Commit | Line | Data |
---|---|---|
e2a58ee5 | 1 | |
602adf40 YS |
2 | /* |
3 | rbd.c -- Export ceph rados objects as a Linux block device | |
4 | ||
5 | ||
6 | based on drivers/block/osdblk.c: | |
7 | ||
8 | Copyright 2009 Red Hat, Inc. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; see the file COPYING. If not, write to | |
21 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
22 | ||
23 | ||
24 | ||
dfc5606d | 25 | For usage instructions, please refer to: |
602adf40 | 26 | |
dfc5606d | 27 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
28 | |
29 | */ | |
30 | ||
31 | #include <linux/ceph/libceph.h> | |
32 | #include <linux/ceph/osd_client.h> | |
33 | #include <linux/ceph/mon_client.h> | |
ed95b21a | 34 | #include <linux/ceph/cls_lock_client.h> |
602adf40 | 35 | #include <linux/ceph/decode.h> |
59c2be1e | 36 | #include <linux/parser.h> |
30d1cff8 | 37 | #include <linux/bsearch.h> |
602adf40 YS |
38 | |
39 | #include <linux/kernel.h> | |
40 | #include <linux/device.h> | |
41 | #include <linux/module.h> | |
7ad18afa | 42 | #include <linux/blk-mq.h> |
602adf40 YS |
43 | #include <linux/fs.h> |
44 | #include <linux/blkdev.h> | |
1c2a9dfe | 45 | #include <linux/slab.h> |
f8a22fc2 | 46 | #include <linux/idr.h> |
bc1ecc65 | 47 | #include <linux/workqueue.h> |
602adf40 YS |
48 | |
49 | #include "rbd_types.h" | |
50 | ||
aafb230e AE |
51 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
52 | ||
593a9e7b AE |
53 | /* |
54 | * The basic unit of block I/O is a sector. It is interpreted in a | |
55 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
56 | * universally 512 bytes. These symbols are just slightly more | |
57 | * meaningful than the bare numbers they represent. | |
58 | */ | |
59 | #define SECTOR_SHIFT 9 | |
60 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
61 | ||
a2acd00e AE |
62 | /* |
63 | * Increment the given counter and return its updated value. | |
64 | * If the counter is already 0 it will not be incremented. | |
65 | * If the counter is already at its maximum value returns | |
66 | * -EINVAL without updating it. | |
67 | */ | |
68 | static int atomic_inc_return_safe(atomic_t *v) | |
69 | { | |
70 | unsigned int counter; | |
71 | ||
72 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
73 | if (counter <= (unsigned int)INT_MAX) | |
74 | return (int)counter; | |
75 | ||
76 | atomic_dec(v); | |
77 | ||
78 | return -EINVAL; | |
79 | } | |
80 | ||
81 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
82 | static int atomic_dec_return_safe(atomic_t *v) | |
83 | { | |
84 | int counter; | |
85 | ||
86 | counter = atomic_dec_return(v); | |
87 | if (counter >= 0) | |
88 | return counter; | |
89 | ||
90 | atomic_inc(v); | |
91 | ||
92 | return -EINVAL; | |
93 | } | |
94 | ||
f0f8cef5 | 95 | #define RBD_DRV_NAME "rbd" |
602adf40 | 96 | |
7e513d43 ID |
97 | #define RBD_MINORS_PER_MAJOR 256 |
98 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 99 | |
6d69bb53 ID |
100 | #define RBD_MAX_PARENT_CHAIN_LEN 16 |
101 | ||
d4b125e9 AE |
102 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
103 | #define RBD_MAX_SNAP_NAME_LEN \ | |
104 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
105 | ||
35d489f9 | 106 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
107 | |
108 | #define RBD_SNAP_HEAD_NAME "-" | |
109 | ||
9682fc6d AE |
110 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
111 | ||
9e15b77d AE |
112 | /* This allows a single page to hold an image name sent by OSD */ |
113 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 114 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 115 | |
1e130199 | 116 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 117 | |
ed95b21a | 118 | #define RBD_NOTIFY_TIMEOUT 5 /* seconds */ |
99d16943 ID |
119 | #define RBD_RETRY_DELAY msecs_to_jiffies(1000) |
120 | ||
d889140c AE |
121 | /* Feature bits */ |
122 | ||
5cbf6f12 AE |
123 | #define RBD_FEATURE_LAYERING (1<<0) |
124 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
ed95b21a ID |
125 | #define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2) |
126 | #define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \ | |
127 | RBD_FEATURE_STRIPINGV2 | \ | |
128 | RBD_FEATURE_EXCLUSIVE_LOCK) | |
d889140c AE |
129 | |
130 | /* Features supported by this (client software) implementation. */ | |
131 | ||
770eba6e | 132 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 133 | |
81a89793 AE |
134 | /* |
135 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
136 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
81a89793 | 137 | */ |
602adf40 YS |
138 | #define DEV_NAME_LEN 32 |
139 | ||
140 | /* | |
141 | * block device image metadata (in-memory version) | |
142 | */ | |
143 | struct rbd_image_header { | |
f35a4dee | 144 | /* These six fields never change for a given rbd image */ |
849b4260 | 145 | char *object_prefix; |
602adf40 YS |
146 | __u8 obj_order; |
147 | __u8 crypt_type; | |
148 | __u8 comp_type; | |
f35a4dee AE |
149 | u64 stripe_unit; |
150 | u64 stripe_count; | |
151 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 152 | |
f84344f3 AE |
153 | /* The remaining fields need to be updated occasionally */ |
154 | u64 image_size; | |
155 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
156 | char *snap_names; /* format 1 only */ |
157 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
158 | }; |
159 | ||
0d7dbfce AE |
160 | /* |
161 | * An rbd image specification. | |
162 | * | |
163 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
164 | * identify an image. Each rbd_dev structure includes a pointer to |
165 | * an rbd_spec structure that encapsulates this identity. | |
166 | * | |
167 | * Each of the id's in an rbd_spec has an associated name. For a | |
168 | * user-mapped image, the names are supplied and the id's associated | |
169 | * with them are looked up. For a layered image, a parent image is | |
170 | * defined by the tuple, and the names are looked up. | |
171 | * | |
172 | * An rbd_dev structure contains a parent_spec pointer which is | |
173 | * non-null if the image it represents is a child in a layered | |
174 | * image. This pointer will refer to the rbd_spec structure used | |
175 | * by the parent rbd_dev for its own identity (i.e., the structure | |
176 | * is shared between the parent and child). | |
177 | * | |
178 | * Since these structures are populated once, during the discovery | |
179 | * phase of image construction, they are effectively immutable so | |
180 | * we make no effort to synchronize access to them. | |
181 | * | |
182 | * Note that code herein does not assume the image name is known (it | |
183 | * could be a null pointer). | |
0d7dbfce AE |
184 | */ |
185 | struct rbd_spec { | |
186 | u64 pool_id; | |
ecb4dc22 | 187 | const char *pool_name; |
0d7dbfce | 188 | |
ecb4dc22 AE |
189 | const char *image_id; |
190 | const char *image_name; | |
0d7dbfce AE |
191 | |
192 | u64 snap_id; | |
ecb4dc22 | 193 | const char *snap_name; |
0d7dbfce AE |
194 | |
195 | struct kref kref; | |
196 | }; | |
197 | ||
602adf40 | 198 | /* |
f0f8cef5 | 199 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
200 | */ |
201 | struct rbd_client { | |
202 | struct ceph_client *client; | |
203 | struct kref kref; | |
204 | struct list_head node; | |
205 | }; | |
206 | ||
bf0d5f50 AE |
207 | struct rbd_img_request; |
208 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
209 | ||
210 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
211 | ||
212 | struct rbd_obj_request; | |
213 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
214 | ||
9969ebc5 AE |
215 | enum obj_request_type { |
216 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
217 | }; | |
bf0d5f50 | 218 | |
6d2940c8 GZ |
219 | enum obj_operation_type { |
220 | OBJ_OP_WRITE, | |
221 | OBJ_OP_READ, | |
90e98c52 | 222 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
223 | }; |
224 | ||
926f9b3f AE |
225 | enum obj_req_flags { |
226 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 227 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
228 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
229 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
230 | }; |
231 | ||
bf0d5f50 AE |
232 | struct rbd_obj_request { |
233 | const char *object_name; | |
234 | u64 offset; /* object start byte */ | |
235 | u64 length; /* bytes from offset */ | |
926f9b3f | 236 | unsigned long flags; |
bf0d5f50 | 237 | |
c5b5ef6c AE |
238 | /* |
239 | * An object request associated with an image will have its | |
240 | * img_data flag set; a standalone object request will not. | |
241 | * | |
242 | * A standalone object request will have which == BAD_WHICH | |
243 | * and a null obj_request pointer. | |
244 | * | |
245 | * An object request initiated in support of a layered image | |
246 | * object (to check for its existence before a write) will | |
247 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
248 | * | |
249 | * Finally, an object request for rbd image data will have | |
250 | * which != BAD_WHICH, and will have a non-null img_request | |
251 | * pointer. The value of which will be in the range | |
252 | * 0..(img_request->obj_request_count-1). | |
253 | */ | |
254 | union { | |
255 | struct rbd_obj_request *obj_request; /* STAT op */ | |
256 | struct { | |
257 | struct rbd_img_request *img_request; | |
258 | u64 img_offset; | |
259 | /* links for img_request->obj_requests list */ | |
260 | struct list_head links; | |
261 | }; | |
262 | }; | |
bf0d5f50 AE |
263 | u32 which; /* posn image request list */ |
264 | ||
265 | enum obj_request_type type; | |
788e2df3 AE |
266 | union { |
267 | struct bio *bio_list; | |
268 | struct { | |
269 | struct page **pages; | |
270 | u32 page_count; | |
271 | }; | |
272 | }; | |
0eefd470 | 273 | struct page **copyup_pages; |
ebda6408 | 274 | u32 copyup_page_count; |
bf0d5f50 AE |
275 | |
276 | struct ceph_osd_request *osd_req; | |
277 | ||
278 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 279 | int result; |
bf0d5f50 AE |
280 | |
281 | rbd_obj_callback_t callback; | |
788e2df3 | 282 | struct completion completion; |
bf0d5f50 AE |
283 | |
284 | struct kref kref; | |
285 | }; | |
286 | ||
0c425248 | 287 | enum img_req_flags { |
9849e986 AE |
288 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
289 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 290 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 291 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
292 | }; |
293 | ||
bf0d5f50 | 294 | struct rbd_img_request { |
bf0d5f50 AE |
295 | struct rbd_device *rbd_dev; |
296 | u64 offset; /* starting image byte offset */ | |
297 | u64 length; /* byte count from offset */ | |
0c425248 | 298 | unsigned long flags; |
bf0d5f50 | 299 | union { |
9849e986 | 300 | u64 snap_id; /* for reads */ |
bf0d5f50 | 301 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
302 | }; |
303 | union { | |
304 | struct request *rq; /* block request */ | |
305 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 306 | }; |
3d7efd18 | 307 | struct page **copyup_pages; |
ebda6408 | 308 | u32 copyup_page_count; |
bf0d5f50 AE |
309 | spinlock_t completion_lock;/* protects next_completion */ |
310 | u32 next_completion; | |
311 | rbd_img_callback_t callback; | |
55f27e09 | 312 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 313 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
314 | |
315 | u32 obj_request_count; | |
316 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
317 | ||
318 | struct kref kref; | |
319 | }; | |
320 | ||
321 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 322 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 323 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 324 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 325 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 326 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 327 | |
99d16943 ID |
328 | enum rbd_watch_state { |
329 | RBD_WATCH_STATE_UNREGISTERED, | |
330 | RBD_WATCH_STATE_REGISTERED, | |
331 | RBD_WATCH_STATE_ERROR, | |
332 | }; | |
333 | ||
ed95b21a ID |
334 | enum rbd_lock_state { |
335 | RBD_LOCK_STATE_UNLOCKED, | |
336 | RBD_LOCK_STATE_LOCKED, | |
337 | RBD_LOCK_STATE_RELEASING, | |
338 | }; | |
339 | ||
340 | /* WatchNotify::ClientId */ | |
341 | struct rbd_client_id { | |
342 | u64 gid; | |
343 | u64 handle; | |
344 | }; | |
345 | ||
f84344f3 | 346 | struct rbd_mapping { |
99c1f08f | 347 | u64 size; |
34b13184 | 348 | u64 features; |
f84344f3 AE |
349 | bool read_only; |
350 | }; | |
351 | ||
602adf40 YS |
352 | /* |
353 | * a single device | |
354 | */ | |
355 | struct rbd_device { | |
de71a297 | 356 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
357 | |
358 | int major; /* blkdev assigned major */ | |
dd82fff1 | 359 | int minor; |
602adf40 | 360 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 361 | |
a30b71b9 | 362 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
363 | struct rbd_client *rbd_client; |
364 | ||
365 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
366 | ||
b82d167b | 367 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
368 | |
369 | struct rbd_image_header header; | |
b82d167b | 370 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 371 | struct rbd_spec *spec; |
d147543d | 372 | struct rbd_options *opts; |
0d6d1e9c | 373 | char *config_info; /* add{,_single_major} string */ |
602adf40 | 374 | |
c41d13a3 | 375 | struct ceph_object_id header_oid; |
922dab61 | 376 | struct ceph_object_locator header_oloc; |
971f839a | 377 | |
1643dfa4 | 378 | struct ceph_file_layout layout; /* used for all rbd requests */ |
0903e875 | 379 | |
99d16943 ID |
380 | struct mutex watch_mutex; |
381 | enum rbd_watch_state watch_state; | |
922dab61 | 382 | struct ceph_osd_linger_request *watch_handle; |
99d16943 ID |
383 | u64 watch_cookie; |
384 | struct delayed_work watch_dwork; | |
59c2be1e | 385 | |
ed95b21a ID |
386 | struct rw_semaphore lock_rwsem; |
387 | enum rbd_lock_state lock_state; | |
388 | struct rbd_client_id owner_cid; | |
389 | struct work_struct acquired_lock_work; | |
390 | struct work_struct released_lock_work; | |
391 | struct delayed_work lock_dwork; | |
392 | struct work_struct unlock_work; | |
393 | wait_queue_head_t lock_waitq; | |
394 | ||
1643dfa4 ID |
395 | struct workqueue_struct *task_wq; |
396 | ||
86b00e0d AE |
397 | struct rbd_spec *parent_spec; |
398 | u64 parent_overlap; | |
a2acd00e | 399 | atomic_t parent_ref; |
2f82ee54 | 400 | struct rbd_device *parent; |
86b00e0d | 401 | |
7ad18afa CH |
402 | /* Block layer tags. */ |
403 | struct blk_mq_tag_set tag_set; | |
404 | ||
c666601a JD |
405 | /* protects updating the header */ |
406 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
407 | |
408 | struct rbd_mapping mapping; | |
602adf40 YS |
409 | |
410 | struct list_head node; | |
dfc5606d | 411 | |
dfc5606d YS |
412 | /* sysfs related */ |
413 | struct device dev; | |
b82d167b | 414 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
415 | }; |
416 | ||
b82d167b AE |
417 | /* |
418 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
419 | * rbd_dev->lock is used to protect access. | |
420 | * | |
421 | * Currently, only the "removing" flag (which is coupled with the | |
422 | * "open_count" field) requires atomic access. | |
423 | */ | |
6d292906 AE |
424 | enum rbd_dev_flags { |
425 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 426 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
427 | }; |
428 | ||
cfbf6377 | 429 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 430 | |
602adf40 | 431 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
432 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
433 | ||
432b8587 AE |
434 | static LIST_HEAD(rbd_client_list); /* clients */ |
435 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 436 | |
78c2a44a AE |
437 | /* Slab caches for frequently-allocated structures */ |
438 | ||
1c2a9dfe | 439 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 440 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 441 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 442 | |
9b60e70b | 443 | static int rbd_major; |
f8a22fc2 ID |
444 | static DEFINE_IDA(rbd_dev_id_ida); |
445 | ||
f5ee37bd ID |
446 | static struct workqueue_struct *rbd_wq; |
447 | ||
9b60e70b ID |
448 | /* |
449 | * Default to false for now, as single-major requires >= 0.75 version of | |
450 | * userspace rbd utility. | |
451 | */ | |
452 | static bool single_major = false; | |
453 | module_param(single_major, bool, S_IRUGO); | |
454 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
455 | ||
3d7efd18 AE |
456 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
457 | ||
f0f8cef5 AE |
458 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
459 | size_t count); | |
460 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
461 | size_t count); | |
9b60e70b ID |
462 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
463 | size_t count); | |
464 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
465 | size_t count); | |
6d69bb53 | 466 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
a2acd00e | 467 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 468 | |
9b60e70b ID |
469 | static int rbd_dev_id_to_minor(int dev_id) |
470 | { | |
7e513d43 | 471 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
472 | } |
473 | ||
474 | static int minor_to_rbd_dev_id(int minor) | |
475 | { | |
7e513d43 | 476 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
477 | } |
478 | ||
ed95b21a ID |
479 | static bool rbd_is_lock_supported(struct rbd_device *rbd_dev) |
480 | { | |
481 | return (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) && | |
482 | rbd_dev->spec->snap_id == CEPH_NOSNAP && | |
483 | !rbd_dev->mapping.read_only; | |
484 | } | |
485 | ||
486 | static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev) | |
487 | { | |
488 | return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED || | |
489 | rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING; | |
490 | } | |
491 | ||
492 | static bool rbd_is_lock_owner(struct rbd_device *rbd_dev) | |
493 | { | |
494 | bool is_lock_owner; | |
495 | ||
496 | down_read(&rbd_dev->lock_rwsem); | |
497 | is_lock_owner = __rbd_is_lock_owner(rbd_dev); | |
498 | up_read(&rbd_dev->lock_rwsem); | |
499 | return is_lock_owner; | |
500 | } | |
501 | ||
b15a21dd GKH |
502 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
503 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
504 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
505 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
506 | |
507 | static struct attribute *rbd_bus_attrs[] = { | |
508 | &bus_attr_add.attr, | |
509 | &bus_attr_remove.attr, | |
9b60e70b ID |
510 | &bus_attr_add_single_major.attr, |
511 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 512 | NULL, |
f0f8cef5 | 513 | }; |
92c76dc0 ID |
514 | |
515 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
516 | struct attribute *attr, int index) | |
517 | { | |
9b60e70b ID |
518 | if (!single_major && |
519 | (attr == &bus_attr_add_single_major.attr || | |
520 | attr == &bus_attr_remove_single_major.attr)) | |
521 | return 0; | |
522 | ||
92c76dc0 ID |
523 | return attr->mode; |
524 | } | |
525 | ||
526 | static const struct attribute_group rbd_bus_group = { | |
527 | .attrs = rbd_bus_attrs, | |
528 | .is_visible = rbd_bus_is_visible, | |
529 | }; | |
530 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
531 | |
532 | static struct bus_type rbd_bus_type = { | |
533 | .name = "rbd", | |
b15a21dd | 534 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
535 | }; |
536 | ||
537 | static void rbd_root_dev_release(struct device *dev) | |
538 | { | |
539 | } | |
540 | ||
541 | static struct device rbd_root_dev = { | |
542 | .init_name = "rbd", | |
543 | .release = rbd_root_dev_release, | |
544 | }; | |
545 | ||
06ecc6cb AE |
546 | static __printf(2, 3) |
547 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
548 | { | |
549 | struct va_format vaf; | |
550 | va_list args; | |
551 | ||
552 | va_start(args, fmt); | |
553 | vaf.fmt = fmt; | |
554 | vaf.va = &args; | |
555 | ||
556 | if (!rbd_dev) | |
557 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
558 | else if (rbd_dev->disk) | |
559 | printk(KERN_WARNING "%s: %s: %pV\n", | |
560 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
561 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
562 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
563 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
564 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
565 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
566 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
567 | else /* punt */ | |
568 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
569 | RBD_DRV_NAME, rbd_dev, &vaf); | |
570 | va_end(args); | |
571 | } | |
572 | ||
aafb230e AE |
573 | #ifdef RBD_DEBUG |
574 | #define rbd_assert(expr) \ | |
575 | if (unlikely(!(expr))) { \ | |
576 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
577 | "at line %d:\n\n" \ | |
578 | "\trbd_assert(%s);\n\n", \ | |
579 | __func__, __LINE__, #expr); \ | |
580 | BUG(); \ | |
581 | } | |
582 | #else /* !RBD_DEBUG */ | |
583 | # define rbd_assert(expr) ((void) 0) | |
584 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 585 | |
2761713d | 586 | static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request); |
b454e36d | 587 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
588 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
589 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 590 | |
cc4a38bd | 591 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 592 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 593 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 594 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
595 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
596 | u64 snap_id); | |
2ad3d716 AE |
597 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
598 | u8 *order, u64 *snap_size); | |
599 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
600 | u64 *snap_features); | |
59c2be1e | 601 | |
602adf40 YS |
602 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
603 | { | |
f0f8cef5 | 604 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 605 | bool removing = false; |
602adf40 | 606 | |
f84344f3 | 607 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
608 | return -EROFS; |
609 | ||
a14ea269 | 610 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
611 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
612 | removing = true; | |
613 | else | |
614 | rbd_dev->open_count++; | |
a14ea269 | 615 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
616 | if (removing) |
617 | return -ENOENT; | |
618 | ||
c3e946ce | 619 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 620 | |
602adf40 YS |
621 | return 0; |
622 | } | |
623 | ||
db2a144b | 624 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
625 | { |
626 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
627 | unsigned long open_count_before; |
628 | ||
a14ea269 | 629 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 630 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 631 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 632 | rbd_assert(open_count_before > 0); |
dfc5606d | 633 | |
c3e946ce | 634 | put_device(&rbd_dev->dev); |
dfc5606d YS |
635 | } |
636 | ||
131fd9f6 GZ |
637 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
638 | { | |
77f33c03 | 639 | int ret = 0; |
131fd9f6 GZ |
640 | int val; |
641 | bool ro; | |
77f33c03 | 642 | bool ro_changed = false; |
131fd9f6 | 643 | |
77f33c03 | 644 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
645 | if (get_user(val, (int __user *)(arg))) |
646 | return -EFAULT; | |
647 | ||
648 | ro = val ? true : false; | |
649 | /* Snapshot doesn't allow to write*/ | |
650 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
651 | return -EROFS; | |
652 | ||
77f33c03 JD |
653 | spin_lock_irq(&rbd_dev->lock); |
654 | /* prevent others open this device */ | |
655 | if (rbd_dev->open_count > 1) { | |
656 | ret = -EBUSY; | |
657 | goto out; | |
658 | } | |
659 | ||
131fd9f6 GZ |
660 | if (rbd_dev->mapping.read_only != ro) { |
661 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 662 | ro_changed = true; |
131fd9f6 GZ |
663 | } |
664 | ||
77f33c03 JD |
665 | out: |
666 | spin_unlock_irq(&rbd_dev->lock); | |
667 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
668 | if (ret == 0 && ro_changed) | |
669 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
670 | ||
671 | return ret; | |
131fd9f6 GZ |
672 | } |
673 | ||
674 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
675 | unsigned int cmd, unsigned long arg) | |
676 | { | |
677 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
678 | int ret = 0; | |
679 | ||
131fd9f6 GZ |
680 | switch (cmd) { |
681 | case BLKROSET: | |
682 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
683 | break; | |
684 | default: | |
685 | ret = -ENOTTY; | |
686 | } | |
687 | ||
131fd9f6 GZ |
688 | return ret; |
689 | } | |
690 | ||
691 | #ifdef CONFIG_COMPAT | |
692 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
693 | unsigned int cmd, unsigned long arg) | |
694 | { | |
695 | return rbd_ioctl(bdev, mode, cmd, arg); | |
696 | } | |
697 | #endif /* CONFIG_COMPAT */ | |
698 | ||
602adf40 YS |
699 | static const struct block_device_operations rbd_bd_ops = { |
700 | .owner = THIS_MODULE, | |
701 | .open = rbd_open, | |
dfc5606d | 702 | .release = rbd_release, |
131fd9f6 GZ |
703 | .ioctl = rbd_ioctl, |
704 | #ifdef CONFIG_COMPAT | |
705 | .compat_ioctl = rbd_compat_ioctl, | |
706 | #endif | |
602adf40 YS |
707 | }; |
708 | ||
709 | /* | |
7262cfca | 710 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 711 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 712 | */ |
f8c38929 | 713 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
714 | { |
715 | struct rbd_client *rbdc; | |
716 | int ret = -ENOMEM; | |
717 | ||
37206ee5 | 718 | dout("%s:\n", __func__); |
602adf40 YS |
719 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
720 | if (!rbdc) | |
721 | goto out_opt; | |
722 | ||
723 | kref_init(&rbdc->kref); | |
724 | INIT_LIST_HEAD(&rbdc->node); | |
725 | ||
43ae4701 | 726 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 727 | if (IS_ERR(rbdc->client)) |
08f75463 | 728 | goto out_rbdc; |
43ae4701 | 729 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
730 | |
731 | ret = ceph_open_session(rbdc->client); | |
732 | if (ret < 0) | |
08f75463 | 733 | goto out_client; |
602adf40 | 734 | |
432b8587 | 735 | spin_lock(&rbd_client_list_lock); |
602adf40 | 736 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 737 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 738 | |
37206ee5 | 739 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 740 | |
602adf40 | 741 | return rbdc; |
08f75463 | 742 | out_client: |
602adf40 | 743 | ceph_destroy_client(rbdc->client); |
08f75463 | 744 | out_rbdc: |
602adf40 YS |
745 | kfree(rbdc); |
746 | out_opt: | |
43ae4701 AE |
747 | if (ceph_opts) |
748 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
749 | dout("%s: error %d\n", __func__, ret); |
750 | ||
28f259b7 | 751 | return ERR_PTR(ret); |
602adf40 YS |
752 | } |
753 | ||
2f82ee54 AE |
754 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
755 | { | |
756 | kref_get(&rbdc->kref); | |
757 | ||
758 | return rbdc; | |
759 | } | |
760 | ||
602adf40 | 761 | /* |
1f7ba331 AE |
762 | * Find a ceph client with specific addr and configuration. If |
763 | * found, bump its reference count. | |
602adf40 | 764 | */ |
1f7ba331 | 765 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
766 | { |
767 | struct rbd_client *client_node; | |
1f7ba331 | 768 | bool found = false; |
602adf40 | 769 | |
43ae4701 | 770 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
771 | return NULL; |
772 | ||
1f7ba331 AE |
773 | spin_lock(&rbd_client_list_lock); |
774 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
775 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
776 | __rbd_get_client(client_node); |
777 | ||
1f7ba331 AE |
778 | found = true; |
779 | break; | |
780 | } | |
781 | } | |
782 | spin_unlock(&rbd_client_list_lock); | |
783 | ||
784 | return found ? client_node : NULL; | |
602adf40 YS |
785 | } |
786 | ||
59c2be1e | 787 | /* |
210c104c | 788 | * (Per device) rbd map options |
59c2be1e YS |
789 | */ |
790 | enum { | |
b5584180 | 791 | Opt_queue_depth, |
59c2be1e YS |
792 | Opt_last_int, |
793 | /* int args above */ | |
794 | Opt_last_string, | |
795 | /* string args above */ | |
cc0538b6 AE |
796 | Opt_read_only, |
797 | Opt_read_write, | |
210c104c | 798 | Opt_err |
59c2be1e YS |
799 | }; |
800 | ||
43ae4701 | 801 | static match_table_t rbd_opts_tokens = { |
b5584180 | 802 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
803 | /* int args above */ |
804 | /* string args above */ | |
be466c1c | 805 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
806 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
807 | {Opt_read_write, "read_write"}, | |
808 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 809 | {Opt_err, NULL} |
59c2be1e YS |
810 | }; |
811 | ||
98571b5a | 812 | struct rbd_options { |
b5584180 | 813 | int queue_depth; |
98571b5a AE |
814 | bool read_only; |
815 | }; | |
816 | ||
b5584180 | 817 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a AE |
818 | #define RBD_READ_ONLY_DEFAULT false |
819 | ||
59c2be1e YS |
820 | static int parse_rbd_opts_token(char *c, void *private) |
821 | { | |
43ae4701 | 822 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
823 | substring_t argstr[MAX_OPT_ARGS]; |
824 | int token, intval, ret; | |
825 | ||
43ae4701 | 826 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
827 | if (token < Opt_last_int) { |
828 | ret = match_int(&argstr[0], &intval); | |
829 | if (ret < 0) { | |
210c104c | 830 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
831 | return ret; |
832 | } | |
833 | dout("got int token %d val %d\n", token, intval); | |
834 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 835 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
836 | } else { |
837 | dout("got token %d\n", token); | |
838 | } | |
839 | ||
840 | switch (token) { | |
b5584180 ID |
841 | case Opt_queue_depth: |
842 | if (intval < 1) { | |
843 | pr_err("queue_depth out of range\n"); | |
844 | return -EINVAL; | |
845 | } | |
846 | rbd_opts->queue_depth = intval; | |
847 | break; | |
cc0538b6 AE |
848 | case Opt_read_only: |
849 | rbd_opts->read_only = true; | |
850 | break; | |
851 | case Opt_read_write: | |
852 | rbd_opts->read_only = false; | |
853 | break; | |
59c2be1e | 854 | default: |
210c104c ID |
855 | /* libceph prints "bad option" msg */ |
856 | return -EINVAL; | |
59c2be1e | 857 | } |
210c104c | 858 | |
59c2be1e YS |
859 | return 0; |
860 | } | |
861 | ||
6d2940c8 GZ |
862 | static char* obj_op_name(enum obj_operation_type op_type) |
863 | { | |
864 | switch (op_type) { | |
865 | case OBJ_OP_READ: | |
866 | return "read"; | |
867 | case OBJ_OP_WRITE: | |
868 | return "write"; | |
90e98c52 GZ |
869 | case OBJ_OP_DISCARD: |
870 | return "discard"; | |
6d2940c8 GZ |
871 | default: |
872 | return "???"; | |
873 | } | |
874 | } | |
875 | ||
602adf40 YS |
876 | /* |
877 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
878 | * not exist create it. Either way, ceph_opts is consumed by this |
879 | * function. | |
602adf40 | 880 | */ |
9d3997fd | 881 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 882 | { |
f8c38929 | 883 | struct rbd_client *rbdc; |
59c2be1e | 884 | |
cfbf6377 | 885 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 886 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 887 | if (rbdc) /* using an existing client */ |
43ae4701 | 888 | ceph_destroy_options(ceph_opts); |
9d3997fd | 889 | else |
f8c38929 | 890 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 891 | mutex_unlock(&client_mutex); |
602adf40 | 892 | |
9d3997fd | 893 | return rbdc; |
602adf40 YS |
894 | } |
895 | ||
896 | /* | |
897 | * Destroy ceph client | |
d23a4b3f | 898 | * |
432b8587 | 899 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
900 | */ |
901 | static void rbd_client_release(struct kref *kref) | |
902 | { | |
903 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
904 | ||
37206ee5 | 905 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 906 | spin_lock(&rbd_client_list_lock); |
602adf40 | 907 | list_del(&rbdc->node); |
cd9d9f5d | 908 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
909 | |
910 | ceph_destroy_client(rbdc->client); | |
911 | kfree(rbdc); | |
912 | } | |
913 | ||
914 | /* | |
915 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
916 | * it. | |
917 | */ | |
9d3997fd | 918 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 919 | { |
c53d5893 AE |
920 | if (rbdc) |
921 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
922 | } |
923 | ||
a30b71b9 AE |
924 | static bool rbd_image_format_valid(u32 image_format) |
925 | { | |
926 | return image_format == 1 || image_format == 2; | |
927 | } | |
928 | ||
8e94af8e AE |
929 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
930 | { | |
103a150f AE |
931 | size_t size; |
932 | u32 snap_count; | |
933 | ||
934 | /* The header has to start with the magic rbd header text */ | |
935 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
936 | return false; | |
937 | ||
db2388b6 AE |
938 | /* The bio layer requires at least sector-sized I/O */ |
939 | ||
940 | if (ondisk->options.order < SECTOR_SHIFT) | |
941 | return false; | |
942 | ||
943 | /* If we use u64 in a few spots we may be able to loosen this */ | |
944 | ||
945 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
946 | return false; | |
947 | ||
103a150f AE |
948 | /* |
949 | * The size of a snapshot header has to fit in a size_t, and | |
950 | * that limits the number of snapshots. | |
951 | */ | |
952 | snap_count = le32_to_cpu(ondisk->snap_count); | |
953 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
954 | if (snap_count > size / sizeof (__le64)) | |
955 | return false; | |
956 | ||
957 | /* | |
958 | * Not only that, but the size of the entire the snapshot | |
959 | * header must also be representable in a size_t. | |
960 | */ | |
961 | size -= snap_count * sizeof (__le64); | |
962 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
963 | return false; | |
964 | ||
965 | return true; | |
8e94af8e AE |
966 | } |
967 | ||
602adf40 | 968 | /* |
bb23e37a AE |
969 | * Fill an rbd image header with information from the given format 1 |
970 | * on-disk header. | |
602adf40 | 971 | */ |
662518b1 | 972 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 973 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 974 | { |
662518b1 | 975 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
976 | bool first_time = header->object_prefix == NULL; |
977 | struct ceph_snap_context *snapc; | |
978 | char *object_prefix = NULL; | |
979 | char *snap_names = NULL; | |
980 | u64 *snap_sizes = NULL; | |
ccece235 | 981 | u32 snap_count; |
d2bb24e5 | 982 | size_t size; |
bb23e37a | 983 | int ret = -ENOMEM; |
621901d6 | 984 | u32 i; |
602adf40 | 985 | |
bb23e37a | 986 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 987 | |
bb23e37a AE |
988 | if (first_time) { |
989 | size_t len; | |
103a150f | 990 | |
bb23e37a AE |
991 | len = strnlen(ondisk->object_prefix, |
992 | sizeof (ondisk->object_prefix)); | |
993 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
994 | if (!object_prefix) | |
995 | return -ENOMEM; | |
996 | memcpy(object_prefix, ondisk->object_prefix, len); | |
997 | object_prefix[len] = '\0'; | |
998 | } | |
00f1f36f | 999 | |
bb23e37a | 1000 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 1001 | |
bb23e37a AE |
1002 | snap_count = le32_to_cpu(ondisk->snap_count); |
1003 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
1004 | if (!snapc) | |
1005 | goto out_err; | |
1006 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 1007 | if (snap_count) { |
bb23e37a | 1008 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
1009 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
1010 | ||
bb23e37a | 1011 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 1012 | |
bb23e37a AE |
1013 | if (snap_names_len > (u64)SIZE_MAX) |
1014 | goto out_2big; | |
1015 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
1016 | if (!snap_names) | |
6a52325f AE |
1017 | goto out_err; |
1018 | ||
bb23e37a | 1019 | /* ...as well as the array of their sizes. */ |
621901d6 | 1020 | |
d2bb24e5 | 1021 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
1022 | snap_sizes = kmalloc(size, GFP_KERNEL); |
1023 | if (!snap_sizes) | |
6a52325f | 1024 | goto out_err; |
bb23e37a | 1025 | |
f785cc1d | 1026 | /* |
bb23e37a AE |
1027 | * Copy the names, and fill in each snapshot's id |
1028 | * and size. | |
1029 | * | |
99a41ebc | 1030 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 1031 | * ondisk buffer we're working with has |
f785cc1d AE |
1032 | * snap_names_len bytes beyond the end of the |
1033 | * snapshot id array, this memcpy() is safe. | |
1034 | */ | |
bb23e37a AE |
1035 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
1036 | snaps = ondisk->snaps; | |
1037 | for (i = 0; i < snap_count; i++) { | |
1038 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
1039 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
1040 | } | |
602adf40 | 1041 | } |
6a52325f | 1042 | |
bb23e37a | 1043 | /* We won't fail any more, fill in the header */ |
621901d6 | 1044 | |
bb23e37a AE |
1045 | if (first_time) { |
1046 | header->object_prefix = object_prefix; | |
1047 | header->obj_order = ondisk->options.order; | |
1048 | header->crypt_type = ondisk->options.crypt_type; | |
1049 | header->comp_type = ondisk->options.comp_type; | |
1050 | /* The rest aren't used for format 1 images */ | |
1051 | header->stripe_unit = 0; | |
1052 | header->stripe_count = 0; | |
1053 | header->features = 0; | |
602adf40 | 1054 | } else { |
662518b1 AE |
1055 | ceph_put_snap_context(header->snapc); |
1056 | kfree(header->snap_names); | |
1057 | kfree(header->snap_sizes); | |
602adf40 | 1058 | } |
849b4260 | 1059 | |
bb23e37a | 1060 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1061 | |
f84344f3 | 1062 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1063 | header->snapc = snapc; |
1064 | header->snap_names = snap_names; | |
1065 | header->snap_sizes = snap_sizes; | |
468521c1 | 1066 | |
602adf40 | 1067 | return 0; |
bb23e37a AE |
1068 | out_2big: |
1069 | ret = -EIO; | |
6a52325f | 1070 | out_err: |
bb23e37a AE |
1071 | kfree(snap_sizes); |
1072 | kfree(snap_names); | |
1073 | ceph_put_snap_context(snapc); | |
1074 | kfree(object_prefix); | |
ccece235 | 1075 | |
bb23e37a | 1076 | return ret; |
602adf40 YS |
1077 | } |
1078 | ||
9682fc6d AE |
1079 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1080 | { | |
1081 | const char *snap_name; | |
1082 | ||
1083 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1084 | ||
1085 | /* Skip over names until we find the one we are looking for */ | |
1086 | ||
1087 | snap_name = rbd_dev->header.snap_names; | |
1088 | while (which--) | |
1089 | snap_name += strlen(snap_name) + 1; | |
1090 | ||
1091 | return kstrdup(snap_name, GFP_KERNEL); | |
1092 | } | |
1093 | ||
30d1cff8 AE |
1094 | /* |
1095 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1096 | * Note that result is for snapshots in *descending* order. | |
1097 | */ | |
1098 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1099 | { | |
1100 | u64 snap_id1 = *(u64 *)s1; | |
1101 | u64 snap_id2 = *(u64 *)s2; | |
1102 | ||
1103 | if (snap_id1 < snap_id2) | |
1104 | return 1; | |
1105 | return snap_id1 == snap_id2 ? 0 : -1; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Search a snapshot context to see if the given snapshot id is | |
1110 | * present. | |
1111 | * | |
1112 | * Returns the position of the snapshot id in the array if it's found, | |
1113 | * or BAD_SNAP_INDEX otherwise. | |
1114 | * | |
1115 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1116 | * reverse order, highest snapshot id first. | |
1117 | */ | |
9682fc6d AE |
1118 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1119 | { | |
1120 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1121 | u64 *found; |
9682fc6d | 1122 | |
30d1cff8 AE |
1123 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1124 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1125 | |
30d1cff8 | 1126 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1127 | } |
1128 | ||
2ad3d716 AE |
1129 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1130 | u64 snap_id) | |
9e15b77d | 1131 | { |
54cac61f | 1132 | u32 which; |
da6a6b63 | 1133 | const char *snap_name; |
9e15b77d | 1134 | |
54cac61f AE |
1135 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1136 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1137 | return ERR_PTR(-ENOENT); |
54cac61f | 1138 | |
da6a6b63 JD |
1139 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1140 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1141 | } |
1142 | ||
1143 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1144 | { | |
9e15b77d AE |
1145 | if (snap_id == CEPH_NOSNAP) |
1146 | return RBD_SNAP_HEAD_NAME; | |
1147 | ||
54cac61f AE |
1148 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1149 | if (rbd_dev->image_format == 1) | |
1150 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1151 | |
54cac61f | 1152 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1153 | } |
1154 | ||
2ad3d716 AE |
1155 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1156 | u64 *snap_size) | |
602adf40 | 1157 | { |
2ad3d716 AE |
1158 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1159 | if (snap_id == CEPH_NOSNAP) { | |
1160 | *snap_size = rbd_dev->header.image_size; | |
1161 | } else if (rbd_dev->image_format == 1) { | |
1162 | u32 which; | |
602adf40 | 1163 | |
2ad3d716 AE |
1164 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1165 | if (which == BAD_SNAP_INDEX) | |
1166 | return -ENOENT; | |
e86924a8 | 1167 | |
2ad3d716 AE |
1168 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1169 | } else { | |
1170 | u64 size = 0; | |
1171 | int ret; | |
1172 | ||
1173 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1174 | if (ret) | |
1175 | return ret; | |
1176 | ||
1177 | *snap_size = size; | |
1178 | } | |
1179 | return 0; | |
602adf40 YS |
1180 | } |
1181 | ||
2ad3d716 AE |
1182 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1183 | u64 *snap_features) | |
602adf40 | 1184 | { |
2ad3d716 AE |
1185 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1186 | if (snap_id == CEPH_NOSNAP) { | |
1187 | *snap_features = rbd_dev->header.features; | |
1188 | } else if (rbd_dev->image_format == 1) { | |
1189 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1190 | } else { |
2ad3d716 AE |
1191 | u64 features = 0; |
1192 | int ret; | |
8b0241f8 | 1193 | |
2ad3d716 AE |
1194 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1195 | if (ret) | |
1196 | return ret; | |
1197 | ||
1198 | *snap_features = features; | |
1199 | } | |
1200 | return 0; | |
1201 | } | |
1202 | ||
1203 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1204 | { | |
8f4b7d98 | 1205 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1206 | u64 size = 0; |
1207 | u64 features = 0; | |
1208 | int ret; | |
1209 | ||
2ad3d716 AE |
1210 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1211 | if (ret) | |
1212 | return ret; | |
1213 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1214 | if (ret) | |
1215 | return ret; | |
1216 | ||
1217 | rbd_dev->mapping.size = size; | |
1218 | rbd_dev->mapping.features = features; | |
1219 | ||
8b0241f8 | 1220 | return 0; |
602adf40 YS |
1221 | } |
1222 | ||
d1cf5788 AE |
1223 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1224 | { | |
1225 | rbd_dev->mapping.size = 0; | |
1226 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1227 | } |
1228 | ||
7d5079aa HS |
1229 | static void rbd_segment_name_free(const char *name) |
1230 | { | |
1231 | /* The explicit cast here is needed to drop the const qualifier */ | |
1232 | ||
1233 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1234 | } | |
1235 | ||
98571b5a | 1236 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1237 | { |
65ccfe21 AE |
1238 | char *name; |
1239 | u64 segment; | |
1240 | int ret; | |
3a96d5cd | 1241 | char *name_format; |
602adf40 | 1242 | |
78c2a44a | 1243 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1244 | if (!name) |
1245 | return NULL; | |
1246 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1247 | name_format = "%s.%012llx"; |
1248 | if (rbd_dev->image_format == 2) | |
1249 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1250 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1251 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1252 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1253 | pr_err("error formatting segment name for #%llu (%d)\n", |
1254 | segment, ret); | |
7d5079aa | 1255 | rbd_segment_name_free(name); |
65ccfe21 AE |
1256 | name = NULL; |
1257 | } | |
602adf40 | 1258 | |
65ccfe21 AE |
1259 | return name; |
1260 | } | |
602adf40 | 1261 | |
65ccfe21 AE |
1262 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1263 | { | |
1264 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1265 | |
65ccfe21 AE |
1266 | return offset & (segment_size - 1); |
1267 | } | |
1268 | ||
1269 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1270 | u64 offset, u64 length) | |
1271 | { | |
1272 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1273 | ||
1274 | offset &= segment_size - 1; | |
1275 | ||
aafb230e | 1276 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1277 | if (offset + length > segment_size) |
1278 | length = segment_size - offset; | |
1279 | ||
1280 | return length; | |
602adf40 YS |
1281 | } |
1282 | ||
029bcbd8 JD |
1283 | /* |
1284 | * returns the size of an object in the image | |
1285 | */ | |
1286 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1287 | { | |
1288 | return 1 << header->obj_order; | |
1289 | } | |
1290 | ||
602adf40 YS |
1291 | /* |
1292 | * bio helpers | |
1293 | */ | |
1294 | ||
1295 | static void bio_chain_put(struct bio *chain) | |
1296 | { | |
1297 | struct bio *tmp; | |
1298 | ||
1299 | while (chain) { | |
1300 | tmp = chain; | |
1301 | chain = chain->bi_next; | |
1302 | bio_put(tmp); | |
1303 | } | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * zeros a bio chain, starting at specific offset | |
1308 | */ | |
1309 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1310 | { | |
7988613b KO |
1311 | struct bio_vec bv; |
1312 | struct bvec_iter iter; | |
602adf40 YS |
1313 | unsigned long flags; |
1314 | void *buf; | |
602adf40 YS |
1315 | int pos = 0; |
1316 | ||
1317 | while (chain) { | |
7988613b KO |
1318 | bio_for_each_segment(bv, chain, iter) { |
1319 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1320 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1321 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1322 | memset(buf + remainder, 0, |
7988613b KO |
1323 | bv.bv_len - remainder); |
1324 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1325 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1326 | } |
7988613b | 1327 | pos += bv.bv_len; |
602adf40 YS |
1328 | } |
1329 | ||
1330 | chain = chain->bi_next; | |
1331 | } | |
1332 | } | |
1333 | ||
b9434c5b AE |
1334 | /* |
1335 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1336 | * starting at the given byte offset from the start of the array and | |
1337 | * continuing up to the given end offset. The pages array is | |
1338 | * assumed to be big enough to hold all bytes up to the end. | |
1339 | */ | |
1340 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1341 | { | |
1342 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1343 | ||
1344 | rbd_assert(end > offset); | |
1345 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1346 | while (offset < end) { | |
1347 | size_t page_offset; | |
1348 | size_t length; | |
1349 | unsigned long flags; | |
1350 | void *kaddr; | |
1351 | ||
491205a8 GU |
1352 | page_offset = offset & ~PAGE_MASK; |
1353 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1354 | local_irq_save(flags); |
1355 | kaddr = kmap_atomic(*page); | |
1356 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1357 | flush_dcache_page(*page); |
b9434c5b AE |
1358 | kunmap_atomic(kaddr); |
1359 | local_irq_restore(flags); | |
1360 | ||
1361 | offset += length; | |
1362 | page++; | |
1363 | } | |
1364 | } | |
1365 | ||
602adf40 | 1366 | /* |
f7760dad AE |
1367 | * Clone a portion of a bio, starting at the given byte offset |
1368 | * and continuing for the number of bytes indicated. | |
602adf40 | 1369 | */ |
f7760dad AE |
1370 | static struct bio *bio_clone_range(struct bio *bio_src, |
1371 | unsigned int offset, | |
1372 | unsigned int len, | |
1373 | gfp_t gfpmask) | |
602adf40 | 1374 | { |
f7760dad AE |
1375 | struct bio *bio; |
1376 | ||
5341a627 | 1377 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1378 | if (!bio) |
1379 | return NULL; /* ENOMEM */ | |
602adf40 | 1380 | |
5341a627 | 1381 | bio_advance(bio, offset); |
4f024f37 | 1382 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1383 | |
1384 | return bio; | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Clone a portion of a bio chain, starting at the given byte offset | |
1389 | * into the first bio in the source chain and continuing for the | |
1390 | * number of bytes indicated. The result is another bio chain of | |
1391 | * exactly the given length, or a null pointer on error. | |
1392 | * | |
1393 | * The bio_src and offset parameters are both in-out. On entry they | |
1394 | * refer to the first source bio and the offset into that bio where | |
1395 | * the start of data to be cloned is located. | |
1396 | * | |
1397 | * On return, bio_src is updated to refer to the bio in the source | |
1398 | * chain that contains first un-cloned byte, and *offset will | |
1399 | * contain the offset of that byte within that bio. | |
1400 | */ | |
1401 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1402 | unsigned int *offset, | |
1403 | unsigned int len, | |
1404 | gfp_t gfpmask) | |
1405 | { | |
1406 | struct bio *bi = *bio_src; | |
1407 | unsigned int off = *offset; | |
1408 | struct bio *chain = NULL; | |
1409 | struct bio **end; | |
1410 | ||
1411 | /* Build up a chain of clone bios up to the limit */ | |
1412 | ||
4f024f37 | 1413 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1414 | return NULL; /* Nothing to clone */ |
602adf40 | 1415 | |
f7760dad AE |
1416 | end = &chain; |
1417 | while (len) { | |
1418 | unsigned int bi_size; | |
1419 | struct bio *bio; | |
1420 | ||
f5400b7a AE |
1421 | if (!bi) { |
1422 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1423 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1424 | } |
4f024f37 | 1425 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1426 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1427 | if (!bio) | |
1428 | goto out_err; /* ENOMEM */ | |
1429 | ||
1430 | *end = bio; | |
1431 | end = &bio->bi_next; | |
602adf40 | 1432 | |
f7760dad | 1433 | off += bi_size; |
4f024f37 | 1434 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1435 | bi = bi->bi_next; |
1436 | off = 0; | |
1437 | } | |
1438 | len -= bi_size; | |
1439 | } | |
1440 | *bio_src = bi; | |
1441 | *offset = off; | |
1442 | ||
1443 | return chain; | |
1444 | out_err: | |
1445 | bio_chain_put(chain); | |
602adf40 | 1446 | |
602adf40 YS |
1447 | return NULL; |
1448 | } | |
1449 | ||
926f9b3f AE |
1450 | /* |
1451 | * The default/initial value for all object request flags is 0. For | |
1452 | * each flag, once its value is set to 1 it is never reset to 0 | |
1453 | * again. | |
1454 | */ | |
57acbaa7 | 1455 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1456 | { |
57acbaa7 | 1457 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1458 | struct rbd_device *rbd_dev; |
1459 | ||
57acbaa7 | 1460 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1461 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1462 | obj_request); |
1463 | } | |
1464 | } | |
1465 | ||
57acbaa7 | 1466 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1467 | { |
1468 | smp_mb(); | |
57acbaa7 | 1469 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1470 | } |
1471 | ||
57acbaa7 | 1472 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1473 | { |
57acbaa7 AE |
1474 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1475 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1476 | |
57acbaa7 AE |
1477 | if (obj_request_img_data_test(obj_request)) |
1478 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1479 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1480 | obj_request); |
1481 | } | |
1482 | } | |
1483 | ||
57acbaa7 | 1484 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1485 | { |
1486 | smp_mb(); | |
57acbaa7 | 1487 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1488 | } |
1489 | ||
5679c59f AE |
1490 | /* |
1491 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1492 | * flag. The latter is set based on the "exists" value provided. | |
1493 | * | |
1494 | * Note that for our purposes once an object exists it never goes | |
1495 | * away again. It's possible that the response from two existence | |
1496 | * checks are separated by the creation of the target object, and | |
1497 | * the first ("doesn't exist") response arrives *after* the second | |
1498 | * ("does exist"). In that case we ignore the second one. | |
1499 | */ | |
1500 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1501 | bool exists) | |
1502 | { | |
1503 | if (exists) | |
1504 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1505 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1506 | smp_mb(); | |
1507 | } | |
1508 | ||
1509 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1510 | { | |
1511 | smp_mb(); | |
1512 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1513 | } | |
1514 | ||
1515 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1516 | { | |
1517 | smp_mb(); | |
1518 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1519 | } | |
1520 | ||
9638556a ID |
1521 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1522 | { | |
1523 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1524 | ||
1525 | return obj_request->img_offset < | |
1526 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1527 | } | |
1528 | ||
bf0d5f50 AE |
1529 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1530 | { | |
37206ee5 AE |
1531 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1532 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1533 | kref_get(&obj_request->kref); |
1534 | } | |
1535 | ||
1536 | static void rbd_obj_request_destroy(struct kref *kref); | |
1537 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1538 | { | |
1539 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1540 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1541 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1542 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1543 | } | |
1544 | ||
0f2d5be7 AE |
1545 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1546 | { | |
1547 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1548 | atomic_read(&img_request->kref.refcount)); | |
1549 | kref_get(&img_request->kref); | |
1550 | } | |
1551 | ||
e93f3152 AE |
1552 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1553 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1554 | static void rbd_img_request_destroy(struct kref *kref); |
1555 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1556 | { | |
1557 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1558 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1559 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1560 | if (img_request_child_test(img_request)) |
1561 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1562 | else | |
1563 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1564 | } |
1565 | ||
1566 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1567 | struct rbd_obj_request *obj_request) | |
1568 | { | |
25dcf954 AE |
1569 | rbd_assert(obj_request->img_request == NULL); |
1570 | ||
b155e86c | 1571 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1572 | obj_request->img_request = img_request; |
25dcf954 | 1573 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1574 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1575 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1576 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1577 | img_request->obj_request_count++; |
1578 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1579 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1580 | obj_request->which); | |
bf0d5f50 AE |
1581 | } |
1582 | ||
1583 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1584 | struct rbd_obj_request *obj_request) | |
1585 | { | |
1586 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1587 | |
37206ee5 AE |
1588 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1589 | obj_request->which); | |
bf0d5f50 | 1590 | list_del(&obj_request->links); |
25dcf954 AE |
1591 | rbd_assert(img_request->obj_request_count > 0); |
1592 | img_request->obj_request_count--; | |
1593 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1594 | obj_request->which = BAD_WHICH; | |
6365d33a | 1595 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1596 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1597 | obj_request->img_request = NULL; |
25dcf954 | 1598 | obj_request->callback = NULL; |
bf0d5f50 AE |
1599 | rbd_obj_request_put(obj_request); |
1600 | } | |
1601 | ||
1602 | static bool obj_request_type_valid(enum obj_request_type type) | |
1603 | { | |
1604 | switch (type) { | |
9969ebc5 | 1605 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1606 | case OBJ_REQUEST_BIO: |
788e2df3 | 1607 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1608 | return true; |
1609 | default: | |
1610 | return false; | |
1611 | } | |
1612 | } | |
1613 | ||
bf0d5f50 AE |
1614 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1615 | struct rbd_obj_request *obj_request) | |
1616 | { | |
71c20a06 | 1617 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1618 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1619 | } | |
1620 | ||
71c20a06 ID |
1621 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1622 | { | |
1623 | dout("%s %p\n", __func__, obj_request); | |
1624 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * Wait for an object request to complete. If interrupted, cancel the | |
1629 | * underlying osd request. | |
2894e1d7 ID |
1630 | * |
1631 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1632 | */ |
2894e1d7 ID |
1633 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1634 | unsigned long timeout) | |
71c20a06 | 1635 | { |
2894e1d7 | 1636 | long ret; |
71c20a06 ID |
1637 | |
1638 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1639 | ret = wait_for_completion_interruptible_timeout( |
1640 | &obj_request->completion, | |
1641 | ceph_timeout_jiffies(timeout)); | |
1642 | if (ret <= 0) { | |
1643 | if (ret == 0) | |
1644 | ret = -ETIMEDOUT; | |
71c20a06 | 1645 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1646 | } else { |
1647 | ret = 0; | |
71c20a06 ID |
1648 | } |
1649 | ||
2894e1d7 ID |
1650 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1651 | return ret; | |
1652 | } | |
1653 | ||
1654 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1655 | { | |
1656 | return __rbd_obj_request_wait(obj_request, 0); | |
1657 | } | |
1658 | ||
bf0d5f50 AE |
1659 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1660 | { | |
55f27e09 | 1661 | |
37206ee5 | 1662 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1663 | |
1664 | /* | |
1665 | * If no error occurred, compute the aggregate transfer | |
1666 | * count for the image request. We could instead use | |
1667 | * atomic64_cmpxchg() to update it as each object request | |
1668 | * completes; not clear which way is better off hand. | |
1669 | */ | |
1670 | if (!img_request->result) { | |
1671 | struct rbd_obj_request *obj_request; | |
1672 | u64 xferred = 0; | |
1673 | ||
1674 | for_each_obj_request(img_request, obj_request) | |
1675 | xferred += obj_request->xferred; | |
1676 | img_request->xferred = xferred; | |
1677 | } | |
1678 | ||
bf0d5f50 AE |
1679 | if (img_request->callback) |
1680 | img_request->callback(img_request); | |
1681 | else | |
1682 | rbd_img_request_put(img_request); | |
1683 | } | |
1684 | ||
0c425248 AE |
1685 | /* |
1686 | * The default/initial value for all image request flags is 0. Each | |
1687 | * is conditionally set to 1 at image request initialization time | |
1688 | * and currently never change thereafter. | |
1689 | */ | |
1690 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1691 | { | |
1692 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1693 | smp_mb(); | |
1694 | } | |
1695 | ||
1696 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1697 | { | |
1698 | smp_mb(); | |
1699 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1700 | } | |
1701 | ||
90e98c52 GZ |
1702 | /* |
1703 | * Set the discard flag when the img_request is an discard request | |
1704 | */ | |
1705 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1706 | { | |
1707 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1708 | smp_mb(); | |
1709 | } | |
1710 | ||
1711 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1712 | { | |
1713 | smp_mb(); | |
1714 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1715 | } | |
1716 | ||
9849e986 AE |
1717 | static void img_request_child_set(struct rbd_img_request *img_request) |
1718 | { | |
1719 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1720 | smp_mb(); | |
1721 | } | |
1722 | ||
e93f3152 AE |
1723 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1724 | { | |
1725 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1726 | smp_mb(); | |
1727 | } | |
1728 | ||
9849e986 AE |
1729 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1730 | { | |
1731 | smp_mb(); | |
1732 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1733 | } | |
1734 | ||
d0b2e944 AE |
1735 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1736 | { | |
1737 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1738 | smp_mb(); | |
1739 | } | |
1740 | ||
a2acd00e AE |
1741 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1742 | { | |
1743 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1744 | smp_mb(); | |
1745 | } | |
1746 | ||
d0b2e944 AE |
1747 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1748 | { | |
1749 | smp_mb(); | |
1750 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1751 | } | |
1752 | ||
3b434a2a JD |
1753 | static enum obj_operation_type |
1754 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1755 | { | |
1756 | if (img_request_write_test(img_request)) | |
1757 | return OBJ_OP_WRITE; | |
1758 | else if (img_request_discard_test(img_request)) | |
1759 | return OBJ_OP_DISCARD; | |
1760 | else | |
1761 | return OBJ_OP_READ; | |
1762 | } | |
1763 | ||
6e2a4505 AE |
1764 | static void |
1765 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1766 | { | |
b9434c5b AE |
1767 | u64 xferred = obj_request->xferred; |
1768 | u64 length = obj_request->length; | |
1769 | ||
6e2a4505 AE |
1770 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1771 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1772 | xferred, length); |
6e2a4505 | 1773 | /* |
17c1cc1d JD |
1774 | * ENOENT means a hole in the image. We zero-fill the entire |
1775 | * length of the request. A short read also implies zero-fill | |
1776 | * to the end of the request. An error requires the whole | |
1777 | * length of the request to be reported finished with an error | |
1778 | * to the block layer. In each case we update the xferred | |
1779 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1780 | */ |
b9434c5b | 1781 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1782 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1783 | if (obj_request->type == OBJ_REQUEST_BIO) |
1784 | zero_bio_chain(obj_request->bio_list, 0); | |
1785 | else | |
1786 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1787 | obj_request->result = 0; |
b9434c5b AE |
1788 | } else if (xferred < length && !obj_request->result) { |
1789 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1790 | zero_bio_chain(obj_request->bio_list, xferred); | |
1791 | else | |
1792 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1793 | } |
17c1cc1d | 1794 | obj_request->xferred = length; |
6e2a4505 AE |
1795 | obj_request_done_set(obj_request); |
1796 | } | |
1797 | ||
bf0d5f50 AE |
1798 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1799 | { | |
37206ee5 AE |
1800 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1801 | obj_request->callback); | |
bf0d5f50 AE |
1802 | if (obj_request->callback) |
1803 | obj_request->callback(obj_request); | |
788e2df3 AE |
1804 | else |
1805 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1806 | } |
1807 | ||
c47f9371 | 1808 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1809 | { |
57acbaa7 | 1810 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1811 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1812 | bool layered = false; |
1813 | ||
1814 | if (obj_request_img_data_test(obj_request)) { | |
1815 | img_request = obj_request->img_request; | |
1816 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1817 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1818 | } |
8b3e1a56 AE |
1819 | |
1820 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1821 | obj_request, img_request, obj_request->result, | |
1822 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1823 | if (layered && obj_request->result == -ENOENT && |
1824 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1825 | rbd_img_parent_read(obj_request); |
1826 | else if (img_request) | |
6e2a4505 AE |
1827 | rbd_img_obj_request_read_callback(obj_request); |
1828 | else | |
1829 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1830 | } |
1831 | ||
c47f9371 | 1832 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1833 | { |
1b83bef2 SW |
1834 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1835 | obj_request->result, obj_request->length); | |
1836 | /* | |
8b3e1a56 AE |
1837 | * There is no such thing as a successful short write. Set |
1838 | * it to our originally-requested length. | |
1b83bef2 SW |
1839 | */ |
1840 | obj_request->xferred = obj_request->length; | |
07741308 | 1841 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1842 | } |
1843 | ||
90e98c52 GZ |
1844 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1845 | { | |
1846 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1847 | obj_request->result, obj_request->length); | |
1848 | /* | |
1849 | * There is no such thing as a successful short discard. Set | |
1850 | * it to our originally-requested length. | |
1851 | */ | |
1852 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1853 | /* discarding a non-existent object is not a problem */ |
1854 | if (obj_request->result == -ENOENT) | |
1855 | obj_request->result = 0; | |
90e98c52 GZ |
1856 | obj_request_done_set(obj_request); |
1857 | } | |
1858 | ||
fbfab539 AE |
1859 | /* |
1860 | * For a simple stat call there's nothing to do. We'll do more if | |
1861 | * this is part of a write sequence for a layered image. | |
1862 | */ | |
c47f9371 | 1863 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1864 | { |
37206ee5 | 1865 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1866 | obj_request_done_set(obj_request); |
1867 | } | |
1868 | ||
2761713d ID |
1869 | static void rbd_osd_call_callback(struct rbd_obj_request *obj_request) |
1870 | { | |
1871 | dout("%s: obj %p\n", __func__, obj_request); | |
1872 | ||
1873 | if (obj_request_img_data_test(obj_request)) | |
1874 | rbd_osd_copyup_callback(obj_request); | |
1875 | else | |
1876 | obj_request_done_set(obj_request); | |
1877 | } | |
1878 | ||
85e084fe | 1879 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req) |
bf0d5f50 AE |
1880 | { |
1881 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1882 | u16 opcode; |
1883 | ||
85e084fe | 1884 | dout("%s: osd_req %p\n", __func__, osd_req); |
bf0d5f50 | 1885 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1886 | if (obj_request_img_data_test(obj_request)) { |
1887 | rbd_assert(obj_request->img_request); | |
1888 | rbd_assert(obj_request->which != BAD_WHICH); | |
1889 | } else { | |
1890 | rbd_assert(obj_request->which == BAD_WHICH); | |
1891 | } | |
bf0d5f50 | 1892 | |
1b83bef2 SW |
1893 | if (osd_req->r_result < 0) |
1894 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1895 | |
c47f9371 AE |
1896 | /* |
1897 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1898 | * passed to the block layer, which just supports a 32-bit |
1899 | * length field. | |
c47f9371 | 1900 | */ |
7665d85b | 1901 | obj_request->xferred = osd_req->r_ops[0].outdata_len; |
8b3e1a56 | 1902 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1903 | |
79528734 | 1904 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1905 | switch (opcode) { |
1906 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1907 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1908 | break; |
0ccd5926 | 1909 | case CEPH_OSD_OP_SETALLOCHINT: |
e30b7577 ID |
1910 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE || |
1911 | osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL); | |
0ccd5926 | 1912 | /* fall through */ |
bf0d5f50 | 1913 | case CEPH_OSD_OP_WRITE: |
e30b7577 | 1914 | case CEPH_OSD_OP_WRITEFULL: |
c47f9371 | 1915 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1916 | break; |
fbfab539 | 1917 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1918 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1919 | break; |
90e98c52 GZ |
1920 | case CEPH_OSD_OP_DELETE: |
1921 | case CEPH_OSD_OP_TRUNCATE: | |
1922 | case CEPH_OSD_OP_ZERO: | |
1923 | rbd_osd_discard_callback(obj_request); | |
1924 | break; | |
36be9a76 | 1925 | case CEPH_OSD_OP_CALL: |
2761713d ID |
1926 | rbd_osd_call_callback(obj_request); |
1927 | break; | |
bf0d5f50 | 1928 | default: |
9584d508 | 1929 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1930 | obj_request->object_name, (unsigned short) opcode); |
1931 | break; | |
1932 | } | |
1933 | ||
07741308 | 1934 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1935 | rbd_obj_request_complete(obj_request); |
1936 | } | |
1937 | ||
9d4df01f | 1938 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1939 | { |
1940 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1941 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
430c28c3 | 1942 | |
bb873b53 ID |
1943 | if (img_request) |
1944 | osd_req->r_snapid = img_request->snap_id; | |
9d4df01f AE |
1945 | } |
1946 | ||
1947 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1948 | { | |
9d4df01f | 1949 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1950 | |
bb873b53 ID |
1951 | osd_req->r_mtime = CURRENT_TIME; |
1952 | osd_req->r_data_offset = obj_request->offset; | |
430c28c3 AE |
1953 | } |
1954 | ||
0ccd5926 ID |
1955 | /* |
1956 | * Create an osd request. A read request has one osd op (read). | |
1957 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1958 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1959 | * technically osd watch is a write request, hence this distinction.) | |
1960 | */ | |
bf0d5f50 AE |
1961 | static struct ceph_osd_request *rbd_osd_req_create( |
1962 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1963 | enum obj_operation_type op_type, |
deb236b3 | 1964 | unsigned int num_ops, |
430c28c3 | 1965 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1966 | { |
bf0d5f50 AE |
1967 | struct ceph_snap_context *snapc = NULL; |
1968 | struct ceph_osd_client *osdc; | |
1969 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1970 | |
90e98c52 GZ |
1971 | if (obj_request_img_data_test(obj_request) && |
1972 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1973 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1974 | if (op_type == OBJ_OP_WRITE) { |
1975 | rbd_assert(img_request_write_test(img_request)); | |
1976 | } else { | |
1977 | rbd_assert(img_request_discard_test(img_request)); | |
1978 | } | |
6d2940c8 | 1979 | snapc = img_request->snapc; |
bf0d5f50 AE |
1980 | } |
1981 | ||
6d2940c8 | 1982 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1983 | |
1984 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1985 | |
1986 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 | 1987 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
2224d879 | 1988 | GFP_NOIO); |
bf0d5f50 | 1989 | if (!osd_req) |
13d1ad16 | 1990 | goto fail; |
bf0d5f50 | 1991 | |
90e98c52 | 1992 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1993 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1994 | else |
bf0d5f50 | 1995 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1996 | |
1997 | osd_req->r_callback = rbd_osd_req_callback; | |
1998 | osd_req->r_priv = obj_request; | |
1999 | ||
7627151e | 2000 | osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id; |
d30291b9 ID |
2001 | if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s", |
2002 | obj_request->object_name)) | |
2003 | goto fail; | |
bf0d5f50 | 2004 | |
13d1ad16 ID |
2005 | if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO)) |
2006 | goto fail; | |
2007 | ||
bf0d5f50 | 2008 | return osd_req; |
13d1ad16 ID |
2009 | |
2010 | fail: | |
2011 | ceph_osdc_put_request(osd_req); | |
2012 | return NULL; | |
bf0d5f50 AE |
2013 | } |
2014 | ||
0eefd470 | 2015 | /* |
d3246fb0 JD |
2016 | * Create a copyup osd request based on the information in the object |
2017 | * request supplied. A copyup request has two or three osd ops, a | |
2018 | * copyup method call, potentially a hint op, and a write or truncate | |
2019 | * or zero op. | |
0eefd470 AE |
2020 | */ |
2021 | static struct ceph_osd_request * | |
2022 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
2023 | { | |
2024 | struct rbd_img_request *img_request; | |
2025 | struct ceph_snap_context *snapc; | |
2026 | struct rbd_device *rbd_dev; | |
2027 | struct ceph_osd_client *osdc; | |
2028 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 2029 | int num_osd_ops = 3; |
0eefd470 AE |
2030 | |
2031 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2032 | img_request = obj_request->img_request; | |
2033 | rbd_assert(img_request); | |
d3246fb0 JD |
2034 | rbd_assert(img_request_write_test(img_request) || |
2035 | img_request_discard_test(img_request)); | |
0eefd470 | 2036 | |
d3246fb0 JD |
2037 | if (img_request_discard_test(img_request)) |
2038 | num_osd_ops = 2; | |
2039 | ||
2040 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
2041 | |
2042 | snapc = img_request->snapc; | |
2043 | rbd_dev = img_request->rbd_dev; | |
2044 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 | 2045 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
2224d879 | 2046 | false, GFP_NOIO); |
0eefd470 | 2047 | if (!osd_req) |
13d1ad16 | 2048 | goto fail; |
0eefd470 AE |
2049 | |
2050 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
2051 | osd_req->r_callback = rbd_osd_req_callback; | |
2052 | osd_req->r_priv = obj_request; | |
2053 | ||
7627151e | 2054 | osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id; |
d30291b9 ID |
2055 | if (ceph_oid_aprintf(&osd_req->r_base_oid, GFP_NOIO, "%s", |
2056 | obj_request->object_name)) | |
2057 | goto fail; | |
0eefd470 | 2058 | |
13d1ad16 ID |
2059 | if (ceph_osdc_alloc_messages(osd_req, GFP_NOIO)) |
2060 | goto fail; | |
2061 | ||
0eefd470 | 2062 | return osd_req; |
13d1ad16 ID |
2063 | |
2064 | fail: | |
2065 | ceph_osdc_put_request(osd_req); | |
2066 | return NULL; | |
0eefd470 AE |
2067 | } |
2068 | ||
2069 | ||
bf0d5f50 AE |
2070 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2071 | { | |
2072 | ceph_osdc_put_request(osd_req); | |
2073 | } | |
2074 | ||
2075 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2076 | ||
2077 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2078 | u64 offset, u64 length, | |
2079 | enum obj_request_type type) | |
2080 | { | |
2081 | struct rbd_obj_request *obj_request; | |
2082 | size_t size; | |
2083 | char *name; | |
2084 | ||
2085 | rbd_assert(obj_request_type_valid(type)); | |
2086 | ||
2087 | size = strlen(object_name) + 1; | |
5a60e876 | 2088 | name = kmalloc(size, GFP_NOIO); |
f907ad55 | 2089 | if (!name) |
bf0d5f50 AE |
2090 | return NULL; |
2091 | ||
5a60e876 | 2092 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
f907ad55 AE |
2093 | if (!obj_request) { |
2094 | kfree(name); | |
2095 | return NULL; | |
2096 | } | |
2097 | ||
bf0d5f50 AE |
2098 | obj_request->object_name = memcpy(name, object_name, size); |
2099 | obj_request->offset = offset; | |
2100 | obj_request->length = length; | |
926f9b3f | 2101 | obj_request->flags = 0; |
bf0d5f50 AE |
2102 | obj_request->which = BAD_WHICH; |
2103 | obj_request->type = type; | |
2104 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2105 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2106 | kref_init(&obj_request->kref); |
2107 | ||
37206ee5 AE |
2108 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2109 | offset, length, (int)type, obj_request); | |
2110 | ||
bf0d5f50 AE |
2111 | return obj_request; |
2112 | } | |
2113 | ||
2114 | static void rbd_obj_request_destroy(struct kref *kref) | |
2115 | { | |
2116 | struct rbd_obj_request *obj_request; | |
2117 | ||
2118 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2119 | ||
37206ee5 AE |
2120 | dout("%s: obj %p\n", __func__, obj_request); |
2121 | ||
bf0d5f50 AE |
2122 | rbd_assert(obj_request->img_request == NULL); |
2123 | rbd_assert(obj_request->which == BAD_WHICH); | |
2124 | ||
2125 | if (obj_request->osd_req) | |
2126 | rbd_osd_req_destroy(obj_request->osd_req); | |
2127 | ||
2128 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2129 | switch (obj_request->type) { | |
9969ebc5 AE |
2130 | case OBJ_REQUEST_NODATA: |
2131 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2132 | case OBJ_REQUEST_BIO: |
2133 | if (obj_request->bio_list) | |
2134 | bio_chain_put(obj_request->bio_list); | |
2135 | break; | |
788e2df3 AE |
2136 | case OBJ_REQUEST_PAGES: |
2137 | if (obj_request->pages) | |
2138 | ceph_release_page_vector(obj_request->pages, | |
2139 | obj_request->page_count); | |
2140 | break; | |
bf0d5f50 AE |
2141 | } |
2142 | ||
f907ad55 | 2143 | kfree(obj_request->object_name); |
868311b1 AE |
2144 | obj_request->object_name = NULL; |
2145 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2146 | } |
2147 | ||
fb65d228 AE |
2148 | /* It's OK to call this for a device with no parent */ |
2149 | ||
2150 | static void rbd_spec_put(struct rbd_spec *spec); | |
2151 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2152 | { | |
2153 | rbd_dev_remove_parent(rbd_dev); | |
2154 | rbd_spec_put(rbd_dev->parent_spec); | |
2155 | rbd_dev->parent_spec = NULL; | |
2156 | rbd_dev->parent_overlap = 0; | |
2157 | } | |
2158 | ||
a2acd00e AE |
2159 | /* |
2160 | * Parent image reference counting is used to determine when an | |
2161 | * image's parent fields can be safely torn down--after there are no | |
2162 | * more in-flight requests to the parent image. When the last | |
2163 | * reference is dropped, cleaning them up is safe. | |
2164 | */ | |
2165 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2166 | { | |
2167 | int counter; | |
2168 | ||
2169 | if (!rbd_dev->parent_spec) | |
2170 | return; | |
2171 | ||
2172 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2173 | if (counter > 0) | |
2174 | return; | |
2175 | ||
2176 | /* Last reference; clean up parent data structures */ | |
2177 | ||
2178 | if (!counter) | |
2179 | rbd_dev_unparent(rbd_dev); | |
2180 | else | |
9584d508 | 2181 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2182 | } |
2183 | ||
2184 | /* | |
2185 | * If an image has a non-zero parent overlap, get a reference to its | |
2186 | * parent. | |
2187 | * | |
2188 | * Returns true if the rbd device has a parent with a non-zero | |
2189 | * overlap and a reference for it was successfully taken, or | |
2190 | * false otherwise. | |
2191 | */ | |
2192 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2193 | { | |
ae43e9d0 | 2194 | int counter = 0; |
a2acd00e AE |
2195 | |
2196 | if (!rbd_dev->parent_spec) | |
2197 | return false; | |
2198 | ||
ae43e9d0 ID |
2199 | down_read(&rbd_dev->header_rwsem); |
2200 | if (rbd_dev->parent_overlap) | |
2201 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2202 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2203 | |
2204 | if (counter < 0) | |
9584d508 | 2205 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2206 | |
ae43e9d0 | 2207 | return counter > 0; |
a2acd00e AE |
2208 | } |
2209 | ||
bf0d5f50 AE |
2210 | /* |
2211 | * Caller is responsible for filling in the list of object requests | |
2212 | * that comprises the image request, and the Linux request pointer | |
2213 | * (if there is one). | |
2214 | */ | |
cc344fa1 AE |
2215 | static struct rbd_img_request *rbd_img_request_create( |
2216 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2217 | u64 offset, u64 length, |
6d2940c8 | 2218 | enum obj_operation_type op_type, |
4e752f0a | 2219 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2220 | { |
2221 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2222 | |
7a716aac | 2223 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2224 | if (!img_request) |
2225 | return NULL; | |
2226 | ||
bf0d5f50 AE |
2227 | img_request->rq = NULL; |
2228 | img_request->rbd_dev = rbd_dev; | |
2229 | img_request->offset = offset; | |
2230 | img_request->length = length; | |
0c425248 | 2231 | img_request->flags = 0; |
90e98c52 GZ |
2232 | if (op_type == OBJ_OP_DISCARD) { |
2233 | img_request_discard_set(img_request); | |
2234 | img_request->snapc = snapc; | |
2235 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2236 | img_request_write_set(img_request); |
4e752f0a | 2237 | img_request->snapc = snapc; |
0c425248 | 2238 | } else { |
bf0d5f50 | 2239 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2240 | } |
a2acd00e | 2241 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2242 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2243 | spin_lock_init(&img_request->completion_lock); |
2244 | img_request->next_completion = 0; | |
2245 | img_request->callback = NULL; | |
a5a337d4 | 2246 | img_request->result = 0; |
bf0d5f50 AE |
2247 | img_request->obj_request_count = 0; |
2248 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2249 | kref_init(&img_request->kref); | |
2250 | ||
37206ee5 | 2251 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2252 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2253 | |
bf0d5f50 AE |
2254 | return img_request; |
2255 | } | |
2256 | ||
2257 | static void rbd_img_request_destroy(struct kref *kref) | |
2258 | { | |
2259 | struct rbd_img_request *img_request; | |
2260 | struct rbd_obj_request *obj_request; | |
2261 | struct rbd_obj_request *next_obj_request; | |
2262 | ||
2263 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2264 | ||
37206ee5 AE |
2265 | dout("%s: img %p\n", __func__, img_request); |
2266 | ||
bf0d5f50 AE |
2267 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2268 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2269 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2270 | |
a2acd00e AE |
2271 | if (img_request_layered_test(img_request)) { |
2272 | img_request_layered_clear(img_request); | |
2273 | rbd_dev_parent_put(img_request->rbd_dev); | |
2274 | } | |
2275 | ||
bef95455 JD |
2276 | if (img_request_write_test(img_request) || |
2277 | img_request_discard_test(img_request)) | |
812164f8 | 2278 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2279 | |
1c2a9dfe | 2280 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2281 | } |
2282 | ||
e93f3152 AE |
2283 | static struct rbd_img_request *rbd_parent_request_create( |
2284 | struct rbd_obj_request *obj_request, | |
2285 | u64 img_offset, u64 length) | |
2286 | { | |
2287 | struct rbd_img_request *parent_request; | |
2288 | struct rbd_device *rbd_dev; | |
2289 | ||
2290 | rbd_assert(obj_request->img_request); | |
2291 | rbd_dev = obj_request->img_request->rbd_dev; | |
2292 | ||
4e752f0a | 2293 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2294 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2295 | if (!parent_request) |
2296 | return NULL; | |
2297 | ||
2298 | img_request_child_set(parent_request); | |
2299 | rbd_obj_request_get(obj_request); | |
2300 | parent_request->obj_request = obj_request; | |
2301 | ||
2302 | return parent_request; | |
2303 | } | |
2304 | ||
2305 | static void rbd_parent_request_destroy(struct kref *kref) | |
2306 | { | |
2307 | struct rbd_img_request *parent_request; | |
2308 | struct rbd_obj_request *orig_request; | |
2309 | ||
2310 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2311 | orig_request = parent_request->obj_request; | |
2312 | ||
2313 | parent_request->obj_request = NULL; | |
2314 | rbd_obj_request_put(orig_request); | |
2315 | img_request_child_clear(parent_request); | |
2316 | ||
2317 | rbd_img_request_destroy(kref); | |
2318 | } | |
2319 | ||
1217857f AE |
2320 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2321 | { | |
6365d33a | 2322 | struct rbd_img_request *img_request; |
1217857f AE |
2323 | unsigned int xferred; |
2324 | int result; | |
8b3e1a56 | 2325 | bool more; |
1217857f | 2326 | |
6365d33a AE |
2327 | rbd_assert(obj_request_img_data_test(obj_request)); |
2328 | img_request = obj_request->img_request; | |
2329 | ||
1217857f AE |
2330 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2331 | xferred = (unsigned int)obj_request->xferred; | |
2332 | result = obj_request->result; | |
2333 | if (result) { | |
2334 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2335 | enum obj_operation_type op_type; |
2336 | ||
90e98c52 GZ |
2337 | if (img_request_discard_test(img_request)) |
2338 | op_type = OBJ_OP_DISCARD; | |
2339 | else if (img_request_write_test(img_request)) | |
2340 | op_type = OBJ_OP_WRITE; | |
2341 | else | |
2342 | op_type = OBJ_OP_READ; | |
1217857f | 2343 | |
9584d508 | 2344 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2345 | obj_op_name(op_type), obj_request->length, |
2346 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2347 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2348 | result, xferred); |
2349 | if (!img_request->result) | |
2350 | img_request->result = result; | |
082a75da ID |
2351 | /* |
2352 | * Need to end I/O on the entire obj_request worth of | |
2353 | * bytes in case of error. | |
2354 | */ | |
2355 | xferred = obj_request->length; | |
1217857f AE |
2356 | } |
2357 | ||
f1a4739f AE |
2358 | /* Image object requests don't own their page array */ |
2359 | ||
2360 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2361 | obj_request->pages = NULL; | |
2362 | obj_request->page_count = 0; | |
2363 | } | |
2364 | ||
8b3e1a56 AE |
2365 | if (img_request_child_test(img_request)) { |
2366 | rbd_assert(img_request->obj_request != NULL); | |
2367 | more = obj_request->which < img_request->obj_request_count - 1; | |
2368 | } else { | |
2369 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2370 | |
2371 | more = blk_update_request(img_request->rq, result, xferred); | |
2372 | if (!more) | |
2373 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2374 | } |
2375 | ||
2376 | return more; | |
1217857f AE |
2377 | } |
2378 | ||
2169238d AE |
2379 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2380 | { | |
2381 | struct rbd_img_request *img_request; | |
2382 | u32 which = obj_request->which; | |
2383 | bool more = true; | |
2384 | ||
6365d33a | 2385 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2386 | img_request = obj_request->img_request; |
2387 | ||
2388 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2389 | rbd_assert(img_request != NULL); | |
2169238d AE |
2390 | rbd_assert(img_request->obj_request_count > 0); |
2391 | rbd_assert(which != BAD_WHICH); | |
2392 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2393 | |
2394 | spin_lock_irq(&img_request->completion_lock); | |
2395 | if (which != img_request->next_completion) | |
2396 | goto out; | |
2397 | ||
2398 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2399 | rbd_assert(more); |
2400 | rbd_assert(which < img_request->obj_request_count); | |
2401 | ||
2402 | if (!obj_request_done_test(obj_request)) | |
2403 | break; | |
1217857f | 2404 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2405 | which++; |
2406 | } | |
2407 | ||
2408 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2409 | img_request->next_completion = which; | |
2410 | out: | |
2411 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2412 | rbd_img_request_put(img_request); |
2169238d AE |
2413 | |
2414 | if (!more) | |
2415 | rbd_img_request_complete(img_request); | |
2416 | } | |
2417 | ||
3b434a2a JD |
2418 | /* |
2419 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2420 | * them for submission. num_ops is the current number of | |
2421 | * osd operations already to the object request. | |
2422 | */ | |
2423 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2424 | struct ceph_osd_request *osd_request, | |
2425 | enum obj_operation_type op_type, | |
2426 | unsigned int num_ops) | |
2427 | { | |
2428 | struct rbd_img_request *img_request = obj_request->img_request; | |
2429 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2430 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2431 | u64 offset = obj_request->offset; | |
2432 | u64 length = obj_request->length; | |
2433 | u64 img_end; | |
2434 | u16 opcode; | |
2435 | ||
2436 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2437 | if (!offset && length == object_size && |
2438 | (!img_request_layered_test(img_request) || | |
2439 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2440 | opcode = CEPH_OSD_OP_DELETE; |
2441 | } else if ((offset + length == object_size)) { | |
2442 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2443 | } else { | |
2444 | down_read(&rbd_dev->header_rwsem); | |
2445 | img_end = rbd_dev->header.image_size; | |
2446 | up_read(&rbd_dev->header_rwsem); | |
2447 | ||
2448 | if (obj_request->img_offset + length == img_end) | |
2449 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2450 | else | |
2451 | opcode = CEPH_OSD_OP_ZERO; | |
2452 | } | |
2453 | } else if (op_type == OBJ_OP_WRITE) { | |
e30b7577 ID |
2454 | if (!offset && length == object_size) |
2455 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2456 | else | |
2457 | opcode = CEPH_OSD_OP_WRITE; | |
3b434a2a JD |
2458 | osd_req_op_alloc_hint_init(osd_request, num_ops, |
2459 | object_size, object_size); | |
2460 | num_ops++; | |
2461 | } else { | |
2462 | opcode = CEPH_OSD_OP_READ; | |
2463 | } | |
2464 | ||
7e868b6e | 2465 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2466 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2467 | else |
2468 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2469 | offset, length, 0, 0); | |
2470 | ||
3b434a2a JD |
2471 | if (obj_request->type == OBJ_REQUEST_BIO) |
2472 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2473 | obj_request->bio_list, length); | |
2474 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2475 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2476 | obj_request->pages, length, | |
2477 | offset & ~PAGE_MASK, false, false); | |
2478 | ||
2479 | /* Discards are also writes */ | |
2480 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2481 | rbd_osd_req_format_write(obj_request); | |
2482 | else | |
2483 | rbd_osd_req_format_read(obj_request); | |
2484 | } | |
2485 | ||
f1a4739f AE |
2486 | /* |
2487 | * Split up an image request into one or more object requests, each | |
2488 | * to a different object. The "type" parameter indicates whether | |
2489 | * "data_desc" is the pointer to the head of a list of bio | |
2490 | * structures, or the base of a page array. In either case this | |
2491 | * function assumes data_desc describes memory sufficient to hold | |
2492 | * all data described by the image request. | |
2493 | */ | |
2494 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2495 | enum obj_request_type type, | |
2496 | void *data_desc) | |
bf0d5f50 AE |
2497 | { |
2498 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2499 | struct rbd_obj_request *obj_request = NULL; | |
2500 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2501 | struct bio *bio_list = NULL; |
f1a4739f | 2502 | unsigned int bio_offset = 0; |
a158073c | 2503 | struct page **pages = NULL; |
6d2940c8 | 2504 | enum obj_operation_type op_type; |
7da22d29 | 2505 | u64 img_offset; |
bf0d5f50 | 2506 | u64 resid; |
bf0d5f50 | 2507 | |
f1a4739f AE |
2508 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2509 | (int)type, data_desc); | |
37206ee5 | 2510 | |
7da22d29 | 2511 | img_offset = img_request->offset; |
bf0d5f50 | 2512 | resid = img_request->length; |
4dda41d3 | 2513 | rbd_assert(resid > 0); |
3b434a2a | 2514 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2515 | |
2516 | if (type == OBJ_REQUEST_BIO) { | |
2517 | bio_list = data_desc; | |
4f024f37 KO |
2518 | rbd_assert(img_offset == |
2519 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2520 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2521 | pages = data_desc; |
2522 | } | |
2523 | ||
bf0d5f50 | 2524 | while (resid) { |
2fa12320 | 2525 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2526 | const char *object_name; |
bf0d5f50 AE |
2527 | u64 offset; |
2528 | u64 length; | |
2529 | ||
7da22d29 | 2530 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2531 | if (!object_name) |
2532 | goto out_unwind; | |
7da22d29 AE |
2533 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2534 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2535 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2536 | offset, length, type); |
78c2a44a AE |
2537 | /* object request has its own copy of the object name */ |
2538 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2539 | if (!obj_request) |
2540 | goto out_unwind; | |
62054da6 | 2541 | |
03507db6 JD |
2542 | /* |
2543 | * set obj_request->img_request before creating the | |
2544 | * osd_request so that it gets the right snapc | |
2545 | */ | |
2546 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2547 | |
f1a4739f AE |
2548 | if (type == OBJ_REQUEST_BIO) { |
2549 | unsigned int clone_size; | |
2550 | ||
2551 | rbd_assert(length <= (u64)UINT_MAX); | |
2552 | clone_size = (unsigned int)length; | |
2553 | obj_request->bio_list = | |
2554 | bio_chain_clone_range(&bio_list, | |
2555 | &bio_offset, | |
2556 | clone_size, | |
2224d879 | 2557 | GFP_NOIO); |
f1a4739f | 2558 | if (!obj_request->bio_list) |
62054da6 | 2559 | goto out_unwind; |
90e98c52 | 2560 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2561 | unsigned int page_count; |
2562 | ||
2563 | obj_request->pages = pages; | |
2564 | page_count = (u32)calc_pages_for(offset, length); | |
2565 | obj_request->page_count = page_count; | |
2566 | if ((offset + length) & ~PAGE_MASK) | |
2567 | page_count--; /* more on last page */ | |
2568 | pages += page_count; | |
2569 | } | |
bf0d5f50 | 2570 | |
6d2940c8 GZ |
2571 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2572 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2573 | obj_request); | |
2fa12320 | 2574 | if (!osd_req) |
62054da6 | 2575 | goto out_unwind; |
3b434a2a | 2576 | |
2fa12320 | 2577 | obj_request->osd_req = osd_req; |
2169238d | 2578 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2579 | obj_request->img_offset = img_offset; |
9d4df01f | 2580 | |
3b434a2a | 2581 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2582 | |
3b434a2a | 2583 | rbd_img_request_get(img_request); |
bf0d5f50 | 2584 | |
7da22d29 | 2585 | img_offset += length; |
bf0d5f50 AE |
2586 | resid -= length; |
2587 | } | |
2588 | ||
2589 | return 0; | |
2590 | ||
bf0d5f50 AE |
2591 | out_unwind: |
2592 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2593 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2594 | |
2595 | return -ENOMEM; | |
2596 | } | |
2597 | ||
0eefd470 | 2598 | static void |
2761713d | 2599 | rbd_osd_copyup_callback(struct rbd_obj_request *obj_request) |
0eefd470 AE |
2600 | { |
2601 | struct rbd_img_request *img_request; | |
2602 | struct rbd_device *rbd_dev; | |
ebda6408 | 2603 | struct page **pages; |
0eefd470 AE |
2604 | u32 page_count; |
2605 | ||
2761713d ID |
2606 | dout("%s: obj %p\n", __func__, obj_request); |
2607 | ||
d3246fb0 JD |
2608 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2609 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2610 | rbd_assert(obj_request_img_data_test(obj_request)); |
2611 | img_request = obj_request->img_request; | |
2612 | rbd_assert(img_request); | |
2613 | ||
2614 | rbd_dev = img_request->rbd_dev; | |
2615 | rbd_assert(rbd_dev); | |
0eefd470 | 2616 | |
ebda6408 AE |
2617 | pages = obj_request->copyup_pages; |
2618 | rbd_assert(pages != NULL); | |
0eefd470 | 2619 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2620 | page_count = obj_request->copyup_page_count; |
2621 | rbd_assert(page_count); | |
2622 | obj_request->copyup_page_count = 0; | |
2623 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2624 | |
2625 | /* | |
2626 | * We want the transfer count to reflect the size of the | |
2627 | * original write request. There is no such thing as a | |
2628 | * successful short write, so if the request was successful | |
2629 | * we can just set it to the originally-requested length. | |
2630 | */ | |
2631 | if (!obj_request->result) | |
2632 | obj_request->xferred = obj_request->length; | |
2633 | ||
2761713d | 2634 | obj_request_done_set(obj_request); |
0eefd470 AE |
2635 | } |
2636 | ||
3d7efd18 AE |
2637 | static void |
2638 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2639 | { | |
2640 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2641 | struct ceph_osd_request *osd_req; |
2642 | struct ceph_osd_client *osdc; | |
2643 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2644 | struct page **pages; |
d3246fb0 | 2645 | enum obj_operation_type op_type; |
ebda6408 | 2646 | u32 page_count; |
bbea1c1a | 2647 | int img_result; |
ebda6408 | 2648 | u64 parent_length; |
3d7efd18 AE |
2649 | |
2650 | rbd_assert(img_request_child_test(img_request)); | |
2651 | ||
2652 | /* First get what we need from the image request */ | |
2653 | ||
2654 | pages = img_request->copyup_pages; | |
2655 | rbd_assert(pages != NULL); | |
2656 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2657 | page_count = img_request->copyup_page_count; |
2658 | rbd_assert(page_count); | |
2659 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2660 | |
2661 | orig_request = img_request->obj_request; | |
2662 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2663 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2664 | img_result = img_request->result; |
ebda6408 AE |
2665 | parent_length = img_request->length; |
2666 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2667 | rbd_img_request_put(img_request); |
3d7efd18 | 2668 | |
91c6febb AE |
2669 | rbd_assert(orig_request->img_request); |
2670 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2671 | rbd_assert(rbd_dev); |
0eefd470 | 2672 | |
bbea1c1a AE |
2673 | /* |
2674 | * If the overlap has become 0 (most likely because the | |
2675 | * image has been flattened) we need to free the pages | |
2676 | * and re-submit the original write request. | |
2677 | */ | |
2678 | if (!rbd_dev->parent_overlap) { | |
2679 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2680 | |
bbea1c1a AE |
2681 | ceph_release_page_vector(pages, page_count); |
2682 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2683 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2684 | if (!img_result) | |
2685 | return; | |
2686 | } | |
0eefd470 | 2687 | |
bbea1c1a | 2688 | if (img_result) |
0eefd470 | 2689 | goto out_err; |
0eefd470 | 2690 | |
8785b1d4 AE |
2691 | /* |
2692 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2693 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2694 | * request. Allocate the new copyup osd request for the |
2695 | * original request, and release the old one. | |
2696 | */ | |
bbea1c1a | 2697 | img_result = -ENOMEM; |
0eefd470 AE |
2698 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2699 | if (!osd_req) | |
2700 | goto out_err; | |
8785b1d4 | 2701 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2702 | orig_request->osd_req = osd_req; |
2703 | orig_request->copyup_pages = pages; | |
ebda6408 | 2704 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2705 | |
0eefd470 | 2706 | /* Initialize the copyup op */ |
3d7efd18 | 2707 | |
0eefd470 | 2708 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2709 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2710 | false, false); |
3d7efd18 | 2711 | |
d3246fb0 | 2712 | /* Add the other op(s) */ |
0eefd470 | 2713 | |
d3246fb0 JD |
2714 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2715 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2716 | |
2717 | /* All set, send it off. */ | |
2718 | ||
0eefd470 | 2719 | osdc = &rbd_dev->rbd_client->client->osdc; |
bbea1c1a AE |
2720 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2721 | if (!img_result) | |
0eefd470 AE |
2722 | return; |
2723 | out_err: | |
2724 | /* Record the error code and complete the request */ | |
2725 | ||
bbea1c1a | 2726 | orig_request->result = img_result; |
0eefd470 AE |
2727 | orig_request->xferred = 0; |
2728 | obj_request_done_set(orig_request); | |
2729 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2730 | } |
2731 | ||
2732 | /* | |
2733 | * Read from the parent image the range of data that covers the | |
2734 | * entire target of the given object request. This is used for | |
2735 | * satisfying a layered image write request when the target of an | |
2736 | * object request from the image request does not exist. | |
2737 | * | |
2738 | * A page array big enough to hold the returned data is allocated | |
2739 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2740 | * When the read completes, this page array will be transferred to | |
2741 | * the original object request for the copyup operation. | |
2742 | * | |
2743 | * If an error occurs, record it as the result of the original | |
2744 | * object request and mark it done so it gets completed. | |
2745 | */ | |
2746 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2747 | { | |
2748 | struct rbd_img_request *img_request = NULL; | |
2749 | struct rbd_img_request *parent_request = NULL; | |
2750 | struct rbd_device *rbd_dev; | |
2751 | u64 img_offset; | |
2752 | u64 length; | |
2753 | struct page **pages = NULL; | |
2754 | u32 page_count; | |
2755 | int result; | |
2756 | ||
2757 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2758 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2759 | |
2760 | img_request = obj_request->img_request; | |
2761 | rbd_assert(img_request != NULL); | |
2762 | rbd_dev = img_request->rbd_dev; | |
2763 | rbd_assert(rbd_dev->parent != NULL); | |
2764 | ||
2765 | /* | |
2766 | * Determine the byte range covered by the object in the | |
2767 | * child image to which the original request was to be sent. | |
2768 | */ | |
2769 | img_offset = obj_request->img_offset - obj_request->offset; | |
2770 | length = (u64)1 << rbd_dev->header.obj_order; | |
2771 | ||
a9e8ba2c AE |
2772 | /* |
2773 | * There is no defined parent data beyond the parent | |
2774 | * overlap, so limit what we read at that boundary if | |
2775 | * necessary. | |
2776 | */ | |
2777 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2778 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2779 | length = rbd_dev->parent_overlap - img_offset; | |
2780 | } | |
2781 | ||
3d7efd18 AE |
2782 | /* |
2783 | * Allocate a page array big enough to receive the data read | |
2784 | * from the parent. | |
2785 | */ | |
2786 | page_count = (u32)calc_pages_for(0, length); | |
2787 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2788 | if (IS_ERR(pages)) { | |
2789 | result = PTR_ERR(pages); | |
2790 | pages = NULL; | |
2791 | goto out_err; | |
2792 | } | |
2793 | ||
2794 | result = -ENOMEM; | |
e93f3152 AE |
2795 | parent_request = rbd_parent_request_create(obj_request, |
2796 | img_offset, length); | |
3d7efd18 AE |
2797 | if (!parent_request) |
2798 | goto out_err; | |
3d7efd18 AE |
2799 | |
2800 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2801 | if (result) | |
2802 | goto out_err; | |
2803 | parent_request->copyup_pages = pages; | |
ebda6408 | 2804 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2805 | |
2806 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2807 | result = rbd_img_request_submit(parent_request); | |
2808 | if (!result) | |
2809 | return 0; | |
2810 | ||
2811 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2812 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2813 | parent_request->obj_request = NULL; |
2814 | rbd_obj_request_put(obj_request); | |
2815 | out_err: | |
2816 | if (pages) | |
2817 | ceph_release_page_vector(pages, page_count); | |
2818 | if (parent_request) | |
2819 | rbd_img_request_put(parent_request); | |
2820 | obj_request->result = result; | |
2821 | obj_request->xferred = 0; | |
2822 | obj_request_done_set(obj_request); | |
2823 | ||
2824 | return result; | |
2825 | } | |
2826 | ||
c5b5ef6c AE |
2827 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2828 | { | |
c5b5ef6c | 2829 | struct rbd_obj_request *orig_request; |
638f5abe | 2830 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2831 | int result; |
2832 | ||
2833 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2834 | ||
2835 | /* | |
2836 | * All we need from the object request is the original | |
2837 | * request and the result of the STAT op. Grab those, then | |
2838 | * we're done with the request. | |
2839 | */ | |
2840 | orig_request = obj_request->obj_request; | |
2841 | obj_request->obj_request = NULL; | |
912c317d | 2842 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2843 | rbd_assert(orig_request); |
2844 | rbd_assert(orig_request->img_request); | |
2845 | ||
2846 | result = obj_request->result; | |
2847 | obj_request->result = 0; | |
2848 | ||
2849 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2850 | obj_request, orig_request, result, | |
2851 | obj_request->xferred, obj_request->length); | |
2852 | rbd_obj_request_put(obj_request); | |
2853 | ||
638f5abe AE |
2854 | /* |
2855 | * If the overlap has become 0 (most likely because the | |
2856 | * image has been flattened) we need to free the pages | |
2857 | * and re-submit the original write request. | |
2858 | */ | |
2859 | rbd_dev = orig_request->img_request->rbd_dev; | |
2860 | if (!rbd_dev->parent_overlap) { | |
2861 | struct ceph_osd_client *osdc; | |
2862 | ||
638f5abe AE |
2863 | osdc = &rbd_dev->rbd_client->client->osdc; |
2864 | result = rbd_obj_request_submit(osdc, orig_request); | |
2865 | if (!result) | |
2866 | return; | |
2867 | } | |
c5b5ef6c AE |
2868 | |
2869 | /* | |
2870 | * Our only purpose here is to determine whether the object | |
2871 | * exists, and we don't want to treat the non-existence as | |
2872 | * an error. If something else comes back, transfer the | |
2873 | * error to the original request and complete it now. | |
2874 | */ | |
2875 | if (!result) { | |
2876 | obj_request_existence_set(orig_request, true); | |
2877 | } else if (result == -ENOENT) { | |
2878 | obj_request_existence_set(orig_request, false); | |
2879 | } else if (result) { | |
2880 | orig_request->result = result; | |
3d7efd18 | 2881 | goto out; |
c5b5ef6c AE |
2882 | } |
2883 | ||
2884 | /* | |
2885 | * Resubmit the original request now that we have recorded | |
2886 | * whether the target object exists. | |
2887 | */ | |
b454e36d | 2888 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2889 | out: |
c5b5ef6c AE |
2890 | if (orig_request->result) |
2891 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2892 | } |
2893 | ||
2894 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2895 | { | |
2896 | struct rbd_obj_request *stat_request; | |
2897 | struct rbd_device *rbd_dev; | |
2898 | struct ceph_osd_client *osdc; | |
2899 | struct page **pages = NULL; | |
2900 | u32 page_count; | |
2901 | size_t size; | |
2902 | int ret; | |
2903 | ||
2904 | /* | |
2905 | * The response data for a STAT call consists of: | |
2906 | * le64 length; | |
2907 | * struct { | |
2908 | * le32 tv_sec; | |
2909 | * le32 tv_nsec; | |
2910 | * } mtime; | |
2911 | */ | |
2912 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2913 | page_count = (u32)calc_pages_for(0, size); | |
2914 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2915 | if (IS_ERR(pages)) | |
2916 | return PTR_ERR(pages); | |
2917 | ||
2918 | ret = -ENOMEM; | |
2919 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2920 | OBJ_REQUEST_PAGES); | |
2921 | if (!stat_request) | |
2922 | goto out; | |
2923 | ||
2924 | rbd_obj_request_get(obj_request); | |
2925 | stat_request->obj_request = obj_request; | |
2926 | stat_request->pages = pages; | |
2927 | stat_request->page_count = page_count; | |
2928 | ||
2929 | rbd_assert(obj_request->img_request); | |
2930 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2931 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2932 | stat_request); |
c5b5ef6c AE |
2933 | if (!stat_request->osd_req) |
2934 | goto out; | |
2935 | stat_request->callback = rbd_img_obj_exists_callback; | |
2936 | ||
144cba14 | 2937 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2938 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2939 | false, false); | |
9d4df01f | 2940 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2941 | |
2942 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2943 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2944 | out: | |
2945 | if (ret) | |
2946 | rbd_obj_request_put(obj_request); | |
2947 | ||
2948 | return ret; | |
2949 | } | |
2950 | ||
70d045f6 | 2951 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2952 | { |
2953 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2954 | struct rbd_device *rbd_dev; |
b454e36d AE |
2955 | |
2956 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2957 | ||
2958 | img_request = obj_request->img_request; | |
2959 | rbd_assert(img_request); | |
a9e8ba2c | 2960 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2961 | |
70d045f6 | 2962 | /* Reads */ |
1c220881 JD |
2963 | if (!img_request_write_test(img_request) && |
2964 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2965 | return true; |
2966 | ||
2967 | /* Non-layered writes */ | |
2968 | if (!img_request_layered_test(img_request)) | |
2969 | return true; | |
2970 | ||
b454e36d | 2971 | /* |
70d045f6 ID |
2972 | * Layered writes outside of the parent overlap range don't |
2973 | * share any data with the parent. | |
b454e36d | 2974 | */ |
70d045f6 ID |
2975 | if (!obj_request_overlaps_parent(obj_request)) |
2976 | return true; | |
b454e36d | 2977 | |
c622d226 GZ |
2978 | /* |
2979 | * Entire-object layered writes - we will overwrite whatever | |
2980 | * parent data there is anyway. | |
2981 | */ | |
2982 | if (!obj_request->offset && | |
2983 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2984 | return true; | |
2985 | ||
70d045f6 ID |
2986 | /* |
2987 | * If the object is known to already exist, its parent data has | |
2988 | * already been copied. | |
2989 | */ | |
2990 | if (obj_request_known_test(obj_request) && | |
2991 | obj_request_exists_test(obj_request)) | |
2992 | return true; | |
2993 | ||
2994 | return false; | |
2995 | } | |
2996 | ||
2997 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2998 | { | |
2999 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
3000 | struct rbd_device *rbd_dev; |
3001 | struct ceph_osd_client *osdc; | |
3002 | ||
3003 | rbd_dev = obj_request->img_request->rbd_dev; | |
3004 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3005 | ||
3006 | return rbd_obj_request_submit(osdc, obj_request); | |
3007 | } | |
3008 | ||
3009 | /* | |
3d7efd18 AE |
3010 | * It's a layered write. The target object might exist but |
3011 | * we may not know that yet. If we know it doesn't exist, | |
3012 | * start by reading the data for the full target object from | |
3013 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 3014 | */ |
70d045f6 | 3015 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
3016 | return rbd_img_obj_parent_read_full(obj_request); |
3017 | ||
3018 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
3019 | |
3020 | return rbd_img_obj_exists_submit(obj_request); | |
3021 | } | |
3022 | ||
bf0d5f50 AE |
3023 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
3024 | { | |
bf0d5f50 | 3025 | struct rbd_obj_request *obj_request; |
46faeed4 | 3026 | struct rbd_obj_request *next_obj_request; |
663ae2cc | 3027 | int ret = 0; |
bf0d5f50 | 3028 | |
37206ee5 | 3029 | dout("%s: img %p\n", __func__, img_request); |
bf0d5f50 | 3030 | |
663ae2cc ID |
3031 | rbd_img_request_get(img_request); |
3032 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { | |
b454e36d | 3033 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 | 3034 | if (ret) |
663ae2cc | 3035 | goto out_put_ireq; |
bf0d5f50 AE |
3036 | } |
3037 | ||
663ae2cc ID |
3038 | out_put_ireq: |
3039 | rbd_img_request_put(img_request); | |
3040 | return ret; | |
bf0d5f50 | 3041 | } |
8b3e1a56 AE |
3042 | |
3043 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
3044 | { | |
3045 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
3046 | struct rbd_device *rbd_dev; |
3047 | u64 obj_end; | |
02c74fba AE |
3048 | u64 img_xferred; |
3049 | int img_result; | |
8b3e1a56 AE |
3050 | |
3051 | rbd_assert(img_request_child_test(img_request)); | |
3052 | ||
02c74fba AE |
3053 | /* First get what we need from the image request and release it */ |
3054 | ||
8b3e1a56 | 3055 | obj_request = img_request->obj_request; |
02c74fba AE |
3056 | img_xferred = img_request->xferred; |
3057 | img_result = img_request->result; | |
3058 | rbd_img_request_put(img_request); | |
3059 | ||
3060 | /* | |
3061 | * If the overlap has become 0 (most likely because the | |
3062 | * image has been flattened) we need to re-submit the | |
3063 | * original request. | |
3064 | */ | |
a9e8ba2c AE |
3065 | rbd_assert(obj_request); |
3066 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
3067 | rbd_dev = obj_request->img_request->rbd_dev; |
3068 | if (!rbd_dev->parent_overlap) { | |
3069 | struct ceph_osd_client *osdc; | |
3070 | ||
3071 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3072 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
3073 | if (!img_result) | |
3074 | return; | |
3075 | } | |
a9e8ba2c | 3076 | |
02c74fba | 3077 | obj_request->result = img_result; |
a9e8ba2c AE |
3078 | if (obj_request->result) |
3079 | goto out; | |
3080 | ||
3081 | /* | |
3082 | * We need to zero anything beyond the parent overlap | |
3083 | * boundary. Since rbd_img_obj_request_read_callback() | |
3084 | * will zero anything beyond the end of a short read, an | |
3085 | * easy way to do this is to pretend the data from the | |
3086 | * parent came up short--ending at the overlap boundary. | |
3087 | */ | |
3088 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3089 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3090 | if (obj_end > rbd_dev->parent_overlap) { |
3091 | u64 xferred = 0; | |
3092 | ||
3093 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3094 | xferred = rbd_dev->parent_overlap - | |
3095 | obj_request->img_offset; | |
8b3e1a56 | 3096 | |
02c74fba | 3097 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3098 | } else { |
02c74fba | 3099 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3100 | } |
3101 | out: | |
8b3e1a56 AE |
3102 | rbd_img_obj_request_read_callback(obj_request); |
3103 | rbd_obj_request_complete(obj_request); | |
3104 | } | |
3105 | ||
3106 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3107 | { | |
8b3e1a56 AE |
3108 | struct rbd_img_request *img_request; |
3109 | int result; | |
3110 | ||
3111 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3112 | rbd_assert(obj_request->img_request != NULL); | |
3113 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3114 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3115 | |
8b3e1a56 | 3116 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3117 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3118 | obj_request->img_offset, |
e93f3152 | 3119 | obj_request->length); |
8b3e1a56 AE |
3120 | result = -ENOMEM; |
3121 | if (!img_request) | |
3122 | goto out_err; | |
3123 | ||
5b2ab72d AE |
3124 | if (obj_request->type == OBJ_REQUEST_BIO) |
3125 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3126 | obj_request->bio_list); | |
3127 | else | |
3128 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3129 | obj_request->pages); | |
8b3e1a56 AE |
3130 | if (result) |
3131 | goto out_err; | |
3132 | ||
3133 | img_request->callback = rbd_img_parent_read_callback; | |
3134 | result = rbd_img_request_submit(img_request); | |
3135 | if (result) | |
3136 | goto out_err; | |
3137 | ||
3138 | return; | |
3139 | out_err: | |
3140 | if (img_request) | |
3141 | rbd_img_request_put(img_request); | |
3142 | obj_request->result = result; | |
3143 | obj_request->xferred = 0; | |
3144 | obj_request_done_set(obj_request); | |
3145 | } | |
bf0d5f50 | 3146 | |
ed95b21a ID |
3147 | static const struct rbd_client_id rbd_empty_cid; |
3148 | ||
3149 | static bool rbd_cid_equal(const struct rbd_client_id *lhs, | |
3150 | const struct rbd_client_id *rhs) | |
3151 | { | |
3152 | return lhs->gid == rhs->gid && lhs->handle == rhs->handle; | |
3153 | } | |
3154 | ||
3155 | static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev) | |
3156 | { | |
3157 | struct rbd_client_id cid; | |
3158 | ||
3159 | mutex_lock(&rbd_dev->watch_mutex); | |
3160 | cid.gid = ceph_client_gid(rbd_dev->rbd_client->client); | |
3161 | cid.handle = rbd_dev->watch_cookie; | |
3162 | mutex_unlock(&rbd_dev->watch_mutex); | |
3163 | return cid; | |
3164 | } | |
3165 | ||
3166 | /* | |
3167 | * lock_rwsem must be held for write | |
3168 | */ | |
3169 | static void rbd_set_owner_cid(struct rbd_device *rbd_dev, | |
3170 | const struct rbd_client_id *cid) | |
3171 | { | |
3172 | dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev, | |
3173 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle, | |
3174 | cid->gid, cid->handle); | |
3175 | rbd_dev->owner_cid = *cid; /* struct */ | |
3176 | } | |
3177 | ||
3178 | static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf) | |
3179 | { | |
3180 | mutex_lock(&rbd_dev->watch_mutex); | |
3181 | sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie); | |
3182 | mutex_unlock(&rbd_dev->watch_mutex); | |
3183 | } | |
3184 | ||
3185 | /* | |
3186 | * lock_rwsem must be held for write | |
3187 | */ | |
3188 | static int rbd_lock(struct rbd_device *rbd_dev) | |
3189 | { | |
3190 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3191 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); | |
3192 | char cookie[32]; | |
3193 | int ret; | |
3194 | ||
3195 | WARN_ON(__rbd_is_lock_owner(rbd_dev)); | |
3196 | ||
3197 | format_lock_cookie(rbd_dev, cookie); | |
3198 | ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
3199 | RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie, | |
3200 | RBD_LOCK_TAG, "", 0); | |
3201 | if (ret) | |
3202 | return ret; | |
3203 | ||
3204 | rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED; | |
3205 | rbd_set_owner_cid(rbd_dev, &cid); | |
3206 | queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work); | |
3207 | return 0; | |
3208 | } | |
3209 | ||
3210 | /* | |
3211 | * lock_rwsem must be held for write | |
3212 | */ | |
3213 | static int rbd_unlock(struct rbd_device *rbd_dev) | |
b8d70035 | 3214 | { |
922dab61 | 3215 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
ed95b21a | 3216 | char cookie[32]; |
e627db08 | 3217 | int ret; |
b8d70035 | 3218 | |
ed95b21a ID |
3219 | WARN_ON(!__rbd_is_lock_owner(rbd_dev)); |
3220 | ||
3221 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; | |
3222 | ||
3223 | format_lock_cookie(rbd_dev, cookie); | |
3224 | ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc, | |
3225 | RBD_LOCK_NAME, cookie); | |
3226 | if (ret && ret != -ENOENT) { | |
3227 | rbd_warn(rbd_dev, "cls_unlock failed: %d", ret); | |
3228 | return ret; | |
3229 | } | |
3230 | ||
3231 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
3232 | queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work); | |
3233 | return 0; | |
3234 | } | |
3235 | ||
3236 | static int __rbd_notify_op_lock(struct rbd_device *rbd_dev, | |
3237 | enum rbd_notify_op notify_op, | |
3238 | struct page ***preply_pages, | |
3239 | size_t *preply_len) | |
3240 | { | |
3241 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3242 | struct rbd_client_id cid = rbd_get_cid(rbd_dev); | |
3243 | int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN; | |
3244 | char buf[buf_size]; | |
3245 | void *p = buf; | |
3246 | ||
3247 | dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op); | |
3248 | ||
3249 | /* encode *LockPayload NotifyMessage (op + ClientId) */ | |
3250 | ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN); | |
3251 | ceph_encode_32(&p, notify_op); | |
3252 | ceph_encode_64(&p, cid.gid); | |
3253 | ceph_encode_64(&p, cid.handle); | |
3254 | ||
3255 | return ceph_osdc_notify(osdc, &rbd_dev->header_oid, | |
3256 | &rbd_dev->header_oloc, buf, buf_size, | |
3257 | RBD_NOTIFY_TIMEOUT, preply_pages, preply_len); | |
3258 | } | |
3259 | ||
3260 | static void rbd_notify_op_lock(struct rbd_device *rbd_dev, | |
3261 | enum rbd_notify_op notify_op) | |
3262 | { | |
3263 | struct page **reply_pages; | |
3264 | size_t reply_len; | |
3265 | ||
3266 | __rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len); | |
3267 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3268 | } | |
3269 | ||
3270 | static void rbd_notify_acquired_lock(struct work_struct *work) | |
3271 | { | |
3272 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3273 | acquired_lock_work); | |
3274 | ||
3275 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK); | |
3276 | } | |
3277 | ||
3278 | static void rbd_notify_released_lock(struct work_struct *work) | |
3279 | { | |
3280 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3281 | released_lock_work); | |
3282 | ||
3283 | rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK); | |
3284 | } | |
3285 | ||
3286 | static int rbd_request_lock(struct rbd_device *rbd_dev) | |
3287 | { | |
3288 | struct page **reply_pages; | |
3289 | size_t reply_len; | |
3290 | bool lock_owner_responded = false; | |
3291 | int ret; | |
52bb1f9b | 3292 | |
ed95b21a ID |
3293 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
3294 | ||
3295 | ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK, | |
3296 | &reply_pages, &reply_len); | |
3297 | if (ret && ret != -ETIMEDOUT) { | |
3298 | rbd_warn(rbd_dev, "failed to request lock: %d", ret); | |
3299 | goto out; | |
3300 | } | |
3301 | ||
3302 | if (reply_len > 0 && reply_len <= PAGE_SIZE) { | |
3303 | void *p = page_address(reply_pages[0]); | |
3304 | void *const end = p + reply_len; | |
3305 | u32 n; | |
3306 | ||
3307 | ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */ | |
3308 | while (n--) { | |
3309 | u8 struct_v; | |
3310 | u32 len; | |
3311 | ||
3312 | ceph_decode_need(&p, end, 8 + 8, e_inval); | |
3313 | p += 8 + 8; /* skip gid and cookie */ | |
3314 | ||
3315 | ceph_decode_32_safe(&p, end, len, e_inval); | |
3316 | if (!len) | |
3317 | continue; | |
3318 | ||
3319 | if (lock_owner_responded) { | |
3320 | rbd_warn(rbd_dev, | |
3321 | "duplicate lock owners detected"); | |
3322 | ret = -EIO; | |
3323 | goto out; | |
3324 | } | |
3325 | ||
3326 | lock_owner_responded = true; | |
3327 | ret = ceph_start_decoding(&p, end, 1, "ResponseMessage", | |
3328 | &struct_v, &len); | |
3329 | if (ret) { | |
3330 | rbd_warn(rbd_dev, | |
3331 | "failed to decode ResponseMessage: %d", | |
3332 | ret); | |
3333 | goto e_inval; | |
3334 | } | |
3335 | ||
3336 | ret = ceph_decode_32(&p); | |
3337 | } | |
3338 | } | |
3339 | ||
3340 | if (!lock_owner_responded) { | |
3341 | rbd_warn(rbd_dev, "no lock owners detected"); | |
3342 | ret = -ETIMEDOUT; | |
3343 | } | |
3344 | ||
3345 | out: | |
3346 | ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)); | |
3347 | return ret; | |
3348 | ||
3349 | e_inval: | |
3350 | ret = -EINVAL; | |
3351 | goto out; | |
3352 | } | |
3353 | ||
3354 | static void wake_requests(struct rbd_device *rbd_dev, bool wake_all) | |
3355 | { | |
3356 | dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all); | |
3357 | ||
3358 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
3359 | if (wake_all) | |
3360 | wake_up_all(&rbd_dev->lock_waitq); | |
3361 | else | |
3362 | wake_up(&rbd_dev->lock_waitq); | |
3363 | } | |
3364 | ||
3365 | static int get_lock_owner_info(struct rbd_device *rbd_dev, | |
3366 | struct ceph_locker **lockers, u32 *num_lockers) | |
3367 | { | |
3368 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3369 | u8 lock_type; | |
3370 | char *lock_tag; | |
3371 | int ret; | |
3372 | ||
3373 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3374 | ||
3375 | ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid, | |
3376 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
3377 | &lock_type, &lock_tag, lockers, num_lockers); | |
3378 | if (ret) | |
3379 | return ret; | |
3380 | ||
3381 | if (*num_lockers == 0) { | |
3382 | dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev); | |
3383 | goto out; | |
3384 | } | |
3385 | ||
3386 | if (strcmp(lock_tag, RBD_LOCK_TAG)) { | |
3387 | rbd_warn(rbd_dev, "locked by external mechanism, tag %s", | |
3388 | lock_tag); | |
3389 | ret = -EBUSY; | |
3390 | goto out; | |
3391 | } | |
3392 | ||
3393 | if (lock_type == CEPH_CLS_LOCK_SHARED) { | |
3394 | rbd_warn(rbd_dev, "shared lock type detected"); | |
3395 | ret = -EBUSY; | |
3396 | goto out; | |
3397 | } | |
3398 | ||
3399 | if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX, | |
3400 | strlen(RBD_LOCK_COOKIE_PREFIX))) { | |
3401 | rbd_warn(rbd_dev, "locked by external mechanism, cookie %s", | |
3402 | (*lockers)[0].id.cookie); | |
3403 | ret = -EBUSY; | |
3404 | goto out; | |
3405 | } | |
3406 | ||
3407 | out: | |
3408 | kfree(lock_tag); | |
3409 | return ret; | |
3410 | } | |
3411 | ||
3412 | static int find_watcher(struct rbd_device *rbd_dev, | |
3413 | const struct ceph_locker *locker) | |
3414 | { | |
3415 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3416 | struct ceph_watch_item *watchers; | |
3417 | u32 num_watchers; | |
3418 | u64 cookie; | |
3419 | int i; | |
3420 | int ret; | |
3421 | ||
3422 | ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid, | |
3423 | &rbd_dev->header_oloc, &watchers, | |
3424 | &num_watchers); | |
3425 | if (ret) | |
3426 | return ret; | |
3427 | ||
3428 | sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie); | |
3429 | for (i = 0; i < num_watchers; i++) { | |
3430 | if (!memcmp(&watchers[i].addr, &locker->info.addr, | |
3431 | sizeof(locker->info.addr)) && | |
3432 | watchers[i].cookie == cookie) { | |
3433 | struct rbd_client_id cid = { | |
3434 | .gid = le64_to_cpu(watchers[i].name.num), | |
3435 | .handle = cookie, | |
3436 | }; | |
3437 | ||
3438 | dout("%s rbd_dev %p found cid %llu-%llu\n", __func__, | |
3439 | rbd_dev, cid.gid, cid.handle); | |
3440 | rbd_set_owner_cid(rbd_dev, &cid); | |
3441 | ret = 1; | |
3442 | goto out; | |
3443 | } | |
3444 | } | |
3445 | ||
3446 | dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev); | |
3447 | ret = 0; | |
3448 | out: | |
3449 | kfree(watchers); | |
3450 | return ret; | |
3451 | } | |
3452 | ||
3453 | /* | |
3454 | * lock_rwsem must be held for write | |
3455 | */ | |
3456 | static int rbd_try_lock(struct rbd_device *rbd_dev) | |
3457 | { | |
3458 | struct ceph_client *client = rbd_dev->rbd_client->client; | |
3459 | struct ceph_locker *lockers; | |
3460 | u32 num_lockers; | |
3461 | int ret; | |
3462 | ||
3463 | for (;;) { | |
3464 | ret = rbd_lock(rbd_dev); | |
3465 | if (ret != -EBUSY) | |
3466 | return ret; | |
3467 | ||
3468 | /* determine if the current lock holder is still alive */ | |
3469 | ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers); | |
3470 | if (ret) | |
3471 | return ret; | |
3472 | ||
3473 | if (num_lockers == 0) | |
3474 | goto again; | |
3475 | ||
3476 | ret = find_watcher(rbd_dev, lockers); | |
3477 | if (ret) { | |
3478 | if (ret > 0) | |
3479 | ret = 0; /* have to request lock */ | |
3480 | goto out; | |
3481 | } | |
3482 | ||
3483 | rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock", | |
3484 | ENTITY_NAME(lockers[0].id.name)); | |
3485 | ||
3486 | ret = ceph_monc_blacklist_add(&client->monc, | |
3487 | &lockers[0].info.addr); | |
3488 | if (ret) { | |
3489 | rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d", | |
3490 | ENTITY_NAME(lockers[0].id.name), ret); | |
3491 | goto out; | |
3492 | } | |
3493 | ||
3494 | ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid, | |
3495 | &rbd_dev->header_oloc, RBD_LOCK_NAME, | |
3496 | lockers[0].id.cookie, | |
3497 | &lockers[0].id.name); | |
3498 | if (ret && ret != -ENOENT) | |
3499 | goto out; | |
3500 | ||
3501 | again: | |
3502 | ceph_free_lockers(lockers, num_lockers); | |
3503 | } | |
3504 | ||
3505 | out: | |
3506 | ceph_free_lockers(lockers, num_lockers); | |
3507 | return ret; | |
3508 | } | |
3509 | ||
3510 | /* | |
3511 | * ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED | |
3512 | */ | |
3513 | static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev, | |
3514 | int *pret) | |
3515 | { | |
3516 | enum rbd_lock_state lock_state; | |
3517 | ||
3518 | down_read(&rbd_dev->lock_rwsem); | |
3519 | dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev, | |
3520 | rbd_dev->lock_state); | |
3521 | if (__rbd_is_lock_owner(rbd_dev)) { | |
3522 | lock_state = rbd_dev->lock_state; | |
3523 | up_read(&rbd_dev->lock_rwsem); | |
3524 | return lock_state; | |
3525 | } | |
3526 | ||
3527 | up_read(&rbd_dev->lock_rwsem); | |
3528 | down_write(&rbd_dev->lock_rwsem); | |
3529 | dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev, | |
3530 | rbd_dev->lock_state); | |
3531 | if (!__rbd_is_lock_owner(rbd_dev)) { | |
3532 | *pret = rbd_try_lock(rbd_dev); | |
3533 | if (*pret) | |
3534 | rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret); | |
3535 | } | |
3536 | ||
3537 | lock_state = rbd_dev->lock_state; | |
3538 | up_write(&rbd_dev->lock_rwsem); | |
3539 | return lock_state; | |
3540 | } | |
3541 | ||
3542 | static void rbd_acquire_lock(struct work_struct *work) | |
3543 | { | |
3544 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
3545 | struct rbd_device, lock_dwork); | |
3546 | enum rbd_lock_state lock_state; | |
3547 | int ret; | |
3548 | ||
3549 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3550 | again: | |
3551 | lock_state = rbd_try_acquire_lock(rbd_dev, &ret); | |
3552 | if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) { | |
3553 | if (lock_state == RBD_LOCK_STATE_LOCKED) | |
3554 | wake_requests(rbd_dev, true); | |
3555 | dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__, | |
3556 | rbd_dev, lock_state, ret); | |
3557 | return; | |
3558 | } | |
3559 | ||
3560 | ret = rbd_request_lock(rbd_dev); | |
3561 | if (ret == -ETIMEDOUT) { | |
3562 | goto again; /* treat this as a dead client */ | |
3563 | } else if (ret < 0) { | |
3564 | rbd_warn(rbd_dev, "error requesting lock: %d", ret); | |
3565 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
3566 | RBD_RETRY_DELAY); | |
3567 | } else { | |
3568 | /* | |
3569 | * lock owner acked, but resend if we don't see them | |
3570 | * release the lock | |
3571 | */ | |
3572 | dout("%s rbd_dev %p requeueing lock_dwork\n", __func__, | |
3573 | rbd_dev); | |
3574 | mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, | |
3575 | msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC)); | |
3576 | } | |
3577 | } | |
3578 | ||
3579 | /* | |
3580 | * lock_rwsem must be held for write | |
3581 | */ | |
3582 | static bool rbd_release_lock(struct rbd_device *rbd_dev) | |
3583 | { | |
3584 | dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev, | |
3585 | rbd_dev->lock_state); | |
3586 | if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED) | |
3587 | return false; | |
3588 | ||
3589 | rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING; | |
3590 | downgrade_write(&rbd_dev->lock_rwsem); | |
52bb1f9b | 3591 | /* |
ed95b21a | 3592 | * Ensure that all in-flight IO is flushed. |
52bb1f9b | 3593 | * |
ed95b21a ID |
3594 | * FIXME: ceph_osdc_sync() flushes the entire OSD client, which |
3595 | * may be shared with other devices. | |
52bb1f9b | 3596 | */ |
ed95b21a ID |
3597 | ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc); |
3598 | up_read(&rbd_dev->lock_rwsem); | |
3599 | ||
3600 | down_write(&rbd_dev->lock_rwsem); | |
3601 | dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev, | |
3602 | rbd_dev->lock_state); | |
3603 | if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING) | |
3604 | return false; | |
3605 | ||
3606 | if (!rbd_unlock(rbd_dev)) | |
3607 | /* | |
3608 | * Give others a chance to grab the lock - we would re-acquire | |
3609 | * almost immediately if we got new IO during ceph_osdc_sync() | |
3610 | * otherwise. We need to ack our own notifications, so this | |
3611 | * lock_dwork will be requeued from rbd_wait_state_locked() | |
3612 | * after wake_requests() in rbd_handle_released_lock(). | |
3613 | */ | |
3614 | cancel_delayed_work(&rbd_dev->lock_dwork); | |
3615 | ||
3616 | return true; | |
3617 | } | |
3618 | ||
3619 | static void rbd_release_lock_work(struct work_struct *work) | |
3620 | { | |
3621 | struct rbd_device *rbd_dev = container_of(work, struct rbd_device, | |
3622 | unlock_work); | |
3623 | ||
3624 | down_write(&rbd_dev->lock_rwsem); | |
3625 | rbd_release_lock(rbd_dev); | |
3626 | up_write(&rbd_dev->lock_rwsem); | |
3627 | } | |
3628 | ||
3629 | static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3630 | void **p) | |
3631 | { | |
3632 | struct rbd_client_id cid = { 0 }; | |
3633 | ||
3634 | if (struct_v >= 2) { | |
3635 | cid.gid = ceph_decode_64(p); | |
3636 | cid.handle = ceph_decode_64(p); | |
3637 | } | |
3638 | ||
3639 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3640 | cid.handle); | |
3641 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
3642 | down_write(&rbd_dev->lock_rwsem); | |
3643 | if (rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
3644 | /* | |
3645 | * we already know that the remote client is | |
3646 | * the owner | |
3647 | */ | |
3648 | up_write(&rbd_dev->lock_rwsem); | |
3649 | return; | |
3650 | } | |
3651 | ||
3652 | rbd_set_owner_cid(rbd_dev, &cid); | |
3653 | downgrade_write(&rbd_dev->lock_rwsem); | |
3654 | } else { | |
3655 | down_read(&rbd_dev->lock_rwsem); | |
3656 | } | |
3657 | ||
3658 | if (!__rbd_is_lock_owner(rbd_dev)) | |
3659 | wake_requests(rbd_dev, false); | |
3660 | up_read(&rbd_dev->lock_rwsem); | |
3661 | } | |
3662 | ||
3663 | static void rbd_handle_released_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3664 | void **p) | |
3665 | { | |
3666 | struct rbd_client_id cid = { 0 }; | |
3667 | ||
3668 | if (struct_v >= 2) { | |
3669 | cid.gid = ceph_decode_64(p); | |
3670 | cid.handle = ceph_decode_64(p); | |
3671 | } | |
3672 | ||
3673 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3674 | cid.handle); | |
3675 | if (!rbd_cid_equal(&cid, &rbd_empty_cid)) { | |
3676 | down_write(&rbd_dev->lock_rwsem); | |
3677 | if (!rbd_cid_equal(&cid, &rbd_dev->owner_cid)) { | |
3678 | dout("%s rbd_dev %p unexpected owner, cid %llu-%llu != owner_cid %llu-%llu\n", | |
3679 | __func__, rbd_dev, cid.gid, cid.handle, | |
3680 | rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle); | |
3681 | up_write(&rbd_dev->lock_rwsem); | |
3682 | return; | |
3683 | } | |
3684 | ||
3685 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
3686 | downgrade_write(&rbd_dev->lock_rwsem); | |
3687 | } else { | |
3688 | down_read(&rbd_dev->lock_rwsem); | |
3689 | } | |
3690 | ||
3691 | if (!__rbd_is_lock_owner(rbd_dev)) | |
3692 | wake_requests(rbd_dev, false); | |
3693 | up_read(&rbd_dev->lock_rwsem); | |
3694 | } | |
3695 | ||
3696 | static bool rbd_handle_request_lock(struct rbd_device *rbd_dev, u8 struct_v, | |
3697 | void **p) | |
3698 | { | |
3699 | struct rbd_client_id my_cid = rbd_get_cid(rbd_dev); | |
3700 | struct rbd_client_id cid = { 0 }; | |
3701 | bool need_to_send; | |
3702 | ||
3703 | if (struct_v >= 2) { | |
3704 | cid.gid = ceph_decode_64(p); | |
3705 | cid.handle = ceph_decode_64(p); | |
3706 | } | |
3707 | ||
3708 | dout("%s rbd_dev %p cid %llu-%llu\n", __func__, rbd_dev, cid.gid, | |
3709 | cid.handle); | |
3710 | if (rbd_cid_equal(&cid, &my_cid)) | |
3711 | return false; | |
3712 | ||
3713 | down_read(&rbd_dev->lock_rwsem); | |
3714 | need_to_send = __rbd_is_lock_owner(rbd_dev); | |
3715 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) { | |
3716 | if (!rbd_cid_equal(&rbd_dev->owner_cid, &rbd_empty_cid)) { | |
3717 | dout("%s rbd_dev %p queueing unlock_work\n", __func__, | |
3718 | rbd_dev); | |
3719 | queue_work(rbd_dev->task_wq, &rbd_dev->unlock_work); | |
3720 | } | |
3721 | } | |
3722 | up_read(&rbd_dev->lock_rwsem); | |
3723 | return need_to_send; | |
3724 | } | |
3725 | ||
3726 | static void __rbd_acknowledge_notify(struct rbd_device *rbd_dev, | |
3727 | u64 notify_id, u64 cookie, s32 *result) | |
3728 | { | |
3729 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3730 | int buf_size = 4 + CEPH_ENCODING_START_BLK_LEN; | |
3731 | char buf[buf_size]; | |
3732 | int ret; | |
3733 | ||
3734 | if (result) { | |
3735 | void *p = buf; | |
3736 | ||
3737 | /* encode ResponseMessage */ | |
3738 | ceph_start_encoding(&p, 1, 1, | |
3739 | buf_size - CEPH_ENCODING_START_BLK_LEN); | |
3740 | ceph_encode_32(&p, *result); | |
3741 | } else { | |
3742 | buf_size = 0; | |
3743 | } | |
b8d70035 | 3744 | |
922dab61 ID |
3745 | ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid, |
3746 | &rbd_dev->header_oloc, notify_id, cookie, | |
ed95b21a | 3747 | buf, buf_size); |
52bb1f9b | 3748 | if (ret) |
ed95b21a ID |
3749 | rbd_warn(rbd_dev, "acknowledge_notify failed: %d", ret); |
3750 | } | |
3751 | ||
3752 | static void rbd_acknowledge_notify(struct rbd_device *rbd_dev, u64 notify_id, | |
3753 | u64 cookie) | |
3754 | { | |
3755 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3756 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, NULL); | |
3757 | } | |
3758 | ||
3759 | static void rbd_acknowledge_notify_result(struct rbd_device *rbd_dev, | |
3760 | u64 notify_id, u64 cookie, s32 result) | |
3761 | { | |
3762 | dout("%s rbd_dev %p result %d\n", __func__, rbd_dev, result); | |
3763 | __rbd_acknowledge_notify(rbd_dev, notify_id, cookie, &result); | |
3764 | } | |
3765 | ||
3766 | static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie, | |
3767 | u64 notifier_id, void *data, size_t data_len) | |
3768 | { | |
3769 | struct rbd_device *rbd_dev = arg; | |
3770 | void *p = data; | |
3771 | void *const end = p + data_len; | |
3772 | u8 struct_v; | |
3773 | u32 len; | |
3774 | u32 notify_op; | |
3775 | int ret; | |
3776 | ||
3777 | dout("%s rbd_dev %p cookie %llu notify_id %llu data_len %zu\n", | |
3778 | __func__, rbd_dev, cookie, notify_id, data_len); | |
3779 | if (data_len) { | |
3780 | ret = ceph_start_decoding(&p, end, 1, "NotifyMessage", | |
3781 | &struct_v, &len); | |
3782 | if (ret) { | |
3783 | rbd_warn(rbd_dev, "failed to decode NotifyMessage: %d", | |
3784 | ret); | |
3785 | return; | |
3786 | } | |
3787 | ||
3788 | notify_op = ceph_decode_32(&p); | |
3789 | } else { | |
3790 | /* legacy notification for header updates */ | |
3791 | notify_op = RBD_NOTIFY_OP_HEADER_UPDATE; | |
3792 | len = 0; | |
3793 | } | |
3794 | ||
3795 | dout("%s rbd_dev %p notify_op %u\n", __func__, rbd_dev, notify_op); | |
3796 | switch (notify_op) { | |
3797 | case RBD_NOTIFY_OP_ACQUIRED_LOCK: | |
3798 | rbd_handle_acquired_lock(rbd_dev, struct_v, &p); | |
3799 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3800 | break; | |
3801 | case RBD_NOTIFY_OP_RELEASED_LOCK: | |
3802 | rbd_handle_released_lock(rbd_dev, struct_v, &p); | |
3803 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3804 | break; | |
3805 | case RBD_NOTIFY_OP_REQUEST_LOCK: | |
3806 | if (rbd_handle_request_lock(rbd_dev, struct_v, &p)) | |
3807 | /* | |
3808 | * send ResponseMessage(0) back so the client | |
3809 | * can detect a missing owner | |
3810 | */ | |
3811 | rbd_acknowledge_notify_result(rbd_dev, notify_id, | |
3812 | cookie, 0); | |
3813 | else | |
3814 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3815 | break; | |
3816 | case RBD_NOTIFY_OP_HEADER_UPDATE: | |
3817 | ret = rbd_dev_refresh(rbd_dev); | |
3818 | if (ret) | |
3819 | rbd_warn(rbd_dev, "refresh failed: %d", ret); | |
3820 | ||
3821 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3822 | break; | |
3823 | default: | |
3824 | if (rbd_is_lock_owner(rbd_dev)) | |
3825 | rbd_acknowledge_notify_result(rbd_dev, notify_id, | |
3826 | cookie, -EOPNOTSUPP); | |
3827 | else | |
3828 | rbd_acknowledge_notify(rbd_dev, notify_id, cookie); | |
3829 | break; | |
3830 | } | |
b8d70035 AE |
3831 | } |
3832 | ||
99d16943 ID |
3833 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev); |
3834 | ||
922dab61 | 3835 | static void rbd_watch_errcb(void *arg, u64 cookie, int err) |
bb040aa0 | 3836 | { |
922dab61 | 3837 | struct rbd_device *rbd_dev = arg; |
bb040aa0 | 3838 | |
922dab61 | 3839 | rbd_warn(rbd_dev, "encountered watch error: %d", err); |
bb040aa0 | 3840 | |
ed95b21a ID |
3841 | down_write(&rbd_dev->lock_rwsem); |
3842 | rbd_set_owner_cid(rbd_dev, &rbd_empty_cid); | |
3843 | up_write(&rbd_dev->lock_rwsem); | |
3844 | ||
99d16943 ID |
3845 | mutex_lock(&rbd_dev->watch_mutex); |
3846 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) { | |
3847 | __rbd_unregister_watch(rbd_dev); | |
3848 | rbd_dev->watch_state = RBD_WATCH_STATE_ERROR; | |
bb040aa0 | 3849 | |
99d16943 | 3850 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->watch_dwork, 0); |
bb040aa0 | 3851 | } |
99d16943 | 3852 | mutex_unlock(&rbd_dev->watch_mutex); |
bb040aa0 ID |
3853 | } |
3854 | ||
9969ebc5 | 3855 | /* |
99d16943 | 3856 | * watch_mutex must be locked |
9969ebc5 | 3857 | */ |
99d16943 | 3858 | static int __rbd_register_watch(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3859 | { |
3860 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
922dab61 | 3861 | struct ceph_osd_linger_request *handle; |
9969ebc5 | 3862 | |
922dab61 | 3863 | rbd_assert(!rbd_dev->watch_handle); |
99d16943 | 3864 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
9969ebc5 | 3865 | |
922dab61 ID |
3866 | handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid, |
3867 | &rbd_dev->header_oloc, rbd_watch_cb, | |
3868 | rbd_watch_errcb, rbd_dev); | |
3869 | if (IS_ERR(handle)) | |
3870 | return PTR_ERR(handle); | |
8eb87565 | 3871 | |
922dab61 | 3872 | rbd_dev->watch_handle = handle; |
b30a01f2 | 3873 | return 0; |
b30a01f2 ID |
3874 | } |
3875 | ||
99d16943 ID |
3876 | /* |
3877 | * watch_mutex must be locked | |
3878 | */ | |
3879 | static void __rbd_unregister_watch(struct rbd_device *rbd_dev) | |
b30a01f2 | 3880 | { |
922dab61 ID |
3881 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3882 | int ret; | |
b30a01f2 | 3883 | |
99d16943 ID |
3884 | rbd_assert(rbd_dev->watch_handle); |
3885 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
b30a01f2 | 3886 | |
922dab61 ID |
3887 | ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle); |
3888 | if (ret) | |
3889 | rbd_warn(rbd_dev, "failed to unwatch: %d", ret); | |
76756a51 | 3890 | |
922dab61 | 3891 | rbd_dev->watch_handle = NULL; |
c525f036 ID |
3892 | } |
3893 | ||
99d16943 ID |
3894 | static int rbd_register_watch(struct rbd_device *rbd_dev) |
3895 | { | |
3896 | int ret; | |
3897 | ||
3898 | mutex_lock(&rbd_dev->watch_mutex); | |
3899 | rbd_assert(rbd_dev->watch_state == RBD_WATCH_STATE_UNREGISTERED); | |
3900 | ret = __rbd_register_watch(rbd_dev); | |
3901 | if (ret) | |
3902 | goto out; | |
3903 | ||
3904 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; | |
3905 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
3906 | ||
3907 | out: | |
3908 | mutex_unlock(&rbd_dev->watch_mutex); | |
3909 | return ret; | |
3910 | } | |
3911 | ||
3912 | static void cancel_tasks_sync(struct rbd_device *rbd_dev) | |
c525f036 | 3913 | { |
99d16943 ID |
3914 | dout("%s rbd_dev %p\n", __func__, rbd_dev); |
3915 | ||
3916 | cancel_delayed_work_sync(&rbd_dev->watch_dwork); | |
ed95b21a ID |
3917 | cancel_work_sync(&rbd_dev->acquired_lock_work); |
3918 | cancel_work_sync(&rbd_dev->released_lock_work); | |
3919 | cancel_delayed_work_sync(&rbd_dev->lock_dwork); | |
3920 | cancel_work_sync(&rbd_dev->unlock_work); | |
99d16943 ID |
3921 | } |
3922 | ||
3923 | static void rbd_unregister_watch(struct rbd_device *rbd_dev) | |
3924 | { | |
ed95b21a | 3925 | WARN_ON(waitqueue_active(&rbd_dev->lock_waitq)); |
99d16943 ID |
3926 | cancel_tasks_sync(rbd_dev); |
3927 | ||
3928 | mutex_lock(&rbd_dev->watch_mutex); | |
3929 | if (rbd_dev->watch_state == RBD_WATCH_STATE_REGISTERED) | |
3930 | __rbd_unregister_watch(rbd_dev); | |
3931 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
3932 | mutex_unlock(&rbd_dev->watch_mutex); | |
811c6688 | 3933 | |
811c6688 | 3934 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); |
fca27065 ID |
3935 | } |
3936 | ||
99d16943 ID |
3937 | static void rbd_reregister_watch(struct work_struct *work) |
3938 | { | |
3939 | struct rbd_device *rbd_dev = container_of(to_delayed_work(work), | |
3940 | struct rbd_device, watch_dwork); | |
ed95b21a | 3941 | bool was_lock_owner = false; |
99d16943 ID |
3942 | int ret; |
3943 | ||
3944 | dout("%s rbd_dev %p\n", __func__, rbd_dev); | |
3945 | ||
ed95b21a ID |
3946 | down_write(&rbd_dev->lock_rwsem); |
3947 | if (rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED) | |
3948 | was_lock_owner = rbd_release_lock(rbd_dev); | |
3949 | ||
99d16943 ID |
3950 | mutex_lock(&rbd_dev->watch_mutex); |
3951 | if (rbd_dev->watch_state != RBD_WATCH_STATE_ERROR) | |
3952 | goto fail_unlock; | |
3953 | ||
3954 | ret = __rbd_register_watch(rbd_dev); | |
3955 | if (ret) { | |
3956 | rbd_warn(rbd_dev, "failed to reregister watch: %d", ret); | |
3957 | if (ret != -EBLACKLISTED) | |
3958 | queue_delayed_work(rbd_dev->task_wq, | |
3959 | &rbd_dev->watch_dwork, | |
3960 | RBD_RETRY_DELAY); | |
3961 | goto fail_unlock; | |
3962 | } | |
3963 | ||
3964 | rbd_dev->watch_state = RBD_WATCH_STATE_REGISTERED; | |
3965 | rbd_dev->watch_cookie = rbd_dev->watch_handle->linger_id; | |
3966 | mutex_unlock(&rbd_dev->watch_mutex); | |
3967 | ||
3968 | ret = rbd_dev_refresh(rbd_dev); | |
3969 | if (ret) | |
3970 | rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret); | |
3971 | ||
ed95b21a ID |
3972 | if (was_lock_owner) { |
3973 | ret = rbd_try_lock(rbd_dev); | |
3974 | if (ret) | |
3975 | rbd_warn(rbd_dev, "reregisteration lock failed: %d", | |
3976 | ret); | |
3977 | } | |
3978 | ||
3979 | up_write(&rbd_dev->lock_rwsem); | |
3980 | wake_requests(rbd_dev, true); | |
99d16943 ID |
3981 | return; |
3982 | ||
3983 | fail_unlock: | |
3984 | mutex_unlock(&rbd_dev->watch_mutex); | |
ed95b21a | 3985 | up_write(&rbd_dev->lock_rwsem); |
99d16943 ID |
3986 | } |
3987 | ||
36be9a76 | 3988 | /* |
f40eb349 AE |
3989 | * Synchronous osd object method call. Returns the number of bytes |
3990 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3991 | */ |
3992 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3993 | const char *object_name, | |
3994 | const char *class_name, | |
3995 | const char *method_name, | |
4157976b | 3996 | const void *outbound, |
36be9a76 | 3997 | size_t outbound_size, |
4157976b | 3998 | void *inbound, |
e2a58ee5 | 3999 | size_t inbound_size) |
36be9a76 | 4000 | { |
2169238d | 4001 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 4002 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
4003 | struct page **pages; |
4004 | u32 page_count; | |
4005 | int ret; | |
4006 | ||
4007 | /* | |
6010a451 AE |
4008 | * Method calls are ultimately read operations. The result |
4009 | * should placed into the inbound buffer provided. They | |
4010 | * also supply outbound data--parameters for the object | |
4011 | * method. Currently if this is present it will be a | |
4012 | * snapshot id. | |
36be9a76 | 4013 | */ |
57385b51 | 4014 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
4015 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
4016 | if (IS_ERR(pages)) | |
4017 | return PTR_ERR(pages); | |
4018 | ||
4019 | ret = -ENOMEM; | |
6010a451 | 4020 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
4021 | OBJ_REQUEST_PAGES); |
4022 | if (!obj_request) | |
4023 | goto out; | |
4024 | ||
4025 | obj_request->pages = pages; | |
4026 | obj_request->page_count = page_count; | |
4027 | ||
6d2940c8 | 4028 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 4029 | obj_request); |
36be9a76 AE |
4030 | if (!obj_request->osd_req) |
4031 | goto out; | |
4032 | ||
c99d2d4a | 4033 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
4034 | class_name, method_name); |
4035 | if (outbound_size) { | |
4036 | struct ceph_pagelist *pagelist; | |
4037 | ||
4038 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
4039 | if (!pagelist) | |
4040 | goto out; | |
4041 | ||
4042 | ceph_pagelist_init(pagelist); | |
4043 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
4044 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
4045 | pagelist); | |
4046 | } | |
a4ce40a9 AE |
4047 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
4048 | obj_request->pages, inbound_size, | |
44cd188d | 4049 | 0, false, false); |
9d4df01f | 4050 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 4051 | |
36be9a76 AE |
4052 | ret = rbd_obj_request_submit(osdc, obj_request); |
4053 | if (ret) | |
4054 | goto out; | |
4055 | ret = rbd_obj_request_wait(obj_request); | |
4056 | if (ret) | |
4057 | goto out; | |
4058 | ||
4059 | ret = obj_request->result; | |
4060 | if (ret < 0) | |
4061 | goto out; | |
57385b51 AE |
4062 | |
4063 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
4064 | ret = (int)obj_request->xferred; | |
903bb32e | 4065 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
4066 | out: |
4067 | if (obj_request) | |
4068 | rbd_obj_request_put(obj_request); | |
4069 | else | |
4070 | ceph_release_page_vector(pages, page_count); | |
4071 | ||
4072 | return ret; | |
4073 | } | |
4074 | ||
ed95b21a ID |
4075 | /* |
4076 | * lock_rwsem must be held for read | |
4077 | */ | |
4078 | static void rbd_wait_state_locked(struct rbd_device *rbd_dev) | |
4079 | { | |
4080 | DEFINE_WAIT(wait); | |
4081 | ||
4082 | do { | |
4083 | /* | |
4084 | * Note the use of mod_delayed_work() in rbd_acquire_lock() | |
4085 | * and cancel_delayed_work() in wake_requests(). | |
4086 | */ | |
4087 | dout("%s rbd_dev %p queueing lock_dwork\n", __func__, rbd_dev); | |
4088 | queue_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0); | |
4089 | prepare_to_wait_exclusive(&rbd_dev->lock_waitq, &wait, | |
4090 | TASK_UNINTERRUPTIBLE); | |
4091 | up_read(&rbd_dev->lock_rwsem); | |
4092 | schedule(); | |
4093 | down_read(&rbd_dev->lock_rwsem); | |
4094 | } while (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED); | |
4095 | finish_wait(&rbd_dev->lock_waitq, &wait); | |
4096 | } | |
4097 | ||
7ad18afa | 4098 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 4099 | { |
7ad18afa CH |
4100 | struct request *rq = blk_mq_rq_from_pdu(work); |
4101 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 4102 | struct rbd_img_request *img_request; |
4e752f0a | 4103 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
4104 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
4105 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 4106 | enum obj_operation_type op_type; |
4e752f0a | 4107 | u64 mapping_size; |
ed95b21a | 4108 | bool must_be_locked = false; |
bf0d5f50 AE |
4109 | int result; |
4110 | ||
7ad18afa CH |
4111 | if (rq->cmd_type != REQ_TYPE_FS) { |
4112 | dout("%s: non-fs request type %d\n", __func__, | |
4113 | (int) rq->cmd_type); | |
4114 | result = -EIO; | |
4115 | goto err; | |
4116 | } | |
4117 | ||
c2df40df | 4118 | if (req_op(rq) == REQ_OP_DISCARD) |
90e98c52 | 4119 | op_type = OBJ_OP_DISCARD; |
c2df40df | 4120 | else if (req_op(rq) == REQ_OP_WRITE) |
6d2940c8 GZ |
4121 | op_type = OBJ_OP_WRITE; |
4122 | else | |
4123 | op_type = OBJ_OP_READ; | |
4124 | ||
bc1ecc65 | 4125 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 4126 | |
bc1ecc65 ID |
4127 | if (!length) { |
4128 | dout("%s: zero-length request\n", __func__); | |
4129 | result = 0; | |
4130 | goto err_rq; | |
4131 | } | |
bf0d5f50 | 4132 | |
6d2940c8 | 4133 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 4134 | |
6d2940c8 | 4135 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
4136 | if (rbd_dev->mapping.read_only) { |
4137 | result = -EROFS; | |
4138 | goto err_rq; | |
4dda41d3 | 4139 | } |
bc1ecc65 ID |
4140 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
4141 | } | |
4dda41d3 | 4142 | |
bc1ecc65 ID |
4143 | /* |
4144 | * Quit early if the mapped snapshot no longer exists. It's | |
4145 | * still possible the snapshot will have disappeared by the | |
4146 | * time our request arrives at the osd, but there's no sense in | |
4147 | * sending it if we already know. | |
4148 | */ | |
4149 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
4150 | dout("request for non-existent snapshot"); | |
4151 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
4152 | result = -ENXIO; | |
4153 | goto err_rq; | |
4154 | } | |
4dda41d3 | 4155 | |
bc1ecc65 ID |
4156 | if (offset && length > U64_MAX - offset + 1) { |
4157 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
4158 | length); | |
4159 | result = -EINVAL; | |
4160 | goto err_rq; /* Shouldn't happen */ | |
4161 | } | |
4dda41d3 | 4162 | |
7ad18afa CH |
4163 | blk_mq_start_request(rq); |
4164 | ||
4e752f0a JD |
4165 | down_read(&rbd_dev->header_rwsem); |
4166 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 4167 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
4168 | snapc = rbd_dev->header.snapc; |
4169 | ceph_get_snap_context(snapc); | |
ed95b21a | 4170 | must_be_locked = rbd_is_lock_supported(rbd_dev); |
4e752f0a JD |
4171 | } |
4172 | up_read(&rbd_dev->header_rwsem); | |
4173 | ||
4174 | if (offset + length > mapping_size) { | |
bc1ecc65 | 4175 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 4176 | length, mapping_size); |
bc1ecc65 ID |
4177 | result = -EIO; |
4178 | goto err_rq; | |
4179 | } | |
bf0d5f50 | 4180 | |
ed95b21a ID |
4181 | if (must_be_locked) { |
4182 | down_read(&rbd_dev->lock_rwsem); | |
4183 | if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED) | |
4184 | rbd_wait_state_locked(rbd_dev); | |
4185 | } | |
4186 | ||
6d2940c8 | 4187 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 4188 | snapc); |
bc1ecc65 ID |
4189 | if (!img_request) { |
4190 | result = -ENOMEM; | |
ed95b21a | 4191 | goto err_unlock; |
bc1ecc65 ID |
4192 | } |
4193 | img_request->rq = rq; | |
70b16db8 | 4194 | snapc = NULL; /* img_request consumes a ref */ |
bf0d5f50 | 4195 | |
90e98c52 GZ |
4196 | if (op_type == OBJ_OP_DISCARD) |
4197 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
4198 | NULL); | |
4199 | else | |
4200 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
4201 | rq->bio); | |
bc1ecc65 ID |
4202 | if (result) |
4203 | goto err_img_request; | |
bf0d5f50 | 4204 | |
bc1ecc65 ID |
4205 | result = rbd_img_request_submit(img_request); |
4206 | if (result) | |
4207 | goto err_img_request; | |
bf0d5f50 | 4208 | |
ed95b21a ID |
4209 | if (must_be_locked) |
4210 | up_read(&rbd_dev->lock_rwsem); | |
bc1ecc65 | 4211 | return; |
bf0d5f50 | 4212 | |
bc1ecc65 ID |
4213 | err_img_request: |
4214 | rbd_img_request_put(img_request); | |
ed95b21a ID |
4215 | err_unlock: |
4216 | if (must_be_locked) | |
4217 | up_read(&rbd_dev->lock_rwsem); | |
bc1ecc65 ID |
4218 | err_rq: |
4219 | if (result) | |
4220 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 4221 | obj_op_name(op_type), length, offset, result); |
e96a650a | 4222 | ceph_put_snap_context(snapc); |
7ad18afa CH |
4223 | err: |
4224 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 4225 | } |
bf0d5f50 | 4226 | |
7ad18afa CH |
4227 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
4228 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 4229 | { |
7ad18afa CH |
4230 | struct request *rq = bd->rq; |
4231 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 4232 | |
7ad18afa CH |
4233 | queue_work(rbd_wq, work); |
4234 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
4235 | } |
4236 | ||
602adf40 YS |
4237 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
4238 | { | |
4239 | struct gendisk *disk = rbd_dev->disk; | |
4240 | ||
4241 | if (!disk) | |
4242 | return; | |
4243 | ||
a0cab924 AE |
4244 | rbd_dev->disk = NULL; |
4245 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 4246 | del_gendisk(disk); |
a0cab924 AE |
4247 | if (disk->queue) |
4248 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 4249 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 4250 | } |
602adf40 YS |
4251 | put_disk(disk); |
4252 | } | |
4253 | ||
788e2df3 AE |
4254 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
4255 | const char *object_name, | |
7097f8df | 4256 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
4257 | |
4258 | { | |
2169238d | 4259 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 4260 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
4261 | struct page **pages = NULL; |
4262 | u32 page_count; | |
1ceae7ef | 4263 | size_t size; |
788e2df3 AE |
4264 | int ret; |
4265 | ||
4266 | page_count = (u32) calc_pages_for(offset, length); | |
4267 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
4268 | if (IS_ERR(pages)) | |
a8d42056 | 4269 | return PTR_ERR(pages); |
788e2df3 AE |
4270 | |
4271 | ret = -ENOMEM; | |
4272 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 4273 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
4274 | if (!obj_request) |
4275 | goto out; | |
4276 | ||
4277 | obj_request->pages = pages; | |
4278 | obj_request->page_count = page_count; | |
4279 | ||
6d2940c8 | 4280 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 4281 | obj_request); |
788e2df3 AE |
4282 | if (!obj_request->osd_req) |
4283 | goto out; | |
4284 | ||
c99d2d4a AE |
4285 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
4286 | offset, length, 0, 0); | |
406e2c9f | 4287 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 4288 | obj_request->pages, |
44cd188d AE |
4289 | obj_request->length, |
4290 | obj_request->offset & ~PAGE_MASK, | |
4291 | false, false); | |
9d4df01f | 4292 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 4293 | |
788e2df3 AE |
4294 | ret = rbd_obj_request_submit(osdc, obj_request); |
4295 | if (ret) | |
4296 | goto out; | |
4297 | ret = rbd_obj_request_wait(obj_request); | |
4298 | if (ret) | |
4299 | goto out; | |
4300 | ||
4301 | ret = obj_request->result; | |
4302 | if (ret < 0) | |
4303 | goto out; | |
1ceae7ef AE |
4304 | |
4305 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
4306 | size = (size_t) obj_request->xferred; | |
903bb32e | 4307 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
4308 | rbd_assert(size <= (size_t)INT_MAX); |
4309 | ret = (int)size; | |
788e2df3 AE |
4310 | out: |
4311 | if (obj_request) | |
4312 | rbd_obj_request_put(obj_request); | |
4313 | else | |
4314 | ceph_release_page_vector(pages, page_count); | |
4315 | ||
4316 | return ret; | |
4317 | } | |
4318 | ||
602adf40 | 4319 | /* |
662518b1 AE |
4320 | * Read the complete header for the given rbd device. On successful |
4321 | * return, the rbd_dev->header field will contain up-to-date | |
4322 | * information about the image. | |
602adf40 | 4323 | */ |
99a41ebc | 4324 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 4325 | { |
4156d998 | 4326 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 4327 | u32 snap_count = 0; |
4156d998 AE |
4328 | u64 names_size = 0; |
4329 | u32 want_count; | |
4330 | int ret; | |
602adf40 | 4331 | |
00f1f36f | 4332 | /* |
4156d998 AE |
4333 | * The complete header will include an array of its 64-bit |
4334 | * snapshot ids, followed by the names of those snapshots as | |
4335 | * a contiguous block of NUL-terminated strings. Note that | |
4336 | * the number of snapshots could change by the time we read | |
4337 | * it in, in which case we re-read it. | |
00f1f36f | 4338 | */ |
4156d998 AE |
4339 | do { |
4340 | size_t size; | |
4341 | ||
4342 | kfree(ondisk); | |
4343 | ||
4344 | size = sizeof (*ondisk); | |
4345 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
4346 | size += names_size; | |
4347 | ondisk = kmalloc(size, GFP_KERNEL); | |
4348 | if (!ondisk) | |
662518b1 | 4349 | return -ENOMEM; |
4156d998 | 4350 | |
c41d13a3 | 4351 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_oid.name, |
7097f8df | 4352 | 0, size, ondisk); |
4156d998 | 4353 | if (ret < 0) |
662518b1 | 4354 | goto out; |
c0cd10db | 4355 | if ((size_t)ret < size) { |
4156d998 | 4356 | ret = -ENXIO; |
06ecc6cb AE |
4357 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
4358 | size, ret); | |
662518b1 | 4359 | goto out; |
4156d998 AE |
4360 | } |
4361 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
4362 | ret = -ENXIO; | |
06ecc6cb | 4363 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 4364 | goto out; |
81e759fb | 4365 | } |
602adf40 | 4366 | |
4156d998 AE |
4367 | names_size = le64_to_cpu(ondisk->snap_names_len); |
4368 | want_count = snap_count; | |
4369 | snap_count = le32_to_cpu(ondisk->snap_count); | |
4370 | } while (snap_count != want_count); | |
00f1f36f | 4371 | |
662518b1 AE |
4372 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
4373 | out: | |
4156d998 AE |
4374 | kfree(ondisk); |
4375 | ||
4376 | return ret; | |
602adf40 YS |
4377 | } |
4378 | ||
15228ede AE |
4379 | /* |
4380 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
4381 | * has disappeared from the (just updated) snapshot context. | |
4382 | */ | |
4383 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
4384 | { | |
4385 | u64 snap_id; | |
4386 | ||
4387 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
4388 | return; | |
4389 | ||
4390 | snap_id = rbd_dev->spec->snap_id; | |
4391 | if (snap_id == CEPH_NOSNAP) | |
4392 | return; | |
4393 | ||
4394 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
4395 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
4396 | } | |
4397 | ||
9875201e JD |
4398 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
4399 | { | |
4400 | sector_t size; | |
9875201e JD |
4401 | |
4402 | /* | |
811c6688 ID |
4403 | * If EXISTS is not set, rbd_dev->disk may be NULL, so don't |
4404 | * try to update its size. If REMOVING is set, updating size | |
4405 | * is just useless work since the device can't be opened. | |
9875201e | 4406 | */ |
811c6688 ID |
4407 | if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) && |
4408 | !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) { | |
9875201e JD |
4409 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; |
4410 | dout("setting size to %llu sectors", (unsigned long long)size); | |
4411 | set_capacity(rbd_dev->disk, size); | |
4412 | revalidate_disk(rbd_dev->disk); | |
4413 | } | |
4414 | } | |
4415 | ||
cc4a38bd | 4416 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 4417 | { |
e627db08 | 4418 | u64 mapping_size; |
1fe5e993 AE |
4419 | int ret; |
4420 | ||
cfbf6377 | 4421 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 4422 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
4423 | |
4424 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 4425 | if (ret) |
73e39e4d | 4426 | goto out; |
15228ede | 4427 | |
e8f59b59 ID |
4428 | /* |
4429 | * If there is a parent, see if it has disappeared due to the | |
4430 | * mapped image getting flattened. | |
4431 | */ | |
4432 | if (rbd_dev->parent) { | |
4433 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4434 | if (ret) | |
73e39e4d | 4435 | goto out; |
e8f59b59 ID |
4436 | } |
4437 | ||
5ff1108c | 4438 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 4439 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
4440 | } else { |
4441 | /* validate mapped snapshot's EXISTS flag */ | |
4442 | rbd_exists_validate(rbd_dev); | |
4443 | } | |
15228ede | 4444 | |
73e39e4d | 4445 | out: |
cfbf6377 | 4446 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 4447 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 4448 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 4449 | |
73e39e4d | 4450 | return ret; |
1fe5e993 AE |
4451 | } |
4452 | ||
7ad18afa CH |
4453 | static int rbd_init_request(void *data, struct request *rq, |
4454 | unsigned int hctx_idx, unsigned int request_idx, | |
4455 | unsigned int numa_node) | |
4456 | { | |
4457 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
4458 | ||
4459 | INIT_WORK(work, rbd_queue_workfn); | |
4460 | return 0; | |
4461 | } | |
4462 | ||
4463 | static struct blk_mq_ops rbd_mq_ops = { | |
4464 | .queue_rq = rbd_queue_rq, | |
4465 | .map_queue = blk_mq_map_queue, | |
4466 | .init_request = rbd_init_request, | |
4467 | }; | |
4468 | ||
602adf40 YS |
4469 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
4470 | { | |
4471 | struct gendisk *disk; | |
4472 | struct request_queue *q; | |
593a9e7b | 4473 | u64 segment_size; |
7ad18afa | 4474 | int err; |
602adf40 | 4475 | |
602adf40 | 4476 | /* create gendisk info */ |
7e513d43 ID |
4477 | disk = alloc_disk(single_major ? |
4478 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
4479 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 4480 | if (!disk) |
1fcdb8aa | 4481 | return -ENOMEM; |
602adf40 | 4482 | |
f0f8cef5 | 4483 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 4484 | rbd_dev->dev_id); |
602adf40 | 4485 | disk->major = rbd_dev->major; |
dd82fff1 | 4486 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
4487 | if (single_major) |
4488 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
4489 | disk->fops = &rbd_bd_ops; |
4490 | disk->private_data = rbd_dev; | |
4491 | ||
7ad18afa CH |
4492 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
4493 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 4494 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 4495 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 4496 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
4497 | rbd_dev->tag_set.nr_hw_queues = 1; |
4498 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
4499 | ||
4500 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
4501 | if (err) | |
602adf40 | 4502 | goto out_disk; |
029bcbd8 | 4503 | |
7ad18afa CH |
4504 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
4505 | if (IS_ERR(q)) { | |
4506 | err = PTR_ERR(q); | |
4507 | goto out_tag_set; | |
4508 | } | |
4509 | ||
d8a2c89c ID |
4510 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
4511 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 4512 | |
029bcbd8 | 4513 | /* set io sizes to object size */ |
593a9e7b AE |
4514 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
4515 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
0d9fde4f | 4516 | q->limits.max_sectors = queue_max_hw_sectors(q); |
d3834fef | 4517 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
4518 | blk_queue_max_segment_size(q, segment_size); |
4519 | blk_queue_io_min(q, segment_size); | |
4520 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 4521 | |
90e98c52 GZ |
4522 | /* enable the discard support */ |
4523 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
4524 | q->limits.discard_granularity = segment_size; | |
4525 | q->limits.discard_alignment = segment_size; | |
2bb4cd5c | 4526 | blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE); |
b76f8239 | 4527 | q->limits.discard_zeroes_data = 1; |
90e98c52 | 4528 | |
bae818ee RH |
4529 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
4530 | q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES; | |
4531 | ||
602adf40 YS |
4532 | disk->queue = q; |
4533 | ||
4534 | q->queuedata = rbd_dev; | |
4535 | ||
4536 | rbd_dev->disk = disk; | |
602adf40 | 4537 | |
602adf40 | 4538 | return 0; |
7ad18afa CH |
4539 | out_tag_set: |
4540 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
4541 | out_disk: |
4542 | put_disk(disk); | |
7ad18afa | 4543 | return err; |
602adf40 YS |
4544 | } |
4545 | ||
dfc5606d YS |
4546 | /* |
4547 | sysfs | |
4548 | */ | |
4549 | ||
593a9e7b AE |
4550 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
4551 | { | |
4552 | return container_of(dev, struct rbd_device, dev); | |
4553 | } | |
4554 | ||
dfc5606d YS |
4555 | static ssize_t rbd_size_show(struct device *dev, |
4556 | struct device_attribute *attr, char *buf) | |
4557 | { | |
593a9e7b | 4558 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 4559 | |
fc71d833 AE |
4560 | return sprintf(buf, "%llu\n", |
4561 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
4562 | } |
4563 | ||
34b13184 AE |
4564 | /* |
4565 | * Note this shows the features for whatever's mapped, which is not | |
4566 | * necessarily the base image. | |
4567 | */ | |
4568 | static ssize_t rbd_features_show(struct device *dev, | |
4569 | struct device_attribute *attr, char *buf) | |
4570 | { | |
4571 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4572 | ||
4573 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 4574 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
4575 | } |
4576 | ||
dfc5606d YS |
4577 | static ssize_t rbd_major_show(struct device *dev, |
4578 | struct device_attribute *attr, char *buf) | |
4579 | { | |
593a9e7b | 4580 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4581 | |
fc71d833 AE |
4582 | if (rbd_dev->major) |
4583 | return sprintf(buf, "%d\n", rbd_dev->major); | |
4584 | ||
4585 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
4586 | } |
4587 | ||
4588 | static ssize_t rbd_minor_show(struct device *dev, | |
4589 | struct device_attribute *attr, char *buf) | |
4590 | { | |
4591 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 4592 | |
dd82fff1 | 4593 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
4594 | } |
4595 | ||
005a07bf ID |
4596 | static ssize_t rbd_client_addr_show(struct device *dev, |
4597 | struct device_attribute *attr, char *buf) | |
4598 | { | |
4599 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4600 | struct ceph_entity_addr *client_addr = | |
4601 | ceph_client_addr(rbd_dev->rbd_client->client); | |
4602 | ||
4603 | return sprintf(buf, "%pISpc/%u\n", &client_addr->in_addr, | |
4604 | le32_to_cpu(client_addr->nonce)); | |
4605 | } | |
4606 | ||
dfc5606d YS |
4607 | static ssize_t rbd_client_id_show(struct device *dev, |
4608 | struct device_attribute *attr, char *buf) | |
602adf40 | 4609 | { |
593a9e7b | 4610 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4611 | |
1dbb4399 | 4612 | return sprintf(buf, "client%lld\n", |
033268a5 | 4613 | ceph_client_gid(rbd_dev->rbd_client->client)); |
602adf40 YS |
4614 | } |
4615 | ||
267fb90b MC |
4616 | static ssize_t rbd_cluster_fsid_show(struct device *dev, |
4617 | struct device_attribute *attr, char *buf) | |
4618 | { | |
4619 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4620 | ||
4621 | return sprintf(buf, "%pU\n", &rbd_dev->rbd_client->client->fsid); | |
4622 | } | |
4623 | ||
0d6d1e9c MC |
4624 | static ssize_t rbd_config_info_show(struct device *dev, |
4625 | struct device_attribute *attr, char *buf) | |
4626 | { | |
4627 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4628 | ||
4629 | return sprintf(buf, "%s\n", rbd_dev->config_info); | |
4630 | } | |
4631 | ||
dfc5606d YS |
4632 | static ssize_t rbd_pool_show(struct device *dev, |
4633 | struct device_attribute *attr, char *buf) | |
602adf40 | 4634 | { |
593a9e7b | 4635 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4636 | |
0d7dbfce | 4637 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
4638 | } |
4639 | ||
9bb2f334 AE |
4640 | static ssize_t rbd_pool_id_show(struct device *dev, |
4641 | struct device_attribute *attr, char *buf) | |
4642 | { | |
4643 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4644 | ||
0d7dbfce | 4645 | return sprintf(buf, "%llu\n", |
fc71d833 | 4646 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
4647 | } |
4648 | ||
dfc5606d YS |
4649 | static ssize_t rbd_name_show(struct device *dev, |
4650 | struct device_attribute *attr, char *buf) | |
4651 | { | |
593a9e7b | 4652 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4653 | |
a92ffdf8 AE |
4654 | if (rbd_dev->spec->image_name) |
4655 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
4656 | ||
4657 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
4658 | } |
4659 | ||
589d30e0 AE |
4660 | static ssize_t rbd_image_id_show(struct device *dev, |
4661 | struct device_attribute *attr, char *buf) | |
4662 | { | |
4663 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4664 | ||
0d7dbfce | 4665 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
4666 | } |
4667 | ||
34b13184 AE |
4668 | /* |
4669 | * Shows the name of the currently-mapped snapshot (or | |
4670 | * RBD_SNAP_HEAD_NAME for the base image). | |
4671 | */ | |
dfc5606d YS |
4672 | static ssize_t rbd_snap_show(struct device *dev, |
4673 | struct device_attribute *attr, | |
4674 | char *buf) | |
4675 | { | |
593a9e7b | 4676 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 4677 | |
0d7dbfce | 4678 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
4679 | } |
4680 | ||
92a58671 MC |
4681 | static ssize_t rbd_snap_id_show(struct device *dev, |
4682 | struct device_attribute *attr, char *buf) | |
4683 | { | |
4684 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4685 | ||
4686 | return sprintf(buf, "%llu\n", rbd_dev->spec->snap_id); | |
4687 | } | |
4688 | ||
86b00e0d | 4689 | /* |
ff96128f ID |
4690 | * For a v2 image, shows the chain of parent images, separated by empty |
4691 | * lines. For v1 images or if there is no parent, shows "(no parent | |
4692 | * image)". | |
86b00e0d AE |
4693 | */ |
4694 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
4695 | struct device_attribute *attr, |
4696 | char *buf) | |
86b00e0d AE |
4697 | { |
4698 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 4699 | ssize_t count = 0; |
86b00e0d | 4700 | |
ff96128f | 4701 | if (!rbd_dev->parent) |
86b00e0d AE |
4702 | return sprintf(buf, "(no parent image)\n"); |
4703 | ||
ff96128f ID |
4704 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
4705 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
4706 | ||
4707 | count += sprintf(&buf[count], "%s" | |
4708 | "pool_id %llu\npool_name %s\n" | |
4709 | "image_id %s\nimage_name %s\n" | |
4710 | "snap_id %llu\nsnap_name %s\n" | |
4711 | "overlap %llu\n", | |
4712 | !count ? "" : "\n", /* first? */ | |
4713 | spec->pool_id, spec->pool_name, | |
4714 | spec->image_id, spec->image_name ?: "(unknown)", | |
4715 | spec->snap_id, spec->snap_name, | |
4716 | rbd_dev->parent_overlap); | |
4717 | } | |
4718 | ||
4719 | return count; | |
86b00e0d AE |
4720 | } |
4721 | ||
dfc5606d YS |
4722 | static ssize_t rbd_image_refresh(struct device *dev, |
4723 | struct device_attribute *attr, | |
4724 | const char *buf, | |
4725 | size_t size) | |
4726 | { | |
593a9e7b | 4727 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 4728 | int ret; |
602adf40 | 4729 | |
cc4a38bd | 4730 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 4731 | if (ret) |
52bb1f9b | 4732 | return ret; |
b813623a | 4733 | |
52bb1f9b | 4734 | return size; |
dfc5606d | 4735 | } |
602adf40 | 4736 | |
dfc5606d | 4737 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 4738 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 4739 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 4740 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
005a07bf | 4741 | static DEVICE_ATTR(client_addr, S_IRUGO, rbd_client_addr_show, NULL); |
dfc5606d | 4742 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
267fb90b | 4743 | static DEVICE_ATTR(cluster_fsid, S_IRUGO, rbd_cluster_fsid_show, NULL); |
0d6d1e9c | 4744 | static DEVICE_ATTR(config_info, S_IRUSR, rbd_config_info_show, NULL); |
dfc5606d | 4745 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); |
9bb2f334 | 4746 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 4747 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 4748 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
4749 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
4750 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
92a58671 | 4751 | static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL); |
86b00e0d | 4752 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
4753 | |
4754 | static struct attribute *rbd_attrs[] = { | |
4755 | &dev_attr_size.attr, | |
34b13184 | 4756 | &dev_attr_features.attr, |
dfc5606d | 4757 | &dev_attr_major.attr, |
dd82fff1 | 4758 | &dev_attr_minor.attr, |
005a07bf | 4759 | &dev_attr_client_addr.attr, |
dfc5606d | 4760 | &dev_attr_client_id.attr, |
267fb90b | 4761 | &dev_attr_cluster_fsid.attr, |
0d6d1e9c | 4762 | &dev_attr_config_info.attr, |
dfc5606d | 4763 | &dev_attr_pool.attr, |
9bb2f334 | 4764 | &dev_attr_pool_id.attr, |
dfc5606d | 4765 | &dev_attr_name.attr, |
589d30e0 | 4766 | &dev_attr_image_id.attr, |
dfc5606d | 4767 | &dev_attr_current_snap.attr, |
92a58671 | 4768 | &dev_attr_snap_id.attr, |
86b00e0d | 4769 | &dev_attr_parent.attr, |
dfc5606d | 4770 | &dev_attr_refresh.attr, |
dfc5606d YS |
4771 | NULL |
4772 | }; | |
4773 | ||
4774 | static struct attribute_group rbd_attr_group = { | |
4775 | .attrs = rbd_attrs, | |
4776 | }; | |
4777 | ||
4778 | static const struct attribute_group *rbd_attr_groups[] = { | |
4779 | &rbd_attr_group, | |
4780 | NULL | |
4781 | }; | |
4782 | ||
6cac4695 | 4783 | static void rbd_dev_release(struct device *dev); |
dfc5606d YS |
4784 | |
4785 | static struct device_type rbd_device_type = { | |
4786 | .name = "rbd", | |
4787 | .groups = rbd_attr_groups, | |
6cac4695 | 4788 | .release = rbd_dev_release, |
dfc5606d YS |
4789 | }; |
4790 | ||
8b8fb99c AE |
4791 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4792 | { | |
4793 | kref_get(&spec->kref); | |
4794 | ||
4795 | return spec; | |
4796 | } | |
4797 | ||
4798 | static void rbd_spec_free(struct kref *kref); | |
4799 | static void rbd_spec_put(struct rbd_spec *spec) | |
4800 | { | |
4801 | if (spec) | |
4802 | kref_put(&spec->kref, rbd_spec_free); | |
4803 | } | |
4804 | ||
4805 | static struct rbd_spec *rbd_spec_alloc(void) | |
4806 | { | |
4807 | struct rbd_spec *spec; | |
4808 | ||
4809 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4810 | if (!spec) | |
4811 | return NULL; | |
04077599 ID |
4812 | |
4813 | spec->pool_id = CEPH_NOPOOL; | |
4814 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4815 | kref_init(&spec->kref); |
4816 | ||
8b8fb99c AE |
4817 | return spec; |
4818 | } | |
4819 | ||
4820 | static void rbd_spec_free(struct kref *kref) | |
4821 | { | |
4822 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4823 | ||
4824 | kfree(spec->pool_name); | |
4825 | kfree(spec->image_id); | |
4826 | kfree(spec->image_name); | |
4827 | kfree(spec->snap_name); | |
4828 | kfree(spec); | |
4829 | } | |
4830 | ||
1643dfa4 | 4831 | static void rbd_dev_free(struct rbd_device *rbd_dev) |
dd5ac32d | 4832 | { |
99d16943 | 4833 | WARN_ON(rbd_dev->watch_state != RBD_WATCH_STATE_UNREGISTERED); |
ed95b21a | 4834 | WARN_ON(rbd_dev->lock_state != RBD_LOCK_STATE_UNLOCKED); |
99d16943 | 4835 | |
c41d13a3 | 4836 | ceph_oid_destroy(&rbd_dev->header_oid); |
6b6dddbe | 4837 | ceph_oloc_destroy(&rbd_dev->header_oloc); |
0d6d1e9c | 4838 | kfree(rbd_dev->config_info); |
c41d13a3 | 4839 | |
dd5ac32d ID |
4840 | rbd_put_client(rbd_dev->rbd_client); |
4841 | rbd_spec_put(rbd_dev->spec); | |
4842 | kfree(rbd_dev->opts); | |
4843 | kfree(rbd_dev); | |
1643dfa4 ID |
4844 | } |
4845 | ||
4846 | static void rbd_dev_release(struct device *dev) | |
4847 | { | |
4848 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4849 | bool need_put = !!rbd_dev->opts; | |
4850 | ||
4851 | if (need_put) { | |
4852 | destroy_workqueue(rbd_dev->task_wq); | |
4853 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
4854 | } | |
4855 | ||
4856 | rbd_dev_free(rbd_dev); | |
dd5ac32d ID |
4857 | |
4858 | /* | |
4859 | * This is racy, but way better than putting module outside of | |
4860 | * the release callback. The race window is pretty small, so | |
4861 | * doing something similar to dm (dm-builtin.c) is overkill. | |
4862 | */ | |
4863 | if (need_put) | |
4864 | module_put(THIS_MODULE); | |
4865 | } | |
4866 | ||
1643dfa4 ID |
4867 | static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc, |
4868 | struct rbd_spec *spec) | |
c53d5893 AE |
4869 | { |
4870 | struct rbd_device *rbd_dev; | |
4871 | ||
1643dfa4 | 4872 | rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL); |
c53d5893 AE |
4873 | if (!rbd_dev) |
4874 | return NULL; | |
4875 | ||
4876 | spin_lock_init(&rbd_dev->lock); | |
4877 | INIT_LIST_HEAD(&rbd_dev->node); | |
c53d5893 AE |
4878 | init_rwsem(&rbd_dev->header_rwsem); |
4879 | ||
c41d13a3 | 4880 | ceph_oid_init(&rbd_dev->header_oid); |
922dab61 | 4881 | ceph_oloc_init(&rbd_dev->header_oloc); |
c41d13a3 | 4882 | |
99d16943 ID |
4883 | mutex_init(&rbd_dev->watch_mutex); |
4884 | rbd_dev->watch_state = RBD_WATCH_STATE_UNREGISTERED; | |
4885 | INIT_DELAYED_WORK(&rbd_dev->watch_dwork, rbd_reregister_watch); | |
4886 | ||
ed95b21a ID |
4887 | init_rwsem(&rbd_dev->lock_rwsem); |
4888 | rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED; | |
4889 | INIT_WORK(&rbd_dev->acquired_lock_work, rbd_notify_acquired_lock); | |
4890 | INIT_WORK(&rbd_dev->released_lock_work, rbd_notify_released_lock); | |
4891 | INIT_DELAYED_WORK(&rbd_dev->lock_dwork, rbd_acquire_lock); | |
4892 | INIT_WORK(&rbd_dev->unlock_work, rbd_release_lock_work); | |
4893 | init_waitqueue_head(&rbd_dev->lock_waitq); | |
4894 | ||
dd5ac32d ID |
4895 | rbd_dev->dev.bus = &rbd_bus_type; |
4896 | rbd_dev->dev.type = &rbd_device_type; | |
4897 | rbd_dev->dev.parent = &rbd_root_dev; | |
dd5ac32d ID |
4898 | device_initialize(&rbd_dev->dev); |
4899 | ||
c53d5893 | 4900 | rbd_dev->rbd_client = rbdc; |
d147543d | 4901 | rbd_dev->spec = spec; |
0903e875 | 4902 | |
7627151e YZ |
4903 | rbd_dev->layout.stripe_unit = 1 << RBD_MAX_OBJ_ORDER; |
4904 | rbd_dev->layout.stripe_count = 1; | |
4905 | rbd_dev->layout.object_size = 1 << RBD_MAX_OBJ_ORDER; | |
4906 | rbd_dev->layout.pool_id = spec->pool_id; | |
30c156d9 | 4907 | RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL); |
0903e875 | 4908 | |
1643dfa4 ID |
4909 | return rbd_dev; |
4910 | } | |
4911 | ||
4912 | /* | |
4913 | * Create a mapping rbd_dev. | |
4914 | */ | |
4915 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, | |
4916 | struct rbd_spec *spec, | |
4917 | struct rbd_options *opts) | |
4918 | { | |
4919 | struct rbd_device *rbd_dev; | |
4920 | ||
4921 | rbd_dev = __rbd_dev_create(rbdc, spec); | |
4922 | if (!rbd_dev) | |
4923 | return NULL; | |
4924 | ||
4925 | rbd_dev->opts = opts; | |
4926 | ||
4927 | /* get an id and fill in device name */ | |
4928 | rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0, | |
4929 | minor_to_rbd_dev_id(1 << MINORBITS), | |
4930 | GFP_KERNEL); | |
4931 | if (rbd_dev->dev_id < 0) | |
4932 | goto fail_rbd_dev; | |
4933 | ||
4934 | sprintf(rbd_dev->name, RBD_DRV_NAME "%d", rbd_dev->dev_id); | |
4935 | rbd_dev->task_wq = alloc_ordered_workqueue("%s-tasks", WQ_MEM_RECLAIM, | |
4936 | rbd_dev->name); | |
4937 | if (!rbd_dev->task_wq) | |
4938 | goto fail_dev_id; | |
dd5ac32d | 4939 | |
1643dfa4 ID |
4940 | /* we have a ref from do_rbd_add() */ |
4941 | __module_get(THIS_MODULE); | |
4942 | ||
4943 | dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id); | |
c53d5893 | 4944 | return rbd_dev; |
1643dfa4 ID |
4945 | |
4946 | fail_dev_id: | |
4947 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); | |
4948 | fail_rbd_dev: | |
4949 | rbd_dev_free(rbd_dev); | |
4950 | return NULL; | |
c53d5893 AE |
4951 | } |
4952 | ||
4953 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4954 | { | |
dd5ac32d ID |
4955 | if (rbd_dev) |
4956 | put_device(&rbd_dev->dev); | |
c53d5893 AE |
4957 | } |
4958 | ||
9d475de5 AE |
4959 | /* |
4960 | * Get the size and object order for an image snapshot, or if | |
4961 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4962 | * image. | |
4963 | */ | |
4964 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4965 | u8 *order, u64 *snap_size) | |
4966 | { | |
4967 | __le64 snapid = cpu_to_le64(snap_id); | |
4968 | int ret; | |
4969 | struct { | |
4970 | u8 order; | |
4971 | __le64 size; | |
4972 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4973 | ||
c41d13a3 | 4974 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
9d475de5 | 4975 | "rbd", "get_size", |
4157976b | 4976 | &snapid, sizeof (snapid), |
e2a58ee5 | 4977 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4978 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4979 | if (ret < 0) |
4980 | return ret; | |
57385b51 AE |
4981 | if (ret < sizeof (size_buf)) |
4982 | return -ERANGE; | |
9d475de5 | 4983 | |
c3545579 | 4984 | if (order) { |
c86f86e9 | 4985 | *order = size_buf.order; |
c3545579 JD |
4986 | dout(" order %u", (unsigned int)*order); |
4987 | } | |
9d475de5 AE |
4988 | *snap_size = le64_to_cpu(size_buf.size); |
4989 | ||
c3545579 JD |
4990 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4991 | (unsigned long long)snap_id, | |
57385b51 | 4992 | (unsigned long long)*snap_size); |
9d475de5 AE |
4993 | |
4994 | return 0; | |
4995 | } | |
4996 | ||
4997 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4998 | { | |
4999 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
5000 | &rbd_dev->header.obj_order, | |
5001 | &rbd_dev->header.image_size); | |
5002 | } | |
5003 | ||
1e130199 AE |
5004 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
5005 | { | |
5006 | void *reply_buf; | |
5007 | int ret; | |
5008 | void *p; | |
5009 | ||
5010 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
5011 | if (!reply_buf) | |
5012 | return -ENOMEM; | |
5013 | ||
c41d13a3 | 5014 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
4157976b | 5015 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 5016 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 5017 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
5018 | if (ret < 0) |
5019 | goto out; | |
5020 | ||
5021 | p = reply_buf; | |
5022 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
5023 | p + ret, NULL, GFP_NOIO); |
5024 | ret = 0; | |
1e130199 AE |
5025 | |
5026 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
5027 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
5028 | rbd_dev->header.object_prefix = NULL; | |
5029 | } else { | |
5030 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
5031 | } | |
1e130199 AE |
5032 | out: |
5033 | kfree(reply_buf); | |
5034 | ||
5035 | return ret; | |
5036 | } | |
5037 | ||
b1b5402a AE |
5038 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
5039 | u64 *snap_features) | |
5040 | { | |
5041 | __le64 snapid = cpu_to_le64(snap_id); | |
5042 | struct { | |
5043 | __le64 features; | |
5044 | __le64 incompat; | |
4157976b | 5045 | } __attribute__ ((packed)) features_buf = { 0 }; |
d3767f0f | 5046 | u64 unsup; |
b1b5402a AE |
5047 | int ret; |
5048 | ||
c41d13a3 | 5049 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
b1b5402a | 5050 | "rbd", "get_features", |
4157976b | 5051 | &snapid, sizeof (snapid), |
e2a58ee5 | 5052 | &features_buf, sizeof (features_buf)); |
36be9a76 | 5053 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
5054 | if (ret < 0) |
5055 | return ret; | |
57385b51 AE |
5056 | if (ret < sizeof (features_buf)) |
5057 | return -ERANGE; | |
d889140c | 5058 | |
d3767f0f ID |
5059 | unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED; |
5060 | if (unsup) { | |
5061 | rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx", | |
5062 | unsup); | |
b8f5c6ed | 5063 | return -ENXIO; |
d3767f0f | 5064 | } |
d889140c | 5065 | |
b1b5402a AE |
5066 | *snap_features = le64_to_cpu(features_buf.features); |
5067 | ||
5068 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
5069 | (unsigned long long)snap_id, |
5070 | (unsigned long long)*snap_features, | |
5071 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
5072 | |
5073 | return 0; | |
5074 | } | |
5075 | ||
5076 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
5077 | { | |
5078 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
5079 | &rbd_dev->header.features); | |
5080 | } | |
5081 | ||
86b00e0d AE |
5082 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
5083 | { | |
5084 | struct rbd_spec *parent_spec; | |
5085 | size_t size; | |
5086 | void *reply_buf = NULL; | |
5087 | __le64 snapid; | |
5088 | void *p; | |
5089 | void *end; | |
642a2537 | 5090 | u64 pool_id; |
86b00e0d | 5091 | char *image_id; |
3b5cf2a2 | 5092 | u64 snap_id; |
86b00e0d | 5093 | u64 overlap; |
86b00e0d AE |
5094 | int ret; |
5095 | ||
5096 | parent_spec = rbd_spec_alloc(); | |
5097 | if (!parent_spec) | |
5098 | return -ENOMEM; | |
5099 | ||
5100 | size = sizeof (__le64) + /* pool_id */ | |
5101 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
5102 | sizeof (__le64) + /* snap_id */ | |
5103 | sizeof (__le64); /* overlap */ | |
5104 | reply_buf = kmalloc(size, GFP_KERNEL); | |
5105 | if (!reply_buf) { | |
5106 | ret = -ENOMEM; | |
5107 | goto out_err; | |
5108 | } | |
5109 | ||
4d9b67cd | 5110 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
c41d13a3 | 5111 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
86b00e0d | 5112 | "rbd", "get_parent", |
4157976b | 5113 | &snapid, sizeof (snapid), |
e2a58ee5 | 5114 | reply_buf, size); |
36be9a76 | 5115 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
5116 | if (ret < 0) |
5117 | goto out_err; | |
5118 | ||
86b00e0d | 5119 | p = reply_buf; |
57385b51 AE |
5120 | end = reply_buf + ret; |
5121 | ret = -ERANGE; | |
642a2537 | 5122 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
5123 | if (pool_id == CEPH_NOPOOL) { |
5124 | /* | |
5125 | * Either the parent never existed, or we have | |
5126 | * record of it but the image got flattened so it no | |
5127 | * longer has a parent. When the parent of a | |
5128 | * layered image disappears we immediately set the | |
5129 | * overlap to 0. The effect of this is that all new | |
5130 | * requests will be treated as if the image had no | |
5131 | * parent. | |
5132 | */ | |
5133 | if (rbd_dev->parent_overlap) { | |
5134 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
5135 | rbd_dev_parent_put(rbd_dev); |
5136 | pr_info("%s: clone image has been flattened\n", | |
5137 | rbd_dev->disk->disk_name); | |
5138 | } | |
5139 | ||
86b00e0d | 5140 | goto out; /* No parent? No problem. */ |
392a9dad | 5141 | } |
86b00e0d | 5142 | |
0903e875 AE |
5143 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5144 | ||
5145 | ret = -EIO; | |
642a2537 | 5146 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 5147 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 5148 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 5149 | goto out_err; |
c0cd10db | 5150 | } |
0903e875 | 5151 | |
979ed480 | 5152 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
5153 | if (IS_ERR(image_id)) { |
5154 | ret = PTR_ERR(image_id); | |
5155 | goto out_err; | |
5156 | } | |
3b5cf2a2 | 5157 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
5158 | ceph_decode_64_safe(&p, end, overlap, out_err); |
5159 | ||
3b5cf2a2 AE |
5160 | /* |
5161 | * The parent won't change (except when the clone is | |
5162 | * flattened, already handled that). So we only need to | |
5163 | * record the parent spec we have not already done so. | |
5164 | */ | |
5165 | if (!rbd_dev->parent_spec) { | |
5166 | parent_spec->pool_id = pool_id; | |
5167 | parent_spec->image_id = image_id; | |
5168 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
5169 | rbd_dev->parent_spec = parent_spec; |
5170 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
5171 | } else { |
5172 | kfree(image_id); | |
3b5cf2a2 AE |
5173 | } |
5174 | ||
5175 | /* | |
cf32bd9c ID |
5176 | * We always update the parent overlap. If it's zero we issue |
5177 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 5178 | */ |
3b5cf2a2 | 5179 | if (!overlap) { |
3b5cf2a2 | 5180 | if (parent_spec) { |
cf32bd9c ID |
5181 | /* refresh, careful to warn just once */ |
5182 | if (rbd_dev->parent_overlap) | |
5183 | rbd_warn(rbd_dev, | |
5184 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 5185 | } else { |
cf32bd9c ID |
5186 | /* initial probe */ |
5187 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 5188 | } |
70cf49cf | 5189 | } |
cf32bd9c ID |
5190 | rbd_dev->parent_overlap = overlap; |
5191 | ||
86b00e0d AE |
5192 | out: |
5193 | ret = 0; | |
5194 | out_err: | |
5195 | kfree(reply_buf); | |
5196 | rbd_spec_put(parent_spec); | |
5197 | ||
5198 | return ret; | |
5199 | } | |
5200 | ||
cc070d59 AE |
5201 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
5202 | { | |
5203 | struct { | |
5204 | __le64 stripe_unit; | |
5205 | __le64 stripe_count; | |
5206 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
5207 | size_t size = sizeof (striping_info_buf); | |
5208 | void *p; | |
5209 | u64 obj_size; | |
5210 | u64 stripe_unit; | |
5211 | u64 stripe_count; | |
5212 | int ret; | |
5213 | ||
c41d13a3 | 5214 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
cc070d59 | 5215 | "rbd", "get_stripe_unit_count", NULL, 0, |
e2a58ee5 | 5216 | (char *)&striping_info_buf, size); |
cc070d59 AE |
5217 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
5218 | if (ret < 0) | |
5219 | return ret; | |
5220 | if (ret < size) | |
5221 | return -ERANGE; | |
5222 | ||
5223 | /* | |
5224 | * We don't actually support the "fancy striping" feature | |
5225 | * (STRIPINGV2) yet, but if the striping sizes are the | |
5226 | * defaults the behavior is the same as before. So find | |
5227 | * out, and only fail if the image has non-default values. | |
5228 | */ | |
5229 | ret = -EINVAL; | |
5230 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
5231 | p = &striping_info_buf; | |
5232 | stripe_unit = ceph_decode_64(&p); | |
5233 | if (stripe_unit != obj_size) { | |
5234 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
5235 | "(got %llu want %llu)", | |
5236 | stripe_unit, obj_size); | |
5237 | return -EINVAL; | |
5238 | } | |
5239 | stripe_count = ceph_decode_64(&p); | |
5240 | if (stripe_count != 1) { | |
5241 | rbd_warn(rbd_dev, "unsupported stripe count " | |
5242 | "(got %llu want 1)", stripe_count); | |
5243 | return -EINVAL; | |
5244 | } | |
500d0c0f AE |
5245 | rbd_dev->header.stripe_unit = stripe_unit; |
5246 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
5247 | |
5248 | return 0; | |
5249 | } | |
5250 | ||
9e15b77d AE |
5251 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
5252 | { | |
5253 | size_t image_id_size; | |
5254 | char *image_id; | |
5255 | void *p; | |
5256 | void *end; | |
5257 | size_t size; | |
5258 | void *reply_buf = NULL; | |
5259 | size_t len = 0; | |
5260 | char *image_name = NULL; | |
5261 | int ret; | |
5262 | ||
5263 | rbd_assert(!rbd_dev->spec->image_name); | |
5264 | ||
69e7a02f AE |
5265 | len = strlen(rbd_dev->spec->image_id); |
5266 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
5267 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
5268 | if (!image_id) | |
5269 | return NULL; | |
5270 | ||
5271 | p = image_id; | |
4157976b | 5272 | end = image_id + image_id_size; |
57385b51 | 5273 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
5274 | |
5275 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
5276 | reply_buf = kmalloc(size, GFP_KERNEL); | |
5277 | if (!reply_buf) | |
5278 | goto out; | |
5279 | ||
36be9a76 | 5280 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
5281 | "rbd", "dir_get_name", |
5282 | image_id, image_id_size, | |
e2a58ee5 | 5283 | reply_buf, size); |
9e15b77d AE |
5284 | if (ret < 0) |
5285 | goto out; | |
5286 | p = reply_buf; | |
f40eb349 AE |
5287 | end = reply_buf + ret; |
5288 | ||
9e15b77d AE |
5289 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
5290 | if (IS_ERR(image_name)) | |
5291 | image_name = NULL; | |
5292 | else | |
5293 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
5294 | out: | |
5295 | kfree(reply_buf); | |
5296 | kfree(image_id); | |
5297 | ||
5298 | return image_name; | |
5299 | } | |
5300 | ||
2ad3d716 AE |
5301 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
5302 | { | |
5303 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
5304 | const char *snap_name; | |
5305 | u32 which = 0; | |
5306 | ||
5307 | /* Skip over names until we find the one we are looking for */ | |
5308 | ||
5309 | snap_name = rbd_dev->header.snap_names; | |
5310 | while (which < snapc->num_snaps) { | |
5311 | if (!strcmp(name, snap_name)) | |
5312 | return snapc->snaps[which]; | |
5313 | snap_name += strlen(snap_name) + 1; | |
5314 | which++; | |
5315 | } | |
5316 | return CEPH_NOSNAP; | |
5317 | } | |
5318 | ||
5319 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
5320 | { | |
5321 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
5322 | u32 which; | |
5323 | bool found = false; | |
5324 | u64 snap_id; | |
5325 | ||
5326 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
5327 | const char *snap_name; | |
5328 | ||
5329 | snap_id = snapc->snaps[which]; | |
5330 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
5331 | if (IS_ERR(snap_name)) { |
5332 | /* ignore no-longer existing snapshots */ | |
5333 | if (PTR_ERR(snap_name) == -ENOENT) | |
5334 | continue; | |
5335 | else | |
5336 | break; | |
5337 | } | |
2ad3d716 AE |
5338 | found = !strcmp(name, snap_name); |
5339 | kfree(snap_name); | |
5340 | } | |
5341 | return found ? snap_id : CEPH_NOSNAP; | |
5342 | } | |
5343 | ||
5344 | /* | |
5345 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
5346 | * no snapshot by that name is found, or if an error occurs. | |
5347 | */ | |
5348 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
5349 | { | |
5350 | if (rbd_dev->image_format == 1) | |
5351 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
5352 | ||
5353 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
5354 | } | |
5355 | ||
9e15b77d | 5356 | /* |
04077599 ID |
5357 | * An image being mapped will have everything but the snap id. |
5358 | */ | |
5359 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
5360 | { | |
5361 | struct rbd_spec *spec = rbd_dev->spec; | |
5362 | ||
5363 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
5364 | rbd_assert(spec->image_id && spec->image_name); | |
5365 | rbd_assert(spec->snap_name); | |
5366 | ||
5367 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
5368 | u64 snap_id; | |
5369 | ||
5370 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
5371 | if (snap_id == CEPH_NOSNAP) | |
5372 | return -ENOENT; | |
5373 | ||
5374 | spec->snap_id = snap_id; | |
5375 | } else { | |
5376 | spec->snap_id = CEPH_NOSNAP; | |
5377 | } | |
5378 | ||
5379 | return 0; | |
5380 | } | |
5381 | ||
5382 | /* | |
5383 | * A parent image will have all ids but none of the names. | |
e1d4213f | 5384 | * |
04077599 ID |
5385 | * All names in an rbd spec are dynamically allocated. It's OK if we |
5386 | * can't figure out the name for an image id. | |
9e15b77d | 5387 | */ |
04077599 | 5388 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 5389 | { |
2e9f7f1c AE |
5390 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
5391 | struct rbd_spec *spec = rbd_dev->spec; | |
5392 | const char *pool_name; | |
5393 | const char *image_name; | |
5394 | const char *snap_name; | |
9e15b77d AE |
5395 | int ret; |
5396 | ||
04077599 ID |
5397 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
5398 | rbd_assert(spec->image_id); | |
5399 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 5400 | |
2e9f7f1c | 5401 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 5402 | |
2e9f7f1c AE |
5403 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
5404 | if (!pool_name) { | |
5405 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
5406 | return -EIO; |
5407 | } | |
2e9f7f1c AE |
5408 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
5409 | if (!pool_name) | |
9e15b77d AE |
5410 | return -ENOMEM; |
5411 | ||
5412 | /* Fetch the image name; tolerate failure here */ | |
5413 | ||
2e9f7f1c AE |
5414 | image_name = rbd_dev_image_name(rbd_dev); |
5415 | if (!image_name) | |
06ecc6cb | 5416 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 5417 | |
04077599 | 5418 | /* Fetch the snapshot name */ |
9e15b77d | 5419 | |
2e9f7f1c | 5420 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
5421 | if (IS_ERR(snap_name)) { |
5422 | ret = PTR_ERR(snap_name); | |
9e15b77d | 5423 | goto out_err; |
2e9f7f1c AE |
5424 | } |
5425 | ||
5426 | spec->pool_name = pool_name; | |
5427 | spec->image_name = image_name; | |
5428 | spec->snap_name = snap_name; | |
9e15b77d AE |
5429 | |
5430 | return 0; | |
04077599 | 5431 | |
9e15b77d | 5432 | out_err: |
2e9f7f1c AE |
5433 | kfree(image_name); |
5434 | kfree(pool_name); | |
9e15b77d AE |
5435 | return ret; |
5436 | } | |
5437 | ||
cc4a38bd | 5438 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
5439 | { |
5440 | size_t size; | |
5441 | int ret; | |
5442 | void *reply_buf; | |
5443 | void *p; | |
5444 | void *end; | |
5445 | u64 seq; | |
5446 | u32 snap_count; | |
5447 | struct ceph_snap_context *snapc; | |
5448 | u32 i; | |
5449 | ||
5450 | /* | |
5451 | * We'll need room for the seq value (maximum snapshot id), | |
5452 | * snapshot count, and array of that many snapshot ids. | |
5453 | * For now we have a fixed upper limit on the number we're | |
5454 | * prepared to receive. | |
5455 | */ | |
5456 | size = sizeof (__le64) + sizeof (__le32) + | |
5457 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
5458 | reply_buf = kzalloc(size, GFP_KERNEL); | |
5459 | if (!reply_buf) | |
5460 | return -ENOMEM; | |
5461 | ||
c41d13a3 | 5462 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
4157976b | 5463 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 5464 | reply_buf, size); |
36be9a76 | 5465 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
5466 | if (ret < 0) |
5467 | goto out; | |
5468 | ||
35d489f9 | 5469 | p = reply_buf; |
57385b51 AE |
5470 | end = reply_buf + ret; |
5471 | ret = -ERANGE; | |
35d489f9 AE |
5472 | ceph_decode_64_safe(&p, end, seq, out); |
5473 | ceph_decode_32_safe(&p, end, snap_count, out); | |
5474 | ||
5475 | /* | |
5476 | * Make sure the reported number of snapshot ids wouldn't go | |
5477 | * beyond the end of our buffer. But before checking that, | |
5478 | * make sure the computed size of the snapshot context we | |
5479 | * allocate is representable in a size_t. | |
5480 | */ | |
5481 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
5482 | / sizeof (u64)) { | |
5483 | ret = -EINVAL; | |
5484 | goto out; | |
5485 | } | |
5486 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
5487 | goto out; | |
468521c1 | 5488 | ret = 0; |
35d489f9 | 5489 | |
812164f8 | 5490 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
5491 | if (!snapc) { |
5492 | ret = -ENOMEM; | |
5493 | goto out; | |
5494 | } | |
35d489f9 | 5495 | snapc->seq = seq; |
35d489f9 AE |
5496 | for (i = 0; i < snap_count; i++) |
5497 | snapc->snaps[i] = ceph_decode_64(&p); | |
5498 | ||
49ece554 | 5499 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
5500 | rbd_dev->header.snapc = snapc; |
5501 | ||
5502 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 5503 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
5504 | out: |
5505 | kfree(reply_buf); | |
5506 | ||
57385b51 | 5507 | return ret; |
35d489f9 AE |
5508 | } |
5509 | ||
54cac61f AE |
5510 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
5511 | u64 snap_id) | |
b8b1e2db AE |
5512 | { |
5513 | size_t size; | |
5514 | void *reply_buf; | |
54cac61f | 5515 | __le64 snapid; |
b8b1e2db AE |
5516 | int ret; |
5517 | void *p; | |
5518 | void *end; | |
b8b1e2db AE |
5519 | char *snap_name; |
5520 | ||
5521 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
5522 | reply_buf = kmalloc(size, GFP_KERNEL); | |
5523 | if (!reply_buf) | |
5524 | return ERR_PTR(-ENOMEM); | |
5525 | ||
54cac61f | 5526 | snapid = cpu_to_le64(snap_id); |
c41d13a3 | 5527 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_oid.name, |
b8b1e2db | 5528 | "rbd", "get_snapshot_name", |
54cac61f | 5529 | &snapid, sizeof (snapid), |
e2a58ee5 | 5530 | reply_buf, size); |
36be9a76 | 5531 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
5532 | if (ret < 0) { |
5533 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 5534 | goto out; |
f40eb349 | 5535 | } |
b8b1e2db AE |
5536 | |
5537 | p = reply_buf; | |
f40eb349 | 5538 | end = reply_buf + ret; |
e5c35534 | 5539 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 5540 | if (IS_ERR(snap_name)) |
b8b1e2db | 5541 | goto out; |
b8b1e2db | 5542 | |
f40eb349 | 5543 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 5544 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
5545 | out: |
5546 | kfree(reply_buf); | |
5547 | ||
f40eb349 | 5548 | return snap_name; |
b8b1e2db AE |
5549 | } |
5550 | ||
2df3fac7 | 5551 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 5552 | { |
2df3fac7 | 5553 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 5554 | int ret; |
117973fb | 5555 | |
1617e40c JD |
5556 | ret = rbd_dev_v2_image_size(rbd_dev); |
5557 | if (ret) | |
cfbf6377 | 5558 | return ret; |
1617e40c | 5559 | |
2df3fac7 AE |
5560 | if (first_time) { |
5561 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
5562 | if (ret) | |
cfbf6377 | 5563 | return ret; |
2df3fac7 AE |
5564 | } |
5565 | ||
cc4a38bd | 5566 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
5567 | if (ret && first_time) { |
5568 | kfree(rbd_dev->header.object_prefix); | |
5569 | rbd_dev->header.object_prefix = NULL; | |
5570 | } | |
117973fb AE |
5571 | |
5572 | return ret; | |
5573 | } | |
5574 | ||
a720ae09 ID |
5575 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
5576 | { | |
5577 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5578 | ||
5579 | if (rbd_dev->image_format == 1) | |
5580 | return rbd_dev_v1_header_info(rbd_dev); | |
5581 | ||
5582 | return rbd_dev_v2_header_info(rbd_dev); | |
5583 | } | |
5584 | ||
e28fff26 AE |
5585 | /* |
5586 | * Skips over white space at *buf, and updates *buf to point to the | |
5587 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
5588 | * the token (string of non-white space characters) found. Note |
5589 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
5590 | */ |
5591 | static inline size_t next_token(const char **buf) | |
5592 | { | |
5593 | /* | |
5594 | * These are the characters that produce nonzero for | |
5595 | * isspace() in the "C" and "POSIX" locales. | |
5596 | */ | |
5597 | const char *spaces = " \f\n\r\t\v"; | |
5598 | ||
5599 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
5600 | ||
5601 | return strcspn(*buf, spaces); /* Return token length */ | |
5602 | } | |
5603 | ||
ea3352f4 AE |
5604 | /* |
5605 | * Finds the next token in *buf, dynamically allocates a buffer big | |
5606 | * enough to hold a copy of it, and copies the token into the new | |
5607 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
5608 | * that a duplicate buffer is created even for a zero-length token. | |
5609 | * | |
5610 | * Returns a pointer to the newly-allocated duplicate, or a null | |
5611 | * pointer if memory for the duplicate was not available. If | |
5612 | * the lenp argument is a non-null pointer, the length of the token | |
5613 | * (not including the '\0') is returned in *lenp. | |
5614 | * | |
5615 | * If successful, the *buf pointer will be updated to point beyond | |
5616 | * the end of the found token. | |
5617 | * | |
5618 | * Note: uses GFP_KERNEL for allocation. | |
5619 | */ | |
5620 | static inline char *dup_token(const char **buf, size_t *lenp) | |
5621 | { | |
5622 | char *dup; | |
5623 | size_t len; | |
5624 | ||
5625 | len = next_token(buf); | |
4caf35f9 | 5626 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
5627 | if (!dup) |
5628 | return NULL; | |
ea3352f4 AE |
5629 | *(dup + len) = '\0'; |
5630 | *buf += len; | |
5631 | ||
5632 | if (lenp) | |
5633 | *lenp = len; | |
5634 | ||
5635 | return dup; | |
5636 | } | |
5637 | ||
a725f65e | 5638 | /* |
859c31df AE |
5639 | * Parse the options provided for an "rbd add" (i.e., rbd image |
5640 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
5641 | * and the data written is passed here via a NUL-terminated buffer. | |
5642 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 5643 | * |
859c31df AE |
5644 | * The information extracted from these options is recorded in |
5645 | * the other parameters which return dynamically-allocated | |
5646 | * structures: | |
5647 | * ceph_opts | |
5648 | * The address of a pointer that will refer to a ceph options | |
5649 | * structure. Caller must release the returned pointer using | |
5650 | * ceph_destroy_options() when it is no longer needed. | |
5651 | * rbd_opts | |
5652 | * Address of an rbd options pointer. Fully initialized by | |
5653 | * this function; caller must release with kfree(). | |
5654 | * spec | |
5655 | * Address of an rbd image specification pointer. Fully | |
5656 | * initialized by this function based on parsed options. | |
5657 | * Caller must release with rbd_spec_put(). | |
5658 | * | |
5659 | * The options passed take this form: | |
5660 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
5661 | * where: | |
5662 | * <mon_addrs> | |
5663 | * A comma-separated list of one or more monitor addresses. | |
5664 | * A monitor address is an ip address, optionally followed | |
5665 | * by a port number (separated by a colon). | |
5666 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
5667 | * <options> | |
5668 | * A comma-separated list of ceph and/or rbd options. | |
5669 | * <pool_name> | |
5670 | * The name of the rados pool containing the rbd image. | |
5671 | * <image_name> | |
5672 | * The name of the image in that pool to map. | |
5673 | * <snap_id> | |
5674 | * An optional snapshot id. If provided, the mapping will | |
5675 | * present data from the image at the time that snapshot was | |
5676 | * created. The image head is used if no snapshot id is | |
5677 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 5678 | */ |
859c31df | 5679 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 5680 | struct ceph_options **ceph_opts, |
859c31df AE |
5681 | struct rbd_options **opts, |
5682 | struct rbd_spec **rbd_spec) | |
e28fff26 | 5683 | { |
d22f76e7 | 5684 | size_t len; |
859c31df | 5685 | char *options; |
0ddebc0c | 5686 | const char *mon_addrs; |
ecb4dc22 | 5687 | char *snap_name; |
0ddebc0c | 5688 | size_t mon_addrs_size; |
859c31df | 5689 | struct rbd_spec *spec = NULL; |
4e9afeba | 5690 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5691 | struct ceph_options *copts; |
dc79b113 | 5692 | int ret; |
e28fff26 AE |
5693 | |
5694 | /* The first four tokens are required */ | |
5695 | ||
7ef3214a | 5696 | len = next_token(&buf); |
4fb5d671 AE |
5697 | if (!len) { |
5698 | rbd_warn(NULL, "no monitor address(es) provided"); | |
5699 | return -EINVAL; | |
5700 | } | |
0ddebc0c | 5701 | mon_addrs = buf; |
f28e565a | 5702 | mon_addrs_size = len + 1; |
7ef3214a | 5703 | buf += len; |
a725f65e | 5704 | |
dc79b113 | 5705 | ret = -EINVAL; |
f28e565a AE |
5706 | options = dup_token(&buf, NULL); |
5707 | if (!options) | |
dc79b113 | 5708 | return -ENOMEM; |
4fb5d671 AE |
5709 | if (!*options) { |
5710 | rbd_warn(NULL, "no options provided"); | |
5711 | goto out_err; | |
5712 | } | |
e28fff26 | 5713 | |
859c31df AE |
5714 | spec = rbd_spec_alloc(); |
5715 | if (!spec) | |
f28e565a | 5716 | goto out_mem; |
859c31df AE |
5717 | |
5718 | spec->pool_name = dup_token(&buf, NULL); | |
5719 | if (!spec->pool_name) | |
5720 | goto out_mem; | |
4fb5d671 AE |
5721 | if (!*spec->pool_name) { |
5722 | rbd_warn(NULL, "no pool name provided"); | |
5723 | goto out_err; | |
5724 | } | |
e28fff26 | 5725 | |
69e7a02f | 5726 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 5727 | if (!spec->image_name) |
f28e565a | 5728 | goto out_mem; |
4fb5d671 AE |
5729 | if (!*spec->image_name) { |
5730 | rbd_warn(NULL, "no image name provided"); | |
5731 | goto out_err; | |
5732 | } | |
d4b125e9 | 5733 | |
f28e565a AE |
5734 | /* |
5735 | * Snapshot name is optional; default is to use "-" | |
5736 | * (indicating the head/no snapshot). | |
5737 | */ | |
3feeb894 | 5738 | len = next_token(&buf); |
820a5f3e | 5739 | if (!len) { |
3feeb894 AE |
5740 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
5741 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 5742 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 5743 | ret = -ENAMETOOLONG; |
f28e565a | 5744 | goto out_err; |
849b4260 | 5745 | } |
ecb4dc22 AE |
5746 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
5747 | if (!snap_name) | |
f28e565a | 5748 | goto out_mem; |
ecb4dc22 AE |
5749 | *(snap_name + len) = '\0'; |
5750 | spec->snap_name = snap_name; | |
e5c35534 | 5751 | |
0ddebc0c | 5752 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 5753 | |
4e9afeba AE |
5754 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
5755 | if (!rbd_opts) | |
5756 | goto out_mem; | |
5757 | ||
5758 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 5759 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
d22f76e7 | 5760 | |
859c31df | 5761 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 5762 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 5763 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
5764 | if (IS_ERR(copts)) { |
5765 | ret = PTR_ERR(copts); | |
dc79b113 AE |
5766 | goto out_err; |
5767 | } | |
859c31df AE |
5768 | kfree(options); |
5769 | ||
5770 | *ceph_opts = copts; | |
4e9afeba | 5771 | *opts = rbd_opts; |
859c31df | 5772 | *rbd_spec = spec; |
0ddebc0c | 5773 | |
dc79b113 | 5774 | return 0; |
f28e565a | 5775 | out_mem: |
dc79b113 | 5776 | ret = -ENOMEM; |
d22f76e7 | 5777 | out_err: |
859c31df AE |
5778 | kfree(rbd_opts); |
5779 | rbd_spec_put(spec); | |
f28e565a | 5780 | kfree(options); |
d22f76e7 | 5781 | |
dc79b113 | 5782 | return ret; |
a725f65e AE |
5783 | } |
5784 | ||
30ba1f02 ID |
5785 | /* |
5786 | * Return pool id (>= 0) or a negative error code. | |
5787 | */ | |
5788 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
5789 | { | |
a319bf56 | 5790 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 5791 | u64 newest_epoch; |
30ba1f02 ID |
5792 | int tries = 0; |
5793 | int ret; | |
5794 | ||
5795 | again: | |
5796 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
5797 | if (ret == -ENOENT && tries++ < 1) { | |
d0b19705 ID |
5798 | ret = ceph_monc_get_version(&rbdc->client->monc, "osdmap", |
5799 | &newest_epoch); | |
30ba1f02 ID |
5800 | if (ret < 0) |
5801 | return ret; | |
5802 | ||
5803 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
7cca78c9 | 5804 | ceph_osdc_maybe_request_map(&rbdc->client->osdc); |
30ba1f02 | 5805 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, |
a319bf56 ID |
5806 | newest_epoch, |
5807 | opts->mount_timeout); | |
30ba1f02 ID |
5808 | goto again; |
5809 | } else { | |
5810 | /* the osdmap we have is new enough */ | |
5811 | return -ENOENT; | |
5812 | } | |
5813 | } | |
5814 | ||
5815 | return ret; | |
5816 | } | |
5817 | ||
589d30e0 AE |
5818 | /* |
5819 | * An rbd format 2 image has a unique identifier, distinct from the | |
5820 | * name given to it by the user. Internally, that identifier is | |
5821 | * what's used to specify the names of objects related to the image. | |
5822 | * | |
5823 | * A special "rbd id" object is used to map an rbd image name to its | |
5824 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5825 | * with the supplied name. | |
5826 | * | |
5827 | * This function will record the given rbd_dev's image_id field if | |
5828 | * it can be determined, and in that case will return 0. If any | |
5829 | * errors occur a negative errno will be returned and the rbd_dev's | |
5830 | * image_id field will be unchanged (and should be NULL). | |
5831 | */ | |
5832 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5833 | { | |
5834 | int ret; | |
5835 | size_t size; | |
5836 | char *object_name; | |
5837 | void *response; | |
c0fba368 | 5838 | char *image_id; |
2f82ee54 | 5839 | |
2c0d0a10 AE |
5840 | /* |
5841 | * When probing a parent image, the image id is already | |
5842 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5843 | * need to fetch the image id again in this case. We |
5844 | * do still need to set the image format though. | |
2c0d0a10 | 5845 | */ |
c0fba368 AE |
5846 | if (rbd_dev->spec->image_id) { |
5847 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5848 | ||
2c0d0a10 | 5849 | return 0; |
c0fba368 | 5850 | } |
2c0d0a10 | 5851 | |
589d30e0 AE |
5852 | /* |
5853 | * First, see if the format 2 image id file exists, and if | |
5854 | * so, get the image's persistent id from it. | |
5855 | */ | |
69e7a02f | 5856 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5857 | object_name = kmalloc(size, GFP_NOIO); |
5858 | if (!object_name) | |
5859 | return -ENOMEM; | |
0d7dbfce | 5860 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5861 | dout("rbd id object name is %s\n", object_name); |
5862 | ||
5863 | /* Response will be an encoded string, which includes a length */ | |
5864 | ||
5865 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5866 | response = kzalloc(size, GFP_NOIO); | |
5867 | if (!response) { | |
5868 | ret = -ENOMEM; | |
5869 | goto out; | |
5870 | } | |
5871 | ||
c0fba368 AE |
5872 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5873 | ||
36be9a76 | 5874 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5875 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5876 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5877 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5878 | if (ret == -ENOENT) { |
5879 | image_id = kstrdup("", GFP_KERNEL); | |
5880 | ret = image_id ? 0 : -ENOMEM; | |
5881 | if (!ret) | |
5882 | rbd_dev->image_format = 1; | |
7dd440c9 | 5883 | } else if (ret >= 0) { |
c0fba368 AE |
5884 | void *p = response; |
5885 | ||
5886 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5887 | NULL, GFP_NOIO); |
461f758a | 5888 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5889 | if (!ret) |
5890 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5891 | } |
5892 | ||
5893 | if (!ret) { | |
5894 | rbd_dev->spec->image_id = image_id; | |
5895 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5896 | } |
5897 | out: | |
5898 | kfree(response); | |
5899 | kfree(object_name); | |
5900 | ||
5901 | return ret; | |
5902 | } | |
5903 | ||
3abef3b3 AE |
5904 | /* |
5905 | * Undo whatever state changes are made by v1 or v2 header info | |
5906 | * call. | |
5907 | */ | |
6fd48b3b AE |
5908 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5909 | { | |
5910 | struct rbd_image_header *header; | |
5911 | ||
e69b8d41 | 5912 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5913 | |
5914 | /* Free dynamic fields from the header, then zero it out */ | |
5915 | ||
5916 | header = &rbd_dev->header; | |
812164f8 | 5917 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5918 | kfree(header->snap_sizes); |
5919 | kfree(header->snap_names); | |
5920 | kfree(header->object_prefix); | |
5921 | memset(header, 0, sizeof (*header)); | |
5922 | } | |
5923 | ||
2df3fac7 | 5924 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5925 | { |
5926 | int ret; | |
a30b71b9 | 5927 | |
1e130199 | 5928 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5929 | if (ret) |
b1b5402a AE |
5930 | goto out_err; |
5931 | ||
2df3fac7 AE |
5932 | /* |
5933 | * Get the and check features for the image. Currently the | |
5934 | * features are assumed to never change. | |
5935 | */ | |
b1b5402a | 5936 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5937 | if (ret) |
9d475de5 | 5938 | goto out_err; |
35d489f9 | 5939 | |
cc070d59 AE |
5940 | /* If the image supports fancy striping, get its parameters */ |
5941 | ||
5942 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5943 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5944 | if (ret < 0) | |
5945 | goto out_err; | |
5946 | } | |
2df3fac7 | 5947 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5948 | |
35152979 | 5949 | return 0; |
9d475de5 | 5950 | out_err: |
642a2537 | 5951 | rbd_dev->header.features = 0; |
1e130199 AE |
5952 | kfree(rbd_dev->header.object_prefix); |
5953 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5954 | |
5955 | return ret; | |
a30b71b9 AE |
5956 | } |
5957 | ||
6d69bb53 ID |
5958 | /* |
5959 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
5960 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
5961 | * length of the already discovered part of the parent chain. | |
5962 | */ | |
5963 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 5964 | { |
2f82ee54 | 5965 | struct rbd_device *parent = NULL; |
124afba2 AE |
5966 | int ret; |
5967 | ||
5968 | if (!rbd_dev->parent_spec) | |
5969 | return 0; | |
124afba2 | 5970 | |
6d69bb53 ID |
5971 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
5972 | pr_info("parent chain is too long (%d)\n", depth); | |
5973 | ret = -EINVAL; | |
5974 | goto out_err; | |
5975 | } | |
5976 | ||
1643dfa4 | 5977 | parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec); |
1f2c6651 ID |
5978 | if (!parent) { |
5979 | ret = -ENOMEM; | |
124afba2 | 5980 | goto out_err; |
1f2c6651 ID |
5981 | } |
5982 | ||
5983 | /* | |
5984 | * Images related by parent/child relationships always share | |
5985 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
5986 | */ | |
5987 | __rbd_get_client(rbd_dev->rbd_client); | |
5988 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 5989 | |
6d69bb53 | 5990 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
5991 | if (ret < 0) |
5992 | goto out_err; | |
1f2c6651 | 5993 | |
124afba2 | 5994 | rbd_dev->parent = parent; |
a2acd00e | 5995 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 5996 | return 0; |
1f2c6651 | 5997 | |
124afba2 | 5998 | out_err: |
1f2c6651 | 5999 | rbd_dev_unparent(rbd_dev); |
1761b229 | 6000 | rbd_dev_destroy(parent); |
124afba2 AE |
6001 | return ret; |
6002 | } | |
6003 | ||
811c6688 ID |
6004 | /* |
6005 | * rbd_dev->header_rwsem must be locked for write and will be unlocked | |
6006 | * upon return. | |
6007 | */ | |
200a6a8b | 6008 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 6009 | { |
83a06263 | 6010 | int ret; |
d1cf5788 | 6011 | |
9b60e70b | 6012 | /* Record our major and minor device numbers. */ |
83a06263 | 6013 | |
9b60e70b ID |
6014 | if (!single_major) { |
6015 | ret = register_blkdev(0, rbd_dev->name); | |
6016 | if (ret < 0) | |
1643dfa4 | 6017 | goto err_out_unlock; |
9b60e70b ID |
6018 | |
6019 | rbd_dev->major = ret; | |
6020 | rbd_dev->minor = 0; | |
6021 | } else { | |
6022 | rbd_dev->major = rbd_major; | |
6023 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
6024 | } | |
83a06263 AE |
6025 | |
6026 | /* Set up the blkdev mapping. */ | |
6027 | ||
6028 | ret = rbd_init_disk(rbd_dev); | |
6029 | if (ret) | |
6030 | goto err_out_blkdev; | |
6031 | ||
f35a4dee | 6032 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
6033 | if (ret) |
6034 | goto err_out_disk; | |
bc1ecc65 | 6035 | |
f35a4dee | 6036 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 6037 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 6038 | |
dd5ac32d ID |
6039 | dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id); |
6040 | ret = device_add(&rbd_dev->dev); | |
f35a4dee | 6041 | if (ret) |
f5ee37bd | 6042 | goto err_out_mapping; |
83a06263 | 6043 | |
83a06263 AE |
6044 | /* Everything's ready. Announce the disk to the world. */ |
6045 | ||
129b79d4 | 6046 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
811c6688 | 6047 | up_write(&rbd_dev->header_rwsem); |
83a06263 | 6048 | |
1643dfa4 ID |
6049 | spin_lock(&rbd_dev_list_lock); |
6050 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
6051 | spin_unlock(&rbd_dev_list_lock); | |
6052 | ||
811c6688 | 6053 | add_disk(rbd_dev->disk); |
ca7909e8 ID |
6054 | pr_info("%s: capacity %llu features 0x%llx\n", rbd_dev->disk->disk_name, |
6055 | (unsigned long long)get_capacity(rbd_dev->disk) << SECTOR_SHIFT, | |
6056 | rbd_dev->header.features); | |
83a06263 AE |
6057 | |
6058 | return ret; | |
2f82ee54 | 6059 | |
f35a4dee AE |
6060 | err_out_mapping: |
6061 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
6062 | err_out_disk: |
6063 | rbd_free_disk(rbd_dev); | |
6064 | err_out_blkdev: | |
9b60e70b ID |
6065 | if (!single_major) |
6066 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
811c6688 ID |
6067 | err_out_unlock: |
6068 | up_write(&rbd_dev->header_rwsem); | |
83a06263 AE |
6069 | return ret; |
6070 | } | |
6071 | ||
332bb12d AE |
6072 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
6073 | { | |
6074 | struct rbd_spec *spec = rbd_dev->spec; | |
c41d13a3 | 6075 | int ret; |
332bb12d AE |
6076 | |
6077 | /* Record the header object name for this rbd image. */ | |
6078 | ||
6079 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
6080 | ||
7627151e | 6081 | rbd_dev->header_oloc.pool = rbd_dev->layout.pool_id; |
332bb12d | 6082 | if (rbd_dev->image_format == 1) |
c41d13a3 ID |
6083 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
6084 | spec->image_name, RBD_SUFFIX); | |
332bb12d | 6085 | else |
c41d13a3 ID |
6086 | ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s", |
6087 | RBD_HEADER_PREFIX, spec->image_id); | |
332bb12d | 6088 | |
c41d13a3 | 6089 | return ret; |
332bb12d AE |
6090 | } |
6091 | ||
200a6a8b AE |
6092 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
6093 | { | |
6fd48b3b | 6094 | rbd_dev_unprobe(rbd_dev); |
6fd48b3b AE |
6095 | rbd_dev->image_format = 0; |
6096 | kfree(rbd_dev->spec->image_id); | |
6097 | rbd_dev->spec->image_id = NULL; | |
6098 | ||
200a6a8b AE |
6099 | rbd_dev_destroy(rbd_dev); |
6100 | } | |
6101 | ||
a30b71b9 AE |
6102 | /* |
6103 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
6104 | * device. If this image is the one being mapped (i.e., not a |
6105 | * parent), initiate a watch on its header object before using that | |
6106 | * object to get detailed information about the rbd image. | |
a30b71b9 | 6107 | */ |
6d69bb53 | 6108 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth) |
a30b71b9 AE |
6109 | { |
6110 | int ret; | |
6111 | ||
6112 | /* | |
3abef3b3 AE |
6113 | * Get the id from the image id object. Unless there's an |
6114 | * error, rbd_dev->spec->image_id will be filled in with | |
6115 | * a dynamically-allocated string, and rbd_dev->image_format | |
6116 | * will be set to either 1 or 2. | |
a30b71b9 AE |
6117 | */ |
6118 | ret = rbd_dev_image_id(rbd_dev); | |
6119 | if (ret) | |
c0fba368 | 6120 | return ret; |
c0fba368 | 6121 | |
332bb12d AE |
6122 | ret = rbd_dev_header_name(rbd_dev); |
6123 | if (ret) | |
6124 | goto err_out_format; | |
6125 | ||
6d69bb53 | 6126 | if (!depth) { |
99d16943 | 6127 | ret = rbd_register_watch(rbd_dev); |
1fe48023 ID |
6128 | if (ret) { |
6129 | if (ret == -ENOENT) | |
6130 | pr_info("image %s/%s does not exist\n", | |
6131 | rbd_dev->spec->pool_name, | |
6132 | rbd_dev->spec->image_name); | |
c41d13a3 | 6133 | goto err_out_format; |
1fe48023 | 6134 | } |
1f3ef788 | 6135 | } |
b644de2b | 6136 | |
a720ae09 | 6137 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 6138 | if (ret) |
b644de2b | 6139 | goto err_out_watch; |
83a06263 | 6140 | |
04077599 ID |
6141 | /* |
6142 | * If this image is the one being mapped, we have pool name and | |
6143 | * id, image name and id, and snap name - need to fill snap id. | |
6144 | * Otherwise this is a parent image, identified by pool, image | |
6145 | * and snap ids - need to fill in names for those ids. | |
6146 | */ | |
6d69bb53 | 6147 | if (!depth) |
04077599 ID |
6148 | ret = rbd_spec_fill_snap_id(rbd_dev); |
6149 | else | |
6150 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
6151 | if (ret) { |
6152 | if (ret == -ENOENT) | |
6153 | pr_info("snap %s/%s@%s does not exist\n", | |
6154 | rbd_dev->spec->pool_name, | |
6155 | rbd_dev->spec->image_name, | |
6156 | rbd_dev->spec->snap_name); | |
33dca39f | 6157 | goto err_out_probe; |
1fe48023 | 6158 | } |
9bb81c9b | 6159 | |
e8f59b59 ID |
6160 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
6161 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
6162 | if (ret) | |
6163 | goto err_out_probe; | |
6164 | ||
6165 | /* | |
6166 | * Need to warn users if this image is the one being | |
6167 | * mapped and has a parent. | |
6168 | */ | |
6d69bb53 | 6169 | if (!depth && rbd_dev->parent_spec) |
e8f59b59 ID |
6170 | rbd_warn(rbd_dev, |
6171 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
6172 | } | |
6173 | ||
6d69bb53 | 6174 | ret = rbd_dev_probe_parent(rbd_dev, depth); |
30d60ba2 AE |
6175 | if (ret) |
6176 | goto err_out_probe; | |
6177 | ||
6178 | dout("discovered format %u image, header name is %s\n", | |
c41d13a3 | 6179 | rbd_dev->image_format, rbd_dev->header_oid.name); |
30d60ba2 | 6180 | return 0; |
e8f59b59 | 6181 | |
6fd48b3b AE |
6182 | err_out_probe: |
6183 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 6184 | err_out_watch: |
6d69bb53 | 6185 | if (!depth) |
99d16943 | 6186 | rbd_unregister_watch(rbd_dev); |
332bb12d AE |
6187 | err_out_format: |
6188 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
6189 | kfree(rbd_dev->spec->image_id); |
6190 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
6191 | return ret; |
6192 | } | |
6193 | ||
9b60e70b ID |
6194 | static ssize_t do_rbd_add(struct bus_type *bus, |
6195 | const char *buf, | |
6196 | size_t count) | |
602adf40 | 6197 | { |
cb8627c7 | 6198 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 6199 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 6200 | struct rbd_options *rbd_opts = NULL; |
859c31df | 6201 | struct rbd_spec *spec = NULL; |
9d3997fd | 6202 | struct rbd_client *rbdc; |
51344a38 | 6203 | bool read_only; |
b51c83c2 | 6204 | int rc; |
602adf40 YS |
6205 | |
6206 | if (!try_module_get(THIS_MODULE)) | |
6207 | return -ENODEV; | |
6208 | ||
602adf40 | 6209 | /* parse add command */ |
859c31df | 6210 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 6211 | if (rc < 0) |
dd5ac32d | 6212 | goto out; |
78cea76e | 6213 | |
9d3997fd AE |
6214 | rbdc = rbd_get_client(ceph_opts); |
6215 | if (IS_ERR(rbdc)) { | |
6216 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 6217 | goto err_out_args; |
9d3997fd | 6218 | } |
602adf40 | 6219 | |
602adf40 | 6220 | /* pick the pool */ |
30ba1f02 | 6221 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
6222 | if (rc < 0) { |
6223 | if (rc == -ENOENT) | |
6224 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 6225 | goto err_out_client; |
1fe48023 | 6226 | } |
c0cd10db | 6227 | spec->pool_id = (u64)rc; |
859c31df | 6228 | |
d147543d | 6229 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
b51c83c2 ID |
6230 | if (!rbd_dev) { |
6231 | rc = -ENOMEM; | |
bd4ba655 | 6232 | goto err_out_client; |
b51c83c2 | 6233 | } |
c53d5893 AE |
6234 | rbdc = NULL; /* rbd_dev now owns this */ |
6235 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 6236 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 6237 | |
0d6d1e9c MC |
6238 | rbd_dev->config_info = kstrdup(buf, GFP_KERNEL); |
6239 | if (!rbd_dev->config_info) { | |
6240 | rc = -ENOMEM; | |
6241 | goto err_out_rbd_dev; | |
6242 | } | |
6243 | ||
811c6688 | 6244 | down_write(&rbd_dev->header_rwsem); |
6d69bb53 | 6245 | rc = rbd_dev_image_probe(rbd_dev, 0); |
0d6d1e9c MC |
6246 | if (rc < 0) { |
6247 | up_write(&rbd_dev->header_rwsem); | |
c53d5893 | 6248 | goto err_out_rbd_dev; |
0d6d1e9c | 6249 | } |
05fd6f6f | 6250 | |
7ce4eef7 AE |
6251 | /* If we are mapping a snapshot it must be marked read-only */ |
6252 | ||
d147543d | 6253 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
6254 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
6255 | read_only = true; | |
6256 | rbd_dev->mapping.read_only = read_only; | |
6257 | ||
b536f69a | 6258 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 6259 | if (rc) { |
e37180c0 | 6260 | /* |
99d16943 | 6261 | * rbd_unregister_watch() can't be moved into |
e37180c0 ID |
6262 | * rbd_dev_image_release() without refactoring, see |
6263 | * commit 1f3ef78861ac. | |
6264 | */ | |
99d16943 | 6265 | rbd_unregister_watch(rbd_dev); |
3abef3b3 | 6266 | rbd_dev_image_release(rbd_dev); |
dd5ac32d | 6267 | goto out; |
3abef3b3 AE |
6268 | } |
6269 | ||
dd5ac32d ID |
6270 | rc = count; |
6271 | out: | |
6272 | module_put(THIS_MODULE); | |
6273 | return rc; | |
b536f69a | 6274 | |
c53d5893 AE |
6275 | err_out_rbd_dev: |
6276 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 6277 | err_out_client: |
9d3997fd | 6278 | rbd_put_client(rbdc); |
0ddebc0c | 6279 | err_out_args: |
859c31df | 6280 | rbd_spec_put(spec); |
d147543d | 6281 | kfree(rbd_opts); |
dd5ac32d | 6282 | goto out; |
602adf40 YS |
6283 | } |
6284 | ||
9b60e70b ID |
6285 | static ssize_t rbd_add(struct bus_type *bus, |
6286 | const char *buf, | |
6287 | size_t count) | |
6288 | { | |
6289 | if (single_major) | |
6290 | return -EINVAL; | |
6291 | ||
6292 | return do_rbd_add(bus, buf, count); | |
6293 | } | |
6294 | ||
6295 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
6296 | const char *buf, | |
6297 | size_t count) | |
6298 | { | |
6299 | return do_rbd_add(bus, buf, count); | |
6300 | } | |
6301 | ||
dd5ac32d | 6302 | static void rbd_dev_device_release(struct rbd_device *rbd_dev) |
602adf40 | 6303 | { |
602adf40 | 6304 | rbd_free_disk(rbd_dev); |
1643dfa4 ID |
6305 | |
6306 | spin_lock(&rbd_dev_list_lock); | |
6307 | list_del_init(&rbd_dev->node); | |
6308 | spin_unlock(&rbd_dev_list_lock); | |
6309 | ||
200a6a8b | 6310 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
dd5ac32d | 6311 | device_del(&rbd_dev->dev); |
6d80b130 | 6312 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
6313 | if (!single_major) |
6314 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
602adf40 YS |
6315 | } |
6316 | ||
05a46afd AE |
6317 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
6318 | { | |
ad945fc1 | 6319 | while (rbd_dev->parent) { |
05a46afd AE |
6320 | struct rbd_device *first = rbd_dev; |
6321 | struct rbd_device *second = first->parent; | |
6322 | struct rbd_device *third; | |
6323 | ||
6324 | /* | |
6325 | * Follow to the parent with no grandparent and | |
6326 | * remove it. | |
6327 | */ | |
6328 | while (second && (third = second->parent)) { | |
6329 | first = second; | |
6330 | second = third; | |
6331 | } | |
ad945fc1 | 6332 | rbd_assert(second); |
8ad42cd0 | 6333 | rbd_dev_image_release(second); |
ad945fc1 AE |
6334 | first->parent = NULL; |
6335 | first->parent_overlap = 0; | |
6336 | ||
6337 | rbd_assert(first->parent_spec); | |
05a46afd AE |
6338 | rbd_spec_put(first->parent_spec); |
6339 | first->parent_spec = NULL; | |
05a46afd AE |
6340 | } |
6341 | } | |
6342 | ||
9b60e70b ID |
6343 | static ssize_t do_rbd_remove(struct bus_type *bus, |
6344 | const char *buf, | |
6345 | size_t count) | |
602adf40 YS |
6346 | { |
6347 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
6348 | struct list_head *tmp; |
6349 | int dev_id; | |
0276dca6 | 6350 | char opt_buf[6]; |
82a442d2 | 6351 | bool already = false; |
0276dca6 | 6352 | bool force = false; |
0d8189e1 | 6353 | int ret; |
602adf40 | 6354 | |
0276dca6 MC |
6355 | dev_id = -1; |
6356 | opt_buf[0] = '\0'; | |
6357 | sscanf(buf, "%d %5s", &dev_id, opt_buf); | |
6358 | if (dev_id < 0) { | |
6359 | pr_err("dev_id out of range\n"); | |
602adf40 | 6360 | return -EINVAL; |
0276dca6 MC |
6361 | } |
6362 | if (opt_buf[0] != '\0') { | |
6363 | if (!strcmp(opt_buf, "force")) { | |
6364 | force = true; | |
6365 | } else { | |
6366 | pr_err("bad remove option at '%s'\n", opt_buf); | |
6367 | return -EINVAL; | |
6368 | } | |
6369 | } | |
602adf40 | 6370 | |
751cc0e3 AE |
6371 | ret = -ENOENT; |
6372 | spin_lock(&rbd_dev_list_lock); | |
6373 | list_for_each(tmp, &rbd_dev_list) { | |
6374 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
6375 | if (rbd_dev->dev_id == dev_id) { | |
6376 | ret = 0; | |
6377 | break; | |
6378 | } | |
42382b70 | 6379 | } |
751cc0e3 AE |
6380 | if (!ret) { |
6381 | spin_lock_irq(&rbd_dev->lock); | |
0276dca6 | 6382 | if (rbd_dev->open_count && !force) |
751cc0e3 AE |
6383 | ret = -EBUSY; |
6384 | else | |
82a442d2 AE |
6385 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
6386 | &rbd_dev->flags); | |
751cc0e3 AE |
6387 | spin_unlock_irq(&rbd_dev->lock); |
6388 | } | |
6389 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 6390 | if (ret < 0 || already) |
1ba0f1e7 | 6391 | return ret; |
751cc0e3 | 6392 | |
0276dca6 MC |
6393 | if (force) { |
6394 | /* | |
6395 | * Prevent new IO from being queued and wait for existing | |
6396 | * IO to complete/fail. | |
6397 | */ | |
6398 | blk_mq_freeze_queue(rbd_dev->disk->queue); | |
6399 | blk_set_queue_dying(rbd_dev->disk->queue); | |
6400 | } | |
6401 | ||
ed95b21a ID |
6402 | down_write(&rbd_dev->lock_rwsem); |
6403 | if (__rbd_is_lock_owner(rbd_dev)) | |
6404 | rbd_unlock(rbd_dev); | |
6405 | up_write(&rbd_dev->lock_rwsem); | |
99d16943 | 6406 | rbd_unregister_watch(rbd_dev); |
fca27065 | 6407 | |
9875201e JD |
6408 | /* |
6409 | * Don't free anything from rbd_dev->disk until after all | |
6410 | * notifies are completely processed. Otherwise | |
6411 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
6412 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
6413 | */ | |
dd5ac32d | 6414 | rbd_dev_device_release(rbd_dev); |
8ad42cd0 | 6415 | rbd_dev_image_release(rbd_dev); |
aafb230e | 6416 | |
1ba0f1e7 | 6417 | return count; |
602adf40 YS |
6418 | } |
6419 | ||
9b60e70b ID |
6420 | static ssize_t rbd_remove(struct bus_type *bus, |
6421 | const char *buf, | |
6422 | size_t count) | |
6423 | { | |
6424 | if (single_major) | |
6425 | return -EINVAL; | |
6426 | ||
6427 | return do_rbd_remove(bus, buf, count); | |
6428 | } | |
6429 | ||
6430 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
6431 | const char *buf, | |
6432 | size_t count) | |
6433 | { | |
6434 | return do_rbd_remove(bus, buf, count); | |
6435 | } | |
6436 | ||
602adf40 YS |
6437 | /* |
6438 | * create control files in sysfs | |
dfc5606d | 6439 | * /sys/bus/rbd/... |
602adf40 YS |
6440 | */ |
6441 | static int rbd_sysfs_init(void) | |
6442 | { | |
dfc5606d | 6443 | int ret; |
602adf40 | 6444 | |
fed4c143 | 6445 | ret = device_register(&rbd_root_dev); |
21079786 | 6446 | if (ret < 0) |
dfc5606d | 6447 | return ret; |
602adf40 | 6448 | |
fed4c143 AE |
6449 | ret = bus_register(&rbd_bus_type); |
6450 | if (ret < 0) | |
6451 | device_unregister(&rbd_root_dev); | |
602adf40 | 6452 | |
602adf40 YS |
6453 | return ret; |
6454 | } | |
6455 | ||
6456 | static void rbd_sysfs_cleanup(void) | |
6457 | { | |
dfc5606d | 6458 | bus_unregister(&rbd_bus_type); |
fed4c143 | 6459 | device_unregister(&rbd_root_dev); |
602adf40 YS |
6460 | } |
6461 | ||
1c2a9dfe AE |
6462 | static int rbd_slab_init(void) |
6463 | { | |
6464 | rbd_assert(!rbd_img_request_cache); | |
03d94406 | 6465 | rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0); |
868311b1 AE |
6466 | if (!rbd_img_request_cache) |
6467 | return -ENOMEM; | |
6468 | ||
6469 | rbd_assert(!rbd_obj_request_cache); | |
03d94406 | 6470 | rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0); |
78c2a44a AE |
6471 | if (!rbd_obj_request_cache) |
6472 | goto out_err; | |
6473 | ||
6474 | rbd_assert(!rbd_segment_name_cache); | |
6475 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 6476 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 6477 | if (rbd_segment_name_cache) |
1c2a9dfe | 6478 | return 0; |
78c2a44a | 6479 | out_err: |
13bf2834 JL |
6480 | kmem_cache_destroy(rbd_obj_request_cache); |
6481 | rbd_obj_request_cache = NULL; | |
1c2a9dfe | 6482 | |
868311b1 AE |
6483 | kmem_cache_destroy(rbd_img_request_cache); |
6484 | rbd_img_request_cache = NULL; | |
6485 | ||
1c2a9dfe AE |
6486 | return -ENOMEM; |
6487 | } | |
6488 | ||
6489 | static void rbd_slab_exit(void) | |
6490 | { | |
78c2a44a AE |
6491 | rbd_assert(rbd_segment_name_cache); |
6492 | kmem_cache_destroy(rbd_segment_name_cache); | |
6493 | rbd_segment_name_cache = NULL; | |
6494 | ||
868311b1 AE |
6495 | rbd_assert(rbd_obj_request_cache); |
6496 | kmem_cache_destroy(rbd_obj_request_cache); | |
6497 | rbd_obj_request_cache = NULL; | |
6498 | ||
1c2a9dfe AE |
6499 | rbd_assert(rbd_img_request_cache); |
6500 | kmem_cache_destroy(rbd_img_request_cache); | |
6501 | rbd_img_request_cache = NULL; | |
6502 | } | |
6503 | ||
cc344fa1 | 6504 | static int __init rbd_init(void) |
602adf40 YS |
6505 | { |
6506 | int rc; | |
6507 | ||
1e32d34c AE |
6508 | if (!libceph_compatible(NULL)) { |
6509 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
6510 | return -EINVAL; |
6511 | } | |
e1b4d96d | 6512 | |
1c2a9dfe | 6513 | rc = rbd_slab_init(); |
602adf40 YS |
6514 | if (rc) |
6515 | return rc; | |
e1b4d96d | 6516 | |
f5ee37bd ID |
6517 | /* |
6518 | * The number of active work items is limited by the number of | |
f77303bd | 6519 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
6520 | */ |
6521 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
6522 | if (!rbd_wq) { | |
6523 | rc = -ENOMEM; | |
6524 | goto err_out_slab; | |
6525 | } | |
6526 | ||
9b60e70b ID |
6527 | if (single_major) { |
6528 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
6529 | if (rbd_major < 0) { | |
6530 | rc = rbd_major; | |
f5ee37bd | 6531 | goto err_out_wq; |
9b60e70b ID |
6532 | } |
6533 | } | |
6534 | ||
1c2a9dfe AE |
6535 | rc = rbd_sysfs_init(); |
6536 | if (rc) | |
9b60e70b ID |
6537 | goto err_out_blkdev; |
6538 | ||
6539 | if (single_major) | |
6540 | pr_info("loaded (major %d)\n", rbd_major); | |
6541 | else | |
6542 | pr_info("loaded\n"); | |
1c2a9dfe | 6543 | |
e1b4d96d ID |
6544 | return 0; |
6545 | ||
9b60e70b ID |
6546 | err_out_blkdev: |
6547 | if (single_major) | |
6548 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
6549 | err_out_wq: |
6550 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
6551 | err_out_slab: |
6552 | rbd_slab_exit(); | |
1c2a9dfe | 6553 | return rc; |
602adf40 YS |
6554 | } |
6555 | ||
cc344fa1 | 6556 | static void __exit rbd_exit(void) |
602adf40 | 6557 | { |
ffe312cf | 6558 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 6559 | rbd_sysfs_cleanup(); |
9b60e70b ID |
6560 | if (single_major) |
6561 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 6562 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 6563 | rbd_slab_exit(); |
602adf40 YS |
6564 | } |
6565 | ||
6566 | module_init(rbd_init); | |
6567 | module_exit(rbd_exit); | |
6568 | ||
d552c619 | 6569 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
6570 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
6571 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
6572 | /* following authorship retained from original osdblk.c */ |
6573 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
6574 | ||
90da258b | 6575 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 6576 | MODULE_LICENSE("GPL"); |