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> |
43df3d35 | 35 | #include <linux/ceph/striper.h> |
602adf40 | 36 | #include <linux/ceph/decode.h> |
59c2be1e | 37 | #include <linux/parser.h> |
30d1cff8 | 38 | #include <linux/bsearch.h> |
602adf40 YS |
39 | |
40 | #include <linux/kernel.h> | |
41 | #include <linux/device.h> | |
42 | #include <linux/module.h> | |
7ad18afa | 43 | #include <linux/blk-mq.h> |
602adf40 YS |
44 | #include <linux/fs.h> |
45 | #include <linux/blkdev.h> | |
1c2a9dfe | 46 | #include <linux/slab.h> |
f8a22fc2 | 47 | #include <linux/idr.h> |
bc1ecc65 | 48 | #include <linux/workqueue.h> |
602adf40 YS |
49 | |
50 | #include "rbd_types.h" | |
51 | ||
aafb230e AE |
52 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
53 | ||
a2acd00e AE |
54 | /* |
55 | * Increment the given counter and return its updated value. | |
56 | * If the counter is already 0 it will not be incremented. | |
57 | * If the counter is already at its maximum value returns | |
58 | * -EINVAL without updating it. | |
59 | */ | |
60 | static int atomic_inc_return_safe(atomic_t *v) | |
61 | { | |
62 | unsigned int counter; | |
63 | ||
bfc18e38 | 64 | counter = (unsigned int)atomic_fetch_add_unless(v, 1, 0); |
a2acd00e AE |
65 | if (counter <= (unsigned int)INT_MAX) |
66 | return (int)counter; | |
67 | ||
68 | atomic_dec(v); | |
69 | ||
70 | return -EINVAL; | |
71 | } | |
72 | ||
73 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
74 | static int atomic_dec_return_safe(atomic_t *v) | |
75 | { | |
76 | int counter; | |
77 | ||
78 | counter = atomic_dec_return(v); | |
79 | if (counter >= 0) | |
80 | return counter; | |
81 | ||
82 | atomic_inc(v); | |
83 | ||
84 | return -EINVAL; | |
85 | } | |
86 | ||
f0f8cef5 | 87 | #define RBD_DRV_NAME "rbd" |
602adf40 | 88 | |
7e513d43 ID |
89 | #define RBD_MINORS_PER_MAJOR 256 |
90 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 91 | |
6d69bb53 ID |
92 | #define RBD_MAX_PARENT_CHAIN_LEN 16 |
93 | ||
d4b125e9 AE |
94 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
95 | #define RBD_MAX_SNAP_NAME_LEN \ | |
96 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
97 | ||
35d489f9 | 98 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
99 | |
100 | #define RBD_SNAP_HEAD_NAME "-" | |
101 | ||
9682fc6d AE |
102 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
103 | ||
9e15b77d AE |
104 | /* This allows a single page to hold an image name sent by OSD */ |
105 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 106 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 107 | |
1e130199 | 108 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 109 | |
ed95b21a | 110 | #define RBD_NOTIFY_TIMEOUT 5 /* seconds */ |
99d16943 ID |
111 | #define RBD_RETRY_DELAY msecs_to_jiffies(1000) |
112 | ||
d889140c AE |
113 | /* Feature bits */ |
114 | ||
8767b293 ID |
115 | #define RBD_FEATURE_LAYERING (1ULL<<0) |
116 | #define RBD_FEATURE_STRIPINGV2 (1ULL<<1) | |
117 | #define RBD_FEATURE_EXCLUSIVE_LOCK (1ULL<<2) | |
22e8bd51 ID |
118 | #define RBD_FEATURE_OBJECT_MAP (1ULL<<3) |
119 | #define RBD_FEATURE_FAST_DIFF (1ULL<<4) | |
b9f6d447 | 120 | #define RBD_FEATURE_DEEP_FLATTEN (1ULL<<5) |
8767b293 | 121 | #define RBD_FEATURE_DATA_POOL (1ULL<<7) |
e573427a | 122 | #define RBD_FEATURE_OPERATIONS (1ULL<<8) |
8767b293 | 123 | |
ed95b21a ID |
124 | #define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \ |
125 | RBD_FEATURE_STRIPINGV2 | \ | |
7e97332e | 126 | RBD_FEATURE_EXCLUSIVE_LOCK | \ |
22e8bd51 ID |
127 | RBD_FEATURE_OBJECT_MAP | \ |
128 | RBD_FEATURE_FAST_DIFF | \ | |
b9f6d447 | 129 | RBD_FEATURE_DEEP_FLATTEN | \ |
e573427a ID |
130 | RBD_FEATURE_DATA_POOL | \ |
131 | RBD_FEATURE_OPERATIONS) | |
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; |
b26c047b | 190 | const char *pool_ns; /* NULL if default, never "" */ |
0d7dbfce | 191 | |
ecb4dc22 AE |
192 | const char *image_id; |
193 | const char *image_name; | |
0d7dbfce AE |
194 | |
195 | u64 snap_id; | |
ecb4dc22 | 196 | const char *snap_name; |
0d7dbfce AE |
197 | |
198 | struct kref kref; | |
199 | }; | |
200 | ||
602adf40 | 201 | /* |
f0f8cef5 | 202 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
203 | */ |
204 | struct rbd_client { | |
205 | struct ceph_client *client; | |
206 | struct kref kref; | |
207 | struct list_head node; | |
208 | }; | |
209 | ||
0192ce2e ID |
210 | struct pending_result { |
211 | int result; /* first nonzero result */ | |
212 | int num_pending; | |
213 | }; | |
214 | ||
bf0d5f50 | 215 | struct rbd_img_request; |
bf0d5f50 | 216 | |
9969ebc5 | 217 | enum obj_request_type { |
a1fbb5e7 | 218 | OBJ_REQUEST_NODATA = 1, |
5359a17d | 219 | OBJ_REQUEST_BIO, /* pointer into provided bio (list) */ |
7e07efb1 | 220 | OBJ_REQUEST_BVECS, /* pointer into provided bio_vec array */ |
afb97888 | 221 | OBJ_REQUEST_OWN_BVECS, /* private bio_vec array, doesn't own pages */ |
9969ebc5 | 222 | }; |
bf0d5f50 | 223 | |
6d2940c8 | 224 | enum obj_operation_type { |
a1fbb5e7 | 225 | OBJ_OP_READ = 1, |
6d2940c8 | 226 | OBJ_OP_WRITE, |
90e98c52 | 227 | OBJ_OP_DISCARD, |
6484cbe9 | 228 | OBJ_OP_ZEROOUT, |
6d2940c8 GZ |
229 | }; |
230 | ||
0ad5d953 ID |
231 | #define RBD_OBJ_FLAG_DELETION (1U << 0) |
232 | #define RBD_OBJ_FLAG_COPYUP_ENABLED (1U << 1) | |
793333a3 | 233 | #define RBD_OBJ_FLAG_COPYUP_ZEROS (1U << 2) |
22e8bd51 ID |
234 | #define RBD_OBJ_FLAG_MAY_EXIST (1U << 3) |
235 | #define RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT (1U << 4) | |
0ad5d953 | 236 | |
a9b67e69 | 237 | enum rbd_obj_read_state { |
85b5e6d1 ID |
238 | RBD_OBJ_READ_START = 1, |
239 | RBD_OBJ_READ_OBJECT, | |
a9b67e69 ID |
240 | RBD_OBJ_READ_PARENT, |
241 | }; | |
242 | ||
3da691bf ID |
243 | /* |
244 | * Writes go through the following state machine to deal with | |
245 | * layering: | |
246 | * | |
89a59c1c ID |
247 | * . . . . . RBD_OBJ_WRITE_GUARD. . . . . . . . . . . . . . |
248 | * . | . | |
249 | * . v . | |
250 | * . RBD_OBJ_WRITE_READ_FROM_PARENT. . . . | |
251 | * . | . . | |
252 | * . v v (deep-copyup . | |
253 | * (image . RBD_OBJ_WRITE_COPYUP_EMPTY_SNAPC . not needed) . | |
254 | * flattened) v | . . | |
255 | * . v . . | |
256 | * . . . .RBD_OBJ_WRITE_COPYUP_OPS. . . . . (copyup . | |
257 | * | not needed) v | |
258 | * v . | |
259 | * done . . . . . . . . . . . . . . . . . . | |
260 | * ^ | |
261 | * | | |
262 | * RBD_OBJ_WRITE_FLAT | |
3da691bf ID |
263 | * |
264 | * Writes start in RBD_OBJ_WRITE_GUARD or _FLAT, depending on whether | |
89a59c1c ID |
265 | * assert_exists guard is needed or not (in some cases it's not needed |
266 | * even if there is a parent). | |
3da691bf ID |
267 | */ |
268 | enum rbd_obj_write_state { | |
85b5e6d1 | 269 | RBD_OBJ_WRITE_START = 1, |
22e8bd51 | 270 | RBD_OBJ_WRITE_PRE_OBJECT_MAP, |
85b5e6d1 | 271 | RBD_OBJ_WRITE_OBJECT, |
793333a3 ID |
272 | __RBD_OBJ_WRITE_COPYUP, |
273 | RBD_OBJ_WRITE_COPYUP, | |
22e8bd51 | 274 | RBD_OBJ_WRITE_POST_OBJECT_MAP, |
793333a3 ID |
275 | }; |
276 | ||
277 | enum rbd_obj_copyup_state { | |
278 | RBD_OBJ_COPYUP_START = 1, | |
279 | RBD_OBJ_COPYUP_READ_PARENT, | |
22e8bd51 ID |
280 | __RBD_OBJ_COPYUP_OBJECT_MAPS, |
281 | RBD_OBJ_COPYUP_OBJECT_MAPS, | |
793333a3 ID |
282 | __RBD_OBJ_COPYUP_WRITE_OBJECT, |
283 | RBD_OBJ_COPYUP_WRITE_OBJECT, | |
926f9b3f AE |
284 | }; |
285 | ||
bf0d5f50 | 286 | struct rbd_obj_request { |
43df3d35 | 287 | struct ceph_object_extent ex; |
0ad5d953 | 288 | unsigned int flags; /* RBD_OBJ_FLAG_* */ |
c5b5ef6c | 289 | union { |
a9b67e69 | 290 | enum rbd_obj_read_state read_state; /* for reads */ |
3da691bf | 291 | enum rbd_obj_write_state write_state; /* for writes */ |
c5b5ef6c | 292 | }; |
bf0d5f50 | 293 | |
51c3509e | 294 | struct rbd_img_request *img_request; |
86bd7998 ID |
295 | struct ceph_file_extent *img_extents; |
296 | u32 num_img_extents; | |
bf0d5f50 | 297 | |
788e2df3 | 298 | union { |
5359a17d | 299 | struct ceph_bio_iter bio_pos; |
788e2df3 | 300 | struct { |
7e07efb1 ID |
301 | struct ceph_bvec_iter bvec_pos; |
302 | u32 bvec_count; | |
afb97888 | 303 | u32 bvec_idx; |
788e2df3 AE |
304 | }; |
305 | }; | |
793333a3 ID |
306 | |
307 | enum rbd_obj_copyup_state copyup_state; | |
7e07efb1 ID |
308 | struct bio_vec *copyup_bvecs; |
309 | u32 copyup_bvec_count; | |
bf0d5f50 | 310 | |
bcbab1db | 311 | struct list_head osd_reqs; /* w/ r_private_item */ |
bf0d5f50 | 312 | |
85b5e6d1 | 313 | struct mutex state_mutex; |
793333a3 | 314 | struct pending_result pending; |
bf0d5f50 AE |
315 | struct kref kref; |
316 | }; | |
317 | ||
0c425248 | 318 | enum img_req_flags { |
9849e986 | 319 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ |
d0b2e944 | 320 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
321 | }; |
322 | ||
0192ce2e ID |
323 | enum rbd_img_state { |
324 | RBD_IMG_START = 1, | |
637cd060 | 325 | RBD_IMG_EXCLUSIVE_LOCK, |
0192ce2e ID |
326 | __RBD_IMG_OBJECT_REQUESTS, |
327 | RBD_IMG_OBJECT_REQUESTS, | |
328 | }; | |
329 | ||
bf0d5f50 | 330 | struct rbd_img_request { |
bf0d5f50 | 331 | struct rbd_device *rbd_dev; |
9bb0248d | 332 | enum obj_operation_type op_type; |
ecc633ca | 333 | enum obj_request_type data_type; |
0c425248 | 334 | unsigned long flags; |
0192ce2e | 335 | enum rbd_img_state state; |
bf0d5f50 | 336 | union { |
9849e986 | 337 | u64 snap_id; /* for reads */ |
bf0d5f50 | 338 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
339 | }; |
340 | union { | |
341 | struct request *rq; /* block request */ | |
342 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 343 | }; |
bf0d5f50 | 344 | |
e1fddc8f | 345 | struct list_head lock_item; |
43df3d35 | 346 | struct list_head object_extents; /* obj_req.ex structs */ |
bf0d5f50 | 347 | |
0192ce2e ID |
348 | struct mutex state_mutex; |
349 | struct pending_result pending; | |
350 | struct work_struct work; | |
351 | int work_result; | |
bf0d5f50 AE |
352 | struct kref kref; |
353 | }; | |
354 | ||
355 | #define for_each_obj_request(ireq, oreq) \ | |
43df3d35 | 356 | list_for_each_entry(oreq, &(ireq)->object_extents, ex.oe_item) |
bf0d5f50 | 357 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
43df3d35 | 358 | list_for_each_entry_safe(oreq, n, &(ireq)->object_extents, ex.oe_item) |
bf0d5f50 | 359 | |
99d16943 ID |
360 | enum rbd_watch_state { |
361 | RBD_WATCH_STATE_UNREGISTERED, | |
362 | RBD_WATCH_STATE_REGISTERED, | |
363 | RBD_WATCH_STATE_ERROR, | |
364 | }; | |
365 | ||
ed95b21a ID |
366 | enum rbd_lock_state { |
367 | RBD_LOCK_STATE_UNLOCKED, | |
368 | RBD_LOCK_STATE_LOCKED, | |
369 | RBD_LOCK_STATE_RELEASING, | |
370 | }; | |
371 | ||
372 | /* WatchNotify::ClientId */ | |
373 | struct rbd_client_id { | |
374 | u64 gid; | |
375 | u64 handle; | |
376 | }; | |
377 | ||
f84344f3 | 378 | struct rbd_mapping { |
99c1f08f | 379 | u64 size; |
34b13184 | 380 | u64 features; |
f84344f3 AE |
381 | }; |
382 | ||
602adf40 YS |
383 | /* |
384 | * a single device | |
385 | */ | |
386 | struct rbd_device { | |
de71a297 | 387 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
388 | |
389 | int major; /* blkdev assigned major */ | |
dd82fff1 | 390 | int minor; |
602adf40 | 391 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 392 | |
a30b71b9 | 393 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
394 | struct rbd_client *rbd_client; |
395 | ||
396 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
397 | ||
b82d167b | 398 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
399 | |
400 | struct rbd_image_header header; | |
b82d167b | 401 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 402 | struct rbd_spec *spec; |
d147543d | 403 | struct rbd_options *opts; |
0d6d1e9c | 404 | char *config_info; /* add{,_single_major} string */ |
602adf40 | 405 | |
c41d13a3 | 406 | struct ceph_object_id header_oid; |
922dab61 | 407 | struct ceph_object_locator header_oloc; |
971f839a | 408 | |
1643dfa4 | 409 | struct ceph_file_layout layout; /* used for all rbd requests */ |
0903e875 | 410 | |
99d16943 ID |
411 | struct mutex watch_mutex; |
412 | enum rbd_watch_state watch_state; | |
922dab61 | 413 | struct ceph_osd_linger_request *watch_handle; |
99d16943 ID |
414 | u64 watch_cookie; |
415 | struct delayed_work watch_dwork; | |
59c2be1e | 416 | |
ed95b21a ID |
417 | struct rw_semaphore lock_rwsem; |
418 | enum rbd_lock_state lock_state; | |
cbbfb0ff | 419 | char lock_cookie[32]; |
ed95b21a ID |
420 | struct rbd_client_id owner_cid; |
421 | struct work_struct acquired_lock_work; | |
422 | struct work_struct released_lock_work; | |
423 | struct delayed_work lock_dwork; | |
424 | struct work_struct unlock_work; | |
e1fddc8f | 425 | spinlock_t lock_lists_lock; |
637cd060 | 426 | struct list_head acquiring_list; |
e1fddc8f | 427 | struct list_head running_list; |
637cd060 ID |
428 | struct completion acquire_wait; |
429 | int acquire_err; | |
e1fddc8f | 430 | struct completion releasing_wait; |
ed95b21a | 431 | |
22e8bd51 ID |
432 | spinlock_t object_map_lock; |
433 | u8 *object_map; | |
434 | u64 object_map_size; /* in objects */ | |
435 | u64 object_map_flags; | |
ed95b21a | 436 | |
1643dfa4 | 437 | struct workqueue_struct *task_wq; |
59c2be1e | 438 | |
86b00e0d AE |
439 | struct rbd_spec *parent_spec; |
440 | u64 parent_overlap; | |
a2acd00e | 441 | atomic_t parent_ref; |
2f82ee54 | 442 | struct rbd_device *parent; |
86b00e0d | 443 | |
7ad18afa CH |
444 | /* Block layer tags. */ |
445 | struct blk_mq_tag_set tag_set; | |
446 | ||
c666601a JD |
447 | /* protects updating the header */ |
448 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
449 | |
450 | struct rbd_mapping mapping; | |
602adf40 YS |
451 | |
452 | struct list_head node; | |
dfc5606d | 453 | |
dfc5606d YS |
454 | /* sysfs related */ |
455 | struct device dev; | |
b82d167b | 456 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
457 | }; |
458 | ||
b82d167b | 459 | /* |
87c0fded ID |
460 | * Flag bits for rbd_dev->flags: |
461 | * - REMOVING (which is coupled with rbd_dev->open_count) is protected | |
462 | * by rbd_dev->lock | |
b82d167b | 463 | */ |
6d292906 AE |
464 | enum rbd_dev_flags { |
465 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 466 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
467 | }; |
468 | ||
cfbf6377 | 469 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 470 | |
602adf40 | 471 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
472 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
473 | ||
432b8587 AE |
474 | static LIST_HEAD(rbd_client_list); /* clients */ |
475 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 476 | |
78c2a44a AE |
477 | /* Slab caches for frequently-allocated structures */ |
478 | ||
1c2a9dfe | 479 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 480 | static struct kmem_cache *rbd_obj_request_cache; |
1c2a9dfe | 481 | |
9b60e70b | 482 | static int rbd_major; |
f8a22fc2 ID |
483 | static DEFINE_IDA(rbd_dev_id_ida); |
484 | ||
f5ee37bd ID |
485 | static struct workqueue_struct *rbd_wq; |
486 | ||
89a59c1c ID |
487 | static struct ceph_snap_context rbd_empty_snapc = { |
488 | .nref = REFCOUNT_INIT(1), | |
489 | }; | |
490 | ||
9b60e70b | 491 | /* |
3cfa3b16 | 492 | * single-major requires >= 0.75 version of userspace rbd utility. |
9b60e70b | 493 | */ |
3cfa3b16 | 494 | static bool single_major = true; |
5657a819 | 495 | module_param(single_major, bool, 0444); |
3cfa3b16 | 496 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: true)"); |
9b60e70b | 497 | |
7e9586ba GKH |
498 | static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count); |
499 | static ssize_t remove_store(struct bus_type *bus, const char *buf, | |
500 | size_t count); | |
501 | static ssize_t add_single_major_store(struct bus_type *bus, const char *buf, | |
502 | size_t count); | |
503 | static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf, | |
504 | size_t count); | |
6d69bb53 | 505 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
f0f8cef5 | 506 | |
9b60e70b ID |
507 | static int rbd_dev_id_to_minor(int dev_id) |
508 | { | |
7e513d43 | 509 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
510 | } |
511 | ||
512 | static int minor_to_rbd_dev_id(int minor) | |
513 | { | |
7e513d43 | 514 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
515 | } |
516 | ||
ed95b21a ID |
517 | static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev) |
518 | { | |
637cd060 ID |
519 | lockdep_assert_held(&rbd_dev->lock_rwsem); |
520 | ||
ed95b21a ID |
521 | return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED || |
522 | rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING; | |
523 | } | |
524 | ||
525 | static bool rbd_is_lock_owner(struct rbd_device *rbd_dev) | |
526 | { | |
527 | bool is_lock_owner; | |
528 | ||
529 | down_read(&rbd_dev->lock_rwsem); | |
530 | is_lock_owner = __rbd_is_lock_owner(rbd_dev); | |
531 | up_read(&rbd_dev->lock_rwsem); | |
532 | return is_lock_owner; | |
533 | } | |
534 | ||
7e9586ba | 535 | static ssize_t supported_features_show(struct bus_type *bus, char *buf) |
8767b293 ID |
536 | { |
537 | return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED); | |
538 | } | |
539 | ||
7e9586ba GKH |
540 | static BUS_ATTR_WO(add); |
541 | static BUS_ATTR_WO(remove); | |
542 | static BUS_ATTR_WO(add_single_major); | |
543 | static BUS_ATTR_WO(remove_single_major); | |
544 | static BUS_ATTR_RO(supported_features); | |
b15a21dd GKH |
545 | |
546 | static struct attribute *rbd_bus_attrs[] = { | |
547 | &bus_attr_add.attr, | |
548 | &bus_attr_remove.attr, | |
9b60e70b ID |
549 | &bus_attr_add_single_major.attr, |
550 | &bus_attr_remove_single_major.attr, | |
8767b293 | 551 | &bus_attr_supported_features.attr, |
b15a21dd | 552 | NULL, |
f0f8cef5 | 553 | }; |
92c76dc0 ID |
554 | |
555 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
556 | struct attribute *attr, int index) | |
557 | { | |
9b60e70b ID |
558 | if (!single_major && |
559 | (attr == &bus_attr_add_single_major.attr || | |
560 | attr == &bus_attr_remove_single_major.attr)) | |
561 | return 0; | |
562 | ||
92c76dc0 ID |
563 | return attr->mode; |
564 | } | |
565 | ||
566 | static const struct attribute_group rbd_bus_group = { | |
567 | .attrs = rbd_bus_attrs, | |
568 | .is_visible = rbd_bus_is_visible, | |
569 | }; | |
570 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
571 | |
572 | static struct bus_type rbd_bus_type = { | |
573 | .name = "rbd", | |
b15a21dd | 574 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
575 | }; |
576 | ||
577 | static void rbd_root_dev_release(struct device *dev) | |
578 | { | |
579 | } | |
580 | ||
581 | static struct device rbd_root_dev = { | |
582 | .init_name = "rbd", | |
583 | .release = rbd_root_dev_release, | |
584 | }; | |
585 | ||
06ecc6cb AE |
586 | static __printf(2, 3) |
587 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
588 | { | |
589 | struct va_format vaf; | |
590 | va_list args; | |
591 | ||
592 | va_start(args, fmt); | |
593 | vaf.fmt = fmt; | |
594 | vaf.va = &args; | |
595 | ||
596 | if (!rbd_dev) | |
597 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
598 | else if (rbd_dev->disk) | |
599 | printk(KERN_WARNING "%s: %s: %pV\n", | |
600 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
601 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
602 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
603 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
604 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
605 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
606 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
607 | else /* punt */ | |
608 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
609 | RBD_DRV_NAME, rbd_dev, &vaf); | |
610 | va_end(args); | |
611 | } | |
612 | ||
aafb230e AE |
613 | #ifdef RBD_DEBUG |
614 | #define rbd_assert(expr) \ | |
615 | if (unlikely(!(expr))) { \ | |
616 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
617 | "at line %d:\n\n" \ | |
618 | "\trbd_assert(%s);\n\n", \ | |
619 | __func__, __LINE__, #expr); \ | |
620 | BUG(); \ | |
621 | } | |
622 | #else /* !RBD_DEBUG */ | |
623 | # define rbd_assert(expr) ((void) 0) | |
624 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 625 | |
05a46afd | 626 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); |
8b3e1a56 | 627 | |
cc4a38bd | 628 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 629 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 630 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 631 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
632 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
633 | u64 snap_id); | |
2ad3d716 AE |
634 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
635 | u8 *order, u64 *snap_size); | |
636 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
637 | u64 *snap_features); | |
22e8bd51 | 638 | static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev); |
59c2be1e | 639 | |
54ab3b24 | 640 | static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result); |
0192ce2e ID |
641 | static void rbd_img_handle_request(struct rbd_img_request *img_req, int result); |
642 | ||
643 | /* | |
644 | * Return true if nothing else is pending. | |
645 | */ | |
646 | static bool pending_result_dec(struct pending_result *pending, int *result) | |
647 | { | |
648 | rbd_assert(pending->num_pending > 0); | |
649 | ||
650 | if (*result && !pending->result) | |
651 | pending->result = *result; | |
652 | if (--pending->num_pending) | |
653 | return false; | |
654 | ||
655 | *result = pending->result; | |
656 | return true; | |
657 | } | |
59c2be1e | 658 | |
602adf40 YS |
659 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
660 | { | |
f0f8cef5 | 661 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 662 | bool removing = false; |
602adf40 | 663 | |
a14ea269 | 664 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
665 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
666 | removing = true; | |
667 | else | |
668 | rbd_dev->open_count++; | |
a14ea269 | 669 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
670 | if (removing) |
671 | return -ENOENT; | |
672 | ||
c3e946ce | 673 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 674 | |
602adf40 YS |
675 | return 0; |
676 | } | |
677 | ||
db2a144b | 678 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
679 | { |
680 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
681 | unsigned long open_count_before; |
682 | ||
a14ea269 | 683 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 684 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 685 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 686 | rbd_assert(open_count_before > 0); |
dfc5606d | 687 | |
c3e946ce | 688 | put_device(&rbd_dev->dev); |
dfc5606d YS |
689 | } |
690 | ||
131fd9f6 GZ |
691 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
692 | { | |
1de797bb | 693 | int ro; |
131fd9f6 | 694 | |
1de797bb | 695 | if (get_user(ro, (int __user *)arg)) |
131fd9f6 GZ |
696 | return -EFAULT; |
697 | ||
1de797bb | 698 | /* Snapshots can't be marked read-write */ |
131fd9f6 GZ |
699 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) |
700 | return -EROFS; | |
701 | ||
1de797bb ID |
702 | /* Let blkdev_roset() handle it */ |
703 | return -ENOTTY; | |
131fd9f6 GZ |
704 | } |
705 | ||
706 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
707 | unsigned int cmd, unsigned long arg) | |
708 | { | |
709 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
1de797bb | 710 | int ret; |
131fd9f6 | 711 | |
131fd9f6 GZ |
712 | switch (cmd) { |
713 | case BLKROSET: | |
714 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
715 | break; | |
716 | default: | |
717 | ret = -ENOTTY; | |
718 | } | |
719 | ||
131fd9f6 GZ |
720 | return ret; |
721 | } | |
722 | ||
723 | #ifdef CONFIG_COMPAT | |
724 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
725 | unsigned int cmd, unsigned long arg) | |
726 | { | |
727 | return rbd_ioctl(bdev, mode, cmd, arg); | |
728 | } | |
729 | #endif /* CONFIG_COMPAT */ | |
730 | ||
602adf40 YS |
731 | static const struct block_device_operations rbd_bd_ops = { |
732 | .owner = THIS_MODULE, | |
733 | .open = rbd_open, | |
dfc5606d | 734 | .release = rbd_release, |
131fd9f6 GZ |
735 | .ioctl = rbd_ioctl, |
736 | #ifdef CONFIG_COMPAT | |
737 | .compat_ioctl = rbd_compat_ioctl, | |
738 | #endif | |
602adf40 YS |
739 | }; |
740 | ||
741 | /* | |
7262cfca | 742 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 743 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 744 | */ |
f8c38929 | 745 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
746 | { |
747 | struct rbd_client *rbdc; | |
748 | int ret = -ENOMEM; | |
749 | ||
37206ee5 | 750 | dout("%s:\n", __func__); |
602adf40 YS |
751 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
752 | if (!rbdc) | |
753 | goto out_opt; | |
754 | ||
755 | kref_init(&rbdc->kref); | |
756 | INIT_LIST_HEAD(&rbdc->node); | |
757 | ||
74da4a0f | 758 | rbdc->client = ceph_create_client(ceph_opts, rbdc); |
602adf40 | 759 | if (IS_ERR(rbdc->client)) |
08f75463 | 760 | goto out_rbdc; |
43ae4701 | 761 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
762 | |
763 | ret = ceph_open_session(rbdc->client); | |
764 | if (ret < 0) | |
08f75463 | 765 | goto out_client; |
602adf40 | 766 | |
432b8587 | 767 | spin_lock(&rbd_client_list_lock); |
602adf40 | 768 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 769 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 770 | |
37206ee5 | 771 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 772 | |
602adf40 | 773 | return rbdc; |
08f75463 | 774 | out_client: |
602adf40 | 775 | ceph_destroy_client(rbdc->client); |
08f75463 | 776 | out_rbdc: |
602adf40 YS |
777 | kfree(rbdc); |
778 | out_opt: | |
43ae4701 AE |
779 | if (ceph_opts) |
780 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
781 | dout("%s: error %d\n", __func__, ret); |
782 | ||
28f259b7 | 783 | return ERR_PTR(ret); |
602adf40 YS |
784 | } |
785 | ||
2f82ee54 AE |
786 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
787 | { | |
788 | kref_get(&rbdc->kref); | |
789 | ||
790 | return rbdc; | |
791 | } | |
792 | ||
602adf40 | 793 | /* |
1f7ba331 AE |
794 | * Find a ceph client with specific addr and configuration. If |
795 | * found, bump its reference count. | |
602adf40 | 796 | */ |
1f7ba331 | 797 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
798 | { |
799 | struct rbd_client *client_node; | |
1f7ba331 | 800 | bool found = false; |
602adf40 | 801 | |
43ae4701 | 802 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
803 | return NULL; |
804 | ||
1f7ba331 AE |
805 | spin_lock(&rbd_client_list_lock); |
806 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
807 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
808 | __rbd_get_client(client_node); |
809 | ||
1f7ba331 AE |
810 | found = true; |
811 | break; | |
812 | } | |
813 | } | |
814 | spin_unlock(&rbd_client_list_lock); | |
815 | ||
816 | return found ? client_node : NULL; | |
602adf40 YS |
817 | } |
818 | ||
59c2be1e | 819 | /* |
210c104c | 820 | * (Per device) rbd map options |
59c2be1e YS |
821 | */ |
822 | enum { | |
b5584180 | 823 | Opt_queue_depth, |
0c93e1b7 | 824 | Opt_alloc_size, |
34f55d0b | 825 | Opt_lock_timeout, |
59c2be1e YS |
826 | Opt_last_int, |
827 | /* int args above */ | |
b26c047b | 828 | Opt_pool_ns, |
59c2be1e YS |
829 | Opt_last_string, |
830 | /* string args above */ | |
cc0538b6 AE |
831 | Opt_read_only, |
832 | Opt_read_write, | |
80de1912 | 833 | Opt_lock_on_read, |
e010dd0a | 834 | Opt_exclusive, |
d9360540 | 835 | Opt_notrim, |
210c104c | 836 | Opt_err |
59c2be1e YS |
837 | }; |
838 | ||
43ae4701 | 839 | static match_table_t rbd_opts_tokens = { |
b5584180 | 840 | {Opt_queue_depth, "queue_depth=%d"}, |
0c93e1b7 | 841 | {Opt_alloc_size, "alloc_size=%d"}, |
34f55d0b | 842 | {Opt_lock_timeout, "lock_timeout=%d"}, |
59c2be1e | 843 | /* int args above */ |
b26c047b | 844 | {Opt_pool_ns, "_pool_ns=%s"}, |
59c2be1e | 845 | /* string args above */ |
be466c1c | 846 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
847 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
848 | {Opt_read_write, "read_write"}, | |
849 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
80de1912 | 850 | {Opt_lock_on_read, "lock_on_read"}, |
e010dd0a | 851 | {Opt_exclusive, "exclusive"}, |
d9360540 | 852 | {Opt_notrim, "notrim"}, |
210c104c | 853 | {Opt_err, NULL} |
59c2be1e YS |
854 | }; |
855 | ||
98571b5a | 856 | struct rbd_options { |
b5584180 | 857 | int queue_depth; |
0c93e1b7 | 858 | int alloc_size; |
34f55d0b | 859 | unsigned long lock_timeout; |
98571b5a | 860 | bool read_only; |
80de1912 | 861 | bool lock_on_read; |
e010dd0a | 862 | bool exclusive; |
d9360540 | 863 | bool trim; |
98571b5a AE |
864 | }; |
865 | ||
b5584180 | 866 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
0c93e1b7 | 867 | #define RBD_ALLOC_SIZE_DEFAULT (64 * 1024) |
34f55d0b | 868 | #define RBD_LOCK_TIMEOUT_DEFAULT 0 /* no timeout */ |
98571b5a | 869 | #define RBD_READ_ONLY_DEFAULT false |
80de1912 | 870 | #define RBD_LOCK_ON_READ_DEFAULT false |
e010dd0a | 871 | #define RBD_EXCLUSIVE_DEFAULT false |
d9360540 | 872 | #define RBD_TRIM_DEFAULT true |
98571b5a | 873 | |
c300156b ID |
874 | struct parse_rbd_opts_ctx { |
875 | struct rbd_spec *spec; | |
876 | struct rbd_options *opts; | |
877 | }; | |
878 | ||
59c2be1e YS |
879 | static int parse_rbd_opts_token(char *c, void *private) |
880 | { | |
c300156b | 881 | struct parse_rbd_opts_ctx *pctx = private; |
59c2be1e YS |
882 | substring_t argstr[MAX_OPT_ARGS]; |
883 | int token, intval, ret; | |
884 | ||
43ae4701 | 885 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
886 | if (token < Opt_last_int) { |
887 | ret = match_int(&argstr[0], &intval); | |
888 | if (ret < 0) { | |
2f56b6ba | 889 | pr_err("bad option arg (not int) at '%s'\n", c); |
59c2be1e YS |
890 | return ret; |
891 | } | |
892 | dout("got int token %d val %d\n", token, intval); | |
893 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 894 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
895 | } else { |
896 | dout("got token %d\n", token); | |
897 | } | |
898 | ||
899 | switch (token) { | |
b5584180 ID |
900 | case Opt_queue_depth: |
901 | if (intval < 1) { | |
902 | pr_err("queue_depth out of range\n"); | |
903 | return -EINVAL; | |
904 | } | |
c300156b | 905 | pctx->opts->queue_depth = intval; |
b5584180 | 906 | break; |
0c93e1b7 | 907 | case Opt_alloc_size: |
16d80c54 | 908 | if (intval < SECTOR_SIZE) { |
0c93e1b7 ID |
909 | pr_err("alloc_size out of range\n"); |
910 | return -EINVAL; | |
911 | } | |
912 | if (!is_power_of_2(intval)) { | |
913 | pr_err("alloc_size must be a power of 2\n"); | |
914 | return -EINVAL; | |
915 | } | |
916 | pctx->opts->alloc_size = intval; | |
917 | break; | |
34f55d0b DY |
918 | case Opt_lock_timeout: |
919 | /* 0 is "wait forever" (i.e. infinite timeout) */ | |
920 | if (intval < 0 || intval > INT_MAX / 1000) { | |
921 | pr_err("lock_timeout out of range\n"); | |
922 | return -EINVAL; | |
923 | } | |
c300156b | 924 | pctx->opts->lock_timeout = msecs_to_jiffies(intval * 1000); |
34f55d0b | 925 | break; |
b26c047b ID |
926 | case Opt_pool_ns: |
927 | kfree(pctx->spec->pool_ns); | |
928 | pctx->spec->pool_ns = match_strdup(argstr); | |
929 | if (!pctx->spec->pool_ns) | |
930 | return -ENOMEM; | |
34f55d0b | 931 | break; |
cc0538b6 | 932 | case Opt_read_only: |
c300156b | 933 | pctx->opts->read_only = true; |
cc0538b6 AE |
934 | break; |
935 | case Opt_read_write: | |
c300156b | 936 | pctx->opts->read_only = false; |
cc0538b6 | 937 | break; |
80de1912 | 938 | case Opt_lock_on_read: |
c300156b | 939 | pctx->opts->lock_on_read = true; |
80de1912 | 940 | break; |
e010dd0a | 941 | case Opt_exclusive: |
c300156b | 942 | pctx->opts->exclusive = true; |
e010dd0a | 943 | break; |
d9360540 | 944 | case Opt_notrim: |
c300156b | 945 | pctx->opts->trim = false; |
d9360540 | 946 | break; |
59c2be1e | 947 | default: |
210c104c ID |
948 | /* libceph prints "bad option" msg */ |
949 | return -EINVAL; | |
59c2be1e | 950 | } |
210c104c | 951 | |
59c2be1e YS |
952 | return 0; |
953 | } | |
954 | ||
6d2940c8 GZ |
955 | static char* obj_op_name(enum obj_operation_type op_type) |
956 | { | |
957 | switch (op_type) { | |
958 | case OBJ_OP_READ: | |
959 | return "read"; | |
960 | case OBJ_OP_WRITE: | |
961 | return "write"; | |
90e98c52 GZ |
962 | case OBJ_OP_DISCARD: |
963 | return "discard"; | |
6484cbe9 ID |
964 | case OBJ_OP_ZEROOUT: |
965 | return "zeroout"; | |
6d2940c8 GZ |
966 | default: |
967 | return "???"; | |
968 | } | |
969 | } | |
970 | ||
602adf40 YS |
971 | /* |
972 | * Destroy ceph client | |
d23a4b3f | 973 | * |
432b8587 | 974 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
975 | */ |
976 | static void rbd_client_release(struct kref *kref) | |
977 | { | |
978 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
979 | ||
37206ee5 | 980 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 981 | spin_lock(&rbd_client_list_lock); |
602adf40 | 982 | list_del(&rbdc->node); |
cd9d9f5d | 983 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
984 | |
985 | ceph_destroy_client(rbdc->client); | |
986 | kfree(rbdc); | |
987 | } | |
988 | ||
989 | /* | |
990 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
991 | * it. | |
992 | */ | |
9d3997fd | 993 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 994 | { |
c53d5893 AE |
995 | if (rbdc) |
996 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
997 | } |
998 | ||
5feb0d8d ID |
999 | /* |
1000 | * Get a ceph client with specific addr and configuration, if one does | |
1001 | * not exist create it. Either way, ceph_opts is consumed by this | |
1002 | * function. | |
1003 | */ | |
1004 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) | |
1005 | { | |
1006 | struct rbd_client *rbdc; | |
dd435855 | 1007 | int ret; |
5feb0d8d | 1008 | |
a32e4143 | 1009 | mutex_lock(&client_mutex); |
5feb0d8d | 1010 | rbdc = rbd_client_find(ceph_opts); |
dd435855 | 1011 | if (rbdc) { |
5feb0d8d | 1012 | ceph_destroy_options(ceph_opts); |
dd435855 ID |
1013 | |
1014 | /* | |
1015 | * Using an existing client. Make sure ->pg_pools is up to | |
1016 | * date before we look up the pool id in do_rbd_add(). | |
1017 | */ | |
9d4a227f ID |
1018 | ret = ceph_wait_for_latest_osdmap(rbdc->client, |
1019 | rbdc->client->options->mount_timeout); | |
dd435855 ID |
1020 | if (ret) { |
1021 | rbd_warn(NULL, "failed to get latest osdmap: %d", ret); | |
1022 | rbd_put_client(rbdc); | |
1023 | rbdc = ERR_PTR(ret); | |
1024 | } | |
1025 | } else { | |
5feb0d8d | 1026 | rbdc = rbd_client_create(ceph_opts); |
dd435855 | 1027 | } |
5feb0d8d ID |
1028 | mutex_unlock(&client_mutex); |
1029 | ||
1030 | return rbdc; | |
1031 | } | |
1032 | ||
a30b71b9 AE |
1033 | static bool rbd_image_format_valid(u32 image_format) |
1034 | { | |
1035 | return image_format == 1 || image_format == 2; | |
1036 | } | |
1037 | ||
8e94af8e AE |
1038 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
1039 | { | |
103a150f AE |
1040 | size_t size; |
1041 | u32 snap_count; | |
1042 | ||
1043 | /* The header has to start with the magic rbd header text */ | |
1044 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
1045 | return false; | |
1046 | ||
db2388b6 AE |
1047 | /* The bio layer requires at least sector-sized I/O */ |
1048 | ||
1049 | if (ondisk->options.order < SECTOR_SHIFT) | |
1050 | return false; | |
1051 | ||
1052 | /* If we use u64 in a few spots we may be able to loosen this */ | |
1053 | ||
1054 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
1055 | return false; | |
1056 | ||
103a150f AE |
1057 | /* |
1058 | * The size of a snapshot header has to fit in a size_t, and | |
1059 | * that limits the number of snapshots. | |
1060 | */ | |
1061 | snap_count = le32_to_cpu(ondisk->snap_count); | |
1062 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
1063 | if (snap_count > size / sizeof (__le64)) | |
1064 | return false; | |
1065 | ||
1066 | /* | |
1067 | * Not only that, but the size of the entire the snapshot | |
1068 | * header must also be representable in a size_t. | |
1069 | */ | |
1070 | size -= snap_count * sizeof (__le64); | |
1071 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
1072 | return false; | |
1073 | ||
1074 | return true; | |
8e94af8e AE |
1075 | } |
1076 | ||
5bc3fb17 ID |
1077 | /* |
1078 | * returns the size of an object in the image | |
1079 | */ | |
1080 | static u32 rbd_obj_bytes(struct rbd_image_header *header) | |
1081 | { | |
1082 | return 1U << header->obj_order; | |
1083 | } | |
1084 | ||
263423f8 ID |
1085 | static void rbd_init_layout(struct rbd_device *rbd_dev) |
1086 | { | |
1087 | if (rbd_dev->header.stripe_unit == 0 || | |
1088 | rbd_dev->header.stripe_count == 0) { | |
1089 | rbd_dev->header.stripe_unit = rbd_obj_bytes(&rbd_dev->header); | |
1090 | rbd_dev->header.stripe_count = 1; | |
1091 | } | |
1092 | ||
1093 | rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit; | |
1094 | rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count; | |
1095 | rbd_dev->layout.object_size = rbd_obj_bytes(&rbd_dev->header); | |
7e97332e ID |
1096 | rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id == CEPH_NOPOOL ? |
1097 | rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id; | |
263423f8 ID |
1098 | RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL); |
1099 | } | |
1100 | ||
602adf40 | 1101 | /* |
bb23e37a AE |
1102 | * Fill an rbd image header with information from the given format 1 |
1103 | * on-disk header. | |
602adf40 | 1104 | */ |
662518b1 | 1105 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 1106 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 1107 | { |
662518b1 | 1108 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
1109 | bool first_time = header->object_prefix == NULL; |
1110 | struct ceph_snap_context *snapc; | |
1111 | char *object_prefix = NULL; | |
1112 | char *snap_names = NULL; | |
1113 | u64 *snap_sizes = NULL; | |
ccece235 | 1114 | u32 snap_count; |
bb23e37a | 1115 | int ret = -ENOMEM; |
621901d6 | 1116 | u32 i; |
602adf40 | 1117 | |
bb23e37a | 1118 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 1119 | |
bb23e37a | 1120 | if (first_time) { |
848d796c ID |
1121 | object_prefix = kstrndup(ondisk->object_prefix, |
1122 | sizeof(ondisk->object_prefix), | |
1123 | GFP_KERNEL); | |
bb23e37a AE |
1124 | if (!object_prefix) |
1125 | return -ENOMEM; | |
bb23e37a | 1126 | } |
00f1f36f | 1127 | |
bb23e37a | 1128 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 1129 | |
bb23e37a AE |
1130 | snap_count = le32_to_cpu(ondisk->snap_count); |
1131 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
1132 | if (!snapc) | |
1133 | goto out_err; | |
1134 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 1135 | if (snap_count) { |
bb23e37a | 1136 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
1137 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
1138 | ||
bb23e37a | 1139 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 1140 | |
bb23e37a AE |
1141 | if (snap_names_len > (u64)SIZE_MAX) |
1142 | goto out_2big; | |
1143 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
1144 | if (!snap_names) | |
6a52325f AE |
1145 | goto out_err; |
1146 | ||
bb23e37a | 1147 | /* ...as well as the array of their sizes. */ |
88a25a5f ME |
1148 | snap_sizes = kmalloc_array(snap_count, |
1149 | sizeof(*header->snap_sizes), | |
1150 | GFP_KERNEL); | |
bb23e37a | 1151 | if (!snap_sizes) |
6a52325f | 1152 | goto out_err; |
bb23e37a | 1153 | |
f785cc1d | 1154 | /* |
bb23e37a AE |
1155 | * Copy the names, and fill in each snapshot's id |
1156 | * and size. | |
1157 | * | |
99a41ebc | 1158 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 1159 | * ondisk buffer we're working with has |
f785cc1d AE |
1160 | * snap_names_len bytes beyond the end of the |
1161 | * snapshot id array, this memcpy() is safe. | |
1162 | */ | |
bb23e37a AE |
1163 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
1164 | snaps = ondisk->snaps; | |
1165 | for (i = 0; i < snap_count; i++) { | |
1166 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
1167 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
1168 | } | |
602adf40 | 1169 | } |
6a52325f | 1170 | |
bb23e37a | 1171 | /* We won't fail any more, fill in the header */ |
621901d6 | 1172 | |
bb23e37a AE |
1173 | if (first_time) { |
1174 | header->object_prefix = object_prefix; | |
1175 | header->obj_order = ondisk->options.order; | |
263423f8 | 1176 | rbd_init_layout(rbd_dev); |
602adf40 | 1177 | } else { |
662518b1 AE |
1178 | ceph_put_snap_context(header->snapc); |
1179 | kfree(header->snap_names); | |
1180 | kfree(header->snap_sizes); | |
602adf40 | 1181 | } |
849b4260 | 1182 | |
bb23e37a | 1183 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1184 | |
f84344f3 | 1185 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1186 | header->snapc = snapc; |
1187 | header->snap_names = snap_names; | |
1188 | header->snap_sizes = snap_sizes; | |
468521c1 | 1189 | |
602adf40 | 1190 | return 0; |
bb23e37a AE |
1191 | out_2big: |
1192 | ret = -EIO; | |
6a52325f | 1193 | out_err: |
bb23e37a AE |
1194 | kfree(snap_sizes); |
1195 | kfree(snap_names); | |
1196 | ceph_put_snap_context(snapc); | |
1197 | kfree(object_prefix); | |
ccece235 | 1198 | |
bb23e37a | 1199 | return ret; |
602adf40 YS |
1200 | } |
1201 | ||
9682fc6d AE |
1202 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1203 | { | |
1204 | const char *snap_name; | |
1205 | ||
1206 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1207 | ||
1208 | /* Skip over names until we find the one we are looking for */ | |
1209 | ||
1210 | snap_name = rbd_dev->header.snap_names; | |
1211 | while (which--) | |
1212 | snap_name += strlen(snap_name) + 1; | |
1213 | ||
1214 | return kstrdup(snap_name, GFP_KERNEL); | |
1215 | } | |
1216 | ||
30d1cff8 AE |
1217 | /* |
1218 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1219 | * Note that result is for snapshots in *descending* order. | |
1220 | */ | |
1221 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1222 | { | |
1223 | u64 snap_id1 = *(u64 *)s1; | |
1224 | u64 snap_id2 = *(u64 *)s2; | |
1225 | ||
1226 | if (snap_id1 < snap_id2) | |
1227 | return 1; | |
1228 | return snap_id1 == snap_id2 ? 0 : -1; | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * Search a snapshot context to see if the given snapshot id is | |
1233 | * present. | |
1234 | * | |
1235 | * Returns the position of the snapshot id in the array if it's found, | |
1236 | * or BAD_SNAP_INDEX otherwise. | |
1237 | * | |
1238 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1239 | * reverse order, highest snapshot id first. | |
1240 | */ | |
9682fc6d AE |
1241 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1242 | { | |
1243 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1244 | u64 *found; |
9682fc6d | 1245 | |
30d1cff8 AE |
1246 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1247 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1248 | |
30d1cff8 | 1249 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1250 | } |
1251 | ||
2ad3d716 AE |
1252 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1253 | u64 snap_id) | |
9e15b77d | 1254 | { |
54cac61f | 1255 | u32 which; |
da6a6b63 | 1256 | const char *snap_name; |
9e15b77d | 1257 | |
54cac61f AE |
1258 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1259 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1260 | return ERR_PTR(-ENOENT); |
54cac61f | 1261 | |
da6a6b63 JD |
1262 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1263 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1264 | } |
1265 | ||
1266 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1267 | { | |
9e15b77d AE |
1268 | if (snap_id == CEPH_NOSNAP) |
1269 | return RBD_SNAP_HEAD_NAME; | |
1270 | ||
54cac61f AE |
1271 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1272 | if (rbd_dev->image_format == 1) | |
1273 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1274 | |
54cac61f | 1275 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1276 | } |
1277 | ||
2ad3d716 AE |
1278 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1279 | u64 *snap_size) | |
602adf40 | 1280 | { |
2ad3d716 AE |
1281 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1282 | if (snap_id == CEPH_NOSNAP) { | |
1283 | *snap_size = rbd_dev->header.image_size; | |
1284 | } else if (rbd_dev->image_format == 1) { | |
1285 | u32 which; | |
602adf40 | 1286 | |
2ad3d716 AE |
1287 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1288 | if (which == BAD_SNAP_INDEX) | |
1289 | return -ENOENT; | |
e86924a8 | 1290 | |
2ad3d716 AE |
1291 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1292 | } else { | |
1293 | u64 size = 0; | |
1294 | int ret; | |
1295 | ||
1296 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1297 | if (ret) | |
1298 | return ret; | |
1299 | ||
1300 | *snap_size = size; | |
1301 | } | |
1302 | return 0; | |
602adf40 YS |
1303 | } |
1304 | ||
2ad3d716 AE |
1305 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1306 | u64 *snap_features) | |
602adf40 | 1307 | { |
2ad3d716 AE |
1308 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1309 | if (snap_id == CEPH_NOSNAP) { | |
1310 | *snap_features = rbd_dev->header.features; | |
1311 | } else if (rbd_dev->image_format == 1) { | |
1312 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1313 | } else { |
2ad3d716 AE |
1314 | u64 features = 0; |
1315 | int ret; | |
8b0241f8 | 1316 | |
2ad3d716 AE |
1317 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1318 | if (ret) | |
1319 | return ret; | |
1320 | ||
1321 | *snap_features = features; | |
1322 | } | |
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1327 | { | |
8f4b7d98 | 1328 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1329 | u64 size = 0; |
1330 | u64 features = 0; | |
1331 | int ret; | |
1332 | ||
2ad3d716 AE |
1333 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1334 | if (ret) | |
1335 | return ret; | |
1336 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1337 | if (ret) | |
1338 | return ret; | |
1339 | ||
1340 | rbd_dev->mapping.size = size; | |
1341 | rbd_dev->mapping.features = features; | |
1342 | ||
8b0241f8 | 1343 | return 0; |
602adf40 YS |
1344 | } |
1345 | ||
d1cf5788 AE |
1346 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1347 | { | |
1348 | rbd_dev->mapping.size = 0; | |
1349 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1350 | } |
1351 | ||
5359a17d | 1352 | static void zero_bvec(struct bio_vec *bv) |
602adf40 | 1353 | { |
602adf40 | 1354 | void *buf; |
5359a17d | 1355 | unsigned long flags; |
602adf40 | 1356 | |
5359a17d ID |
1357 | buf = bvec_kmap_irq(bv, &flags); |
1358 | memset(buf, 0, bv->bv_len); | |
1359 | flush_dcache_page(bv->bv_page); | |
1360 | bvec_kunmap_irq(buf, &flags); | |
602adf40 YS |
1361 | } |
1362 | ||
5359a17d | 1363 | static void zero_bios(struct ceph_bio_iter *bio_pos, u32 off, u32 bytes) |
b9434c5b | 1364 | { |
5359a17d | 1365 | struct ceph_bio_iter it = *bio_pos; |
b9434c5b | 1366 | |
5359a17d ID |
1367 | ceph_bio_iter_advance(&it, off); |
1368 | ceph_bio_iter_advance_step(&it, bytes, ({ | |
1369 | zero_bvec(&bv); | |
1370 | })); | |
b9434c5b AE |
1371 | } |
1372 | ||
7e07efb1 | 1373 | static void zero_bvecs(struct ceph_bvec_iter *bvec_pos, u32 off, u32 bytes) |
602adf40 | 1374 | { |
7e07efb1 | 1375 | struct ceph_bvec_iter it = *bvec_pos; |
602adf40 | 1376 | |
7e07efb1 ID |
1377 | ceph_bvec_iter_advance(&it, off); |
1378 | ceph_bvec_iter_advance_step(&it, bytes, ({ | |
1379 | zero_bvec(&bv); | |
1380 | })); | |
f7760dad AE |
1381 | } |
1382 | ||
1383 | /* | |
3da691bf | 1384 | * Zero a range in @obj_req data buffer defined by a bio (list) or |
afb97888 | 1385 | * (private) bio_vec array. |
f7760dad | 1386 | * |
3da691bf | 1387 | * @off is relative to the start of the data buffer. |
926f9b3f | 1388 | */ |
3da691bf ID |
1389 | static void rbd_obj_zero_range(struct rbd_obj_request *obj_req, u32 off, |
1390 | u32 bytes) | |
926f9b3f | 1391 | { |
54ab3b24 ID |
1392 | dout("%s %p data buf %u~%u\n", __func__, obj_req, off, bytes); |
1393 | ||
ecc633ca | 1394 | switch (obj_req->img_request->data_type) { |
3da691bf ID |
1395 | case OBJ_REQUEST_BIO: |
1396 | zero_bios(&obj_req->bio_pos, off, bytes); | |
1397 | break; | |
1398 | case OBJ_REQUEST_BVECS: | |
afb97888 | 1399 | case OBJ_REQUEST_OWN_BVECS: |
3da691bf ID |
1400 | zero_bvecs(&obj_req->bvec_pos, off, bytes); |
1401 | break; | |
1402 | default: | |
16809372 | 1403 | BUG(); |
6365d33a AE |
1404 | } |
1405 | } | |
1406 | ||
bf0d5f50 AE |
1407 | static void rbd_obj_request_destroy(struct kref *kref); |
1408 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1409 | { | |
1410 | rbd_assert(obj_request != NULL); | |
37206ee5 | 1411 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
2c935bc5 | 1412 | kref_read(&obj_request->kref)); |
bf0d5f50 AE |
1413 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1414 | } | |
1415 | ||
bf0d5f50 AE |
1416 | static void rbd_img_request_destroy(struct kref *kref); |
1417 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1418 | { | |
1419 | rbd_assert(img_request != NULL); | |
37206ee5 | 1420 | dout("%s: img %p (was %d)\n", __func__, img_request, |
2c935bc5 | 1421 | kref_read(&img_request->kref)); |
e93aca0a | 1422 | kref_put(&img_request->kref, rbd_img_request_destroy); |
bf0d5f50 AE |
1423 | } |
1424 | ||
1425 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1426 | struct rbd_obj_request *obj_request) | |
1427 | { | |
25dcf954 AE |
1428 | rbd_assert(obj_request->img_request == NULL); |
1429 | ||
b155e86c | 1430 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1431 | obj_request->img_request = img_request; |
15961b44 | 1432 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
bf0d5f50 AE |
1433 | } |
1434 | ||
1435 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1436 | struct rbd_obj_request *obj_request) | |
1437 | { | |
15961b44 | 1438 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); |
43df3d35 | 1439 | list_del(&obj_request->ex.oe_item); |
bf0d5f50 | 1440 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 AE |
1441 | rbd_obj_request_put(obj_request); |
1442 | } | |
1443 | ||
a086a1b8 | 1444 | static void rbd_osd_submit(struct ceph_osd_request *osd_req) |
bf0d5f50 | 1445 | { |
a086a1b8 | 1446 | struct rbd_obj_request *obj_req = osd_req->r_priv; |
980917fc | 1447 | |
a086a1b8 ID |
1448 | dout("%s osd_req %p for obj_req %p objno %llu %llu~%llu\n", |
1449 | __func__, osd_req, obj_req, obj_req->ex.oe_objno, | |
1450 | obj_req->ex.oe_off, obj_req->ex.oe_len); | |
980917fc | 1451 | ceph_osdc_start_request(osd_req->r_osdc, osd_req, false); |
bf0d5f50 AE |
1452 | } |
1453 | ||
0c425248 AE |
1454 | /* |
1455 | * The default/initial value for all image request flags is 0. Each | |
1456 | * is conditionally set to 1 at image request initialization time | |
1457 | * and currently never change thereafter. | |
1458 | */ | |
d0b2e944 AE |
1459 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1460 | { | |
1461 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1462 | smp_mb(); | |
1463 | } | |
1464 | ||
a2acd00e AE |
1465 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1466 | { | |
1467 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1468 | smp_mb(); | |
1469 | } | |
1470 | ||
d0b2e944 AE |
1471 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1472 | { | |
1473 | smp_mb(); | |
1474 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1475 | } | |
1476 | ||
3da691bf | 1477 | static bool rbd_obj_is_entire(struct rbd_obj_request *obj_req) |
6e2a4505 | 1478 | { |
3da691bf | 1479 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
b9434c5b | 1480 | |
43df3d35 ID |
1481 | return !obj_req->ex.oe_off && |
1482 | obj_req->ex.oe_len == rbd_dev->layout.object_size; | |
6e2a4505 AE |
1483 | } |
1484 | ||
3da691bf | 1485 | static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req) |
bf0d5f50 | 1486 | { |
3da691bf | 1487 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
bf0d5f50 | 1488 | |
43df3d35 | 1489 | return obj_req->ex.oe_off + obj_req->ex.oe_len == |
3da691bf | 1490 | rbd_dev->layout.object_size; |
0dcc685e ID |
1491 | } |
1492 | ||
13488d53 ID |
1493 | /* |
1494 | * Must be called after rbd_obj_calc_img_extents(). | |
1495 | */ | |
1496 | static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req) | |
1497 | { | |
1498 | if (!obj_req->num_img_extents || | |
9b17eb2c ID |
1499 | (rbd_obj_is_entire(obj_req) && |
1500 | !obj_req->img_request->snapc->num_snaps)) | |
13488d53 ID |
1501 | return false; |
1502 | ||
1503 | return true; | |
1504 | } | |
1505 | ||
86bd7998 | 1506 | static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req) |
bf0d5f50 | 1507 | { |
86bd7998 ID |
1508 | return ceph_file_extents_bytes(obj_req->img_extents, |
1509 | obj_req->num_img_extents); | |
bf0d5f50 AE |
1510 | } |
1511 | ||
3da691bf | 1512 | static bool rbd_img_is_write(struct rbd_img_request *img_req) |
bf0d5f50 | 1513 | { |
9bb0248d | 1514 | switch (img_req->op_type) { |
3da691bf ID |
1515 | case OBJ_OP_READ: |
1516 | return false; | |
1517 | case OBJ_OP_WRITE: | |
1518 | case OBJ_OP_DISCARD: | |
6484cbe9 | 1519 | case OBJ_OP_ZEROOUT: |
3da691bf ID |
1520 | return true; |
1521 | default: | |
c6244b3b | 1522 | BUG(); |
3da691bf | 1523 | } |
90e98c52 GZ |
1524 | } |
1525 | ||
85e084fe | 1526 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req) |
bf0d5f50 | 1527 | { |
3da691bf | 1528 | struct rbd_obj_request *obj_req = osd_req->r_priv; |
54ab3b24 | 1529 | int result; |
bf0d5f50 | 1530 | |
3da691bf ID |
1531 | dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req, |
1532 | osd_req->r_result, obj_req); | |
bf0d5f50 | 1533 | |
54ab3b24 ID |
1534 | /* |
1535 | * Writes aren't allowed to return a data payload. In some | |
1536 | * guarded write cases (e.g. stat + zero on an empty object) | |
1537 | * a stat response makes it through, but we don't care. | |
1538 | */ | |
1539 | if (osd_req->r_result > 0 && rbd_img_is_write(obj_req->img_request)) | |
1540 | result = 0; | |
3da691bf | 1541 | else |
54ab3b24 | 1542 | result = osd_req->r_result; |
bf0d5f50 | 1543 | |
54ab3b24 | 1544 | rbd_obj_handle_request(obj_req, result); |
bf0d5f50 AE |
1545 | } |
1546 | ||
bcbab1db | 1547 | static void rbd_osd_format_read(struct ceph_osd_request *osd_req) |
430c28c3 | 1548 | { |
bcbab1db | 1549 | struct rbd_obj_request *obj_request = osd_req->r_priv; |
430c28c3 | 1550 | |
a162b308 | 1551 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
7c84883a | 1552 | osd_req->r_snapid = obj_request->img_request->snap_id; |
9d4df01f AE |
1553 | } |
1554 | ||
bcbab1db | 1555 | static void rbd_osd_format_write(struct ceph_osd_request *osd_req) |
9d4df01f | 1556 | { |
bcbab1db | 1557 | struct rbd_obj_request *obj_request = osd_req->r_priv; |
9d4df01f | 1558 | |
a162b308 | 1559 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE; |
fac02ddf | 1560 | ktime_get_real_ts64(&osd_req->r_mtime); |
43df3d35 | 1561 | osd_req->r_data_offset = obj_request->ex.oe_off; |
430c28c3 AE |
1562 | } |
1563 | ||
bc81207e | 1564 | static struct ceph_osd_request * |
bcbab1db ID |
1565 | __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, |
1566 | struct ceph_snap_context *snapc, int num_ops) | |
bc81207e | 1567 | { |
e28eded5 | 1568 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
bc81207e ID |
1569 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
1570 | struct ceph_osd_request *req; | |
a90bb0c1 ID |
1571 | const char *name_format = rbd_dev->image_format == 1 ? |
1572 | RBD_V1_DATA_FORMAT : RBD_V2_DATA_FORMAT; | |
bcbab1db | 1573 | int ret; |
bc81207e | 1574 | |
e28eded5 | 1575 | req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, GFP_NOIO); |
bc81207e | 1576 | if (!req) |
bcbab1db | 1577 | return ERR_PTR(-ENOMEM); |
bc81207e | 1578 | |
bcbab1db | 1579 | list_add_tail(&req->r_private_item, &obj_req->osd_reqs); |
bc81207e | 1580 | req->r_callback = rbd_osd_req_callback; |
a162b308 | 1581 | req->r_priv = obj_req; |
bc81207e | 1582 | |
b26c047b ID |
1583 | /* |
1584 | * Data objects may be stored in a separate pool, but always in | |
1585 | * the same namespace in that pool as the header in its pool. | |
1586 | */ | |
1587 | ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc); | |
bc81207e | 1588 | req->r_base_oloc.pool = rbd_dev->layout.pool_id; |
b26c047b | 1589 | |
bcbab1db ID |
1590 | ret = ceph_oid_aprintf(&req->r_base_oid, GFP_NOIO, name_format, |
1591 | rbd_dev->header.object_prefix, | |
1592 | obj_req->ex.oe_objno); | |
1593 | if (ret) | |
1594 | return ERR_PTR(ret); | |
bc81207e | 1595 | |
bc81207e | 1596 | return req; |
bc81207e ID |
1597 | } |
1598 | ||
e28eded5 | 1599 | static struct ceph_osd_request * |
bcbab1db | 1600 | rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops) |
bf0d5f50 | 1601 | { |
bcbab1db ID |
1602 | return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc, |
1603 | num_ops); | |
bf0d5f50 AE |
1604 | } |
1605 | ||
ecc633ca | 1606 | static struct rbd_obj_request *rbd_obj_request_create(void) |
bf0d5f50 AE |
1607 | { |
1608 | struct rbd_obj_request *obj_request; | |
bf0d5f50 | 1609 | |
5a60e876 | 1610 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
6c696d85 | 1611 | if (!obj_request) |
f907ad55 | 1612 | return NULL; |
f907ad55 | 1613 | |
43df3d35 | 1614 | ceph_object_extent_init(&obj_request->ex); |
bcbab1db | 1615 | INIT_LIST_HEAD(&obj_request->osd_reqs); |
85b5e6d1 | 1616 | mutex_init(&obj_request->state_mutex); |
bf0d5f50 AE |
1617 | kref_init(&obj_request->kref); |
1618 | ||
67e2b652 | 1619 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1620 | return obj_request; |
1621 | } | |
1622 | ||
1623 | static void rbd_obj_request_destroy(struct kref *kref) | |
1624 | { | |
1625 | struct rbd_obj_request *obj_request; | |
bcbab1db | 1626 | struct ceph_osd_request *osd_req; |
7e07efb1 | 1627 | u32 i; |
bf0d5f50 AE |
1628 | |
1629 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1630 | ||
37206ee5 AE |
1631 | dout("%s: obj %p\n", __func__, obj_request); |
1632 | ||
bcbab1db ID |
1633 | while (!list_empty(&obj_request->osd_reqs)) { |
1634 | osd_req = list_first_entry(&obj_request->osd_reqs, | |
1635 | struct ceph_osd_request, r_private_item); | |
1636 | list_del_init(&osd_req->r_private_item); | |
1637 | ceph_osdc_put_request(osd_req); | |
1638 | } | |
bf0d5f50 | 1639 | |
ecc633ca | 1640 | switch (obj_request->img_request->data_type) { |
9969ebc5 | 1641 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1642 | case OBJ_REQUEST_BIO: |
7e07efb1 | 1643 | case OBJ_REQUEST_BVECS: |
5359a17d | 1644 | break; /* Nothing to do */ |
afb97888 ID |
1645 | case OBJ_REQUEST_OWN_BVECS: |
1646 | kfree(obj_request->bvec_pos.bvecs); | |
788e2df3 | 1647 | break; |
7e07efb1 | 1648 | default: |
16809372 | 1649 | BUG(); |
bf0d5f50 AE |
1650 | } |
1651 | ||
86bd7998 | 1652 | kfree(obj_request->img_extents); |
7e07efb1 ID |
1653 | if (obj_request->copyup_bvecs) { |
1654 | for (i = 0; i < obj_request->copyup_bvec_count; i++) { | |
1655 | if (obj_request->copyup_bvecs[i].bv_page) | |
1656 | __free_page(obj_request->copyup_bvecs[i].bv_page); | |
1657 | } | |
1658 | kfree(obj_request->copyup_bvecs); | |
bf0d5f50 AE |
1659 | } |
1660 | ||
868311b1 | 1661 | kmem_cache_free(rbd_obj_request_cache, obj_request); |
bf0d5f50 AE |
1662 | } |
1663 | ||
fb65d228 AE |
1664 | /* It's OK to call this for a device with no parent */ |
1665 | ||
1666 | static void rbd_spec_put(struct rbd_spec *spec); | |
1667 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
1668 | { | |
1669 | rbd_dev_remove_parent(rbd_dev); | |
1670 | rbd_spec_put(rbd_dev->parent_spec); | |
1671 | rbd_dev->parent_spec = NULL; | |
1672 | rbd_dev->parent_overlap = 0; | |
1673 | } | |
1674 | ||
a2acd00e AE |
1675 | /* |
1676 | * Parent image reference counting is used to determine when an | |
1677 | * image's parent fields can be safely torn down--after there are no | |
1678 | * more in-flight requests to the parent image. When the last | |
1679 | * reference is dropped, cleaning them up is safe. | |
1680 | */ | |
1681 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
1682 | { | |
1683 | int counter; | |
1684 | ||
1685 | if (!rbd_dev->parent_spec) | |
1686 | return; | |
1687 | ||
1688 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
1689 | if (counter > 0) | |
1690 | return; | |
1691 | ||
1692 | /* Last reference; clean up parent data structures */ | |
1693 | ||
1694 | if (!counter) | |
1695 | rbd_dev_unparent(rbd_dev); | |
1696 | else | |
9584d508 | 1697 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
1698 | } |
1699 | ||
1700 | /* | |
1701 | * If an image has a non-zero parent overlap, get a reference to its | |
1702 | * parent. | |
1703 | * | |
1704 | * Returns true if the rbd device has a parent with a non-zero | |
1705 | * overlap and a reference for it was successfully taken, or | |
1706 | * false otherwise. | |
1707 | */ | |
1708 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
1709 | { | |
ae43e9d0 | 1710 | int counter = 0; |
a2acd00e AE |
1711 | |
1712 | if (!rbd_dev->parent_spec) | |
1713 | return false; | |
1714 | ||
ae43e9d0 ID |
1715 | down_read(&rbd_dev->header_rwsem); |
1716 | if (rbd_dev->parent_overlap) | |
1717 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
1718 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
1719 | |
1720 | if (counter < 0) | |
9584d508 | 1721 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 1722 | |
ae43e9d0 | 1723 | return counter > 0; |
a2acd00e AE |
1724 | } |
1725 | ||
bf0d5f50 AE |
1726 | /* |
1727 | * Caller is responsible for filling in the list of object requests | |
1728 | * that comprises the image request, and the Linux request pointer | |
1729 | * (if there is one). | |
1730 | */ | |
cc344fa1 AE |
1731 | static struct rbd_img_request *rbd_img_request_create( |
1732 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1733 | enum obj_operation_type op_type, |
4e752f0a | 1734 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
1735 | { |
1736 | struct rbd_img_request *img_request; | |
bf0d5f50 | 1737 | |
a0c5895b | 1738 | img_request = kmem_cache_zalloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
1739 | if (!img_request) |
1740 | return NULL; | |
1741 | ||
bf0d5f50 | 1742 | img_request->rbd_dev = rbd_dev; |
9bb0248d | 1743 | img_request->op_type = op_type; |
9bb0248d | 1744 | if (!rbd_img_is_write(img_request)) |
bf0d5f50 | 1745 | img_request->snap_id = rbd_dev->spec->snap_id; |
9bb0248d ID |
1746 | else |
1747 | img_request->snapc = snapc; | |
1748 | ||
a2acd00e | 1749 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 1750 | img_request_layered_set(img_request); |
a0c5895b | 1751 | |
e1fddc8f | 1752 | INIT_LIST_HEAD(&img_request->lock_item); |
43df3d35 | 1753 | INIT_LIST_HEAD(&img_request->object_extents); |
0192ce2e | 1754 | mutex_init(&img_request->state_mutex); |
bf0d5f50 AE |
1755 | kref_init(&img_request->kref); |
1756 | ||
bf0d5f50 AE |
1757 | return img_request; |
1758 | } | |
1759 | ||
1760 | static void rbd_img_request_destroy(struct kref *kref) | |
1761 | { | |
1762 | struct rbd_img_request *img_request; | |
1763 | struct rbd_obj_request *obj_request; | |
1764 | struct rbd_obj_request *next_obj_request; | |
1765 | ||
1766 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1767 | ||
37206ee5 AE |
1768 | dout("%s: img %p\n", __func__, img_request); |
1769 | ||
e1fddc8f | 1770 | WARN_ON(!list_empty(&img_request->lock_item)); |
bf0d5f50 AE |
1771 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1772 | rbd_img_obj_request_del(img_request, obj_request); | |
1773 | ||
a2acd00e AE |
1774 | if (img_request_layered_test(img_request)) { |
1775 | img_request_layered_clear(img_request); | |
1776 | rbd_dev_parent_put(img_request->rbd_dev); | |
1777 | } | |
1778 | ||
9bb0248d | 1779 | if (rbd_img_is_write(img_request)) |
812164f8 | 1780 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 1781 | |
1c2a9dfe | 1782 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
1783 | } |
1784 | ||
22e8bd51 ID |
1785 | #define BITS_PER_OBJ 2 |
1786 | #define OBJS_PER_BYTE (BITS_PER_BYTE / BITS_PER_OBJ) | |
1787 | #define OBJ_MASK ((1 << BITS_PER_OBJ) - 1) | |
e93f3152 | 1788 | |
22e8bd51 ID |
1789 | static void __rbd_object_map_index(struct rbd_device *rbd_dev, u64 objno, |
1790 | u64 *index, u8 *shift) | |
1791 | { | |
1792 | u32 off; | |
e93f3152 | 1793 | |
22e8bd51 ID |
1794 | rbd_assert(objno < rbd_dev->object_map_size); |
1795 | *index = div_u64_rem(objno, OBJS_PER_BYTE, &off); | |
1796 | *shift = (OBJS_PER_BYTE - off - 1) * BITS_PER_OBJ; | |
1797 | } | |
e93f3152 | 1798 | |
22e8bd51 ID |
1799 | static u8 __rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno) |
1800 | { | |
1801 | u64 index; | |
1802 | u8 shift; | |
e93f3152 | 1803 | |
22e8bd51 ID |
1804 | lockdep_assert_held(&rbd_dev->object_map_lock); |
1805 | __rbd_object_map_index(rbd_dev, objno, &index, &shift); | |
1806 | return (rbd_dev->object_map[index] >> shift) & OBJ_MASK; | |
e93f3152 AE |
1807 | } |
1808 | ||
22e8bd51 | 1809 | static void __rbd_object_map_set(struct rbd_device *rbd_dev, u64 objno, u8 val) |
e93f3152 | 1810 | { |
22e8bd51 ID |
1811 | u64 index; |
1812 | u8 shift; | |
1813 | u8 *p; | |
e93f3152 | 1814 | |
22e8bd51 ID |
1815 | lockdep_assert_held(&rbd_dev->object_map_lock); |
1816 | rbd_assert(!(val & ~OBJ_MASK)); | |
e93f3152 | 1817 | |
22e8bd51 ID |
1818 | __rbd_object_map_index(rbd_dev, objno, &index, &shift); |
1819 | p = &rbd_dev->object_map[index]; | |
1820 | *p = (*p & ~(OBJ_MASK << shift)) | (val << shift); | |
e93f3152 AE |
1821 | } |
1822 | ||
22e8bd51 | 1823 | static u8 rbd_object_map_get(struct rbd_device *rbd_dev, u64 objno) |
1217857f | 1824 | { |
22e8bd51 ID |
1825 | u8 state; |
1826 | ||
1827 | spin_lock(&rbd_dev->object_map_lock); | |
1828 | state = __rbd_object_map_get(rbd_dev, objno); | |
1829 | spin_unlock(&rbd_dev->object_map_lock); | |
1830 | return state; | |
3da691bf | 1831 | } |
1217857f | 1832 | |
22e8bd51 | 1833 | static bool use_object_map(struct rbd_device *rbd_dev) |
3da691bf | 1834 | { |
22e8bd51 ID |
1835 | return ((rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) && |
1836 | !(rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID)); | |
3da691bf ID |
1837 | } |
1838 | ||
22e8bd51 | 1839 | static bool rbd_object_map_may_exist(struct rbd_device *rbd_dev, u64 objno) |
3da691bf | 1840 | { |
22e8bd51 | 1841 | u8 state; |
8b3e1a56 | 1842 | |
22e8bd51 ID |
1843 | /* fall back to default logic if object map is disabled or invalid */ |
1844 | if (!use_object_map(rbd_dev)) | |
1845 | return true; | |
3da691bf | 1846 | |
22e8bd51 ID |
1847 | state = rbd_object_map_get(rbd_dev, objno); |
1848 | return state != OBJECT_NONEXISTENT; | |
1217857f AE |
1849 | } |
1850 | ||
22e8bd51 ID |
1851 | static void rbd_object_map_name(struct rbd_device *rbd_dev, u64 snap_id, |
1852 | struct ceph_object_id *oid) | |
13488d53 | 1853 | { |
22e8bd51 ID |
1854 | if (snap_id == CEPH_NOSNAP) |
1855 | ceph_oid_printf(oid, "%s%s", RBD_OBJECT_MAP_PREFIX, | |
1856 | rbd_dev->spec->image_id); | |
1857 | else | |
1858 | ceph_oid_printf(oid, "%s%s.%016llx", RBD_OBJECT_MAP_PREFIX, | |
1859 | rbd_dev->spec->image_id, snap_id); | |
13488d53 ID |
1860 | } |
1861 | ||
22e8bd51 | 1862 | static int rbd_object_map_lock(struct rbd_device *rbd_dev) |
2169238d | 1863 | { |
22e8bd51 ID |
1864 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
1865 | CEPH_DEFINE_OID_ONSTACK(oid); | |
1866 | u8 lock_type; | |
1867 | char *lock_tag; | |
1868 | struct ceph_locker *lockers; | |
1869 | u32 num_lockers; | |
1870 | bool broke_lock = false; | |
1871 | int ret; | |
2169238d | 1872 | |
22e8bd51 | 1873 | rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid); |
2169238d | 1874 | |
22e8bd51 ID |
1875 | again: |
1876 | ret = ceph_cls_lock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
1877 | CEPH_CLS_LOCK_EXCLUSIVE, "", "", "", 0); | |
1878 | if (ret != -EBUSY || broke_lock) { | |
1879 | if (ret == -EEXIST) | |
1880 | ret = 0; /* already locked by myself */ | |
1881 | if (ret) | |
1882 | rbd_warn(rbd_dev, "failed to lock object map: %d", ret); | |
1883 | return ret; | |
1884 | } | |
2169238d | 1885 | |
22e8bd51 ID |
1886 | ret = ceph_cls_lock_info(osdc, &oid, &rbd_dev->header_oloc, |
1887 | RBD_LOCK_NAME, &lock_type, &lock_tag, | |
1888 | &lockers, &num_lockers); | |
1889 | if (ret) { | |
1890 | if (ret == -ENOENT) | |
1891 | goto again; | |
3da691bf | 1892 | |
22e8bd51 | 1893 | rbd_warn(rbd_dev, "failed to get object map lockers: %d", ret); |
86bd7998 | 1894 | return ret; |
22e8bd51 | 1895 | } |
86bd7998 | 1896 | |
22e8bd51 ID |
1897 | kfree(lock_tag); |
1898 | if (num_lockers == 0) | |
1899 | goto again; | |
2169238d | 1900 | |
22e8bd51 ID |
1901 | rbd_warn(rbd_dev, "breaking object map lock owned by %s%llu", |
1902 | ENTITY_NAME(lockers[0].id.name)); | |
2169238d | 1903 | |
22e8bd51 ID |
1904 | ret = ceph_cls_break_lock(osdc, &oid, &rbd_dev->header_oloc, |
1905 | RBD_LOCK_NAME, lockers[0].id.cookie, | |
1906 | &lockers[0].id.name); | |
1907 | ceph_free_lockers(lockers, num_lockers); | |
1908 | if (ret) { | |
1909 | if (ret == -ENOENT) | |
1910 | goto again; | |
13488d53 | 1911 | |
22e8bd51 ID |
1912 | rbd_warn(rbd_dev, "failed to break object map lock: %d", ret); |
1913 | return ret; | |
3da691bf ID |
1914 | } |
1915 | ||
22e8bd51 ID |
1916 | broke_lock = true; |
1917 | goto again; | |
2169238d AE |
1918 | } |
1919 | ||
22e8bd51 | 1920 | static void rbd_object_map_unlock(struct rbd_device *rbd_dev) |
6484cbe9 | 1921 | { |
22e8bd51 ID |
1922 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
1923 | CEPH_DEFINE_OID_ONSTACK(oid); | |
1924 | int ret; | |
1925 | ||
1926 | rbd_object_map_name(rbd_dev, CEPH_NOSNAP, &oid); | |
1927 | ||
1928 | ret = ceph_cls_unlock(osdc, &oid, &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
1929 | ""); | |
1930 | if (ret && ret != -ENOENT) | |
1931 | rbd_warn(rbd_dev, "failed to unlock object map: %d", ret); | |
6484cbe9 ID |
1932 | } |
1933 | ||
22e8bd51 | 1934 | static int decode_object_map_header(void **p, void *end, u64 *object_map_size) |
6484cbe9 | 1935 | { |
22e8bd51 ID |
1936 | u8 struct_v; |
1937 | u32 struct_len; | |
1938 | u32 header_len; | |
1939 | void *header_end; | |
6484cbe9 ID |
1940 | int ret; |
1941 | ||
22e8bd51 ID |
1942 | ceph_decode_32_safe(p, end, header_len, e_inval); |
1943 | header_end = *p + header_len; | |
0c93e1b7 | 1944 | |
22e8bd51 ID |
1945 | ret = ceph_start_decoding(p, end, 1, "BitVector header", &struct_v, |
1946 | &struct_len); | |
6484cbe9 ID |
1947 | if (ret) |
1948 | return ret; | |
1949 | ||
22e8bd51 | 1950 | ceph_decode_64_safe(p, end, *object_map_size, e_inval); |
6484cbe9 | 1951 | |
22e8bd51 | 1952 | *p = header_end; |
6484cbe9 | 1953 | return 0; |
22e8bd51 ID |
1954 | |
1955 | e_inval: | |
1956 | return -EINVAL; | |
6484cbe9 ID |
1957 | } |
1958 | ||
22e8bd51 | 1959 | static int __rbd_object_map_load(struct rbd_device *rbd_dev) |
13488d53 | 1960 | { |
22e8bd51 ID |
1961 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
1962 | CEPH_DEFINE_OID_ONSTACK(oid); | |
1963 | struct page **pages; | |
1964 | void *p, *end; | |
1965 | size_t reply_len; | |
1966 | u64 num_objects; | |
1967 | u64 object_map_bytes; | |
1968 | u64 object_map_size; | |
1969 | int num_pages; | |
1970 | int ret; | |
13488d53 | 1971 | |
22e8bd51 | 1972 | rbd_assert(!rbd_dev->object_map && !rbd_dev->object_map_size); |
13488d53 | 1973 | |
22e8bd51 ID |
1974 | num_objects = ceph_get_num_objects(&rbd_dev->layout, |
1975 | rbd_dev->mapping.size); | |
1976 | object_map_bytes = DIV_ROUND_UP_ULL(num_objects * BITS_PER_OBJ, | |
1977 | BITS_PER_BYTE); | |
1978 | num_pages = calc_pages_for(0, object_map_bytes) + 1; | |
1979 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); | |
1980 | if (IS_ERR(pages)) | |
1981 | return PTR_ERR(pages); | |
13488d53 | 1982 | |
22e8bd51 ID |
1983 | reply_len = num_pages * PAGE_SIZE; |
1984 | rbd_object_map_name(rbd_dev, rbd_dev->spec->snap_id, &oid); | |
1985 | ret = ceph_osdc_call(osdc, &oid, &rbd_dev->header_oloc, | |
1986 | "rbd", "object_map_load", CEPH_OSD_FLAG_READ, | |
1987 | NULL, 0, pages, &reply_len); | |
1988 | if (ret) | |
1989 | goto out; | |
3b434a2a | 1990 | |
22e8bd51 ID |
1991 | p = page_address(pages[0]); |
1992 | end = p + min(reply_len, (size_t)PAGE_SIZE); | |
1993 | ret = decode_object_map_header(&p, end, &object_map_size); | |
1994 | if (ret) | |
1995 | goto out; | |
1996 | ||
1997 | if (object_map_size != num_objects) { | |
1998 | rbd_warn(rbd_dev, "object map size mismatch: %llu vs %llu", | |
1999 | object_map_size, num_objects); | |
2000 | ret = -EINVAL; | |
2001 | goto out; | |
3b434a2a JD |
2002 | } |
2003 | ||
22e8bd51 ID |
2004 | if (offset_in_page(p) + object_map_bytes > reply_len) { |
2005 | ret = -EINVAL; | |
2006 | goto out; | |
2007 | } | |
2008 | ||
2009 | rbd_dev->object_map = kvmalloc(object_map_bytes, GFP_KERNEL); | |
2010 | if (!rbd_dev->object_map) { | |
2011 | ret = -ENOMEM; | |
2012 | goto out; | |
2013 | } | |
2014 | ||
2015 | rbd_dev->object_map_size = object_map_size; | |
2016 | ceph_copy_from_page_vector(pages, rbd_dev->object_map, | |
2017 | offset_in_page(p), object_map_bytes); | |
2018 | ||
2019 | out: | |
2020 | ceph_release_page_vector(pages, num_pages); | |
2021 | return ret; | |
2022 | } | |
3da691bf | 2023 | |
22e8bd51 ID |
2024 | static void rbd_object_map_free(struct rbd_device *rbd_dev) |
2025 | { | |
2026 | kvfree(rbd_dev->object_map); | |
2027 | rbd_dev->object_map = NULL; | |
2028 | rbd_dev->object_map_size = 0; | |
3b434a2a JD |
2029 | } |
2030 | ||
22e8bd51 | 2031 | static int rbd_object_map_load(struct rbd_device *rbd_dev) |
bf0d5f50 | 2032 | { |
3da691bf | 2033 | int ret; |
37206ee5 | 2034 | |
22e8bd51 | 2035 | ret = __rbd_object_map_load(rbd_dev); |
86bd7998 ID |
2036 | if (ret) |
2037 | return ret; | |
f1a4739f | 2038 | |
22e8bd51 ID |
2039 | ret = rbd_dev_v2_get_flags(rbd_dev); |
2040 | if (ret) { | |
2041 | rbd_object_map_free(rbd_dev); | |
2042 | return ret; | |
2043 | } | |
2044 | ||
2045 | if (rbd_dev->object_map_flags & RBD_FLAG_OBJECT_MAP_INVALID) | |
2046 | rbd_warn(rbd_dev, "object map is invalid"); | |
2047 | ||
2048 | return 0; | |
2049 | } | |
2050 | ||
2051 | static int rbd_object_map_open(struct rbd_device *rbd_dev) | |
2052 | { | |
2053 | int ret; | |
2054 | ||
2055 | ret = rbd_object_map_lock(rbd_dev); | |
2056 | if (ret) | |
2057 | return ret; | |
2058 | ||
2059 | ret = rbd_object_map_load(rbd_dev); | |
2060 | if (ret) { | |
2061 | rbd_object_map_unlock(rbd_dev); | |
2062 | return ret; | |
2063 | } | |
2064 | ||
2065 | return 0; | |
2066 | } | |
2067 | ||
2068 | static void rbd_object_map_close(struct rbd_device *rbd_dev) | |
2069 | { | |
2070 | rbd_object_map_free(rbd_dev); | |
2071 | rbd_object_map_unlock(rbd_dev); | |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * This function needs snap_id (or more precisely just something to | |
2076 | * distinguish between HEAD and snapshot object maps), new_state and | |
2077 | * current_state that were passed to rbd_object_map_update(). | |
2078 | * | |
2079 | * To avoid allocating and stashing a context we piggyback on the OSD | |
2080 | * request. A HEAD update has two ops (assert_locked). For new_state | |
2081 | * and current_state we decode our own object_map_update op, encoded in | |
2082 | * rbd_cls_object_map_update(). | |
2083 | */ | |
2084 | static int rbd_object_map_update_finish(struct rbd_obj_request *obj_req, | |
2085 | struct ceph_osd_request *osd_req) | |
2086 | { | |
2087 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
2088 | struct ceph_osd_data *osd_data; | |
2089 | u64 objno; | |
2090 | u8 state, new_state, current_state; | |
2091 | bool has_current_state; | |
2092 | void *p; | |
2093 | ||
2094 | if (osd_req->r_result) | |
2095 | return osd_req->r_result; | |
2096 | ||
2097 | /* | |
2098 | * Nothing to do for a snapshot object map. | |
2099 | */ | |
2100 | if (osd_req->r_num_ops == 1) | |
2101 | return 0; | |
2102 | ||
2103 | /* | |
2104 | * Update in-memory HEAD object map. | |
2105 | */ | |
2106 | rbd_assert(osd_req->r_num_ops == 2); | |
2107 | osd_data = osd_req_op_data(osd_req, 1, cls, request_data); | |
2108 | rbd_assert(osd_data->type == CEPH_OSD_DATA_TYPE_PAGES); | |
2109 | ||
2110 | p = page_address(osd_data->pages[0]); | |
2111 | objno = ceph_decode_64(&p); | |
2112 | rbd_assert(objno == obj_req->ex.oe_objno); | |
2113 | rbd_assert(ceph_decode_64(&p) == objno + 1); | |
2114 | new_state = ceph_decode_8(&p); | |
2115 | has_current_state = ceph_decode_8(&p); | |
2116 | if (has_current_state) | |
2117 | current_state = ceph_decode_8(&p); | |
2118 | ||
2119 | spin_lock(&rbd_dev->object_map_lock); | |
2120 | state = __rbd_object_map_get(rbd_dev, objno); | |
2121 | if (!has_current_state || current_state == state || | |
2122 | (current_state == OBJECT_EXISTS && state == OBJECT_EXISTS_CLEAN)) | |
2123 | __rbd_object_map_set(rbd_dev, objno, new_state); | |
2124 | spin_unlock(&rbd_dev->object_map_lock); | |
2125 | ||
2126 | return 0; | |
2127 | } | |
2128 | ||
2129 | static void rbd_object_map_callback(struct ceph_osd_request *osd_req) | |
2130 | { | |
2131 | struct rbd_obj_request *obj_req = osd_req->r_priv; | |
2132 | int result; | |
2133 | ||
2134 | dout("%s osd_req %p result %d for obj_req %p\n", __func__, osd_req, | |
2135 | osd_req->r_result, obj_req); | |
2136 | ||
2137 | result = rbd_object_map_update_finish(obj_req, osd_req); | |
2138 | rbd_obj_handle_request(obj_req, result); | |
2139 | } | |
2140 | ||
2141 | static bool update_needed(struct rbd_device *rbd_dev, u64 objno, u8 new_state) | |
2142 | { | |
2143 | u8 state = rbd_object_map_get(rbd_dev, objno); | |
bf0d5f50 | 2144 | |
22e8bd51 ID |
2145 | if (state == new_state || |
2146 | (new_state == OBJECT_PENDING && state == OBJECT_NONEXISTENT) || | |
2147 | (new_state == OBJECT_NONEXISTENT && state != OBJECT_PENDING)) | |
2148 | return false; | |
2149 | ||
2150 | return true; | |
2151 | } | |
2152 | ||
2153 | static int rbd_cls_object_map_update(struct ceph_osd_request *req, | |
2154 | int which, u64 objno, u8 new_state, | |
2155 | const u8 *current_state) | |
2156 | { | |
2157 | struct page **pages; | |
2158 | void *p, *start; | |
2159 | int ret; | |
2160 | ||
2161 | ret = osd_req_op_cls_init(req, which, "rbd", "object_map_update"); | |
2162 | if (ret) | |
2163 | return ret; | |
2164 | ||
2165 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
2166 | if (IS_ERR(pages)) | |
2167 | return PTR_ERR(pages); | |
2168 | ||
2169 | p = start = page_address(pages[0]); | |
2170 | ceph_encode_64(&p, objno); | |
2171 | ceph_encode_64(&p, objno + 1); | |
2172 | ceph_encode_8(&p, new_state); | |
2173 | if (current_state) { | |
2174 | ceph_encode_8(&p, 1); | |
2175 | ceph_encode_8(&p, *current_state); | |
2176 | } else { | |
2177 | ceph_encode_8(&p, 0); | |
2178 | } | |
2179 | ||
2180 | osd_req_op_cls_request_data_pages(req, which, pages, p - start, 0, | |
2181 | false, true); | |
2182 | return 0; | |
2183 | } | |
2184 | ||
2185 | /* | |
2186 | * Return: | |
2187 | * 0 - object map update sent | |
2188 | * 1 - object map update isn't needed | |
2189 | * <0 - error | |
2190 | */ | |
2191 | static int rbd_object_map_update(struct rbd_obj_request *obj_req, u64 snap_id, | |
2192 | u8 new_state, const u8 *current_state) | |
2193 | { | |
2194 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
2195 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2196 | struct ceph_osd_request *req; | |
2197 | int num_ops = 1; | |
2198 | int which = 0; | |
2199 | int ret; | |
2200 | ||
2201 | if (snap_id == CEPH_NOSNAP) { | |
2202 | if (!update_needed(rbd_dev, obj_req->ex.oe_objno, new_state)) | |
2203 | return 1; | |
2204 | ||
2205 | num_ops++; /* assert_locked */ | |
2206 | } | |
2207 | ||
2208 | req = ceph_osdc_alloc_request(osdc, NULL, num_ops, false, GFP_NOIO); | |
2209 | if (!req) | |
2210 | return -ENOMEM; | |
2211 | ||
2212 | list_add_tail(&req->r_private_item, &obj_req->osd_reqs); | |
2213 | req->r_callback = rbd_object_map_callback; | |
2214 | req->r_priv = obj_req; | |
2215 | ||
2216 | rbd_object_map_name(rbd_dev, snap_id, &req->r_base_oid); | |
2217 | ceph_oloc_copy(&req->r_base_oloc, &rbd_dev->header_oloc); | |
2218 | req->r_flags = CEPH_OSD_FLAG_WRITE; | |
2219 | ktime_get_real_ts64(&req->r_mtime); | |
2220 | ||
2221 | if (snap_id == CEPH_NOSNAP) { | |
2222 | /* | |
2223 | * Protect against possible race conditions during lock | |
2224 | * ownership transitions. | |
2225 | */ | |
2226 | ret = ceph_cls_assert_locked(req, which++, RBD_LOCK_NAME, | |
2227 | CEPH_CLS_LOCK_EXCLUSIVE, "", ""); | |
3da691bf ID |
2228 | if (ret) |
2229 | return ret; | |
22e8bd51 ID |
2230 | } |
2231 | ||
2232 | ret = rbd_cls_object_map_update(req, which, obj_req->ex.oe_objno, | |
2233 | new_state, current_state); | |
2234 | if (ret) | |
2235 | return ret; | |
2236 | ||
2237 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
2238 | if (ret) | |
2239 | return ret; | |
13488d53 | 2240 | |
22e8bd51 ID |
2241 | ceph_osdc_start_request(osdc, req, false); |
2242 | return 0; | |
2243 | } | |
2244 | ||
86bd7998 ID |
2245 | static void prune_extents(struct ceph_file_extent *img_extents, |
2246 | u32 *num_img_extents, u64 overlap) | |
e93f3152 | 2247 | { |
86bd7998 | 2248 | u32 cnt = *num_img_extents; |
e93f3152 | 2249 | |
86bd7998 ID |
2250 | /* drop extents completely beyond the overlap */ |
2251 | while (cnt && img_extents[cnt - 1].fe_off >= overlap) | |
2252 | cnt--; | |
e93f3152 | 2253 | |
86bd7998 ID |
2254 | if (cnt) { |
2255 | struct ceph_file_extent *ex = &img_extents[cnt - 1]; | |
e93f3152 | 2256 | |
86bd7998 ID |
2257 | /* trim final overlapping extent */ |
2258 | if (ex->fe_off + ex->fe_len > overlap) | |
2259 | ex->fe_len = overlap - ex->fe_off; | |
2260 | } | |
e93f3152 | 2261 | |
86bd7998 | 2262 | *num_img_extents = cnt; |
e93f3152 AE |
2263 | } |
2264 | ||
86bd7998 ID |
2265 | /* |
2266 | * Determine the byte range(s) covered by either just the object extent | |
2267 | * or the entire object in the parent image. | |
2268 | */ | |
2269 | static int rbd_obj_calc_img_extents(struct rbd_obj_request *obj_req, | |
2270 | bool entire) | |
e93f3152 | 2271 | { |
86bd7998 ID |
2272 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
2273 | int ret; | |
e93f3152 | 2274 | |
86bd7998 ID |
2275 | if (!rbd_dev->parent_overlap) |
2276 | return 0; | |
e93f3152 | 2277 | |
86bd7998 ID |
2278 | ret = ceph_extent_to_file(&rbd_dev->layout, obj_req->ex.oe_objno, |
2279 | entire ? 0 : obj_req->ex.oe_off, | |
2280 | entire ? rbd_dev->layout.object_size : | |
2281 | obj_req->ex.oe_len, | |
2282 | &obj_req->img_extents, | |
2283 | &obj_req->num_img_extents); | |
2284 | if (ret) | |
2285 | return ret; | |
e93f3152 | 2286 | |
86bd7998 ID |
2287 | prune_extents(obj_req->img_extents, &obj_req->num_img_extents, |
2288 | rbd_dev->parent_overlap); | |
2289 | return 0; | |
e93f3152 AE |
2290 | } |
2291 | ||
bcbab1db | 2292 | static void rbd_osd_setup_data(struct ceph_osd_request *osd_req, int which) |
1217857f | 2293 | { |
bcbab1db ID |
2294 | struct rbd_obj_request *obj_req = osd_req->r_priv; |
2295 | ||
ecc633ca | 2296 | switch (obj_req->img_request->data_type) { |
3da691bf | 2297 | case OBJ_REQUEST_BIO: |
bcbab1db | 2298 | osd_req_op_extent_osd_data_bio(osd_req, which, |
3da691bf | 2299 | &obj_req->bio_pos, |
43df3d35 | 2300 | obj_req->ex.oe_len); |
3da691bf ID |
2301 | break; |
2302 | case OBJ_REQUEST_BVECS: | |
afb97888 | 2303 | case OBJ_REQUEST_OWN_BVECS: |
3da691bf | 2304 | rbd_assert(obj_req->bvec_pos.iter.bi_size == |
43df3d35 | 2305 | obj_req->ex.oe_len); |
afb97888 | 2306 | rbd_assert(obj_req->bvec_idx == obj_req->bvec_count); |
bcbab1db | 2307 | osd_req_op_extent_osd_data_bvec_pos(osd_req, which, |
3da691bf ID |
2308 | &obj_req->bvec_pos); |
2309 | break; | |
2310 | default: | |
16809372 | 2311 | BUG(); |
1217857f | 2312 | } |
3da691bf | 2313 | } |
1217857f | 2314 | |
bcbab1db | 2315 | static int rbd_osd_setup_stat(struct ceph_osd_request *osd_req, int which) |
3da691bf ID |
2316 | { |
2317 | struct page **pages; | |
8b3e1a56 | 2318 | |
3da691bf ID |
2319 | /* |
2320 | * The response data for a STAT call consists of: | |
2321 | * le64 length; | |
2322 | * struct { | |
2323 | * le32 tv_sec; | |
2324 | * le32 tv_nsec; | |
2325 | * } mtime; | |
2326 | */ | |
2327 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
2328 | if (IS_ERR(pages)) | |
2329 | return PTR_ERR(pages); | |
2330 | ||
bcbab1db ID |
2331 | osd_req_op_init(osd_req, which, CEPH_OSD_OP_STAT, 0); |
2332 | osd_req_op_raw_data_in_pages(osd_req, which, pages, | |
3da691bf ID |
2333 | 8 + sizeof(struct ceph_timespec), |
2334 | 0, false, true); | |
2335 | return 0; | |
1217857f AE |
2336 | } |
2337 | ||
b5ae8cbc ID |
2338 | static int rbd_osd_setup_copyup(struct ceph_osd_request *osd_req, int which, |
2339 | u32 bytes) | |
2340 | { | |
2341 | struct rbd_obj_request *obj_req = osd_req->r_priv; | |
2342 | int ret; | |
2343 | ||
2344 | ret = osd_req_op_cls_init(osd_req, which, "rbd", "copyup"); | |
2345 | if (ret) | |
2346 | return ret; | |
2347 | ||
2348 | osd_req_op_cls_request_data_bvecs(osd_req, which, obj_req->copyup_bvecs, | |
2349 | obj_req->copyup_bvec_count, bytes); | |
2350 | return 0; | |
2351 | } | |
2352 | ||
ea9b743c ID |
2353 | static int rbd_obj_init_read(struct rbd_obj_request *obj_req) |
2354 | { | |
2355 | obj_req->read_state = RBD_OBJ_READ_START; | |
2356 | return 0; | |
2357 | } | |
2358 | ||
bcbab1db ID |
2359 | static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req, |
2360 | int which) | |
2169238d | 2361 | { |
bcbab1db | 2362 | struct rbd_obj_request *obj_req = osd_req->r_priv; |
3da691bf ID |
2363 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
2364 | u16 opcode; | |
2169238d | 2365 | |
8b5bec5c ID |
2366 | if (!use_object_map(rbd_dev) || |
2367 | !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) { | |
2368 | osd_req_op_alloc_hint_init(osd_req, which++, | |
2369 | rbd_dev->layout.object_size, | |
2370 | rbd_dev->layout.object_size); | |
2371 | } | |
2169238d | 2372 | |
3da691bf ID |
2373 | if (rbd_obj_is_entire(obj_req)) |
2374 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2375 | else | |
2376 | opcode = CEPH_OSD_OP_WRITE; | |
2169238d | 2377 | |
bcbab1db | 2378 | osd_req_op_extent_init(osd_req, which, opcode, |
43df3d35 | 2379 | obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0); |
bcbab1db | 2380 | rbd_osd_setup_data(osd_req, which); |
3da691bf | 2381 | } |
2169238d | 2382 | |
ea9b743c | 2383 | static int rbd_obj_init_write(struct rbd_obj_request *obj_req) |
3da691bf | 2384 | { |
3da691bf ID |
2385 | int ret; |
2386 | ||
86bd7998 ID |
2387 | /* reverse map the entire object onto the parent */ |
2388 | ret = rbd_obj_calc_img_extents(obj_req, true); | |
2389 | if (ret) | |
2390 | return ret; | |
2391 | ||
0ad5d953 ID |
2392 | if (rbd_obj_copyup_enabled(obj_req)) |
2393 | obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED; | |
2394 | ||
85b5e6d1 | 2395 | obj_req->write_state = RBD_OBJ_WRITE_START; |
3da691bf | 2396 | return 0; |
2169238d AE |
2397 | } |
2398 | ||
6484cbe9 ID |
2399 | static u16 truncate_or_zero_opcode(struct rbd_obj_request *obj_req) |
2400 | { | |
2401 | return rbd_obj_is_tail(obj_req) ? CEPH_OSD_OP_TRUNCATE : | |
2402 | CEPH_OSD_OP_ZERO; | |
2403 | } | |
2404 | ||
27bbd911 ID |
2405 | static void __rbd_osd_setup_discard_ops(struct ceph_osd_request *osd_req, |
2406 | int which) | |
2407 | { | |
2408 | struct rbd_obj_request *obj_req = osd_req->r_priv; | |
2409 | ||
2410 | if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) { | |
2411 | rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION); | |
2412 | osd_req_op_init(osd_req, which, CEPH_OSD_OP_DELETE, 0); | |
13488d53 | 2413 | } else { |
27bbd911 ID |
2414 | osd_req_op_extent_init(osd_req, which, |
2415 | truncate_or_zero_opcode(obj_req), | |
2416 | obj_req->ex.oe_off, obj_req->ex.oe_len, | |
2417 | 0, 0); | |
2418 | } | |
2419 | } | |
2420 | ||
ea9b743c | 2421 | static int rbd_obj_init_discard(struct rbd_obj_request *obj_req) |
6484cbe9 | 2422 | { |
0c93e1b7 | 2423 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
27bbd911 | 2424 | u64 off, next_off; |
6484cbe9 ID |
2425 | int ret; |
2426 | ||
0c93e1b7 ID |
2427 | /* |
2428 | * Align the range to alloc_size boundary and punt on discards | |
2429 | * that are too small to free up any space. | |
2430 | * | |
2431 | * alloc_size == object_size && is_tail() is a special case for | |
2432 | * filestore with filestore_punch_hole = false, needed to allow | |
2433 | * truncate (in addition to delete). | |
2434 | */ | |
2435 | if (rbd_dev->opts->alloc_size != rbd_dev->layout.object_size || | |
2436 | !rbd_obj_is_tail(obj_req)) { | |
27bbd911 ID |
2437 | off = round_up(obj_req->ex.oe_off, rbd_dev->opts->alloc_size); |
2438 | next_off = round_down(obj_req->ex.oe_off + obj_req->ex.oe_len, | |
2439 | rbd_dev->opts->alloc_size); | |
0c93e1b7 ID |
2440 | if (off >= next_off) |
2441 | return 1; | |
27bbd911 ID |
2442 | |
2443 | dout("%s %p %llu~%llu -> %llu~%llu\n", __func__, | |
2444 | obj_req, obj_req->ex.oe_off, obj_req->ex.oe_len, | |
2445 | off, next_off - off); | |
2446 | obj_req->ex.oe_off = off; | |
2447 | obj_req->ex.oe_len = next_off - off; | |
0c93e1b7 ID |
2448 | } |
2449 | ||
6484cbe9 ID |
2450 | /* reverse map the entire object onto the parent */ |
2451 | ret = rbd_obj_calc_img_extents(obj_req, true); | |
2452 | if (ret) | |
2453 | return ret; | |
2454 | ||
22e8bd51 | 2455 | obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT; |
0ad5d953 ID |
2456 | if (rbd_obj_is_entire(obj_req) && !obj_req->num_img_extents) |
2457 | obj_req->flags |= RBD_OBJ_FLAG_DELETION; | |
2458 | ||
85b5e6d1 | 2459 | obj_req->write_state = RBD_OBJ_WRITE_START; |
6484cbe9 ID |
2460 | return 0; |
2461 | } | |
2462 | ||
bcbab1db ID |
2463 | static void __rbd_osd_setup_zeroout_ops(struct ceph_osd_request *osd_req, |
2464 | int which) | |
3da691bf | 2465 | { |
bcbab1db | 2466 | struct rbd_obj_request *obj_req = osd_req->r_priv; |
3b434a2a JD |
2467 | u16 opcode; |
2468 | ||
3da691bf | 2469 | if (rbd_obj_is_entire(obj_req)) { |
86bd7998 | 2470 | if (obj_req->num_img_extents) { |
0ad5d953 | 2471 | if (!(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)) |
bcbab1db | 2472 | osd_req_op_init(osd_req, which++, |
9b17eb2c | 2473 | CEPH_OSD_OP_CREATE, 0); |
3b434a2a JD |
2474 | opcode = CEPH_OSD_OP_TRUNCATE; |
2475 | } else { | |
0ad5d953 | 2476 | rbd_assert(obj_req->flags & RBD_OBJ_FLAG_DELETION); |
bcbab1db | 2477 | osd_req_op_init(osd_req, which++, |
3da691bf ID |
2478 | CEPH_OSD_OP_DELETE, 0); |
2479 | opcode = 0; | |
3b434a2a | 2480 | } |
3b434a2a | 2481 | } else { |
6484cbe9 | 2482 | opcode = truncate_or_zero_opcode(obj_req); |
3b434a2a JD |
2483 | } |
2484 | ||
3da691bf | 2485 | if (opcode) |
bcbab1db | 2486 | osd_req_op_extent_init(osd_req, which, opcode, |
43df3d35 | 2487 | obj_req->ex.oe_off, obj_req->ex.oe_len, |
3da691bf | 2488 | 0, 0); |
3b434a2a JD |
2489 | } |
2490 | ||
ea9b743c | 2491 | static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req) |
bf0d5f50 | 2492 | { |
3da691bf | 2493 | int ret; |
37206ee5 | 2494 | |
86bd7998 ID |
2495 | /* reverse map the entire object onto the parent */ |
2496 | ret = rbd_obj_calc_img_extents(obj_req, true); | |
2497 | if (ret) | |
2498 | return ret; | |
f1a4739f | 2499 | |
0ad5d953 ID |
2500 | if (rbd_obj_copyup_enabled(obj_req)) |
2501 | obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED; | |
2502 | if (!obj_req->num_img_extents) { | |
22e8bd51 | 2503 | obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT; |
0ad5d953 ID |
2504 | if (rbd_obj_is_entire(obj_req)) |
2505 | obj_req->flags |= RBD_OBJ_FLAG_DELETION; | |
3da691bf | 2506 | } |
3b434a2a | 2507 | |
a086a1b8 | 2508 | obj_req->write_state = RBD_OBJ_WRITE_START; |
3da691bf ID |
2509 | return 0; |
2510 | } | |
9d4df01f | 2511 | |
a086a1b8 ID |
2512 | static int count_write_ops(struct rbd_obj_request *obj_req) |
2513 | { | |
8b5bec5c ID |
2514 | struct rbd_img_request *img_req = obj_req->img_request; |
2515 | ||
2516 | switch (img_req->op_type) { | |
a086a1b8 | 2517 | case OBJ_OP_WRITE: |
8b5bec5c ID |
2518 | if (!use_object_map(img_req->rbd_dev) || |
2519 | !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) | |
2520 | return 2; /* setallochint + write/writefull */ | |
2521 | ||
2522 | return 1; /* write/writefull */ | |
a086a1b8 ID |
2523 | case OBJ_OP_DISCARD: |
2524 | return 1; /* delete/truncate/zero */ | |
2525 | case OBJ_OP_ZEROOUT: | |
2526 | if (rbd_obj_is_entire(obj_req) && obj_req->num_img_extents && | |
2527 | !(obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED)) | |
2528 | return 2; /* create + truncate */ | |
bf0d5f50 | 2529 | |
a086a1b8 ID |
2530 | return 1; /* delete/truncate/zero */ |
2531 | default: | |
2532 | BUG(); | |
3da691bf | 2533 | } |
a086a1b8 | 2534 | } |
3b434a2a | 2535 | |
a086a1b8 ID |
2536 | static void rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req, |
2537 | int which) | |
2538 | { | |
2539 | struct rbd_obj_request *obj_req = osd_req->r_priv; | |
2540 | ||
2541 | switch (obj_req->img_request->op_type) { | |
2542 | case OBJ_OP_WRITE: | |
2543 | __rbd_osd_setup_write_ops(osd_req, which); | |
2544 | break; | |
2545 | case OBJ_OP_DISCARD: | |
2546 | __rbd_osd_setup_discard_ops(osd_req, which); | |
2547 | break; | |
2548 | case OBJ_OP_ZEROOUT: | |
2549 | __rbd_osd_setup_zeroout_ops(osd_req, which); | |
2550 | break; | |
2551 | default: | |
2552 | BUG(); | |
2553 | } | |
3da691bf | 2554 | } |
9d4df01f | 2555 | |
3da691bf | 2556 | /* |
a086a1b8 ID |
2557 | * Prune the list of object requests (adjust offset and/or length, drop |
2558 | * redundant requests). Prepare object request state machines and image | |
2559 | * request state machine for execution. | |
3da691bf ID |
2560 | */ |
2561 | static int __rbd_img_fill_request(struct rbd_img_request *img_req) | |
2562 | { | |
0c93e1b7 | 2563 | struct rbd_obj_request *obj_req, *next_obj_req; |
3da691bf | 2564 | int ret; |
430c28c3 | 2565 | |
0c93e1b7 | 2566 | for_each_obj_request_safe(img_req, obj_req, next_obj_req) { |
9bb0248d | 2567 | switch (img_req->op_type) { |
3da691bf | 2568 | case OBJ_OP_READ: |
ea9b743c | 2569 | ret = rbd_obj_init_read(obj_req); |
3da691bf ID |
2570 | break; |
2571 | case OBJ_OP_WRITE: | |
ea9b743c | 2572 | ret = rbd_obj_init_write(obj_req); |
3da691bf ID |
2573 | break; |
2574 | case OBJ_OP_DISCARD: | |
ea9b743c | 2575 | ret = rbd_obj_init_discard(obj_req); |
3da691bf | 2576 | break; |
6484cbe9 | 2577 | case OBJ_OP_ZEROOUT: |
ea9b743c | 2578 | ret = rbd_obj_init_zeroout(obj_req); |
6484cbe9 | 2579 | break; |
3da691bf | 2580 | default: |
16809372 | 2581 | BUG(); |
3da691bf | 2582 | } |
0c93e1b7 | 2583 | if (ret < 0) |
3da691bf | 2584 | return ret; |
0c93e1b7 | 2585 | if (ret > 0) { |
0c93e1b7 ID |
2586 | rbd_img_obj_request_del(img_req, obj_req); |
2587 | continue; | |
2588 | } | |
bf0d5f50 AE |
2589 | } |
2590 | ||
0192ce2e | 2591 | img_req->state = RBD_IMG_START; |
bf0d5f50 | 2592 | return 0; |
3da691bf | 2593 | } |
bf0d5f50 | 2594 | |
5a237819 ID |
2595 | union rbd_img_fill_iter { |
2596 | struct ceph_bio_iter bio_iter; | |
2597 | struct ceph_bvec_iter bvec_iter; | |
2598 | }; | |
bf0d5f50 | 2599 | |
5a237819 ID |
2600 | struct rbd_img_fill_ctx { |
2601 | enum obj_request_type pos_type; | |
2602 | union rbd_img_fill_iter *pos; | |
2603 | union rbd_img_fill_iter iter; | |
2604 | ceph_object_extent_fn_t set_pos_fn; | |
afb97888 ID |
2605 | ceph_object_extent_fn_t count_fn; |
2606 | ceph_object_extent_fn_t copy_fn; | |
5a237819 | 2607 | }; |
bf0d5f50 | 2608 | |
5a237819 | 2609 | static struct ceph_object_extent *alloc_object_extent(void *arg) |
0eefd470 | 2610 | { |
5a237819 ID |
2611 | struct rbd_img_request *img_req = arg; |
2612 | struct rbd_obj_request *obj_req; | |
0eefd470 | 2613 | |
5a237819 ID |
2614 | obj_req = rbd_obj_request_create(); |
2615 | if (!obj_req) | |
2616 | return NULL; | |
2761713d | 2617 | |
5a237819 ID |
2618 | rbd_img_obj_request_add(img_req, obj_req); |
2619 | return &obj_req->ex; | |
2620 | } | |
0eefd470 | 2621 | |
afb97888 ID |
2622 | /* |
2623 | * While su != os && sc == 1 is technically not fancy (it's the same | |
2624 | * layout as su == os && sc == 1), we can't use the nocopy path for it | |
2625 | * because ->set_pos_fn() should be called only once per object. | |
2626 | * ceph_file_to_extents() invokes action_fn once per stripe unit, so | |
2627 | * treat su != os && sc == 1 as fancy. | |
2628 | */ | |
2629 | static bool rbd_layout_is_fancy(struct ceph_file_layout *l) | |
2630 | { | |
2631 | return l->stripe_unit != l->object_size; | |
2632 | } | |
0eefd470 | 2633 | |
afb97888 ID |
2634 | static int rbd_img_fill_request_nocopy(struct rbd_img_request *img_req, |
2635 | struct ceph_file_extent *img_extents, | |
2636 | u32 num_img_extents, | |
2637 | struct rbd_img_fill_ctx *fctx) | |
2638 | { | |
2639 | u32 i; | |
2640 | int ret; | |
2641 | ||
2642 | img_req->data_type = fctx->pos_type; | |
0eefd470 AE |
2643 | |
2644 | /* | |
afb97888 ID |
2645 | * Create object requests and set each object request's starting |
2646 | * position in the provided bio (list) or bio_vec array. | |
0eefd470 | 2647 | */ |
afb97888 ID |
2648 | fctx->iter = *fctx->pos; |
2649 | for (i = 0; i < num_img_extents; i++) { | |
2650 | ret = ceph_file_to_extents(&img_req->rbd_dev->layout, | |
2651 | img_extents[i].fe_off, | |
2652 | img_extents[i].fe_len, | |
2653 | &img_req->object_extents, | |
2654 | alloc_object_extent, img_req, | |
2655 | fctx->set_pos_fn, &fctx->iter); | |
2656 | if (ret) | |
2657 | return ret; | |
2658 | } | |
0eefd470 | 2659 | |
afb97888 | 2660 | return __rbd_img_fill_request(img_req); |
0eefd470 AE |
2661 | } |
2662 | ||
5a237819 ID |
2663 | /* |
2664 | * Map a list of image extents to a list of object extents, create the | |
2665 | * corresponding object requests (normally each to a different object, | |
2666 | * but not always) and add them to @img_req. For each object request, | |
afb97888 | 2667 | * set up its data descriptor to point to the corresponding chunk(s) of |
5a237819 ID |
2668 | * @fctx->pos data buffer. |
2669 | * | |
afb97888 ID |
2670 | * Because ceph_file_to_extents() will merge adjacent object extents |
2671 | * together, each object request's data descriptor may point to multiple | |
2672 | * different chunks of @fctx->pos data buffer. | |
2673 | * | |
5a237819 ID |
2674 | * @fctx->pos data buffer is assumed to be large enough. |
2675 | */ | |
2676 | static int rbd_img_fill_request(struct rbd_img_request *img_req, | |
2677 | struct ceph_file_extent *img_extents, | |
2678 | u32 num_img_extents, | |
2679 | struct rbd_img_fill_ctx *fctx) | |
3d7efd18 | 2680 | { |
afb97888 ID |
2681 | struct rbd_device *rbd_dev = img_req->rbd_dev; |
2682 | struct rbd_obj_request *obj_req; | |
5a237819 ID |
2683 | u32 i; |
2684 | int ret; | |
2685 | ||
afb97888 ID |
2686 | if (fctx->pos_type == OBJ_REQUEST_NODATA || |
2687 | !rbd_layout_is_fancy(&rbd_dev->layout)) | |
2688 | return rbd_img_fill_request_nocopy(img_req, img_extents, | |
2689 | num_img_extents, fctx); | |
3d7efd18 | 2690 | |
afb97888 | 2691 | img_req->data_type = OBJ_REQUEST_OWN_BVECS; |
0eefd470 | 2692 | |
bbea1c1a | 2693 | /* |
afb97888 ID |
2694 | * Create object requests and determine ->bvec_count for each object |
2695 | * request. Note that ->bvec_count sum over all object requests may | |
2696 | * be greater than the number of bio_vecs in the provided bio (list) | |
2697 | * or bio_vec array because when mapped, those bio_vecs can straddle | |
2698 | * stripe unit boundaries. | |
bbea1c1a | 2699 | */ |
5a237819 ID |
2700 | fctx->iter = *fctx->pos; |
2701 | for (i = 0; i < num_img_extents; i++) { | |
afb97888 | 2702 | ret = ceph_file_to_extents(&rbd_dev->layout, |
5a237819 ID |
2703 | img_extents[i].fe_off, |
2704 | img_extents[i].fe_len, | |
2705 | &img_req->object_extents, | |
2706 | alloc_object_extent, img_req, | |
afb97888 ID |
2707 | fctx->count_fn, &fctx->iter); |
2708 | if (ret) | |
2709 | return ret; | |
bbea1c1a | 2710 | } |
0eefd470 | 2711 | |
afb97888 ID |
2712 | for_each_obj_request(img_req, obj_req) { |
2713 | obj_req->bvec_pos.bvecs = kmalloc_array(obj_req->bvec_count, | |
2714 | sizeof(*obj_req->bvec_pos.bvecs), | |
2715 | GFP_NOIO); | |
2716 | if (!obj_req->bvec_pos.bvecs) | |
2717 | return -ENOMEM; | |
2718 | } | |
0eefd470 | 2719 | |
8785b1d4 | 2720 | /* |
afb97888 ID |
2721 | * Fill in each object request's private bio_vec array, splitting and |
2722 | * rearranging the provided bio_vecs in stripe unit chunks as needed. | |
8785b1d4 | 2723 | */ |
afb97888 ID |
2724 | fctx->iter = *fctx->pos; |
2725 | for (i = 0; i < num_img_extents; i++) { | |
2726 | ret = ceph_iterate_extents(&rbd_dev->layout, | |
2727 | img_extents[i].fe_off, | |
2728 | img_extents[i].fe_len, | |
2729 | &img_req->object_extents, | |
2730 | fctx->copy_fn, &fctx->iter); | |
5a237819 ID |
2731 | if (ret) |
2732 | return ret; | |
2733 | } | |
3d7efd18 | 2734 | |
5a237819 ID |
2735 | return __rbd_img_fill_request(img_req); |
2736 | } | |
2737 | ||
2738 | static int rbd_img_fill_nodata(struct rbd_img_request *img_req, | |
2739 | u64 off, u64 len) | |
2740 | { | |
2741 | struct ceph_file_extent ex = { off, len }; | |
2742 | union rbd_img_fill_iter dummy; | |
2743 | struct rbd_img_fill_ctx fctx = { | |
2744 | .pos_type = OBJ_REQUEST_NODATA, | |
2745 | .pos = &dummy, | |
2746 | }; | |
2747 | ||
2748 | return rbd_img_fill_request(img_req, &ex, 1, &fctx); | |
2749 | } | |
2750 | ||
2751 | static void set_bio_pos(struct ceph_object_extent *ex, u32 bytes, void *arg) | |
2752 | { | |
2753 | struct rbd_obj_request *obj_req = | |
2754 | container_of(ex, struct rbd_obj_request, ex); | |
2755 | struct ceph_bio_iter *it = arg; | |
3d7efd18 | 2756 | |
5a237819 ID |
2757 | dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
2758 | obj_req->bio_pos = *it; | |
2759 | ceph_bio_iter_advance(it, bytes); | |
2760 | } | |
3d7efd18 | 2761 | |
afb97888 ID |
2762 | static void count_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
2763 | { | |
2764 | struct rbd_obj_request *obj_req = | |
2765 | container_of(ex, struct rbd_obj_request, ex); | |
2766 | struct ceph_bio_iter *it = arg; | |
0eefd470 | 2767 | |
afb97888 ID |
2768 | dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
2769 | ceph_bio_iter_advance_step(it, bytes, ({ | |
2770 | obj_req->bvec_count++; | |
2771 | })); | |
0eefd470 | 2772 | |
afb97888 | 2773 | } |
0eefd470 | 2774 | |
afb97888 ID |
2775 | static void copy_bio_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
2776 | { | |
2777 | struct rbd_obj_request *obj_req = | |
2778 | container_of(ex, struct rbd_obj_request, ex); | |
2779 | struct ceph_bio_iter *it = arg; | |
0eefd470 | 2780 | |
afb97888 ID |
2781 | dout("%s objno %llu bytes %u\n", __func__, ex->oe_objno, bytes); |
2782 | ceph_bio_iter_advance_step(it, bytes, ({ | |
2783 | obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv; | |
2784 | obj_req->bvec_pos.iter.bi_size += bv.bv_len; | |
2785 | })); | |
3d7efd18 AE |
2786 | } |
2787 | ||
5a237819 ID |
2788 | static int __rbd_img_fill_from_bio(struct rbd_img_request *img_req, |
2789 | struct ceph_file_extent *img_extents, | |
2790 | u32 num_img_extents, | |
2791 | struct ceph_bio_iter *bio_pos) | |
2792 | { | |
2793 | struct rbd_img_fill_ctx fctx = { | |
2794 | .pos_type = OBJ_REQUEST_BIO, | |
2795 | .pos = (union rbd_img_fill_iter *)bio_pos, | |
2796 | .set_pos_fn = set_bio_pos, | |
afb97888 ID |
2797 | .count_fn = count_bio_bvecs, |
2798 | .copy_fn = copy_bio_bvecs, | |
5a237819 | 2799 | }; |
3d7efd18 | 2800 | |
5a237819 ID |
2801 | return rbd_img_fill_request(img_req, img_extents, num_img_extents, |
2802 | &fctx); | |
2803 | } | |
3d7efd18 | 2804 | |
5a237819 ID |
2805 | static int rbd_img_fill_from_bio(struct rbd_img_request *img_req, |
2806 | u64 off, u64 len, struct bio *bio) | |
2807 | { | |
2808 | struct ceph_file_extent ex = { off, len }; | |
2809 | struct ceph_bio_iter it = { .bio = bio, .iter = bio->bi_iter }; | |
3d7efd18 | 2810 | |
5a237819 ID |
2811 | return __rbd_img_fill_from_bio(img_req, &ex, 1, &it); |
2812 | } | |
a9e8ba2c | 2813 | |
5a237819 ID |
2814 | static void set_bvec_pos(struct ceph_object_extent *ex, u32 bytes, void *arg) |
2815 | { | |
2816 | struct rbd_obj_request *obj_req = | |
2817 | container_of(ex, struct rbd_obj_request, ex); | |
2818 | struct ceph_bvec_iter *it = arg; | |
3d7efd18 | 2819 | |
5a237819 ID |
2820 | obj_req->bvec_pos = *it; |
2821 | ceph_bvec_iter_shorten(&obj_req->bvec_pos, bytes); | |
2822 | ceph_bvec_iter_advance(it, bytes); | |
2823 | } | |
3d7efd18 | 2824 | |
afb97888 ID |
2825 | static void count_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
2826 | { | |
2827 | struct rbd_obj_request *obj_req = | |
2828 | container_of(ex, struct rbd_obj_request, ex); | |
2829 | struct ceph_bvec_iter *it = arg; | |
058aa991 | 2830 | |
afb97888 ID |
2831 | ceph_bvec_iter_advance_step(it, bytes, ({ |
2832 | obj_req->bvec_count++; | |
2833 | })); | |
2834 | } | |
058aa991 | 2835 | |
afb97888 ID |
2836 | static void copy_bvecs(struct ceph_object_extent *ex, u32 bytes, void *arg) |
2837 | { | |
2838 | struct rbd_obj_request *obj_req = | |
2839 | container_of(ex, struct rbd_obj_request, ex); | |
2840 | struct ceph_bvec_iter *it = arg; | |
3d7efd18 | 2841 | |
afb97888 ID |
2842 | ceph_bvec_iter_advance_step(it, bytes, ({ |
2843 | obj_req->bvec_pos.bvecs[obj_req->bvec_idx++] = bv; | |
2844 | obj_req->bvec_pos.iter.bi_size += bv.bv_len; | |
2845 | })); | |
3d7efd18 AE |
2846 | } |
2847 | ||
5a237819 ID |
2848 | static int __rbd_img_fill_from_bvecs(struct rbd_img_request *img_req, |
2849 | struct ceph_file_extent *img_extents, | |
2850 | u32 num_img_extents, | |
2851 | struct ceph_bvec_iter *bvec_pos) | |
c5b5ef6c | 2852 | { |
5a237819 ID |
2853 | struct rbd_img_fill_ctx fctx = { |
2854 | .pos_type = OBJ_REQUEST_BVECS, | |
2855 | .pos = (union rbd_img_fill_iter *)bvec_pos, | |
2856 | .set_pos_fn = set_bvec_pos, | |
afb97888 ID |
2857 | .count_fn = count_bvecs, |
2858 | .copy_fn = copy_bvecs, | |
5a237819 | 2859 | }; |
c5b5ef6c | 2860 | |
5a237819 ID |
2861 | return rbd_img_fill_request(img_req, img_extents, num_img_extents, |
2862 | &fctx); | |
2863 | } | |
c5b5ef6c | 2864 | |
5a237819 ID |
2865 | static int rbd_img_fill_from_bvecs(struct rbd_img_request *img_req, |
2866 | struct ceph_file_extent *img_extents, | |
2867 | u32 num_img_extents, | |
2868 | struct bio_vec *bvecs) | |
2869 | { | |
2870 | struct ceph_bvec_iter it = { | |
2871 | .bvecs = bvecs, | |
2872 | .iter = { .bi_size = ceph_file_extents_bytes(img_extents, | |
2873 | num_img_extents) }, | |
2874 | }; | |
c5b5ef6c | 2875 | |
5a237819 ID |
2876 | return __rbd_img_fill_from_bvecs(img_req, img_extents, num_img_extents, |
2877 | &it); | |
2878 | } | |
c5b5ef6c | 2879 | |
0192ce2e | 2880 | static void rbd_img_handle_request_work(struct work_struct *work) |
bf0d5f50 | 2881 | { |
0192ce2e ID |
2882 | struct rbd_img_request *img_req = |
2883 | container_of(work, struct rbd_img_request, work); | |
c5b5ef6c | 2884 | |
0192ce2e ID |
2885 | rbd_img_handle_request(img_req, img_req->work_result); |
2886 | } | |
c2e82414 | 2887 | |
0192ce2e ID |
2888 | static void rbd_img_schedule(struct rbd_img_request *img_req, int result) |
2889 | { | |
2890 | INIT_WORK(&img_req->work, rbd_img_handle_request_work); | |
2891 | img_req->work_result = result; | |
2892 | queue_work(rbd_wq, &img_req->work); | |
c5b5ef6c | 2893 | } |
c2e82414 | 2894 | |
22e8bd51 ID |
2895 | static bool rbd_obj_may_exist(struct rbd_obj_request *obj_req) |
2896 | { | |
2897 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
2898 | ||
2899 | if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) { | |
2900 | obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST; | |
2901 | return true; | |
2902 | } | |
2903 | ||
2904 | dout("%s %p objno %llu assuming dne\n", __func__, obj_req, | |
2905 | obj_req->ex.oe_objno); | |
2906 | return false; | |
2907 | } | |
2908 | ||
85b5e6d1 ID |
2909 | static int rbd_obj_read_object(struct rbd_obj_request *obj_req) |
2910 | { | |
a086a1b8 ID |
2911 | struct ceph_osd_request *osd_req; |
2912 | int ret; | |
2913 | ||
2914 | osd_req = __rbd_obj_add_osd_request(obj_req, NULL, 1); | |
2915 | if (IS_ERR(osd_req)) | |
2916 | return PTR_ERR(osd_req); | |
2917 | ||
2918 | osd_req_op_extent_init(osd_req, 0, CEPH_OSD_OP_READ, | |
2919 | obj_req->ex.oe_off, obj_req->ex.oe_len, 0, 0); | |
2920 | rbd_osd_setup_data(osd_req, 0); | |
2921 | rbd_osd_format_read(osd_req); | |
2922 | ||
2923 | ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); | |
2924 | if (ret) | |
2925 | return ret; | |
2926 | ||
2927 | rbd_osd_submit(osd_req); | |
85b5e6d1 | 2928 | return 0; |
c5b5ef6c AE |
2929 | } |
2930 | ||
86bd7998 | 2931 | static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req) |
c5b5ef6c | 2932 | { |
3da691bf ID |
2933 | struct rbd_img_request *img_req = obj_req->img_request; |
2934 | struct rbd_img_request *child_img_req; | |
c5b5ef6c AE |
2935 | int ret; |
2936 | ||
e93aca0a ID |
2937 | child_img_req = rbd_img_request_create(img_req->rbd_dev->parent, |
2938 | OBJ_OP_READ, NULL); | |
3da691bf | 2939 | if (!child_img_req) |
710214e3 ID |
2940 | return -ENOMEM; |
2941 | ||
e93aca0a ID |
2942 | __set_bit(IMG_REQ_CHILD, &child_img_req->flags); |
2943 | child_img_req->obj_request = obj_req; | |
a90bb0c1 | 2944 | |
21ed05a8 ID |
2945 | dout("%s child_img_req %p for obj_req %p\n", __func__, child_img_req, |
2946 | obj_req); | |
2947 | ||
3da691bf | 2948 | if (!rbd_img_is_write(img_req)) { |
ecc633ca | 2949 | switch (img_req->data_type) { |
3da691bf | 2950 | case OBJ_REQUEST_BIO: |
5a237819 ID |
2951 | ret = __rbd_img_fill_from_bio(child_img_req, |
2952 | obj_req->img_extents, | |
2953 | obj_req->num_img_extents, | |
2954 | &obj_req->bio_pos); | |
3da691bf ID |
2955 | break; |
2956 | case OBJ_REQUEST_BVECS: | |
afb97888 | 2957 | case OBJ_REQUEST_OWN_BVECS: |
5a237819 ID |
2958 | ret = __rbd_img_fill_from_bvecs(child_img_req, |
2959 | obj_req->img_extents, | |
2960 | obj_req->num_img_extents, | |
2961 | &obj_req->bvec_pos); | |
3da691bf ID |
2962 | break; |
2963 | default: | |
d342a15b | 2964 | BUG(); |
3da691bf ID |
2965 | } |
2966 | } else { | |
5a237819 ID |
2967 | ret = rbd_img_fill_from_bvecs(child_img_req, |
2968 | obj_req->img_extents, | |
2969 | obj_req->num_img_extents, | |
2970 | obj_req->copyup_bvecs); | |
3da691bf ID |
2971 | } |
2972 | if (ret) { | |
2973 | rbd_img_request_put(child_img_req); | |
2974 | return ret; | |
2975 | } | |
2976 | ||
0192ce2e ID |
2977 | /* avoid parent chain recursion */ |
2978 | rbd_img_schedule(child_img_req, 0); | |
3da691bf ID |
2979 | return 0; |
2980 | } | |
2981 | ||
85b5e6d1 | 2982 | static bool rbd_obj_advance_read(struct rbd_obj_request *obj_req, int *result) |
3da691bf ID |
2983 | { |
2984 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
2985 | int ret; | |
2986 | ||
22e8bd51 | 2987 | again: |
a9b67e69 | 2988 | switch (obj_req->read_state) { |
85b5e6d1 ID |
2989 | case RBD_OBJ_READ_START: |
2990 | rbd_assert(!*result); | |
2991 | ||
22e8bd51 ID |
2992 | if (!rbd_obj_may_exist(obj_req)) { |
2993 | *result = -ENOENT; | |
2994 | obj_req->read_state = RBD_OBJ_READ_OBJECT; | |
2995 | goto again; | |
2996 | } | |
2997 | ||
85b5e6d1 | 2998 | ret = rbd_obj_read_object(obj_req); |
3da691bf | 2999 | if (ret) { |
85b5e6d1 | 3000 | *result = ret; |
3da691bf ID |
3001 | return true; |
3002 | } | |
85b5e6d1 ID |
3003 | obj_req->read_state = RBD_OBJ_READ_OBJECT; |
3004 | return false; | |
a9b67e69 ID |
3005 | case RBD_OBJ_READ_OBJECT: |
3006 | if (*result == -ENOENT && rbd_dev->parent_overlap) { | |
3007 | /* reverse map this object extent onto the parent */ | |
3008 | ret = rbd_obj_calc_img_extents(obj_req, false); | |
86bd7998 | 3009 | if (ret) { |
54ab3b24 | 3010 | *result = ret; |
86bd7998 ID |
3011 | return true; |
3012 | } | |
a9b67e69 ID |
3013 | if (obj_req->num_img_extents) { |
3014 | ret = rbd_obj_read_from_parent(obj_req); | |
3015 | if (ret) { | |
3016 | *result = ret; | |
3017 | return true; | |
3018 | } | |
3019 | obj_req->read_state = RBD_OBJ_READ_PARENT; | |
3020 | return false; | |
3021 | } | |
86bd7998 | 3022 | } |
710214e3 | 3023 | |
a9b67e69 ID |
3024 | /* |
3025 | * -ENOENT means a hole in the image -- zero-fill the entire | |
3026 | * length of the request. A short read also implies zero-fill | |
3027 | * to the end of the request. | |
3028 | */ | |
3029 | if (*result == -ENOENT) { | |
3030 | rbd_obj_zero_range(obj_req, 0, obj_req->ex.oe_len); | |
3031 | *result = 0; | |
3032 | } else if (*result >= 0) { | |
3033 | if (*result < obj_req->ex.oe_len) | |
3034 | rbd_obj_zero_range(obj_req, *result, | |
3035 | obj_req->ex.oe_len - *result); | |
3036 | else | |
3037 | rbd_assert(*result == obj_req->ex.oe_len); | |
3038 | *result = 0; | |
3039 | } | |
3040 | return true; | |
3041 | case RBD_OBJ_READ_PARENT: | |
d435c9a7 ID |
3042 | /* |
3043 | * The parent image is read only up to the overlap -- zero-fill | |
3044 | * from the overlap to the end of the request. | |
3045 | */ | |
3046 | if (!*result) { | |
3047 | u32 obj_overlap = rbd_obj_img_extents_bytes(obj_req); | |
3048 | ||
3049 | if (obj_overlap < obj_req->ex.oe_len) | |
3050 | rbd_obj_zero_range(obj_req, obj_overlap, | |
3051 | obj_req->ex.oe_len - obj_overlap); | |
3052 | } | |
a9b67e69 ID |
3053 | return true; |
3054 | default: | |
3055 | BUG(); | |
710214e3 | 3056 | } |
3da691bf | 3057 | } |
c5b5ef6c | 3058 | |
22e8bd51 ID |
3059 | static bool rbd_obj_write_is_noop(struct rbd_obj_request *obj_req) |
3060 | { | |
3061 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
3062 | ||
3063 | if (rbd_object_map_may_exist(rbd_dev, obj_req->ex.oe_objno)) | |
3064 | obj_req->flags |= RBD_OBJ_FLAG_MAY_EXIST; | |
3065 | ||
3066 | if (!(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST) && | |
3067 | (obj_req->flags & RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT)) { | |
3068 | dout("%s %p noop for nonexistent\n", __func__, obj_req); | |
3069 | return true; | |
3070 | } | |
3071 | ||
3072 | return false; | |
3073 | } | |
3074 | ||
3075 | /* | |
3076 | * Return: | |
3077 | * 0 - object map update sent | |
3078 | * 1 - object map update isn't needed | |
3079 | * <0 - error | |
3080 | */ | |
3081 | static int rbd_obj_write_pre_object_map(struct rbd_obj_request *obj_req) | |
3082 | { | |
3083 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
3084 | u8 new_state; | |
3085 | ||
3086 | if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) | |
3087 | return 1; | |
3088 | ||
3089 | if (obj_req->flags & RBD_OBJ_FLAG_DELETION) | |
3090 | new_state = OBJECT_PENDING; | |
3091 | else | |
3092 | new_state = OBJECT_EXISTS; | |
3093 | ||
3094 | return rbd_object_map_update(obj_req, CEPH_NOSNAP, new_state, NULL); | |
3095 | } | |
3096 | ||
85b5e6d1 ID |
3097 | static int rbd_obj_write_object(struct rbd_obj_request *obj_req) |
3098 | { | |
a086a1b8 ID |
3099 | struct ceph_osd_request *osd_req; |
3100 | int num_ops = count_write_ops(obj_req); | |
3101 | int which = 0; | |
3102 | int ret; | |
710214e3 | 3103 | |
a086a1b8 ID |
3104 | if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) |
3105 | num_ops++; /* stat */ | |
3106 | ||
3107 | osd_req = rbd_obj_add_osd_request(obj_req, num_ops); | |
3108 | if (IS_ERR(osd_req)) | |
3109 | return PTR_ERR(osd_req); | |
3110 | ||
3111 | if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) { | |
3112 | ret = rbd_osd_setup_stat(osd_req, which++); | |
3113 | if (ret) | |
3114 | return ret; | |
710214e3 | 3115 | } |
c5b5ef6c | 3116 | |
a086a1b8 ID |
3117 | rbd_osd_setup_write_ops(osd_req, which); |
3118 | rbd_osd_format_write(osd_req); | |
3119 | ||
3120 | ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); | |
3121 | if (ret) | |
3122 | return ret; | |
3123 | ||
3124 | rbd_osd_submit(osd_req); | |
85b5e6d1 | 3125 | return 0; |
3da691bf | 3126 | } |
c5b5ef6c | 3127 | |
3da691bf ID |
3128 | /* |
3129 | * copyup_bvecs pages are never highmem pages | |
3130 | */ | |
3131 | static bool is_zero_bvecs(struct bio_vec *bvecs, u32 bytes) | |
3132 | { | |
3133 | struct ceph_bvec_iter it = { | |
3134 | .bvecs = bvecs, | |
3135 | .iter = { .bi_size = bytes }, | |
3136 | }; | |
c5b5ef6c | 3137 | |
3da691bf ID |
3138 | ceph_bvec_iter_advance_step(&it, bytes, ({ |
3139 | if (memchr_inv(page_address(bv.bv_page) + bv.bv_offset, 0, | |
3140 | bv.bv_len)) | |
3141 | return false; | |
3142 | })); | |
3143 | return true; | |
c5b5ef6c AE |
3144 | } |
3145 | ||
3a482501 ID |
3146 | #define MODS_ONLY U32_MAX |
3147 | ||
793333a3 ID |
3148 | static int rbd_obj_copyup_empty_snapc(struct rbd_obj_request *obj_req, |
3149 | u32 bytes) | |
b454e36d | 3150 | { |
bcbab1db | 3151 | struct ceph_osd_request *osd_req; |
fe943d50 | 3152 | int ret; |
70d045f6 | 3153 | |
3da691bf | 3154 | dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes); |
89a59c1c | 3155 | rbd_assert(bytes > 0 && bytes != MODS_ONLY); |
70d045f6 | 3156 | |
bcbab1db ID |
3157 | osd_req = __rbd_obj_add_osd_request(obj_req, &rbd_empty_snapc, 1); |
3158 | if (IS_ERR(osd_req)) | |
3159 | return PTR_ERR(osd_req); | |
b454e36d | 3160 | |
b5ae8cbc | 3161 | ret = rbd_osd_setup_copyup(osd_req, 0, bytes); |
fe943d50 CX |
3162 | if (ret) |
3163 | return ret; | |
3164 | ||
bcbab1db | 3165 | rbd_osd_format_write(osd_req); |
3da691bf | 3166 | |
bcbab1db | 3167 | ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
89a59c1c ID |
3168 | if (ret) |
3169 | return ret; | |
3170 | ||
a086a1b8 | 3171 | rbd_osd_submit(osd_req); |
89a59c1c ID |
3172 | return 0; |
3173 | } | |
3174 | ||
793333a3 ID |
3175 | static int rbd_obj_copyup_current_snapc(struct rbd_obj_request *obj_req, |
3176 | u32 bytes) | |
b454e36d | 3177 | { |
bcbab1db | 3178 | struct ceph_osd_request *osd_req; |
a086a1b8 ID |
3179 | int num_ops = count_write_ops(obj_req); |
3180 | int which = 0; | |
fe943d50 | 3181 | int ret; |
70d045f6 | 3182 | |
3da691bf | 3183 | dout("%s obj_req %p bytes %u\n", __func__, obj_req, bytes); |
70d045f6 | 3184 | |
a086a1b8 ID |
3185 | if (bytes != MODS_ONLY) |
3186 | num_ops++; /* copyup */ | |
13488d53 | 3187 | |
a086a1b8 | 3188 | osd_req = rbd_obj_add_osd_request(obj_req, num_ops); |
bcbab1db ID |
3189 | if (IS_ERR(osd_req)) |
3190 | return PTR_ERR(osd_req); | |
b454e36d | 3191 | |
3a482501 | 3192 | if (bytes != MODS_ONLY) { |
b5ae8cbc | 3193 | ret = rbd_osd_setup_copyup(osd_req, which++, bytes); |
3a482501 ID |
3194 | if (ret) |
3195 | return ret; | |
3da691bf | 3196 | } |
3da691bf | 3197 | |
a086a1b8 ID |
3198 | rbd_osd_setup_write_ops(osd_req, which); |
3199 | rbd_osd_format_write(osd_req); | |
70d045f6 | 3200 | |
bcbab1db | 3201 | ret = ceph_osdc_alloc_messages(osd_req, GFP_NOIO); |
26f887e0 ID |
3202 | if (ret) |
3203 | return ret; | |
3204 | ||
a086a1b8 | 3205 | rbd_osd_submit(osd_req); |
3da691bf | 3206 | return 0; |
70d045f6 ID |
3207 | } |
3208 | ||
7e07efb1 | 3209 | static int setup_copyup_bvecs(struct rbd_obj_request *obj_req, u64 obj_overlap) |
70d045f6 | 3210 | { |
7e07efb1 | 3211 | u32 i; |
b454e36d | 3212 | |
7e07efb1 ID |
3213 | rbd_assert(!obj_req->copyup_bvecs); |
3214 | obj_req->copyup_bvec_count = calc_pages_for(0, obj_overlap); | |
3215 | obj_req->copyup_bvecs = kcalloc(obj_req->copyup_bvec_count, | |
3216 | sizeof(*obj_req->copyup_bvecs), | |
3217 | GFP_NOIO); | |
3218 | if (!obj_req->copyup_bvecs) | |
3219 | return -ENOMEM; | |
b454e36d | 3220 | |
7e07efb1 ID |
3221 | for (i = 0; i < obj_req->copyup_bvec_count; i++) { |
3222 | unsigned int len = min(obj_overlap, (u64)PAGE_SIZE); | |
3223 | ||
3224 | obj_req->copyup_bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
3225 | if (!obj_req->copyup_bvecs[i].bv_page) | |
3226 | return -ENOMEM; | |
3d7efd18 | 3227 | |
7e07efb1 ID |
3228 | obj_req->copyup_bvecs[i].bv_offset = 0; |
3229 | obj_req->copyup_bvecs[i].bv_len = len; | |
3230 | obj_overlap -= len; | |
3231 | } | |
b454e36d | 3232 | |
7e07efb1 ID |
3233 | rbd_assert(!obj_overlap); |
3234 | return 0; | |
b454e36d AE |
3235 | } |
3236 | ||
0ad5d953 ID |
3237 | /* |
3238 | * The target object doesn't exist. Read the data for the entire | |
3239 | * target object up to the overlap point (if any) from the parent, | |
3240 | * so we can use it for a copyup. | |
3241 | */ | |
793333a3 | 3242 | static int rbd_obj_copyup_read_parent(struct rbd_obj_request *obj_req) |
bf0d5f50 | 3243 | { |
3da691bf | 3244 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
3da691bf | 3245 | int ret; |
bf0d5f50 | 3246 | |
86bd7998 ID |
3247 | rbd_assert(obj_req->num_img_extents); |
3248 | prune_extents(obj_req->img_extents, &obj_req->num_img_extents, | |
3249 | rbd_dev->parent_overlap); | |
3250 | if (!obj_req->num_img_extents) { | |
3da691bf ID |
3251 | /* |
3252 | * The overlap has become 0 (most likely because the | |
3a482501 ID |
3253 | * image has been flattened). Re-submit the original write |
3254 | * request -- pass MODS_ONLY since the copyup isn't needed | |
3255 | * anymore. | |
3da691bf | 3256 | */ |
793333a3 | 3257 | return rbd_obj_copyup_current_snapc(obj_req, MODS_ONLY); |
bf0d5f50 AE |
3258 | } |
3259 | ||
86bd7998 | 3260 | ret = setup_copyup_bvecs(obj_req, rbd_obj_img_extents_bytes(obj_req)); |
3da691bf ID |
3261 | if (ret) |
3262 | return ret; | |
3263 | ||
86bd7998 | 3264 | return rbd_obj_read_from_parent(obj_req); |
bf0d5f50 | 3265 | } |
8b3e1a56 | 3266 | |
22e8bd51 ID |
3267 | static void rbd_obj_copyup_object_maps(struct rbd_obj_request *obj_req) |
3268 | { | |
3269 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
3270 | struct ceph_snap_context *snapc = obj_req->img_request->snapc; | |
3271 | u8 new_state; | |
3272 | u32 i; | |
3273 | int ret; | |
3274 | ||
3275 | rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending); | |
3276 | ||
3277 | if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) | |
3278 | return; | |
3279 | ||
3280 | if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS) | |
3281 | return; | |
3282 | ||
3283 | for (i = 0; i < snapc->num_snaps; i++) { | |
3284 | if ((rbd_dev->header.features & RBD_FEATURE_FAST_DIFF) && | |
3285 | i + 1 < snapc->num_snaps) | |
3286 | new_state = OBJECT_EXISTS_CLEAN; | |
3287 | else | |
3288 | new_state = OBJECT_EXISTS; | |
3289 | ||
3290 | ret = rbd_object_map_update(obj_req, snapc->snaps[i], | |
3291 | new_state, NULL); | |
3292 | if (ret < 0) { | |
3293 | obj_req->pending.result = ret; | |
3294 | return; | |
3295 | } | |
3296 | ||
3297 | rbd_assert(!ret); | |
3298 | obj_req->pending.num_pending++; | |
3299 | } | |
3300 | } | |
3301 | ||
793333a3 ID |
3302 | static void rbd_obj_copyup_write_object(struct rbd_obj_request *obj_req) |
3303 | { | |
3304 | u32 bytes = rbd_obj_img_extents_bytes(obj_req); | |
3305 | int ret; | |
3306 | ||
3307 | rbd_assert(!obj_req->pending.result && !obj_req->pending.num_pending); | |
3308 | ||
3309 | /* | |
3310 | * Only send non-zero copyup data to save some I/O and network | |
3311 | * bandwidth -- zero copyup data is equivalent to the object not | |
3312 | * existing. | |
3313 | */ | |
3314 | if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ZEROS) | |
3315 | bytes = 0; | |
3316 | ||
3317 | if (obj_req->img_request->snapc->num_snaps && bytes > 0) { | |
3318 | /* | |
3319 | * Send a copyup request with an empty snapshot context to | |
3320 | * deep-copyup the object through all existing snapshots. | |
3321 | * A second request with the current snapshot context will be | |
3322 | * sent for the actual modification. | |
3323 | */ | |
3324 | ret = rbd_obj_copyup_empty_snapc(obj_req, bytes); | |
3325 | if (ret) { | |
3326 | obj_req->pending.result = ret; | |
3327 | return; | |
3328 | } | |
3329 | ||
3330 | obj_req->pending.num_pending++; | |
3331 | bytes = MODS_ONLY; | |
3332 | } | |
3333 | ||
3334 | ret = rbd_obj_copyup_current_snapc(obj_req, bytes); | |
3335 | if (ret) { | |
3336 | obj_req->pending.result = ret; | |
3337 | return; | |
3338 | } | |
3339 | ||
3340 | obj_req->pending.num_pending++; | |
3341 | } | |
3342 | ||
3343 | static bool rbd_obj_advance_copyup(struct rbd_obj_request *obj_req, int *result) | |
3344 | { | |
22e8bd51 | 3345 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
793333a3 ID |
3346 | int ret; |
3347 | ||
3348 | again: | |
3349 | switch (obj_req->copyup_state) { | |
3350 | case RBD_OBJ_COPYUP_START: | |
3351 | rbd_assert(!*result); | |
3352 | ||
3353 | ret = rbd_obj_copyup_read_parent(obj_req); | |
3354 | if (ret) { | |
3355 | *result = ret; | |
3356 | return true; | |
3357 | } | |
3358 | if (obj_req->num_img_extents) | |
3359 | obj_req->copyup_state = RBD_OBJ_COPYUP_READ_PARENT; | |
3360 | else | |
3361 | obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT; | |
3362 | return false; | |
3363 | case RBD_OBJ_COPYUP_READ_PARENT: | |
3364 | if (*result) | |
3365 | return true; | |
3366 | ||
3367 | if (is_zero_bvecs(obj_req->copyup_bvecs, | |
3368 | rbd_obj_img_extents_bytes(obj_req))) { | |
3369 | dout("%s %p detected zeros\n", __func__, obj_req); | |
3370 | obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ZEROS; | |
3371 | } | |
3372 | ||
22e8bd51 ID |
3373 | rbd_obj_copyup_object_maps(obj_req); |
3374 | if (!obj_req->pending.num_pending) { | |
3375 | *result = obj_req->pending.result; | |
3376 | obj_req->copyup_state = RBD_OBJ_COPYUP_OBJECT_MAPS; | |
3377 | goto again; | |
3378 | } | |
3379 | obj_req->copyup_state = __RBD_OBJ_COPYUP_OBJECT_MAPS; | |
3380 | return false; | |
3381 | case __RBD_OBJ_COPYUP_OBJECT_MAPS: | |
3382 | if (!pending_result_dec(&obj_req->pending, result)) | |
3383 | return false; | |
3384 | /* fall through */ | |
3385 | case RBD_OBJ_COPYUP_OBJECT_MAPS: | |
3386 | if (*result) { | |
3387 | rbd_warn(rbd_dev, "snap object map update failed: %d", | |
3388 | *result); | |
3389 | return true; | |
3390 | } | |
3391 | ||
793333a3 ID |
3392 | rbd_obj_copyup_write_object(obj_req); |
3393 | if (!obj_req->pending.num_pending) { | |
3394 | *result = obj_req->pending.result; | |
3395 | obj_req->copyup_state = RBD_OBJ_COPYUP_WRITE_OBJECT; | |
3396 | goto again; | |
3397 | } | |
3398 | obj_req->copyup_state = __RBD_OBJ_COPYUP_WRITE_OBJECT; | |
3399 | return false; | |
3400 | case __RBD_OBJ_COPYUP_WRITE_OBJECT: | |
3401 | if (!pending_result_dec(&obj_req->pending, result)) | |
3402 | return false; | |
3403 | /* fall through */ | |
3404 | case RBD_OBJ_COPYUP_WRITE_OBJECT: | |
3405 | return true; | |
3406 | default: | |
3407 | BUG(); | |
3408 | } | |
3409 | } | |
3410 | ||
22e8bd51 ID |
3411 | /* |
3412 | * Return: | |
3413 | * 0 - object map update sent | |
3414 | * 1 - object map update isn't needed | |
3415 | * <0 - error | |
3416 | */ | |
3417 | static int rbd_obj_write_post_object_map(struct rbd_obj_request *obj_req) | |
3418 | { | |
3419 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; | |
3420 | u8 current_state = OBJECT_PENDING; | |
3421 | ||
3422 | if (!(rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) | |
3423 | return 1; | |
3424 | ||
3425 | if (!(obj_req->flags & RBD_OBJ_FLAG_DELETION)) | |
3426 | return 1; | |
3427 | ||
3428 | return rbd_object_map_update(obj_req, CEPH_NOSNAP, OBJECT_NONEXISTENT, | |
3429 | ¤t_state); | |
3430 | } | |
3431 | ||
85b5e6d1 | 3432 | static bool rbd_obj_advance_write(struct rbd_obj_request *obj_req, int *result) |
8b3e1a56 | 3433 | { |
793333a3 | 3434 | struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev; |
3da691bf | 3435 | int ret; |
8b3e1a56 | 3436 | |
793333a3 | 3437 | again: |
3da691bf | 3438 | switch (obj_req->write_state) { |
85b5e6d1 ID |
3439 | case RBD_OBJ_WRITE_START: |
3440 | rbd_assert(!*result); | |
3441 | ||
22e8bd51 ID |
3442 | if (rbd_obj_write_is_noop(obj_req)) |
3443 | return true; | |
3444 | ||
3445 | ret = rbd_obj_write_pre_object_map(obj_req); | |
3446 | if (ret < 0) { | |
3447 | *result = ret; | |
3448 | return true; | |
3449 | } | |
3450 | obj_req->write_state = RBD_OBJ_WRITE_PRE_OBJECT_MAP; | |
3451 | if (ret > 0) | |
3452 | goto again; | |
3453 | return false; | |
3454 | case RBD_OBJ_WRITE_PRE_OBJECT_MAP: | |
3455 | if (*result) { | |
3456 | rbd_warn(rbd_dev, "pre object map update failed: %d", | |
3457 | *result); | |
3458 | return true; | |
3459 | } | |
85b5e6d1 ID |
3460 | ret = rbd_obj_write_object(obj_req); |
3461 | if (ret) { | |
3462 | *result = ret; | |
3463 | return true; | |
3464 | } | |
3465 | obj_req->write_state = RBD_OBJ_WRITE_OBJECT; | |
3466 | return false; | |
0ad5d953 | 3467 | case RBD_OBJ_WRITE_OBJECT: |
54ab3b24 | 3468 | if (*result == -ENOENT) { |
0ad5d953 | 3469 | if (obj_req->flags & RBD_OBJ_FLAG_COPYUP_ENABLED) { |
793333a3 ID |
3470 | *result = 0; |
3471 | obj_req->copyup_state = RBD_OBJ_COPYUP_START; | |
3472 | obj_req->write_state = __RBD_OBJ_WRITE_COPYUP; | |
3473 | goto again; | |
0ad5d953 | 3474 | } |
3da691bf | 3475 | /* |
0ad5d953 ID |
3476 | * On a non-existent object: |
3477 | * delete - -ENOENT, truncate/zero - 0 | |
3da691bf | 3478 | */ |
0ad5d953 ID |
3479 | if (obj_req->flags & RBD_OBJ_FLAG_DELETION) |
3480 | *result = 0; | |
3da691bf | 3481 | } |
a9b67e69 | 3482 | if (*result) |
3a482501 | 3483 | return true; |
8b3e1a56 | 3484 | |
793333a3 ID |
3485 | obj_req->write_state = RBD_OBJ_WRITE_COPYUP; |
3486 | goto again; | |
3487 | case __RBD_OBJ_WRITE_COPYUP: | |
3488 | if (!rbd_obj_advance_copyup(obj_req, result)) | |
3489 | return false; | |
3490 | /* fall through */ | |
3491 | case RBD_OBJ_WRITE_COPYUP: | |
22e8bd51 | 3492 | if (*result) { |
793333a3 | 3493 | rbd_warn(rbd_dev, "copyup failed: %d", *result); |
22e8bd51 ID |
3494 | return true; |
3495 | } | |
3496 | ret = rbd_obj_write_post_object_map(obj_req); | |
3497 | if (ret < 0) { | |
3498 | *result = ret; | |
3499 | return true; | |
3500 | } | |
3501 | obj_req->write_state = RBD_OBJ_WRITE_POST_OBJECT_MAP; | |
3502 | if (ret > 0) | |
3503 | goto again; | |
3504 | return false; | |
3505 | case RBD_OBJ_WRITE_POST_OBJECT_MAP: | |
3506 | if (*result) | |
3507 | rbd_warn(rbd_dev, "post object map update failed: %d", | |
3508 | *result); | |
793333a3 | 3509 | return true; |
3da691bf | 3510 | default: |
c6244b3b | 3511 | BUG(); |
3da691bf ID |
3512 | } |
3513 | } | |
02c74fba | 3514 | |
3da691bf | 3515 | /* |
0ad5d953 | 3516 | * Return true if @obj_req is completed. |
3da691bf | 3517 | */ |
54ab3b24 ID |
3518 | static bool __rbd_obj_handle_request(struct rbd_obj_request *obj_req, |
3519 | int *result) | |
3da691bf | 3520 | { |
0ad5d953 | 3521 | struct rbd_img_request *img_req = obj_req->img_request; |
0192ce2e | 3522 | struct rbd_device *rbd_dev = img_req->rbd_dev; |
0ad5d953 ID |
3523 | bool done; |
3524 | ||
85b5e6d1 | 3525 | mutex_lock(&obj_req->state_mutex); |
0ad5d953 | 3526 | if (!rbd_img_is_write(img_req)) |
85b5e6d1 | 3527 | done = rbd_obj_advance_read(obj_req, result); |
0ad5d953 | 3528 | else |
85b5e6d1 ID |
3529 | done = rbd_obj_advance_write(obj_req, result); |
3530 | mutex_unlock(&obj_req->state_mutex); | |
0ad5d953 | 3531 | |
0192ce2e ID |
3532 | if (done && *result) { |
3533 | rbd_assert(*result < 0); | |
3534 | rbd_warn(rbd_dev, "%s at objno %llu %llu~%llu result %d", | |
3535 | obj_op_name(img_req->op_type), obj_req->ex.oe_objno, | |
3536 | obj_req->ex.oe_off, obj_req->ex.oe_len, *result); | |
3537 | } | |
0ad5d953 | 3538 | return done; |
3da691bf | 3539 | } |
02c74fba | 3540 | |
0192ce2e ID |
3541 | /* |
3542 | * This is open-coded in rbd_img_handle_request() to avoid parent chain | |
3543 | * recursion. | |
3544 | */ | |
3545 | static void rbd_obj_handle_request(struct rbd_obj_request *obj_req, int result) | |
3546 | { | |
3547 | if (__rbd_obj_handle_request(obj_req, &result)) | |
3548 | rbd_img_handle_request(obj_req->img_request, result); | |
3549 | } | |
3550 | ||
e1fddc8f ID |
3551 | static bool need_exclusive_lock(struct rbd_img_request *img_req) |
3552 | { | |
3553 | struct rbd_device *rbd_dev = img_req->rbd_dev; | |
3554 | ||
3555 | if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) | |
3556 | return false; | |
3557 | ||
3558 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
3559 | return false; | |
3560 | ||
3561 | rbd_assert(!test_bit(IMG_REQ_CHILD, &img_req->flags)); | |
22e8bd51 ID |
3562 | if (rbd_dev->opts->lock_on_read || |
3563 | (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) | |
e1fddc8f ID |
3564 | return true; |
3565 | ||
3566 | return rbd_img_is_write(img_req); | |
3567 | } | |
3568 | ||
637cd060 | 3569 | static bool rbd_lock_add_request(struct rbd_img_request *img_req) |
e1fddc8f ID |
3570 | { |
3571 | struct rbd_device *rbd_dev = img_req->rbd_dev; | |
637cd060 | 3572 | bool locked; |
e1fddc8f ID |
3573 | |
3574 | lockdep_assert_held(&rbd_dev->lock_rwsem); | |
637cd060 | 3575 | locked = rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED; |
e1fddc8f ID |
3576 | spin_lock(&rbd_dev->lock_lists_lock); |
3577 | rbd_assert(list_empty(&img_req->lock_item)); | |
637cd060 ID |
3578 | if (!locked) |
3579 | list_add_tail(&img_req->lock_item, &rbd_dev->acquiring_list); | |
3580 | else | |
3581 | list_add_tail(&img_req->lock_item, &rbd_dev->running_list); | |
e1fddc8f | 3582 | spin_unlock(&rbd_dev->lock_lists_lock); |
637cd060 | 3583 | return locked; |
e1fddc8f ID |
3584 | } |
3585 | ||
3586 | static void rbd_lock_del_request(struct rbd_img_request *img_req) | |
3587 | { | |
3588 | struct rbd_device *rbd_dev = img_req->rbd_dev; | |
3589 | bool need_wakeup; | |
3590 | ||
3591 | lockdep_assert_held(&rbd_dev->lock_rwsem); | |
3592 | spin_lock(&rbd_dev->lock_lists_lock); | |
3593 | rbd_assert(!list_empty(&img_req->lock_item)); | |
3594 | list_del_init(&img_req->lock_item); | |
3595 | need_wakeup = (rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING && | |
3596 | list_empty(&rbd_dev->running_list)); | |
3597 | spin_unlock(&rbd_dev->lock_lists_lock); | |
3598 | if (need_wakeup) | |
3599 | complete(&rbd_dev->releasing_wait); | |
3600 | } | |
3601 | ||
637cd060 ID |
3602 | static int rbd_img_exclusive_lock(struct rbd_img_request *img_req) |
3603 | { | |
3604 | struct rbd_device *rbd_dev = img_req->rbd_dev; | |
3605 | ||
3606 | if (!need_exclusive_lock(img_req)) | |
3607 | return 1; | |
3608 | ||
3609 | if (rbd_lock_add_request(img_req)) | |
3610 | return 1; | |
3611 | ||
3612 | if (rbd_dev->opts->exclusive) { | |
3613 | WARN_ON(1); /* lock got released? */ | |
3614 | return -EROFS; | |
3615 | } | |
3616 | ||
3617 | /* | |
3618 | * Note the use of mod_delayed_work() in rbd_acquire_lock() | |
3619 | * and cancel_delayed_work() in wake_lock_waiters(). | |
3620 | */ | |
3621 | dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev); | |
3622 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); | |
3623 | return 0; | |
3624 | } | |
3625 | ||
0192ce2e | 3626 | static void rbd_img_object_requests(struct rbd_img_request *img_req) |
7114edac | 3627 | { |
0192ce2e | 3628 | struct rbd_obj_request *obj_req; |
7114edac | 3629 | |
0192ce2e ID |
3630 | rbd_assert(!img_req->pending.result && !img_req->pending.num_pending); |
3631 | ||
3632 | for_each_obj_request(img_req, obj_req) { | |
3633 | int result = 0; | |
a9e8ba2c | 3634 | |
0192ce2e ID |
3635 | if (__rbd_obj_handle_request(obj_req, &result)) { |
3636 | if (result) { | |
3637 | img_req->pending.result = result; | |
3638 | return; | |
3639 | } | |
3640 | } else { | |
3641 | img_req->pending.num_pending++; | |
3642 | } | |
3643 | } | |
8b3e1a56 AE |
3644 | } |
3645 | ||
0192ce2e | 3646 | static bool rbd_img_advance(struct rbd_img_request *img_req, int *result) |
8b3e1a56 | 3647 | { |
637cd060 | 3648 | struct rbd_device *rbd_dev = img_req->rbd_dev; |
3da691bf | 3649 | int ret; |
8b3e1a56 | 3650 | |
0192ce2e ID |
3651 | again: |
3652 | switch (img_req->state) { | |
3653 | case RBD_IMG_START: | |
3654 | rbd_assert(!*result); | |
8b3e1a56 | 3655 | |
637cd060 ID |
3656 | ret = rbd_img_exclusive_lock(img_req); |
3657 | if (ret < 0) { | |
3658 | *result = ret; | |
3da691bf ID |
3659 | return true; |
3660 | } | |
637cd060 ID |
3661 | img_req->state = RBD_IMG_EXCLUSIVE_LOCK; |
3662 | if (ret > 0) | |
3663 | goto again; | |
3da691bf | 3664 | return false; |
637cd060 ID |
3665 | case RBD_IMG_EXCLUSIVE_LOCK: |
3666 | if (*result) | |
89a59c1c ID |
3667 | return true; |
3668 | ||
637cd060 ID |
3669 | rbd_assert(!need_exclusive_lock(img_req) || |
3670 | __rbd_is_lock_owner(rbd_dev)); | |
3671 | ||
0192ce2e ID |
3672 | rbd_img_object_requests(img_req); |
3673 | if (!img_req->pending.num_pending) { | |
3674 | *result = img_req->pending.result; | |
3675 | img_req->state = RBD_IMG_OBJECT_REQUESTS; | |
3676 | goto again; | |
3da691bf | 3677 | } |
0192ce2e | 3678 | img_req->state = __RBD_IMG_OBJECT_REQUESTS; |
3da691bf | 3679 | return false; |
0192ce2e ID |
3680 | case __RBD_IMG_OBJECT_REQUESTS: |
3681 | if (!pending_result_dec(&img_req->pending, result)) | |
3682 | return false; | |
3683 | /* fall through */ | |
3684 | case RBD_IMG_OBJECT_REQUESTS: | |
3685 | return true; | |
3da691bf | 3686 | default: |
c6244b3b | 3687 | BUG(); |
3da691bf ID |
3688 | } |
3689 | } | |
02c74fba | 3690 | |
3da691bf | 3691 | /* |
0192ce2e | 3692 | * Return true if @img_req is completed. |
3da691bf | 3693 | */ |
0192ce2e ID |
3694 | static bool __rbd_img_handle_request(struct rbd_img_request *img_req, |
3695 | int *result) | |
7114edac | 3696 | { |
0192ce2e ID |
3697 | struct rbd_device *rbd_dev = img_req->rbd_dev; |
3698 | bool done; | |
7114edac | 3699 | |
e1fddc8f ID |
3700 | if (need_exclusive_lock(img_req)) { |
3701 | down_read(&rbd_dev->lock_rwsem); | |
3702 | mutex_lock(&img_req->state_mutex); | |
3703 | done = rbd_img_advance(img_req, result); | |
3704 | if (done) | |
3705 | rbd_lock_del_request(img_req); | |
3706 | mutex_unlock(&img_req->state_mutex); | |
3707 | up_read(&rbd_dev->lock_rwsem); | |
3708 | } else { | |
3709 | mutex_lock(&img_req->state_mutex); | |
3710 | done = rbd_img_advance(img_req, result); | |
3711 | mutex_unlock(&img_req->state_mutex); | |
02c74fba | 3712 | } |
a9e8ba2c | 3713 | |
0192ce2e ID |
3714 | if (done && *result) { |
3715 | rbd_assert(*result < 0); | |
3716 | rbd_warn(rbd_dev, "%s%s result %d", | |
3717 | test_bit(IMG_REQ_CHILD, &img_req->flags) ? "child " : "", | |
3718 | obj_op_name(img_req->op_type), *result); | |
7114edac | 3719 | } |
0192ce2e | 3720 | return done; |
7114edac | 3721 | } |
a9e8ba2c | 3722 | |
0192ce2e | 3723 | static void rbd_img_handle_request(struct rbd_img_request *img_req, int result) |
3da691bf | 3724 | { |
7114edac | 3725 | again: |
0192ce2e | 3726 | if (!__rbd_img_handle_request(img_req, &result)) |
7114edac | 3727 | return; |
8b3e1a56 | 3728 | |
7114edac | 3729 | if (test_bit(IMG_REQ_CHILD, &img_req->flags)) { |
0192ce2e ID |
3730 | struct rbd_obj_request *obj_req = img_req->obj_request; |
3731 | ||
54ab3b24 | 3732 | rbd_img_request_put(img_req); |
0192ce2e ID |
3733 | if (__rbd_obj_handle_request(obj_req, &result)) { |
3734 | img_req = obj_req->img_request; | |
3735 | goto again; | |
3736 | } | |
3737 | } else { | |
3738 | struct request *rq = img_req->rq; | |
3739 | ||
3740 | rbd_img_request_put(img_req); | |
3741 | blk_mq_end_request(rq, errno_to_blk_status(result)); | |
7114edac | 3742 | } |
8b3e1a56 | 3743 | } |
bf0d5f50 | 3744 | |
ed95b21a | 3745 | static const struct rbd_client_id rbd_empty_cid; |
b8d70035 | 3746 | |
ed95b21a ID |
3747 | static bool rbd_cid_equal(const struct rbd_client_id *lhs, |
3748 | const struct rbd_client_id *rhs) | |
3749 | { | |
3750 | return lhs->gid == rhs->gid && lhs->handle == rhs->handle; | |
3751 | } | |
3752 | ||
3753 | static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev) | |
3754 | { | |
3755 | struct rbd_client_id cid; | |
3756 | ||
3757 | mutex_lock(&rbd_dev->watch_mutex); | |
3758 | cid.gid = ceph_client_gid(rbd_dev->rbd_client->client); | |
3759 | cid.handle = rbd_dev->watch_cookie; | |
3760 | mutex_unlock(&rbd_dev->watch_mutex); | |
3761 | return cid; | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | * lock_rwsem must be held for write | |
3766 | */ | |
3767 | static void rbd_set_owner_cid(struct rbd_device *rbd_dev, | |
3768 | const struct rbd_client_id *cid) | |
3769 | { | |
3770 | dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev, | |
3771 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle, | |
3772 | cid->gid, cid->handle); | |
3773 | rbd_dev->owner_cid = *cid; /* struct */ | |
3774 | } | |
3775 | ||
3776 | static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf) | |
3777 | { | |
3778 | mutex_lock(&rbd_dev->watch_mutex); | |
3779 | sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie); | |
3780 | mutex_unlock(&rbd_dev->watch_mutex); | |
3781 | } | |
3782 | ||
edd8ca80 FM |
3783 | static void __rbd_lock(struct rbd_device *rbd_dev, const char *cookie) |
3784 | { | |
3785 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); | |
3786 | ||
a2b1da09 | 3787 | rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED; |
edd8ca80 FM |
3788 | strcpy(rbd_dev->lock_cookie, cookie); |
3789 | rbd_set_owner_cid(rbd_dev, &cid); | |
3790 | queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work); | |
3791 | } | |
3792 | ||
ed95b21a ID |
3793 | /* |
3794 | * lock_rwsem must be held for write | |
3795 | */ | |
3796 | static int rbd_lock(struct rbd_device *rbd_dev) | |
b8d70035 | 3797 | { |
922dab61 | 3798 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
ed95b21a | 3799 | char cookie[32]; |
e627db08 | 3800 | int ret; |
b8d70035 | 3801 | |
cbbfb0ff ID |
3802 | WARN_ON(__rbd_is_lock_owner(rbd_dev) || |
3803 | rbd_dev->lock_cookie[0] != '\0'); | |
52bb1f9b | 3804 | |
ed95b21a ID |
3805 | format_lock_cookie(rbd_dev, cookie); |
3806 | ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
3807 | RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie, | |
3808 | RBD_LOCK_TAG, "", 0); | |
e627db08 | 3809 | if (ret) |
ed95b21a | 3810 | return ret; |
b8d70035 | 3811 | |
edd8ca80 | 3812 | __rbd_lock(rbd_dev, cookie); |
ed95b21a | 3813 | return 0; |
b8d70035 AE |
3814 | } |
3815 | ||
ed95b21a ID |
3816 | /* |
3817 | * lock_rwsem must be held for write | |
3818 | */ | |
bbead745 | 3819 | static void rbd_unlock(struct rbd_device *rbd_dev) |
bb040aa0 | 3820 | { |
922dab61 | 3821 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
bb040aa0 ID |
3822 | int ret; |
3823 | ||
cbbfb0ff ID |
3824 | WARN_ON(!__rbd_is_lock_owner(rbd_dev) || |
3825 | rbd_dev->lock_cookie[0] == '\0'); | |
bb040aa0 | 3826 | |
ed95b21a | 3827 | ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, |
cbbfb0ff | 3828 | RBD_LOCK_NAME, rbd_dev->lock_cookie); |
bbead745 | 3829 | if (ret && ret != -ENOENT) |
637cd060 | 3830 | rbd_warn(rbd_dev, "failed to unlock header: %d", ret); |
bb040aa0 | 3831 | |
bbead745 ID |
3832 | /* treat errors as the image is unlocked */ |
3833 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; | |
cbbfb0ff | 3834 | rbd_dev->lock_cookie[0] = '\0'; |
ed95b21a ID |
3835 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); |
3836 | queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work); | |
bb040aa0 ID |
3837 | } |
3838 | ||
ed95b21a ID |
3839 | static int __rbd_notify_op_lock(struct rbd_device *rbd_dev, |
3840 | enum rbd_notify_op notify_op, | |
3841 | struct page ***preply_pages, | |
3842 | size_t *preply_len) | |
9969ebc5 AE |
3843 | { |
3844 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
ed95b21a | 3845 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); |
08a79102 KS |
3846 | char buf[4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN]; |
3847 | int buf_size = sizeof(buf); | |
ed95b21a | 3848 | void *p = buf; |
9969ebc5 | 3849 | |
ed95b21a | 3850 | dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op); |
9969ebc5 | 3851 | |
ed95b21a ID |
3852 | /* encode *LockPayload NotifyMessage (op + ClientId) */ |
3853 | ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN); | |
3854 | ceph_encode_32(&p, notify_op); | |
3855 | ceph_encode_64(&p, cid.gid); | |
3856 | ceph_encode_64(&p, cid.handle); | |
8eb87565 | 3857 | |
ed95b21a ID |
3858 | return ceph_osdc_notify(osdc, &rbd_dev->header_oid, |
3859 | &rbd_dev->header_oloc, buf, buf_size, | |
3860 | RBD_NOTIFY_TIMEOUT, preply_pages, preply_len); | |
b30a01f2 ID |
3861 | } |
3862 | ||
ed95b21a ID |
3863 | static void rbd_notify_op_lock(struct rbd_device *rbd_dev, |
3864 | enum rbd_notify_op notify_op) | |
b30a01f2 | 3865 | { |
ed95b21a ID |
3866 | struct page **reply_pages; |
3867 | size_t reply_len; | |
b30a01f2 | 3868 | |
ed95b21a ID |
3869 | __rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len); |
3870 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3871 | } | |
b30a01f2 | 3872 | |
ed95b21a ID |
3873 | static void rbd_notify_acquired_lock(struct work_struct *work) |
3874 | { | |
3875 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3876 | acquired_lock_work); | |
76756a51 | 3877 | |
ed95b21a | 3878 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK); |
c525f036 ID |
3879 | } |
3880 | ||
ed95b21a | 3881 | static void rbd_notify_released_lock(struct work_struct *work) |
c525f036 | 3882 | { |
ed95b21a ID |
3883 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, |
3884 | released_lock_work); | |
811c6688 | 3885 | |
ed95b21a | 3886 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK); |
fca27065 ID |
3887 | } |
3888 | ||
ed95b21a | 3889 | static int rbd_request_lock(struct rbd_device *rbd_dev) |
36be9a76 | 3890 | { |
ed95b21a ID |
3891 | struct page **reply_pages; |
3892 | size_t reply_len; | |
3893 | bool lock_owner_responded = false; | |
36be9a76 AE |
3894 | int ret; |
3895 | ||
ed95b21a | 3896 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
36be9a76 | 3897 | |
ed95b21a ID |
3898 | ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK, |
3899 | &reply_pages, &reply_len); | |
3900 | if (ret && ret != -ETIMEDOUT) { | |
3901 | rbd_warn(rbd_dev, "failed to request lock: %d", ret); | |
36be9a76 | 3902 | goto out; |
ed95b21a | 3903 | } |
36be9a76 | 3904 | |
ed95b21a ID |
3905 | if (reply_len > 0 && reply_len <= PAGE_SIZE) { |
3906 | void *p = page_address(reply_pages[0]); | |
3907 | void *const end = p + reply_len; | |
3908 | u32 n; | |
36be9a76 | 3909 | |
ed95b21a ID |
3910 | ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */ |
3911 | while (n--) { | |
3912 | u8 struct_v; | |
3913 | u32 len; | |
36be9a76 | 3914 | |
ed95b21a ID |
3915 | ceph_decode_need(&p, end, 8 + 8, e_inval); |
3916 | p += 8 + 8; /* skip gid and cookie */ | |
04017e29 | 3917 | |
ed95b21a ID |
3918 | ceph_decode_32_safe(&p, end, len, e_inval); |
3919 | if (!len) | |
3920 | continue; | |
3921 | ||
3922 | if (lock_owner_responded) { | |
3923 | rbd_warn(rbd_dev, | |
3924 | "duplicate lock owners detected"); | |
3925 | ret = -EIO; | |
3926 | goto out; | |
3927 | } | |
3928 | ||
3929 | lock_owner_responded = true; | |
3930 | ret = ceph_start_decoding(&p, end, 1, "ResponseMessage", | |
3931 | &struct_v, &len); | |
3932 | if (ret) { | |
3933 | rbd_warn(rbd_dev, | |
3934 | "failed to decode ResponseMessage: %d", | |
3935 | ret); | |
3936 | goto e_inval; | |
3937 | } | |
3938 | ||
3939 | ret = ceph_decode_32(&p); | |
3940 | } | |
3941 | } | |
3942 | ||
3943 | if (!lock_owner_responded) { | |
3944 | rbd_warn(rbd_dev, "no lock owners detected"); | |
3945 | ret = -ETIMEDOUT; | |
3946 | } | |
3947 | ||
3948 | out: | |
3949 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3950 | return ret; | |
3951 | ||
3952 | e_inval: | |
3953 | ret = -EINVAL; | |
3954 | goto out; | |
3955 | } | |
3956 | ||
637cd060 ID |
3957 | /* |
3958 | * Either image request state machine(s) or rbd_add_acquire_lock() | |
3959 | * (i.e. "rbd map"). | |
3960 | */ | |
3961 | static void wake_lock_waiters(struct rbd_device *rbd_dev, int result) | |
ed95b21a | 3962 | { |
637cd060 ID |
3963 | struct rbd_img_request *img_req; |
3964 | ||
3965 | dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result); | |
d9b9c893 | 3966 | lockdep_assert_held_write(&rbd_dev->lock_rwsem); |
ed95b21a ID |
3967 | |
3968 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
637cd060 ID |
3969 | if (!completion_done(&rbd_dev->acquire_wait)) { |
3970 | rbd_assert(list_empty(&rbd_dev->acquiring_list) && | |
3971 | list_empty(&rbd_dev->running_list)); | |
3972 | rbd_dev->acquire_err = result; | |
3973 | complete_all(&rbd_dev->acquire_wait); | |
3974 | return; | |
3975 | } | |
3976 | ||
3977 | list_for_each_entry(img_req, &rbd_dev->acquiring_list, lock_item) { | |
3978 | mutex_lock(&img_req->state_mutex); | |
3979 | rbd_assert(img_req->state == RBD_IMG_EXCLUSIVE_LOCK); | |
3980 | rbd_img_schedule(img_req, result); | |
3981 | mutex_unlock(&img_req->state_mutex); | |
3982 | } | |
3983 | ||
3984 | list_splice_tail_init(&rbd_dev->acquiring_list, &rbd_dev->running_list); | |
ed95b21a ID |
3985 | } |
3986 | ||
3987 | static int get_lock_owner_info(struct rbd_device *rbd_dev, | |
3988 | struct ceph_locker **lockers, u32 *num_lockers) | |
3989 | { | |
3990 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3991 | u8 lock_type; | |
3992 | char *lock_tag; | |
3993 | int ret; | |
3994 | ||
3995 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3996 | ||
3997 | ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid, | |
3998 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
3999 | &lock_type, &lock_tag, lockers, num_lockers); | |
4000 | if (ret) | |
4001 | return ret; | |
4002 | ||
4003 | if (*num_lockers == 0) { | |
4004 | dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev); | |
4005 | goto out; | |
4006 | } | |
4007 | ||
4008 | if (strcmp(lock_tag, RBD_LOCK_TAG)) { | |
4009 | rbd_warn(rbd_dev, "locked by external mechanism, tag %s", | |
4010 | lock_tag); | |
4011 | ret = -EBUSY; | |
4012 | goto out; | |
4013 | } | |
4014 | ||
4015 | if (lock_type == CEPH_CLS_LOCK_SHARED) { | |
4016 | rbd_warn(rbd_dev, "shared lock type detected"); | |
4017 | ret = -EBUSY; | |
4018 | goto out; | |
4019 | } | |
4020 | ||
4021 | if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX, | |
4022 | strlen(RBD_LOCK_COOKIE_PREFIX))) { | |
4023 | rbd_warn(rbd_dev, "locked by external mechanism, cookie %s", | |
4024 | (*lockers)[0].id.cookie); | |
4025 | ret = -EBUSY; | |
4026 | goto out; | |
4027 | } | |
4028 | ||
4029 | out: | |
4030 | kfree(lock_tag); | |
4031 | return ret; | |
4032 | } | |
4033 | ||
4034 | static int find_watcher(struct rbd_device *rbd_dev, | |
4035 | const struct ceph_locker *locker) | |
4036 | { | |
4037 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
4038 | struct ceph_watch_item *watchers; | |
4039 | u32 num_watchers; | |
4040 | u64 cookie; | |
4041 | int i; | |
4042 | int ret; | |
4043 | ||
4044 | ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid, | |
4045 | &rbd_dev->header_oloc, &watchers, | |
4046 | &num_watchers); | |
4047 | if (ret) | |
4048 | return ret; | |
4049 | ||
4050 | sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie); | |
4051 | for (i = 0; i < num_watchers; i++) { | |
4052 | if (!memcmp(&watchers[i].addr, &locker->info.addr, | |
4053 | sizeof(locker->info.addr)) && | |
4054 | watchers[i].cookie == cookie) { | |
4055 | struct rbd_client_id cid = { | |
4056 | .gid = le64_to_cpu(watchers[i].name.num), | |
4057 | .handle = cookie, | |
4058 | }; | |
4059 | ||
4060 | dout("%s rbd_dev %p found cid %llu-%llu\n", __func__, | |
4061 | rbd_dev, cid.gid, cid.handle); | |
4062 | rbd_set_owner_cid(rbd_dev, &cid); | |
4063 | ret = 1; | |
4064 | goto out; | |
4065 | } | |
4066 | } | |
4067 | ||
4068 | dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev); | |
4069 | ret = 0; | |
4070 | out: | |
4071 | kfree(watchers); | |
4072 | return ret; | |
4073 | } | |
4074 | ||
4075 | /* | |
4076 | * lock_rwsem must be held for write | |
4077 | */ | |
4078 | static int rbd_try_lock(struct rbd_device *rbd_dev) | |
4079 | { | |
4080 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
4081 | struct ceph_locker *lockers; | |
4082 | u32 num_lockers; | |
4083 | int ret; | |
4084 | ||
4085 | for (;;) { | |
4086 | ret = rbd_lock(rbd_dev); | |
4087 | if (ret != -EBUSY) | |
4088 | return ret; | |
4089 | ||
4090 | /* determine if the current lock holder is still alive */ | |
4091 | ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers); | |
4092 | if (ret) | |
4093 | return ret; | |
4094 | ||
4095 | if (num_lockers == 0) | |
4096 | goto again; | |
4097 | ||
4098 | ret = find_watcher(rbd_dev, lockers); | |
637cd060 ID |
4099 | if (ret) |
4100 | goto out; /* request lock or error */ | |
ed95b21a | 4101 | |
22e8bd51 | 4102 | rbd_warn(rbd_dev, "breaking header lock owned by %s%llu", |
ed95b21a ID |
4103 | ENTITY_NAME(lockers[0].id.name)); |
4104 | ||
4105 | ret = ceph_monc_blacklist_add(&client->monc, | |
4106 | &lockers[0].info.addr); | |
4107 | if (ret) { | |
4108 | rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d", | |
4109 | ENTITY_NAME(lockers[0].id.name), ret); | |
4110 | goto out; | |
4111 | } | |
4112 | ||
4113 | ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid, | |
4114 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
4115 | lockers[0].id.cookie, | |
4116 | &lockers[0].id.name); | |
4117 | if (ret && ret != -ENOENT) | |
4118 | goto out; | |
4119 | ||
4120 | again: | |
4121 | ceph_free_lockers(lockers, num_lockers); | |
4122 | } | |
4123 | ||
4124 | out: | |
4125 | ceph_free_lockers(lockers, num_lockers); | |
4126 | return ret; | |
4127 | } | |
4128 | ||
22e8bd51 ID |
4129 | static int rbd_post_acquire_action(struct rbd_device *rbd_dev) |
4130 | { | |
4131 | int ret; | |
4132 | ||
4133 | if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) { | |
4134 | ret = rbd_object_map_open(rbd_dev); | |
4135 | if (ret) | |
4136 | return ret; | |
4137 | } | |
4138 | ||
4139 | return 0; | |
4140 | } | |
4141 | ||
ed95b21a | 4142 | /* |
637cd060 ID |
4143 | * Return: |
4144 | * 0 - lock acquired | |
4145 | * 1 - caller should call rbd_request_lock() | |
4146 | * <0 - error | |
ed95b21a | 4147 | */ |
637cd060 | 4148 | static int rbd_try_acquire_lock(struct rbd_device *rbd_dev) |
ed95b21a | 4149 | { |
637cd060 | 4150 | int ret; |
ed95b21a ID |
4151 | |
4152 | down_read(&rbd_dev->lock_rwsem); | |
4153 | dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev, | |
4154 | rbd_dev->lock_state); | |
4155 | if (__rbd_is_lock_owner(rbd_dev)) { | |
ed95b21a | 4156 | up_read(&rbd_dev->lock_rwsem); |
637cd060 | 4157 | return 0; |
ed95b21a ID |
4158 | } |
4159 | ||
4160 | up_read(&rbd_dev->lock_rwsem); | |
4161 | down_write(&rbd_dev->lock_rwsem); | |
4162 | dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev, | |
4163 | rbd_dev->lock_state); | |
637cd060 ID |
4164 | if (__rbd_is_lock_owner(rbd_dev)) { |
4165 | up_write(&rbd_dev->lock_rwsem); | |
4166 | return 0; | |
ed95b21a ID |
4167 | } |
4168 | ||
637cd060 ID |
4169 | ret = rbd_try_lock(rbd_dev); |
4170 | if (ret < 0) { | |
4171 | rbd_warn(rbd_dev, "failed to lock header: %d", ret); | |
4172 | if (ret == -EBLACKLISTED) | |
4173 | goto out; | |
4174 | ||
4175 | ret = 1; /* request lock anyway */ | |
4176 | } | |
4177 | if (ret > 0) { | |
4178 | up_write(&rbd_dev->lock_rwsem); | |
4179 | return ret; | |
4180 | } | |
4181 | ||
4182 | rbd_assert(rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED); | |
4183 | rbd_assert(list_empty(&rbd_dev->running_list)); | |
4184 | ||
22e8bd51 ID |
4185 | ret = rbd_post_acquire_action(rbd_dev); |
4186 | if (ret) { | |
4187 | rbd_warn(rbd_dev, "post-acquire action failed: %d", ret); | |
4188 | /* | |
4189 | * Can't stay in RBD_LOCK_STATE_LOCKED because | |
4190 | * rbd_lock_add_request() would let the request through, | |
4191 | * assuming that e.g. object map is locked and loaded. | |
4192 | */ | |
4193 | rbd_unlock(rbd_dev); | |
ed95b21a ID |
4194 | } |
4195 | ||
637cd060 ID |
4196 | out: |
4197 | wake_lock_waiters(rbd_dev, ret); | |
ed95b21a | 4198 | up_write(&rbd_dev->lock_rwsem); |
637cd060 | 4199 | return ret; |
ed95b21a ID |
4200 | } |
4201 | ||
4202 | static void rbd_acquire_lock(struct work_struct *work) | |
4203 | { | |
4204 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
4205 | struct rbd_device, lock_dwork); | |
637cd060 | 4206 | int ret; |
ed95b21a ID |
4207 | |
4208 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
4209 | again: | |
637cd060 ID |
4210 | ret = rbd_try_acquire_lock(rbd_dev); |
4211 | if (ret <= 0) { | |
4212 | dout("%s rbd_dev %p ret %d - done\n", __func__, rbd_dev, ret); | |
ed95b21a ID |
4213 | return; |
4214 | } | |
4215 | ||
4216 | ret = rbd_request_lock(rbd_dev); | |
4217 | if (ret == -ETIMEDOUT) { | |
4218 | goto again; /* treat this as a dead client */ | |
e010dd0a ID |
4219 | } else if (ret == -EROFS) { |
4220 | rbd_warn(rbd_dev, "peer will not release lock"); | |
637cd060 ID |
4221 | down_write(&rbd_dev->lock_rwsem); |
4222 | wake_lock_waiters(rbd_dev, ret); | |
4223 | up_write(&rbd_dev->lock_rwsem); | |
ed95b21a ID |
4224 | } else if (ret < 0) { |
4225 | rbd_warn(rbd_dev, "error requesting lock: %d", ret); | |
4226 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
4227 | RBD_RETRY_DELAY); | |
4228 | } else { | |
4229 | /* | |
4230 | * lock owner acked, but resend if we don't see them | |
4231 | * release the lock | |
4232 | */ | |
4233 | dout("%s rbd_dev %p requeueing lock_dwork\n", __func__, | |
4234 | rbd_dev); | |
4235 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
4236 | msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC)); | |
4237 | } | |
4238 | } | |
4239 | ||
a2b1da09 | 4240 | static bool rbd_quiesce_lock(struct rbd_device *rbd_dev) |
ed95b21a | 4241 | { |
e1fddc8f ID |
4242 | bool need_wait; |
4243 | ||
a2b1da09 | 4244 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
d9b9c893 | 4245 | lockdep_assert_held_write(&rbd_dev->lock_rwsem); |
a2b1da09 | 4246 | |
ed95b21a ID |
4247 | if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED) |
4248 | return false; | |
4249 | ||
52bb1f9b | 4250 | /* |
ed95b21a | 4251 | * Ensure that all in-flight IO is flushed. |
52bb1f9b | 4252 | */ |
e1fddc8f ID |
4253 | rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING; |
4254 | rbd_assert(!completion_done(&rbd_dev->releasing_wait)); | |
4255 | need_wait = !list_empty(&rbd_dev->running_list); | |
4256 | downgrade_write(&rbd_dev->lock_rwsem); | |
4257 | if (need_wait) | |
4258 | wait_for_completion(&rbd_dev->releasing_wait); | |
ed95b21a ID |
4259 | up_read(&rbd_dev->lock_rwsem); |
4260 | ||
4261 | down_write(&rbd_dev->lock_rwsem); | |
ed95b21a ID |
4262 | if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING) |
4263 | return false; | |
4264 | ||
e1fddc8f | 4265 | rbd_assert(list_empty(&rbd_dev->running_list)); |
a2b1da09 ID |
4266 | return true; |
4267 | } | |
4268 | ||
22e8bd51 ID |
4269 | static void rbd_pre_release_action(struct rbd_device *rbd_dev) |
4270 | { | |
4271 | if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) | |
4272 | rbd_object_map_close(rbd_dev); | |
4273 | } | |
4274 | ||
e1fddc8f ID |
4275 | static void __rbd_release_lock(struct rbd_device *rbd_dev) |
4276 | { | |
4277 | rbd_assert(list_empty(&rbd_dev->running_list)); | |
4278 | ||
22e8bd51 | 4279 | rbd_pre_release_action(rbd_dev); |
bbead745 | 4280 | rbd_unlock(rbd_dev); |
e1fddc8f ID |
4281 | } |
4282 | ||
a2b1da09 ID |
4283 | /* |
4284 | * lock_rwsem must be held for write | |
4285 | */ | |
4286 | static void rbd_release_lock(struct rbd_device *rbd_dev) | |
4287 | { | |
4288 | if (!rbd_quiesce_lock(rbd_dev)) | |
4289 | return; | |
4290 | ||
e1fddc8f | 4291 | __rbd_release_lock(rbd_dev); |
a2b1da09 | 4292 | |
bbead745 ID |
4293 | /* |
4294 | * Give others a chance to grab the lock - we would re-acquire | |
637cd060 ID |
4295 | * almost immediately if we got new IO while draining the running |
4296 | * list otherwise. We need to ack our own notifications, so this | |
4297 | * lock_dwork will be requeued from rbd_handle_released_lock() by | |
4298 | * way of maybe_kick_acquire(). | |
bbead745 ID |
4299 | */ |
4300 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
ed95b21a ID |
4301 | } |
4302 | ||
4303 | static void rbd_release_lock_work(struct work_struct *work) | |
4304 | { | |
4305 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
4306 | unlock_work); | |
4307 | ||
4308 | down_write(&rbd_dev->lock_rwsem); | |
4309 | rbd_release_lock(rbd_dev); | |
4310 | up_write(&rbd_dev->lock_rwsem); | |
4311 | } | |
4312 | ||
637cd060 ID |
4313 | static void maybe_kick_acquire(struct rbd_device *rbd_dev) |
4314 | { | |
4315 | bool have_requests; | |
4316 | ||
4317 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
4318 | if (__rbd_is_lock_owner(rbd_dev)) | |
4319 | return; | |
4320 | ||
4321 | spin_lock(&rbd_dev->lock_lists_lock); | |
4322 | have_requests = !list_empty(&rbd_dev->acquiring_list); | |
4323 | spin_unlock(&rbd_dev->lock_lists_lock); | |
4324 | if (have_requests || delayed_work_pending(&rbd_dev->lock_dwork)) { | |
4325 | dout("%s rbd_dev %p kicking lock_dwork\n", __func__, rbd_dev); | |
4326 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); | |
4327 | } | |
4328 | } | |
4329 | ||
ed95b21a ID |
4330 | static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v, |
4331 | void **p) | |
4332 | { | |
4333 | struct rbd_client_id cid = { 0 }; | |
4334 | ||
4335 | if (struct_v >= 2) { | |
4336 | cid.gid = ceph_decode_64(p); | |
4337 | cid.handle = ceph_decode_64(p); | |
4338 | } | |
4339 | ||
4340 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
4341 | cid.handle); | |
4342 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
4343 | down_write(&rbd_dev->lock_rwsem); | |
4344 | if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
4345 | /* | |
4346 | * we already know that the remote client is | |
4347 | * the owner | |
4348 | */ | |
4349 | up_write(&rbd_dev->lock_rwsem); | |
4350 | return; | |
4351 | } | |
4352 | ||
4353 | rbd_set_owner_cid(rbd_dev, &cid); | |
4354 | downgrade_write(&rbd_dev->lock_rwsem); | |
4355 | } else { | |
4356 | down_read(&rbd_dev->lock_rwsem); | |
4357 | } | |
4358 | ||
637cd060 | 4359 | maybe_kick_acquire(rbd_dev); |
ed95b21a ID |
4360 | up_read(&rbd_dev->lock_rwsem); |
4361 | } | |
4362 | ||
4363 | static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
4364 | void **p) | |
4365 | { | |
4366 | struct rbd_client_id cid = { 0 }; | |
4367 | ||
4368 | if (struct_v >= 2) { | |
4369 | cid.gid = ceph_decode_64(p); | |
4370 | cid.handle = ceph_decode_64(p); | |
4371 | } | |
4372 | ||
4373 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
4374 | cid.handle); | |
4375 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
4376 | down_write(&rbd_dev->lock_rwsem); | |
4377 | if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
4378 | dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n", | |
4379 | __func__, rbd_dev, cid.gid, cid.handle, | |
4380 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle); | |
4381 | up_write(&rbd_dev->lock_rwsem); | |
4382 | return; | |
4383 | } | |
4384 | ||
4385 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
4386 | downgrade_write(&rbd_dev->lock_rwsem); | |
4387 | } else { | |
4388 | down_read(&rbd_dev->lock_rwsem); | |
4389 | } | |
4390 | ||
637cd060 | 4391 | maybe_kick_acquire(rbd_dev); |
ed95b21a ID |
4392 | up_read(&rbd_dev->lock_rwsem); |
4393 | } | |
4394 | ||
3b77faa0 ID |
4395 | /* |
4396 | * Returns result for ResponseMessage to be encoded (<= 0), or 1 if no | |
4397 | * ResponseMessage is needed. | |
4398 | */ | |
4399 | static int rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
4400 | void **p) | |
ed95b21a ID |
4401 | { |
4402 | struct rbd_client_id my_cid = rbd_get_cid(rbd_dev); | |
4403 | struct rbd_client_id cid = { 0 }; | |
3b77faa0 | 4404 | int result = 1; |
ed95b21a ID |
4405 | |
4406 | if (struct_v >= 2) { | |
4407 | cid.gid = ceph_decode_64(p); | |
4408 | cid.handle = ceph_decode_64(p); | |
4409 | } | |
4410 | ||
4411 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
4412 | cid.handle); | |
4413 | if (rbd_cid_equal(&cid, &my_cid)) | |
3b77faa0 | 4414 | return result; |
ed95b21a ID |
4415 | |
4416 | down_read(&rbd_dev->lock_rwsem); | |
3b77faa0 ID |
4417 | if (__rbd_is_lock_owner(rbd_dev)) { |
4418 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED && | |
4419 | rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid)) | |
4420 | goto out_unlock; | |
4421 | ||
4422 | /* | |
4423 | * encode ResponseMessage(0) so the peer can detect | |
4424 | * a missing owner | |
4425 | */ | |
4426 | result = 0; | |
4427 | ||
4428 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) { | |
e010dd0a ID |
4429 | if (!rbd_dev->opts->exclusive) { |
4430 | dout("%s rbd_dev %p queueing unlock_work\n", | |
4431 | __func__, rbd_dev); | |
4432 | queue_work(rbd_dev->task_wq, | |
4433 | &rbd_dev->unlock_work); | |
4434 | } else { | |
4435 | /* refuse to release the lock */ | |
4436 | result = -EROFS; | |
4437 | } | |
ed95b21a ID |
4438 | } |
4439 | } | |
3b77faa0 ID |
4440 | |
4441 | out_unlock: | |
ed95b21a | 4442 | up_read(&rbd_dev->lock_rwsem); |
3b77faa0 | 4443 | return result; |
ed95b21a ID |
4444 | } |
4445 | ||
4446 | static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev, | |
4447 | u64 notify_id, u64 cookie, s32 *result) | |
4448 | { | |
4449 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
08a79102 KS |
4450 | char buf[4 + CEPH_ENCODING_START_BLK_LEN]; |
4451 | int buf_size = sizeof(buf); | |
ed95b21a ID |
4452 | int ret; |
4453 | ||
4454 | if (result) { | |
4455 | void *p = buf; | |
4456 | ||
4457 | /* encode ResponseMessage */ | |
4458 | ceph_start_encoding(&p, 1, 1, | |
4459 | buf_size - CEPH_ENCODING_START_BLK_LEN); | |
4460 | ceph_encode_32(&p, *result); | |
4461 | } else { | |
4462 | buf_size = 0; | |
4463 | } | |
b8d70035 | 4464 | |
922dab61 ID |
4465 | ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid, |
4466 | &rbd_dev->header_oloc, notify_id, cookie, | |
ed95b21a | 4467 | buf, buf_size); |
52bb1f9b | 4468 | if (ret) |
ed95b21a ID |
4469 | rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret); |
4470 | } | |
4471 | ||
4472 | static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id, | |
4473 | u64 cookie) | |
4474 | { | |
4475 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
4476 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL); | |
4477 | } | |
4478 | ||
4479 | static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev, | |
4480 | u64 notify_id, u64 cookie, s32 result) | |
4481 | { | |
4482 | dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result); | |
4483 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result); | |
4484 | } | |
4485 | ||
4486 | static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie, | |
4487 | u64 notifier_id, void *data, size_t data_len) | |
4488 | { | |
4489 | struct rbd_device *rbd_dev = arg; | |
4490 | void *p = data; | |
4491 | void *const end = p + data_len; | |
d4c2269b | 4492 | u8 struct_v = 0; |
ed95b21a ID |
4493 | u32 len; |
4494 | u32 notify_op; | |
4495 | int ret; | |
4496 | ||
4497 | dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n", | |
4498 | __func__, rbd_dev, cookie, notify_id, data_len); | |
4499 | if (data_len) { | |
4500 | ret = ceph_start_decoding(&p, end, 1, "NotifyMessage", | |
4501 | &struct_v, &len); | |
4502 | if (ret) { | |
4503 | rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d", | |
4504 | ret); | |
4505 | return; | |
4506 | } | |
4507 | ||
4508 | notify_op = ceph_decode_32(&p); | |
4509 | } else { | |
4510 | /* legacy notification for header updates */ | |
4511 | notify_op = RBD_NOTIFY_OP_HEADER_UPDATE; | |
4512 | len = 0; | |
4513 | } | |
4514 | ||
4515 | dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op); | |
4516 | switch (notify_op) { | |
4517 | case RBD_NOTIFY_OP_ACQUIRED_LOCK: | |
4518 | rbd_handle_acquired_lock(rbd_dev, struct_v, &p); | |
4519 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
4520 | break; | |
4521 | case RBD_NOTIFY_OP_RELEASED_LOCK: | |
4522 | rbd_handle_released_lock(rbd_dev, struct_v, &p); | |
4523 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
4524 | break; | |
4525 | case RBD_NOTIFY_OP_REQUEST_LOCK: | |
3b77faa0 ID |
4526 | ret = rbd_handle_request_lock(rbd_dev, struct_v, &p); |
4527 | if (ret <= 0) | |
ed95b21a | 4528 | rbd_acknowledge_notify_result(rbd_dev, notify_id, |
3b77faa0 | 4529 | cookie, ret); |
ed95b21a ID |
4530 | else |
4531 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
4532 | break; | |
4533 | case RBD_NOTIFY_OP_HEADER_UPDATE: | |
4534 | ret = rbd_dev_refresh(rbd_dev); | |
4535 | if (ret) | |
4536 | rbd_warn(rbd_dev, "refresh failed: %d", ret); | |
4537 | ||
4538 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
4539 | break; | |
4540 | default: | |
4541 | if (rbd_is_lock_owner(rbd_dev)) | |
4542 | rbd_acknowledge_notify_result(rbd_dev, notify_id, | |
4543 | cookie, -EOPNOTSUPP); | |
4544 | else | |
4545 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
4546 | break; | |
4547 | } | |
b8d70035 AE |
4548 | } |
4549 | ||
99d16943 ID |
4550 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev); |
4551 | ||
922dab61 | 4552 | static void rbd_watch_errcb(void *arg, u64 cookie, int err) |
bb040aa0 | 4553 | { |
922dab61 | 4554 | struct rbd_device *rbd_dev = arg; |
bb040aa0 | 4555 | |
922dab61 | 4556 | rbd_warn(rbd_dev, "encountered watch error: %d", err); |
bb040aa0 | 4557 | |
ed95b21a ID |
4558 | down_write(&rbd_dev->lock_rwsem); |
4559 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
4560 | up_write(&rbd_dev->lock_rwsem); | |
4561 | ||
99d16943 ID |
4562 | mutex_lock(&rbd_dev->watch_mutex); |
4563 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) { | |
4564 | __rbd_unregister_watch(rbd_dev); | |
4565 | rbd_dev->watch_state = RBD_WATCH_STATE_ERROR; | |
bb040aa0 | 4566 | |
99d16943 | 4567 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0); |
bb040aa0 | 4568 | } |
99d16943 | 4569 | mutex_unlock(&rbd_dev->watch_mutex); |
bb040aa0 ID |
4570 | } |
4571 | ||
9969ebc5 | 4572 | /* |
99d16943 | 4573 | * watch_mutex must be locked |
9969ebc5 | 4574 | */ |
99d16943 | 4575 | static int __rbd_register_watch(struct rbd_device *rbd_dev) |
9969ebc5 AE |
4576 | { |
4577 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
922dab61 | 4578 | struct ceph_osd_linger_request *handle; |
9969ebc5 | 4579 | |
922dab61 | 4580 | rbd_assert(!rbd_dev->watch_handle); |
99d16943 | 4581 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
9969ebc5 | 4582 | |
922dab61 ID |
4583 | handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid, |
4584 | &rbd_dev->header_oloc, rbd_watch_cb, | |
4585 | rbd_watch_errcb, rbd_dev); | |
4586 | if (IS_ERR(handle)) | |
4587 | return PTR_ERR(handle); | |
8eb87565 | 4588 | |
922dab61 | 4589 | rbd_dev->watch_handle = handle; |
b30a01f2 | 4590 | return 0; |
b30a01f2 ID |
4591 | } |
4592 | ||
99d16943 ID |
4593 | /* |
4594 | * watch_mutex must be locked | |
4595 | */ | |
4596 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev) | |
b30a01f2 | 4597 | { |
922dab61 ID |
4598 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4599 | int ret; | |
b30a01f2 | 4600 | |
99d16943 ID |
4601 | rbd_assert(rbd_dev->watch_handle); |
4602 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
b30a01f2 | 4603 | |
922dab61 ID |
4604 | ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle); |
4605 | if (ret) | |
4606 | rbd_warn(rbd_dev, "failed to unwatch: %d", ret); | |
76756a51 | 4607 | |
922dab61 | 4608 | rbd_dev->watch_handle = NULL; |
c525f036 ID |
4609 | } |
4610 | ||
99d16943 ID |
4611 | static int rbd_register_watch(struct rbd_device *rbd_dev) |
4612 | { | |
4613 | int ret; | |
4614 | ||
4615 | mutex_lock(&rbd_dev->watch_mutex); | |
4616 | rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED); | |
4617 | ret = __rbd_register_watch(rbd_dev); | |
4618 | if (ret) | |
4619 | goto out; | |
4620 | ||
4621 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; | |
4622 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
4623 | ||
4624 | out: | |
4625 | mutex_unlock(&rbd_dev->watch_mutex); | |
4626 | return ret; | |
4627 | } | |
4628 | ||
4629 | static void cancel_tasks_sync(struct rbd_device *rbd_dev) | |
c525f036 | 4630 | { |
99d16943 ID |
4631 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
4632 | ||
ed95b21a ID |
4633 | cancel_work_sync(&rbd_dev->acquired_lock_work); |
4634 | cancel_work_sync(&rbd_dev->released_lock_work); | |
4635 | cancel_delayed_work_sync(&rbd_dev->lock_dwork); | |
4636 | cancel_work_sync(&rbd_dev->unlock_work); | |
99d16943 ID |
4637 | } |
4638 | ||
4639 | static void rbd_unregister_watch(struct rbd_device *rbd_dev) | |
4640 | { | |
4641 | cancel_tasks_sync(rbd_dev); | |
4642 | ||
4643 | mutex_lock(&rbd_dev->watch_mutex); | |
4644 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) | |
4645 | __rbd_unregister_watch(rbd_dev); | |
4646 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
4647 | mutex_unlock(&rbd_dev->watch_mutex); | |
811c6688 | 4648 | |
23edca86 | 4649 | cancel_delayed_work_sync(&rbd_dev->watch_dwork); |
811c6688 | 4650 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); |
fca27065 ID |
4651 | } |
4652 | ||
14bb211d ID |
4653 | /* |
4654 | * lock_rwsem must be held for write | |
4655 | */ | |
4656 | static void rbd_reacquire_lock(struct rbd_device *rbd_dev) | |
4657 | { | |
4658 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
4659 | char cookie[32]; | |
4660 | int ret; | |
4661 | ||
a2b1da09 ID |
4662 | if (!rbd_quiesce_lock(rbd_dev)) |
4663 | return; | |
14bb211d ID |
4664 | |
4665 | format_lock_cookie(rbd_dev, cookie); | |
4666 | ret = ceph_cls_set_cookie(osdc, &rbd_dev->header_oid, | |
4667 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
4668 | CEPH_CLS_LOCK_EXCLUSIVE, rbd_dev->lock_cookie, | |
4669 | RBD_LOCK_TAG, cookie); | |
4670 | if (ret) { | |
4671 | if (ret != -EOPNOTSUPP) | |
4672 | rbd_warn(rbd_dev, "failed to update lock cookie: %d", | |
4673 | ret); | |
4674 | ||
4675 | /* | |
4676 | * Lock cookie cannot be updated on older OSDs, so do | |
4677 | * a manual release and queue an acquire. | |
4678 | */ | |
e1fddc8f | 4679 | __rbd_release_lock(rbd_dev); |
a2b1da09 | 4680 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); |
14bb211d | 4681 | } else { |
edd8ca80 | 4682 | __rbd_lock(rbd_dev, cookie); |
637cd060 | 4683 | wake_lock_waiters(rbd_dev, 0); |
14bb211d ID |
4684 | } |
4685 | } | |
4686 | ||
99d16943 ID |
4687 | static void rbd_reregister_watch(struct work_struct *work) |
4688 | { | |
4689 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
4690 | struct rbd_device, watch_dwork); | |
4691 | int ret; | |
4692 | ||
4693 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
4694 | ||
4695 | mutex_lock(&rbd_dev->watch_mutex); | |
87c0fded ID |
4696 | if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) { |
4697 | mutex_unlock(&rbd_dev->watch_mutex); | |
14bb211d | 4698 | return; |
87c0fded | 4699 | } |
99d16943 ID |
4700 | |
4701 | ret = __rbd_register_watch(rbd_dev); | |
4702 | if (ret) { | |
4703 | rbd_warn(rbd_dev, "failed to reregister watch: %d", ret); | |
637cd060 | 4704 | if (ret != -EBLACKLISTED && ret != -ENOENT) { |
99d16943 ID |
4705 | queue_delayed_work(rbd_dev->task_wq, |
4706 | &rbd_dev->watch_dwork, | |
4707 | RBD_RETRY_DELAY); | |
637cd060 ID |
4708 | mutex_unlock(&rbd_dev->watch_mutex); |
4709 | return; | |
87c0fded | 4710 | } |
637cd060 | 4711 | |
87c0fded | 4712 | mutex_unlock(&rbd_dev->watch_mutex); |
637cd060 ID |
4713 | down_write(&rbd_dev->lock_rwsem); |
4714 | wake_lock_waiters(rbd_dev, ret); | |
4715 | up_write(&rbd_dev->lock_rwsem); | |
14bb211d | 4716 | return; |
99d16943 ID |
4717 | } |
4718 | ||
4719 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; | |
4720 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
4721 | mutex_unlock(&rbd_dev->watch_mutex); | |
4722 | ||
14bb211d ID |
4723 | down_write(&rbd_dev->lock_rwsem); |
4724 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) | |
4725 | rbd_reacquire_lock(rbd_dev); | |
4726 | up_write(&rbd_dev->lock_rwsem); | |
4727 | ||
99d16943 ID |
4728 | ret = rbd_dev_refresh(rbd_dev); |
4729 | if (ret) | |
f6870cc9 | 4730 | rbd_warn(rbd_dev, "reregistration refresh failed: %d", ret); |
99d16943 ID |
4731 | } |
4732 | ||
36be9a76 | 4733 | /* |
f40eb349 AE |
4734 | * Synchronous osd object method call. Returns the number of bytes |
4735 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
4736 | */ |
4737 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
ecd4a68a ID |
4738 | struct ceph_object_id *oid, |
4739 | struct ceph_object_locator *oloc, | |
36be9a76 | 4740 | const char *method_name, |
4157976b | 4741 | const void *outbound, |
36be9a76 | 4742 | size_t outbound_size, |
4157976b | 4743 | void *inbound, |
e2a58ee5 | 4744 | size_t inbound_size) |
36be9a76 | 4745 | { |
ecd4a68a ID |
4746 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4747 | struct page *req_page = NULL; | |
4748 | struct page *reply_page; | |
36be9a76 AE |
4749 | int ret; |
4750 | ||
4751 | /* | |
6010a451 AE |
4752 | * Method calls are ultimately read operations. The result |
4753 | * should placed into the inbound buffer provided. They | |
4754 | * also supply outbound data--parameters for the object | |
4755 | * method. Currently if this is present it will be a | |
4756 | * snapshot id. | |
36be9a76 | 4757 | */ |
ecd4a68a ID |
4758 | if (outbound) { |
4759 | if (outbound_size > PAGE_SIZE) | |
4760 | return -E2BIG; | |
36be9a76 | 4761 | |
ecd4a68a ID |
4762 | req_page = alloc_page(GFP_KERNEL); |
4763 | if (!req_page) | |
4764 | return -ENOMEM; | |
04017e29 | 4765 | |
ecd4a68a | 4766 | memcpy(page_address(req_page), outbound, outbound_size); |
04017e29 | 4767 | } |
36be9a76 | 4768 | |
ecd4a68a ID |
4769 | reply_page = alloc_page(GFP_KERNEL); |
4770 | if (!reply_page) { | |
4771 | if (req_page) | |
4772 | __free_page(req_page); | |
4773 | return -ENOMEM; | |
4774 | } | |
57385b51 | 4775 | |
ecd4a68a ID |
4776 | ret = ceph_osdc_call(osdc, oid, oloc, RBD_DRV_NAME, method_name, |
4777 | CEPH_OSD_FLAG_READ, req_page, outbound_size, | |
68ada915 | 4778 | &reply_page, &inbound_size); |
ecd4a68a ID |
4779 | if (!ret) { |
4780 | memcpy(inbound, page_address(reply_page), inbound_size); | |
4781 | ret = inbound_size; | |
4782 | } | |
36be9a76 | 4783 | |
ecd4a68a ID |
4784 | if (req_page) |
4785 | __free_page(req_page); | |
4786 | __free_page(reply_page); | |
36be9a76 AE |
4787 | return ret; |
4788 | } | |
4789 | ||
7ad18afa | 4790 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 4791 | { |
7ad18afa CH |
4792 | struct request *rq = blk_mq_rq_from_pdu(work); |
4793 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 4794 | struct rbd_img_request *img_request; |
4e752f0a | 4795 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
4796 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
4797 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 4798 | enum obj_operation_type op_type; |
4e752f0a | 4799 | u64 mapping_size; |
bf0d5f50 AE |
4800 | int result; |
4801 | ||
aebf526b CH |
4802 | switch (req_op(rq)) { |
4803 | case REQ_OP_DISCARD: | |
90e98c52 | 4804 | op_type = OBJ_OP_DISCARD; |
aebf526b | 4805 | break; |
6484cbe9 ID |
4806 | case REQ_OP_WRITE_ZEROES: |
4807 | op_type = OBJ_OP_ZEROOUT; | |
4808 | break; | |
aebf526b | 4809 | case REQ_OP_WRITE: |
6d2940c8 | 4810 | op_type = OBJ_OP_WRITE; |
aebf526b CH |
4811 | break; |
4812 | case REQ_OP_READ: | |
6d2940c8 | 4813 | op_type = OBJ_OP_READ; |
aebf526b CH |
4814 | break; |
4815 | default: | |
4816 | dout("%s: non-fs request type %d\n", __func__, req_op(rq)); | |
4817 | result = -EIO; | |
4818 | goto err; | |
4819 | } | |
6d2940c8 | 4820 | |
bc1ecc65 | 4821 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 4822 | |
bc1ecc65 ID |
4823 | if (!length) { |
4824 | dout("%s: zero-length request\n", __func__); | |
4825 | result = 0; | |
4826 | goto err_rq; | |
4827 | } | |
bf0d5f50 | 4828 | |
b91a7bdc ID |
4829 | if (op_type != OBJ_OP_READ && rbd_dev->spec->snap_id != CEPH_NOSNAP) { |
4830 | rbd_warn(rbd_dev, "%s on read-only snapshot", | |
4831 | obj_op_name(op_type)); | |
4832 | result = -EIO; | |
4833 | goto err; | |
4834 | } | |
4dda41d3 | 4835 | |
bc1ecc65 ID |
4836 | /* |
4837 | * Quit early if the mapped snapshot no longer exists. It's | |
4838 | * still possible the snapshot will have disappeared by the | |
4839 | * time our request arrives at the osd, but there's no sense in | |
4840 | * sending it if we already know. | |
4841 | */ | |
4842 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
4843 | dout("request for non-existent snapshot"); | |
4844 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
4845 | result = -ENXIO; | |
4846 | goto err_rq; | |
4847 | } | |
4dda41d3 | 4848 | |
bc1ecc65 ID |
4849 | if (offset && length > U64_MAX - offset + 1) { |
4850 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
4851 | length); | |
4852 | result = -EINVAL; | |
4853 | goto err_rq; /* Shouldn't happen */ | |
4854 | } | |
4dda41d3 | 4855 | |
7ad18afa CH |
4856 | blk_mq_start_request(rq); |
4857 | ||
4e752f0a JD |
4858 | down_read(&rbd_dev->header_rwsem); |
4859 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 4860 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
4861 | snapc = rbd_dev->header.snapc; |
4862 | ceph_get_snap_context(snapc); | |
4863 | } | |
4864 | up_read(&rbd_dev->header_rwsem); | |
4865 | ||
4866 | if (offset + length > mapping_size) { | |
bc1ecc65 | 4867 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 4868 | length, mapping_size); |
bc1ecc65 ID |
4869 | result = -EIO; |
4870 | goto err_rq; | |
4871 | } | |
bf0d5f50 | 4872 | |
dfd9875f | 4873 | img_request = rbd_img_request_create(rbd_dev, op_type, snapc); |
bc1ecc65 ID |
4874 | if (!img_request) { |
4875 | result = -ENOMEM; | |
637cd060 | 4876 | goto err_rq; |
bc1ecc65 ID |
4877 | } |
4878 | img_request->rq = rq; | |
70b16db8 | 4879 | snapc = NULL; /* img_request consumes a ref */ |
bf0d5f50 | 4880 | |
21ed05a8 ID |
4881 | dout("%s rbd_dev %p img_req %p %s %llu~%llu\n", __func__, rbd_dev, |
4882 | img_request, obj_op_name(op_type), offset, length); | |
4883 | ||
6484cbe9 | 4884 | if (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_ZEROOUT) |
5a237819 | 4885 | result = rbd_img_fill_nodata(img_request, offset, length); |
90e98c52 | 4886 | else |
5a237819 ID |
4887 | result = rbd_img_fill_from_bio(img_request, offset, length, |
4888 | rq->bio); | |
0192ce2e | 4889 | if (result) |
bc1ecc65 | 4890 | goto err_img_request; |
bf0d5f50 | 4891 | |
e1fddc8f | 4892 | rbd_img_handle_request(img_request, 0); |
bc1ecc65 | 4893 | return; |
bf0d5f50 | 4894 | |
bc1ecc65 ID |
4895 | err_img_request: |
4896 | rbd_img_request_put(img_request); | |
4897 | err_rq: | |
4898 | if (result) | |
4899 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 4900 | obj_op_name(op_type), length, offset, result); |
e96a650a | 4901 | ceph_put_snap_context(snapc); |
7ad18afa | 4902 | err: |
2a842aca | 4903 | blk_mq_end_request(rq, errno_to_blk_status(result)); |
bc1ecc65 | 4904 | } |
bf0d5f50 | 4905 | |
fc17b653 | 4906 | static blk_status_t rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
7ad18afa | 4907 | const struct blk_mq_queue_data *bd) |
bc1ecc65 | 4908 | { |
7ad18afa CH |
4909 | struct request *rq = bd->rq; |
4910 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 4911 | |
7ad18afa | 4912 | queue_work(rbd_wq, work); |
fc17b653 | 4913 | return BLK_STS_OK; |
bf0d5f50 AE |
4914 | } |
4915 | ||
602adf40 YS |
4916 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
4917 | { | |
5769ed0c ID |
4918 | blk_cleanup_queue(rbd_dev->disk->queue); |
4919 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
4920 | put_disk(rbd_dev->disk); | |
a0cab924 | 4921 | rbd_dev->disk = NULL; |
602adf40 YS |
4922 | } |
4923 | ||
788e2df3 | 4924 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
fe5478e0 ID |
4925 | struct ceph_object_id *oid, |
4926 | struct ceph_object_locator *oloc, | |
4927 | void *buf, int buf_len) | |
788e2df3 AE |
4928 | |
4929 | { | |
fe5478e0 ID |
4930 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4931 | struct ceph_osd_request *req; | |
4932 | struct page **pages; | |
4933 | int num_pages = calc_pages_for(0, buf_len); | |
788e2df3 AE |
4934 | int ret; |
4935 | ||
fe5478e0 ID |
4936 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL); |
4937 | if (!req) | |
4938 | return -ENOMEM; | |
788e2df3 | 4939 | |
fe5478e0 ID |
4940 | ceph_oid_copy(&req->r_base_oid, oid); |
4941 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
4942 | req->r_flags = CEPH_OSD_FLAG_READ; | |
430c28c3 | 4943 | |
fe5478e0 ID |
4944 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
4945 | if (IS_ERR(pages)) { | |
4946 | ret = PTR_ERR(pages); | |
4947 | goto out_req; | |
4948 | } | |
1ceae7ef | 4949 | |
fe5478e0 ID |
4950 | osd_req_op_extent_init(req, 0, CEPH_OSD_OP_READ, 0, buf_len, 0, 0); |
4951 | osd_req_op_extent_osd_data_pages(req, 0, pages, buf_len, 0, false, | |
4952 | true); | |
4953 | ||
26f887e0 ID |
4954 | ret = ceph_osdc_alloc_messages(req, GFP_KERNEL); |
4955 | if (ret) | |
4956 | goto out_req; | |
4957 | ||
fe5478e0 ID |
4958 | ceph_osdc_start_request(osdc, req, false); |
4959 | ret = ceph_osdc_wait_request(osdc, req); | |
4960 | if (ret >= 0) | |
4961 | ceph_copy_from_page_vector(pages, buf, 0, ret); | |
788e2df3 | 4962 | |
fe5478e0 ID |
4963 | out_req: |
4964 | ceph_osdc_put_request(req); | |
788e2df3 AE |
4965 | return ret; |
4966 | } | |
4967 | ||
602adf40 | 4968 | /* |
662518b1 AE |
4969 | * Read the complete header for the given rbd device. On successful |
4970 | * return, the rbd_dev->header field will contain up-to-date | |
4971 | * information about the image. | |
602adf40 | 4972 | */ |
99a41ebc | 4973 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 4974 | { |
4156d998 | 4975 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 4976 | u32 snap_count = 0; |
4156d998 AE |
4977 | u64 names_size = 0; |
4978 | u32 want_count; | |
4979 | int ret; | |
602adf40 | 4980 | |
00f1f36f | 4981 | /* |
4156d998 AE |
4982 | * The complete header will include an array of its 64-bit |
4983 | * snapshot ids, followed by the names of those snapshots as | |
4984 | * a contiguous block of NUL-terminated strings. Note that | |
4985 | * the number of snapshots could change by the time we read | |
4986 | * it in, in which case we re-read it. | |
00f1f36f | 4987 | */ |
4156d998 AE |
4988 | do { |
4989 | size_t size; | |
4990 | ||
4991 | kfree(ondisk); | |
4992 | ||
4993 | size = sizeof (*ondisk); | |
4994 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
4995 | size += names_size; | |
4996 | ondisk = kmalloc(size, GFP_KERNEL); | |
4997 | if (!ondisk) | |
662518b1 | 4998 | return -ENOMEM; |
4156d998 | 4999 | |
fe5478e0 ID |
5000 | ret = rbd_obj_read_sync(rbd_dev, &rbd_dev->header_oid, |
5001 | &rbd_dev->header_oloc, ondisk, size); | |
4156d998 | 5002 | if (ret < 0) |
662518b1 | 5003 | goto out; |
c0cd10db | 5004 | if ((size_t)ret < size) { |
4156d998 | 5005 | ret = -ENXIO; |
06ecc6cb AE |
5006 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
5007 | size, ret); | |
662518b1 | 5008 | goto out; |
4156d998 AE |
5009 | } |
5010 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
5011 | ret = -ENXIO; | |
06ecc6cb | 5012 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 5013 | goto out; |
81e759fb | 5014 | } |
602adf40 | 5015 | |
4156d998 AE |
5016 | names_size = le64_to_cpu(ondisk->snap_names_len); |
5017 | want_count = snap_count; | |
5018 | snap_count = le32_to_cpu(ondisk->snap_count); | |
5019 | } while (snap_count != want_count); | |
00f1f36f | 5020 | |
662518b1 AE |
5021 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
5022 | out: | |
4156d998 AE |
5023 | kfree(ondisk); |
5024 | ||
5025 | return ret; | |
602adf40 YS |
5026 | } |
5027 | ||
15228ede AE |
5028 | /* |
5029 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
5030 | * has disappeared from the (just updated) snapshot context. | |
5031 | */ | |
5032 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
5033 | { | |
5034 | u64 snap_id; | |
5035 | ||
5036 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
5037 | return; | |
5038 | ||
5039 | snap_id = rbd_dev->spec->snap_id; | |
5040 | if (snap_id == CEPH_NOSNAP) | |
5041 | return; | |
5042 | ||
5043 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
5044 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
5045 | } | |
5046 | ||
9875201e JD |
5047 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
5048 | { | |
5049 | sector_t size; | |
9875201e JD |
5050 | |
5051 | /* | |
811c6688 ID |
5052 | * If EXISTS is not set, rbd_dev->disk may be NULL, so don't |
5053 | * try to update its size. If REMOVING is set, updating size | |
5054 | * is just useless work since the device can't be opened. | |
9875201e | 5055 | */ |
811c6688 ID |
5056 | if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) && |
5057 | !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) { | |
9875201e JD |
5058 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; |
5059 | dout("setting size to %llu sectors", (unsigned long long)size); | |
5060 | set_capacity(rbd_dev->disk, size); | |
5061 | revalidate_disk(rbd_dev->disk); | |
5062 | } | |
5063 | } | |
5064 | ||
cc4a38bd | 5065 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 5066 | { |
e627db08 | 5067 | u64 mapping_size; |
1fe5e993 AE |
5068 | int ret; |
5069 | ||
cfbf6377 | 5070 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 5071 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
5072 | |
5073 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 5074 | if (ret) |
73e39e4d | 5075 | goto out; |
15228ede | 5076 | |
e8f59b59 ID |
5077 | /* |
5078 | * If there is a parent, see if it has disappeared due to the | |
5079 | * mapped image getting flattened. | |
5080 | */ | |
5081 | if (rbd_dev->parent) { | |
5082 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5083 | if (ret) | |
73e39e4d | 5084 | goto out; |
e8f59b59 ID |
5085 | } |
5086 | ||
5ff1108c | 5087 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 5088 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
5089 | } else { |
5090 | /* validate mapped snapshot's EXISTS flag */ | |
5091 | rbd_exists_validate(rbd_dev); | |
5092 | } | |
15228ede | 5093 | |
73e39e4d | 5094 | out: |
cfbf6377 | 5095 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 5096 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 5097 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 5098 | |
73e39e4d | 5099 | return ret; |
1fe5e993 AE |
5100 | } |
5101 | ||
d6296d39 CH |
5102 | static int rbd_init_request(struct blk_mq_tag_set *set, struct request *rq, |
5103 | unsigned int hctx_idx, unsigned int numa_node) | |
7ad18afa CH |
5104 | { |
5105 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
5106 | ||
5107 | INIT_WORK(work, rbd_queue_workfn); | |
5108 | return 0; | |
5109 | } | |
5110 | ||
f363b089 | 5111 | static const struct blk_mq_ops rbd_mq_ops = { |
7ad18afa | 5112 | .queue_rq = rbd_queue_rq, |
7ad18afa CH |
5113 | .init_request = rbd_init_request, |
5114 | }; | |
5115 | ||
602adf40 YS |
5116 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
5117 | { | |
5118 | struct gendisk *disk; | |
5119 | struct request_queue *q; | |
420efbdf ID |
5120 | unsigned int objset_bytes = |
5121 | rbd_dev->layout.object_size * rbd_dev->layout.stripe_count; | |
7ad18afa | 5122 | int err; |
602adf40 | 5123 | |
602adf40 | 5124 | /* create gendisk info */ |
7e513d43 ID |
5125 | disk = alloc_disk(single_major ? |
5126 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
5127 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 5128 | if (!disk) |
1fcdb8aa | 5129 | return -ENOMEM; |
602adf40 | 5130 | |
f0f8cef5 | 5131 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 5132 | rbd_dev->dev_id); |
602adf40 | 5133 | disk->major = rbd_dev->major; |
dd82fff1 | 5134 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
5135 | if (single_major) |
5136 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
5137 | disk->fops = &rbd_bd_ops; |
5138 | disk->private_data = rbd_dev; | |
5139 | ||
7ad18afa CH |
5140 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
5141 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 5142 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 5143 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
56d18f62 | 5144 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; |
7ad18afa CH |
5145 | rbd_dev->tag_set.nr_hw_queues = 1; |
5146 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
5147 | ||
5148 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
5149 | if (err) | |
602adf40 | 5150 | goto out_disk; |
029bcbd8 | 5151 | |
7ad18afa CH |
5152 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
5153 | if (IS_ERR(q)) { | |
5154 | err = PTR_ERR(q); | |
5155 | goto out_tag_set; | |
5156 | } | |
5157 | ||
8b904b5b | 5158 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q); |
d8a2c89c | 5159 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ |
593a9e7b | 5160 | |
420efbdf | 5161 | blk_queue_max_hw_sectors(q, objset_bytes >> SECTOR_SHIFT); |
0d9fde4f | 5162 | q->limits.max_sectors = queue_max_hw_sectors(q); |
21acdf45 | 5163 | blk_queue_max_segments(q, USHRT_MAX); |
24f1df60 | 5164 | blk_queue_max_segment_size(q, UINT_MAX); |
16d80c54 ID |
5165 | blk_queue_io_min(q, rbd_dev->opts->alloc_size); |
5166 | blk_queue_io_opt(q, rbd_dev->opts->alloc_size); | |
029bcbd8 | 5167 | |
d9360540 ID |
5168 | if (rbd_dev->opts->trim) { |
5169 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); | |
16d80c54 | 5170 | q->limits.discard_granularity = rbd_dev->opts->alloc_size; |
d9360540 ID |
5171 | blk_queue_max_discard_sectors(q, objset_bytes >> SECTOR_SHIFT); |
5172 | blk_queue_max_write_zeroes_sectors(q, objset_bytes >> SECTOR_SHIFT); | |
5173 | } | |
90e98c52 | 5174 | |
bae818ee | 5175 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
dc3b17cc | 5176 | q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES; |
bae818ee | 5177 | |
5769ed0c ID |
5178 | /* |
5179 | * disk_release() expects a queue ref from add_disk() and will | |
5180 | * put it. Hold an extra ref until add_disk() is called. | |
5181 | */ | |
5182 | WARN_ON(!blk_get_queue(q)); | |
602adf40 | 5183 | disk->queue = q; |
602adf40 YS |
5184 | q->queuedata = rbd_dev; |
5185 | ||
5186 | rbd_dev->disk = disk; | |
602adf40 | 5187 | |
602adf40 | 5188 | return 0; |
7ad18afa CH |
5189 | out_tag_set: |
5190 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
5191 | out_disk: |
5192 | put_disk(disk); | |
7ad18afa | 5193 | return err; |
602adf40 YS |
5194 | } |
5195 | ||
dfc5606d YS |
5196 | /* |
5197 | sysfs | |
5198 | */ | |
5199 | ||
593a9e7b AE |
5200 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
5201 | { | |
5202 | return container_of(dev, struct rbd_device, dev); | |
5203 | } | |
5204 | ||
dfc5606d YS |
5205 | static ssize_t rbd_size_show(struct device *dev, |
5206 | struct device_attribute *attr, char *buf) | |
5207 | { | |
593a9e7b | 5208 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 5209 | |
fc71d833 AE |
5210 | return sprintf(buf, "%llu\n", |
5211 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
5212 | } |
5213 | ||
34b13184 AE |
5214 | /* |
5215 | * Note this shows the features for whatever's mapped, which is not | |
5216 | * necessarily the base image. | |
5217 | */ | |
5218 | static ssize_t rbd_features_show(struct device *dev, | |
5219 | struct device_attribute *attr, char *buf) | |
5220 | { | |
5221 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5222 | ||
5223 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 5224 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
5225 | } |
5226 | ||
dfc5606d YS |
5227 | static ssize_t rbd_major_show(struct device *dev, |
5228 | struct device_attribute *attr, char *buf) | |
5229 | { | |
593a9e7b | 5230 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5231 | |
fc71d833 AE |
5232 | if (rbd_dev->major) |
5233 | return sprintf(buf, "%d\n", rbd_dev->major); | |
5234 | ||
5235 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
5236 | } |
5237 | ||
5238 | static ssize_t rbd_minor_show(struct device *dev, | |
5239 | struct device_attribute *attr, char *buf) | |
5240 | { | |
5241 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 5242 | |
dd82fff1 | 5243 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
5244 | } |
5245 | ||
005a07bf ID |
5246 | static ssize_t rbd_client_addr_show(struct device *dev, |
5247 | struct device_attribute *attr, char *buf) | |
5248 | { | |
5249 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5250 | struct ceph_entity_addr *client_addr = | |
5251 | ceph_client_addr(rbd_dev->rbd_client->client); | |
5252 | ||
5253 | return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr, | |
5254 | le32_to_cpu(client_addr->nonce)); | |
5255 | } | |
5256 | ||
dfc5606d YS |
5257 | static ssize_t rbd_client_id_show(struct device *dev, |
5258 | struct device_attribute *attr, char *buf) | |
602adf40 | 5259 | { |
593a9e7b | 5260 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 5261 | |
1dbb4399 | 5262 | return sprintf(buf, "client%lld\n", |
033268a5 | 5263 | ceph_client_gid(rbd_dev->rbd_client->client)); |
602adf40 YS |
5264 | } |
5265 | ||
267fb90b MC |
5266 | static ssize_t rbd_cluster_fsid_show(struct device *dev, |
5267 | struct device_attribute *attr, char *buf) | |
5268 | { | |
5269 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5270 | ||
5271 | return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid); | |
5272 | } | |
5273 | ||
0d6d1e9c MC |
5274 | static ssize_t rbd_config_info_show(struct device *dev, |
5275 | struct device_attribute *attr, char *buf) | |
5276 | { | |
5277 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5278 | ||
5279 | return sprintf(buf, "%s\n", rbd_dev->config_info); | |
602adf40 YS |
5280 | } |
5281 | ||
dfc5606d YS |
5282 | static ssize_t rbd_pool_show(struct device *dev, |
5283 | struct device_attribute *attr, char *buf) | |
602adf40 | 5284 | { |
593a9e7b | 5285 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 5286 | |
0d7dbfce | 5287 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
5288 | } |
5289 | ||
9bb2f334 AE |
5290 | static ssize_t rbd_pool_id_show(struct device *dev, |
5291 | struct device_attribute *attr, char *buf) | |
5292 | { | |
5293 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5294 | ||
0d7dbfce | 5295 | return sprintf(buf, "%llu\n", |
fc71d833 | 5296 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
5297 | } |
5298 | ||
b26c047b ID |
5299 | static ssize_t rbd_pool_ns_show(struct device *dev, |
5300 | struct device_attribute *attr, char *buf) | |
5301 | { | |
5302 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5303 | ||
5304 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_ns ?: ""); | |
5305 | } | |
5306 | ||
dfc5606d YS |
5307 | static ssize_t rbd_name_show(struct device *dev, |
5308 | struct device_attribute *attr, char *buf) | |
5309 | { | |
593a9e7b | 5310 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 5311 | |
a92ffdf8 AE |
5312 | if (rbd_dev->spec->image_name) |
5313 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
5314 | ||
5315 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
5316 | } |
5317 | ||
589d30e0 AE |
5318 | static ssize_t rbd_image_id_show(struct device *dev, |
5319 | struct device_attribute *attr, char *buf) | |
5320 | { | |
5321 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5322 | ||
0d7dbfce | 5323 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
5324 | } |
5325 | ||
34b13184 AE |
5326 | /* |
5327 | * Shows the name of the currently-mapped snapshot (or | |
5328 | * RBD_SNAP_HEAD_NAME for the base image). | |
5329 | */ | |
dfc5606d YS |
5330 | static ssize_t rbd_snap_show(struct device *dev, |
5331 | struct device_attribute *attr, | |
5332 | char *buf) | |
5333 | { | |
593a9e7b | 5334 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 5335 | |
0d7dbfce | 5336 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
5337 | } |
5338 | ||
92a58671 MC |
5339 | static ssize_t rbd_snap_id_show(struct device *dev, |
5340 | struct device_attribute *attr, char *buf) | |
5341 | { | |
5342 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5343 | ||
5344 | return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id); | |
5345 | } | |
5346 | ||
86b00e0d | 5347 | /* |
ff96128f ID |
5348 | * For a v2 image, shows the chain of parent images, separated by empty |
5349 | * lines. For v1 images or if there is no parent, shows "(no parent | |
5350 | * image)". | |
86b00e0d AE |
5351 | */ |
5352 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
5353 | struct device_attribute *attr, |
5354 | char *buf) | |
86b00e0d AE |
5355 | { |
5356 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 5357 | ssize_t count = 0; |
86b00e0d | 5358 | |
ff96128f | 5359 | if (!rbd_dev->parent) |
86b00e0d AE |
5360 | return sprintf(buf, "(no parent image)\n"); |
5361 | ||
ff96128f ID |
5362 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
5363 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
5364 | ||
5365 | count += sprintf(&buf[count], "%s" | |
5366 | "pool_id %llu\npool_name %s\n" | |
e92c0eaf | 5367 | "pool_ns %s\n" |
ff96128f ID |
5368 | "image_id %s\nimage_name %s\n" |
5369 | "snap_id %llu\nsnap_name %s\n" | |
5370 | "overlap %llu\n", | |
5371 | !count ? "" : "\n", /* first? */ | |
5372 | spec->pool_id, spec->pool_name, | |
e92c0eaf | 5373 | spec->pool_ns ?: "", |
ff96128f ID |
5374 | spec->image_id, spec->image_name ?: "(unknown)", |
5375 | spec->snap_id, spec->snap_name, | |
5376 | rbd_dev->parent_overlap); | |
5377 | } | |
5378 | ||
5379 | return count; | |
86b00e0d AE |
5380 | } |
5381 | ||
dfc5606d YS |
5382 | static ssize_t rbd_image_refresh(struct device *dev, |
5383 | struct device_attribute *attr, | |
5384 | const char *buf, | |
5385 | size_t size) | |
5386 | { | |
593a9e7b | 5387 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 5388 | int ret; |
602adf40 | 5389 | |
cc4a38bd | 5390 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 5391 | if (ret) |
52bb1f9b | 5392 | return ret; |
b813623a | 5393 | |
52bb1f9b | 5394 | return size; |
dfc5606d | 5395 | } |
602adf40 | 5396 | |
5657a819 JP |
5397 | static DEVICE_ATTR(size, 0444, rbd_size_show, NULL); |
5398 | static DEVICE_ATTR(features, 0444, rbd_features_show, NULL); | |
5399 | static DEVICE_ATTR(major, 0444, rbd_major_show, NULL); | |
5400 | static DEVICE_ATTR(minor, 0444, rbd_minor_show, NULL); | |
5401 | static DEVICE_ATTR(client_addr, 0444, rbd_client_addr_show, NULL); | |
5402 | static DEVICE_ATTR(client_id, 0444, rbd_client_id_show, NULL); | |
5403 | static DEVICE_ATTR(cluster_fsid, 0444, rbd_cluster_fsid_show, NULL); | |
5404 | static DEVICE_ATTR(config_info, 0400, rbd_config_info_show, NULL); | |
5405 | static DEVICE_ATTR(pool, 0444, rbd_pool_show, NULL); | |
5406 | static DEVICE_ATTR(pool_id, 0444, rbd_pool_id_show, NULL); | |
b26c047b | 5407 | static DEVICE_ATTR(pool_ns, 0444, rbd_pool_ns_show, NULL); |
5657a819 JP |
5408 | static DEVICE_ATTR(name, 0444, rbd_name_show, NULL); |
5409 | static DEVICE_ATTR(image_id, 0444, rbd_image_id_show, NULL); | |
5410 | static DEVICE_ATTR(refresh, 0200, NULL, rbd_image_refresh); | |
5411 | static DEVICE_ATTR(current_snap, 0444, rbd_snap_show, NULL); | |
5412 | static DEVICE_ATTR(snap_id, 0444, rbd_snap_id_show, NULL); | |
5413 | static DEVICE_ATTR(parent, 0444, rbd_parent_show, NULL); | |
dfc5606d YS |
5414 | |
5415 | static struct attribute *rbd_attrs[] = { | |
5416 | &dev_attr_size.attr, | |
34b13184 | 5417 | &dev_attr_features.attr, |
dfc5606d | 5418 | &dev_attr_major.attr, |
dd82fff1 | 5419 | &dev_attr_minor.attr, |
005a07bf | 5420 | &dev_attr_client_addr.attr, |
dfc5606d | 5421 | &dev_attr_client_id.attr, |
267fb90b | 5422 | &dev_attr_cluster_fsid.attr, |
0d6d1e9c | 5423 | &dev_attr_config_info.attr, |
dfc5606d | 5424 | &dev_attr_pool.attr, |
9bb2f334 | 5425 | &dev_attr_pool_id.attr, |
b26c047b | 5426 | &dev_attr_pool_ns.attr, |
dfc5606d | 5427 | &dev_attr_name.attr, |
589d30e0 | 5428 | &dev_attr_image_id.attr, |
dfc5606d | 5429 | &dev_attr_current_snap.attr, |
92a58671 | 5430 | &dev_attr_snap_id.attr, |
86b00e0d | 5431 | &dev_attr_parent.attr, |
dfc5606d | 5432 | &dev_attr_refresh.attr, |
dfc5606d YS |
5433 | NULL |
5434 | }; | |
5435 | ||
5436 | static struct attribute_group rbd_attr_group = { | |
5437 | .attrs = rbd_attrs, | |
5438 | }; | |
5439 | ||
5440 | static const struct attribute_group *rbd_attr_groups[] = { | |
5441 | &rbd_attr_group, | |
5442 | NULL | |
5443 | }; | |
5444 | ||
6cac4695 | 5445 | static void rbd_dev_release(struct device *dev); |
dfc5606d | 5446 | |
b9942bc9 | 5447 | static const struct device_type rbd_device_type = { |
dfc5606d YS |
5448 | .name = "rbd", |
5449 | .groups = rbd_attr_groups, | |
6cac4695 | 5450 | .release = rbd_dev_release, |
dfc5606d YS |
5451 | }; |
5452 | ||
8b8fb99c AE |
5453 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
5454 | { | |
5455 | kref_get(&spec->kref); | |
5456 | ||
5457 | return spec; | |
5458 | } | |
5459 | ||
5460 | static void rbd_spec_free(struct kref *kref); | |
5461 | static void rbd_spec_put(struct rbd_spec *spec) | |
5462 | { | |
5463 | if (spec) | |
5464 | kref_put(&spec->kref, rbd_spec_free); | |
5465 | } | |
5466 | ||
5467 | static struct rbd_spec *rbd_spec_alloc(void) | |
5468 | { | |
5469 | struct rbd_spec *spec; | |
5470 | ||
5471 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
5472 | if (!spec) | |
5473 | return NULL; | |
04077599 ID |
5474 | |
5475 | spec->pool_id = CEPH_NOPOOL; | |
5476 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
5477 | kref_init(&spec->kref); |
5478 | ||
8b8fb99c AE |
5479 | return spec; |
5480 | } | |
5481 | ||
5482 | static void rbd_spec_free(struct kref *kref) | |
5483 | { | |
5484 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
5485 | ||
5486 | kfree(spec->pool_name); | |
b26c047b | 5487 | kfree(spec->pool_ns); |
8b8fb99c AE |
5488 | kfree(spec->image_id); |
5489 | kfree(spec->image_name); | |
5490 | kfree(spec->snap_name); | |
5491 | kfree(spec); | |
5492 | } | |
5493 | ||
1643dfa4 | 5494 | static void rbd_dev_free(struct rbd_device *rbd_dev) |
dd5ac32d | 5495 | { |
99d16943 | 5496 | WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED); |
ed95b21a | 5497 | WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED); |
dd5ac32d | 5498 | |
c41d13a3 | 5499 | ceph_oid_destroy(&rbd_dev->header_oid); |
6b6dddbe | 5500 | ceph_oloc_destroy(&rbd_dev->header_oloc); |
0d6d1e9c | 5501 | kfree(rbd_dev->config_info); |
c41d13a3 | 5502 | |
dd5ac32d ID |
5503 | rbd_put_client(rbd_dev->rbd_client); |
5504 | rbd_spec_put(rbd_dev->spec); | |
5505 | kfree(rbd_dev->opts); | |
5506 | kfree(rbd_dev); | |
1643dfa4 ID |
5507 | } |
5508 | ||
5509 | static void rbd_dev_release(struct device *dev) | |
5510 | { | |
5511 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
5512 | bool need_put = !!rbd_dev->opts; | |
5513 | ||
5514 | if (need_put) { | |
5515 | destroy_workqueue(rbd_dev->task_wq); | |
5516 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
5517 | } | |
5518 | ||
5519 | rbd_dev_free(rbd_dev); | |
dd5ac32d ID |
5520 | |
5521 | /* | |
5522 | * This is racy, but way better than putting module outside of | |
5523 | * the release callback. The race window is pretty small, so | |
5524 | * doing something similar to dm (dm-builtin.c) is overkill. | |
5525 | */ | |
5526 | if (need_put) | |
5527 | module_put(THIS_MODULE); | |
5528 | } | |
5529 | ||
1643dfa4 ID |
5530 | static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc, |
5531 | struct rbd_spec *spec) | |
c53d5893 AE |
5532 | { |
5533 | struct rbd_device *rbd_dev; | |
5534 | ||
1643dfa4 | 5535 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
c53d5893 AE |
5536 | if (!rbd_dev) |
5537 | return NULL; | |
5538 | ||
5539 | spin_lock_init(&rbd_dev->lock); | |
5540 | INIT_LIST_HEAD(&rbd_dev->node); | |
c53d5893 AE |
5541 | init_rwsem(&rbd_dev->header_rwsem); |
5542 | ||
7e97332e | 5543 | rbd_dev->header.data_pool_id = CEPH_NOPOOL; |
c41d13a3 | 5544 | ceph_oid_init(&rbd_dev->header_oid); |
431a02cd | 5545 | rbd_dev->header_oloc.pool = spec->pool_id; |
b26c047b ID |
5546 | if (spec->pool_ns) { |
5547 | WARN_ON(!*spec->pool_ns); | |
5548 | rbd_dev->header_oloc.pool_ns = | |
5549 | ceph_find_or_create_string(spec->pool_ns, | |
5550 | strlen(spec->pool_ns)); | |
5551 | } | |
c41d13a3 | 5552 | |
99d16943 ID |
5553 | mutex_init(&rbd_dev->watch_mutex); |
5554 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
5555 | INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch); | |
5556 | ||
ed95b21a ID |
5557 | init_rwsem(&rbd_dev->lock_rwsem); |
5558 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; | |
5559 | INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock); | |
5560 | INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock); | |
5561 | INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock); | |
5562 | INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work); | |
e1fddc8f | 5563 | spin_lock_init(&rbd_dev->lock_lists_lock); |
637cd060 | 5564 | INIT_LIST_HEAD(&rbd_dev->acquiring_list); |
e1fddc8f | 5565 | INIT_LIST_HEAD(&rbd_dev->running_list); |
637cd060 | 5566 | init_completion(&rbd_dev->acquire_wait); |
e1fddc8f | 5567 | init_completion(&rbd_dev->releasing_wait); |
ed95b21a | 5568 | |
22e8bd51 | 5569 | spin_lock_init(&rbd_dev->object_map_lock); |
ed95b21a | 5570 | |
dd5ac32d ID |
5571 | rbd_dev->dev.bus = &rbd_bus_type; |
5572 | rbd_dev->dev.type = &rbd_device_type; | |
5573 | rbd_dev->dev.parent = &rbd_root_dev; | |
dd5ac32d ID |
5574 | device_initialize(&rbd_dev->dev); |
5575 | ||
c53d5893 | 5576 | rbd_dev->rbd_client = rbdc; |
d147543d | 5577 | rbd_dev->spec = spec; |
0903e875 | 5578 | |
1643dfa4 ID |
5579 | return rbd_dev; |
5580 | } | |
5581 | ||
5582 | /* | |
5583 | * Create a mapping rbd_dev. | |
5584 | */ | |
5585 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, | |
5586 | struct rbd_spec *spec, | |
5587 | struct rbd_options *opts) | |
5588 | { | |
5589 | struct rbd_device *rbd_dev; | |
5590 | ||
5591 | rbd_dev = __rbd_dev_create(rbdc, spec); | |
5592 | if (!rbd_dev) | |
5593 | return NULL; | |
5594 | ||
5595 | rbd_dev->opts = opts; | |
5596 | ||
5597 | /* get an id and fill in device name */ | |
5598 | rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0, | |
5599 | minor_to_rbd_dev_id(1 << MINORBITS), | |
5600 | GFP_KERNEL); | |
5601 | if (rbd_dev->dev_id < 0) | |
5602 | goto fail_rbd_dev; | |
5603 | ||
5604 | sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id); | |
5605 | rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM, | |
5606 | rbd_dev->name); | |
5607 | if (!rbd_dev->task_wq) | |
5608 | goto fail_dev_id; | |
dd5ac32d | 5609 | |
1643dfa4 ID |
5610 | /* we have a ref from do_rbd_add() */ |
5611 | __module_get(THIS_MODULE); | |
dd5ac32d | 5612 | |
1643dfa4 | 5613 | dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id); |
c53d5893 | 5614 | return rbd_dev; |
1643dfa4 ID |
5615 | |
5616 | fail_dev_id: | |
5617 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
5618 | fail_rbd_dev: | |
5619 | rbd_dev_free(rbd_dev); | |
5620 | return NULL; | |
c53d5893 AE |
5621 | } |
5622 | ||
5623 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
5624 | { | |
dd5ac32d ID |
5625 | if (rbd_dev) |
5626 | put_device(&rbd_dev->dev); | |
c53d5893 AE |
5627 | } |
5628 | ||
9d475de5 AE |
5629 | /* |
5630 | * Get the size and object order for an image snapshot, or if | |
5631 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
5632 | * image. | |
5633 | */ | |
5634 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
5635 | u8 *order, u64 *snap_size) | |
5636 | { | |
5637 | __le64 snapid = cpu_to_le64(snap_id); | |
5638 | int ret; | |
5639 | struct { | |
5640 | u8 order; | |
5641 | __le64 size; | |
5642 | } __attribute__ ((packed)) size_buf = { 0 }; | |
5643 | ||
ecd4a68a ID |
5644 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5645 | &rbd_dev->header_oloc, "get_size", | |
5646 | &snapid, sizeof(snapid), | |
5647 | &size_buf, sizeof(size_buf)); | |
36be9a76 | 5648 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
5649 | if (ret < 0) |
5650 | return ret; | |
57385b51 AE |
5651 | if (ret < sizeof (size_buf)) |
5652 | return -ERANGE; | |
9d475de5 | 5653 | |
c3545579 | 5654 | if (order) { |
c86f86e9 | 5655 | *order = size_buf.order; |
c3545579 JD |
5656 | dout(" order %u", (unsigned int)*order); |
5657 | } | |
9d475de5 AE |
5658 | *snap_size = le64_to_cpu(size_buf.size); |
5659 | ||
c3545579 JD |
5660 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
5661 | (unsigned long long)snap_id, | |
57385b51 | 5662 | (unsigned long long)*snap_size); |
9d475de5 AE |
5663 | |
5664 | return 0; | |
5665 | } | |
5666 | ||
5667 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
5668 | { | |
5669 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
5670 | &rbd_dev->header.obj_order, | |
5671 | &rbd_dev->header.image_size); | |
5672 | } | |
5673 | ||
1e130199 AE |
5674 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
5675 | { | |
5435d206 | 5676 | size_t size; |
1e130199 AE |
5677 | void *reply_buf; |
5678 | int ret; | |
5679 | void *p; | |
5680 | ||
5435d206 DY |
5681 | /* Response will be an encoded string, which includes a length */ |
5682 | size = sizeof(__le32) + RBD_OBJ_PREFIX_LEN_MAX; | |
5683 | reply_buf = kzalloc(size, GFP_KERNEL); | |
1e130199 AE |
5684 | if (!reply_buf) |
5685 | return -ENOMEM; | |
5686 | ||
ecd4a68a ID |
5687 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5688 | &rbd_dev->header_oloc, "get_object_prefix", | |
5435d206 | 5689 | NULL, 0, reply_buf, size); |
36be9a76 | 5690 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
5691 | if (ret < 0) |
5692 | goto out; | |
5693 | ||
5694 | p = reply_buf; | |
5695 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
5696 | p + ret, NULL, GFP_NOIO); |
5697 | ret = 0; | |
1e130199 AE |
5698 | |
5699 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
5700 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
5701 | rbd_dev->header.object_prefix = NULL; | |
5702 | } else { | |
5703 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
5704 | } | |
1e130199 AE |
5705 | out: |
5706 | kfree(reply_buf); | |
5707 | ||
5708 | return ret; | |
5709 | } | |
5710 | ||
b1b5402a AE |
5711 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
5712 | u64 *snap_features) | |
5713 | { | |
5714 | __le64 snapid = cpu_to_le64(snap_id); | |
5715 | struct { | |
5716 | __le64 features; | |
5717 | __le64 incompat; | |
4157976b | 5718 | } __attribute__ ((packed)) features_buf = { 0 }; |
d3767f0f | 5719 | u64 unsup; |
b1b5402a AE |
5720 | int ret; |
5721 | ||
ecd4a68a ID |
5722 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
5723 | &rbd_dev->header_oloc, "get_features", | |
5724 | &snapid, sizeof(snapid), | |
5725 | &features_buf, sizeof(features_buf)); | |
36be9a76 | 5726 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
5727 | if (ret < 0) |
5728 | return ret; | |
57385b51 AE |
5729 | if (ret < sizeof (features_buf)) |
5730 | return -ERANGE; | |
d889140c | 5731 | |
d3767f0f ID |
5732 | unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED; |
5733 | if (unsup) { | |
5734 | rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx", | |
5735 | unsup); | |
b8f5c6ed | 5736 | return -ENXIO; |
d3767f0f | 5737 | } |
d889140c | 5738 | |
b1b5402a AE |
5739 | *snap_features = le64_to_cpu(features_buf.features); |
5740 | ||
5741 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
5742 | (unsigned long long)snap_id, |
5743 | (unsigned long long)*snap_features, | |
5744 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
5745 | |
5746 | return 0; | |
5747 | } | |
5748 | ||
5749 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
5750 | { | |
5751 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
5752 | &rbd_dev->header.features); | |
5753 | } | |
5754 | ||
22e8bd51 ID |
5755 | /* |
5756 | * These are generic image flags, but since they are used only for | |
5757 | * object map, store them in rbd_dev->object_map_flags. | |
5758 | * | |
5759 | * For the same reason, this function is called only on object map | |
5760 | * (re)load and not on header refresh. | |
5761 | */ | |
5762 | static int rbd_dev_v2_get_flags(struct rbd_device *rbd_dev) | |
5763 | { | |
5764 | __le64 snapid = cpu_to_le64(rbd_dev->spec->snap_id); | |
5765 | __le64 flags; | |
5766 | int ret; | |
5767 | ||
5768 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, | |
5769 | &rbd_dev->header_oloc, "get_flags", | |
5770 | &snapid, sizeof(snapid), | |
5771 | &flags, sizeof(flags)); | |
5772 | if (ret < 0) | |
5773 | return ret; | |
5774 | if (ret < sizeof(flags)) | |
5775 | return -EBADMSG; | |
5776 | ||
5777 | rbd_dev->object_map_flags = le64_to_cpu(flags); | |
5778 | return 0; | |
5779 | } | |
5780 | ||
eb3b2d6b ID |
5781 | struct parent_image_info { |
5782 | u64 pool_id; | |
e92c0eaf | 5783 | const char *pool_ns; |
eb3b2d6b ID |
5784 | const char *image_id; |
5785 | u64 snap_id; | |
5786 | ||
e92c0eaf | 5787 | bool has_overlap; |
eb3b2d6b ID |
5788 | u64 overlap; |
5789 | }; | |
5790 | ||
e92c0eaf ID |
5791 | /* |
5792 | * The caller is responsible for @pii. | |
5793 | */ | |
5794 | static int decode_parent_image_spec(void **p, void *end, | |
5795 | struct parent_image_info *pii) | |
5796 | { | |
5797 | u8 struct_v; | |
5798 | u32 struct_len; | |
5799 | int ret; | |
5800 | ||
5801 | ret = ceph_start_decoding(p, end, 1, "ParentImageSpec", | |
5802 | &struct_v, &struct_len); | |
5803 | if (ret) | |
5804 | return ret; | |
5805 | ||
5806 | ceph_decode_64_safe(p, end, pii->pool_id, e_inval); | |
5807 | pii->pool_ns = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL); | |
5808 | if (IS_ERR(pii->pool_ns)) { | |
5809 | ret = PTR_ERR(pii->pool_ns); | |
5810 | pii->pool_ns = NULL; | |
5811 | return ret; | |
5812 | } | |
5813 | pii->image_id = ceph_extract_encoded_string(p, end, NULL, GFP_KERNEL); | |
5814 | if (IS_ERR(pii->image_id)) { | |
5815 | ret = PTR_ERR(pii->image_id); | |
5816 | pii->image_id = NULL; | |
5817 | return ret; | |
5818 | } | |
5819 | ceph_decode_64_safe(p, end, pii->snap_id, e_inval); | |
5820 | return 0; | |
5821 | ||
5822 | e_inval: | |
5823 | return -EINVAL; | |
5824 | } | |
5825 | ||
5826 | static int __get_parent_info(struct rbd_device *rbd_dev, | |
5827 | struct page *req_page, | |
5828 | struct page *reply_page, | |
5829 | struct parent_image_info *pii) | |
5830 | { | |
5831 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
5832 | size_t reply_len = PAGE_SIZE; | |
5833 | void *p, *end; | |
5834 | int ret; | |
5835 | ||
5836 | ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
5837 | "rbd", "parent_get", CEPH_OSD_FLAG_READ, | |
68ada915 | 5838 | req_page, sizeof(u64), &reply_page, &reply_len); |
e92c0eaf ID |
5839 | if (ret) |
5840 | return ret == -EOPNOTSUPP ? 1 : ret; | |
5841 | ||
5842 | p = page_address(reply_page); | |
5843 | end = p + reply_len; | |
5844 | ret = decode_parent_image_spec(&p, end, pii); | |
5845 | if (ret) | |
5846 | return ret; | |
5847 | ||
5848 | ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
5849 | "rbd", "parent_overlap_get", CEPH_OSD_FLAG_READ, | |
68ada915 | 5850 | req_page, sizeof(u64), &reply_page, &reply_len); |
e92c0eaf ID |
5851 | if (ret) |
5852 | return ret; | |
5853 | ||
5854 | p = page_address(reply_page); | |
5855 | end = p + reply_len; | |
5856 | ceph_decode_8_safe(&p, end, pii->has_overlap, e_inval); | |
5857 | if (pii->has_overlap) | |
5858 | ceph_decode_64_safe(&p, end, pii->overlap, e_inval); | |
5859 | ||
5860 | return 0; | |
5861 | ||
5862 | e_inval: | |
5863 | return -EINVAL; | |
5864 | } | |
5865 | ||
eb3b2d6b ID |
5866 | /* |
5867 | * The caller is responsible for @pii. | |
5868 | */ | |
5869 | static int __get_parent_info_legacy(struct rbd_device *rbd_dev, | |
5870 | struct page *req_page, | |
5871 | struct page *reply_page, | |
5872 | struct parent_image_info *pii) | |
5873 | { | |
5874 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
5875 | size_t reply_len = PAGE_SIZE; | |
5876 | void *p, *end; | |
5877 | int ret; | |
5878 | ||
5879 | ret = ceph_osdc_call(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
5880 | "rbd", "get_parent", CEPH_OSD_FLAG_READ, | |
68ada915 | 5881 | req_page, sizeof(u64), &reply_page, &reply_len); |
eb3b2d6b ID |
5882 | if (ret) |
5883 | return ret; | |
5884 | ||
5885 | p = page_address(reply_page); | |
5886 | end = p + reply_len; | |
5887 | ceph_decode_64_safe(&p, end, pii->pool_id, e_inval); | |
5888 | pii->image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); | |
5889 | if (IS_ERR(pii->image_id)) { | |
5890 | ret = PTR_ERR(pii->image_id); | |
5891 | pii->image_id = NULL; | |
5892 | return ret; | |
5893 | } | |
5894 | ceph_decode_64_safe(&p, end, pii->snap_id, e_inval); | |
e92c0eaf | 5895 | pii->has_overlap = true; |
eb3b2d6b ID |
5896 | ceph_decode_64_safe(&p, end, pii->overlap, e_inval); |
5897 | ||
5898 | return 0; | |
5899 | ||
5900 | e_inval: | |
5901 | return -EINVAL; | |
5902 | } | |
5903 | ||
5904 | static int get_parent_info(struct rbd_device *rbd_dev, | |
5905 | struct parent_image_info *pii) | |
5906 | { | |
5907 | struct page *req_page, *reply_page; | |
5908 | void *p; | |
5909 | int ret; | |
5910 | ||
5911 | req_page = alloc_page(GFP_KERNEL); | |
5912 | if (!req_page) | |
5913 | return -ENOMEM; | |
5914 | ||
5915 | reply_page = alloc_page(GFP_KERNEL); | |
5916 | if (!reply_page) { | |
5917 | __free_page(req_page); | |
5918 | return -ENOMEM; | |
5919 | } | |
5920 | ||
5921 | p = page_address(req_page); | |
5922 | ceph_encode_64(&p, rbd_dev->spec->snap_id); | |
e92c0eaf ID |
5923 | ret = __get_parent_info(rbd_dev, req_page, reply_page, pii); |
5924 | if (ret > 0) | |
5925 | ret = __get_parent_info_legacy(rbd_dev, req_page, reply_page, | |
5926 | pii); | |
eb3b2d6b ID |
5927 | |
5928 | __free_page(req_page); | |
5929 | __free_page(reply_page); | |
5930 | return ret; | |
5931 | } | |
5932 | ||
86b00e0d AE |
5933 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
5934 | { | |
5935 | struct rbd_spec *parent_spec; | |
eb3b2d6b | 5936 | struct parent_image_info pii = { 0 }; |
86b00e0d AE |
5937 | int ret; |
5938 | ||
5939 | parent_spec = rbd_spec_alloc(); | |
5940 | if (!parent_spec) | |
5941 | return -ENOMEM; | |
5942 | ||
eb3b2d6b ID |
5943 | ret = get_parent_info(rbd_dev, &pii); |
5944 | if (ret) | |
86b00e0d | 5945 | goto out_err; |
86b00e0d | 5946 | |
e92c0eaf ID |
5947 | dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n", |
5948 | __func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id, | |
5949 | pii.has_overlap, pii.overlap); | |
86b00e0d | 5950 | |
e92c0eaf | 5951 | if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) { |
392a9dad AE |
5952 | /* |
5953 | * Either the parent never existed, or we have | |
5954 | * record of it but the image got flattened so it no | |
5955 | * longer has a parent. When the parent of a | |
5956 | * layered image disappears we immediately set the | |
5957 | * overlap to 0. The effect of this is that all new | |
5958 | * requests will be treated as if the image had no | |
5959 | * parent. | |
e92c0eaf ID |
5960 | * |
5961 | * If !pii.has_overlap, the parent image spec is not | |
5962 | * applicable. It's there to avoid duplication in each | |
5963 | * snapshot record. | |
392a9dad AE |
5964 | */ |
5965 | if (rbd_dev->parent_overlap) { | |
5966 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
5967 | rbd_dev_parent_put(rbd_dev); |
5968 | pr_info("%s: clone image has been flattened\n", | |
5969 | rbd_dev->disk->disk_name); | |
5970 | } | |
5971 | ||
86b00e0d | 5972 | goto out; /* No parent? No problem. */ |
392a9dad | 5973 | } |
86b00e0d | 5974 | |
0903e875 AE |
5975 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5976 | ||
5977 | ret = -EIO; | |
eb3b2d6b | 5978 | if (pii.pool_id > (u64)U32_MAX) { |
9584d508 | 5979 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
eb3b2d6b | 5980 | (unsigned long long)pii.pool_id, U32_MAX); |
86b00e0d AE |
5981 | goto out_err; |
5982 | } | |
86b00e0d | 5983 | |
3b5cf2a2 AE |
5984 | /* |
5985 | * The parent won't change (except when the clone is | |
5986 | * flattened, already handled that). So we only need to | |
5987 | * record the parent spec we have not already done so. | |
5988 | */ | |
5989 | if (!rbd_dev->parent_spec) { | |
eb3b2d6b | 5990 | parent_spec->pool_id = pii.pool_id; |
e92c0eaf ID |
5991 | if (pii.pool_ns && *pii.pool_ns) { |
5992 | parent_spec->pool_ns = pii.pool_ns; | |
5993 | pii.pool_ns = NULL; | |
5994 | } | |
eb3b2d6b ID |
5995 | parent_spec->image_id = pii.image_id; |
5996 | pii.image_id = NULL; | |
5997 | parent_spec->snap_id = pii.snap_id; | |
b26c047b | 5998 | |
70cf49cf AE |
5999 | rbd_dev->parent_spec = parent_spec; |
6000 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3b5cf2a2 AE |
6001 | } |
6002 | ||
6003 | /* | |
cf32bd9c ID |
6004 | * We always update the parent overlap. If it's zero we issue |
6005 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 6006 | */ |
eb3b2d6b | 6007 | if (!pii.overlap) { |
3b5cf2a2 | 6008 | if (parent_spec) { |
cf32bd9c ID |
6009 | /* refresh, careful to warn just once */ |
6010 | if (rbd_dev->parent_overlap) | |
6011 | rbd_warn(rbd_dev, | |
6012 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 6013 | } else { |
cf32bd9c ID |
6014 | /* initial probe */ |
6015 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 6016 | } |
70cf49cf | 6017 | } |
eb3b2d6b | 6018 | rbd_dev->parent_overlap = pii.overlap; |
cf32bd9c | 6019 | |
86b00e0d AE |
6020 | out: |
6021 | ret = 0; | |
6022 | out_err: | |
e92c0eaf | 6023 | kfree(pii.pool_ns); |
eb3b2d6b | 6024 | kfree(pii.image_id); |
86b00e0d | 6025 | rbd_spec_put(parent_spec); |
86b00e0d AE |
6026 | return ret; |
6027 | } | |
6028 | ||
cc070d59 AE |
6029 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
6030 | { | |
6031 | struct { | |
6032 | __le64 stripe_unit; | |
6033 | __le64 stripe_count; | |
6034 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
6035 | size_t size = sizeof (striping_info_buf); | |
6036 | void *p; | |
cc070d59 AE |
6037 | int ret; |
6038 | ||
ecd4a68a ID |
6039 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
6040 | &rbd_dev->header_oloc, "get_stripe_unit_count", | |
6041 | NULL, 0, &striping_info_buf, size); | |
cc070d59 AE |
6042 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
6043 | if (ret < 0) | |
6044 | return ret; | |
6045 | if (ret < size) | |
6046 | return -ERANGE; | |
6047 | ||
cc070d59 | 6048 | p = &striping_info_buf; |
b1331852 ID |
6049 | rbd_dev->header.stripe_unit = ceph_decode_64(&p); |
6050 | rbd_dev->header.stripe_count = ceph_decode_64(&p); | |
cc070d59 AE |
6051 | return 0; |
6052 | } | |
6053 | ||
7e97332e ID |
6054 | static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev) |
6055 | { | |
6056 | __le64 data_pool_id; | |
6057 | int ret; | |
6058 | ||
6059 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, | |
6060 | &rbd_dev->header_oloc, "get_data_pool", | |
6061 | NULL, 0, &data_pool_id, sizeof(data_pool_id)); | |
6062 | if (ret < 0) | |
6063 | return ret; | |
6064 | if (ret < sizeof(data_pool_id)) | |
6065 | return -EBADMSG; | |
6066 | ||
6067 | rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id); | |
6068 | WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL); | |
6069 | return 0; | |
6070 | } | |
6071 | ||
9e15b77d AE |
6072 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
6073 | { | |
ecd4a68a | 6074 | CEPH_DEFINE_OID_ONSTACK(oid); |
9e15b77d AE |
6075 | size_t image_id_size; |
6076 | char *image_id; | |
6077 | void *p; | |
6078 | void *end; | |
6079 | size_t size; | |
6080 | void *reply_buf = NULL; | |
6081 | size_t len = 0; | |
6082 | char *image_name = NULL; | |
6083 | int ret; | |
6084 | ||
6085 | rbd_assert(!rbd_dev->spec->image_name); | |
6086 | ||
69e7a02f AE |
6087 | len = strlen(rbd_dev->spec->image_id); |
6088 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
6089 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
6090 | if (!image_id) | |
6091 | return NULL; | |
6092 | ||
6093 | p = image_id; | |
4157976b | 6094 | end = image_id + image_id_size; |
57385b51 | 6095 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
6096 | |
6097 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
6098 | reply_buf = kmalloc(size, GFP_KERNEL); | |
6099 | if (!reply_buf) | |
6100 | goto out; | |
6101 | ||
ecd4a68a ID |
6102 | ceph_oid_printf(&oid, "%s", RBD_DIRECTORY); |
6103 | ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc, | |
6104 | "dir_get_name", image_id, image_id_size, | |
6105 | reply_buf, size); | |
9e15b77d AE |
6106 | if (ret < 0) |
6107 | goto out; | |
6108 | p = reply_buf; | |
f40eb349 AE |
6109 | end = reply_buf + ret; |
6110 | ||
9e15b77d AE |
6111 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
6112 | if (IS_ERR(image_name)) | |
6113 | image_name = NULL; | |
6114 | else | |
6115 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
6116 | out: | |
6117 | kfree(reply_buf); | |
6118 | kfree(image_id); | |
6119 | ||
6120 | return image_name; | |
6121 | } | |
6122 | ||
2ad3d716 AE |
6123 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
6124 | { | |
6125 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
6126 | const char *snap_name; | |
6127 | u32 which = 0; | |
6128 | ||
6129 | /* Skip over names until we find the one we are looking for */ | |
6130 | ||
6131 | snap_name = rbd_dev->header.snap_names; | |
6132 | while (which < snapc->num_snaps) { | |
6133 | if (!strcmp(name, snap_name)) | |
6134 | return snapc->snaps[which]; | |
6135 | snap_name += strlen(snap_name) + 1; | |
6136 | which++; | |
6137 | } | |
6138 | return CEPH_NOSNAP; | |
6139 | } | |
6140 | ||
6141 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
6142 | { | |
6143 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
6144 | u32 which; | |
6145 | bool found = false; | |
6146 | u64 snap_id; | |
6147 | ||
6148 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
6149 | const char *snap_name; | |
6150 | ||
6151 | snap_id = snapc->snaps[which]; | |
6152 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
6153 | if (IS_ERR(snap_name)) { |
6154 | /* ignore no-longer existing snapshots */ | |
6155 | if (PTR_ERR(snap_name) == -ENOENT) | |
6156 | continue; | |
6157 | else | |
6158 | break; | |
6159 | } | |
2ad3d716 AE |
6160 | found = !strcmp(name, snap_name); |
6161 | kfree(snap_name); | |
6162 | } | |
6163 | return found ? snap_id : CEPH_NOSNAP; | |
6164 | } | |
6165 | ||
6166 | /* | |
6167 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
6168 | * no snapshot by that name is found, or if an error occurs. | |
6169 | */ | |
6170 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
6171 | { | |
6172 | if (rbd_dev->image_format == 1) | |
6173 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
6174 | ||
6175 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
6176 | } | |
6177 | ||
9e15b77d | 6178 | /* |
04077599 ID |
6179 | * An image being mapped will have everything but the snap id. |
6180 | */ | |
6181 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
6182 | { | |
6183 | struct rbd_spec *spec = rbd_dev->spec; | |
6184 | ||
6185 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
6186 | rbd_assert(spec->image_id && spec->image_name); | |
6187 | rbd_assert(spec->snap_name); | |
6188 | ||
6189 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
6190 | u64 snap_id; | |
6191 | ||
6192 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
6193 | if (snap_id == CEPH_NOSNAP) | |
6194 | return -ENOENT; | |
6195 | ||
6196 | spec->snap_id = snap_id; | |
6197 | } else { | |
6198 | spec->snap_id = CEPH_NOSNAP; | |
6199 | } | |
6200 | ||
6201 | return 0; | |
6202 | } | |
6203 | ||
6204 | /* | |
6205 | * A parent image will have all ids but none of the names. | |
e1d4213f | 6206 | * |
04077599 ID |
6207 | * All names in an rbd spec are dynamically allocated. It's OK if we |
6208 | * can't figure out the name for an image id. | |
9e15b77d | 6209 | */ |
04077599 | 6210 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 6211 | { |
2e9f7f1c AE |
6212 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
6213 | struct rbd_spec *spec = rbd_dev->spec; | |
6214 | const char *pool_name; | |
6215 | const char *image_name; | |
6216 | const char *snap_name; | |
9e15b77d AE |
6217 | int ret; |
6218 | ||
04077599 ID |
6219 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
6220 | rbd_assert(spec->image_id); | |
6221 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 6222 | |
2e9f7f1c | 6223 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 6224 | |
2e9f7f1c AE |
6225 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
6226 | if (!pool_name) { | |
6227 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
6228 | return -EIO; |
6229 | } | |
2e9f7f1c AE |
6230 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
6231 | if (!pool_name) | |
9e15b77d AE |
6232 | return -ENOMEM; |
6233 | ||
6234 | /* Fetch the image name; tolerate failure here */ | |
6235 | ||
2e9f7f1c AE |
6236 | image_name = rbd_dev_image_name(rbd_dev); |
6237 | if (!image_name) | |
06ecc6cb | 6238 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 6239 | |
04077599 | 6240 | /* Fetch the snapshot name */ |
9e15b77d | 6241 | |
2e9f7f1c | 6242 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
6243 | if (IS_ERR(snap_name)) { |
6244 | ret = PTR_ERR(snap_name); | |
9e15b77d | 6245 | goto out_err; |
2e9f7f1c AE |
6246 | } |
6247 | ||
6248 | spec->pool_name = pool_name; | |
6249 | spec->image_name = image_name; | |
6250 | spec->snap_name = snap_name; | |
9e15b77d AE |
6251 | |
6252 | return 0; | |
04077599 | 6253 | |
9e15b77d | 6254 | out_err: |
2e9f7f1c AE |
6255 | kfree(image_name); |
6256 | kfree(pool_name); | |
9e15b77d AE |
6257 | return ret; |
6258 | } | |
6259 | ||
cc4a38bd | 6260 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
6261 | { |
6262 | size_t size; | |
6263 | int ret; | |
6264 | void *reply_buf; | |
6265 | void *p; | |
6266 | void *end; | |
6267 | u64 seq; | |
6268 | u32 snap_count; | |
6269 | struct ceph_snap_context *snapc; | |
6270 | u32 i; | |
6271 | ||
6272 | /* | |
6273 | * We'll need room for the seq value (maximum snapshot id), | |
6274 | * snapshot count, and array of that many snapshot ids. | |
6275 | * For now we have a fixed upper limit on the number we're | |
6276 | * prepared to receive. | |
6277 | */ | |
6278 | size = sizeof (__le64) + sizeof (__le32) + | |
6279 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
6280 | reply_buf = kzalloc(size, GFP_KERNEL); | |
6281 | if (!reply_buf) | |
6282 | return -ENOMEM; | |
6283 | ||
ecd4a68a ID |
6284 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
6285 | &rbd_dev->header_oloc, "get_snapcontext", | |
6286 | NULL, 0, reply_buf, size); | |
36be9a76 | 6287 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
6288 | if (ret < 0) |
6289 | goto out; | |
6290 | ||
35d489f9 | 6291 | p = reply_buf; |
57385b51 AE |
6292 | end = reply_buf + ret; |
6293 | ret = -ERANGE; | |
35d489f9 AE |
6294 | ceph_decode_64_safe(&p, end, seq, out); |
6295 | ceph_decode_32_safe(&p, end, snap_count, out); | |
6296 | ||
6297 | /* | |
6298 | * Make sure the reported number of snapshot ids wouldn't go | |
6299 | * beyond the end of our buffer. But before checking that, | |
6300 | * make sure the computed size of the snapshot context we | |
6301 | * allocate is representable in a size_t. | |
6302 | */ | |
6303 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
6304 | / sizeof (u64)) { | |
6305 | ret = -EINVAL; | |
6306 | goto out; | |
6307 | } | |
6308 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
6309 | goto out; | |
468521c1 | 6310 | ret = 0; |
35d489f9 | 6311 | |
812164f8 | 6312 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
6313 | if (!snapc) { |
6314 | ret = -ENOMEM; | |
6315 | goto out; | |
6316 | } | |
35d489f9 | 6317 | snapc->seq = seq; |
35d489f9 AE |
6318 | for (i = 0; i < snap_count; i++) |
6319 | snapc->snaps[i] = ceph_decode_64(&p); | |
6320 | ||
49ece554 | 6321 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
6322 | rbd_dev->header.snapc = snapc; |
6323 | ||
6324 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 6325 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
6326 | out: |
6327 | kfree(reply_buf); | |
6328 | ||
57385b51 | 6329 | return ret; |
35d489f9 AE |
6330 | } |
6331 | ||
54cac61f AE |
6332 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
6333 | u64 snap_id) | |
b8b1e2db AE |
6334 | { |
6335 | size_t size; | |
6336 | void *reply_buf; | |
54cac61f | 6337 | __le64 snapid; |
b8b1e2db AE |
6338 | int ret; |
6339 | void *p; | |
6340 | void *end; | |
b8b1e2db AE |
6341 | char *snap_name; |
6342 | ||
6343 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
6344 | reply_buf = kmalloc(size, GFP_KERNEL); | |
6345 | if (!reply_buf) | |
6346 | return ERR_PTR(-ENOMEM); | |
6347 | ||
54cac61f | 6348 | snapid = cpu_to_le64(snap_id); |
ecd4a68a ID |
6349 | ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid, |
6350 | &rbd_dev->header_oloc, "get_snapshot_name", | |
6351 | &snapid, sizeof(snapid), reply_buf, size); | |
36be9a76 | 6352 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
6353 | if (ret < 0) { |
6354 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 6355 | goto out; |
f40eb349 | 6356 | } |
b8b1e2db AE |
6357 | |
6358 | p = reply_buf; | |
f40eb349 | 6359 | end = reply_buf + ret; |
e5c35534 | 6360 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 6361 | if (IS_ERR(snap_name)) |
b8b1e2db | 6362 | goto out; |
b8b1e2db | 6363 | |
f40eb349 | 6364 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 6365 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
6366 | out: |
6367 | kfree(reply_buf); | |
6368 | ||
f40eb349 | 6369 | return snap_name; |
b8b1e2db AE |
6370 | } |
6371 | ||
2df3fac7 | 6372 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 6373 | { |
2df3fac7 | 6374 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 6375 | int ret; |
117973fb | 6376 | |
1617e40c JD |
6377 | ret = rbd_dev_v2_image_size(rbd_dev); |
6378 | if (ret) | |
cfbf6377 | 6379 | return ret; |
1617e40c | 6380 | |
2df3fac7 AE |
6381 | if (first_time) { |
6382 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
6383 | if (ret) | |
cfbf6377 | 6384 | return ret; |
2df3fac7 AE |
6385 | } |
6386 | ||
cc4a38bd | 6387 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
6388 | if (ret && first_time) { |
6389 | kfree(rbd_dev->header.object_prefix); | |
6390 | rbd_dev->header.object_prefix = NULL; | |
6391 | } | |
117973fb AE |
6392 | |
6393 | return ret; | |
6394 | } | |
6395 | ||
a720ae09 ID |
6396 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
6397 | { | |
6398 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
6399 | ||
6400 | if (rbd_dev->image_format == 1) | |
6401 | return rbd_dev_v1_header_info(rbd_dev); | |
6402 | ||
6403 | return rbd_dev_v2_header_info(rbd_dev); | |
6404 | } | |
6405 | ||
e28fff26 AE |
6406 | /* |
6407 | * Skips over white space at *buf, and updates *buf to point to the | |
6408 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
6409 | * the token (string of non-white space characters) found. Note |
6410 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
6411 | */ |
6412 | static inline size_t next_token(const char **buf) | |
6413 | { | |
6414 | /* | |
6415 | * These are the characters that produce nonzero for | |
6416 | * isspace() in the "C" and "POSIX" locales. | |
6417 | */ | |
6418 | const char *spaces = " \f\n\r\t\v"; | |
6419 | ||
6420 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
6421 | ||
6422 | return strcspn(*buf, spaces); /* Return token length */ | |
6423 | } | |
6424 | ||
ea3352f4 AE |
6425 | /* |
6426 | * Finds the next token in *buf, dynamically allocates a buffer big | |
6427 | * enough to hold a copy of it, and copies the token into the new | |
6428 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
6429 | * that a duplicate buffer is created even for a zero-length token. | |
6430 | * | |
6431 | * Returns a pointer to the newly-allocated duplicate, or a null | |
6432 | * pointer if memory for the duplicate was not available. If | |
6433 | * the lenp argument is a non-null pointer, the length of the token | |
6434 | * (not including the '\0') is returned in *lenp. | |
6435 | * | |
6436 | * If successful, the *buf pointer will be updated to point beyond | |
6437 | * the end of the found token. | |
6438 | * | |
6439 | * Note: uses GFP_KERNEL for allocation. | |
6440 | */ | |
6441 | static inline char *dup_token(const char **buf, size_t *lenp) | |
6442 | { | |
6443 | char *dup; | |
6444 | size_t len; | |
6445 | ||
6446 | len = next_token(buf); | |
4caf35f9 | 6447 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
6448 | if (!dup) |
6449 | return NULL; | |
ea3352f4 AE |
6450 | *(dup + len) = '\0'; |
6451 | *buf += len; | |
6452 | ||
6453 | if (lenp) | |
6454 | *lenp = len; | |
6455 | ||
6456 | return dup; | |
6457 | } | |
6458 | ||
a725f65e | 6459 | /* |
859c31df AE |
6460 | * Parse the options provided for an "rbd add" (i.e., rbd image |
6461 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
6462 | * and the data written is passed here via a NUL-terminated buffer. | |
6463 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 6464 | * |
859c31df AE |
6465 | * The information extracted from these options is recorded in |
6466 | * the other parameters which return dynamically-allocated | |
6467 | * structures: | |
6468 | * ceph_opts | |
6469 | * The address of a pointer that will refer to a ceph options | |
6470 | * structure. Caller must release the returned pointer using | |
6471 | * ceph_destroy_options() when it is no longer needed. | |
6472 | * rbd_opts | |
6473 | * Address of an rbd options pointer. Fully initialized by | |
6474 | * this function; caller must release with kfree(). | |
6475 | * spec | |
6476 | * Address of an rbd image specification pointer. Fully | |
6477 | * initialized by this function based on parsed options. | |
6478 | * Caller must release with rbd_spec_put(). | |
6479 | * | |
6480 | * The options passed take this form: | |
6481 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
6482 | * where: | |
6483 | * <mon_addrs> | |
6484 | * A comma-separated list of one or more monitor addresses. | |
6485 | * A monitor address is an ip address, optionally followed | |
6486 | * by a port number (separated by a colon). | |
6487 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
6488 | * <options> | |
6489 | * A comma-separated list of ceph and/or rbd options. | |
6490 | * <pool_name> | |
6491 | * The name of the rados pool containing the rbd image. | |
6492 | * <image_name> | |
6493 | * The name of the image in that pool to map. | |
6494 | * <snap_id> | |
6495 | * An optional snapshot id. If provided, the mapping will | |
6496 | * present data from the image at the time that snapshot was | |
6497 | * created. The image head is used if no snapshot id is | |
6498 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 6499 | */ |
859c31df | 6500 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 6501 | struct ceph_options **ceph_opts, |
859c31df AE |
6502 | struct rbd_options **opts, |
6503 | struct rbd_spec **rbd_spec) | |
e28fff26 | 6504 | { |
d22f76e7 | 6505 | size_t len; |
859c31df | 6506 | char *options; |
0ddebc0c | 6507 | const char *mon_addrs; |
ecb4dc22 | 6508 | char *snap_name; |
0ddebc0c | 6509 | size_t mon_addrs_size; |
c300156b | 6510 | struct parse_rbd_opts_ctx pctx = { 0 }; |
859c31df | 6511 | struct ceph_options *copts; |
dc79b113 | 6512 | int ret; |
e28fff26 AE |
6513 | |
6514 | /* The first four tokens are required */ | |
6515 | ||
7ef3214a | 6516 | len = next_token(&buf); |
4fb5d671 AE |
6517 | if (!len) { |
6518 | rbd_warn(NULL, "no monitor address(es) provided"); | |
6519 | return -EINVAL; | |
6520 | } | |
0ddebc0c | 6521 | mon_addrs = buf; |
f28e565a | 6522 | mon_addrs_size = len + 1; |
7ef3214a | 6523 | buf += len; |
a725f65e | 6524 | |
dc79b113 | 6525 | ret = -EINVAL; |
f28e565a AE |
6526 | options = dup_token(&buf, NULL); |
6527 | if (!options) | |
dc79b113 | 6528 | return -ENOMEM; |
4fb5d671 AE |
6529 | if (!*options) { |
6530 | rbd_warn(NULL, "no options provided"); | |
6531 | goto out_err; | |
6532 | } | |
e28fff26 | 6533 | |
c300156b ID |
6534 | pctx.spec = rbd_spec_alloc(); |
6535 | if (!pctx.spec) | |
f28e565a | 6536 | goto out_mem; |
859c31df | 6537 | |
c300156b ID |
6538 | pctx.spec->pool_name = dup_token(&buf, NULL); |
6539 | if (!pctx.spec->pool_name) | |
859c31df | 6540 | goto out_mem; |
c300156b | 6541 | if (!*pctx.spec->pool_name) { |
4fb5d671 AE |
6542 | rbd_warn(NULL, "no pool name provided"); |
6543 | goto out_err; | |
6544 | } | |
e28fff26 | 6545 | |
c300156b ID |
6546 | pctx.spec->image_name = dup_token(&buf, NULL); |
6547 | if (!pctx.spec->image_name) | |
f28e565a | 6548 | goto out_mem; |
c300156b | 6549 | if (!*pctx.spec->image_name) { |
4fb5d671 AE |
6550 | rbd_warn(NULL, "no image name provided"); |
6551 | goto out_err; | |
6552 | } | |
d4b125e9 | 6553 | |
f28e565a AE |
6554 | /* |
6555 | * Snapshot name is optional; default is to use "-" | |
6556 | * (indicating the head/no snapshot). | |
6557 | */ | |
3feeb894 | 6558 | len = next_token(&buf); |
820a5f3e | 6559 | if (!len) { |
3feeb894 AE |
6560 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
6561 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 6562 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 6563 | ret = -ENAMETOOLONG; |
f28e565a | 6564 | goto out_err; |
849b4260 | 6565 | } |
ecb4dc22 AE |
6566 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
6567 | if (!snap_name) | |
f28e565a | 6568 | goto out_mem; |
ecb4dc22 | 6569 | *(snap_name + len) = '\0'; |
c300156b | 6570 | pctx.spec->snap_name = snap_name; |
e5c35534 | 6571 | |
0ddebc0c | 6572 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 6573 | |
c300156b ID |
6574 | pctx.opts = kzalloc(sizeof(*pctx.opts), GFP_KERNEL); |
6575 | if (!pctx.opts) | |
4e9afeba AE |
6576 | goto out_mem; |
6577 | ||
c300156b ID |
6578 | pctx.opts->read_only = RBD_READ_ONLY_DEFAULT; |
6579 | pctx.opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; | |
0c93e1b7 | 6580 | pctx.opts->alloc_size = RBD_ALLOC_SIZE_DEFAULT; |
c300156b ID |
6581 | pctx.opts->lock_timeout = RBD_LOCK_TIMEOUT_DEFAULT; |
6582 | pctx.opts->lock_on_read = RBD_LOCK_ON_READ_DEFAULT; | |
6583 | pctx.opts->exclusive = RBD_EXCLUSIVE_DEFAULT; | |
6584 | pctx.opts->trim = RBD_TRIM_DEFAULT; | |
d22f76e7 | 6585 | |
859c31df | 6586 | copts = ceph_parse_options(options, mon_addrs, |
c300156b ID |
6587 | mon_addrs + mon_addrs_size - 1, |
6588 | parse_rbd_opts_token, &pctx); | |
859c31df AE |
6589 | if (IS_ERR(copts)) { |
6590 | ret = PTR_ERR(copts); | |
dc79b113 AE |
6591 | goto out_err; |
6592 | } | |
859c31df AE |
6593 | kfree(options); |
6594 | ||
6595 | *ceph_opts = copts; | |
c300156b ID |
6596 | *opts = pctx.opts; |
6597 | *rbd_spec = pctx.spec; | |
0ddebc0c | 6598 | |
dc79b113 | 6599 | return 0; |
f28e565a | 6600 | out_mem: |
dc79b113 | 6601 | ret = -ENOMEM; |
d22f76e7 | 6602 | out_err: |
c300156b ID |
6603 | kfree(pctx.opts); |
6604 | rbd_spec_put(pctx.spec); | |
f28e565a | 6605 | kfree(options); |
d22f76e7 | 6606 | |
dc79b113 | 6607 | return ret; |
a725f65e AE |
6608 | } |
6609 | ||
e010dd0a ID |
6610 | static void rbd_dev_image_unlock(struct rbd_device *rbd_dev) |
6611 | { | |
6612 | down_write(&rbd_dev->lock_rwsem); | |
6613 | if (__rbd_is_lock_owner(rbd_dev)) | |
e1fddc8f | 6614 | __rbd_release_lock(rbd_dev); |
e010dd0a ID |
6615 | up_write(&rbd_dev->lock_rwsem); |
6616 | } | |
6617 | ||
637cd060 ID |
6618 | /* |
6619 | * If the wait is interrupted, an error is returned even if the lock | |
6620 | * was successfully acquired. rbd_dev_image_unlock() will release it | |
6621 | * if needed. | |
6622 | */ | |
e010dd0a ID |
6623 | static int rbd_add_acquire_lock(struct rbd_device *rbd_dev) |
6624 | { | |
637cd060 | 6625 | long ret; |
2f18d466 | 6626 | |
e010dd0a | 6627 | if (!(rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK)) { |
637cd060 ID |
6628 | if (!rbd_dev->opts->exclusive && !rbd_dev->opts->lock_on_read) |
6629 | return 0; | |
6630 | ||
e010dd0a ID |
6631 | rbd_warn(rbd_dev, "exclusive-lock feature is not enabled"); |
6632 | return -EINVAL; | |
6633 | } | |
6634 | ||
637cd060 ID |
6635 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
6636 | return 0; | |
6637 | ||
6638 | rbd_assert(!rbd_is_lock_owner(rbd_dev)); | |
6639 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); | |
6640 | ret = wait_for_completion_killable_timeout(&rbd_dev->acquire_wait, | |
6641 | ceph_timeout_jiffies(rbd_dev->opts->lock_timeout)); | |
6642 | if (ret > 0) | |
6643 | ret = rbd_dev->acquire_err; | |
6644 | else if (!ret) | |
6645 | ret = -ETIMEDOUT; | |
6646 | ||
2f18d466 | 6647 | if (ret) { |
637cd060 ID |
6648 | rbd_warn(rbd_dev, "failed to acquire exclusive lock: %ld", ret); |
6649 | return ret; | |
e010dd0a ID |
6650 | } |
6651 | ||
637cd060 ID |
6652 | /* |
6653 | * The lock may have been released by now, unless automatic lock | |
6654 | * transitions are disabled. | |
6655 | */ | |
6656 | rbd_assert(!rbd_dev->opts->exclusive || rbd_is_lock_owner(rbd_dev)); | |
e010dd0a ID |
6657 | return 0; |
6658 | } | |
6659 | ||
589d30e0 AE |
6660 | /* |
6661 | * An rbd format 2 image has a unique identifier, distinct from the | |
6662 | * name given to it by the user. Internally, that identifier is | |
6663 | * what's used to specify the names of objects related to the image. | |
6664 | * | |
6665 | * A special "rbd id" object is used to map an rbd image name to its | |
6666 | * id. If that object doesn't exist, then there is no v2 rbd image | |
6667 | * with the supplied name. | |
6668 | * | |
6669 | * This function will record the given rbd_dev's image_id field if | |
6670 | * it can be determined, and in that case will return 0. If any | |
6671 | * errors occur a negative errno will be returned and the rbd_dev's | |
6672 | * image_id field will be unchanged (and should be NULL). | |
6673 | */ | |
6674 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
6675 | { | |
6676 | int ret; | |
6677 | size_t size; | |
ecd4a68a | 6678 | CEPH_DEFINE_OID_ONSTACK(oid); |
589d30e0 | 6679 | void *response; |
c0fba368 | 6680 | char *image_id; |
2f82ee54 | 6681 | |
2c0d0a10 AE |
6682 | /* |
6683 | * When probing a parent image, the image id is already | |
6684 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
6685 | * need to fetch the image id again in this case. We |
6686 | * do still need to set the image format though. | |
2c0d0a10 | 6687 | */ |
c0fba368 AE |
6688 | if (rbd_dev->spec->image_id) { |
6689 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
6690 | ||
2c0d0a10 | 6691 | return 0; |
c0fba368 | 6692 | } |
2c0d0a10 | 6693 | |
589d30e0 AE |
6694 | /* |
6695 | * First, see if the format 2 image id file exists, and if | |
6696 | * so, get the image's persistent id from it. | |
6697 | */ | |
ecd4a68a ID |
6698 | ret = ceph_oid_aprintf(&oid, GFP_KERNEL, "%s%s", RBD_ID_PREFIX, |
6699 | rbd_dev->spec->image_name); | |
6700 | if (ret) | |
6701 | return ret; | |
6702 | ||
6703 | dout("rbd id object name is %s\n", oid.name); | |
589d30e0 AE |
6704 | |
6705 | /* Response will be an encoded string, which includes a length */ | |
589d30e0 AE |
6706 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; |
6707 | response = kzalloc(size, GFP_NOIO); | |
6708 | if (!response) { | |
6709 | ret = -ENOMEM; | |
6710 | goto out; | |
6711 | } | |
6712 | ||
c0fba368 AE |
6713 | /* If it doesn't exist we'll assume it's a format 1 image */ |
6714 | ||
ecd4a68a ID |
6715 | ret = rbd_obj_method_sync(rbd_dev, &oid, &rbd_dev->header_oloc, |
6716 | "get_id", NULL, 0, | |
5435d206 | 6717 | response, size); |
36be9a76 | 6718 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
6719 | if (ret == -ENOENT) { |
6720 | image_id = kstrdup("", GFP_KERNEL); | |
6721 | ret = image_id ? 0 : -ENOMEM; | |
6722 | if (!ret) | |
6723 | rbd_dev->image_format = 1; | |
7dd440c9 | 6724 | } else if (ret >= 0) { |
c0fba368 AE |
6725 | void *p = response; |
6726 | ||
6727 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 6728 | NULL, GFP_NOIO); |
461f758a | 6729 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
6730 | if (!ret) |
6731 | rbd_dev->image_format = 2; | |
c0fba368 AE |
6732 | } |
6733 | ||
6734 | if (!ret) { | |
6735 | rbd_dev->spec->image_id = image_id; | |
6736 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
6737 | } |
6738 | out: | |
6739 | kfree(response); | |
ecd4a68a | 6740 | ceph_oid_destroy(&oid); |
589d30e0 AE |
6741 | return ret; |
6742 | } | |
6743 | ||
3abef3b3 AE |
6744 | /* |
6745 | * Undo whatever state changes are made by v1 or v2 header info | |
6746 | * call. | |
6747 | */ | |
6fd48b3b AE |
6748 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
6749 | { | |
6750 | struct rbd_image_header *header; | |
6751 | ||
e69b8d41 | 6752 | rbd_dev_parent_put(rbd_dev); |
22e8bd51 | 6753 | rbd_object_map_free(rbd_dev); |
da5ef6be | 6754 | rbd_dev_mapping_clear(rbd_dev); |
6fd48b3b AE |
6755 | |
6756 | /* Free dynamic fields from the header, then zero it out */ | |
6757 | ||
6758 | header = &rbd_dev->header; | |
812164f8 | 6759 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
6760 | kfree(header->snap_sizes); |
6761 | kfree(header->snap_names); | |
6762 | kfree(header->object_prefix); | |
6763 | memset(header, 0, sizeof (*header)); | |
6764 | } | |
6765 | ||
2df3fac7 | 6766 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
6767 | { |
6768 | int ret; | |
a30b71b9 | 6769 | |
1e130199 | 6770 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 6771 | if (ret) |
b1b5402a AE |
6772 | goto out_err; |
6773 | ||
2df3fac7 AE |
6774 | /* |
6775 | * Get the and check features for the image. Currently the | |
6776 | * features are assumed to never change. | |
6777 | */ | |
b1b5402a | 6778 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 6779 | if (ret) |
9d475de5 | 6780 | goto out_err; |
35d489f9 | 6781 | |
cc070d59 AE |
6782 | /* If the image supports fancy striping, get its parameters */ |
6783 | ||
6784 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
6785 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
6786 | if (ret < 0) | |
6787 | goto out_err; | |
6788 | } | |
a30b71b9 | 6789 | |
7e97332e ID |
6790 | if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) { |
6791 | ret = rbd_dev_v2_data_pool(rbd_dev); | |
6792 | if (ret) | |
6793 | goto out_err; | |
6794 | } | |
6795 | ||
263423f8 | 6796 | rbd_init_layout(rbd_dev); |
35152979 | 6797 | return 0; |
263423f8 | 6798 | |
9d475de5 | 6799 | out_err: |
642a2537 | 6800 | rbd_dev->header.features = 0; |
1e130199 AE |
6801 | kfree(rbd_dev->header.object_prefix); |
6802 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 | 6803 | return ret; |
a30b71b9 AE |
6804 | } |
6805 | ||
6d69bb53 ID |
6806 | /* |
6807 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
6808 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
6809 | * length of the already discovered part of the parent chain. | |
6810 | */ | |
6811 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 6812 | { |
2f82ee54 | 6813 | struct rbd_device *parent = NULL; |
124afba2 AE |
6814 | int ret; |
6815 | ||
6816 | if (!rbd_dev->parent_spec) | |
6817 | return 0; | |
124afba2 | 6818 | |
6d69bb53 ID |
6819 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
6820 | pr_info("parent chain is too long (%d)\n", depth); | |
6821 | ret = -EINVAL; | |
6822 | goto out_err; | |
6823 | } | |
6824 | ||
1643dfa4 | 6825 | parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec); |
1f2c6651 ID |
6826 | if (!parent) { |
6827 | ret = -ENOMEM; | |
124afba2 | 6828 | goto out_err; |
1f2c6651 ID |
6829 | } |
6830 | ||
6831 | /* | |
6832 | * Images related by parent/child relationships always share | |
6833 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
6834 | */ | |
6835 | __rbd_get_client(rbd_dev->rbd_client); | |
6836 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 6837 | |
6d69bb53 | 6838 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
6839 | if (ret < 0) |
6840 | goto out_err; | |
1f2c6651 | 6841 | |
124afba2 | 6842 | rbd_dev->parent = parent; |
a2acd00e | 6843 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 6844 | return 0; |
1f2c6651 | 6845 | |
124afba2 | 6846 | out_err: |
1f2c6651 | 6847 | rbd_dev_unparent(rbd_dev); |
1761b229 | 6848 | rbd_dev_destroy(parent); |
124afba2 AE |
6849 | return ret; |
6850 | } | |
6851 | ||
5769ed0c ID |
6852 | static void rbd_dev_device_release(struct rbd_device *rbd_dev) |
6853 | { | |
6854 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
5769ed0c ID |
6855 | rbd_free_disk(rbd_dev); |
6856 | if (!single_major) | |
6857 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
6858 | } | |
6859 | ||
811c6688 ID |
6860 | /* |
6861 | * rbd_dev->header_rwsem must be locked for write and will be unlocked | |
6862 | * upon return. | |
6863 | */ | |
200a6a8b | 6864 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 6865 | { |
83a06263 | 6866 | int ret; |
d1cf5788 | 6867 | |
9b60e70b | 6868 | /* Record our major and minor device numbers. */ |
83a06263 | 6869 | |
9b60e70b ID |
6870 | if (!single_major) { |
6871 | ret = register_blkdev(0, rbd_dev->name); | |
6872 | if (ret < 0) | |
1643dfa4 | 6873 | goto err_out_unlock; |
9b60e70b ID |
6874 | |
6875 | rbd_dev->major = ret; | |
6876 | rbd_dev->minor = 0; | |
6877 | } else { | |
6878 | rbd_dev->major = rbd_major; | |
6879 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
6880 | } | |
83a06263 AE |
6881 | |
6882 | /* Set up the blkdev mapping. */ | |
6883 | ||
6884 | ret = rbd_init_disk(rbd_dev); | |
6885 | if (ret) | |
6886 | goto err_out_blkdev; | |
6887 | ||
f35a4dee | 6888 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
9568c93e | 6889 | set_disk_ro(rbd_dev->disk, rbd_dev->opts->read_only); |
f35a4dee | 6890 | |
5769ed0c | 6891 | ret = dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id); |
f35a4dee | 6892 | if (ret) |
da5ef6be | 6893 | goto err_out_disk; |
83a06263 | 6894 | |
129b79d4 | 6895 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
811c6688 | 6896 | up_write(&rbd_dev->header_rwsem); |
5769ed0c | 6897 | return 0; |
2f82ee54 | 6898 | |
83a06263 AE |
6899 | err_out_disk: |
6900 | rbd_free_disk(rbd_dev); | |
6901 | err_out_blkdev: | |
9b60e70b ID |
6902 | if (!single_major) |
6903 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
811c6688 ID |
6904 | err_out_unlock: |
6905 | up_write(&rbd_dev->header_rwsem); | |
83a06263 AE |
6906 | return ret; |
6907 | } | |
6908 | ||
332bb12d AE |
6909 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
6910 | { | |
6911 | struct rbd_spec *spec = rbd_dev->spec; | |
c41d13a3 | 6912 | int ret; |
332bb12d AE |
6913 | |
6914 | /* Record the header object name for this rbd image. */ | |
6915 | ||
6916 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
332bb12d | 6917 | if (rbd_dev->image_format == 1) |
c41d13a3 ID |
6918 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
6919 | spec->image_name, RBD_SUFFIX); | |
332bb12d | 6920 | else |
c41d13a3 ID |
6921 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
6922 | RBD_HEADER_PREFIX, spec->image_id); | |
332bb12d | 6923 | |
c41d13a3 | 6924 | return ret; |
332bb12d AE |
6925 | } |
6926 | ||
200a6a8b AE |
6927 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
6928 | { | |
6fd48b3b | 6929 | rbd_dev_unprobe(rbd_dev); |
fd22aef8 ID |
6930 | if (rbd_dev->opts) |
6931 | rbd_unregister_watch(rbd_dev); | |
6fd48b3b AE |
6932 | rbd_dev->image_format = 0; |
6933 | kfree(rbd_dev->spec->image_id); | |
6934 | rbd_dev->spec->image_id = NULL; | |
200a6a8b AE |
6935 | } |
6936 | ||
a30b71b9 AE |
6937 | /* |
6938 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
6939 | * device. If this image is the one being mapped (i.e., not a |
6940 | * parent), initiate a watch on its header object before using that | |
6941 | * object to get detailed information about the rbd image. | |
a30b71b9 | 6942 | */ |
6d69bb53 | 6943 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth) |
a30b71b9 AE |
6944 | { |
6945 | int ret; | |
6946 | ||
6947 | /* | |
3abef3b3 AE |
6948 | * Get the id from the image id object. Unless there's an |
6949 | * error, rbd_dev->spec->image_id will be filled in with | |
6950 | * a dynamically-allocated string, and rbd_dev->image_format | |
6951 | * will be set to either 1 or 2. | |
a30b71b9 AE |
6952 | */ |
6953 | ret = rbd_dev_image_id(rbd_dev); | |
6954 | if (ret) | |
c0fba368 | 6955 | return ret; |
c0fba368 | 6956 | |
332bb12d AE |
6957 | ret = rbd_dev_header_name(rbd_dev); |
6958 | if (ret) | |
6959 | goto err_out_format; | |
6960 | ||
6d69bb53 | 6961 | if (!depth) { |
99d16943 | 6962 | ret = rbd_register_watch(rbd_dev); |
1fe48023 ID |
6963 | if (ret) { |
6964 | if (ret == -ENOENT) | |
b26c047b | 6965 | pr_info("image %s/%s%s%s does not exist\n", |
1fe48023 | 6966 | rbd_dev->spec->pool_name, |
b26c047b ID |
6967 | rbd_dev->spec->pool_ns ?: "", |
6968 | rbd_dev->spec->pool_ns ? "/" : "", | |
1fe48023 | 6969 | rbd_dev->spec->image_name); |
c41d13a3 | 6970 | goto err_out_format; |
1fe48023 | 6971 | } |
1f3ef788 | 6972 | } |
b644de2b | 6973 | |
a720ae09 | 6974 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 6975 | if (ret) |
b644de2b | 6976 | goto err_out_watch; |
83a06263 | 6977 | |
04077599 ID |
6978 | /* |
6979 | * If this image is the one being mapped, we have pool name and | |
6980 | * id, image name and id, and snap name - need to fill snap id. | |
6981 | * Otherwise this is a parent image, identified by pool, image | |
6982 | * and snap ids - need to fill in names for those ids. | |
6983 | */ | |
6d69bb53 | 6984 | if (!depth) |
04077599 ID |
6985 | ret = rbd_spec_fill_snap_id(rbd_dev); |
6986 | else | |
6987 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
6988 | if (ret) { |
6989 | if (ret == -ENOENT) | |
b26c047b | 6990 | pr_info("snap %s/%s%s%s@%s does not exist\n", |
1fe48023 | 6991 | rbd_dev->spec->pool_name, |
b26c047b ID |
6992 | rbd_dev->spec->pool_ns ?: "", |
6993 | rbd_dev->spec->pool_ns ? "/" : "", | |
1fe48023 ID |
6994 | rbd_dev->spec->image_name, |
6995 | rbd_dev->spec->snap_name); | |
33dca39f | 6996 | goto err_out_probe; |
1fe48023 | 6997 | } |
9bb81c9b | 6998 | |
da5ef6be ID |
6999 | ret = rbd_dev_mapping_set(rbd_dev); |
7000 | if (ret) | |
7001 | goto err_out_probe; | |
7002 | ||
22e8bd51 ID |
7003 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && |
7004 | (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP)) { | |
7005 | ret = rbd_object_map_load(rbd_dev); | |
7006 | if (ret) | |
7007 | goto err_out_probe; | |
7008 | } | |
7009 | ||
e8f59b59 ID |
7010 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
7011 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
7012 | if (ret) | |
7013 | goto err_out_probe; | |
e8f59b59 ID |
7014 | } |
7015 | ||
6d69bb53 | 7016 | ret = rbd_dev_probe_parent(rbd_dev, depth); |
30d60ba2 AE |
7017 | if (ret) |
7018 | goto err_out_probe; | |
7019 | ||
7020 | dout("discovered format %u image, header name is %s\n", | |
c41d13a3 | 7021 | rbd_dev->image_format, rbd_dev->header_oid.name); |
30d60ba2 | 7022 | return 0; |
e8f59b59 | 7023 | |
6fd48b3b AE |
7024 | err_out_probe: |
7025 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 7026 | err_out_watch: |
6d69bb53 | 7027 | if (!depth) |
99d16943 | 7028 | rbd_unregister_watch(rbd_dev); |
332bb12d AE |
7029 | err_out_format: |
7030 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
7031 | kfree(rbd_dev->spec->image_id); |
7032 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
7033 | return ret; |
7034 | } | |
7035 | ||
9b60e70b ID |
7036 | static ssize_t do_rbd_add(struct bus_type *bus, |
7037 | const char *buf, | |
7038 | size_t count) | |
602adf40 | 7039 | { |
cb8627c7 | 7040 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 7041 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 7042 | struct rbd_options *rbd_opts = NULL; |
859c31df | 7043 | struct rbd_spec *spec = NULL; |
9d3997fd | 7044 | struct rbd_client *rbdc; |
b51c83c2 | 7045 | int rc; |
602adf40 YS |
7046 | |
7047 | if (!try_module_get(THIS_MODULE)) | |
7048 | return -ENODEV; | |
7049 | ||
602adf40 | 7050 | /* parse add command */ |
859c31df | 7051 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 7052 | if (rc < 0) |
dd5ac32d | 7053 | goto out; |
78cea76e | 7054 | |
9d3997fd AE |
7055 | rbdc = rbd_get_client(ceph_opts); |
7056 | if (IS_ERR(rbdc)) { | |
7057 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 7058 | goto err_out_args; |
9d3997fd | 7059 | } |
602adf40 | 7060 | |
602adf40 | 7061 | /* pick the pool */ |
dd435855 | 7062 | rc = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, spec->pool_name); |
1fe48023 ID |
7063 | if (rc < 0) { |
7064 | if (rc == -ENOENT) | |
7065 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 7066 | goto err_out_client; |
1fe48023 | 7067 | } |
c0cd10db | 7068 | spec->pool_id = (u64)rc; |
859c31df | 7069 | |
d147543d | 7070 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
b51c83c2 ID |
7071 | if (!rbd_dev) { |
7072 | rc = -ENOMEM; | |
bd4ba655 | 7073 | goto err_out_client; |
b51c83c2 | 7074 | } |
c53d5893 AE |
7075 | rbdc = NULL; /* rbd_dev now owns this */ |
7076 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 7077 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 7078 | |
0d6d1e9c MC |
7079 | rbd_dev->config_info = kstrdup(buf, GFP_KERNEL); |
7080 | if (!rbd_dev->config_info) { | |
7081 | rc = -ENOMEM; | |
7082 | goto err_out_rbd_dev; | |
7083 | } | |
7084 | ||
811c6688 | 7085 | down_write(&rbd_dev->header_rwsem); |
6d69bb53 | 7086 | rc = rbd_dev_image_probe(rbd_dev, 0); |
0d6d1e9c MC |
7087 | if (rc < 0) { |
7088 | up_write(&rbd_dev->header_rwsem); | |
c53d5893 | 7089 | goto err_out_rbd_dev; |
0d6d1e9c | 7090 | } |
05fd6f6f | 7091 | |
7ce4eef7 | 7092 | /* If we are mapping a snapshot it must be marked read-only */ |
7ce4eef7 | 7093 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9568c93e | 7094 | rbd_dev->opts->read_only = true; |
7ce4eef7 | 7095 | |
0c93e1b7 ID |
7096 | if (rbd_dev->opts->alloc_size > rbd_dev->layout.object_size) { |
7097 | rbd_warn(rbd_dev, "alloc_size adjusted to %u", | |
7098 | rbd_dev->layout.object_size); | |
7099 | rbd_dev->opts->alloc_size = rbd_dev->layout.object_size; | |
7100 | } | |
7101 | ||
b536f69a | 7102 | rc = rbd_dev_device_setup(rbd_dev); |
fd22aef8 | 7103 | if (rc) |
8b679ec5 | 7104 | goto err_out_image_probe; |
3abef3b3 | 7105 | |
637cd060 ID |
7106 | rc = rbd_add_acquire_lock(rbd_dev); |
7107 | if (rc) | |
7108 | goto err_out_image_lock; | |
3abef3b3 | 7109 | |
5769ed0c ID |
7110 | /* Everything's ready. Announce the disk to the world. */ |
7111 | ||
7112 | rc = device_add(&rbd_dev->dev); | |
7113 | if (rc) | |
e010dd0a | 7114 | goto err_out_image_lock; |
5769ed0c ID |
7115 | |
7116 | add_disk(rbd_dev->disk); | |
7117 | /* see rbd_init_disk() */ | |
7118 | blk_put_queue(rbd_dev->disk->queue); | |
7119 | ||
7120 | spin_lock(&rbd_dev_list_lock); | |
7121 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
7122 | spin_unlock(&rbd_dev_list_lock); | |
7123 | ||
7124 | pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name, | |
7125 | (unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT, | |
7126 | rbd_dev->header.features); | |
dd5ac32d ID |
7127 | rc = count; |
7128 | out: | |
7129 | module_put(THIS_MODULE); | |
7130 | return rc; | |
b536f69a | 7131 | |
e010dd0a ID |
7132 | err_out_image_lock: |
7133 | rbd_dev_image_unlock(rbd_dev); | |
5769ed0c | 7134 | rbd_dev_device_release(rbd_dev); |
8b679ec5 ID |
7135 | err_out_image_probe: |
7136 | rbd_dev_image_release(rbd_dev); | |
c53d5893 AE |
7137 | err_out_rbd_dev: |
7138 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 7139 | err_out_client: |
9d3997fd | 7140 | rbd_put_client(rbdc); |
0ddebc0c | 7141 | err_out_args: |
859c31df | 7142 | rbd_spec_put(spec); |
d147543d | 7143 | kfree(rbd_opts); |
dd5ac32d | 7144 | goto out; |
602adf40 YS |
7145 | } |
7146 | ||
7e9586ba | 7147 | static ssize_t add_store(struct bus_type *bus, const char *buf, size_t count) |
9b60e70b ID |
7148 | { |
7149 | if (single_major) | |
7150 | return -EINVAL; | |
7151 | ||
7152 | return do_rbd_add(bus, buf, count); | |
7153 | } | |
7154 | ||
7e9586ba GKH |
7155 | static ssize_t add_single_major_store(struct bus_type *bus, const char *buf, |
7156 | size_t count) | |
9b60e70b ID |
7157 | { |
7158 | return do_rbd_add(bus, buf, count); | |
7159 | } | |
7160 | ||
05a46afd AE |
7161 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
7162 | { | |
ad945fc1 | 7163 | while (rbd_dev->parent) { |
05a46afd AE |
7164 | struct rbd_device *first = rbd_dev; |
7165 | struct rbd_device *second = first->parent; | |
7166 | struct rbd_device *third; | |
7167 | ||
7168 | /* | |
7169 | * Follow to the parent with no grandparent and | |
7170 | * remove it. | |
7171 | */ | |
7172 | while (second && (third = second->parent)) { | |
7173 | first = second; | |
7174 | second = third; | |
7175 | } | |
ad945fc1 | 7176 | rbd_assert(second); |
8ad42cd0 | 7177 | rbd_dev_image_release(second); |
8b679ec5 | 7178 | rbd_dev_destroy(second); |
ad945fc1 AE |
7179 | first->parent = NULL; |
7180 | first->parent_overlap = 0; | |
7181 | ||
7182 | rbd_assert(first->parent_spec); | |
05a46afd AE |
7183 | rbd_spec_put(first->parent_spec); |
7184 | first->parent_spec = NULL; | |
05a46afd AE |
7185 | } |
7186 | } | |
7187 | ||
9b60e70b ID |
7188 | static ssize_t do_rbd_remove(struct bus_type *bus, |
7189 | const char *buf, | |
7190 | size_t count) | |
602adf40 YS |
7191 | { |
7192 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
7193 | struct list_head *tmp; |
7194 | int dev_id; | |
0276dca6 | 7195 | char opt_buf[6]; |
0276dca6 | 7196 | bool force = false; |
0d8189e1 | 7197 | int ret; |
602adf40 | 7198 | |
0276dca6 MC |
7199 | dev_id = -1; |
7200 | opt_buf[0] = '\0'; | |
7201 | sscanf(buf, "%d %5s", &dev_id, opt_buf); | |
7202 | if (dev_id < 0) { | |
7203 | pr_err("dev_id out of range\n"); | |
602adf40 | 7204 | return -EINVAL; |
0276dca6 MC |
7205 | } |
7206 | if (opt_buf[0] != '\0') { | |
7207 | if (!strcmp(opt_buf, "force")) { | |
7208 | force = true; | |
7209 | } else { | |
7210 | pr_err("bad remove option at '%s'\n", opt_buf); | |
7211 | return -EINVAL; | |
7212 | } | |
7213 | } | |
602adf40 | 7214 | |
751cc0e3 AE |
7215 | ret = -ENOENT; |
7216 | spin_lock(&rbd_dev_list_lock); | |
7217 | list_for_each(tmp, &rbd_dev_list) { | |
7218 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
7219 | if (rbd_dev->dev_id == dev_id) { | |
7220 | ret = 0; | |
7221 | break; | |
7222 | } | |
42382b70 | 7223 | } |
751cc0e3 AE |
7224 | if (!ret) { |
7225 | spin_lock_irq(&rbd_dev->lock); | |
0276dca6 | 7226 | if (rbd_dev->open_count && !force) |
751cc0e3 | 7227 | ret = -EBUSY; |
85f5a4d6 ID |
7228 | else if (test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
7229 | &rbd_dev->flags)) | |
7230 | ret = -EINPROGRESS; | |
751cc0e3 AE |
7231 | spin_unlock_irq(&rbd_dev->lock); |
7232 | } | |
7233 | spin_unlock(&rbd_dev_list_lock); | |
85f5a4d6 | 7234 | if (ret) |
1ba0f1e7 | 7235 | return ret; |
751cc0e3 | 7236 | |
0276dca6 MC |
7237 | if (force) { |
7238 | /* | |
7239 | * Prevent new IO from being queued and wait for existing | |
7240 | * IO to complete/fail. | |
7241 | */ | |
7242 | blk_mq_freeze_queue(rbd_dev->disk->queue); | |
7243 | blk_set_queue_dying(rbd_dev->disk->queue); | |
7244 | } | |
7245 | ||
5769ed0c ID |
7246 | del_gendisk(rbd_dev->disk); |
7247 | spin_lock(&rbd_dev_list_lock); | |
7248 | list_del_init(&rbd_dev->node); | |
7249 | spin_unlock(&rbd_dev_list_lock); | |
7250 | device_del(&rbd_dev->dev); | |
fca27065 | 7251 | |
e010dd0a | 7252 | rbd_dev_image_unlock(rbd_dev); |
dd5ac32d | 7253 | rbd_dev_device_release(rbd_dev); |
8ad42cd0 | 7254 | rbd_dev_image_release(rbd_dev); |
8b679ec5 | 7255 | rbd_dev_destroy(rbd_dev); |
1ba0f1e7 | 7256 | return count; |
602adf40 YS |
7257 | } |
7258 | ||
7e9586ba | 7259 | static ssize_t remove_store(struct bus_type *bus, const char *buf, size_t count) |
9b60e70b ID |
7260 | { |
7261 | if (single_major) | |
7262 | return -EINVAL; | |
7263 | ||
7264 | return do_rbd_remove(bus, buf, count); | |
7265 | } | |
7266 | ||
7e9586ba GKH |
7267 | static ssize_t remove_single_major_store(struct bus_type *bus, const char *buf, |
7268 | size_t count) | |
9b60e70b ID |
7269 | { |
7270 | return do_rbd_remove(bus, buf, count); | |
7271 | } | |
7272 | ||
602adf40 YS |
7273 | /* |
7274 | * create control files in sysfs | |
dfc5606d | 7275 | * /sys/bus/rbd/... |
602adf40 | 7276 | */ |
7d8dc534 | 7277 | static int __init rbd_sysfs_init(void) |
602adf40 | 7278 | { |
dfc5606d | 7279 | int ret; |
602adf40 | 7280 | |
fed4c143 | 7281 | ret = device_register(&rbd_root_dev); |
21079786 | 7282 | if (ret < 0) |
dfc5606d | 7283 | return ret; |
602adf40 | 7284 | |
fed4c143 AE |
7285 | ret = bus_register(&rbd_bus_type); |
7286 | if (ret < 0) | |
7287 | device_unregister(&rbd_root_dev); | |
602adf40 | 7288 | |
602adf40 YS |
7289 | return ret; |
7290 | } | |
7291 | ||
7d8dc534 | 7292 | static void __exit rbd_sysfs_cleanup(void) |
602adf40 | 7293 | { |
dfc5606d | 7294 | bus_unregister(&rbd_bus_type); |
fed4c143 | 7295 | device_unregister(&rbd_root_dev); |
602adf40 YS |
7296 | } |
7297 | ||
7d8dc534 | 7298 | static int __init rbd_slab_init(void) |
1c2a9dfe AE |
7299 | { |
7300 | rbd_assert(!rbd_img_request_cache); | |
03d94406 | 7301 | rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0); |
868311b1 AE |
7302 | if (!rbd_img_request_cache) |
7303 | return -ENOMEM; | |
7304 | ||
7305 | rbd_assert(!rbd_obj_request_cache); | |
03d94406 | 7306 | rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0); |
78c2a44a AE |
7307 | if (!rbd_obj_request_cache) |
7308 | goto out_err; | |
7309 | ||
6c696d85 | 7310 | return 0; |
1c2a9dfe | 7311 | |
6c696d85 | 7312 | out_err: |
868311b1 AE |
7313 | kmem_cache_destroy(rbd_img_request_cache); |
7314 | rbd_img_request_cache = NULL; | |
1c2a9dfe AE |
7315 | return -ENOMEM; |
7316 | } | |
7317 | ||
7318 | static void rbd_slab_exit(void) | |
7319 | { | |
868311b1 AE |
7320 | rbd_assert(rbd_obj_request_cache); |
7321 | kmem_cache_destroy(rbd_obj_request_cache); | |
7322 | rbd_obj_request_cache = NULL; | |
7323 | ||
1c2a9dfe AE |
7324 | rbd_assert(rbd_img_request_cache); |
7325 | kmem_cache_destroy(rbd_img_request_cache); | |
7326 | rbd_img_request_cache = NULL; | |
7327 | } | |
7328 | ||
cc344fa1 | 7329 | static int __init rbd_init(void) |
602adf40 YS |
7330 | { |
7331 | int rc; | |
7332 | ||
1e32d34c AE |
7333 | if (!libceph_compatible(NULL)) { |
7334 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
7335 | return -EINVAL; |
7336 | } | |
e1b4d96d | 7337 | |
1c2a9dfe | 7338 | rc = rbd_slab_init(); |
602adf40 YS |
7339 | if (rc) |
7340 | return rc; | |
e1b4d96d | 7341 | |
f5ee37bd ID |
7342 | /* |
7343 | * The number of active work items is limited by the number of | |
f77303bd | 7344 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
7345 | */ |
7346 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
7347 | if (!rbd_wq) { | |
7348 | rc = -ENOMEM; | |
7349 | goto err_out_slab; | |
7350 | } | |
7351 | ||
9b60e70b ID |
7352 | if (single_major) { |
7353 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
7354 | if (rbd_major < 0) { | |
7355 | rc = rbd_major; | |
f5ee37bd | 7356 | goto err_out_wq; |
9b60e70b ID |
7357 | } |
7358 | } | |
7359 | ||
1c2a9dfe AE |
7360 | rc = rbd_sysfs_init(); |
7361 | if (rc) | |
9b60e70b ID |
7362 | goto err_out_blkdev; |
7363 | ||
7364 | if (single_major) | |
7365 | pr_info("loaded (major %d)\n", rbd_major); | |
7366 | else | |
7367 | pr_info("loaded\n"); | |
1c2a9dfe | 7368 | |
e1b4d96d ID |
7369 | return 0; |
7370 | ||
9b60e70b ID |
7371 | err_out_blkdev: |
7372 | if (single_major) | |
7373 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
7374 | err_out_wq: |
7375 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
7376 | err_out_slab: |
7377 | rbd_slab_exit(); | |
1c2a9dfe | 7378 | return rc; |
602adf40 YS |
7379 | } |
7380 | ||
cc344fa1 | 7381 | static void __exit rbd_exit(void) |
602adf40 | 7382 | { |
ffe312cf | 7383 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 7384 | rbd_sysfs_cleanup(); |
9b60e70b ID |
7385 | if (single_major) |
7386 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 7387 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 7388 | rbd_slab_exit(); |
602adf40 YS |
7389 | } |
7390 | ||
7391 | module_init(rbd_init); | |
7392 | module_exit(rbd_exit); | |
7393 | ||
d552c619 | 7394 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
7395 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
7396 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
7397 | /* following authorship retained from original osdblk.c */ |
7398 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
7399 | ||
90da258b | 7400 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 7401 | MODULE_LICENSE("GPL"); |