
131 | RBD_FEATURE_DATA_POOL) | |
d889140c AE |
132 | |
133 | /* Features supported by this (client software) implementation. */ | |
134 | ||
770eba6e | 135 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 136 | |
81a89793 AE |
137 | /* |
138 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
139 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
81a89793 | 140 | */ |
602adf40 YS |
141 | #define DEV_NAME_LEN 32 |
142 | ||
143 | /* | |
144 | * block device image metadata (in-memory version) | |
145 | */ | |
146 | struct rbd_image_header { | |
f35a4dee | 147 | /* These six fields never change for a given rbd image */ |
849b4260 | 148 | char *object_prefix; |
602adf40 | 149 | __u8 obj_order; |
f35a4dee AE |
150 | u64 stripe_unit; |
151 | u64 stripe_count; | |
7e97332e | 152 | s64 data_pool_id; |
f35a4dee | 153 | u64 features; /* Might be changeable someday? */ |
602adf40 | 154 | |
f84344f3 AE |
155 | /* The remaining fields need to be updated occasionally */ |
156 | u64 image_size; | |
157 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
158 | char *snap_names; /* format 1 only */ |
159 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
160 | }; |
161 | ||
0d7dbfce AE |
162 | /* |
163 | * An rbd image specification. | |
164 | * | |
165 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
166 | * identify an image. Each rbd_dev structure includes a pointer to |
167 | * an rbd_spec structure that encapsulates this identity. | |
168 | * | |
169 | * Each of the id's in an rbd_spec has an associated name. For a | |
170 | * user-mapped image, the names are supplied and the id's associated | |
171 | * with them are looked up. For a layered image, a parent image is | |
172 | * defined by the tuple, and the names are looked up. | |
173 | * | |
174 | * An rbd_dev structure contains a parent_spec pointer which is | |
175 | * non-null if the image it represents is a child in a layered | |
176 | * image. This pointer will refer to the rbd_spec structure used | |
177 | * by the parent rbd_dev for its own identity (i.e., the structure | |
178 | * is shared between the parent and child). | |
179 | * | |
180 | * Since these structures are populated once, during the discovery | |
181 | * phase of image construction, they are effectively immutable so | |
182 | * we make no effort to synchronize access to them. | |
183 | * | |
184 | * Note that code herein does not assume the image name is known (it | |
185 | * could be a null pointer). | |
0d7dbfce AE |
186 | */ |
187 | struct rbd_spec { | |
188 | u64 pool_id; | |
ecb4dc22 | 189 | const char *pool_name; |
0d7dbfce | 190 | |
ecb4dc22 AE |
191 | const char *image_id; |
192 | const char *image_name; | |
0d7dbfce AE |
193 | |
194 | u64 snap_id; | |
ecb4dc22 | 195 | const char *snap_name; |
0d7dbfce AE |
196 | |
197 | struct kref kref; | |
198 | }; | |
199 | ||
602adf40 | 200 | /* |
f0f8cef5 | 201 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
202 | */ |
203 | struct rbd_client { | |
204 | struct ceph_client *client; | |
205 | struct kref kref; | |
206 | struct list_head node; | |
207 | }; | |
208 | ||
bf0d5f50 AE |
209 | struct rbd_img_request; |
210 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
211 | ||
212 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
213 | ||
214 | struct rbd_obj_request; | |
215 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
216 | ||
9969ebc5 AE |
217 | enum obj_request_type { |
218 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
219 | }; | |
bf0d5f50 | 220 | |
6d2940c8 GZ |
221 | enum obj_operation_type { |
222 | OBJ_OP_WRITE, | |
223 | OBJ_OP_READ, | |
90e98c52 | 224 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
225 | }; |
226 | ||
926f9b3f AE |
227 | enum obj_req_flags { |
228 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 229 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
230 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
231 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
232 | }; |
233 | ||
bf0d5f50 | 234 | struct rbd_obj_request { |
a90bb0c1 | 235 | u64 object_no; |
bf0d5f50 AE |
236 | u64 offset; /* object start byte */ |
237 | u64 length; /* bytes from offset */ | |
926f9b3f | 238 | unsigned long flags; |
bf0d5f50 | 239 | |
c5b5ef6c AE |
240 | /* |
241 | * An object request associated with an image will have its | |
242 | * img_data flag set; a standalone object request will not. | |
243 | * | |
244 | * A standalone object request will have which == BAD_WHICH | |
245 | * and a null obj_request pointer. | |
246 | * | |
247 | * An object request initiated in support of a layered image | |
248 | * object (to check for its existence before a write) will | |
249 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
250 | * | |
251 | * Finally, an object request for rbd image data will have | |
252 | * which != BAD_WHICH, and will have a non-null img_request | |
253 | * pointer. The value of which will be in the range | |
254 | * 0..(img_request->obj_request_count-1). | |
255 | */ | |
256 | union { | |
257 | struct rbd_obj_request *obj_request; /* STAT op */ | |
258 | struct { | |
259 | struct rbd_img_request *img_request; | |
260 | u64 img_offset; | |
261 | /* links for img_request->obj_requests list */ | |
262 | struct list_head links; | |
263 | }; | |
264 | }; | |
bf0d5f50 AE |
265 | u32 which; /* posn image request list */ |
266 | ||
267 | enum obj_request_type type; | |
788e2df3 AE |
268 | union { |
269 | struct bio *bio_list; | |
270 | struct { | |
271 | struct page **pages; | |
272 | u32 page_count; | |
273 | }; | |
274 | }; | |
0eefd470 | 275 | struct page **copyup_pages; |
ebda6408 | 276 | u32 copyup_page_count; |
bf0d5f50 AE |
277 | |
278 | struct ceph_osd_request *osd_req; | |
279 | ||
280 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 281 | int result; |
bf0d5f50 AE |
282 | |
283 | rbd_obj_callback_t callback; | |
788e2df3 | 284 | struct completion completion; |
bf0d5f50 AE |
285 | |
286 | struct kref kref; | |
287 | }; | |
288 | ||
0c425248 | 289 | enum img_req_flags { |
9849e986 AE |
290 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
291 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 292 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 293 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
294 | }; |
295 | ||
bf0d5f50 | 296 | struct rbd_img_request { |
bf0d5f50 AE |
297 | struct rbd_device *rbd_dev; |
298 | u64 offset; /* starting image byte offset */ | |
299 | u64 length; /* byte count from offset */ | |
0c425248 | 300 | unsigned long flags; |
bf0d5f50 | 301 | union { |
9849e986 | 302 | u64 snap_id; /* for reads */ |
bf0d5f50 | 303 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
304 | }; |
305 | union { | |
306 | struct request *rq; /* block request */ | |
307 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 308 | }; |
3d7efd18 | 309 | struct page **copyup_pages; |
ebda6408 | 310 | u32 copyup_page_count; |
bf0d5f50 AE |
311 | spinlock_t completion_lock;/* protects next_completion */ |
312 | u32 next_completion; | |
313 | rbd_img_callback_t callback; | |
55f27e09 | 314 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 315 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
316 | |
317 | u32 obj_request_count; | |
318 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
319 | ||
320 | struct kref kref; | |
321 | }; | |
322 | ||
323 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 324 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 325 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 326 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 327 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 328 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 329 | |
99d16943 ID |
330 | enum rbd_watch_state { |
331 | RBD_WATCH_STATE_UNREGISTERED, | |
332 | RBD_WATCH_STATE_REGISTERED, | |
333 | RBD_WATCH_STATE_ERROR, | |
334 | }; | |
335 | ||
ed95b21a ID |
336 | enum rbd_lock_state { |
337 | RBD_LOCK_STATE_UNLOCKED, | |
338 | RBD_LOCK_STATE_LOCKED, | |
339 | RBD_LOCK_STATE_RELEASING, | |
340 | }; | |
341 | ||
342 | /* WatchNotify::ClientId */ | |
343 | struct rbd_client_id { | |
344 | u64 gid; | |
345 | u64 handle; | |
346 | }; | |
347 | ||
f84344f3 | 348 | struct rbd_mapping { |
99c1f08f | 349 | u64 size; |
34b13184 | 350 | u64 features; |
f84344f3 AE |
351 | bool read_only; |
352 | }; | |
353 | ||
602adf40 YS |
354 | /* |
355 | * a single device | |
356 | */ | |
357 | struct rbd_device { | |
de71a297 | 358 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
359 | |
360 | int major; /* blkdev assigned major */ | |
dd82fff1 | 361 | int minor; |
602adf40 | 362 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 363 | |
a30b71b9 | 364 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
365 | struct rbd_client *rbd_client; |
366 | ||
367 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
368 | ||
b82d167b | 369 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
370 | |
371 | struct rbd_image_header header; | |
b82d167b | 372 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 373 | struct rbd_spec *spec; |
d147543d | 374 | struct rbd_options *opts; |
0d6d1e9c | 375 | char *config_info; /* add{,_single_major} string */ |
602adf40 | 376 | |
c41d13a3 | 377 | struct ceph_object_id header_oid; |
922dab61 | 378 | struct ceph_object_locator header_oloc; |
971f839a | 379 | |
1643dfa4 | 380 | struct ceph_file_layout layout; /* used for all rbd requests */ |
0903e875 | 381 | |
99d16943 ID |
382 | struct mutex watch_mutex; |
383 | enum rbd_watch_state watch_state; | |
922dab61 | 384 | struct ceph_osd_linger_request *watch_handle; |
99d16943 ID |
385 | u64 watch_cookie; |
386 | struct delayed_work watch_dwork; | |
59c2be1e | 387 | |
ed95b21a ID |
388 | struct rw_semaphore lock_rwsem; |
389 | enum rbd_lock_state lock_state; | |
390 | struct rbd_client_id owner_cid; | |
391 | struct work_struct acquired_lock_work; | |
392 | struct work_struct released_lock_work; | |
393 | struct delayed_work lock_dwork; | |
394 | struct work_struct unlock_work; | |
395 | wait_queue_head_t lock_waitq; | |
396 | ||
1643dfa4 | 397 | struct workqueue_struct *task_wq; |
59c2be1e | 398 | |
86b00e0d AE |
399 | struct rbd_spec *parent_spec; |
400 | u64 parent_overlap; | |
a2acd00e | 401 | atomic_t parent_ref; |
2f82ee54 | 402 | struct rbd_device *parent; |
86b00e0d | 403 | |
7ad18afa CH |
404 | /* Block layer tags. */ |
405 | struct blk_mq_tag_set tag_set; | |
406 | ||
c666601a JD |
407 | /* protects updating the header */ |
408 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
409 | |
410 | struct rbd_mapping mapping; | |
602adf40 YS |
411 | |
412 | struct list_head node; | |
dfc5606d | 413 | |
dfc5606d YS |
414 | /* sysfs related */ |
415 | struct device dev; | |
b82d167b | 416 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
417 | }; |
418 | ||
b82d167b | 419 | /* |
87c0fded ID |
420 | * Flag bits for rbd_dev->flags: |
421 | * - REMOVING (which is coupled with rbd_dev->open_count) is protected | |
422 | * by rbd_dev->lock | |
423 | * - BLACKLISTED is protected by rbd_dev->lock_rwsem | |
b82d167b | 424 | */ |
6d292906 AE |
425 | enum rbd_dev_flags { |
426 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 427 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
87c0fded | 428 | RBD_DEV_FLAG_BLACKLISTED, /* our ceph_client is blacklisted */ |
6d292906 AE |
429 | }; |
430 | ||
cfbf6377 | 431 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 432 | |
602adf40 | 433 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
434 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
435 | ||
432b8587 AE |
436 | static LIST_HEAD(rbd_client_list); /* clients */ |
437 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 438 | |
78c2a44a AE |
439 | /* Slab caches for frequently-allocated structures */ |
440 | ||
1c2a9dfe | 441 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 442 | static struct kmem_cache *rbd_obj_request_cache; |
1c2a9dfe | 443 | |
9b60e70b | 444 | static int rbd_major; |
f8a22fc2 ID |
445 | static DEFINE_IDA(rbd_dev_id_ida); |
446 | ||
f5ee37bd ID |
447 | static struct workqueue_struct *rbd_wq; |
448 | ||
9b60e70b ID |
449 | /* |
450 | * Default to false for now, as single-major requires >= 0.75 version of | |
451 | * userspace rbd utility. | |
452 | */ | |
453 | static bool single_major = false; | |
454 | module_param(single_major, bool, S_IRUGO); | |
455 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
456 | ||
3d7efd18 AE |
457 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
458 | ||
f0f8cef5 AE |
459 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
460 | size_t count); | |
461 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
462 | size_t count); | |
9b60e70b ID |
463 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
464 | size_t count); | |
465 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
466 | size_t count); | |
6d69bb53 | 467 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
a2acd00e | 468 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 469 | |
9b60e70b ID |
470 | static int rbd_dev_id_to_minor(int dev_id) |
471 | { | |
7e513d43 | 472 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
473 | } |
474 | ||
475 | static int minor_to_rbd_dev_id(int minor) | |
476 | { | |
7e513d43 | 477 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
478 | } |
479 | ||
ed95b21a ID |
480 | static bool rbd_is_lock_supported(struct rbd_device *rbd_dev) |
481 | { | |
482 | return (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) && | |
483 | rbd_dev->spec->snap_id == CEPH_NOSNAP && | |
484 | !rbd_dev->mapping.read_only; | |
485 | } | |
486 | ||
487 | static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev) | |
488 | { | |
489 | return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED || | |
490 | rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING; | |
491 | } | |
492 | ||
493 | static bool rbd_is_lock_owner(struct rbd_device *rbd_dev) | |
494 | { | |
495 | bool is_lock_owner; | |
496 | ||
497 | down_read(&rbd_dev->lock_rwsem); | |
498 | is_lock_owner = __rbd_is_lock_owner(rbd_dev); | |
499 | up_read(&rbd_dev->lock_rwsem); | |
500 | return is_lock_owner; | |
501 | } | |
502 | ||
8767b293 ID |
503 | static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf) |
504 | { | |
505 | return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED); | |
506 | } | |
507 | ||
b15a21dd GKH |
508 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
509 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
510 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
511 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
8767b293 | 512 | static BUS_ATTR(supported_features, S_IRUGO, rbd_supported_features_show, NULL); |
b15a21dd GKH |
513 | |
514 | static struct attribute *rbd_bus_attrs[] = { | |
515 | &bus_attr_add.attr, | |
516 | &bus_attr_remove.attr, | |
9b60e70b ID |
517 | &bus_attr_add_single_major.attr, |
518 | &bus_attr_remove_single_major.attr, | |
8767b293 | 519 | &bus_attr_supported_features.attr, |
b15a21dd | 520 | NULL, |
f0f8cef5 | 521 | }; |
92c76dc0 ID |
522 | |
523 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
524 | struct attribute *attr, int index) | |
525 | { | |
9b60e70b ID |
526 | if (!single_major && |
527 | (attr == &bus_attr_add_single_major.attr || | |
528 | attr == &bus_attr_remove_single_major.attr)) | |
529 | return 0; | |
530 | ||
92c76dc0 ID |
531 | return attr->mode; |
532 | } | |
533 | ||
534 | static const struct attribute_group rbd_bus_group = { | |
535 | .attrs = rbd_bus_attrs, | |
536 | .is_visible = rbd_bus_is_visible, | |
537 | }; | |
538 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
539 | |
540 | static struct bus_type rbd_bus_type = { | |
541 | .name = "rbd", | |
b15a21dd | 542 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
543 | }; |
544 | ||
545 | static void rbd_root_dev_release(struct device *dev) | |
546 | { | |
547 | } | |
548 | ||
549 | static struct device rbd_root_dev = { | |
550 | .init_name = "rbd", | |
551 | .release = rbd_root_dev_release, | |
552 | }; | |
553 | ||
06ecc6cb AE |
554 | static __printf(2, 3) |
555 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
556 | { | |
557 | struct va_format vaf; | |
558 | va_list args; | |
559 | ||
560 | va_start(args, fmt); | |
561 | vaf.fmt = fmt; | |
562 | vaf.va = &args; | |
563 | ||
564 | if (!rbd_dev) | |
565 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
566 | else if (rbd_dev->disk) | |
567 | printk(KERN_WARNING "%s: %s: %pV\n", | |
568 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
569 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
570 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
571 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
572 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
573 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
574 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
575 | else /* punt */ | |
576 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
577 | RBD_DRV_NAME, rbd_dev, &vaf); | |
578 | va_end(args); | |
579 | } | |
580 | ||
aafb230e AE |
581 | #ifdef RBD_DEBUG |
582 | #define rbd_assert(expr) \ | |
583 | if (unlikely(!(expr))) { \ | |
584 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
585 | "at line %d:\n\n" \ | |
586 | "\trbd_assert(%s);\n\n", \ | |
587 | __func__, __LINE__, #expr); \ | |
588 | BUG(); \ | |
589 | } | |
590 | #else /* !RBD_DEBUG */ | |
591 | # define rbd_assert(expr) ((void) 0) | |
592 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 593 | |
2761713d | 594 | static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request); |
b454e36d | 595 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
596 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
597 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 598 | |
cc4a38bd | 599 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 600 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 601 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 602 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
603 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
604 | u64 snap_id); | |
2ad3d716 AE |
605 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
606 | u8 *order, u64 *snap_size); | |
607 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
608 | u64 *snap_features); | |
59c2be1e | 609 | |
602adf40 YS |
610 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
611 | { | |
f0f8cef5 | 612 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 613 | bool removing = false; |
602adf40 | 614 | |
f84344f3 | 615 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
616 | return -EROFS; |
617 | ||
a14ea269 | 618 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
619 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
620 | removing = true; | |
621 | else | |
622 | rbd_dev->open_count++; | |
a14ea269 | 623 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
624 | if (removing) |
625 | return -ENOENT; | |
626 | ||
c3e946ce | 627 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 628 | |
602adf40 YS |
629 | return 0; |
630 | } | |
631 | ||
db2a144b | 632 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
633 | { |
634 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
635 | unsigned long open_count_before; |
636 | ||
a14ea269 | 637 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 638 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 639 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 640 | rbd_assert(open_count_before > 0); |
dfc5606d | 641 | |
c3e946ce | 642 | put_device(&rbd_dev->dev); |
dfc5606d YS |
643 | } |
644 | ||
131fd9f6 GZ |
645 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
646 | { | |
77f33c03 | 647 | int ret = 0; |
131fd9f6 GZ |
648 | int val; |
649 | bool ro; | |
77f33c03 | 650 | bool ro_changed = false; |
131fd9f6 | 651 | |
77f33c03 | 652 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
653 | if (get_user(val, (int __user *)(arg))) |
654 | return -EFAULT; | |
655 | ||
656 | ro = val ? true : false; | |
657 | /* Snapshot doesn't allow to write*/ | |
658 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
659 | return -EROFS; | |
660 | ||
77f33c03 JD |
661 | spin_lock_irq(&rbd_dev->lock); |
662 | /* prevent others open this device */ | |
663 | if (rbd_dev->open_count > 1) { | |
664 | ret = -EBUSY; | |
665 | goto out; | |
666 | } | |
667 | ||
131fd9f6 GZ |
668 | if (rbd_dev->mapping.read_only != ro) { |
669 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 670 | ro_changed = true; |
131fd9f6 GZ |
671 | } |
672 | ||
77f33c03 JD |
673 | out: |
674 | spin_unlock_irq(&rbd_dev->lock); | |
675 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
676 | if (ret == 0 && ro_changed) | |
677 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
678 | ||
679 | return ret; | |
131fd9f6 GZ |
680 | } |
681 | ||
682 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
683 | unsigned int cmd, unsigned long arg) | |
684 | { | |
685 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
686 | int ret = 0; | |
687 | ||
131fd9f6 GZ |
688 | switch (cmd) { |
689 | case BLKROSET: | |
690 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
691 | break; | |
692 | default: | |
693 | ret = -ENOTTY; | |
694 | } | |
695 | ||
131fd9f6 GZ |
696 | return ret; |
697 | } | |
698 | ||
699 | #ifdef CONFIG_COMPAT | |
700 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
701 | unsigned int cmd, unsigned long arg) | |
702 | { | |
703 | return rbd_ioctl(bdev, mode, cmd, arg); | |
704 | } | |
705 | #endif /* CONFIG_COMPAT */ | |
706 | ||
602adf40 YS |
707 | static const struct block_device_operations rbd_bd_ops = { |
708 | .owner = THIS_MODULE, | |
709 | .open = rbd_open, | |
dfc5606d | 710 | .release = rbd_release, |
131fd9f6 GZ |
711 | .ioctl = rbd_ioctl, |
712 | #ifdef CONFIG_COMPAT | |
713 | .compat_ioctl = rbd_compat_ioctl, | |
714 | #endif | |
602adf40 YS |
715 | }; |
716 | ||
717 | /* | |
7262cfca | 718 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 719 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 720 | */ |
f8c38929 | 721 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
722 | { |
723 | struct rbd_client *rbdc; | |
724 | int ret = -ENOMEM; | |
725 | ||
37206ee5 | 726 | dout("%s:\n", __func__); |
602adf40 YS |
727 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
728 | if (!rbdc) | |
729 | goto out_opt; | |
730 | ||
731 | kref_init(&rbdc->kref); | |
732 | INIT_LIST_HEAD(&rbdc->node); | |
733 | ||
43ae4701 | 734 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 735 | if (IS_ERR(rbdc->client)) |
08f75463 | 736 | goto out_rbdc; |
43ae4701 | 737 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
738 | |
739 | ret = ceph_open_session(rbdc->client); | |
740 | if (ret < 0) | |
08f75463 | 741 | goto out_client; |
602adf40 | 742 | |
432b8587 | 743 | spin_lock(&rbd_client_list_lock); |
602adf40 | 744 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 745 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 746 | |
37206ee5 | 747 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 748 | |
602adf40 | 749 | return rbdc; |
08f75463 | 750 | out_client: |
602adf40 | 751 | ceph_destroy_client(rbdc->client); |
08f75463 | 752 | out_rbdc: |
602adf40 YS |
753 | kfree(rbdc); |
754 | out_opt: | |
43ae4701 AE |
755 | if (ceph_opts) |
756 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
757 | dout("%s: error %d\n", __func__, ret); |
758 | ||
28f259b7 | 759 | return ERR_PTR(ret); |
602adf40 YS |
760 | } |
761 | ||
2f82ee54 AE |
762 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
763 | { | |
764 | kref_get(&rbdc->kref); | |
765 | ||
766 | return rbdc; | |
767 | } | |
768 | ||
602adf40 | 769 | /* |
1f7ba331 AE |
770 | * Find a ceph client with specific addr and configuration. If |
771 | * found, bump its reference count. | |
602adf40 | 772 | */ |
1f7ba331 | 773 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
774 | { |
775 | struct rbd_client *client_node; | |
1f7ba331 | 776 | bool found = false; |
602adf40 | 777 | |
43ae4701 | 778 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
779 | return NULL; |
780 | ||
1f7ba331 AE |
781 | spin_lock(&rbd_client_list_lock); |
782 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
783 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
784 | __rbd_get_client(client_node); |
785 | ||
1f7ba331 AE |
786 | found = true; |
787 | break; | |
788 | } | |
789 | } | |
790 | spin_unlock(&rbd_client_list_lock); | |
791 | ||
792 | return found ? client_node : NULL; | |
602adf40 YS |
793 | } |
794 | ||
59c2be1e | 795 | /* |
210c104c | 796 | * (Per device) rbd map options |
59c2be1e YS |
797 | */ |
798 | enum { | |
b5584180 | 799 | Opt_queue_depth, |
59c2be1e YS |
800 | Opt_last_int, |
801 | /* int args above */ | |
802 | Opt_last_string, | |
803 | /* string args above */ | |
cc0538b6 AE |
804 | Opt_read_only, |
805 | Opt_read_write, | |
80de1912 | 806 | Opt_lock_on_read, |
210c104c | 807 | Opt_err |
59c2be1e YS |
808 | }; |
809 | ||
43ae4701 | 810 | static match_table_t rbd_opts_tokens = { |
b5584180 | 811 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
812 | /* int args above */ |
813 | /* string args above */ | |
be466c1c | 814 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
815 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
816 | {Opt_read_write, "read_write"}, | |
817 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
80de1912 | 818 | {Opt_lock_on_read, "lock_on_read"}, |
210c104c | 819 | {Opt_err, NULL} |
59c2be1e YS |
820 | }; |
821 | ||
98571b5a | 822 | struct rbd_options { |
b5584180 | 823 | int queue_depth; |
98571b5a | 824 | bool read_only; |
80de1912 | 825 | bool lock_on_read; |
98571b5a AE |
826 | }; |
827 | ||
b5584180 | 828 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a | 829 | #define RBD_READ_ONLY_DEFAULT false |
80de1912 | 830 | #define RBD_LOCK_ON_READ_DEFAULT false |
98571b5a | 831 | |
59c2be1e YS |
832 | static int parse_rbd_opts_token(char *c, void *private) |
833 | { | |
43ae4701 | 834 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
835 | substring_t argstr[MAX_OPT_ARGS]; |
836 | int token, intval, ret; | |
837 | ||
43ae4701 | 838 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
839 | if (token < Opt_last_int) { |
840 | ret = match_int(&argstr[0], &intval); | |
841 | if (ret < 0) { | |
210c104c | 842 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
843 | return ret; |
844 | } | |
845 | dout("got int token %d val %d\n", token, intval); | |
846 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 847 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
848 | } else { |
849 | dout("got token %d\n", token); | |
850 | } | |
851 | ||
852 | switch (token) { | |
b5584180 ID |
853 | case Opt_queue_depth: |
854 | if (intval < 1) { | |
855 | pr_err("queue_depth out of range\n"); | |
856 | return -EINVAL; | |
857 | } | |
858 | rbd_opts->queue_depth = intval; | |
859 | break; | |
cc0538b6 AE |
860 | case Opt_read_only: |
861 | rbd_opts->read_only = true; | |
862 | break; | |
863 | case Opt_read_write: | |
864 | rbd_opts->read_only = false; | |
865 | break; | |
80de1912 ID |
866 | case Opt_lock_on_read: |
867 | rbd_opts->lock_on_read = true; | |
868 | break; | |
59c2be1e | 869 | default: |
210c104c ID |
870 | /* libceph prints "bad option" msg */ |
871 | return -EINVAL; | |
59c2be1e | 872 | } |
210c104c | 873 | |
59c2be1e YS |
874 | return 0; |
875 | } | |
876 | ||
6d2940c8 GZ |
877 | static char* obj_op_name(enum obj_operation_type op_type) |
878 | { | |
879 | switch (op_type) { | |
880 | case OBJ_OP_READ: | |
881 | return "read"; | |
882 | case OBJ_OP_WRITE: | |
883 | return "write"; | |
90e98c52 GZ |
884 | case OBJ_OP_DISCARD: |
885 | return "discard"; | |
6d2940c8 GZ |
886 | default: |
887 | return "???"; | |
888 | } | |
889 | } | |
890 | ||
602adf40 YS |
891 | /* |
892 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
893 | * not exist create it. Either way, ceph_opts is consumed by this |
894 | * function. | |
602adf40 | 895 | */ |
9d3997fd | 896 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 897 | { |
f8c38929 | 898 | struct rbd_client *rbdc; |
59c2be1e | 899 | |
cfbf6377 | 900 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 901 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 902 | if (rbdc) /* using an existing client */ |
43ae4701 | 903 | ceph_destroy_options(ceph_opts); |
9d3997fd | 904 | else |
f8c38929 | 905 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 906 | mutex_unlock(&client_mutex); |
602adf40 | 907 | |
9d3997fd | 908 | return rbdc; |
602adf40 YS |
909 | } |
910 | ||
911 | /* | |
912 | * Destroy ceph client | |
d23a4b3f | 913 | * |
432b8587 | 914 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
915 | */ |
916 | static void rbd_client_release(struct kref *kref) | |
917 | { | |
918 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
919 | ||
37206ee5 | 920 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 921 | spin_lock(&rbd_client_list_lock); |
602adf40 | 922 | list_del(&rbdc->node); |
cd9d9f5d | 923 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
924 | |
925 | ceph_destroy_client(rbdc->client); | |
926 | kfree(rbdc); | |
927 | } | |
928 | ||
929 | /* | |
930 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
931 | * it. | |
932 | */ | |
9d3997fd | 933 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 934 | { |
c53d5893 AE |
935 | if (rbdc) |
936 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
937 | } |
938 | ||
a30b71b9 AE |
939 | static bool rbd_image_format_valid(u32 image_format) |
940 | { | |
941 | return image_format == 1 || image_format == 2; | |
942 | } | |
943 | ||
8e94af8e AE |
944 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
945 | { | |
103a150f AE |
946 | size_t size; |
947 | u32 snap_count; | |
948 | ||
949 | /* The header has to start with the magic rbd header text */ | |
950 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
951 | return false; | |
952 | ||
db2388b6 AE |
953 | /* The bio layer requires at least sector-sized I/O */ |
954 | ||
955 | if (ondisk->options.order < SECTOR_SHIFT) | |
956 | return false; | |
957 | ||
958 | /* If we use u64 in a few spots we may be able to loosen this */ | |
959 | ||
960 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
961 | return false; | |
962 | ||
103a150f AE |
963 | /* |
964 | * The size of a snapshot header has to fit in a size_t, and | |
965 | * that limits the number of snapshots. | |
966 | */ | |
967 | snap_count = le32_to_cpu(ondisk->snap_count); | |
968 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
969 | if (snap_count > size / sizeof (__le64)) | |
970 | return false; | |
971 | ||
972 | /* | |
973 | * Not only that, but the size of the entire the snapshot | |
974 | * header must also be representable in a size_t. | |
975 | */ | |
976 | size -= snap_count * sizeof (__le64); | |
977 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
978 | return false; | |
979 | ||
980 | return true; | |
8e94af8e AE |
981 | } |
982 | ||
5bc3fb17 ID |
983 | /* |
984 | * returns the size of an object in the image | |
985 | */ | |
986 | static u32 rbd_obj_bytes(struct rbd_image_header *header) | |
987 | { | |
988 | return 1U << header->obj_order; | |
989 | } | |
990 | ||
263423f8 ID |
991 | static void rbd_init_layout(struct rbd_device *rbd_dev) |
992 | { | |
993 | if (rbd_dev->header.stripe_unit == 0 || | |
994 | rbd_dev->header.stripe_count == 0) { | |
995 | rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header); | |
996 | rbd_dev->header.stripe_count = 1; | |
997 | } | |
998 | ||
999 | rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit; | |
1000 | rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count; | |
1001 | rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header); | |
7e97332e ID |
1002 | rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ? |
1003 | rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id; | |
263423f8 ID |
1004 | RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL); |
1005 | } | |
1006 | ||
602adf40 | 1007 | /* |
bb23e37a AE |
1008 | * Fill an rbd image header with information from the given format 1 |
1009 | * on-disk header. | |
602adf40 | 1010 | */ |
662518b1 | 1011 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 1012 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 1013 | { |
662518b1 | 1014 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
1015 | bool first_time = header->object_prefix == NULL; |
1016 | struct ceph_snap_context *snapc; | |
1017 | char *object_prefix = NULL; | |
1018 | char *snap_names = NULL; | |
1019 | u64 *snap_sizes = NULL; | |
ccece235 | 1020 | u32 snap_count; |
bb23e37a | 1021 | int ret = -ENOMEM; |
621901d6 | 1022 | u32 i; |
602adf40 | 1023 | |
bb23e37a | 1024 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 1025 | |
bb23e37a | 1026 | if (first_time) { |
848d796c ID |
1027 | object_prefix = kstrndup(ondisk->object_prefix, |
1028 | sizeof(ondisk->object_prefix), | |
1029 | GFP_KERNEL); | |
bb23e37a AE |
1030 | if (!object_prefix) |
1031 | return -ENOMEM; | |
bb23e37a | 1032 | } |
00f1f36f | 1033 | |
bb23e37a | 1034 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 1035 | |
bb23e37a AE |
1036 | snap_count = le32_to_cpu(ondisk->snap_count); |
1037 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
1038 | if (!snapc) | |
1039 | goto out_err; | |
1040 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 1041 | if (snap_count) { |
bb23e37a | 1042 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
1043 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
1044 | ||
bb23e37a | 1045 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 1046 | |
bb23e37a AE |
1047 | if (snap_names_len > (u64)SIZE_MAX) |
1048 | goto out_2big; | |
1049 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
1050 | if (!snap_names) | |
6a52325f AE |
1051 | goto out_err; |
1052 | ||
bb23e37a | 1053 | /* ...as well as the array of their sizes. */ |
88a25a5f ME |
1054 | snap_sizes = kmalloc_array(snap_count, |
1055 | sizeof(*header->snap_sizes), | |
1056 | GFP_KERNEL); | |
bb23e37a | 1057 | if (!snap_sizes) |
6a52325f | 1058 | goto out_err; |
bb23e37a | 1059 | |
f785cc1d | 1060 | /* |
bb23e37a AE |
1061 | * Copy the names, and fill in each snapshot's id |
1062 | * and size. | |
1063 | * | |
99a41ebc | 1064 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 1065 | * ondisk buffer we're working with has |
f785cc1d AE |
1066 | * snap_names_len bytes beyond the end of the |
1067 | * snapshot id array, this memcpy() is safe. | |
1068 | */ | |
bb23e37a AE |
1069 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
1070 | snaps = ondisk->snaps; | |
1071 | for (i = 0; i < snap_count; i++) { | |
1072 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
1073 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
1074 | } | |
602adf40 | 1075 | } |
6a52325f | 1076 | |
bb23e37a | 1077 | /* We won't fail any more, fill in the header */ |
621901d6 | 1078 | |
bb23e37a AE |
1079 | if (first_time) { |
1080 | header->object_prefix = object_prefix; | |
1081 | header->obj_order = ondisk->options.order; | |
263423f8 | 1082 | rbd_init_layout(rbd_dev); |
602adf40 | 1083 | } else { |
662518b1 AE |
1084 | ceph_put_snap_context(header->snapc); |
1085 | kfree(header->snap_names); | |
1086 | kfree(header->snap_sizes); | |
602adf40 | 1087 | } |
849b4260 | 1088 | |
bb23e37a | 1089 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1090 | |
f84344f3 | 1091 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1092 | header->snapc = snapc; |
1093 | header->snap_names = snap_names; | |
1094 | header->snap_sizes = snap_sizes; | |
468521c1 | 1095 | |
602adf40 | 1096 | return 0; |
bb23e37a AE |
1097 | out_2big: |
1098 | ret = -EIO; | |
6a52325f | 1099 | out_err: |
bb23e37a AE |
1100 | kfree(snap_sizes); |
1101 | kfree(snap_names); | |
1102 | ceph_put_snap_context(snapc); | |
1103 | kfree(object_prefix); | |
ccece235 | 1104 | |
bb23e37a | 1105 | return ret; |
602adf40 YS |
1106 | } |
1107 | ||
9682fc6d AE |
1108 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1109 | { | |
1110 | const char *snap_name; | |
1111 | ||
1112 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1113 | ||
1114 | /* Skip over names until we find the one we are looking for */ | |
1115 | ||
1116 | snap_name = rbd_dev->header.snap_names; | |
1117 | while (which--) | |
1118 | snap_name += strlen(snap_name) + 1; | |
1119 | ||
1120 | return kstrdup(snap_name, GFP_KERNEL); | |
1121 | } | |
1122 | ||
30d1cff8 AE |
1123 | /* |
1124 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1125 | * Note that result is for snapshots in *descending* order. | |
1126 | */ | |
1127 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1128 | { | |
1129 | u64 snap_id1 = *(u64 *)s1; | |
1130 | u64 snap_id2 = *(u64 *)s2; | |
1131 | ||
1132 | if (snap_id1 < snap_id2) | |
1133 | return 1; | |
1134 | return snap_id1 == snap_id2 ? 0 : -1; | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Search a snapshot context to see if the given snapshot id is | |
1139 | * present. | |
1140 | * | |
1141 | * Returns the position of the snapshot id in the array if it's found, | |
1142 | * or BAD_SNAP_INDEX otherwise. | |
1143 | * | |
1144 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1145 | * reverse order, highest snapshot id first. | |
1146 | */ | |
9682fc6d AE |
1147 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1148 | { | |
1149 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1150 | u64 *found; |
9682fc6d | 1151 | |
30d1cff8 AE |
1152 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1153 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1154 | |
30d1cff8 | 1155 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1156 | } |
1157 | ||
2ad3d716 AE |
1158 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1159 | u64 snap_id) | |
9e15b77d | 1160 | { |
54cac61f | 1161 | u32 which; |
da6a6b63 | 1162 | const char *snap_name; |
9e15b77d | 1163 | |
54cac61f AE |
1164 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1165 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1166 | return ERR_PTR(-ENOENT); |
54cac61f | 1167 | |
da6a6b63 JD |
1168 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1169 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1170 | } |
1171 | ||
1172 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1173 | { | |
9e15b77d AE |
1174 | if (snap_id == CEPH_NOSNAP) |
1175 | return RBD_SNAP_HEAD_NAME; | |
1176 | ||
54cac61f AE |
1177 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1178 | if (rbd_dev->image_format == 1) | |
1179 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1180 | |
54cac61f | 1181 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1182 | } |
1183 | ||
2ad3d716 AE |
1184 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1185 | u64 *snap_size) | |
602adf40 | 1186 | { |
2ad3d716 AE |
1187 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1188 | if (snap_id == CEPH_NOSNAP) { | |
1189 | *snap_size = rbd_dev->header.image_size; | |
1190 | } else if (rbd_dev->image_format == 1) { | |
1191 | u32 which; | |
602adf40 | 1192 | |
2ad3d716 AE |
1193 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1194 | if (which == BAD_SNAP_INDEX) | |
1195 | return -ENOENT; | |
e86924a8 | 1196 | |
2ad3d716 AE |
1197 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1198 | } else { | |
1199 | u64 size = 0; | |
1200 | int ret; | |
1201 | ||
1202 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1203 | if (ret) | |
1204 | return ret; | |
1205 | ||
1206 | *snap_size = size; | |
1207 | } | |
1208 | return 0; | |
602adf40 YS |
1209 | } |
1210 | ||
2ad3d716 AE |
1211 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1212 | u64 *snap_features) | |
602adf40 | 1213 | { |
2ad3d716 AE |
1214 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1215 | if (snap_id == CEPH_NOSNAP) { | |
1216 | *snap_features = rbd_dev->header.features; | |
1217 | } else if (rbd_dev->image_format == 1) { | |
1218 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1219 | } else { |
2ad3d716 AE |
1220 | u64 features = 0; |
1221 | int ret; | |
8b0241f8 | 1222 | |
2ad3d716 AE |
1223 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1224 | if (ret) | |
1225 | return ret; | |
1226 | ||
1227 | *snap_features = features; | |
1228 | } | |
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1233 | { | |
8f4b7d98 | 1234 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1235 | u64 size = 0; |
1236 | u64 features = 0; | |
1237 | int ret; | |
1238 | ||
2ad3d716 AE |
1239 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1240 | if (ret) | |
1241 | return ret; | |
1242 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1243 | if (ret) | |
1244 | return ret; | |
1245 | ||
1246 | rbd_dev->mapping.size = size; | |
1247 | rbd_dev->mapping.features = features; | |
1248 | ||
8b0241f8 | 1249 | return 0; |
602adf40 YS |
1250 | } |
1251 | ||
d1cf5788 AE |
1252 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1253 | { | |
1254 | rbd_dev->mapping.size = 0; | |
1255 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1256 | } |
1257 | ||
65ccfe21 AE |
1258 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1259 | { | |
5bc3fb17 | 1260 | u64 segment_size = rbd_obj_bytes(&rbd_dev->header); |
602adf40 | 1261 | |
65ccfe21 AE |
1262 | return offset & (segment_size - 1); |
1263 | } | |
1264 | ||
1265 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1266 | u64 offset, u64 length) | |
1267 | { | |
5bc3fb17 | 1268 | u64 segment_size = rbd_obj_bytes(&rbd_dev->header); |
65ccfe21 AE |
1269 | |
1270 | offset &= segment_size - 1; | |
1271 | ||
aafb230e | 1272 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1273 | if (offset + length > segment_size) |
1274 | length = segment_size - offset; | |
1275 | ||
1276 | return length; | |
602adf40 YS |
1277 | } |
1278 | ||
1279 | /* | |
1280 | * bio helpers | |
1281 | */ | |
1282 | ||
1283 | static void bio_chain_put(struct bio *chain) | |
1284 | { | |
1285 | struct bio *tmp; | |
1286 | ||
1287 | while (chain) { | |
1288 | tmp = chain; | |
1289 | chain = chain->bi_next; | |
1290 | bio_put(tmp); | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * zeros a bio chain, starting at specific offset | |
1296 | */ | |
1297 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1298 | { | |
7988613b KO |
1299 | struct bio_vec bv; |
1300 | struct bvec_iter iter; | |
602adf40 YS |
1301 | unsigned long flags; |
1302 | void *buf; | |
602adf40 YS |
1303 | int pos = 0; |
1304 | ||
1305 | while (chain) { | |
7988613b KO |
1306 | bio_for_each_segment(bv, chain, iter) { |
1307 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1308 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1309 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1310 | memset(buf + remainder, 0, |
7988613b KO |
1311 | bv.bv_len - remainder); |
1312 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1313 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1314 | } |
7988613b | 1315 | pos += bv.bv_len; |
602adf40 YS |
1316 | } |
1317 | ||
1318 | chain = chain->bi_next; | |
1319 | } | |
1320 | } | |
1321 | ||
b9434c5b AE |
1322 | /* |
1323 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1324 | * starting at the given byte offset from the start of the array and | |
1325 | * continuing up to the given end offset. The pages array is | |
1326 | * assumed to be big enough to hold all bytes up to the end. | |
1327 | */ | |
1328 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1329 | { | |
1330 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1331 | ||
1332 | rbd_assert(end > offset); | |
1333 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1334 | while (offset < end) { | |
1335 | size_t page_offset; | |
1336 | size_t length; | |
1337 | unsigned long flags; | |
1338 | void *kaddr; | |
1339 | ||
491205a8 GU |
1340 | page_offset = offset & ~PAGE_MASK; |
1341 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1342 | local_irq_save(flags); |
1343 | kaddr = kmap_atomic(*page); | |
1344 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1345 | flush_dcache_page(*page); |
b9434c5b AE |
1346 | kunmap_atomic(kaddr); |
1347 | local_irq_restore(flags); | |
1348 | ||
1349 | offset += length; | |
1350 | page++; | |
1351 | } | |
1352 | } | |
1353 | ||
602adf40 | 1354 | /* |
f7760dad AE |
1355 | * Clone a portion of a bio, starting at the given byte offset |
1356 | * and continuing for the number of bytes indicated. | |
602adf40 | 1357 | */ |
f7760dad AE |
1358 | static struct bio *bio_clone_range(struct bio *bio_src, |
1359 | unsigned int offset, | |
1360 | unsigned int len, | |
1361 | gfp_t gfpmask) | |
602adf40 | 1362 | { |
f7760dad AE |
1363 | struct bio *bio; |
1364 | ||
5341a627 | 1365 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1366 | if (!bio) |
1367 | return NULL; /* ENOMEM */ | |
602adf40 | 1368 | |
5341a627 | 1369 | bio_advance(bio, offset); |
4f024f37 | 1370 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1371 | |
1372 | return bio; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Clone a portion of a bio chain, starting at the given byte offset | |
1377 | * into the first bio in the source chain and continuing for the | |
1378 | * number of bytes indicated. The result is another bio chain of | |
1379 | * exactly the given length, or a null pointer on error. | |
1380 | * | |
1381 | * The bio_src and offset parameters are both in-out. On entry they | |
1382 | * refer to the first source bio and the offset into that bio where | |
1383 | * the start of data to be cloned is located. | |
1384 | * | |
1385 | * On return, bio_src is updated to refer to the bio in the source | |
1386 | * chain that contains first un-cloned byte, and *offset will | |
1387 | * contain the offset of that byte within that bio. | |
1388 | */ | |
1389 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1390 | unsigned int *offset, | |
1391 | unsigned int len, | |
1392 | gfp_t gfpmask) | |
1393 | { | |
1394 | struct bio *bi = *bio_src; | |
1395 | unsigned int off = *offset; | |
1396 | struct bio *chain = NULL; | |
1397 | struct bio **end; | |
1398 | ||
1399 | /* Build up a chain of clone bios up to the limit */ | |
1400 | ||
4f024f37 | 1401 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1402 | return NULL; /* Nothing to clone */ |
602adf40 | 1403 | |
f7760dad AE |
1404 | end = &chain; |
1405 | while (len) { | |
1406 | unsigned int bi_size; | |
1407 | struct bio *bio; | |
1408 | ||
f5400b7a AE |
1409 | if (!bi) { |
1410 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1411 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1412 | } |
4f024f37 | 1413 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1414 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1415 | if (!bio) | |
1416 | goto out_err; /* ENOMEM */ | |
1417 | ||
1418 | *end = bio; | |
1419 | end = &bio->bi_next; | |
602adf40 | 1420 | |
f7760dad | 1421 | off += bi_size; |
4f024f37 | 1422 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1423 | bi = bi->bi_next; |
1424 | off = 0; | |
1425 | } | |
1426 | len -= bi_size; | |
1427 | } | |
1428 | *bio_src = bi; | |
1429 | *offset = off; | |
1430 | ||
1431 | return chain; | |
1432 | out_err: | |
1433 | bio_chain_put(chain); | |
602adf40 | 1434 | |
602adf40 YS |
1435 | return NULL; |
1436 | } | |
1437 | ||
926f9b3f AE |
1438 | /* |
1439 | * The default/initial value for all object request flags is 0. For | |
1440 | * each flag, once its value is set to 1 it is never reset to 0 | |
1441 | * again. | |
1442 | */ | |
57acbaa7 | 1443 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1444 | { |
57acbaa7 | 1445 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1446 | struct rbd_device *rbd_dev; |
1447 | ||
57acbaa7 | 1448 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1449 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1450 | obj_request); |
1451 | } | |
1452 | } | |
1453 | ||
57acbaa7 | 1454 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1455 | { |
1456 | smp_mb(); | |
57acbaa7 | 1457 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1458 | } |
1459 | ||
57acbaa7 | 1460 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1461 | { |
57acbaa7 AE |
1462 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1463 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1464 | |
57acbaa7 AE |
1465 | if (obj_request_img_data_test(obj_request)) |
1466 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1467 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1468 | obj_request); |
1469 | } | |
1470 | } | |
1471 | ||
57acbaa7 | 1472 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1473 | { |
1474 | smp_mb(); | |
57acbaa7 | 1475 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1476 | } |
1477 | ||
5679c59f AE |
1478 | /* |
1479 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1480 | * flag. The latter is set based on the "exists" value provided. | |
1481 | * | |
1482 | * Note that for our purposes once an object exists it never goes | |
1483 | * away again. It's possible that the response from two existence | |
1484 | * checks are separated by the creation of the target object, and | |
1485 | * the first ("doesn't exist") response arrives *after* the second | |
1486 | * ("does exist"). In that case we ignore the second one. | |
1487 | */ | |
1488 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1489 | bool exists) | |
1490 | { | |
1491 | if (exists) | |
1492 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1493 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1494 | smp_mb(); | |
1495 | } | |
1496 | ||
1497 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1498 | { | |
1499 | smp_mb(); | |
1500 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1501 | } | |
1502 | ||
1503 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1504 | { | |
1505 | smp_mb(); | |
1506 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1507 | } | |
1508 | ||
9638556a ID |
1509 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1510 | { | |
1511 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1512 | ||
1513 | return obj_request->img_offset < | |
1514 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1515 | } | |
1516 | ||
bf0d5f50 AE |
1517 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1518 | { | |
37206ee5 | 1519 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
2c935bc5 | 1520 | kref_read(&obj_request->kref)); |
bf0d5f50 AE |
1521 | kref_get(&obj_request->kref); |
1522 | } | |
1523 | ||
1524 | static void rbd_obj_request_destroy(struct kref *kref); | |
1525 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1526 | { | |
1527 | rbd_assert(obj_request != NULL); | |
37206ee5 | 1528 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
2c935bc5 | 1529 | kref_read(&obj_request->kref)); |
bf0d5f50 AE |
1530 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1531 | } | |
1532 | ||
0f2d5be7 AE |
1533 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1534 | { | |
1535 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
2c935bc5 | 1536 | kref_read(&img_request->kref)); |
0f2d5be7 AE |
1537 | kref_get(&img_request->kref); |
1538 | } | |
1539 | ||
e93f3152 AE |
1540 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1541 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1542 | static void rbd_img_request_destroy(struct kref *kref); |
1543 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1544 | { | |
1545 | rbd_assert(img_request != NULL); | |
37206ee5 | 1546 | dout("%s: img %p (was %d)\n", __func__, img_request, |
2c935bc5 | 1547 | kref_read(&img_request->kref)); |
e93f3152 AE |
1548 | if (img_request_child_test(img_request)) |
1549 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1550 | else | |
1551 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1552 | } |
1553 | ||
1554 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1555 | struct rbd_obj_request *obj_request) | |
1556 | { | |
25dcf954 AE |
1557 | rbd_assert(obj_request->img_request == NULL); |
1558 | ||
b155e86c | 1559 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1560 | obj_request->img_request = img_request; |
25dcf954 | 1561 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1562 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1563 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1564 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1565 | img_request->obj_request_count++; |
1566 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1567 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1568 | obj_request->which); | |
bf0d5f50 AE |
1569 | } |
1570 | ||
1571 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1572 | struct rbd_obj_request *obj_request) | |
1573 | { | |
1574 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1575 | |
37206ee5 AE |
1576 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1577 | obj_request->which); | |
bf0d5f50 | 1578 | list_del(&obj_request->links); |
25dcf954 AE |
1579 | rbd_assert(img_request->obj_request_count > 0); |
1580 | img_request->obj_request_count--; | |
1581 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1582 | obj_request->which = BAD_WHICH; | |
6365d33a | 1583 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1584 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1585 | obj_request->img_request = NULL; |
25dcf954 | 1586 | obj_request->callback = NULL; |
bf0d5f50 AE |
1587 | rbd_obj_request_put(obj_request); |
1588 | } | |
1589 | ||
1590 | static bool obj_request_type_valid(enum obj_request_type type) | |
1591 | { | |
1592 | switch (type) { | |
9969ebc5 | 1593 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1594 | case OBJ_REQUEST_BIO: |
788e2df3 | 1595 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1596 | return true; |
1597 | default: | |
1598 | return false; | |
1599 | } | |
1600 | } | |
1601 | ||
4a17dadc ID |
1602 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request); |
1603 | ||
980917fc | 1604 | static void rbd_obj_request_submit(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1605 | { |
980917fc ID |
1606 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
1607 | ||
a90bb0c1 ID |
1608 | dout("%s %p object_no %016llx %llu~%llu osd_req %p\n", __func__, |
1609 | obj_request, obj_request->object_no, obj_request->offset, | |
67e2b652 | 1610 | obj_request->length, osd_req); |
4a17dadc ID |
1611 | if (obj_request_img_data_test(obj_request)) { |
1612 | WARN_ON(obj_request->callback != rbd_img_obj_callback); | |
1613 | rbd_img_request_get(obj_request->img_request); | |
1614 | } | |
980917fc | 1615 | ceph_osdc_start_request(osd_req->r_osdc, osd_req, false); |
bf0d5f50 AE |
1616 | } |
1617 | ||
1618 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1619 | { | |
55f27e09 | 1620 | |
37206ee5 | 1621 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1622 | |
1623 | /* | |
1624 | * If no error occurred, compute the aggregate transfer | |
1625 | * count for the image request. We could instead use | |
1626 | * atomic64_cmpxchg() to update it as each object request | |
1627 | * completes; not clear which way is better off hand. | |
1628 | */ | |
1629 | if (!img_request->result) { | |
1630 | struct rbd_obj_request *obj_request; | |
1631 | u64 xferred = 0; | |
1632 | ||
1633 | for_each_obj_request(img_request, obj_request) | |
1634 | xferred += obj_request->xferred; | |
1635 | img_request->xferred = xferred; | |
1636 | } | |
1637 | ||
bf0d5f50 AE |
1638 | if (img_request->callback) |
1639 | img_request->callback(img_request); | |
1640 | else | |
1641 | rbd_img_request_put(img_request); | |
1642 | } | |
1643 | ||
0c425248 AE |
1644 | /* |
1645 | * The default/initial value for all image request flags is 0. Each | |
1646 | * is conditionally set to 1 at image request initialization time | |
1647 | * and currently never change thereafter. | |
1648 | */ | |
1649 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1650 | { | |
1651 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1652 | smp_mb(); | |
1653 | } | |
1654 | ||
1655 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1656 | { | |
1657 | smp_mb(); | |
1658 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1659 | } | |
1660 | ||
90e98c52 GZ |
1661 | /* |
1662 | * Set the discard flag when the img_request is an discard request | |
1663 | */ | |
1664 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1665 | { | |
1666 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1667 | smp_mb(); | |
1668 | } | |
1669 | ||
1670 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1671 | { | |
1672 | smp_mb(); | |
1673 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1674 | } | |
1675 | ||
9849e986 AE |
1676 | static void img_request_child_set(struct rbd_img_request *img_request) |
1677 | { | |
1678 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1679 | smp_mb(); | |
1680 | } | |
1681 | ||
e93f3152 AE |
1682 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1683 | { | |
1684 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1685 | smp_mb(); | |
1686 | } | |
1687 | ||
9849e986 AE |
1688 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1689 | { | |
1690 | smp_mb(); | |
1691 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1692 | } | |
1693 | ||
d0b2e944 AE |
1694 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1695 | { | |
1696 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1697 | smp_mb(); | |
1698 | } | |
1699 | ||
a2acd00e AE |
1700 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1701 | { | |
1702 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1703 | smp_mb(); | |
1704 | } | |
1705 | ||
d0b2e944 AE |
1706 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1707 | { | |
1708 | smp_mb(); | |
1709 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1710 | } | |
1711 | ||
3b434a2a JD |
1712 | static enum obj_operation_type |
1713 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1714 | { | |
1715 | if (img_request_write_test(img_request)) | |
1716 | return OBJ_OP_WRITE; | |
1717 | else if (img_request_discard_test(img_request)) | |
1718 | return OBJ_OP_DISCARD; | |
1719 | else | |
1720 | return OBJ_OP_READ; | |
1721 | } | |
1722 | ||
6e2a4505 AE |
1723 | static void |
1724 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1725 | { | |
b9434c5b AE |
1726 | u64 xferred = obj_request->xferred; |
1727 | u64 length = obj_request->length; | |
1728 | ||
6e2a4505 AE |
1729 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1730 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1731 | xferred, length); |
6e2a4505 | 1732 | /* |
17c1cc1d JD |
1733 | * ENOENT means a hole in the image. We zero-fill the entire |
1734 | * length of the request. A short read also implies zero-fill | |
1735 | * to the end of the request. An error requires the whole | |
1736 | * length of the request to be reported finished with an error | |
1737 | * to the block layer. In each case we update the xferred | |
1738 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1739 | */ |
b9434c5b | 1740 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1741 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1742 | if (obj_request->type == OBJ_REQUEST_BIO) |
1743 | zero_bio_chain(obj_request->bio_list, 0); | |
1744 | else | |
1745 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1746 | obj_request->result = 0; |
b9434c5b AE |
1747 | } else if (xferred < length && !obj_request->result) { |
1748 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1749 | zero_bio_chain(obj_request->bio_list, xferred); | |
1750 | else | |
1751 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1752 | } |
17c1cc1d | 1753 | obj_request->xferred = length; |
6e2a4505 AE |
1754 | obj_request_done_set(obj_request); |
1755 | } | |
1756 | ||
bf0d5f50 AE |
1757 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1758 | { | |
37206ee5 AE |
1759 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1760 | obj_request->callback); | |
bf0d5f50 AE |
1761 | if (obj_request->callback) |
1762 | obj_request->callback(obj_request); | |
788e2df3 AE |
1763 | else |
1764 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1765 | } |
1766 | ||
0dcc685e ID |
1767 | static void rbd_obj_request_error(struct rbd_obj_request *obj_request, int err) |
1768 | { | |
1769 | obj_request->result = err; | |
1770 | obj_request->xferred = 0; | |
1771 | /* | |
1772 | * kludge - mirror rbd_obj_request_submit() to match a put in | |
1773 | * rbd_img_obj_callback() | |
1774 | */ | |
1775 | if (obj_request_img_data_test(obj_request)) { | |
1776 | WARN_ON(obj_request->callback != rbd_img_obj_callback); | |
1777 | rbd_img_request_get(obj_request->img_request); | |
1778 | } | |
1779 | obj_request_done_set(obj_request); | |
1780 | rbd_obj_request_complete(obj_request); | |
1781 | } | |
1782 | ||
c47f9371 | 1783 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1784 | { |
57acbaa7 | 1785 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1786 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1787 | bool layered = false; |
1788 | ||
1789 | if (obj_request_img_data_test(obj_request)) { | |
1790 | img_request = obj_request->img_request; | |
1791 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1792 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1793 | } |
8b3e1a56 AE |
1794 | |
1795 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1796 | obj_request, img_request, obj_request->result, | |
1797 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1798 | if (layered && obj_request->result == -ENOENT && |
1799 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1800 | rbd_img_parent_read(obj_request); |
1801 | else if (img_request) | |
6e2a4505 AE |
1802 | rbd_img_obj_request_read_callback(obj_request); |
1803 | else | |
1804 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1805 | } |
1806 | ||
c47f9371 | 1807 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1808 | { |
1b83bef2 SW |
1809 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1810 | obj_request->result, obj_request->length); | |
1811 | /* | |
8b3e1a56 AE |
1812 | * There is no such thing as a successful short write. Set |
1813 | * it to our originally-requested length. | |
1b83bef2 SW |
1814 | */ |
1815 | obj_request->xferred = obj_request->length; | |
07741308 | 1816 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1817 | } |
1818 | ||
90e98c52 GZ |
1819 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1820 | { | |
1821 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1822 | obj_request->result, obj_request->length); | |
1823 | /* | |
1824 | * There is no such thing as a successful short discard. Set | |
1825 | * it to our originally-requested length. | |
1826 | */ | |
1827 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1828 | /* discarding a non-existent object is not a problem */ |
1829 | if (obj_request->result == -ENOENT) | |
1830 | obj_request->result = 0; | |
90e98c52 GZ |
1831 | obj_request_done_set(obj_request); |
1832 | } | |
1833 | ||
fbfab539 AE |
1834 | /* |
1835 | * For a simple stat call there's nothing to do. We'll do more if | |
1836 | * this is part of a write sequence for a layered image. | |
1837 | */ | |
c47f9371 | 1838 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1839 | { |
37206ee5 | 1840 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1841 | obj_request_done_set(obj_request); |
1842 | } | |
1843 | ||
2761713d ID |
1844 | static void rbd_osd_call_callback(struct rbd_obj_request *obj_request) |
1845 | { | |
1846 | dout("%s: obj %p\n", __func__, obj_request); | |
1847 | ||
1848 | if (obj_request_img_data_test(obj_request)) | |
1849 | rbd_osd_copyup_callback(obj_request); | |
1850 | else | |
1851 | obj_request_done_set(obj_request); | |
1852 | } | |
1853 | ||
85e084fe | 1854 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req) |
bf0d5f50 AE |
1855 | { |
1856 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1857 | u16 opcode; |
1858 | ||
85e084fe | 1859 | dout("%s: osd_req %p\n", __func__, osd_req); |
bf0d5f50 | 1860 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1861 | if (obj_request_img_data_test(obj_request)) { |
1862 | rbd_assert(obj_request->img_request); | |
1863 | rbd_assert(obj_request->which != BAD_WHICH); | |
1864 | } else { | |
1865 | rbd_assert(obj_request->which == BAD_WHICH); | |
1866 | } | |
bf0d5f50 | 1867 | |
1b83bef2 SW |
1868 | if (osd_req->r_result < 0) |
1869 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1870 | |
c47f9371 AE |
1871 | /* |
1872 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1873 | * passed to the block layer, which just supports a 32-bit |
1874 | * length field. | |
c47f9371 | 1875 | */ |
7665d85b | 1876 | obj_request->xferred = osd_req->r_ops[0].outdata_len; |
8b3e1a56 | 1877 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1878 | |
79528734 | 1879 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1880 | switch (opcode) { |
1881 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1882 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1883 | break; |
0ccd5926 | 1884 | case CEPH_OSD_OP_SETALLOCHINT: |
e30b7577 ID |
1885 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE || |
1886 | osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL); | |
0ccd5926 | 1887 | /* fall through */ |
bf0d5f50 | 1888 | case CEPH_OSD_OP_WRITE: |
e30b7577 | 1889 | case CEPH_OSD_OP_WRITEFULL: |
c47f9371 | 1890 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1891 | break; |
fbfab539 | 1892 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1893 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1894 | break; |
90e98c52 GZ |
1895 | case CEPH_OSD_OP_DELETE: |
1896 | case CEPH_OSD_OP_TRUNCATE: | |
1897 | case CEPH_OSD_OP_ZERO: | |
1898 | rbd_osd_discard_callback(obj_request); | |
1899 | break; | |
36be9a76 | 1900 | case CEPH_OSD_OP_CALL: |
2761713d ID |
1901 | rbd_osd_call_callback(obj_request); |
1902 | break; | |
bf0d5f50 | 1903 | default: |
a90bb0c1 ID |
1904 | rbd_warn(NULL, "unexpected OSD op: object_no %016llx opcode %d", |
1905 | obj_request->object_no, opcode); | |
bf0d5f50 AE |
1906 | break; |
1907 | } | |
1908 | ||
07741308 | 1909 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1910 | rbd_obj_request_complete(obj_request); |
1911 | } | |
1912 | ||
9d4df01f | 1913 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 | 1914 | { |
8c042b0d | 1915 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
430c28c3 | 1916 | |
7c84883a ID |
1917 | rbd_assert(obj_request_img_data_test(obj_request)); |
1918 | osd_req->r_snapid = obj_request->img_request->snap_id; | |
9d4df01f AE |
1919 | } |
1920 | ||
1921 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1922 | { | |
9d4df01f | 1923 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1924 | |
bb873b53 ID |
1925 | osd_req->r_mtime = CURRENT_TIME; |
1926 | osd_req->r_data_offset = obj_request->offset; | |
430c28c3 AE |
1927 | } |
1928 | ||
bc81207e ID |
1929 | static struct ceph_osd_request * |
1930 | __rbd_osd_req_create(struct rbd_device *rbd_dev, | |
1931 | struct ceph_snap_context *snapc, | |
1932 | int num_ops, unsigned int flags, | |
1933 | struct rbd_obj_request *obj_request) | |
1934 | { | |
1935 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
1936 | struct ceph_osd_request *req; | |
a90bb0c1 ID |
1937 | const char *name_format = rbd_dev->image_format == 1 ? |
1938 | RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT; | |
bc81207e ID |
1939 | |
1940 | req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO); | |
1941 | if (!req) | |
1942 | return NULL; | |
1943 | ||
1944 | req->r_flags = flags; | |
1945 | req->r_callback = rbd_osd_req_callback; | |
1946 | req->r_priv = obj_request; | |
1947 | ||
1948 | req->r_base_oloc.pool = rbd_dev->layout.pool_id; | |
a90bb0c1 ID |
1949 | if (ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format, |
1950 | rbd_dev->header.object_prefix, obj_request->object_no)) | |
bc81207e ID |
1951 | goto err_req; |
1952 | ||
1953 | if (ceph_osdc_alloc_messages(req, GFP_NOIO)) | |
1954 | goto err_req; | |
1955 | ||
1956 | return req; | |
1957 | ||
1958 | err_req: | |
1959 | ceph_osdc_put_request(req); | |
1960 | return NULL; | |
1961 | } | |
1962 | ||
0ccd5926 ID |
1963 | /* |
1964 | * Create an osd request. A read request has one osd op (read). | |
1965 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1966 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1967 | * technically osd watch is a write request, hence this distinction.) | |
1968 | */ | |
bf0d5f50 AE |
1969 | static struct ceph_osd_request *rbd_osd_req_create( |
1970 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1971 | enum obj_operation_type op_type, |
deb236b3 | 1972 | unsigned int num_ops, |
430c28c3 | 1973 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1974 | { |
bf0d5f50 | 1975 | struct ceph_snap_context *snapc = NULL; |
bf0d5f50 | 1976 | |
90e98c52 GZ |
1977 | if (obj_request_img_data_test(obj_request) && |
1978 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1979 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1980 | if (op_type == OBJ_OP_WRITE) { |
1981 | rbd_assert(img_request_write_test(img_request)); | |
1982 | } else { | |
1983 | rbd_assert(img_request_discard_test(img_request)); | |
1984 | } | |
6d2940c8 | 1985 | snapc = img_request->snapc; |
bf0d5f50 AE |
1986 | } |
1987 | ||
6d2940c8 | 1988 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 | 1989 | |
bc81207e ID |
1990 | return __rbd_osd_req_create(rbd_dev, snapc, num_ops, |
1991 | (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) ? | |
54ea0046 | 1992 | CEPH_OSD_FLAG_WRITE : CEPH_OSD_FLAG_READ, obj_request); |
bf0d5f50 AE |
1993 | } |
1994 | ||
0eefd470 | 1995 | /* |
d3246fb0 JD |
1996 | * Create a copyup osd request based on the information in the object |
1997 | * request supplied. A copyup request has two or three osd ops, a | |
1998 | * copyup method call, potentially a hint op, and a write or truncate | |
1999 | * or zero op. | |
0eefd470 AE |
2000 | */ |
2001 | static struct ceph_osd_request * | |
2002 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
2003 | { | |
2004 | struct rbd_img_request *img_request; | |
d3246fb0 | 2005 | int num_osd_ops = 3; |
0eefd470 AE |
2006 | |
2007 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2008 | img_request = obj_request->img_request; | |
2009 | rbd_assert(img_request); | |
d3246fb0 JD |
2010 | rbd_assert(img_request_write_test(img_request) || |
2011 | img_request_discard_test(img_request)); | |
0eefd470 | 2012 | |
d3246fb0 JD |
2013 | if (img_request_discard_test(img_request)) |
2014 | num_osd_ops = 2; | |
2015 | ||
bc81207e ID |
2016 | return __rbd_osd_req_create(img_request->rbd_dev, |
2017 | img_request->snapc, num_osd_ops, | |
54ea0046 | 2018 | CEPH_OSD_FLAG_WRITE, obj_request); |
0eefd470 AE |
2019 | } |
2020 | ||
bf0d5f50 AE |
2021 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2022 | { | |
2023 | ceph_osdc_put_request(osd_req); | |
2024 | } | |
2025 | ||
6c696d85 ID |
2026 | static struct rbd_obj_request * |
2027 | rbd_obj_request_create(enum obj_request_type type) | |
bf0d5f50 AE |
2028 | { |
2029 | struct rbd_obj_request *obj_request; | |
bf0d5f50 AE |
2030 | |
2031 | rbd_assert(obj_request_type_valid(type)); | |
2032 | ||
5a60e876 | 2033 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
6c696d85 | 2034 | if (!obj_request) |
f907ad55 | 2035 | return NULL; |
f907ad55 | 2036 | |
bf0d5f50 AE |
2037 | obj_request->which = BAD_WHICH; |
2038 | obj_request->type = type; | |
2039 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2040 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2041 | kref_init(&obj_request->kref); |
2042 | ||
67e2b652 | 2043 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
2044 | return obj_request; |
2045 | } | |
2046 | ||
2047 | static void rbd_obj_request_destroy(struct kref *kref) | |
2048 | { | |
2049 | struct rbd_obj_request *obj_request; | |
2050 | ||
2051 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2052 | ||
37206ee5 AE |
2053 | dout("%s: obj %p\n", __func__, obj_request); |
2054 | ||
bf0d5f50 AE |
2055 | rbd_assert(obj_request->img_request == NULL); |
2056 | rbd_assert(obj_request->which == BAD_WHICH); | |
2057 | ||
2058 | if (obj_request->osd_req) | |
2059 | rbd_osd_req_destroy(obj_request->osd_req); | |
2060 | ||
2061 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2062 | switch (obj_request->type) { | |
9969ebc5 AE |
2063 | case OBJ_REQUEST_NODATA: |
2064 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2065 | case OBJ_REQUEST_BIO: |
2066 | if (obj_request->bio_list) | |
2067 | bio_chain_put(obj_request->bio_list); | |
2068 | break; | |
788e2df3 | 2069 | case OBJ_REQUEST_PAGES: |
04dc923c ID |
2070 | /* img_data requests don't own their page array */ |
2071 | if (obj_request->pages && | |
2072 | !obj_request_img_data_test(obj_request)) | |
788e2df3 AE |
2073 | ceph_release_page_vector(obj_request->pages, |
2074 | obj_request->page_count); | |
2075 | break; | |
bf0d5f50 AE |
2076 | } |
2077 | ||
868311b1 | 2078 | kmem_cache_free(rbd_obj_request_cache, obj_request); |
bf0d5f50 AE |
2079 | } |
2080 | ||
fb65d228 AE |
2081 | /* It's OK to call this for a device with no parent */ |
2082 | ||
2083 | static void rbd_spec_put(struct rbd_spec *spec); | |
2084 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2085 | { | |
2086 | rbd_dev_remove_parent(rbd_dev); | |
2087 | rbd_spec_put(rbd_dev->parent_spec); | |
2088 | rbd_dev->parent_spec = NULL; | |
2089 | rbd_dev->parent_overlap = 0; | |
2090 | } | |
2091 | ||
a2acd00e AE |
2092 | /* |
2093 | * Parent image reference counting is used to determine when an | |
2094 | * image's parent fields can be safely torn down--after there are no | |
2095 | * more in-flight requests to the parent image. When the last | |
2096 | * reference is dropped, cleaning them up is safe. | |
2097 | */ | |
2098 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2099 | { | |
2100 | int counter; | |
2101 | ||
2102 | if (!rbd_dev->parent_spec) | |
2103 | return; | |
2104 | ||
2105 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2106 | if (counter > 0) | |
2107 | return; | |
2108 | ||
2109 | /* Last reference; clean up parent data structures */ | |
2110 | ||
2111 | if (!counter) | |
2112 | rbd_dev_unparent(rbd_dev); | |
2113 | else | |
9584d508 | 2114 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2115 | } |
2116 | ||
2117 | /* | |
2118 | * If an image has a non-zero parent overlap, get a reference to its | |
2119 | * parent. | |
2120 | * | |
2121 | * Returns true if the rbd device has a parent with a non-zero | |
2122 | * overlap and a reference for it was successfully taken, or | |
2123 | * false otherwise. | |
2124 | */ | |
2125 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2126 | { | |
ae43e9d0 | 2127 | int counter = 0; |
a2acd00e AE |
2128 | |
2129 | if (!rbd_dev->parent_spec) | |
2130 | return false; | |
2131 | ||
ae43e9d0 ID |
2132 | down_read(&rbd_dev->header_rwsem); |
2133 | if (rbd_dev->parent_overlap) | |
2134 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2135 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2136 | |
2137 | if (counter < 0) | |
9584d508 | 2138 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2139 | |
ae43e9d0 | 2140 | return counter > 0; |
a2acd00e AE |
2141 | } |
2142 | ||
bf0d5f50 AE |
2143 | /* |
2144 | * Caller is responsible for filling in the list of object requests | |
2145 | * that comprises the image request, and the Linux request pointer | |
2146 | * (if there is one). | |
2147 | */ | |
cc344fa1 AE |
2148 | static struct rbd_img_request *rbd_img_request_create( |
2149 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2150 | u64 offset, u64 length, |
6d2940c8 | 2151 | enum obj_operation_type op_type, |
4e752f0a | 2152 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2153 | { |
2154 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2155 | |
7a716aac | 2156 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2157 | if (!img_request) |
2158 | return NULL; | |
2159 | ||
bf0d5f50 AE |
2160 | img_request->rq = NULL; |
2161 | img_request->rbd_dev = rbd_dev; | |
2162 | img_request->offset = offset; | |
2163 | img_request->length = length; | |
0c425248 | 2164 | img_request->flags = 0; |
90e98c52 GZ |
2165 | if (op_type == OBJ_OP_DISCARD) { |
2166 | img_request_discard_set(img_request); | |
2167 | img_request->snapc = snapc; | |
2168 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2169 | img_request_write_set(img_request); |
4e752f0a | 2170 | img_request->snapc = snapc; |
0c425248 | 2171 | } else { |
bf0d5f50 | 2172 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2173 | } |
a2acd00e | 2174 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2175 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2176 | spin_lock_init(&img_request->completion_lock); |
2177 | img_request->next_completion = 0; | |
2178 | img_request->callback = NULL; | |
a5a337d4 | 2179 | img_request->result = 0; |
bf0d5f50 AE |
2180 | img_request->obj_request_count = 0; |
2181 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2182 | kref_init(&img_request->kref); | |
2183 | ||
37206ee5 | 2184 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2185 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2186 | |
bf0d5f50 AE |
2187 | return img_request; |
2188 | } | |
2189 | ||
2190 | static void rbd_img_request_destroy(struct kref *kref) | |
2191 | { | |
2192 | struct rbd_img_request *img_request; | |
2193 | struct rbd_obj_request *obj_request; | |
2194 | struct rbd_obj_request *next_obj_request; | |
2195 | ||
2196 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2197 | ||
37206ee5 AE |
2198 | dout("%s: img %p\n", __func__, img_request); |
2199 | ||
bf0d5f50 AE |
2200 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2201 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2202 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2203 | |
a2acd00e AE |
2204 | if (img_request_layered_test(img_request)) { |
2205 | img_request_layered_clear(img_request); | |
2206 | rbd_dev_parent_put(img_request->rbd_dev); | |
2207 | } | |
2208 | ||
bef95455 JD |
2209 | if (img_request_write_test(img_request) || |
2210 | img_request_discard_test(img_request)) | |
812164f8 | 2211 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2212 | |
1c2a9dfe | 2213 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2214 | } |
2215 | ||
e93f3152 AE |
2216 | static struct rbd_img_request *rbd_parent_request_create( |
2217 | struct rbd_obj_request *obj_request, | |
2218 | u64 img_offset, u64 length) | |
2219 | { | |
2220 | struct rbd_img_request *parent_request; | |
2221 | struct rbd_device *rbd_dev; | |
2222 | ||
2223 | rbd_assert(obj_request->img_request); | |
2224 | rbd_dev = obj_request->img_request->rbd_dev; | |
2225 | ||
4e752f0a | 2226 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2227 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2228 | if (!parent_request) |
2229 | return NULL; | |
2230 | ||
2231 | img_request_child_set(parent_request); | |
2232 | rbd_obj_request_get(obj_request); | |
2233 | parent_request->obj_request = obj_request; | |
2234 | ||
2235 | return parent_request; | |
2236 | } | |
2237 | ||
2238 | static void rbd_parent_request_destroy(struct kref *kref) | |
2239 | { | |
2240 | struct rbd_img_request *parent_request; | |
2241 | struct rbd_obj_request *orig_request; | |
2242 | ||
2243 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2244 | orig_request = parent_request->obj_request; | |
2245 | ||
2246 | parent_request->obj_request = NULL; | |
2247 | rbd_obj_request_put(orig_request); | |
2248 | img_request_child_clear(parent_request); | |
2249 | ||
2250 | rbd_img_request_destroy(kref); | |
2251 | } | |
2252 | ||
1217857f AE |
2253 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2254 | { | |
6365d33a | 2255 | struct rbd_img_request *img_request; |
1217857f AE |
2256 | unsigned int xferred; |
2257 | int result; | |
8b3e1a56 | 2258 | bool more; |
1217857f | 2259 | |
6365d33a AE |
2260 | rbd_assert(obj_request_img_data_test(obj_request)); |
2261 | img_request = obj_request->img_request; | |
2262 | ||
1217857f AE |
2263 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2264 | xferred = (unsigned int)obj_request->xferred; | |
2265 | result = obj_request->result; | |
2266 | if (result) { | |
2267 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2268 | enum obj_operation_type op_type; |
2269 | ||
90e98c52 GZ |
2270 | if (img_request_discard_test(img_request)) |
2271 | op_type = OBJ_OP_DISCARD; | |
2272 | else if (img_request_write_test(img_request)) | |
2273 | op_type = OBJ_OP_WRITE; | |
2274 | else | |
2275 | op_type = OBJ_OP_READ; | |
1217857f | 2276 | |
9584d508 | 2277 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2278 | obj_op_name(op_type), obj_request->length, |
2279 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2280 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2281 | result, xferred); |
2282 | if (!img_request->result) | |
2283 | img_request->result = result; | |
082a75da ID |
2284 | /* |
2285 | * Need to end I/O on the entire obj_request worth of | |
2286 | * bytes in case of error. | |
2287 | */ | |
2288 | xferred = obj_request->length; | |
1217857f AE |
2289 | } |
2290 | ||
8b3e1a56 AE |
2291 | if (img_request_child_test(img_request)) { |
2292 | rbd_assert(img_request->obj_request != NULL); | |
2293 | more = obj_request->which < img_request->obj_request_count - 1; | |
2294 | } else { | |
2295 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2296 | |
2297 | more = blk_update_request(img_request->rq, result, xferred); | |
2298 | if (!more) | |
2299 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2300 | } |
2301 | ||
2302 | return more; | |
1217857f AE |
2303 | } |
2304 | ||
2169238d AE |
2305 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2306 | { | |
2307 | struct rbd_img_request *img_request; | |
2308 | u32 which = obj_request->which; | |
2309 | bool more = true; | |
2310 | ||
6365d33a | 2311 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2312 | img_request = obj_request->img_request; |
2313 | ||
2314 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2315 | rbd_assert(img_request != NULL); | |
2169238d AE |
2316 | rbd_assert(img_request->obj_request_count > 0); |
2317 | rbd_assert(which != BAD_WHICH); | |
2318 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2319 | |
2320 | spin_lock_irq(&img_request->completion_lock); | |
2321 | if (which != img_request->next_completion) | |
2322 | goto out; | |
2323 | ||
2324 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2325 | rbd_assert(more); |
2326 | rbd_assert(which < img_request->obj_request_count); | |
2327 | ||
2328 | if (!obj_request_done_test(obj_request)) | |
2329 | break; | |
1217857f | 2330 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2331 | which++; |
2332 | } | |
2333 | ||
2334 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2335 | img_request->next_completion = which; | |
2336 | out: | |
2337 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2338 | rbd_img_request_put(img_request); |
2169238d AE |
2339 | |
2340 | if (!more) | |
2341 | rbd_img_request_complete(img_request); | |
2342 | } | |
2343 | ||
3b434a2a JD |
2344 | /* |
2345 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2346 | * them for submission. num_ops is the current number of | |
2347 | * osd operations already to the object request. | |
2348 | */ | |
2349 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2350 | struct ceph_osd_request *osd_request, | |
2351 | enum obj_operation_type op_type, | |
2352 | unsigned int num_ops) | |
2353 | { | |
2354 | struct rbd_img_request *img_request = obj_request->img_request; | |
2355 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2356 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2357 | u64 offset = obj_request->offset; | |
2358 | u64 length = obj_request->length; | |
2359 | u64 img_end; | |
2360 | u16 opcode; | |
2361 | ||
2362 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2363 | if (!offset && length == object_size && |
2364 | (!img_request_layered_test(img_request) || | |
2365 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2366 | opcode = CEPH_OSD_OP_DELETE; |
2367 | } else if ((offset + length == object_size)) { | |
2368 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2369 | } else { | |
2370 | down_read(&rbd_dev->header_rwsem); | |
2371 | img_end = rbd_dev->header.image_size; | |
2372 | up_read(&rbd_dev->header_rwsem); | |
2373 | ||
2374 | if (obj_request->img_offset + length == img_end) | |
2375 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2376 | else | |
2377 | opcode = CEPH_OSD_OP_ZERO; | |
2378 | } | |
2379 | } else if (op_type == OBJ_OP_WRITE) { | |
e30b7577 ID |
2380 | if (!offset && length == object_size) |
2381 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2382 | else | |
2383 | opcode = CEPH_OSD_OP_WRITE; | |
3b434a2a JD |
2384 | osd_req_op_alloc_hint_init(osd_request, num_ops, |
2385 | object_size, object_size); | |
2386 | num_ops++; | |
2387 | } else { | |
2388 | opcode = CEPH_OSD_OP_READ; | |
2389 | } | |
2390 | ||
7e868b6e | 2391 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2392 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2393 | else |
2394 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2395 | offset, length, 0, 0); | |
2396 | ||
3b434a2a JD |
2397 | if (obj_request->type == OBJ_REQUEST_BIO) |
2398 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2399 | obj_request->bio_list, length); | |
2400 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2401 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2402 | obj_request->pages, length, | |
2403 | offset & ~PAGE_MASK, false, false); | |
2404 | ||
2405 | /* Discards are also writes */ | |
2406 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2407 | rbd_osd_req_format_write(obj_request); | |
2408 | else | |
2409 | rbd_osd_req_format_read(obj_request); | |
2410 | } | |
2411 | ||
f1a4739f AE |
2412 | /* |
2413 | * Split up an image request into one or more object requests, each | |
2414 | * to a different object. The "type" parameter indicates whether | |
2415 | * "data_desc" is the pointer to the head of a list of bio | |
2416 | * structures, or the base of a page array. In either case this | |
2417 | * function assumes data_desc describes memory sufficient to hold | |
2418 | * all data described by the image request. | |
2419 | */ | |
2420 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2421 | enum obj_request_type type, | |
2422 | void *data_desc) | |
bf0d5f50 AE |
2423 | { |
2424 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2425 | struct rbd_obj_request *obj_request = NULL; | |
2426 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2427 | struct bio *bio_list = NULL; |
f1a4739f | 2428 | unsigned int bio_offset = 0; |
a158073c | 2429 | struct page **pages = NULL; |
6d2940c8 | 2430 | enum obj_operation_type op_type; |
7da22d29 | 2431 | u64 img_offset; |
bf0d5f50 | 2432 | u64 resid; |
bf0d5f50 | 2433 | |
f1a4739f AE |
2434 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2435 | (int)type, data_desc); | |
37206ee5 | 2436 | |
7da22d29 | 2437 | img_offset = img_request->offset; |
bf0d5f50 | 2438 | resid = img_request->length; |
4dda41d3 | 2439 | rbd_assert(resid > 0); |
3b434a2a | 2440 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2441 | |
2442 | if (type == OBJ_REQUEST_BIO) { | |
2443 | bio_list = data_desc; | |
4f024f37 KO |
2444 | rbd_assert(img_offset == |
2445 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2446 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2447 | pages = data_desc; |
2448 | } | |
2449 | ||
bf0d5f50 | 2450 | while (resid) { |
2fa12320 | 2451 | struct ceph_osd_request *osd_req; |
a90bb0c1 | 2452 | u64 object_no = img_offset >> rbd_dev->header.obj_order; |
67e2b652 ID |
2453 | u64 offset = rbd_segment_offset(rbd_dev, img_offset); |
2454 | u64 length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2455 | |
6c696d85 | 2456 | obj_request = rbd_obj_request_create(type); |
bf0d5f50 AE |
2457 | if (!obj_request) |
2458 | goto out_unwind; | |
62054da6 | 2459 | |
a90bb0c1 | 2460 | obj_request->object_no = object_no; |
67e2b652 ID |
2461 | obj_request->offset = offset; |
2462 | obj_request->length = length; | |
2463 | ||
03507db6 JD |
2464 | /* |
2465 | * set obj_request->img_request before creating the | |
2466 | * osd_request so that it gets the right snapc | |
2467 | */ | |
2468 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2469 | |
f1a4739f AE |
2470 | if (type == OBJ_REQUEST_BIO) { |
2471 | unsigned int clone_size; | |
2472 | ||
2473 | rbd_assert(length <= (u64)UINT_MAX); | |
2474 | clone_size = (unsigned int)length; | |
2475 | obj_request->bio_list = | |
2476 | bio_chain_clone_range(&bio_list, | |
2477 | &bio_offset, | |
2478 | clone_size, | |
2224d879 | 2479 | GFP_NOIO); |
f1a4739f | 2480 | if (!obj_request->bio_list) |
62054da6 | 2481 | goto out_unwind; |
90e98c52 | 2482 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2483 | unsigned int page_count; |
2484 | ||
2485 | obj_request->pages = pages; | |
2486 | page_count = (u32)calc_pages_for(offset, length); | |
2487 | obj_request->page_count = page_count; | |
2488 | if ((offset + length) & ~PAGE_MASK) | |
2489 | page_count--; /* more on last page */ | |
2490 | pages += page_count; | |
2491 | } | |
bf0d5f50 | 2492 | |
6d2940c8 GZ |
2493 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2494 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2495 | obj_request); | |
2fa12320 | 2496 | if (!osd_req) |
62054da6 | 2497 | goto out_unwind; |
3b434a2a | 2498 | |
2fa12320 | 2499 | obj_request->osd_req = osd_req; |
2169238d | 2500 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2501 | obj_request->img_offset = img_offset; |
9d4df01f | 2502 | |
3b434a2a | 2503 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2504 | |
7da22d29 | 2505 | img_offset += length; |
bf0d5f50 AE |
2506 | resid -= length; |
2507 | } | |
2508 | ||
2509 | return 0; | |
2510 | ||
bf0d5f50 AE |
2511 | out_unwind: |
2512 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2513 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2514 | |
2515 | return -ENOMEM; | |
2516 | } | |
2517 | ||
0eefd470 | 2518 | static void |
2761713d | 2519 | rbd_osd_copyup_callback(struct rbd_obj_request *obj_request) |
0eefd470 AE |
2520 | { |
2521 | struct rbd_img_request *img_request; | |
2522 | struct rbd_device *rbd_dev; | |
ebda6408 | 2523 | struct page **pages; |
0eefd470 AE |
2524 | u32 page_count; |
2525 | ||
2761713d ID |
2526 | dout("%s: obj %p\n", __func__, obj_request); |
2527 | ||
d3246fb0 JD |
2528 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2529 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2530 | rbd_assert(obj_request_img_data_test(obj_request)); |
2531 | img_request = obj_request->img_request; | |
2532 | rbd_assert(img_request); | |
2533 | ||
2534 | rbd_dev = img_request->rbd_dev; | |
2535 | rbd_assert(rbd_dev); | |
0eefd470 | 2536 | |
ebda6408 AE |
2537 | pages = obj_request->copyup_pages; |
2538 | rbd_assert(pages != NULL); | |
0eefd470 | 2539 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2540 | page_count = obj_request->copyup_page_count; |
2541 | rbd_assert(page_count); | |
2542 | obj_request->copyup_page_count = 0; | |
2543 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2544 | |
2545 | /* | |
2546 | * We want the transfer count to reflect the size of the | |
2547 | * original write request. There is no such thing as a | |
2548 | * successful short write, so if the request was successful | |
2549 | * we can just set it to the originally-requested length. | |
2550 | */ | |
2551 | if (!obj_request->result) | |
2552 | obj_request->xferred = obj_request->length; | |
2553 | ||
2761713d | 2554 | obj_request_done_set(obj_request); |
0eefd470 AE |
2555 | } |
2556 | ||
3d7efd18 AE |
2557 | static void |
2558 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2559 | { | |
2560 | struct rbd_obj_request *orig_request; | |
0eefd470 | 2561 | struct ceph_osd_request *osd_req; |
0eefd470 | 2562 | struct rbd_device *rbd_dev; |
3d7efd18 | 2563 | struct page **pages; |
d3246fb0 | 2564 | enum obj_operation_type op_type; |
ebda6408 | 2565 | u32 page_count; |
bbea1c1a | 2566 | int img_result; |
ebda6408 | 2567 | u64 parent_length; |
3d7efd18 AE |
2568 | |
2569 | rbd_assert(img_request_child_test(img_request)); | |
2570 | ||
2571 | /* First get what we need from the image request */ | |
2572 | ||
2573 | pages = img_request->copyup_pages; | |
2574 | rbd_assert(pages != NULL); | |
2575 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2576 | page_count = img_request->copyup_page_count; |
2577 | rbd_assert(page_count); | |
2578 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2579 | |
2580 | orig_request = img_request->obj_request; | |
2581 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2582 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2583 | img_result = img_request->result; |
ebda6408 | 2584 | parent_length = img_request->length; |
fa355112 | 2585 | rbd_assert(img_result || parent_length == img_request->xferred); |
91c6febb | 2586 | rbd_img_request_put(img_request); |
3d7efd18 | 2587 | |
91c6febb AE |
2588 | rbd_assert(orig_request->img_request); |
2589 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2590 | rbd_assert(rbd_dev); |
0eefd470 | 2591 | |
bbea1c1a AE |
2592 | /* |
2593 | * If the overlap has become 0 (most likely because the | |
2594 | * image has been flattened) we need to free the pages | |
2595 | * and re-submit the original write request. | |
2596 | */ | |
2597 | if (!rbd_dev->parent_overlap) { | |
bbea1c1a | 2598 | ceph_release_page_vector(pages, page_count); |
980917fc ID |
2599 | rbd_obj_request_submit(orig_request); |
2600 | return; | |
bbea1c1a | 2601 | } |
0eefd470 | 2602 | |
bbea1c1a | 2603 | if (img_result) |
0eefd470 | 2604 | goto out_err; |
0eefd470 | 2605 | |
8785b1d4 AE |
2606 | /* |
2607 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2608 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2609 | * request. Allocate the new copyup osd request for the |
2610 | * original request, and release the old one. | |
2611 | */ | |
bbea1c1a | 2612 | img_result = -ENOMEM; |
0eefd470 AE |
2613 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2614 | if (!osd_req) | |
2615 | goto out_err; | |
8785b1d4 | 2616 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2617 | orig_request->osd_req = osd_req; |
2618 | orig_request->copyup_pages = pages; | |
ebda6408 | 2619 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2620 | |
0eefd470 | 2621 | /* Initialize the copyup op */ |
3d7efd18 | 2622 | |
0eefd470 | 2623 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2624 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2625 | false, false); |
3d7efd18 | 2626 | |
d3246fb0 | 2627 | /* Add the other op(s) */ |
0eefd470 | 2628 | |
d3246fb0 JD |
2629 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2630 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2631 | |
2632 | /* All set, send it off. */ | |
2633 | ||
980917fc ID |
2634 | rbd_obj_request_submit(orig_request); |
2635 | return; | |
0eefd470 | 2636 | |
0eefd470 | 2637 | out_err: |
fa355112 | 2638 | ceph_release_page_vector(pages, page_count); |
0dcc685e | 2639 | rbd_obj_request_error(orig_request, img_result); |
3d7efd18 AE |
2640 | } |
2641 | ||
2642 | /* | |
2643 | * Read from the parent image the range of data that covers the | |
2644 | * entire target of the given object request. This is used for | |
2645 | * satisfying a layered image write request when the target of an | |
2646 | * object request from the image request does not exist. | |
2647 | * | |
2648 | * A page array big enough to hold the returned data is allocated | |
2649 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2650 | * When the read completes, this page array will be transferred to | |
2651 | * the original object request for the copyup operation. | |
2652 | * | |
c2e82414 ID |
2653 | * If an error occurs, it is recorded as the result of the original |
2654 | * object request in rbd_img_obj_exists_callback(). | |
3d7efd18 AE |
2655 | */ |
2656 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2657 | { | |
058aa991 | 2658 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; |
3d7efd18 | 2659 | struct rbd_img_request *parent_request = NULL; |
3d7efd18 AE |
2660 | u64 img_offset; |
2661 | u64 length; | |
2662 | struct page **pages = NULL; | |
2663 | u32 page_count; | |
2664 | int result; | |
2665 | ||
3d7efd18 AE |
2666 | rbd_assert(rbd_dev->parent != NULL); |
2667 | ||
2668 | /* | |
2669 | * Determine the byte range covered by the object in the | |
2670 | * child image to which the original request was to be sent. | |
2671 | */ | |
2672 | img_offset = obj_request->img_offset - obj_request->offset; | |
5bc3fb17 | 2673 | length = rbd_obj_bytes(&rbd_dev->header); |
3d7efd18 | 2674 | |
a9e8ba2c AE |
2675 | /* |
2676 | * There is no defined parent data beyond the parent | |
2677 | * overlap, so limit what we read at that boundary if | |
2678 | * necessary. | |
2679 | */ | |
2680 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2681 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2682 | length = rbd_dev->parent_overlap - img_offset; | |
2683 | } | |
2684 | ||
3d7efd18 AE |
2685 | /* |
2686 | * Allocate a page array big enough to receive the data read | |
2687 | * from the parent. | |
2688 | */ | |
2689 | page_count = (u32)calc_pages_for(0, length); | |
2690 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2691 | if (IS_ERR(pages)) { | |
2692 | result = PTR_ERR(pages); | |
2693 | pages = NULL; | |
2694 | goto out_err; | |
2695 | } | |
2696 | ||
2697 | result = -ENOMEM; | |
e93f3152 AE |
2698 | parent_request = rbd_parent_request_create(obj_request, |
2699 | img_offset, length); | |
3d7efd18 AE |
2700 | if (!parent_request) |
2701 | goto out_err; | |
3d7efd18 AE |
2702 | |
2703 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2704 | if (result) | |
2705 | goto out_err; | |
058aa991 | 2706 | |
3d7efd18 | 2707 | parent_request->copyup_pages = pages; |
ebda6408 | 2708 | parent_request->copyup_page_count = page_count; |
3d7efd18 | 2709 | parent_request->callback = rbd_img_obj_parent_read_full_callback; |
058aa991 | 2710 | |
3d7efd18 AE |
2711 | result = rbd_img_request_submit(parent_request); |
2712 | if (!result) | |
2713 | return 0; | |
2714 | ||
2715 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2716 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2717 | parent_request->obj_request = NULL; |
2718 | rbd_obj_request_put(obj_request); | |
2719 | out_err: | |
2720 | if (pages) | |
2721 | ceph_release_page_vector(pages, page_count); | |
2722 | if (parent_request) | |
2723 | rbd_img_request_put(parent_request); | |
3d7efd18 AE |
2724 | return result; |
2725 | } | |
2726 | ||
c5b5ef6c AE |
2727 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2728 | { | |
c5b5ef6c | 2729 | struct rbd_obj_request *orig_request; |
638f5abe | 2730 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2731 | int result; |
2732 | ||
2733 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2734 | ||
2735 | /* | |
2736 | * All we need from the object request is the original | |
2737 | * request and the result of the STAT op. Grab those, then | |
2738 | * we're done with the request. | |
2739 | */ | |
2740 | orig_request = obj_request->obj_request; | |
2741 | obj_request->obj_request = NULL; | |
912c317d | 2742 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2743 | rbd_assert(orig_request); |
2744 | rbd_assert(orig_request->img_request); | |
2745 | ||
2746 | result = obj_request->result; | |
2747 | obj_request->result = 0; | |
2748 | ||
2749 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2750 | obj_request, orig_request, result, | |
2751 | obj_request->xferred, obj_request->length); | |
2752 | rbd_obj_request_put(obj_request); | |
2753 | ||
638f5abe AE |
2754 | /* |
2755 | * If the overlap has become 0 (most likely because the | |
980917fc ID |
2756 | * image has been flattened) we need to re-submit the |
2757 | * original request. | |
638f5abe AE |
2758 | */ |
2759 | rbd_dev = orig_request->img_request->rbd_dev; | |
2760 | if (!rbd_dev->parent_overlap) { | |
980917fc ID |
2761 | rbd_obj_request_submit(orig_request); |
2762 | return; | |
638f5abe | 2763 | } |
c5b5ef6c AE |
2764 | |
2765 | /* | |
2766 | * Our only purpose here is to determine whether the object | |
2767 | * exists, and we don't want to treat the non-existence as | |
2768 | * an error. If something else comes back, transfer the | |
2769 | * error to the original request and complete it now. | |
2770 | */ | |
2771 | if (!result) { | |
2772 | obj_request_existence_set(orig_request, true); | |
2773 | } else if (result == -ENOENT) { | |
2774 | obj_request_existence_set(orig_request, false); | |
c2e82414 ID |
2775 | } else { |
2776 | goto fail_orig_request; | |
c5b5ef6c AE |
2777 | } |
2778 | ||
2779 | /* | |
2780 | * Resubmit the original request now that we have recorded | |
2781 | * whether the target object exists. | |
2782 | */ | |
c2e82414 ID |
2783 | result = rbd_img_obj_request_submit(orig_request); |
2784 | if (result) | |
2785 | goto fail_orig_request; | |
2786 | ||
2787 | return; | |
2788 | ||
2789 | fail_orig_request: | |
0dcc685e | 2790 | rbd_obj_request_error(orig_request, result); |
c5b5ef6c AE |
2791 | } |
2792 | ||
2793 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2794 | { | |
058aa991 | 2795 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; |
c5b5ef6c | 2796 | struct rbd_obj_request *stat_request; |
710214e3 | 2797 | struct page **pages; |
c5b5ef6c AE |
2798 | u32 page_count; |
2799 | size_t size; | |
2800 | int ret; | |
2801 | ||
6c696d85 | 2802 | stat_request = rbd_obj_request_create(OBJ_REQUEST_PAGES); |
710214e3 ID |
2803 | if (!stat_request) |
2804 | return -ENOMEM; | |
2805 | ||
a90bb0c1 ID |
2806 | stat_request->object_no = obj_request->object_no; |
2807 | ||
710214e3 ID |
2808 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
2809 | stat_request); | |
2810 | if (!stat_request->osd_req) { | |
2811 | ret = -ENOMEM; | |
2812 | goto fail_stat_request; | |
2813 | } | |
2814 | ||
c5b5ef6c AE |
2815 | /* |
2816 | * The response data for a STAT call consists of: | |
2817 | * le64 length; | |
2818 | * struct { | |
2819 | * le32 tv_sec; | |
2820 | * le32 tv_nsec; | |
2821 | * } mtime; | |
2822 | */ | |
2823 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2824 | page_count = (u32)calc_pages_for(0, size); | |
2825 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
710214e3 ID |
2826 | if (IS_ERR(pages)) { |
2827 | ret = PTR_ERR(pages); | |
2828 | goto fail_stat_request; | |
2829 | } | |
c5b5ef6c | 2830 | |
710214e3 ID |
2831 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
2832 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2833 | false, false); | |
c5b5ef6c AE |
2834 | |
2835 | rbd_obj_request_get(obj_request); | |
2836 | stat_request->obj_request = obj_request; | |
2837 | stat_request->pages = pages; | |
2838 | stat_request->page_count = page_count; | |
c5b5ef6c AE |
2839 | stat_request->callback = rbd_img_obj_exists_callback; |
2840 | ||
980917fc ID |
2841 | rbd_obj_request_submit(stat_request); |
2842 | return 0; | |
c5b5ef6c | 2843 | |
710214e3 ID |
2844 | fail_stat_request: |
2845 | rbd_obj_request_put(stat_request); | |
c5b5ef6c AE |
2846 | return ret; |
2847 | } | |
2848 | ||
70d045f6 | 2849 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d | 2850 | { |
058aa991 ID |
2851 | struct rbd_img_request *img_request = obj_request->img_request; |
2852 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
b454e36d | 2853 | |
70d045f6 | 2854 | /* Reads */ |
1c220881 JD |
2855 | if (!img_request_write_test(img_request) && |
2856 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2857 | return true; |
2858 | ||
2859 | /* Non-layered writes */ | |
2860 | if (!img_request_layered_test(img_request)) | |
2861 | return true; | |
2862 | ||
b454e36d | 2863 | /* |
70d045f6 ID |
2864 | * Layered writes outside of the parent overlap range don't |
2865 | * share any data with the parent. | |
b454e36d | 2866 | */ |
70d045f6 ID |
2867 | if (!obj_request_overlaps_parent(obj_request)) |
2868 | return true; | |
b454e36d | 2869 | |
c622d226 GZ |
2870 | /* |
2871 | * Entire-object layered writes - we will overwrite whatever | |
2872 | * parent data there is anyway. | |
2873 | */ | |
2874 | if (!obj_request->offset && | |
2875 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2876 | return true; | |
2877 | ||
70d045f6 ID |
2878 | /* |
2879 | * If the object is known to already exist, its parent data has | |
2880 | * already been copied. | |
2881 | */ | |
2882 | if (obj_request_known_test(obj_request) && | |
2883 | obj_request_exists_test(obj_request)) | |
2884 | return true; | |
2885 | ||
2886 | return false; | |
2887 | } | |
2888 | ||
2889 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2890 | { | |
058aa991 ID |
2891 | rbd_assert(obj_request_img_data_test(obj_request)); |
2892 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2893 | rbd_assert(obj_request->img_request); | |
b454e36d | 2894 | |
70d045f6 | 2895 | if (img_obj_request_simple(obj_request)) { |
980917fc ID |
2896 | rbd_obj_request_submit(obj_request); |
2897 | return 0; | |
b454e36d AE |
2898 | } |
2899 | ||
2900 | /* | |
3d7efd18 AE |
2901 | * It's a layered write. The target object might exist but |
2902 | * we may not know that yet. If we know it doesn't exist, | |
2903 | * start by reading the data for the full target object from | |
2904 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2905 | */ |
70d045f6 | 2906 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2907 | return rbd_img_obj_parent_read_full(obj_request); |
2908 | ||
2909 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2910 | |
2911 | return rbd_img_obj_exists_submit(obj_request); | |
2912 | } | |
2913 | ||
bf0d5f50 AE |
2914 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2915 | { | |
bf0d5f50 | 2916 | struct rbd_obj_request *obj_request; |
46faeed4 | 2917 | struct rbd_obj_request *next_obj_request; |
663ae2cc | 2918 | int ret = 0; |
bf0d5f50 | 2919 | |
37206ee5 | 2920 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 | 2921 | |
663ae2cc ID |
2922 | rbd_img_request_get(img_request); |
2923 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { | |
b454e36d | 2924 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 | 2925 | if (ret) |
663ae2cc | 2926 | goto out_put_ireq; |
bf0d5f50 AE |
2927 | } |
2928 | ||
663ae2cc ID |
2929 | out_put_ireq: |
2930 | rbd_img_request_put(img_request); | |
2931 | return ret; | |
bf0d5f50 | 2932 | } |
8b3e1a56 AE |
2933 | |
2934 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2935 | { | |
2936 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2937 | struct rbd_device *rbd_dev; |
2938 | u64 obj_end; | |
02c74fba AE |
2939 | u64 img_xferred; |
2940 | int img_result; | |
8b3e1a56 AE |
2941 | |
2942 | rbd_assert(img_request_child_test(img_request)); | |
2943 | ||
02c74fba AE |
2944 | /* First get what we need from the image request and release it */ |
2945 | ||
8b3e1a56 | 2946 | obj_request = img_request->obj_request; |
02c74fba AE |
2947 | img_xferred = img_request->xferred; |
2948 | img_result = img_request->result; | |
2949 | rbd_img_request_put(img_request); | |
2950 | ||
2951 | /* | |
2952 | * If the overlap has become 0 (most likely because the | |
2953 | * image has been flattened) we need to re-submit the | |
2954 | * original request. | |
2955 | */ | |
a9e8ba2c AE |
2956 | rbd_assert(obj_request); |
2957 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2958 | rbd_dev = obj_request->img_request->rbd_dev; |
2959 | if (!rbd_dev->parent_overlap) { | |
980917fc ID |
2960 | rbd_obj_request_submit(obj_request); |
2961 | return; | |
02c74fba | 2962 | } |
a9e8ba2c | 2963 | |
02c74fba | 2964 | obj_request->result = img_result; |
a9e8ba2c AE |
2965 | if (obj_request->result) |
2966 | goto out; | |
2967 | ||
2968 | /* | |
2969 | * We need to zero anything beyond the parent overlap | |
2970 | * boundary. Since rbd_img_obj_request_read_callback() | |
2971 | * will zero anything beyond the end of a short read, an | |
2972 | * easy way to do this is to pretend the data from the | |
2973 | * parent came up short--ending at the overlap boundary. | |
2974 | */ | |
2975 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2976 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2977 | if (obj_end > rbd_dev->parent_overlap) { |
2978 | u64 xferred = 0; | |
2979 | ||
2980 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2981 | xferred = rbd_dev->parent_overlap - | |
2982 | obj_request->img_offset; | |
8b3e1a56 | 2983 | |
02c74fba | 2984 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2985 | } else { |
02c74fba | 2986 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2987 | } |
2988 | out: | |
8b3e1a56 AE |
2989 | rbd_img_obj_request_read_callback(obj_request); |
2990 | rbd_obj_request_complete(obj_request); | |
2991 | } | |
2992 | ||
2993 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2994 | { | |
8b3e1a56 AE |
2995 | struct rbd_img_request *img_request; |
2996 | int result; | |
2997 | ||
2998 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2999 | rbd_assert(obj_request->img_request != NULL); | |
3000 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3001 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3002 | |
8b3e1a56 | 3003 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3004 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3005 | obj_request->img_offset, |
e93f3152 | 3006 | obj_request->length); |
8b3e1a56 AE |
3007 | result = -ENOMEM; |
3008 | if (!img_request) | |
3009 | goto out_err; | |
3010 | ||
5b2ab72d AE |
3011 | if (obj_request->type == OBJ_REQUEST_BIO) |
3012 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3013 | obj_request->bio_list); | |
3014 | else | |
3015 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3016 | obj_request->pages); | |
8b3e1a56 AE |
3017 | if (result) |
3018 | goto out_err; | |
3019 | ||
3020 | img_request->callback = rbd_img_parent_read_callback; | |
3021 | result = rbd_img_request_submit(img_request); | |
3022 | if (result) | |
3023 | goto out_err; | |
3024 | ||
3025 | return; | |
3026 | out_err: | |
3027 | if (img_request) | |
3028 | rbd_img_request_put(img_request); | |
3029 | obj_request->result = result; | |
3030 | obj_request->xferred = 0; | |
3031 | obj_request_done_set(obj_request); | |
3032 | } | |
bf0d5f50 | 3033 | |
ed95b21a | 3034 | static const struct rbd_client_id rbd_empty_cid; |
b8d70035 | 3035 | |
ed95b21a ID |
3036 | static bool rbd_cid_equal(const struct rbd_client_id *lhs, |
3037 | const struct rbd_client_id *rhs) | |
3038 | { | |
3039 | return lhs->gid == rhs->gid && lhs->handle == rhs->handle; | |
3040 | } | |
3041 | ||
3042 | static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev) | |
3043 | { | |
3044 | struct rbd_client_id cid; | |
3045 | ||
3046 | mutex_lock(&rbd_dev->watch_mutex); | |
3047 | cid.gid = ceph_client_gid(rbd_dev->rbd_client->client); | |
3048 | cid.handle = rbd_dev->watch_cookie; | |
3049 | mutex_unlock(&rbd_dev->watch_mutex); | |
3050 | return cid; | |
3051 | } | |
3052 | ||
3053 | /* | |
3054 | * lock_rwsem must be held for write | |
3055 | */ | |
3056 | static void rbd_set_owner_cid(struct rbd_device *rbd_dev, | |
3057 | const struct rbd_client_id *cid) | |
3058 | { | |
3059 | dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev, | |
3060 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle, | |
3061 | cid->gid, cid->handle); | |
3062 | rbd_dev->owner_cid = *cid; /* struct */ | |
3063 | } | |
3064 | ||
3065 | static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf) | |
3066 | { | |
3067 | mutex_lock(&rbd_dev->watch_mutex); | |
3068 | sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie); | |
3069 | mutex_unlock(&rbd_dev->watch_mutex); | |
3070 | } | |
3071 | ||
3072 | /* | |
3073 | * lock_rwsem must be held for write | |
3074 | */ | |
3075 | static int rbd_lock(struct rbd_device *rbd_dev) | |
b8d70035 | 3076 | { |
922dab61 | 3077 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
ed95b21a ID |
3078 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); |
3079 | char cookie[32]; | |
e627db08 | 3080 | int ret; |
b8d70035 | 3081 | |
ed95b21a | 3082 | WARN_ON(__rbd_is_lock_owner(rbd_dev)); |
52bb1f9b | 3083 | |
ed95b21a ID |
3084 | format_lock_cookie(rbd_dev, cookie); |
3085 | ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
3086 | RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie, | |
3087 | RBD_LOCK_TAG, "", 0); | |
e627db08 | 3088 | if (ret) |
ed95b21a | 3089 | return ret; |
b8d70035 | 3090 | |
ed95b21a ID |
3091 | rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED; |
3092 | rbd_set_owner_cid(rbd_dev, &cid); | |
3093 | queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work); | |
3094 | return 0; | |
b8d70035 AE |
3095 | } |
3096 | ||
ed95b21a ID |
3097 | /* |
3098 | * lock_rwsem must be held for write | |
3099 | */ | |
3100 | static int rbd_unlock(struct rbd_device *rbd_dev) | |
bb040aa0 | 3101 | { |
922dab61 | 3102 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
ed95b21a | 3103 | char cookie[32]; |
bb040aa0 ID |
3104 | int ret; |
3105 | ||
ed95b21a | 3106 | WARN_ON(!__rbd_is_lock_owner(rbd_dev)); |
bb040aa0 | 3107 | |
ed95b21a | 3108 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; |
bb040aa0 | 3109 | |
ed95b21a ID |
3110 | format_lock_cookie(rbd_dev, cookie); |
3111 | ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
3112 | RBD_LOCK_NAME, cookie); | |
3113 | if (ret && ret != -ENOENT) { | |
3114 | rbd_warn(rbd_dev, "cls_unlock failed: %d", ret); | |
3115 | return ret; | |
bb040aa0 ID |
3116 | } |
3117 | ||
ed95b21a ID |
3118 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); |
3119 | queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work); | |
3120 | return 0; | |
bb040aa0 ID |
3121 | } |
3122 | ||
ed95b21a ID |
3123 | static int __rbd_notify_op_lock(struct rbd_device *rbd_dev, |
3124 | enum rbd_notify_op notify_op, | |
3125 | struct page ***preply_pages, | |
3126 | size_t *preply_len) | |
9969ebc5 AE |
3127 | { |
3128 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
ed95b21a ID |
3129 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); |
3130 | int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN; | |
3131 | char buf[buf_size]; | |
3132 | void *p = buf; | |
9969ebc5 | 3133 | |
ed95b21a | 3134 | dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op); |
9969ebc5 | 3135 | |
ed95b21a ID |
3136 | /* encode *LockPayload NotifyMessage (op + ClientId) */ |
3137 | ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN); | |
3138 | ceph_encode_32(&p, notify_op); | |
3139 | ceph_encode_64(&p, cid.gid); | |
3140 | ceph_encode_64(&p, cid.handle); | |
8eb87565 | 3141 | |
ed95b21a ID |
3142 | return ceph_osdc_notify(osdc, &rbd_dev->header_oid, |
3143 | &rbd_dev->header_oloc, buf, buf_size, | |
3144 | RBD_NOTIFY_TIMEOUT, preply_pages, preply_len); | |
b30a01f2 ID |
3145 | } |
3146 | ||
ed95b21a ID |
3147 | static void rbd_notify_op_lock(struct rbd_device *rbd_dev, |
3148 | enum rbd_notify_op notify_op) | |
b30a01f2 | 3149 | { |
ed95b21a ID |
3150 | struct page **reply_pages; |
3151 | size_t reply_len; | |
b30a01f2 | 3152 | |
ed95b21a ID |
3153 | __rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len); |
3154 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3155 | } | |
b30a01f2 | 3156 | |
ed95b21a ID |
3157 | static void rbd_notify_acquired_lock(struct work_struct *work) |
3158 | { | |
3159 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3160 | acquired_lock_work); | |
76756a51 | 3161 | |
ed95b21a | 3162 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK); |
c525f036 ID |
3163 | } |
3164 | ||
ed95b21a | 3165 | static void rbd_notify_released_lock(struct work_struct *work) |
c525f036 | 3166 | { |
ed95b21a ID |
3167 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, |
3168 | released_lock_work); | |
811c6688 | 3169 | |
ed95b21a | 3170 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK); |
fca27065 ID |
3171 | } |
3172 | ||
ed95b21a | 3173 | static int rbd_request_lock(struct rbd_device *rbd_dev) |
36be9a76 | 3174 | { |
ed95b21a ID |
3175 | struct page **reply_pages; |
3176 | size_t reply_len; | |
3177 | bool lock_owner_responded = false; | |
36be9a76 AE |
3178 | int ret; |
3179 | ||
ed95b21a | 3180 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
36be9a76 | 3181 | |
ed95b21a ID |
3182 | ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK, |
3183 | &reply_pages, &reply_len); | |
3184 | if (ret && ret != -ETIMEDOUT) { | |
3185 | rbd_warn(rbd_dev, "failed to request lock: %d", ret); | |
36be9a76 | 3186 | goto out; |
ed95b21a | 3187 | } |
36be9a76 | 3188 | |
ed95b21a ID |
3189 | if (reply_len > 0 && reply_len <= PAGE_SIZE) { |
3190 | void *p = page_address(reply_pages[0]); | |
3191 | void *const end = p + reply_len; | |
3192 | u32 n; | |
36be9a76 | 3193 | |
ed95b21a ID |
3194 | ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */ |
3195 | while (n--) { | |
3196 | u8 struct_v; | |
3197 | u32 len; | |
36be9a76 | 3198 | |
ed95b21a ID |
3199 | ceph_decode_need(&p, end, 8 + 8, e_inval); |
3200 | p += 8 + 8; /* skip gid and cookie */ | |
04017e29 | 3201 | |
ed95b21a ID |
3202 | ceph_decode_32_safe(&p, end, len, e_inval); |
3203 | if (!len) | |
3204 | continue; | |
3205 | ||
3206 | if (lock_owner_responded) { | |
3207 | rbd_warn(rbd_dev, | |
3208 | "duplicate lock owners detected"); | |
3209 | ret = -EIO; | |
3210 | goto out; | |
3211 | } | |
3212 | ||
3213 | lock_owner_responded = true; | |
3214 | ret = ceph_start_decoding(&p, end, 1, "ResponseMessage", | |
3215 | &struct_v, &len); | |
3216 | if (ret) { | |
3217 | rbd_warn(rbd_dev, | |
3218 | "failed to decode ResponseMessage: %d", | |
3219 | ret); | |
3220 | goto e_inval; | |
3221 | } | |
3222 | ||
3223 | ret = ceph_decode_32(&p); | |
3224 | } | |
3225 | } | |
3226 | ||
3227 | if (!lock_owner_responded) { | |
3228 | rbd_warn(rbd_dev, "no lock owners detected"); | |
3229 | ret = -ETIMEDOUT; | |
3230 | } | |
3231 | ||
3232 | out: | |
3233 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3234 | return ret; | |
3235 | ||
3236 | e_inval: | |
3237 | ret = -EINVAL; | |
3238 | goto out; | |
3239 | } | |
3240 | ||
3241 | static void wake_requests(struct rbd_device *rbd_dev, bool wake_all) | |
3242 | { | |
3243 | dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all); | |
3244 | ||
3245 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
3246 | if (wake_all) | |
3247 | wake_up_all(&rbd_dev->lock_waitq); | |
3248 | else | |
3249 | wake_up(&rbd_dev->lock_waitq); | |
3250 | } | |
3251 | ||
3252 | static int get_lock_owner_info(struct rbd_device *rbd_dev, | |
3253 | struct ceph_locker **lockers, u32 *num_lockers) | |
3254 | { | |
3255 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3256 | u8 lock_type; | |
3257 | char *lock_tag; | |
3258 | int ret; | |
3259 | ||
3260 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3261 | ||
3262 | ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid, | |
3263 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
3264 | &lock_type, &lock_tag, lockers, num_lockers); | |
3265 | if (ret) | |
3266 | return ret; | |
3267 | ||
3268 | if (*num_lockers == 0) { | |
3269 | dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev); | |
3270 | goto out; | |
3271 | } | |
3272 | ||
3273 | if (strcmp(lock_tag, RBD_LOCK_TAG)) { | |
3274 | rbd_warn(rbd_dev, "locked by external mechanism, tag %s", | |
3275 | lock_tag); | |
3276 | ret = -EBUSY; | |
3277 | goto out; | |
3278 | } | |
3279 | ||
3280 | if (lock_type == CEPH_CLS_LOCK_SHARED) { | |
3281 | rbd_warn(rbd_dev, "shared lock type detected"); | |
3282 | ret = -EBUSY; | |
3283 | goto out; | |
3284 | } | |
3285 | ||
3286 | if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX, | |
3287 | strlen(RBD_LOCK_COOKIE_PREFIX))) { | |
3288 | rbd_warn(rbd_dev, "locked by external mechanism, cookie %s", | |
3289 | (*lockers)[0].id.cookie); | |
3290 | ret = -EBUSY; | |
3291 | goto out; | |
3292 | } | |
3293 | ||
3294 | out: | |
3295 | kfree(lock_tag); | |
3296 | return ret; | |
3297 | } | |
3298 | ||
3299 | static int find_watcher(struct rbd_device *rbd_dev, | |
3300 | const struct ceph_locker *locker) | |
3301 | { | |
3302 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3303 | struct ceph_watch_item *watchers; | |
3304 | u32 num_watchers; | |
3305 | u64 cookie; | |
3306 | int i; | |
3307 | int ret; | |
3308 | ||
3309 | ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid, | |
3310 | &rbd_dev->header_oloc, &watchers, | |
3311 | &num_watchers); | |
3312 | if (ret) | |
3313 | return ret; | |
3314 | ||
3315 | sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie); | |
3316 | for (i = 0; i < num_watchers; i++) { | |
3317 | if (!memcmp(&watchers[i].addr, &locker->info.addr, | |
3318 | sizeof(locker->info.addr)) && | |
3319 | watchers[i].cookie == cookie) { | |
3320 | struct rbd_client_id cid = { | |
3321 | .gid = le64_to_cpu(watchers[i].name.num), | |
3322 | .handle = cookie, | |
3323 | }; | |
3324 | ||
3325 | dout("%s rbd_dev %p found cid %llu-%llu\n", __func__, | |
3326 | rbd_dev, cid.gid, cid.handle); | |
3327 | rbd_set_owner_cid(rbd_dev, &cid); | |
3328 | ret = 1; | |
3329 | goto out; | |
3330 | } | |
3331 | } | |
3332 | ||
3333 | dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev); | |
3334 | ret = 0; | |
3335 | out: | |
3336 | kfree(watchers); | |
3337 | return ret; | |
3338 | } | |
3339 | ||
3340 | /* | |
3341 | * lock_rwsem must be held for write | |
3342 | */ | |
3343 | static int rbd_try_lock(struct rbd_device *rbd_dev) | |
3344 | { | |
3345 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
3346 | struct ceph_locker *lockers; | |
3347 | u32 num_lockers; | |
3348 | int ret; | |
3349 | ||
3350 | for (;;) { | |
3351 | ret = rbd_lock(rbd_dev); | |
3352 | if (ret != -EBUSY) | |
3353 | return ret; | |
3354 | ||
3355 | /* determine if the current lock holder is still alive */ | |
3356 | ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers); | |
3357 | if (ret) | |
3358 | return ret; | |
3359 | ||
3360 | if (num_lockers == 0) | |
3361 | goto again; | |
3362 | ||
3363 | ret = find_watcher(rbd_dev, lockers); | |
3364 | if (ret) { | |
3365 | if (ret > 0) | |
3366 | ret = 0; /* have to request lock */ | |
3367 | goto out; | |
3368 | } | |
3369 | ||
3370 | rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock", | |
3371 | ENTITY_NAME(lockers[0].id.name)); | |
3372 | ||
3373 | ret = ceph_monc_blacklist_add(&client->monc, | |
3374 | &lockers[0].info.addr); | |
3375 | if (ret) { | |
3376 | rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d", | |
3377 | ENTITY_NAME(lockers[0].id.name), ret); | |
3378 | goto out; | |
3379 | } | |
3380 | ||
3381 | ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid, | |
3382 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
3383 | lockers[0].id.cookie, | |
3384 | &lockers[0].id.name); | |
3385 | if (ret && ret != -ENOENT) | |
3386 | goto out; | |
3387 | ||
3388 | again: | |
3389 | ceph_free_lockers(lockers, num_lockers); | |
3390 | } | |
3391 | ||
3392 | out: | |
3393 | ceph_free_lockers(lockers, num_lockers); | |
3394 | return ret; | |
3395 | } | |
3396 | ||
3397 | /* | |
3398 | * ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED | |
3399 | */ | |
3400 | static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev, | |
3401 | int *pret) | |
3402 | { | |
3403 | enum rbd_lock_state lock_state; | |
3404 | ||
3405 | down_read(&rbd_dev->lock_rwsem); | |
3406 | dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev, | |
3407 | rbd_dev->lock_state); | |
3408 | if (__rbd_is_lock_owner(rbd_dev)) { | |
3409 | lock_state = rbd_dev->lock_state; | |
3410 | up_read(&rbd_dev->lock_rwsem); | |
3411 | return lock_state; | |
3412 | } | |
3413 | ||
3414 | up_read(&rbd_dev->lock_rwsem); | |
3415 | down_write(&rbd_dev->lock_rwsem); | |
3416 | dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev, | |
3417 | rbd_dev->lock_state); | |
3418 | if (!__rbd_is_lock_owner(rbd_dev)) { | |
3419 | *pret = rbd_try_lock(rbd_dev); | |
3420 | if (*pret) | |
3421 | rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret); | |
3422 | } | |
3423 | ||
3424 | lock_state = rbd_dev->lock_state; | |
3425 | up_write(&rbd_dev->lock_rwsem); | |
3426 | return lock_state; | |
3427 | } | |
3428 | ||
3429 | static void rbd_acquire_lock(struct work_struct *work) | |
3430 | { | |
3431 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
3432 | struct rbd_device, lock_dwork); | |
3433 | enum rbd_lock_state lock_state; | |
3434 | int ret; | |
3435 | ||
3436 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3437 | again: | |
3438 | lock_state = rbd_try_acquire_lock(rbd_dev, &ret); | |
3439 | if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) { | |
3440 | if (lock_state == RBD_LOCK_STATE_LOCKED) | |
3441 | wake_requests(rbd_dev, true); | |
3442 | dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__, | |
3443 | rbd_dev, lock_state, ret); | |
3444 | return; | |
3445 | } | |
3446 | ||
3447 | ret = rbd_request_lock(rbd_dev); | |
3448 | if (ret == -ETIMEDOUT) { | |
3449 | goto again; /* treat this as a dead client */ | |
3450 | } else if (ret < 0) { | |
3451 | rbd_warn(rbd_dev, "error requesting lock: %d", ret); | |
3452 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
3453 | RBD_RETRY_DELAY); | |
3454 | } else { | |
3455 | /* | |
3456 | * lock owner acked, but resend if we don't see them | |
3457 | * release the lock | |
3458 | */ | |
3459 | dout("%s rbd_dev %p requeueing lock_dwork\n", __func__, | |
3460 | rbd_dev); | |
3461 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
3462 | msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC)); | |
3463 | } | |
3464 | } | |
3465 | ||
3466 | /* | |
3467 | * lock_rwsem must be held for write | |
3468 | */ | |
3469 | static bool rbd_release_lock(struct rbd_device *rbd_dev) | |
3470 | { | |
3471 | dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev, | |
3472 | rbd_dev->lock_state); | |
3473 | if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED) | |
3474 | return false; | |
3475 | ||
3476 | rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING; | |
3477 | downgrade_write(&rbd_dev->lock_rwsem); | |
52bb1f9b | 3478 | /* |
ed95b21a | 3479 | * Ensure that all in-flight IO is flushed. |
52bb1f9b | 3480 | * |
ed95b21a ID |
3481 | * FIXME: ceph_osdc_sync() flushes the entire OSD client, which |
3482 | * may be shared with other devices. | |
52bb1f9b | 3483 | */ |
ed95b21a ID |
3484 | ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc); |
3485 | up_read(&rbd_dev->lock_rwsem); | |
3486 | ||
3487 | down_write(&rbd_dev->lock_rwsem); | |
3488 | dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev, | |
3489 | rbd_dev->lock_state); | |
3490 | if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING) | |
3491 | return false; | |
3492 | ||
3493 | if (!rbd_unlock(rbd_dev)) | |
3494 | /* | |
3495 | * Give others a chance to grab the lock - we would re-acquire | |
3496 | * almost immediately if we got new IO during ceph_osdc_sync() | |
3497 | * otherwise. We need to ack our own notifications, so this | |
3498 | * lock_dwork will be requeued from rbd_wait_state_locked() | |
3499 | * after wake_requests() in rbd_handle_released_lock(). | |
3500 | */ | |
3501 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
3502 | ||
3503 | return true; | |
3504 | } | |
3505 | ||
3506 | static void rbd_release_lock_work(struct work_struct *work) | |
3507 | { | |
3508 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3509 | unlock_work); | |
3510 | ||
3511 | down_write(&rbd_dev->lock_rwsem); | |
3512 | rbd_release_lock(rbd_dev); | |
3513 | up_write(&rbd_dev->lock_rwsem); | |
3514 | } | |
3515 | ||
3516 | static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3517 | void **p) | |
3518 | { | |
3519 | struct rbd_client_id cid = { 0 }; | |
3520 | ||
3521 | if (struct_v >= 2) { | |
3522 | cid.gid = ceph_decode_64(p); | |
3523 | cid.handle = ceph_decode_64(p); | |
3524 | } | |
3525 | ||
3526 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3527 | cid.handle); | |
3528 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
3529 | down_write(&rbd_dev->lock_rwsem); | |
3530 | if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
3531 | /* | |
3532 | * we already know that the remote client is | |
3533 | * the owner | |
3534 | */ | |
3535 | up_write(&rbd_dev->lock_rwsem); | |
3536 | return; | |
3537 | } | |
3538 | ||
3539 | rbd_set_owner_cid(rbd_dev, &cid); | |
3540 | downgrade_write(&rbd_dev->lock_rwsem); | |
3541 | } else { | |
3542 | down_read(&rbd_dev->lock_rwsem); | |
3543 | } | |
3544 | ||
3545 | if (!__rbd_is_lock_owner(rbd_dev)) | |
3546 | wake_requests(rbd_dev, false); | |
3547 | up_read(&rbd_dev->lock_rwsem); | |
3548 | } | |
3549 | ||
3550 | static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3551 | void **p) | |
3552 | { | |
3553 | struct rbd_client_id cid = { 0 }; | |
3554 | ||
3555 | if (struct_v >= 2) { | |
3556 | cid.gid = ceph_decode_64(p); | |
3557 | cid.handle = ceph_decode_64(p); | |
3558 | } | |
3559 | ||
3560 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3561 | cid.handle); | |
3562 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
3563 | down_write(&rbd_dev->lock_rwsem); | |
3564 | if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
3565 | dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n", | |
3566 | __func__, rbd_dev, cid.gid, cid.handle, | |
3567 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle); | |
3568 | up_write(&rbd_dev->lock_rwsem); | |
3569 | return; | |
3570 | } | |
3571 | ||
3572 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
3573 | downgrade_write(&rbd_dev->lock_rwsem); | |
3574 | } else { | |
3575 | down_read(&rbd_dev->lock_rwsem); | |
3576 | } | |
3577 | ||
3578 | if (!__rbd_is_lock_owner(rbd_dev)) | |
3579 | wake_requests(rbd_dev, false); | |
3580 | up_read(&rbd_dev->lock_rwsem); | |
3581 | } | |
3582 | ||
3583 | static bool rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3584 | void **p) | |
3585 | { | |
3586 | struct rbd_client_id my_cid = rbd_get_cid(rbd_dev); | |
3587 | struct rbd_client_id cid = { 0 }; | |
3588 | bool need_to_send; | |
3589 | ||
3590 | if (struct_v >= 2) { | |
3591 | cid.gid = ceph_decode_64(p); | |
3592 | cid.handle = ceph_decode_64(p); | |
3593 | } | |
3594 | ||
3595 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3596 | cid.handle); | |
3597 | if (rbd_cid_equal(&cid, &my_cid)) | |
3598 | return false; | |
3599 | ||
3600 | down_read(&rbd_dev->lock_rwsem); | |
3601 | need_to_send = __rbd_is_lock_owner(rbd_dev); | |
3602 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) { | |
3603 | if (!rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid)) { | |
3604 | dout("%s rbd_dev %p queueing unlock_work\n", __func__, | |
3605 | rbd_dev); | |
3606 | queue_work(rbd_dev->task_wq, &rbd_dev->unlock_work); | |
3607 | } | |
3608 | } | |
3609 | up_read(&rbd_dev->lock_rwsem); | |
3610 | return need_to_send; | |
3611 | } | |
3612 | ||
3613 | static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev, | |
3614 | u64 notify_id, u64 cookie, s32 *result) | |
3615 | { | |
3616 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3617 | int buf_size = 4 + CEPH_ENCODING_START_BLK_LEN; | |
3618 | char buf[buf_size]; | |
3619 | int ret; | |
3620 | ||
3621 | if (result) { | |
3622 | void *p = buf; | |
3623 | ||
3624 | /* encode ResponseMessage */ | |
3625 | ceph_start_encoding(&p, 1, 1, | |
3626 | buf_size - CEPH_ENCODING_START_BLK_LEN); | |
3627 | ceph_encode_32(&p, *result); | |
3628 | } else { | |
3629 | buf_size = 0; | |
3630 | } | |
b8d70035 | 3631 | |
922dab61 ID |
3632 | ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid, |
3633 | &rbd_dev->header_oloc, notify_id, cookie, | |
ed95b21a | 3634 | buf, buf_size); |
52bb1f9b | 3635 | if (ret) |
ed95b21a ID |
3636 | rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret); |
3637 | } | |
3638 | ||
3639 | static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id, | |
3640 | u64 cookie) | |
3641 | { | |
3642 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3643 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL); | |
3644 | } | |
3645 | ||
3646 | static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev, | |
3647 | u64 notify_id, u64 cookie, s32 result) | |
3648 | { | |
3649 | dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result); | |
3650 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result); | |
3651 | } | |
3652 | ||
3653 | static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie, | |
3654 | u64 notifier_id, void *data, size_t data_len) | |
3655 | { | |
3656 | struct rbd_device *rbd_dev = arg; | |
3657 | void *p = data; | |
3658 | void *const end = p + data_len; | |
d4c2269b | 3659 | u8 struct_v = 0; |
ed95b21a ID |
3660 | u32 len; |
3661 | u32 notify_op; | |
3662 | int ret; | |
3663 | ||
3664 | dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n", | |
3665 | __func__, rbd_dev, cookie, notify_id, data_len); | |
3666 | if (data_len) { | |
3667 | ret = ceph_start_decoding(&p, end, 1, "NotifyMessage", | |
3668 | &struct_v, &len); | |
3669 | if (ret) { | |
3670 | rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d", | |
3671 | ret); | |
3672 | return; | |
3673 | } | |
3674 | ||
3675 | notify_op = ceph_decode_32(&p); | |
3676 | } else { | |
3677 | /* legacy notification for header updates */ | |
3678 | notify_op = RBD_NOTIFY_OP_HEADER_UPDATE; | |
3679 | len = 0; | |
3680 | } | |
3681 | ||
3682 | dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op); | |
3683 | switch (notify_op) { | |
3684 | case RBD_NOTIFY_OP_ACQUIRED_LOCK: | |
3685 | rbd_handle_acquired_lock(rbd_dev, struct_v, &p); | |
3686 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3687 | break; | |
3688 | case RBD_NOTIFY_OP_RELEASED_LOCK: | |
3689 | rbd_handle_released_lock(rbd_dev, struct_v, &p); | |
3690 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3691 | break; | |
3692 | case RBD_NOTIFY_OP_REQUEST_LOCK: | |
3693 | if (rbd_handle_request_lock(rbd_dev, struct_v, &p)) | |
3694 | /* | |
3695 | * send ResponseMessage(0) back so the client | |
3696 | * can detect a missing owner | |
3697 | */ | |
3698 | rbd_acknowledge_notify_result(rbd_dev, notify_id, | |
3699 | cookie, 0); | |
3700 | else | |
3701 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3702 | break; | |
3703 | case RBD_NOTIFY_OP_HEADER_UPDATE: | |
3704 | ret = rbd_dev_refresh(rbd_dev); | |
3705 | if (ret) | |
3706 | rbd_warn(rbd_dev, "refresh failed: %d", ret); | |
3707 | ||
3708 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3709 | break; | |
3710 | default: | |
3711 | if (rbd_is_lock_owner(rbd_dev)) | |
3712 | rbd_acknowledge_notify_result(rbd_dev, notify_id, | |
3713 | cookie, -EOPNOTSUPP); | |
3714 | else | |
3715 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3716 | break; | |
3717 | } | |
b8d70035 AE |
3718 | } |
3719 | ||
99d16943 ID |
3720 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev); |
3721 | ||
922dab61 | 3722 | static void rbd_watch_errcb(void *arg, u64 cookie, int err) |
bb040aa0 | 3723 | { |
922dab61 | 3724 | struct rbd_device *rbd_dev = arg; |
bb040aa0 | 3725 | |
922dab61 | 3726 | rbd_warn(rbd_dev, "encountered watch error: %d", err); |
bb040aa0 | 3727 | |
ed95b21a ID |
3728 | down_write(&rbd_dev->lock_rwsem); |
3729 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
3730 | up_write(&rbd_dev->lock_rwsem); | |
3731 | ||
99d16943 ID |
3732 | mutex_lock(&rbd_dev->watch_mutex); |
3733 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) { | |
3734 | __rbd_unregister_watch(rbd_dev); | |
3735 | rbd_dev->watch_state = RBD_WATCH_STATE_ERROR; | |
bb040aa0 | 3736 | |
99d16943 | 3737 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0); |
bb040aa0 | 3738 | } |
99d16943 | 3739 | mutex_unlock(&rbd_dev->watch_mutex); |
bb040aa0 ID |
3740 | } |
3741 | ||
9969ebc5 | 3742 | /* |
99d16943 | 3743 | * watch_mutex must be locked |
9969ebc5 | 3744 | */ |
99d16943 | 3745 | static int __rbd_register_watch(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3746 | { |
3747 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
922dab61 | 3748 | struct ceph_osd_linger_request *handle; |
9969ebc5 | 3749 | |
922dab61 | 3750 | rbd_assert(!rbd_dev->watch_handle); |
99d16943 | 3751 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
9969ebc5 | 3752 | |
922dab61 ID |
3753 | handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid, |
3754 | &rbd_dev->header_oloc, rbd_watch_cb, | |
3755 | rbd_watch_errcb, rbd_dev); | |
3756 | if (IS_ERR(handle)) | |
3757 | return PTR_ERR(handle); | |
8eb87565 | 3758 | |
922dab61 | 3759 | rbd_dev->watch_handle = handle; |
b30a01f2 | 3760 | return 0; |
b30a01f2 ID |
3761 | } |
3762 | ||
99d16943 ID |
3763 | /* |
3764 | * watch_mutex must be locked | |
3765 | */ | |
3766 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev) | |
b30a01f2 | 3767 | { |
922dab61 ID |
3768 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3769 | int ret; | |
b30a01f2 | 3770 | |
99d16943 ID |
3771 | rbd_assert(rbd_dev->watch_handle); |
3772 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
b30a01f2 | 3773 | |
922dab61 ID |
3774 | ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle); |
3775 | if (ret) | |
3776 | rbd_warn(rbd_dev, "failed to unwatch: %d", ret); | |
76756a51 | 3777 | |
922dab61 | 3778 | rbd_dev->watch_handle = NULL; |
c525f036 ID |
3779 | } |
3780 | ||
99d16943 ID |
3781 | static int rbd_register_watch(struct rbd_device *rbd_dev) |
3782 | { | |
3783 | int ret; | |
3784 | ||
3785 | mutex_lock(&rbd_dev->watch_mutex); | |
3786 | rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED); | |
3787 | ret = __rbd_register_watch(rbd_dev); | |
3788 | if (ret) | |
3789 | goto out; | |
3790 | ||
3791 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; | |
3792 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
3793 | ||
3794 | out: | |
3795 | mutex_unlock(&rbd_dev->watch_mutex); | |
3796 | return ret; | |
3797 | } | |
3798 | ||
3799 | static void cancel_tasks_sync(struct rbd_device *rbd_dev) | |
c525f036 | 3800 | { |
99d16943 ID |
3801 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
3802 | ||
3803 | cancel_delayed_work_sync(&rbd_dev->watch_dwork); | |
ed95b21a ID |
3804 | cancel_work_sync(&rbd_dev->acquired_lock_work); |
3805 | cancel_work_sync(&rbd_dev->released_lock_work); | |
3806 | cancel_delayed_work_sync(&rbd_dev->lock_dwork); | |
3807 | cancel_work_sync(&rbd_dev->unlock_work); | |
99d16943 ID |
3808 | } |
3809 | ||
3810 | static void rbd_unregister_watch(struct rbd_device *rbd_dev) | |
3811 | { | |
ed95b21a | 3812 | WARN_ON(waitqueue_active(&rbd_dev->lock_waitq)); |
99d16943 ID |
3813 | cancel_tasks_sync(rbd_dev); |
3814 | ||
3815 | mutex_lock(&rbd_dev->watch_mutex); | |
3816 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) | |
3817 | __rbd_unregister_watch(rbd_dev); | |
3818 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
3819 | mutex_unlock(&rbd_dev->watch_mutex); | |
811c6688 | 3820 | |
811c6688 | 3821 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); |
fca27065 ID |
3822 | } |
3823 | ||
99d16943 ID |
3824 | static void rbd_reregister_watch(struct work_struct *work) |
3825 | { | |
3826 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
3827 | struct rbd_device, watch_dwork); | |
ed95b21a | 3828 | bool was_lock_owner = false; |
87c0fded | 3829 | bool need_to_wake = false; |
99d16943 ID |
3830 | int ret; |
3831 | ||
3832 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3833 | ||
ed95b21a ID |
3834 | down_write(&rbd_dev->lock_rwsem); |
3835 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) | |
3836 | was_lock_owner = rbd_release_lock(rbd_dev); | |
3837 | ||
99d16943 | 3838 | mutex_lock(&rbd_dev->watch_mutex); |
87c0fded ID |
3839 | if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) { |
3840 | mutex_unlock(&rbd_dev->watch_mutex); | |
3841 | goto out; | |
3842 | } | |
99d16943 ID |
3843 | |
3844 | ret = __rbd_register_watch(rbd_dev); | |
3845 | if (ret) { | |
3846 | rbd_warn(rbd_dev, "failed to reregister watch: %d", ret); | |
4d73644b | 3847 | if (ret == -EBLACKLISTED || ret == -ENOENT) { |
87c0fded ID |
3848 | set_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags); |
3849 | need_to_wake = true; | |
3850 | } else { | |
99d16943 ID |
3851 | queue_delayed_work(rbd_dev->task_wq, |
3852 | &rbd_dev->watch_dwork, | |
3853 | RBD_RETRY_DELAY); | |
87c0fded ID |
3854 | } |
3855 | mutex_unlock(&rbd_dev->watch_mutex); | |
3856 | goto out; | |
99d16943 ID |
3857 | } |
3858 | ||
87c0fded | 3859 | need_to_wake = true; |
99d16943 ID |
3860 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; |
3861 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
3862 | mutex_unlock(&rbd_dev->watch_mutex); | |
3863 | ||
3864 | ret = rbd_dev_refresh(rbd_dev); | |
3865 | if (ret) | |
3866 | rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret); | |
3867 | ||
ed95b21a ID |
3868 | if (was_lock_owner) { |
3869 | ret = rbd_try_lock(rbd_dev); | |
3870 | if (ret) | |
3871 | rbd_warn(rbd_dev, "reregisteration lock failed: %d", | |
3872 | ret); | |
3873 | } | |
3874 | ||
87c0fded | 3875 | out: |
ed95b21a | 3876 | up_write(&rbd_dev->lock_rwsem); |
87c0fded ID |
3877 | if (need_to_wake) |
3878 | wake_requests(rbd_dev, true); | |
99d16943 ID |
3879 | } |
3880 | ||
36be9a76 | 3881 | /* |
f40eb349 AE |
3882 | * Synchronous osd object method call. Returns the number of bytes |
3883 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3884 | */ |
3885 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
ecd4a68a ID |
3886 | struct ceph_object_id *oid, |
3887 | struct ceph_object_locator *oloc, | |
36be9a76 | 3888 | const char *method_name, |
4157976b | 3889 | const void *outbound, |
36be9a76 | 3890 | size_t outbound_size, |
4157976b | 3891 | void *inbound, |
e2a58ee5 | 3892 | size_t inbound_size) |
36be9a76 | 3893 | { |
ecd4a68a ID |
3894 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3895 | struct page *req_page = NULL; | |
3896 | struct page *reply_page; | |
36be9a76 AE |
3897 | int ret; |
3898 | ||
3899 | /* | |
6010a451 AE |
3900 | * Method calls are ultimately read operations. The result |
3901 | * should placed into the inbound buffer provided. They | |
3902 | * also supply outbound data--parameters for the object | |
3903 | * method. Currently if this is present it will be a | |
3904 | * snapshot id. | |
36be9a76 | 3905 | */ |
ecd4a68a ID |
3906 | if (outbound) { |
3907 | if (outbound_size > PAGE_SIZE) | |
3908 | return -E2BIG; | |
36be9a76 | 3909 | |
ecd4a68a ID |
3910 | req_page = alloc_page(GFP_KERNEL); |
3911 | if (!req_page) | |
3912 | return -ENOMEM; | |
04017e29 | 3913 | |
ecd4a68a | 3914 | memcpy(page_address(req_page), outbound, outbound_size); |
04017e29 | 3915 | } |
36be9a76 | 3916 | |
ecd4a68a ID |
3917 | reply_page = alloc_page(GFP_KERNEL); |
3918 | if (!reply_page) { | |
3919 | if (req_page) | |
3920 | __free_page(req_page); | |
3921 | return -ENOMEM; | |
3922 | } | |
57385b51 | 3923 | |
ecd4a68a ID |
3924 | ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name, |
3925 | CEPH_OSD_FLAG_READ, req_page, outbound_size, | |
3926 | reply_page, &inbound_size); | |
3927 | if (!ret) { | |
3928 | memcpy(inbound, page_address(reply_page), inbound_size); | |
3929 | ret = inbound_size; | |
3930 | } | |
36be9a76 | 3931 | |
ecd4a68a ID |
3932 | if (req_page) |
3933 | __free_page(req_page); | |
3934 | __free_page(reply_page); | |
36be9a76 AE |
3935 | return ret; |
3936 | } | |
3937 | ||
ed95b21a ID |
3938 | /* |
3939 | * lock_rwsem must be held for read | |
3940 | */ | |
3941 | static void rbd_wait_state_locked(struct rbd_device *rbd_dev) | |
3942 | { | |
3943 | DEFINE_WAIT(wait); | |
3944 | ||
3945 | do { | |
3946 | /* | |
3947 | * Note the use of mod_delayed_work() in rbd_acquire_lock() | |
3948 | * and cancel_delayed_work() in wake_requests(). | |
3949 | */ | |
3950 | dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev); | |
3951 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); | |
3952 | prepare_to_wait_exclusive(&rbd_dev->lock_waitq, &wait, | |
3953 | TASK_UNINTERRUPTIBLE); | |
3954 | up_read(&rbd_dev->lock_rwsem); | |
3955 | schedule(); | |
3956 | down_read(&rbd_dev->lock_rwsem); | |
87c0fded ID |
3957 | } while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED && |
3958 | !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)); | |
3959 | ||
ed95b21a ID |
3960 | finish_wait(&rbd_dev->lock_waitq, &wait); |
3961 | } | |
3962 | ||
7ad18afa | 3963 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3964 | { |
7ad18afa CH |
3965 | struct request *rq = blk_mq_rq_from_pdu(work); |
3966 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3967 | struct rbd_img_request *img_request; |
4e752f0a | 3968 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3969 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3970 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3971 | enum obj_operation_type op_type; |
4e752f0a | 3972 | u64 mapping_size; |
80de1912 | 3973 | bool must_be_locked; |
bf0d5f50 AE |
3974 | int result; |
3975 | ||
aebf526b CH |
3976 | switch (req_op(rq)) { |
3977 | case REQ_OP_DISCARD: | |
90e98c52 | 3978 | op_type = OBJ_OP_DISCARD; |
aebf526b CH |
3979 | break; |
3980 | case REQ_OP_WRITE: | |
6d2940c8 | 3981 | op_type = OBJ_OP_WRITE; |
aebf526b CH |
3982 | break; |
3983 | case REQ_OP_READ: | |
6d2940c8 | 3984 | op_type = OBJ_OP_READ; |
aebf526b CH |
3985 | break; |
3986 | default: | |
3987 | dout("%s: non-fs request type %d\n", __func__, req_op(rq)); | |
3988 | result = -EIO; | |
3989 | goto err; | |
3990 | } | |
6d2940c8 | 3991 | |
bc1ecc65 | 3992 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3993 | |
bc1ecc65 ID |
3994 | if (!length) { |
3995 | dout("%s: zero-length request\n", __func__); | |
3996 | result = 0; | |
3997 | goto err_rq; | |
3998 | } | |
bf0d5f50 | 3999 | |
6d2940c8 | 4000 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 4001 | |
6d2940c8 | 4002 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
4003 | if (rbd_dev->mapping.read_only) { |
4004 | result = -EROFS; | |
4005 | goto err_rq; | |
4dda41d3 | 4006 | } |
bc1ecc65 ID |
4007 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
4008 | } | |
4dda41d3 | 4009 | |
bc1ecc65 ID |
4010 | /* |
4011 | * Quit early if the mapped snapshot no longer exists. It's | |
4012 | * still possible the snapshot will have disappeared by the | |
4013 | * time our request arrives at the osd, but there's no sense in | |
4014 | * sending it if we already know. | |
4015 | */ | |
4016 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
4017 | dout("request for non-existent snapshot"); | |
4018 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
4019 | result = -ENXIO; | |
4020 | goto err_rq; | |
4021 | } | |
4dda41d3 | 4022 | |
bc1ecc65 ID |
4023 | if (offset && length > U64_MAX - offset + 1) { |
4024 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
4025 | length); | |
4026 | result = -EINVAL; | |
4027 | goto err_rq; /* Shouldn't happen */ | |
4028 | } | |
4dda41d3 | 4029 | |
7ad18afa CH |
4030 | blk_mq_start_request(rq); |
4031 | ||
4e752f0a JD |
4032 | down_read(&rbd_dev->header_rwsem); |
4033 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 4034 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
4035 | snapc = rbd_dev->header.snapc; |
4036 | ceph_get_snap_context(snapc); | |
ed95b21a | 4037 | must_be_locked = rbd_is_lock_supported(rbd_dev); |
80de1912 ID |
4038 | } else { |
4039 | must_be_locked = rbd_dev->opts->lock_on_read && | |
4040 | rbd_is_lock_supported(rbd_dev); | |
4e752f0a JD |
4041 | } |
4042 | up_read(&rbd_dev->header_rwsem); | |
4043 | ||
4044 | if (offset + length > mapping_size) { | |
bc1ecc65 | 4045 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 4046 | length, mapping_size); |
bc1ecc65 ID |
4047 | result = -EIO; |
4048 | goto err_rq; | |
4049 | } | |
bf0d5f50 | 4050 | |
ed95b21a ID |
4051 | if (must_be_locked) { |
4052 | down_read(&rbd_dev->lock_rwsem); | |
87c0fded ID |
4053 | if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED && |
4054 | !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) | |
ed95b21a | 4055 | rbd_wait_state_locked(rbd_dev); |
87c0fded ID |
4056 | |
4057 | WARN_ON((rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) ^ | |
4058 | !test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)); | |
4059 | if (test_bit(RBD_DEV_FLAG_BLACKLISTED, &rbd_dev->flags)) { | |
4060 | result = -EBLACKLISTED; | |
4061 | goto err_unlock; | |
4062 | } | |
ed95b21a ID |
4063 | } |
4064 | ||
6d2940c8 | 4065 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 4066 | snapc); |
bc1ecc65 ID |
4067 | if (!img_request) { |
4068 | result = -ENOMEM; | |
ed95b21a | 4069 | goto err_unlock; |
bc1ecc65 ID |
4070 | } |
4071 | img_request->rq = rq; | |
70b16db8 | 4072 | snapc = NULL; /* img_request consumes a ref */ |
bf0d5f50 | 4073 | |
90e98c52 GZ |
4074 | if (op_type == OBJ_OP_DISCARD) |
4075 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
4076 | NULL); | |
4077 | else | |
4078 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
4079 | rq->bio); | |
bc1ecc65 ID |
4080 | if (result) |
4081 | goto err_img_request; | |
bf0d5f50 | 4082 | |
bc1ecc65 ID |
4083 | result = rbd_img_request_submit(img_request); |
4084 | if (result) | |
4085 | goto err_img_request; | |
bf0d5f50 | 4086 | |
ed95b21a ID |
4087 | if (must_be_locked) |
4088 | up_read(&rbd_dev->lock_rwsem); | |
bc1ecc65 | 4089 | return; |
bf0d5f50 | 4090 | |
bc1ecc65 ID |
4091 | err_img_request: |
4092 | rbd_img_request_put(img_request); | |
ed95b21a ID |
4093 | err_unlock: |
4094 | if (must_be_locked) | |
4095 | up_read(&rbd_dev->lock_rwsem); | |
bc1ecc65 ID |
4096 | err_rq: |
4097 | if (result) | |
4098 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 4099 | obj_op_name(op_type), length, offset, result); |
e96a650a | 4100 | ceph_put_snap_context(snapc); |
7ad18afa CH |
4101 | err: |
4102 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 4103 | } |
bf0d5f50 | 4104 | |
7ad18afa CH |
4105 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
4106 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 4107 | { |
7ad18afa CH |
4108 | struct request *rq = bd->rq; |
4109 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 4110 | |
7ad18afa CH |
4111 | queue_work(rbd_wq, work); |
4112 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
4113 | } |
4114 | ||
602adf40 YS |
4115 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
4116 | { | |
4117 | struct gendisk *disk = rbd_dev->disk; | |
4118 | ||
4119 | if (!disk) | |
4120 | return; | |
4121 | ||
a0cab924 AE |
4122 | rbd_dev->disk = NULL; |
4123 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 4124 | del_gendisk(disk); |
a0cab924 AE |
4125 | if (disk->queue) |
4126 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 4127 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 4128 | } |
602adf40 YS |
4129 | put_disk(disk); |
4130 | } | |
4131 | ||
788e2df3 | 4132 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
fe5478e0 ID |
4133 | struct ceph_object_id *oid, |
4134 | struct ceph_object_locator *oloc, | |
4135 | void *buf, int buf_len) | |
788e2df3 AE |
4136 | |
4137 | { | |
fe5478e0 ID |
4138 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4139 | struct ceph_osd_request *req; | |
4140 | struct page **pages; | |
4141 | int num_pages = calc_pages_for(0, buf_len); | |
788e2df3 AE |
4142 | int ret; |
4143 | ||
fe5478e0 ID |
4144 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL); |
4145 | if (!req) | |
4146 | return -ENOMEM; | |
788e2df3 | 4147 | |
fe5478e0 ID |
4148 | ceph_oid_copy(&req->r_base_oid, oid); |
4149 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
4150 | req->r_flags = CEPH_OSD_FLAG_READ; | |
430c28c3 | 4151 | |
fe5478e0 | 4152 | ret = ceph_osdc_alloc_messages(req, GFP_KERNEL); |
788e2df3 | 4153 | if (ret) |
fe5478e0 | 4154 | goto out_req; |
788e2df3 | 4155 | |
fe5478e0 ID |
4156 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
4157 | if (IS_ERR(pages)) { | |
4158 | ret = PTR_ERR(pages); | |
4159 | goto out_req; | |
4160 | } | |
1ceae7ef | 4161 | |
fe5478e0 ID |
4162 | osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0); |
4163 | osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false, | |
4164 | true); | |
4165 | ||
4166 | ceph_osdc_start_request(osdc, req, false); | |
4167 | ret = ceph_osdc_wait_request(osdc, req); | |
4168 | if (ret >= 0) | |
4169 | ceph_copy_from_page_vector(pages, buf, 0, ret); | |
788e2df3 | 4170 | |
fe5478e0 ID |
4171 | out_req: |
4172 | ceph_osdc_put_request(req); | |
788e2df3 AE |
4173 | return ret; |
4174 | } | |
4175 | ||
602adf40 | 4176 | /* |
662518b1 AE |
4177 | * Read the complete header for the given rbd device. On successful |
4178 | * return, the rbd_dev->header field will contain up-to-date | |
4179 | * information about the image. | |
602adf40 | 4180 | */ |
99a41ebc | 4181 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 4182 | { |
4156d998 | 4183 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 4184 | u32 snap_count = 0; |
4156d998 AE |
4185 | u64 names_size = 0; |
4186 | u32 want_count; | |
4187 | int ret; | |
602adf40 | 4188 | |
00f1f36f | 4189 | /* |
4156d998 AE |
4190 | * The complete header will include an array of its 64-bit |
4191 | * snapshot ids, followed by the names of those snapshots as | |
4192 | * a contiguous block of NUL-terminated strings. Note that | |
4193 | * the number of snapshots could change by the time we read | |
4194 | * it in, in which case we re-read it. | |
00f1f36f | 4195 | */ |
4156d998 AE |
4196 | do { |
4197 | size_t size; | |
4198 | ||
4199 | kfree(ondisk); | |
4200 | ||
4201 | size = sizeof (*ondisk); | |
4202 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
4203 | size += names_size; | |
4204 | ondisk = kmalloc(size, GFP_KERNEL); | |
4205 | if (!ondisk) | |
662518b1 | 4206 | return -ENOMEM; |
4156d998 | 4207 | |
fe5478e0 ID |
4208 | ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid, |
4209 | &rbd_dev->header_oloc, ondisk, size); | |
4156d998 | 4210 | if (ret < 0) |
662518b1 | 4211 | goto out; |
c0cd10db | 4212 | if ((size_t)ret < size) { |
4156d998 | 4213 | ret = -ENXIO; |
06ecc6cb AE |
4214 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
4215 | size, ret); | |
662518b1 | 4216 | goto out; |
4156d998 AE |
4217 | } |
4218 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
4219 | ret = -ENXIO; | |
06ecc6cb | 4220 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 4221 | goto out; |
81e759fb | 4222 | } |
602adf40 | 4223 | |
4156d998 AE |
4224 | names_size = le64_to_cpu(ondisk->snap_names_len); |
4225 | want_count = snap_count; | |
4226 | snap_count = le32_to_cpu(ondisk->snap_count); | |
4227 | } while (snap_count != want_count); | |
00f1f36f | 4228 | |
662518b1 AE |
4229 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
4230 | out: | |
4156d998 AE |
4231 | kfree(ondisk); |
4232 | ||
4233 | return ret; | |
602adf40 YS |
4234 | } |
4235 | ||
15228ede AE |
4236 | /* |
4237 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
4238 | * has disappeared from the (just updated) snapshot context. | |
4239 | */ | |
4240 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
4241 | { | |
4242 | u64 snap_id; | |
4243 | ||
4244 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
4245 | return; | |
4246 | ||
4247 | snap_id = rbd_dev->spec->snap_id; | |
4248 | if (snap_id == CEPH_NOSNAP) | |
4249 | return; | |
4250 | ||
4251 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
4252 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
4253 | } | |
4254 | ||
9875201e JD |
4255 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
4256 | { | |
4257 | sector_t size; | |
9875201e JD |
4258 | |
4259 | /* | |
811c6688 ID |
4260 | * If EXISTS is not set, rbd_dev->disk may be NULL, so don't |
4261 | * try to update its size. If REMOVING is set, updating size | |
4262 | * is just useless work since the device can't be opened. | |
9875201e | 4263 | */ |
811c6688 ID |
4264 | if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) && |
4265 | !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) { | |
9875201e JD |
4266 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; |
4267 | dout("setting size to %llu sectors", (unsigned long long)size); | |
4268 | set_capacity(rbd_dev->disk, size); | |
4269 | revalidate_disk(rbd_dev->disk); | |
4270 | } | |
4271 | } | |
4272 | ||
cc4a38bd | 4273 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 4274 | { |
e627db08 | 4275 | u64 mapping_size; |
1fe5e993 AE |
4276 | int ret; |
4277 | ||
cfbf6377 | 4278 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 4279 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
4280 | |
4281 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 4282 | if (ret) |
73e39e4d | 4283 | goto out; |
15228ede | 4284 | |
e8f59b59 ID |
4285 | /* |
4286 | * If there is a parent, see if it has disappeared due to the | |
4287 | * mapped image getting flattened. | |
4288 | */ | |
4289 | if (rbd_dev->parent) { | |
4290 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4291 | if (ret) | |
73e39e4d | 4292 | goto out; |
e8f59b59 ID |
4293 | } |
4294 | ||
5ff1108c | 4295 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 4296 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
4297 | } else { |
4298 | /* validate mapped snapshot's EXISTS flag */ | |
4299 | rbd_exists_validate(rbd_dev); | |
4300 | } | |
15228ede | 4301 | |
73e39e4d | 4302 | out: |
cfbf6377 | 4303 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 4304 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 4305 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 4306 | |
73e39e4d | 4307 | return ret; |
1fe5e993 AE |
4308 | } |
4309 | ||
7ad18afa CH |
4310 | static int rbd_init_request(void *data, struct request *rq, |
4311 | unsigned int hctx_idx, unsigned int request_idx, | |
4312 | unsigned int numa_node) | |
4313 | { | |
4314 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
4315 | ||
4316 | INIT_WORK(work, rbd_queue_workfn); | |
4317 | return 0; | |
4318 | } | |
4319 | ||
f363b089 | 4320 | static const struct blk_mq_ops rbd_mq_ops = { |
7ad18afa | 4321 | .queue_rq = rbd_queue_rq, |
7ad18afa CH |
4322 | .init_request = rbd_init_request, |
4323 | }; | |
4324 | ||
602adf40 YS |
4325 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
4326 | { | |
4327 | struct gendisk *disk; | |
4328 | struct request_queue *q; | |
593a9e7b | 4329 | u64 segment_size; |
7ad18afa | 4330 | int err; |
602adf40 | 4331 | |
602adf40 | 4332 | /* create gendisk info */ |
7e513d43 ID |
4333 | disk = alloc_disk(single_major ? |
4334 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
4335 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 4336 | if (!disk) |
1fcdb8aa | 4337 | return -ENOMEM; |
602adf40 | 4338 | |
f0f8cef5 | 4339 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 4340 | rbd_dev->dev_id); |
602adf40 | 4341 | disk->major = rbd_dev->major; |
dd82fff1 | 4342 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
4343 | if (single_major) |
4344 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
4345 | disk->fops = &rbd_bd_ops; |
4346 | disk->private_data = rbd_dev; | |
4347 | ||
7ad18afa CH |
4348 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
4349 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 4350 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 4351 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 4352 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
4353 | rbd_dev->tag_set.nr_hw_queues = 1; |
4354 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
4355 | ||
4356 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
4357 | if (err) | |
602adf40 | 4358 | goto out_disk; |
029bcbd8 | 4359 | |
7ad18afa CH |
4360 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
4361 | if (IS_ERR(q)) { | |
4362 | err = PTR_ERR(q); | |
4363 | goto out_tag_set; | |
4364 | } | |
4365 | ||
d8a2c89c ID |
4366 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
4367 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 4368 | |
029bcbd8 | 4369 | /* set io sizes to object size */ |
593a9e7b AE |
4370 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
4371 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
0d9fde4f | 4372 | q->limits.max_sectors = queue_max_hw_sectors(q); |
d3834fef | 4373 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
4374 | blk_queue_max_segment_size(q, segment_size); |
4375 | blk_queue_io_min(q, segment_size); | |
4376 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 4377 | |
90e98c52 GZ |
4378 | /* enable the discard support */ |
4379 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
4380 | q->limits.discard_granularity = segment_size; | |
4381 | q->limits.discard_alignment = segment_size; | |
2bb4cd5c | 4382 | blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE); |
90e98c52 | 4383 | |
bae818ee | 4384 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
dc3b17cc | 4385 | q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES; |
bae818ee | 4386 | |
602adf40 YS |
4387 | disk->queue = q; |
4388 | ||
4389 | q->queuedata = rbd_dev; | |
4390 | ||
4391 | rbd_dev->disk = disk; | |
602adf40 | 4392 | |
602adf40 | 4393 | return 0; |
7ad18afa CH |
4394 | out_tag_set: |
4395 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
4396 | out_disk: |
4397 | put_disk(disk); | |
7ad18afa | 4398 | return err; |
602adf40 YS |
4399 | } |
4400 | ||
dfc5606d YS |
4401 | /* |
4402 | sysfs | |
4403 | */ | |
4404 | ||
593a9e7b AE |
4405 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
4406 | { | |
4407 | return container_of(dev, struct rbd_device, dev); | |
4408 | } | |
4409 | ||
dfc5606d YS |
4410 | static ssize_t rbd_size_show(struct device *dev, |
4411 | struct device_attribute *attr, char *buf) | |
4412 | { | |
593a9e7b | 4413 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 4414 | |
fc71d833 AE |
4415 | return sprintf(buf, "%llu\n", |
4416 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
4417 | } |
4418 | ||
34b13184 AE |
4419 | /* |
4420 | * Note this shows the features for whatever's mapped, which is not | |
4421 | * necessarily the base image. | |
4422 | */ | |
4423 | static ssize_t rbd_features_show(struct device *dev, | |
4424 | struct device_attribute *attr, char *buf) | |
4425 | { | |
4426 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4427 | ||
4428 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 4429 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
4430 | } |
4431 | ||
dfc5606d YS |
4432 | static ssize_t rbd_major_show(struct device *dev, |
4433 | struct device_attribute *attr, char *buf) | |
4434 | { | |
593a9e7b | 4435 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4436 | |
fc71d833 AE |
4437 | if (rbd_dev->major) |
4438 | return sprintf(buf, "%d\n", rbd_dev->major); | |
4439 | ||
4440 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
4441 | } |
4442 | ||
4443 | static ssize_t rbd_minor_show(struct device *dev, | |
4444 | struct device_attribute *attr, char *buf) | |
4445 | { | |
4446 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 4447 | |
dd82fff1 | 4448 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
4449 | } |
4450 | ||
005a07bf ID |
4451 | static ssize_t rbd_client_addr_show(struct device *dev, |
4452 | struct device_attribute *attr, char *buf) | |
4453 | { | |
4454 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4455 | struct ceph_entity_addr *client_addr = | |
4456 | ceph_client_addr(rbd_dev->rbd_client->client); | |
4457 | ||
4458 | return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr, | |
4459 | le32_to_cpu(client_addr->nonce)); | |
4460 | } | |
4461 | ||
dfc5606d YS |
4462 | static ssize_t rbd_client_id_show(struct device *dev, |
4463 | struct device_attribute *attr, char *buf) | |
602adf40 | 4464 | { |
593a9e7b | 4465 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4466 | |
1dbb4399 | 4467 | return sprintf(buf, "client%lld\n", |
033268a5 | 4468 | ceph_client_gid(rbd_dev->rbd_client->client)); |
602adf40 YS |
4469 | } |
4470 | ||
267fb90b MC |
4471 | static ssize_t rbd_cluster_fsid_show(struct device *dev, |
4472 | struct device_attribute *attr, char *buf) | |
4473 | { | |
4474 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4475 | ||
4476 | return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid); | |
4477 | } | |
4478 | ||
0d6d1e9c MC |
4479 | static ssize_t rbd_config_info_show(struct device *dev, |
4480 | struct device_attribute *attr, char *buf) | |
4481 | { | |
4482 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4483 | ||
4484 | return sprintf(buf, "%s\n", rbd_dev->config_info); | |
602adf40 YS |
4485 | } |
4486 | ||
dfc5606d YS |
4487 | static ssize_t rbd_pool_show(struct device *dev, |
4488 | struct device_attribute *attr, char *buf) | |
602adf40 | 4489 | { |
593a9e7b | 4490 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4491 | |
0d7dbfce | 4492 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
4493 | } |
4494 | ||
9bb2f334 AE |
4495 | static ssize_t rbd_pool_id_show(struct device *dev, |
4496 | struct device_attribute *attr, char *buf) | |
4497 | { | |
4498 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4499 | ||
0d7dbfce | 4500 | return sprintf(buf, "%llu\n", |
fc71d833 | 4501 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
4502 | } |
4503 | ||
dfc5606d YS |
4504 | static ssize_t rbd_name_show(struct device *dev, |
4505 | struct device_attribute *attr, char *buf) | |
4506 | { | |
593a9e7b | 4507 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4508 | |
a92ffdf8 AE |
4509 | if (rbd_dev->spec->image_name) |
4510 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
4511 | ||
4512 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
4513 | } |
4514 | ||
589d30e0 AE |
4515 | static ssize_t rbd_image_id_show(struct device *dev, |
4516 | struct device_attribute *attr, char *buf) | |
4517 | { | |
4518 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4519 | ||
0d7dbfce | 4520 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
4521 | } |
4522 | ||
34b13184 AE |
4523 | /* |
4524 | * Shows the name of the currently-mapped snapshot (or | |
4525 | * RBD_SNAP_HEAD_NAME for the base image). | |
4526 | */ | |
dfc5606d YS |
4527 | static ssize_t rbd_snap_show(struct device *dev, |
4528 | struct device_attribute *attr, | |
4529 | char *buf) | |
4530 | { | |
593a9e7b | 4531 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4532 | |
0d7dbfce | 4533 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
4534 | } |
4535 | ||
92a58671 MC |
4536 | static ssize_t rbd_snap_id_show(struct device *dev, |
4537 | struct device_attribute *attr, char *buf) | |
4538 | { | |
4539 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4540 | ||
4541 | return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id); | |
4542 | } | |
4543 | ||
86b00e0d | 4544 | /* |
ff96128f ID |
4545 | * For a v2 image, shows the chain of parent images, separated by empty |
4546 | * lines. For v1 images or if there is no parent, shows "(no parent | |
4547 | * image)". | |
86b00e0d AE |
4548 | */ |
4549 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
4550 | struct device_attribute *attr, |
4551 | char *buf) | |
86b00e0d AE |
4552 | { |
4553 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 4554 | ssize_t count = 0; |
86b00e0d | 4555 | |
ff96128f | 4556 | if (!rbd_dev->parent) |
86b00e0d AE |
4557 | return sprintf(buf, "(no parent image)\n"); |
4558 | ||
ff96128f ID |
4559 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
4560 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
4561 | ||
4562 | count += sprintf(&buf[count], "%s" | |
4563 | "pool_id %llu\npool_name %s\n" | |
4564 | "image_id %s\nimage_name %s\n" | |
4565 | "snap_id %llu\nsnap_name %s\n" | |
4566 | "overlap %llu\n", | |
4567 | !count ? "" : "\n", /* first? */ | |
4568 | spec->pool_id, spec->pool_name, | |
4569 | spec->image_id, spec->image_name ?: "(unknown)", | |
4570 | spec->snap_id, spec->snap_name, | |
4571 | rbd_dev->parent_overlap); | |
4572 | } | |
4573 | ||
4574 | return count; | |
86b00e0d AE |
4575 | } |
4576 | ||
dfc5606d YS |
4577 | static ssize_t rbd_image_refresh(struct device *dev, |
4578 | struct device_attribute *attr, | |
4579 | const char *buf, | |
4580 | size_t size) | |
4581 | { | |
593a9e7b | 4582 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 4583 | int ret; |
602adf40 | 4584 | |
cc4a38bd | 4585 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 4586 | if (ret) |
52bb1f9b | 4587 | return ret; |
b813623a | 4588 | |
52bb1f9b | 4589 | return size; |
dfc5606d | 4590 | } |
602adf40 | 4591 | |
dfc5606d | 4592 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 4593 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 4594 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 4595 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
005a07bf | 4596 | static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL); |
dfc5606d | 4597 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
267fb90b | 4598 | static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL); |
0d6d1e9c | 4599 | static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL); |
dfc5606d | 4600 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); |
9bb2f334 | 4601 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 4602 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 4603 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
4604 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
4605 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
92a58671 | 4606 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); |
86b00e0d | 4607 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
4608 | |
4609 | static struct attribute *rbd_attrs[] = { | |
4610 | &dev_attr_size.attr, | |
34b13184 | 4611 | &dev_attr_features.attr, |
dfc5606d | 4612 | &dev_attr_major.attr, |
dd82fff1 | 4613 | &dev_attr_minor.attr, |
005a07bf | 4614 | &dev_attr_client_addr.attr, |
dfc5606d | 4615 | &dev_attr_client_id.attr, |
267fb90b | 4616 | &dev_attr_cluster_fsid.attr, |
0d6d1e9c | 4617 | &dev_attr_config_info.attr, |
dfc5606d | 4618 | &dev_attr_pool.attr, |
9bb2f334 | 4619 | &dev_attr_pool_id.attr, |
dfc5606d | 4620 | &dev_attr_name.attr, |
589d30e0 | 4621 | &dev_attr_image_id.attr, |
dfc5606d | 4622 | &dev_attr_current_snap.attr, |
92a58671 | 4623 | &dev_attr_snap_id.attr, |
86b00e0d | 4624 | &dev_attr_parent.attr, |
dfc5606d | 4625 | &dev_attr_refresh.attr, |
dfc5606d YS |
4626 | NULL |
4627 | }; | |
4628 | ||
4629 | static struct attribute_group rbd_attr_group = { | |
4630 | .attrs = rbd_attrs, | |
4631 | }; | |
4632 | ||
4633 | static const struct attribute_group *rbd_attr_groups[] = { | |
4634 | &rbd_attr_group, | |
4635 | NULL | |
4636 | }; | |
4637 | ||
6cac4695 | 4638 | static void rbd_dev_release(struct device *dev); |
dfc5606d | 4639 | |
b9942bc9 | 4640 | static const struct device_type rbd_device_type = { |
dfc5606d YS |
4641 | .name = "rbd", |
4642 | .groups = rbd_attr_groups, | |
6cac4695 | 4643 | .release = rbd_dev_release, |
dfc5606d YS |
4644 | }; |
4645 | ||
8b8fb99c AE |
4646 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4647 | { | |
4648 | kref_get(&spec->kref); | |
4649 | ||
4650 | return spec; | |
4651 | } | |
4652 | ||
4653 | static void rbd_spec_free(struct kref *kref); | |
4654 | static void rbd_spec_put(struct rbd_spec *spec) | |
4655 | { | |
4656 | if (spec) | |
4657 | kref_put(&spec->kref, rbd_spec_free); | |
4658 | } | |
4659 | ||
4660 | static struct rbd_spec *rbd_spec_alloc(void) | |
4661 | { | |
4662 | struct rbd_spec *spec; | |
4663 | ||
4664 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4665 | if (!spec) | |
4666 | return NULL; | |
04077599 ID |
4667 | |
4668 | spec->pool_id = CEPH_NOPOOL; | |
4669 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4670 | kref_init(&spec->kref); |
4671 | ||
8b8fb99c AE |
4672 | return spec; |
4673 | } | |
4674 | ||
4675 | static void rbd_spec_free(struct kref *kref) | |
4676 | { | |
4677 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4678 | ||
4679 | kfree(spec->pool_name); | |
4680 | kfree(spec->image_id); | |
4681 | kfree(spec->image_name); | |
4682 | kfree(spec->snap_name); | |
4683 | kfree(spec); | |
4684 | } | |
4685 | ||
1643dfa4 | 4686 | static void rbd_dev_free(struct rbd_device *rbd_dev) |
dd5ac32d | 4687 | { |
99d16943 | 4688 | WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED); |
ed95b21a | 4689 | WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED); |
dd5ac32d | 4690 | |
c41d13a3 | 4691 | ceph_oid_destroy(&rbd_dev->header_oid); |
6b6dddbe | 4692 | ceph_oloc_destroy(&rbd_dev->header_oloc); |
0d6d1e9c | 4693 | kfree(rbd_dev->config_info); |
c41d13a3 | 4694 | |
dd5ac32d ID |
4695 | rbd_put_client(rbd_dev->rbd_client); |
4696 | rbd_spec_put(rbd_dev->spec); | |
4697 | kfree(rbd_dev->opts); | |
4698 | kfree(rbd_dev); | |
1643dfa4 ID |
4699 | } |
4700 | ||
4701 | static void rbd_dev_release(struct device *dev) | |
4702 | { | |
4703 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4704 | bool need_put = !!rbd_dev->opts; | |
4705 | ||
4706 | if (need_put) { | |
4707 | destroy_workqueue(rbd_dev->task_wq); | |
4708 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
4709 | } | |
4710 | ||
4711 | rbd_dev_free(rbd_dev); | |
dd5ac32d ID |
4712 | |
4713 | /* | |
4714 | * This is racy, but way better than putting module outside of | |
4715 | * the release callback. The race window is pretty small, so | |
4716 | * doing something similar to dm (dm-builtin.c) is overkill. | |
4717 | */ | |
4718 | if (need_put) | |
4719 | module_put(THIS_MODULE); | |
4720 | } | |
4721 | ||
1643dfa4 ID |
4722 | static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc, |
4723 | struct rbd_spec *spec) | |
c53d5893 AE |
4724 | { |
4725 | struct rbd_device *rbd_dev; | |
4726 | ||
1643dfa4 | 4727 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
c53d5893 AE |
4728 | if (!rbd_dev) |
4729 | return NULL; | |
4730 | ||
4731 | spin_lock_init(&rbd_dev->lock); | |
4732 | INIT_LIST_HEAD(&rbd_dev->node); | |
c53d5893 AE |
4733 | init_rwsem(&rbd_dev->header_rwsem); |
4734 | ||
7e97332e | 4735 | rbd_dev->header.data_pool_id = CEPH_NOPOOL; |
c41d13a3 | 4736 | ceph_oid_init(&rbd_dev->header_oid); |
431a02cd | 4737 | rbd_dev->header_oloc.pool = spec->pool_id; |
c41d13a3 | 4738 | |
99d16943 ID |
4739 | mutex_init(&rbd_dev->watch_mutex); |
4740 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
4741 | INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch); | |
4742 | ||
ed95b21a ID |
4743 | init_rwsem(&rbd_dev->lock_rwsem); |
4744 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; | |
4745 | INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock); | |
4746 | INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock); | |
4747 | INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock); | |
4748 | INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work); | |
4749 | init_waitqueue_head(&rbd_dev->lock_waitq); | |
4750 | ||
dd5ac32d ID |
4751 | rbd_dev->dev.bus = &rbd_bus_type; |
4752 | rbd_dev->dev.type = &rbd_device_type; | |
4753 | rbd_dev->dev.parent = &rbd_root_dev; | |
dd5ac32d ID |
4754 | device_initialize(&rbd_dev->dev); |
4755 | ||
c53d5893 | 4756 | rbd_dev->rbd_client = rbdc; |
d147543d | 4757 | rbd_dev->spec = spec; |
0903e875 | 4758 | |
1643dfa4 ID |
4759 | return rbd_dev; |
4760 | } | |
4761 | ||
4762 | /* | |
4763 | * Create a mapping rbd_dev. | |
4764 | */ | |
4765 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, | |
4766 | struct rbd_spec *spec, | |
4767 | struct rbd_options *opts) | |
4768 | { | |
4769 | struct rbd_device *rbd_dev; | |
4770 | ||
4771 | rbd_dev = __rbd_dev_create(rbdc, spec); | |
4772 | if (!rbd_dev) | |
4773 | return NULL; | |
4774 | ||
4775 | rbd_dev->opts = opts; | |
4776 | ||
4777 | /* get an id and fill in device name */ | |
4778 | rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0, | |
4779 | minor_to_rbd_dev_id(1 << MINORBITS), | |
4780 | GFP_KERNEL); | |
4781 | if (rbd_dev->dev_id < 0) | |
4782 | goto fail_rbd_dev; | |
4783 | ||
4784 | sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id); | |
4785 | rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM, | |
4786 | rbd_dev->name); | |
4787 | if (!rbd_dev->task_wq) | |
4788 | goto fail_dev_id; | |
dd5ac32d | 4789 | |
1643dfa4 ID |
4790 | /* we have a ref from do_rbd_add() */ |
4791 | __module_get(THIS_MODULE); | |
dd5ac32d | 4792 | |
1643dfa4 | 4793 | dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id); |
c53d5893 | 4794 | return rbd_dev; |
1643dfa4 ID |
4795 | |
4796 | fail_dev_id: | |
4797 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
4798 | fail_rbd_dev: | |
4799 | rbd_dev_free(rbd_dev); | |
4800 | return NULL; | |
c53d5893 AE |
4801 | } |
4802 | ||
4803 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4804 | { | |
dd5ac32d ID |
4805 | if (rbd_dev) |
4806 | put_device(&rbd_dev->dev); | |
c53d5893 AE |
4807 | } |
4808 | ||
9d475de5 AE |
4809 | /* |
4810 | * Get the size and object order for an image snapshot, or if | |
4811 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4812 | * image. | |
4813 | */ | |
4814 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4815 | u8 *order, u64 *snap_size) | |
4816 | { | |
4817 | __le64 snapid = cpu_to_le64(snap_id); | |
4818 | int ret; | |
4819 | struct { | |
4820 | u8 order; | |
4821 | __le64 size; | |
4822 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4823 | ||
ecd4a68a ID |
4824 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
4825 | &rbd_dev->header_oloc, "get_size", | |
4826 | &snapid, sizeof(snapid), | |
4827 | &size_buf, sizeof(size_buf)); | |
36be9a76 | 4828 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4829 | if (ret < 0) |
4830 | return ret; | |
57385b51 AE |
4831 | if (ret < sizeof (size_buf)) |
4832 | return -ERANGE; | |
9d475de5 | 4833 | |
c3545579 | 4834 | if (order) { |
c86f86e9 | 4835 | *order = size_buf.order; |
c3545579 JD |
4836 | dout(" order %u", (unsigned int)*order); |
4837 | } | |
9d475de5 AE |
4838 | *snap_size = le64_to_cpu(size_buf.size); |
4839 | ||
c3545579 JD |
4840 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4841 | (unsigned long long)snap_id, | |
57385b51 | 4842 | (unsigned long long)*snap_size); |
9d475de5 AE |
4843 | |
4844 | return 0; | |
4845 | } | |
4846 | ||
4847 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4848 | { | |
4849 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4850 | &rbd_dev->header.obj_order, | |
4851 | &rbd_dev->header.image_size); | |
4852 | } | |
4853 | ||
1e130199 AE |
4854 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4855 | { | |
4856 | void *reply_buf; | |
4857 | int ret; | |
4858 | void *p; | |
4859 | ||
4860 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4861 | if (!reply_buf) | |
4862 | return -ENOMEM; | |
4863 | ||
ecd4a68a ID |
4864 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
4865 | &rbd_dev->header_oloc, "get_object_prefix", | |
4866 | NULL, 0, reply_buf, RBD_OBJ_PREFIX_LEN_MAX); | |
36be9a76 | 4867 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4868 | if (ret < 0) |
4869 | goto out; | |
4870 | ||
4871 | p = reply_buf; | |
4872 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4873 | p + ret, NULL, GFP_NOIO); |
4874 | ret = 0; | |
1e130199 AE |
4875 | |
4876 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4877 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4878 | rbd_dev->header.object_prefix = NULL; | |
4879 | } else { | |
4880 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4881 | } | |
1e130199 AE |
4882 | out: |
4883 | kfree(reply_buf); | |
4884 | ||
4885 | return ret; | |
4886 | } | |
4887 | ||
b1b5402a AE |
4888 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4889 | u64 *snap_features) | |
4890 | { | |
4891 | __le64 snapid = cpu_to_le64(snap_id); | |
4892 | struct { | |
4893 | __le64 features; | |
4894 | __le64 incompat; | |
4157976b | 4895 | } __attribute__ ((packed)) features_buf = { 0 }; |
d3767f0f | 4896 | u64 unsup; |
b1b5402a AE |
4897 | int ret; |
4898 | ||
ecd4a68a ID |
4899 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
4900 | &rbd_dev->header_oloc, "get_features", | |
4901 | &snapid, sizeof(snapid), | |
4902 | &features_buf, sizeof(features_buf)); | |
36be9a76 | 4903 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4904 | if (ret < 0) |
4905 | return ret; | |
57385b51 AE |
4906 | if (ret < sizeof (features_buf)) |
4907 | return -ERANGE; | |
d889140c | 4908 | |
d3767f0f ID |
4909 | unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED; |
4910 | if (unsup) { | |
4911 | rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx", | |
4912 | unsup); | |
b8f5c6ed | 4913 | return -ENXIO; |
d3767f0f | 4914 | } |
d889140c | 4915 | |
b1b5402a AE |
4916 | *snap_features = le64_to_cpu(features_buf.features); |
4917 | ||
4918 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4919 | (unsigned long long)snap_id, |
4920 | (unsigned long long)*snap_features, | |
4921 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4922 | |
4923 | return 0; | |
4924 | } | |
4925 | ||
4926 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4927 | { | |
4928 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4929 | &rbd_dev->header.features); | |
4930 | } | |
4931 | ||
86b00e0d AE |
4932 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4933 | { | |
4934 | struct rbd_spec *parent_spec; | |
4935 | size_t size; | |
4936 | void *reply_buf = NULL; | |
4937 | __le64 snapid; | |
4938 | void *p; | |
4939 | void *end; | |
642a2537 | 4940 | u64 pool_id; |
86b00e0d | 4941 | char *image_id; |
3b5cf2a2 | 4942 | u64 snap_id; |
86b00e0d | 4943 | u64 overlap; |
86b00e0d AE |
4944 | int ret; |
4945 | ||
4946 | parent_spec = rbd_spec_alloc(); | |
4947 | if (!parent_spec) | |
4948 | return -ENOMEM; | |
4949 | ||
4950 | size = sizeof (__le64) + /* pool_id */ | |
4951 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4952 | sizeof (__le64) + /* snap_id */ | |
4953 | sizeof (__le64); /* overlap */ | |
4954 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4955 | if (!reply_buf) { | |
4956 | ret = -ENOMEM; | |
4957 | goto out_err; | |
4958 | } | |
4959 | ||
4d9b67cd | 4960 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
ecd4a68a ID |
4961 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
4962 | &rbd_dev->header_oloc, "get_parent", | |
4963 | &snapid, sizeof(snapid), reply_buf, size); | |
36be9a76 | 4964 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4965 | if (ret < 0) |
4966 | goto out_err; | |
4967 | ||
86b00e0d | 4968 | p = reply_buf; |
57385b51 AE |
4969 | end = reply_buf + ret; |
4970 | ret = -ERANGE; | |
642a2537 | 4971 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4972 | if (pool_id == CEPH_NOPOOL) { |
4973 | /* | |
4974 | * Either the parent never existed, or we have | |
4975 | * record of it but the image got flattened so it no | |
4976 | * longer has a parent. When the parent of a | |
4977 | * layered image disappears we immediately set the | |
4978 | * overlap to 0. The effect of this is that all new | |
4979 | * requests will be treated as if the image had no | |
4980 | * parent. | |
4981 | */ | |
4982 | if (rbd_dev->parent_overlap) { | |
4983 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4984 | rbd_dev_parent_put(rbd_dev); |
4985 | pr_info("%s: clone image has been flattened\n", | |
4986 | rbd_dev->disk->disk_name); | |
4987 | } | |
4988 | ||
86b00e0d | 4989 | goto out; /* No parent? No problem. */ |
392a9dad | 4990 | } |
86b00e0d | 4991 | |
0903e875 AE |
4992 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4993 | ||
4994 | ret = -EIO; | |
642a2537 | 4995 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4996 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4997 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4998 | goto out_err; |
c0cd10db | 4999 | } |
0903e875 | 5000 | |
979ed480 | 5001 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
5002 | if (IS_ERR(image_id)) { |
5003 | ret = PTR_ERR(image_id); | |
5004 | goto out_err; | |
5005 | } | |
3b5cf2a2 | 5006 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
5007 | ceph_decode_64_safe(&p, end, overlap, out_err); |
5008 | ||
3b5cf2a2 AE |
5009 | /* |
5010 | * The parent won't change (except when the clone is | |
5011 | * flattened, already handled that). So we only need to | |
5012 | * record the parent spec we have not already done so. | |
5013 | */ | |
5014 | if (!rbd_dev->parent_spec) { | |
5015 | parent_spec->pool_id = pool_id; | |
5016 | parent_spec->image_id = image_id; | |
5017 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
5018 | rbd_dev->parent_spec = parent_spec; |
5019 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
5020 | } else { |
5021 | kfree(image_id); | |
3b5cf2a2 AE |
5022 | } |
5023 | ||
5024 | /* | |
cf32bd9c ID |
5025 | * We always update the parent overlap. If it's zero we issue |
5026 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 5027 | */ |
3b5cf2a2 | 5028 | if (!overlap) { |
3b5cf2a2 | 5029 | if (parent_spec) { |
cf32bd9c ID |
5030 | /* refresh, careful to warn just once */ |
5031 | if (rbd_dev->parent_overlap) | |
5032 | rbd_warn(rbd_dev, | |
5033 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 5034 | } else { |
cf32bd9c ID |
5035 | /* initial probe */ |
5036 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 5037 | } |
70cf49cf | 5038 | } |
cf32bd9c ID |
5039 | rbd_dev->parent_overlap = overlap; |
5040 | ||
86b00e0d AE |
5041 | out: |
5042 | ret = 0; | |
5043 | out_err: | |
5044 | kfree(reply_buf); | |
5045 | rbd_spec_put(parent_spec); | |
5046 | ||
5047 | return ret; | |
5048 | } | |
5049 | ||
cc070d59 AE |
5050 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
5051 | { | |
5052 | struct { | |
5053 | __le64 stripe_unit; | |
5054 | __le64 stripe_count; | |
5055 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
5056 | size_t size = sizeof (striping_info_buf); | |
5057 | void *p; | |
5058 | u64 obj_size; | |
5059 | u64 stripe_unit; | |
5060 | u64 stripe_count; | |
5061 | int ret; | |
5062 | ||
ecd4a68a ID |
5063 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5064 | &rbd_dev->header_oloc, "get_stripe_unit_count", | |
5065 | NULL, 0, &striping_info_buf, size); | |
cc070d59 AE |
5066 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
5067 | if (ret < 0) | |
5068 | return ret; | |
5069 | if (ret < size) | |
5070 | return -ERANGE; | |
5071 | ||
5072 | /* | |
5073 | * We don't actually support the "fancy striping" feature | |
5074 | * (STRIPINGV2) yet, but if the striping sizes are the | |
5075 | * defaults the behavior is the same as before. So find | |
5076 | * out, and only fail if the image has non-default values. | |
5077 | */ | |
5078 | ret = -EINVAL; | |
5bc3fb17 | 5079 | obj_size = rbd_obj_bytes(&rbd_dev->header); |
cc070d59 AE |
5080 | p = &striping_info_buf; |
5081 | stripe_unit = ceph_decode_64(&p); | |
5082 | if (stripe_unit != obj_size) { | |
5083 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
5084 | "(got %llu want %llu)", | |
5085 | stripe_unit, obj_size); | |
5086 | return -EINVAL; | |
5087 | } | |
5088 | stripe_count = ceph_decode_64(&p); | |
5089 | if (stripe_count != 1) { | |
5090 | rbd_warn(rbd_dev, "unsupported stripe count " | |
5091 | "(got %llu want 1)", stripe_count); | |
5092 | return -EINVAL; | |
5093 | } | |
500d0c0f AE |
5094 | rbd_dev->header.stripe_unit = stripe_unit; |
5095 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
5096 | |
5097 | return 0; | |
5098 | } | |
5099 | ||
7e97332e ID |
5100 | static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev) |
5101 | { | |
5102 | __le64 data_pool_id; | |
5103 | int ret; | |
5104 | ||
5105 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, | |
5106 | &rbd_dev->header_oloc, "get_data_pool", | |
5107 | NULL, 0, &data_pool_id, sizeof(data_pool_id)); | |
5108 | if (ret < 0) | |
5109 | return ret; | |
5110 | if (ret < sizeof(data_pool_id)) | |
5111 | return -EBADMSG; | |
5112 | ||
5113 | rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id); | |
5114 | WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL); | |
5115 | return 0; | |
5116 | } | |
5117 | ||
9e15b77d AE |
5118 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
5119 | { | |
ecd4a68a | 5120 | CEPH_DEFINE_OID_ONSTACK(oid); |
9e15b77d AE |
5121 | size_t image_id_size; |
5122 | char *image_id; | |
5123 | void *p; | |
5124 | void *end; | |
5125 | size_t size; | |
5126 | void *reply_buf = NULL; | |
5127 | size_t len = 0; | |
5128 | char *image_name = NULL; | |
5129 | int ret; | |
5130 | ||
5131 | rbd_assert(!rbd_dev->spec->image_name); | |
5132 | ||
69e7a02f AE |
5133 | len = strlen(rbd_dev->spec->image_id); |
5134 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
5135 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
5136 | if (!image_id) | |
5137 | return NULL; | |
5138 | ||
5139 | p = image_id; | |
4157976b | 5140 | end = image_id + image_id_size; |
57385b51 | 5141 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
5142 | |
5143 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
5144 | reply_buf = kmalloc(size, GFP_KERNEL); | |
5145 | if (!reply_buf) | |
5146 | goto out; | |
5147 | ||
ecd4a68a ID |
5148 | ceph_oid_printf(&oid, "%s", RBD_DIRECTORY); |
5149 | ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc, | |
5150 | "dir_get_name", image_id, image_id_size, | |
5151 | reply_buf, size); | |
9e15b77d AE |
5152 | if (ret < 0) |
5153 | goto out; | |
5154 | p = reply_buf; | |
f40eb349 AE |
5155 | end = reply_buf + ret; |
5156 | ||
9e15b77d AE |
5157 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
5158 | if (IS_ERR(image_name)) | |
5159 | image_name = NULL; | |
5160 | else | |
5161 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
5162 | out: | |
5163 | kfree(reply_buf); | |
5164 | kfree(image_id); | |
5165 | ||
5166 | return image_name; | |
5167 | } | |
5168 | ||
2ad3d716 AE |
5169 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
5170 | { | |
5171 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
5172 | const char *snap_name; | |
5173 | u32 which = 0; | |
5174 | ||
5175 | /* Skip over names until we find the one we are looking for */ | |
5176 | ||
5177 | snap_name = rbd_dev->header.snap_names; | |
5178 | while (which < snapc->num_snaps) { | |
5179 | if (!strcmp(name, snap_name)) | |
5180 | return snapc->snaps[which]; | |
5181 | snap_name += strlen(snap_name) + 1; | |
5182 | which++; | |
5183 | } | |
5184 | return CEPH_NOSNAP; | |
5185 | } | |
5186 | ||
5187 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
5188 | { | |
5189 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
5190 | u32 which; | |
5191 | bool found = false; | |
5192 | u64 snap_id; | |
5193 | ||
5194 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
5195 | const char *snap_name; | |
5196 | ||
5197 | snap_id = snapc->snaps[which]; | |
5198 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
5199 | if (IS_ERR(snap_name)) { |
5200 | /* ignore no-longer existing snapshots */ | |
5201 | if (PTR_ERR(snap_name) == -ENOENT) | |
5202 | continue; | |
5203 | else | |
5204 | break; | |
5205 | } | |
2ad3d716 AE |
5206 | found = !strcmp(name, snap_name); |
5207 | kfree(snap_name); | |
5208 | } | |
5209 | return found ? snap_id : CEPH_NOSNAP; | |
5210 | } | |
5211 | ||
5212 | /* | |
5213 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
5214 | * no snapshot by that name is found, or if an error occurs. | |
5215 | */ | |
5216 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
5217 | { | |
5218 | if (rbd_dev->image_format == 1) | |
5219 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
5220 | ||
5221 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
5222 | } | |
5223 | ||
9e15b77d | 5224 | /* |
04077599 ID |
5225 | * An image being mapped will have everything but the snap id. |
5226 | */ | |
5227 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
5228 | { | |
5229 | struct rbd_spec *spec = rbd_dev->spec; | |
5230 | ||
5231 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
5232 | rbd_assert(spec->image_id && spec->image_name); | |
5233 | rbd_assert(spec->snap_name); | |
5234 | ||
5235 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
5236 | u64 snap_id; | |
5237 | ||
5238 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
5239 | if (snap_id == CEPH_NOSNAP) | |
5240 | return -ENOENT; | |
5241 | ||
5242 | spec->snap_id = snap_id; | |
5243 | } else { | |
5244 | spec->snap_id = CEPH_NOSNAP; | |
5245 | } | |
5246 | ||
5247 | return 0; | |
5248 | } | |
5249 | ||
5250 | /* | |
5251 | * A parent image will have all ids but none of the names. | |
e1d4213f | 5252 | * |
04077599 ID |
5253 | * All names in an rbd spec are dynamically allocated. It's OK if we |
5254 | * can't figure out the name for an image id. | |
9e15b77d | 5255 | */ |
04077599 | 5256 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 5257 | { |
2e9f7f1c AE |
5258 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
5259 | struct rbd_spec *spec = rbd_dev->spec; | |
5260 | const char *pool_name; | |
5261 | const char *image_name; | |
5262 | const char *snap_name; | |
9e15b77d AE |
5263 | int ret; |
5264 | ||
04077599 ID |
5265 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
5266 | rbd_assert(spec->image_id); | |
5267 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 5268 | |
2e9f7f1c | 5269 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 5270 | |
2e9f7f1c AE |
5271 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
5272 | if (!pool_name) { | |
5273 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
5274 | return -EIO; |
5275 | } | |
2e9f7f1c AE |
5276 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
5277 | if (!pool_name) | |
9e15b77d AE |
5278 | return -ENOMEM; |
5279 | ||
5280 | /* Fetch the image name; tolerate failure here */ | |
5281 | ||
2e9f7f1c AE |
5282 | image_name = rbd_dev_image_name(rbd_dev); |
5283 | if (!image_name) | |
06ecc6cb | 5284 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 5285 | |
04077599 | 5286 | /* Fetch the snapshot name */ |
9e15b77d | 5287 | |
2e9f7f1c | 5288 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
5289 | if (IS_ERR(snap_name)) { |
5290 | ret = PTR_ERR(snap_name); | |
9e15b77d | 5291 | goto out_err; |
2e9f7f1c AE |
5292 | } |
5293 | ||
5294 | spec->pool_name = pool_name; | |
5295 | spec->image_name = image_name; | |
5296 | spec->snap_name = snap_name; | |
9e15b77d AE |
5297 | |
5298 | return 0; | |
04077599 | 5299 | |
9e15b77d | 5300 | out_err: |
2e9f7f1c AE |
5301 | kfree(image_name); |
5302 | kfree(pool_name); | |
9e15b77d AE |
5303 | return ret; |
5304 | } | |
5305 | ||
cc4a38bd | 5306 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
5307 | { |
5308 | size_t size; | |
5309 | int ret; | |
5310 | void *reply_buf; | |
5311 | void *p; | |
5312 | void *end; | |
5313 | u64 seq; | |
5314 | u32 snap_count; | |
5315 | struct ceph_snap_context *snapc; | |
5316 | u32 i; | |
5317 | ||
5318 | /* | |
5319 | * We'll need room for the seq value (maximum snapshot id), | |
5320 | * snapshot count, and array of that many snapshot ids. | |
5321 | * For now we have a fixed upper limit on the number we're | |
5322 | * prepared to receive. | |
5323 | */ | |
5324 | size = sizeof (__le64) + sizeof (__le32) + | |
5325 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
5326 | reply_buf = kzalloc(size, GFP_KERNEL); | |
5327 | if (!reply_buf) | |
5328 | return -ENOMEM; | |
5329 | ||
ecd4a68a ID |
5330 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5331 | &rbd_dev->header_oloc, "get_snapcontext", | |
5332 | NULL, 0, reply_buf, size); | |
36be9a76 | 5333 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
5334 | if (ret < 0) |
5335 | goto out; | |
5336 | ||
35d489f9 | 5337 | p = reply_buf; |
57385b51 AE |
5338 | end = reply_buf + ret; |
5339 | ret = -ERANGE; | |
35d489f9 AE |
5340 | ceph_decode_64_safe(&p, end, seq, out); |
5341 | ceph_decode_32_safe(&p, end, snap_count, out); | |
5342 | ||
5343 | /* | |
5344 | * Make sure the reported number of snapshot ids wouldn't go | |
5345 | * beyond the end of our buffer. But before checking that, | |
5346 | * make sure the computed size of the snapshot context we | |
5347 | * allocate is representable in a size_t. | |
5348 | */ | |
5349 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
5350 | / sizeof (u64)) { | |
5351 | ret = -EINVAL; | |
5352 | goto out; | |
5353 | } | |
5354 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
5355 | goto out; | |
468521c1 | 5356 | ret = 0; |
35d489f9 | 5357 | |
812164f8 | 5358 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
5359 | if (!snapc) { |
5360 | ret = -ENOMEM; | |
5361 | goto out; | |
5362 | } | |
35d489f9 | 5363 | snapc->seq = seq; |
35d489f9 AE |
5364 | for (i = 0; i < snap_count; i++) |
5365 | snapc->snaps[i] = ceph_decode_64(&p); | |
5366 | ||
49ece554 | 5367 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
5368 | rbd_dev->header.snapc = snapc; |
5369 | ||
5370 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 5371 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
5372 | out: |
5373 | kfree(reply_buf); | |
5374 | ||
57385b51 | 5375 | return ret; |
35d489f9 AE |
5376 | } |
5377 | ||
54cac61f AE |
5378 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
5379 | u64 snap_id) | |
b8b1e2db AE |
5380 | { |
5381 | size_t size; | |
5382 | void *reply_buf; | |
54cac61f | 5383 | __le64 snapid; |
b8b1e2db AE |
5384 | int ret; |
5385 | void *p; | |
5386 | void *end; | |
b8b1e2db AE |
5387 | char *snap_name; |
5388 | ||
5389 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
5390 | reply_buf = kmalloc(size, GFP_KERNEL); | |
5391 | if (!reply_buf) | |
5392 | return ERR_PTR(-ENOMEM); | |
5393 | ||
54cac61f | 5394 | snapid = cpu_to_le64(snap_id); |
ecd4a68a ID |
5395 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5396 | &rbd_dev->header_oloc, "get_snapshot_name", | |
5397 | &snapid, sizeof(snapid), reply_buf, size); | |
36be9a76 | 5398 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
5399 | if (ret < 0) { |
5400 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 5401 | goto out; |
f40eb349 | 5402 | } |
b8b1e2db AE |
5403 | |
5404 | p = reply_buf; | |
f40eb349 | 5405 | end = reply_buf + ret; |
e5c35534 | 5406 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 5407 | if (IS_ERR(snap_name)) |
b8b1e2db | 5408 | goto out; |
b8b1e2db | 5409 | |
f40eb349 | 5410 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 5411 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
5412 | out: |
5413 | kfree(reply_buf); | |
5414 | ||
f40eb349 | 5415 | return snap_name; |
b8b1e2db AE |
5416 | } |
5417 | ||
2df3fac7 | 5418 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 5419 | { |
2df3fac7 | 5420 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 5421 | int ret; |
117973fb | 5422 | |
1617e40c JD |
5423 | ret = rbd_dev_v2_image_size(rbd_dev); |
5424 | if (ret) | |
cfbf6377 | 5425 | return ret; |
1617e40c | 5426 | |
2df3fac7 AE |
5427 | if (first_time) { |
5428 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
5429 | if (ret) | |
cfbf6377 | 5430 | return ret; |
2df3fac7 AE |
5431 | } |
5432 | ||
cc4a38bd | 5433 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
5434 | if (ret && first_time) { |
5435 | kfree(rbd_dev->header.object_prefix); | |
5436 | rbd_dev->header.object_prefix = NULL; | |
5437 | } | |
117973fb AE |
5438 | |
5439 | return ret; | |
5440 | } | |
5441 | ||
a720ae09 ID |
5442 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
5443 | { | |
5444 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5445 | ||
5446 | if (rbd_dev->image_format == 1) | |
5447 | return rbd_dev_v1_header_info(rbd_dev); | |
5448 | ||
5449 | return rbd_dev_v2_header_info(rbd_dev); | |
5450 | } | |
5451 | ||
e28fff26 AE |
5452 | /* |
5453 | * Skips over white space at *buf, and updates *buf to point to the | |
5454 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
5455 | * the token (string of non-white space characters) found. Note |
5456 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
5457 | */ |
5458 | static inline size_t next_token(const char **buf) | |
5459 | { | |
5460 | /* | |
5461 | * These are the characters that produce nonzero for | |
5462 | * isspace() in the "C" and "POSIX" locales. | |
5463 | */ | |
5464 | const char *spaces = " \f\n\r\t\v"; | |
5465 | ||
5466 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
5467 | ||
5468 | return strcspn(*buf, spaces); /* Return token length */ | |
5469 | } | |
5470 | ||
ea3352f4 AE |
5471 | /* |
5472 | * Finds the next token in *buf, dynamically allocates a buffer big | |
5473 | * enough to hold a copy of it, and copies the token into the new | |
5474 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
5475 | * that a duplicate buffer is created even for a zero-length token. | |
5476 | * | |
5477 | * Returns a pointer to the newly-allocated duplicate, or a null | |
5478 | * pointer if memory for the duplicate was not available. If | |
5479 | * the lenp argument is a non-null pointer, the length of the token | |
5480 | * (not including the '\0') is returned in *lenp. | |
5481 | * | |
5482 | * If successful, the *buf pointer will be updated to point beyond | |
5483 | * the end of the found token. | |
5484 | * | |
5485 | * Note: uses GFP_KERNEL for allocation. | |
5486 | */ | |
5487 | static inline char *dup_token(const char **buf, size_t *lenp) | |
5488 | { | |
5489 | char *dup; | |
5490 | size_t len; | |
5491 | ||
5492 | len = next_token(buf); | |
4caf35f9 | 5493 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
5494 | if (!dup) |
5495 | return NULL; | |
ea3352f4 AE |
5496 | *(dup + len) = '\0'; |
5497 | *buf += len; | |
5498 | ||
5499 | if (lenp) | |
5500 | *lenp = len; | |
5501 | ||
5502 | return dup; | |
5503 | } | |
5504 | ||
a725f65e | 5505 | /* |
859c31df AE |
5506 | * Parse the options provided for an "rbd add" (i.e., rbd image |
5507 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
5508 | * and the data written is passed here via a NUL-terminated buffer. | |
5509 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 5510 | * |
859c31df AE |
5511 | * The information extracted from these options is recorded in |
5512 | * the other parameters which return dynamically-allocated | |
5513 | * structures: | |
5514 | * ceph_opts | |
5515 | * The address of a pointer that will refer to a ceph options | |
5516 | * structure. Caller must release the returned pointer using | |
5517 | * ceph_destroy_options() when it is no longer needed. | |
5518 | * rbd_opts | |
5519 | * Address of an rbd options pointer. Fully initialized by | |
5520 | * this function; caller must release with kfree(). | |
5521 | * spec | |
5522 | * Address of an rbd image specification pointer. Fully | |
5523 | * initialized by this function based on parsed options. | |
5524 | * Caller must release with rbd_spec_put(). | |
5525 | * | |
5526 | * The options passed take this form: | |
5527 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
5528 | * where: | |
5529 | * <mon_addrs> | |
5530 | * A comma-separated list of one or more monitor addresses. | |
5531 | * A monitor address is an ip address, optionally followed | |
5532 | * by a port number (separated by a colon). | |
5533 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
5534 | * <options> | |
5535 | * A comma-separated list of ceph and/or rbd options. | |
5536 | * <pool_name> | |
5537 | * The name of the rados pool containing the rbd image. | |
5538 | * <image_name> | |
5539 | * The name of the image in that pool to map. | |
5540 | * <snap_id> | |
5541 | * An optional snapshot id. If provided, the mapping will | |
5542 | * present data from the image at the time that snapshot was | |
5543 | * created. The image head is used if no snapshot id is | |
5544 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 5545 | */ |
859c31df | 5546 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 5547 | struct ceph_options **ceph_opts, |
859c31df AE |
5548 | struct rbd_options **opts, |
5549 | struct rbd_spec **rbd_spec) | |
e28fff26 | 5550 | { |
d22f76e7 | 5551 | size_t len; |
859c31df | 5552 | char *options; |
0ddebc0c | 5553 | const char *mon_addrs; |
ecb4dc22 | 5554 | char *snap_name; |
0ddebc0c | 5555 | size_t mon_addrs_size; |
859c31df | 5556 | struct rbd_spec *spec = NULL; |
4e9afeba | 5557 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5558 | struct ceph_options *copts; |
dc79b113 | 5559 | int ret; |
e28fff26 AE |
5560 | |
5561 | /* The first four tokens are required */ | |
5562 | ||
7ef3214a | 5563 | len = next_token(&buf); |
4fb5d671 AE |
5564 | if (!len) { |
5565 | rbd_warn(NULL, "no monitor address(es) provided"); | |
5566 | return -EINVAL; | |
5567 | } | |
0ddebc0c | 5568 | mon_addrs = buf; |
f28e565a | 5569 | mon_addrs_size = len + 1; |
7ef3214a | 5570 | buf += len; |
a725f65e | 5571 | |
dc79b113 | 5572 | ret = -EINVAL; |
f28e565a AE |
5573 | options = dup_token(&buf, NULL); |
5574 | if (!options) | |
dc79b113 | 5575 | return -ENOMEM; |
4fb5d671 AE |
5576 | if (!*options) { |
5577 | rbd_warn(NULL, "no options provided"); | |
5578 | goto out_err; | |
5579 | } | |
e28fff26 | 5580 | |
859c31df AE |
5581 | spec = rbd_spec_alloc(); |
5582 | if (!spec) | |
f28e565a | 5583 | goto out_mem; |
859c31df AE |
5584 | |
5585 | spec->pool_name = dup_token(&buf, NULL); | |
5586 | if (!spec->pool_name) | |
5587 | goto out_mem; | |
4fb5d671 AE |
5588 | if (!*spec->pool_name) { |
5589 | rbd_warn(NULL, "no pool name provided"); | |
5590 | goto out_err; | |
5591 | } | |
e28fff26 | 5592 | |
69e7a02f | 5593 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 5594 | if (!spec->image_name) |
f28e565a | 5595 | goto out_mem; |
4fb5d671 AE |
5596 | if (!*spec->image_name) { |
5597 | rbd_warn(NULL, "no image name provided"); | |
5598 | goto out_err; | |
5599 | } | |
d4b125e9 | 5600 | |
f28e565a AE |
5601 | /* |
5602 | * Snapshot name is optional; default is to use "-" | |
5603 | * (indicating the head/no snapshot). | |
5604 | */ | |
3feeb894 | 5605 | len = next_token(&buf); |
820a5f3e | 5606 | if (!len) { |
3feeb894 AE |
5607 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
5608 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 5609 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 5610 | ret = -ENAMETOOLONG; |
f28e565a | 5611 | goto out_err; |
849b4260 | 5612 | } |
ecb4dc22 AE |
5613 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
5614 | if (!snap_name) | |
f28e565a | 5615 | goto out_mem; |
ecb4dc22 AE |
5616 | *(snap_name + len) = '\0'; |
5617 | spec->snap_name = snap_name; | |
e5c35534 | 5618 | |
0ddebc0c | 5619 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 5620 | |
4e9afeba AE |
5621 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
5622 | if (!rbd_opts) | |
5623 | goto out_mem; | |
5624 | ||
5625 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 5626 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
80de1912 | 5627 | rbd_opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT; |
d22f76e7 | 5628 | |
859c31df | 5629 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 5630 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 5631 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
5632 | if (IS_ERR(copts)) { |
5633 | ret = PTR_ERR(copts); | |
dc79b113 AE |
5634 | goto out_err; |
5635 | } | |
859c31df AE |
5636 | kfree(options); |
5637 | ||
5638 | *ceph_opts = copts; | |
4e9afeba | 5639 | *opts = rbd_opts; |
859c31df | 5640 | *rbd_spec = spec; |
0ddebc0c | 5641 | |
dc79b113 | 5642 | return 0; |
f28e565a | 5643 | out_mem: |
dc79b113 | 5644 | ret = -ENOMEM; |
d22f76e7 | 5645 | out_err: |
859c31df AE |
5646 | kfree(rbd_opts); |
5647 | rbd_spec_put(spec); | |
f28e565a | 5648 | kfree(options); |
d22f76e7 | 5649 | |
dc79b113 | 5650 | return ret; |
a725f65e AE |
5651 | } |
5652 | ||
30ba1f02 ID |
5653 | /* |
5654 | * Return pool id (>= 0) or a negative error code. | |
5655 | */ | |
5656 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
5657 | { | |
a319bf56 | 5658 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 5659 | u64 newest_epoch; |
30ba1f02 ID |
5660 | int tries = 0; |
5661 | int ret; | |
5662 | ||
5663 | again: | |
5664 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
5665 | if (ret == -ENOENT && tries++ < 1) { | |
d0b19705 ID |
5666 | ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap", |
5667 | &newest_epoch); | |
30ba1f02 ID |
5668 | if (ret < 0) |
5669 | return ret; | |
5670 | ||
5671 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
7cca78c9 | 5672 | ceph_osdc_maybe_request_map(&rbdc->client->osdc); |
30ba1f02 | 5673 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, |
a319bf56 ID |
5674 | newest_epoch, |
5675 | opts->mount_timeout); | |
30ba1f02 ID |
5676 | goto again; |
5677 | } else { | |
5678 | /* the osdmap we have is new enough */ | |
5679 | return -ENOENT; | |
5680 | } | |
5681 | } | |
5682 | ||
5683 | return ret; | |
5684 | } | |
5685 | ||
589d30e0 AE |
5686 | /* |
5687 | * An rbd format 2 image has a unique identifier, distinct from the | |
5688 | * name given to it by the user. Internally, that identifier is | |
5689 | * what's used to specify the names of objects related to the image. | |
5690 | * | |
5691 | * A special "rbd id" object is used to map an rbd image name to its | |
5692 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5693 | * with the supplied name. | |
5694 | * | |
5695 | * This function will record the given rbd_dev's image_id field if | |
5696 | * it can be determined, and in that case will return 0. If any | |
5697 | * errors occur a negative errno will be returned and the rbd_dev's | |
5698 | * image_id field will be unchanged (and should be NULL). | |
5699 | */ | |
5700 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5701 | { | |
5702 | int ret; | |
5703 | size_t size; | |
ecd4a68a | 5704 | CEPH_DEFINE_OID_ONSTACK(oid); |
589d30e0 | 5705 | void *response; |
c0fba368 | 5706 | char *image_id; |
2f82ee54 | 5707 | |
2c0d0a10 AE |
5708 | /* |
5709 | * When probing a parent image, the image id is already | |
5710 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5711 | * need to fetch the image id again in this case. We |
5712 | * do still need to set the image format though. | |
2c0d0a10 | 5713 | */ |
c0fba368 AE |
5714 | if (rbd_dev->spec->image_id) { |
5715 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5716 | ||
2c0d0a10 | 5717 | return 0; |
c0fba368 | 5718 | } |
2c0d0a10 | 5719 | |
589d30e0 AE |
5720 | /* |
5721 | * First, see if the format 2 image id file exists, and if | |
5722 | * so, get the image's persistent id from it. | |
5723 | */ | |
ecd4a68a ID |
5724 | ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX, |
5725 | rbd_dev->spec->image_name); | |
5726 | if (ret) | |
5727 | return ret; | |
5728 | ||
5729 | dout("rbd id object name is %s\n", oid.name); | |
589d30e0 AE |
5730 | |
5731 | /* Response will be an encoded string, which includes a length */ | |
5732 | ||
5733 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5734 | response = kzalloc(size, GFP_NOIO); | |
5735 | if (!response) { | |
5736 | ret = -ENOMEM; | |
5737 | goto out; | |
5738 | } | |
5739 | ||
c0fba368 AE |
5740 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5741 | ||
ecd4a68a ID |
5742 | ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc, |
5743 | "get_id", NULL, 0, | |
5744 | response, RBD_IMAGE_ID_LEN_MAX); | |
36be9a76 | 5745 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5746 | if (ret == -ENOENT) { |
5747 | image_id = kstrdup("", GFP_KERNEL); | |
5748 | ret = image_id ? 0 : -ENOMEM; | |
5749 | if (!ret) | |
5750 | rbd_dev->image_format = 1; | |
7dd440c9 | 5751 | } else if (ret >= 0) { |
c0fba368 AE |
5752 | void *p = response; |
5753 | ||
5754 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5755 | NULL, GFP_NOIO); |
461f758a | 5756 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5757 | if (!ret) |
5758 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5759 | } |
5760 | ||
5761 | if (!ret) { | |
5762 | rbd_dev->spec->image_id = image_id; | |
5763 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5764 | } |
5765 | out: | |
5766 | kfree(response); | |
ecd4a68a | 5767 | ceph_oid_destroy(&oid); |
589d30e0 AE |
5768 | return ret; |
5769 | } | |
5770 | ||
3abef3b3 AE |
5771 | /* |
5772 | * Undo whatever state changes are made by v1 or v2 header info | |
5773 | * call. | |
5774 | */ | |
6fd48b3b AE |
5775 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5776 | { | |
5777 | struct rbd_image_header *header; | |
5778 | ||
e69b8d41 | 5779 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5780 | |
5781 | /* Free dynamic fields from the header, then zero it out */ | |
5782 | ||
5783 | header = &rbd_dev->header; | |
812164f8 | 5784 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5785 | kfree(header->snap_sizes); |
5786 | kfree(header->snap_names); | |
5787 | kfree(header->object_prefix); | |
5788 | memset(header, 0, sizeof (*header)); | |
5789 | } | |
5790 | ||
2df3fac7 | 5791 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5792 | { |
5793 | int ret; | |
a30b71b9 | 5794 | |
1e130199 | 5795 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5796 | if (ret) |
b1b5402a AE |
5797 | goto out_err; |
5798 | ||
2df3fac7 AE |
5799 | /* |
5800 | * Get the and check features for the image. Currently the | |
5801 | * features are assumed to never change. | |
5802 | */ | |
b1b5402a | 5803 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5804 | if (ret) |
9d475de5 | 5805 | goto out_err; |
35d489f9 | 5806 | |
cc070d59 AE |
5807 | /* If the image supports fancy striping, get its parameters */ |
5808 | ||
5809 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5810 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5811 | if (ret < 0) | |
5812 | goto out_err; | |
5813 | } | |
a30b71b9 | 5814 | |
7e97332e ID |
5815 | if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) { |
5816 | ret = rbd_dev_v2_data_pool(rbd_dev); | |
5817 | if (ret) | |
5818 | goto out_err; | |
5819 | } | |
5820 | ||
263423f8 | 5821 | rbd_init_layout(rbd_dev); |
35152979 | 5822 | return 0; |
263423f8 | 5823 | |
9d475de5 | 5824 | out_err: |
642a2537 | 5825 | rbd_dev->header.features = 0; |
1e130199 AE |
5826 | kfree(rbd_dev->header.object_prefix); |
5827 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 | 5828 | return ret; |
a30b71b9 AE |
5829 | } |
5830 | ||
6d69bb53 ID |
5831 | /* |
5832 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
5833 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
5834 | * length of the already discovered part of the parent chain. | |
5835 | */ | |
5836 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 5837 | { |
2f82ee54 | 5838 | struct rbd_device *parent = NULL; |
124afba2 AE |
5839 | int ret; |
5840 | ||
5841 | if (!rbd_dev->parent_spec) | |
5842 | return 0; | |
124afba2 | 5843 | |
6d69bb53 ID |
5844 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
5845 | pr_info("parent chain is too long (%d)\n", depth); | |
5846 | ret = -EINVAL; | |
5847 | goto out_err; | |
5848 | } | |
5849 | ||
1643dfa4 | 5850 | parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec); |
1f2c6651 ID |
5851 | if (!parent) { |
5852 | ret = -ENOMEM; | |
124afba2 | 5853 | goto out_err; |
1f2c6651 ID |
5854 | } |
5855 | ||
5856 | /* | |
5857 | * Images related by parent/child relationships always share | |
5858 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
5859 | */ | |
5860 | __rbd_get_client(rbd_dev->rbd_client); | |
5861 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 5862 | |
6d69bb53 | 5863 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
5864 | if (ret < 0) |
5865 | goto out_err; | |
1f2c6651 | 5866 | |
124afba2 | 5867 | rbd_dev->parent = parent; |
a2acd00e | 5868 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 5869 | return 0; |
1f2c6651 | 5870 | |
124afba2 | 5871 | out_err: |
1f2c6651 | 5872 | rbd_dev_unparent(rbd_dev); |
1761b229 | 5873 | rbd_dev_destroy(parent); |
124afba2 AE |
5874 | return ret; |
5875 | } | |
5876 | ||
811c6688 ID |
5877 | /* |
5878 | * rbd_dev->header_rwsem must be locked for write and will be unlocked | |
5879 | * upon return. | |
5880 | */ | |
200a6a8b | 5881 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5882 | { |
83a06263 | 5883 | int ret; |
d1cf5788 | 5884 | |
9b60e70b | 5885 | /* Record our major and minor device numbers. */ |
83a06263 | 5886 | |
9b60e70b ID |
5887 | if (!single_major) { |
5888 | ret = register_blkdev(0, rbd_dev->name); | |
5889 | if (ret < 0) | |
1643dfa4 | 5890 | goto err_out_unlock; |
9b60e70b ID |
5891 | |
5892 | rbd_dev->major = ret; | |
5893 | rbd_dev->minor = 0; | |
5894 | } else { | |
5895 | rbd_dev->major = rbd_major; | |
5896 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5897 | } | |
83a06263 AE |
5898 | |
5899 | /* Set up the blkdev mapping. */ | |
5900 | ||
5901 | ret = rbd_init_disk(rbd_dev); | |
5902 | if (ret) | |
5903 | goto err_out_blkdev; | |
5904 | ||
f35a4dee | 5905 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5906 | if (ret) |
5907 | goto err_out_disk; | |
bc1ecc65 | 5908 | |
f35a4dee | 5909 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5910 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 5911 | |
dd5ac32d ID |
5912 | dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id); |
5913 | ret = device_add(&rbd_dev->dev); | |
f35a4dee | 5914 | if (ret) |
f5ee37bd | 5915 | goto err_out_mapping; |
83a06263 | 5916 | |
83a06263 AE |
5917 | /* Everything's ready. Announce the disk to the world. */ |
5918 | ||
129b79d4 | 5919 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
811c6688 | 5920 | up_write(&rbd_dev->header_rwsem); |
83a06263 | 5921 | |
1643dfa4 ID |
5922 | spin_lock(&rbd_dev_list_lock); |
5923 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
5924 | spin_unlock(&rbd_dev_list_lock); | |
5925 | ||
811c6688 | 5926 | add_disk(rbd_dev->disk); |
ca7909e8 ID |
5927 | pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name, |
5928 | (unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT, | |
5929 | rbd_dev->header.features); | |
83a06263 AE |
5930 | |
5931 | return ret; | |
2f82ee54 | 5932 | |
f35a4dee AE |
5933 | err_out_mapping: |
5934 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5935 | err_out_disk: |
5936 | rbd_free_disk(rbd_dev); | |
5937 | err_out_blkdev: | |
9b60e70b ID |
5938 | if (!single_major) |
5939 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
811c6688 ID |
5940 | err_out_unlock: |
5941 | up_write(&rbd_dev->header_rwsem); | |
83a06263 AE |
5942 | return ret; |
5943 | } | |
5944 | ||
332bb12d AE |
5945 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5946 | { | |
5947 | struct rbd_spec *spec = rbd_dev->spec; | |
c41d13a3 | 5948 | int ret; |
332bb12d AE |
5949 | |
5950 | /* Record the header object name for this rbd image. */ | |
5951 | ||
5952 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
332bb12d | 5953 | if (rbd_dev->image_format == 1) |
c41d13a3 ID |
5954 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
5955 | spec->image_name, RBD_SUFFIX); | |
332bb12d | 5956 | else |
c41d13a3 ID |
5957 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
5958 | RBD_HEADER_PREFIX, spec->image_id); | |
332bb12d | 5959 | |
c41d13a3 | 5960 | return ret; |
332bb12d AE |
5961 | } |
5962 | ||
200a6a8b AE |
5963 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5964 | { | |
6fd48b3b | 5965 | rbd_dev_unprobe(rbd_dev); |
6fd48b3b AE |
5966 | rbd_dev->image_format = 0; |
5967 | kfree(rbd_dev->spec->image_id); | |
5968 | rbd_dev->spec->image_id = NULL; | |
5969 | ||
200a6a8b AE |
5970 | rbd_dev_destroy(rbd_dev); |
5971 | } | |
5972 | ||
a30b71b9 AE |
5973 | /* |
5974 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5975 | * device. If this image is the one being mapped (i.e., not a |
5976 | * parent), initiate a watch on its header object before using that | |
5977 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5978 | */ |
6d69bb53 | 5979 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth) |
a30b71b9 AE |
5980 | { |
5981 | int ret; | |
5982 | ||
5983 | /* | |
3abef3b3 AE |
5984 | * Get the id from the image id object. Unless there's an |
5985 | * error, rbd_dev->spec->image_id will be filled in with | |
5986 | * a dynamically-allocated string, and rbd_dev->image_format | |
5987 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5988 | */ |
5989 | ret = rbd_dev_image_id(rbd_dev); | |
5990 | if (ret) | |
c0fba368 | 5991 | return ret; |
c0fba368 | 5992 | |
332bb12d AE |
5993 | ret = rbd_dev_header_name(rbd_dev); |
5994 | if (ret) | |
5995 | goto err_out_format; | |
5996 | ||
6d69bb53 | 5997 | if (!depth) { |
99d16943 | 5998 | ret = rbd_register_watch(rbd_dev); |
1fe48023 ID |
5999 | if (ret) { |
6000 | if (ret == -ENOENT) | |
6001 | pr_info("image %s/%s does not exist\n", | |
6002 | rbd_dev->spec->pool_name, | |
6003 | rbd_dev->spec->image_name); | |
c41d13a3 | 6004 | goto err_out_format; |
1fe48023 | 6005 | } |
1f3ef788 | 6006 | } |
b644de2b | 6007 | |
a720ae09 | 6008 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 6009 | if (ret) |
b644de2b | 6010 | goto err_out_watch; |
83a06263 | 6011 | |
04077599 ID |
6012 | /* |
6013 | * If this image is the one being mapped, we have pool name and | |
6014 | * id, image name and id, and snap name - need to fill snap id. | |
6015 | * Otherwise this is a parent image, identified by pool, image | |
6016 | * and snap ids - need to fill in names for those ids. | |
6017 | */ | |
6d69bb53 | 6018 | if (!depth) |
04077599 ID |
6019 | ret = rbd_spec_fill_snap_id(rbd_dev); |
6020 | else | |
6021 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
6022 | if (ret) { |
6023 | if (ret == -ENOENT) | |
6024 | pr_info("snap %s/%s@%s does not exist\n", | |
6025 | rbd_dev->spec->pool_name, | |
6026 | rbd_dev->spec->image_name, | |
6027 | rbd_dev->spec->snap_name); | |
33dca39f | 6028 | goto err_out_probe; |
1fe48023 | 6029 | } |
9bb81c9b | 6030 | |
e8f59b59 ID |
6031 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
6032 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
6033 | if (ret) | |
6034 | goto err_out_probe; | |
6035 | ||
6036 | /* | |
6037 | * Need to warn users if this image is the one being | |
6038 | * mapped and has a parent. | |
6039 | */ | |
6d69bb53 | 6040 | if (!depth && rbd_dev->parent_spec) |
e8f59b59 ID |
6041 | rbd_warn(rbd_dev, |
6042 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
6043 | } | |
6044 | ||
6d69bb53 | 6045 | ret = rbd_dev_probe_parent(rbd_dev, depth); |
30d60ba2 AE |
6046 | if (ret) |
6047 | goto err_out_probe; | |
6048 | ||
6049 | dout("discovered format %u image, header name is %s\n", | |
c41d13a3 | 6050 | rbd_dev->image_format, rbd_dev->header_oid.name); |
30d60ba2 | 6051 | return 0; |
e8f59b59 | 6052 | |
6fd48b3b AE |
6053 | err_out_probe: |
6054 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 6055 | err_out_watch: |
6d69bb53 | 6056 | if (!depth) |
99d16943 | 6057 | rbd_unregister_watch(rbd_dev); |
332bb12d AE |
6058 | err_out_format: |
6059 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
6060 | kfree(rbd_dev->spec->image_id); |
6061 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
6062 | return ret; |
6063 | } | |
6064 | ||
9b60e70b ID |
6065 | static ssize_t do_rbd_add(struct bus_type *bus, |
6066 | const char *buf, | |
6067 | size_t count) | |
602adf40 | 6068 | { |
cb8627c7 | 6069 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 6070 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 6071 | struct rbd_options *rbd_opts = NULL; |
859c31df | 6072 | struct rbd_spec *spec = NULL; |
9d3997fd | 6073 | struct rbd_client *rbdc; |
51344a38 | 6074 | bool read_only; |
b51c83c2 | 6075 | int rc; |
602adf40 YS |
6076 | |
6077 | if (!try_module_get(THIS_MODULE)) | |
6078 | return -ENODEV; | |
6079 | ||
602adf40 | 6080 | /* parse add command */ |
859c31df | 6081 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 6082 | if (rc < 0) |
dd5ac32d | 6083 | goto out; |
78cea76e | 6084 | |
9d3997fd AE |
6085 | rbdc = rbd_get_client(ceph_opts); |
6086 | if (IS_ERR(rbdc)) { | |
6087 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 6088 | goto err_out_args; |
9d3997fd | 6089 | } |
602adf40 | 6090 | |
602adf40 | 6091 | /* pick the pool */ |
30ba1f02 | 6092 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
6093 | if (rc < 0) { |
6094 | if (rc == -ENOENT) | |
6095 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 6096 | goto err_out_client; |
1fe48023 | 6097 | } |
c0cd10db | 6098 | spec->pool_id = (u64)rc; |
859c31df | 6099 | |
d147543d | 6100 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
b51c83c2 ID |
6101 | if (!rbd_dev) { |
6102 | rc = -ENOMEM; | |
bd4ba655 | 6103 | goto err_out_client; |
b51c83c2 | 6104 | } |
c53d5893 AE |
6105 | rbdc = NULL; /* rbd_dev now owns this */ |
6106 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 6107 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 6108 | |
0d6d1e9c MC |
6109 | rbd_dev->config_info = kstrdup(buf, GFP_KERNEL); |
6110 | if (!rbd_dev->config_info) { | |
6111 | rc = -ENOMEM; | |
6112 | goto err_out_rbd_dev; | |
6113 | } | |
6114 | ||
811c6688 | 6115 | down_write(&rbd_dev->header_rwsem); |
6d69bb53 | 6116 | rc = rbd_dev_image_probe(rbd_dev, 0); |
0d6d1e9c MC |
6117 | if (rc < 0) { |
6118 | up_write(&rbd_dev->header_rwsem); | |
c53d5893 | 6119 | goto err_out_rbd_dev; |
0d6d1e9c | 6120 | } |
05fd6f6f | 6121 | |
7ce4eef7 AE |
6122 | /* If we are mapping a snapshot it must be marked read-only */ |
6123 | ||
d147543d | 6124 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
6125 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
6126 | read_only = true; | |
6127 | rbd_dev->mapping.read_only = read_only; | |
6128 | ||
b536f69a | 6129 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 6130 | if (rc) { |
e37180c0 | 6131 | /* |
99d16943 | 6132 | * rbd_unregister_watch() can't be moved into |
e37180c0 ID |
6133 | * rbd_dev_image_release() without refactoring, see |
6134 | * commit 1f3ef78861ac. | |
6135 | */ | |
99d16943 | 6136 | rbd_unregister_watch(rbd_dev); |
3abef3b3 | 6137 | rbd_dev_image_release(rbd_dev); |
dd5ac32d | 6138 | goto out; |
3abef3b3 AE |
6139 | } |
6140 | ||
dd5ac32d ID |
6141 | rc = count; |
6142 | out: | |
6143 | module_put(THIS_MODULE); | |
6144 | return rc; | |
b536f69a | 6145 | |
c53d5893 AE |
6146 | err_out_rbd_dev: |
6147 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 6148 | err_out_client: |
9d3997fd | 6149 | rbd_put_client(rbdc); |
0ddebc0c | 6150 | err_out_args: |
859c31df | 6151 | rbd_spec_put(spec); |
d147543d | 6152 | kfree(rbd_opts); |
dd5ac32d | 6153 | goto out; |
602adf40 YS |
6154 | } |
6155 | ||
9b60e70b ID |
6156 | static ssize_t rbd_add(struct bus_type *bus, |
6157 | const char *buf, | |
6158 | size_t count) | |
6159 | { | |
6160 | if (single_major) | |
6161 | return -EINVAL; | |
6162 | ||
6163 | return do_rbd_add(bus, buf, count); | |
6164 | } | |
6165 | ||
6166 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
6167 | const char *buf, | |
6168 | size_t count) | |
6169 | { | |
6170 | return do_rbd_add(bus, buf, count); | |
6171 | } | |
6172 | ||
dd5ac32d | 6173 | static void rbd_dev_device_release(struct rbd_device *rbd_dev) |
602adf40 | 6174 | { |
602adf40 | 6175 | rbd_free_disk(rbd_dev); |
1643dfa4 ID |
6176 | |
6177 | spin_lock(&rbd_dev_list_lock); | |
6178 | list_del_init(&rbd_dev->node); | |
6179 | spin_unlock(&rbd_dev_list_lock); | |
6180 | ||
200a6a8b | 6181 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
dd5ac32d | 6182 | device_del(&rbd_dev->dev); |
6d80b130 | 6183 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
6184 | if (!single_major) |
6185 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
602adf40 YS |
6186 | } |
6187 | ||
05a46afd AE |
6188 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
6189 | { | |
ad945fc1 | 6190 | while (rbd_dev->parent) { |
05a46afd AE |
6191 | struct rbd_device *first = rbd_dev; |
6192 | struct rbd_device *second = first->parent; | |
6193 | struct rbd_device *third; | |
6194 | ||
6195 | /* | |
6196 | * Follow to the parent with no grandparent and | |
6197 | * remove it. | |
6198 | */ | |
6199 | while (second && (third = second->parent)) { | |
6200 | first = second; | |
6201 | second = third; | |
6202 | } | |
ad945fc1 | 6203 | rbd_assert(second); |
8ad42cd0 | 6204 | rbd_dev_image_release(second); |
ad945fc1 AE |
6205 | first->parent = NULL; |
6206 | first->parent_overlap = 0; | |
6207 | ||
6208 | rbd_assert(first->parent_spec); | |
05a46afd AE |
6209 | rbd_spec_put(first->parent_spec); |
6210 | first->parent_spec = NULL; | |
05a46afd AE |
6211 | } |
6212 | } | |
6213 | ||
9b60e70b ID |
6214 | static ssize_t do_rbd_remove(struct bus_type *bus, |
6215 | const char *buf, | |
6216 | size_t count) | |
602adf40 YS |
6217 | { |
6218 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
6219 | struct list_head *tmp; |
6220 | int dev_id; | |
0276dca6 | 6221 | char opt_buf[6]; |
82a442d2 | 6222 | bool already = false; |
0276dca6 | 6223 | bool force = false; |
0d8189e1 | 6224 | int ret; |
602adf40 | 6225 | |
0276dca6 MC |
6226 | dev_id = -1; |
6227 | opt_buf[0] = '\0'; | |
6228 | sscanf(buf, "%d %5s", &dev_id, opt_buf); | |
6229 | if (dev_id < 0) { | |
6230 | pr_err("dev_id out of range\n"); | |
602adf40 | 6231 | return -EINVAL; |
0276dca6 MC |
6232 | } |
6233 | if (opt_buf[0] != '\0') { | |
6234 | if (!strcmp(opt_buf, "force")) { | |
6235 | force = true; | |
6236 | } else { | |
6237 | pr_err("bad remove option at '%s'\n", opt_buf); | |
6238 | return -EINVAL; | |
6239 | } | |
6240 | } | |
602adf40 | 6241 | |
751cc0e3 AE |
6242 | ret = -ENOENT; |
6243 | spin_lock(&rbd_dev_list_lock); | |
6244 | list_for_each(tmp, &rbd_dev_list) { | |
6245 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
6246 | if (rbd_dev->dev_id == dev_id) { | |
6247 | ret = 0; | |
6248 | break; | |
6249 | } | |
42382b70 | 6250 | } |
751cc0e3 AE |
6251 | if (!ret) { |
6252 | spin_lock_irq(&rbd_dev->lock); | |
0276dca6 | 6253 | if (rbd_dev->open_count && !force) |
751cc0e3 AE |
6254 | ret = -EBUSY; |
6255 | else | |
82a442d2 AE |
6256 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
6257 | &rbd_dev->flags); | |
751cc0e3 AE |
6258 | spin_unlock_irq(&rbd_dev->lock); |
6259 | } | |
6260 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 6261 | if (ret < 0 || already) |
1ba0f1e7 | 6262 | return ret; |
751cc0e3 | 6263 | |
0276dca6 MC |
6264 | if (force) { |
6265 | /* | |
6266 | * Prevent new IO from being queued and wait for existing | |
6267 | * IO to complete/fail. | |
6268 | */ | |
6269 | blk_mq_freeze_queue(rbd_dev->disk->queue); | |
6270 | blk_set_queue_dying(rbd_dev->disk->queue); | |
6271 | } | |
6272 | ||
ed95b21a ID |
6273 | down_write(&rbd_dev->lock_rwsem); |
6274 | if (__rbd_is_lock_owner(rbd_dev)) | |
6275 | rbd_unlock(rbd_dev); | |
6276 | up_write(&rbd_dev->lock_rwsem); | |
99d16943 | 6277 | rbd_unregister_watch(rbd_dev); |
fca27065 | 6278 | |
9875201e JD |
6279 | /* |
6280 | * Don't free anything from rbd_dev->disk until after all | |
6281 | * notifies are completely processed. Otherwise | |
6282 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
6283 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
6284 | */ | |
dd5ac32d | 6285 | rbd_dev_device_release(rbd_dev); |
8ad42cd0 | 6286 | rbd_dev_image_release(rbd_dev); |
aafb230e | 6287 | |
1ba0f1e7 | 6288 | return count; |
602adf40 YS |
6289 | } |
6290 | ||
9b60e70b ID |
6291 | static ssize_t rbd_remove(struct bus_type *bus, |
6292 | const char *buf, | |
6293 | size_t count) | |
6294 | { | |
6295 | if (single_major) | |
6296 | return -EINVAL; | |
6297 | ||
6298 | return do_rbd_remove(bus, buf, count); | |
6299 | } | |
6300 | ||
6301 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
6302 | const char *buf, | |
6303 | size_t count) | |
6304 | { | |
6305 | return do_rbd_remove(bus, buf, count); | |
6306 | } | |
6307 | ||
602adf40 YS |
6308 | /* |
6309 | * create control files in sysfs | |
dfc5606d | 6310 | * /sys/bus/rbd/... |
602adf40 YS |
6311 | */ |
6312 | static int rbd_sysfs_init(void) | |
6313 | { | |
dfc5606d | 6314 | int ret; |
602adf40 | 6315 | |
fed4c143 | 6316 | ret = device_register(&rbd_root_dev); |
21079786 | 6317 | if (ret < 0) |
dfc5606d | 6318 | return ret; |
602adf40 | 6319 | |
fed4c143 AE |
6320 | ret = bus_register(&rbd_bus_type); |
6321 | if (ret < 0) | |
6322 | device_unregister(&rbd_root_dev); | |
602adf40 | 6323 | |
602adf40 YS |
6324 | return ret; |
6325 | } | |
6326 | ||
6327 | static void rbd_sysfs_cleanup(void) | |
6328 | { | |
dfc5606d | 6329 | bus_unregister(&rbd_bus_type); |
fed4c143 | 6330 | device_unregister(&rbd_root_dev); |
602adf40 YS |
6331 | } |
6332 | ||
1c2a9dfe AE |
6333 | static int rbd_slab_init(void) |
6334 | { | |
6335 | rbd_assert(!rbd_img_request_cache); | |
03d94406 | 6336 | rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0); |
868311b1 AE |
6337 | if (!rbd_img_request_cache) |
6338 | return -ENOMEM; | |
6339 | ||
6340 | rbd_assert(!rbd_obj_request_cache); | |
03d94406 | 6341 | rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0); |
78c2a44a AE |
6342 | if (!rbd_obj_request_cache) |
6343 | goto out_err; | |
6344 | ||
6c696d85 | 6345 | return 0; |
1c2a9dfe | 6346 | |
6c696d85 | 6347 | out_err: |
868311b1 AE |
6348 | kmem_cache_destroy(rbd_img_request_cache); |
6349 | rbd_img_request_cache = NULL; | |
1c2a9dfe AE |
6350 | return -ENOMEM; |
6351 | } | |
6352 | ||
6353 | static void rbd_slab_exit(void) | |
6354 | { | |
868311b1 AE |
6355 | rbd_assert(rbd_obj_request_cache); |
6356 | kmem_cache_destroy(rbd_obj_request_cache); | |
6357 | rbd_obj_request_cache = NULL; | |
6358 | ||
1c2a9dfe AE |
6359 | rbd_assert(rbd_img_request_cache); |
6360 | kmem_cache_destroy(rbd_img_request_cache); | |
6361 | rbd_img_request_cache = NULL; | |
6362 | } | |
6363 | ||
cc344fa1 | 6364 | static int __init rbd_init(void) |
602adf40 YS |
6365 | { |
6366 | int rc; | |
6367 | ||
1e32d34c AE |
6368 | if (!libceph_compatible(NULL)) { |
6369 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
6370 | return -EINVAL; |
6371 | } | |
e1b4d96d | 6372 | |
1c2a9dfe | 6373 | rc = rbd_slab_init(); |
602adf40 YS |
6374 | if (rc) |
6375 | return rc; | |
e1b4d96d | 6376 | |
f5ee37bd ID |
6377 | /* |
6378 | * The number of active work items is limited by the number of | |
f77303bd | 6379 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
6380 | */ |
6381 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
6382 | if (!rbd_wq) { | |
6383 | rc = -ENOMEM; | |
6384 | goto err_out_slab; | |
6385 | } | |
6386 | ||
9b60e70b ID |
6387 | if (single_major) { |
6388 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
6389 | if (rbd_major < 0) { | |
6390 | rc = rbd_major; | |
f5ee37bd | 6391 | goto err_out_wq; |
9b60e70b ID |
6392 | } |
6393 | } | |
6394 | ||
1c2a9dfe AE |
6395 | rc = rbd_sysfs_init(); |
6396 | if (rc) | |
9b60e70b ID |
6397 | goto err_out_blkdev; |
6398 | ||
6399 | if (single_major) | |
6400 | pr_info("loaded (major %d)\n", rbd_major); | |
6401 | else | |
6402 | pr_info("loaded\n"); | |
1c2a9dfe | 6403 | |
e1b4d96d ID |
6404 | return 0; |
6405 | ||
9b60e70b ID |
6406 | err_out_blkdev: |
6407 | if (single_major) | |
6408 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
6409 | err_out_wq: |
6410 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
6411 | err_out_slab: |
6412 | rbd_slab_exit(); | |
1c2a9dfe | 6413 | return rc; |
602adf40 YS |
6414 | } |
6415 | ||
cc344fa1 | 6416 | static void __exit rbd_exit(void) |
602adf40 | 6417 | { |
ffe312cf | 6418 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 6419 | rbd_sysfs_cleanup(); |
9b60e70b ID |
6420 | if (single_major) |
6421 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 6422 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 6423 | rbd_slab_exit(); |
602adf40 YS |
6424 | } |
6425 | ||
6426 | module_init(rbd_init); | |
6427 | module_exit(rbd_exit); | |
6428 | ||
d552c619 | 6429 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
6430 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
6431 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
6432 | /* following authorship retained from original osdblk.c */ |
6433 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
6434 | ||
90da258b | 6435 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 6436 | MODULE_LICENSE("GPL"); |