]>
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> | |
34 | #include <linux/ceph/decode.h> | |
59c2be1e | 35 | #include <linux/parser.h> |
30d1cff8 | 36 | #include <linux/bsearch.h> |
602adf40 YS |
37 | |
38 | #include <linux/kernel.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/module.h> | |
41 | #include <linux/fs.h> | |
42 | #include <linux/blkdev.h> | |
1c2a9dfe | 43 | #include <linux/slab.h> |
602adf40 YS |
44 | |
45 | #include "rbd_types.h" | |
46 | ||
aafb230e AE |
47 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
48 | ||
593a9e7b AE |
49 | /* |
50 | * The basic unit of block I/O is a sector. It is interpreted in a | |
51 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
52 | * universally 512 bytes. These symbols are just slightly more | |
53 | * meaningful than the bare numbers they represent. | |
54 | */ | |
55 | #define SECTOR_SHIFT 9 | |
56 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
57 | ||
a2acd00e AE |
58 | /* |
59 | * Increment the given counter and return its updated value. | |
60 | * If the counter is already 0 it will not be incremented. | |
61 | * If the counter is already at its maximum value returns | |
62 | * -EINVAL without updating it. | |
63 | */ | |
64 | static int atomic_inc_return_safe(atomic_t *v) | |
65 | { | |
66 | unsigned int counter; | |
67 | ||
68 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
69 | if (counter <= (unsigned int)INT_MAX) | |
70 | return (int)counter; | |
71 | ||
72 | atomic_dec(v); | |
73 | ||
74 | return -EINVAL; | |
75 | } | |
76 | ||
77 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
78 | static int atomic_dec_return_safe(atomic_t *v) | |
79 | { | |
80 | int counter; | |
81 | ||
82 | counter = atomic_dec_return(v); | |
83 | if (counter >= 0) | |
84 | return counter; | |
85 | ||
86 | atomic_inc(v); | |
87 | ||
88 | return -EINVAL; | |
89 | } | |
90 | ||
f0f8cef5 AE |
91 | #define RBD_DRV_NAME "rbd" |
92 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
93 | |
94 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
95 | ||
d4b125e9 AE |
96 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
97 | #define RBD_MAX_SNAP_NAME_LEN \ | |
98 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
99 | ||
35d489f9 | 100 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
101 | |
102 | #define RBD_SNAP_HEAD_NAME "-" | |
103 | ||
9682fc6d AE |
104 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
105 | ||
9e15b77d AE |
106 | /* This allows a single page to hold an image name sent by OSD */ |
107 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 108 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 109 | |
1e130199 | 110 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 111 | |
d889140c AE |
112 | /* Feature bits */ |
113 | ||
5cbf6f12 AE |
114 | #define RBD_FEATURE_LAYERING (1<<0) |
115 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
116 | #define RBD_FEATURES_ALL \ | |
117 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
118 | |
119 | /* Features supported by this (client software) implementation. */ | |
120 | ||
770eba6e | 121 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 122 | |
81a89793 AE |
123 | /* |
124 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
125 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
126 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
127 | * enough to hold all possible device names. | |
128 | */ | |
602adf40 | 129 | #define DEV_NAME_LEN 32 |
81a89793 | 130 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
131 | |
132 | /* | |
133 | * block device image metadata (in-memory version) | |
134 | */ | |
135 | struct rbd_image_header { | |
f35a4dee | 136 | /* These six fields never change for a given rbd image */ |
849b4260 | 137 | char *object_prefix; |
602adf40 YS |
138 | __u8 obj_order; |
139 | __u8 crypt_type; | |
140 | __u8 comp_type; | |
f35a4dee AE |
141 | u64 stripe_unit; |
142 | u64 stripe_count; | |
143 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 144 | |
f84344f3 AE |
145 | /* The remaining fields need to be updated occasionally */ |
146 | u64 image_size; | |
147 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
148 | char *snap_names; /* format 1 only */ |
149 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
150 | }; |
151 | ||
0d7dbfce AE |
152 | /* |
153 | * An rbd image specification. | |
154 | * | |
155 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
156 | * identify an image. Each rbd_dev structure includes a pointer to |
157 | * an rbd_spec structure that encapsulates this identity. | |
158 | * | |
159 | * Each of the id's in an rbd_spec has an associated name. For a | |
160 | * user-mapped image, the names are supplied and the id's associated | |
161 | * with them are looked up. For a layered image, a parent image is | |
162 | * defined by the tuple, and the names are looked up. | |
163 | * | |
164 | * An rbd_dev structure contains a parent_spec pointer which is | |
165 | * non-null if the image it represents is a child in a layered | |
166 | * image. This pointer will refer to the rbd_spec structure used | |
167 | * by the parent rbd_dev for its own identity (i.e., the structure | |
168 | * is shared between the parent and child). | |
169 | * | |
170 | * Since these structures are populated once, during the discovery | |
171 | * phase of image construction, they are effectively immutable so | |
172 | * we make no effort to synchronize access to them. | |
173 | * | |
174 | * Note that code herein does not assume the image name is known (it | |
175 | * could be a null pointer). | |
0d7dbfce AE |
176 | */ |
177 | struct rbd_spec { | |
178 | u64 pool_id; | |
ecb4dc22 | 179 | const char *pool_name; |
0d7dbfce | 180 | |
ecb4dc22 AE |
181 | const char *image_id; |
182 | const char *image_name; | |
0d7dbfce AE |
183 | |
184 | u64 snap_id; | |
ecb4dc22 | 185 | const char *snap_name; |
0d7dbfce AE |
186 | |
187 | struct kref kref; | |
188 | }; | |
189 | ||
602adf40 | 190 | /* |
f0f8cef5 | 191 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
192 | */ |
193 | struct rbd_client { | |
194 | struct ceph_client *client; | |
195 | struct kref kref; | |
196 | struct list_head node; | |
197 | }; | |
198 | ||
bf0d5f50 AE |
199 | struct rbd_img_request; |
200 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
201 | ||
202 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
203 | ||
204 | struct rbd_obj_request; | |
205 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
206 | ||
9969ebc5 AE |
207 | enum obj_request_type { |
208 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
209 | }; | |
bf0d5f50 | 210 | |
926f9b3f AE |
211 | enum obj_req_flags { |
212 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 213 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
214 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
215 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
216 | }; |
217 | ||
bf0d5f50 AE |
218 | struct rbd_obj_request { |
219 | const char *object_name; | |
220 | u64 offset; /* object start byte */ | |
221 | u64 length; /* bytes from offset */ | |
926f9b3f | 222 | unsigned long flags; |
bf0d5f50 | 223 | |
c5b5ef6c AE |
224 | /* |
225 | * An object request associated with an image will have its | |
226 | * img_data flag set; a standalone object request will not. | |
227 | * | |
228 | * A standalone object request will have which == BAD_WHICH | |
229 | * and a null obj_request pointer. | |
230 | * | |
231 | * An object request initiated in support of a layered image | |
232 | * object (to check for its existence before a write) will | |
233 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
234 | * | |
235 | * Finally, an object request for rbd image data will have | |
236 | * which != BAD_WHICH, and will have a non-null img_request | |
237 | * pointer. The value of which will be in the range | |
238 | * 0..(img_request->obj_request_count-1). | |
239 | */ | |
240 | union { | |
241 | struct rbd_obj_request *obj_request; /* STAT op */ | |
242 | struct { | |
243 | struct rbd_img_request *img_request; | |
244 | u64 img_offset; | |
245 | /* links for img_request->obj_requests list */ | |
246 | struct list_head links; | |
247 | }; | |
248 | }; | |
bf0d5f50 AE |
249 | u32 which; /* posn image request list */ |
250 | ||
251 | enum obj_request_type type; | |
788e2df3 AE |
252 | union { |
253 | struct bio *bio_list; | |
254 | struct { | |
255 | struct page **pages; | |
256 | u32 page_count; | |
257 | }; | |
258 | }; | |
0eefd470 | 259 | struct page **copyup_pages; |
ebda6408 | 260 | u32 copyup_page_count; |
bf0d5f50 AE |
261 | |
262 | struct ceph_osd_request *osd_req; | |
263 | ||
264 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 265 | int result; |
bf0d5f50 AE |
266 | |
267 | rbd_obj_callback_t callback; | |
788e2df3 | 268 | struct completion completion; |
bf0d5f50 AE |
269 | |
270 | struct kref kref; | |
271 | }; | |
272 | ||
0c425248 | 273 | enum img_req_flags { |
9849e986 AE |
274 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
275 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 276 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
277 | }; |
278 | ||
bf0d5f50 | 279 | struct rbd_img_request { |
bf0d5f50 AE |
280 | struct rbd_device *rbd_dev; |
281 | u64 offset; /* starting image byte offset */ | |
282 | u64 length; /* byte count from offset */ | |
0c425248 | 283 | unsigned long flags; |
bf0d5f50 | 284 | union { |
9849e986 | 285 | u64 snap_id; /* for reads */ |
bf0d5f50 | 286 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
287 | }; |
288 | union { | |
289 | struct request *rq; /* block request */ | |
290 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 291 | }; |
3d7efd18 | 292 | struct page **copyup_pages; |
ebda6408 | 293 | u32 copyup_page_count; |
bf0d5f50 AE |
294 | spinlock_t completion_lock;/* protects next_completion */ |
295 | u32 next_completion; | |
296 | rbd_img_callback_t callback; | |
55f27e09 | 297 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 298 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
299 | |
300 | u32 obj_request_count; | |
301 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
302 | ||
303 | struct kref kref; | |
304 | }; | |
305 | ||
306 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 307 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 308 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 309 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 310 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 311 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 312 | |
f84344f3 | 313 | struct rbd_mapping { |
99c1f08f | 314 | u64 size; |
34b13184 | 315 | u64 features; |
f84344f3 AE |
316 | bool read_only; |
317 | }; | |
318 | ||
602adf40 YS |
319 | /* |
320 | * a single device | |
321 | */ | |
322 | struct rbd_device { | |
de71a297 | 323 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
324 | |
325 | int major; /* blkdev assigned major */ | |
326 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 327 | |
a30b71b9 | 328 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
329 | struct rbd_client *rbd_client; |
330 | ||
331 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
332 | ||
b82d167b | 333 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
334 | |
335 | struct rbd_image_header header; | |
b82d167b | 336 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 337 | struct rbd_spec *spec; |
602adf40 | 338 | |
0d7dbfce | 339 | char *header_name; |
971f839a | 340 | |
0903e875 AE |
341 | struct ceph_file_layout layout; |
342 | ||
59c2be1e | 343 | struct ceph_osd_event *watch_event; |
975241af | 344 | struct rbd_obj_request *watch_request; |
59c2be1e | 345 | |
86b00e0d AE |
346 | struct rbd_spec *parent_spec; |
347 | u64 parent_overlap; | |
a2acd00e | 348 | atomic_t parent_ref; |
2f82ee54 | 349 | struct rbd_device *parent; |
86b00e0d | 350 | |
c666601a JD |
351 | /* protects updating the header */ |
352 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
353 | |
354 | struct rbd_mapping mapping; | |
602adf40 YS |
355 | |
356 | struct list_head node; | |
dfc5606d | 357 | |
dfc5606d YS |
358 | /* sysfs related */ |
359 | struct device dev; | |
b82d167b | 360 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
361 | }; |
362 | ||
b82d167b AE |
363 | /* |
364 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
365 | * rbd_dev->lock is used to protect access. | |
366 | * | |
367 | * Currently, only the "removing" flag (which is coupled with the | |
368 | * "open_count" field) requires atomic access. | |
369 | */ | |
6d292906 AE |
370 | enum rbd_dev_flags { |
371 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 372 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
373 | }; |
374 | ||
602adf40 | 375 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 376 | |
602adf40 | 377 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
378 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
379 | ||
432b8587 AE |
380 | static LIST_HEAD(rbd_client_list); /* clients */ |
381 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 382 | |
78c2a44a AE |
383 | /* Slab caches for frequently-allocated structures */ |
384 | ||
1c2a9dfe | 385 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 386 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 387 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 388 | |
3d7efd18 AE |
389 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
390 | ||
200a6a8b | 391 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 392 | |
f0f8cef5 AE |
393 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
394 | size_t count); | |
395 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
396 | size_t count); | |
1f3ef788 | 397 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping); |
a2acd00e | 398 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 AE |
399 | |
400 | static struct bus_attribute rbd_bus_attrs[] = { | |
401 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
402 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
403 | __ATTR_NULL | |
404 | }; | |
405 | ||
406 | static struct bus_type rbd_bus_type = { | |
407 | .name = "rbd", | |
408 | .bus_attrs = rbd_bus_attrs, | |
409 | }; | |
410 | ||
411 | static void rbd_root_dev_release(struct device *dev) | |
412 | { | |
413 | } | |
414 | ||
415 | static struct device rbd_root_dev = { | |
416 | .init_name = "rbd", | |
417 | .release = rbd_root_dev_release, | |
418 | }; | |
419 | ||
06ecc6cb AE |
420 | static __printf(2, 3) |
421 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
422 | { | |
423 | struct va_format vaf; | |
424 | va_list args; | |
425 | ||
426 | va_start(args, fmt); | |
427 | vaf.fmt = fmt; | |
428 | vaf.va = &args; | |
429 | ||
430 | if (!rbd_dev) | |
431 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
432 | else if (rbd_dev->disk) | |
433 | printk(KERN_WARNING "%s: %s: %pV\n", | |
434 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
435 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
436 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
437 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
438 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
439 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
440 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
441 | else /* punt */ | |
442 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
443 | RBD_DRV_NAME, rbd_dev, &vaf); | |
444 | va_end(args); | |
445 | } | |
446 | ||
aafb230e AE |
447 | #ifdef RBD_DEBUG |
448 | #define rbd_assert(expr) \ | |
449 | if (unlikely(!(expr))) { \ | |
450 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
451 | "at line %d:\n\n" \ | |
452 | "\trbd_assert(%s);\n\n", \ | |
453 | __func__, __LINE__, #expr); \ | |
454 | BUG(); \ | |
455 | } | |
456 | #else /* !RBD_DEBUG */ | |
457 | # define rbd_assert(expr) ((void) 0) | |
458 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 459 | |
b454e36d | 460 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
461 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
462 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 463 | |
cc4a38bd | 464 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 AE |
465 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
466 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev); | |
54cac61f AE |
467 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
468 | u64 snap_id); | |
2ad3d716 AE |
469 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
470 | u8 *order, u64 *snap_size); | |
471 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
472 | u64 *snap_features); | |
473 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 474 | |
602adf40 YS |
475 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
476 | { | |
f0f8cef5 | 477 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 478 | bool removing = false; |
602adf40 | 479 | |
f84344f3 | 480 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
481 | return -EROFS; |
482 | ||
a14ea269 | 483 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
484 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
485 | removing = true; | |
486 | else | |
487 | rbd_dev->open_count++; | |
a14ea269 | 488 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
489 | if (removing) |
490 | return -ENOENT; | |
491 | ||
42382b70 | 492 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 493 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 494 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 495 | mutex_unlock(&ctl_mutex); |
340c7a2b | 496 | |
602adf40 YS |
497 | return 0; |
498 | } | |
499 | ||
dfc5606d YS |
500 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
501 | { | |
502 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
503 | unsigned long open_count_before; |
504 | ||
a14ea269 | 505 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 506 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 507 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 508 | rbd_assert(open_count_before > 0); |
dfc5606d | 509 | |
42382b70 | 510 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 511 | put_device(&rbd_dev->dev); |
42382b70 | 512 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
513 | |
514 | return 0; | |
515 | } | |
516 | ||
602adf40 YS |
517 | static const struct block_device_operations rbd_bd_ops = { |
518 | .owner = THIS_MODULE, | |
519 | .open = rbd_open, | |
dfc5606d | 520 | .release = rbd_release, |
602adf40 YS |
521 | }; |
522 | ||
523 | /* | |
7262cfca AE |
524 | * Initialize an rbd client instance. Success or not, this function |
525 | * consumes ceph_opts. | |
602adf40 | 526 | */ |
f8c38929 | 527 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
528 | { |
529 | struct rbd_client *rbdc; | |
530 | int ret = -ENOMEM; | |
531 | ||
37206ee5 | 532 | dout("%s:\n", __func__); |
602adf40 YS |
533 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
534 | if (!rbdc) | |
535 | goto out_opt; | |
536 | ||
537 | kref_init(&rbdc->kref); | |
538 | INIT_LIST_HEAD(&rbdc->node); | |
539 | ||
bc534d86 AE |
540 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
541 | ||
43ae4701 | 542 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 543 | if (IS_ERR(rbdc->client)) |
bc534d86 | 544 | goto out_mutex; |
43ae4701 | 545 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
546 | |
547 | ret = ceph_open_session(rbdc->client); | |
548 | if (ret < 0) | |
549 | goto out_err; | |
550 | ||
432b8587 | 551 | spin_lock(&rbd_client_list_lock); |
602adf40 | 552 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 553 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 554 | |
bc534d86 | 555 | mutex_unlock(&ctl_mutex); |
37206ee5 | 556 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 557 | |
602adf40 YS |
558 | return rbdc; |
559 | ||
560 | out_err: | |
561 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
562 | out_mutex: |
563 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
564 | kfree(rbdc); |
565 | out_opt: | |
43ae4701 AE |
566 | if (ceph_opts) |
567 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
568 | dout("%s: error %d\n", __func__, ret); |
569 | ||
28f259b7 | 570 | return ERR_PTR(ret); |
602adf40 YS |
571 | } |
572 | ||
2f82ee54 AE |
573 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
574 | { | |
575 | kref_get(&rbdc->kref); | |
576 | ||
577 | return rbdc; | |
578 | } | |
579 | ||
602adf40 | 580 | /* |
1f7ba331 AE |
581 | * Find a ceph client with specific addr and configuration. If |
582 | * found, bump its reference count. | |
602adf40 | 583 | */ |
1f7ba331 | 584 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
585 | { |
586 | struct rbd_client *client_node; | |
1f7ba331 | 587 | bool found = false; |
602adf40 | 588 | |
43ae4701 | 589 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
590 | return NULL; |
591 | ||
1f7ba331 AE |
592 | spin_lock(&rbd_client_list_lock); |
593 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
594 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
595 | __rbd_get_client(client_node); |
596 | ||
1f7ba331 AE |
597 | found = true; |
598 | break; | |
599 | } | |
600 | } | |
601 | spin_unlock(&rbd_client_list_lock); | |
602 | ||
603 | return found ? client_node : NULL; | |
602adf40 YS |
604 | } |
605 | ||
59c2be1e YS |
606 | /* |
607 | * mount options | |
608 | */ | |
609 | enum { | |
59c2be1e YS |
610 | Opt_last_int, |
611 | /* int args above */ | |
612 | Opt_last_string, | |
613 | /* string args above */ | |
cc0538b6 AE |
614 | Opt_read_only, |
615 | Opt_read_write, | |
616 | /* Boolean args above */ | |
617 | Opt_last_bool, | |
59c2be1e YS |
618 | }; |
619 | ||
43ae4701 | 620 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
621 | /* int args above */ |
622 | /* string args above */ | |
be466c1c | 623 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
624 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
625 | {Opt_read_write, "read_write"}, | |
626 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
627 | /* Boolean args above */ | |
59c2be1e YS |
628 | {-1, NULL} |
629 | }; | |
630 | ||
98571b5a AE |
631 | struct rbd_options { |
632 | bool read_only; | |
633 | }; | |
634 | ||
635 | #define RBD_READ_ONLY_DEFAULT false | |
636 | ||
59c2be1e YS |
637 | static int parse_rbd_opts_token(char *c, void *private) |
638 | { | |
43ae4701 | 639 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
640 | substring_t argstr[MAX_OPT_ARGS]; |
641 | int token, intval, ret; | |
642 | ||
43ae4701 | 643 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
644 | if (token < 0) |
645 | return -EINVAL; | |
646 | ||
647 | if (token < Opt_last_int) { | |
648 | ret = match_int(&argstr[0], &intval); | |
649 | if (ret < 0) { | |
650 | pr_err("bad mount option arg (not int) " | |
651 | "at '%s'\n", c); | |
652 | return ret; | |
653 | } | |
654 | dout("got int token %d val %d\n", token, intval); | |
655 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
656 | dout("got string token %d val %s\n", token, | |
657 | argstr[0].from); | |
cc0538b6 AE |
658 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
659 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
660 | } else { |
661 | dout("got token %d\n", token); | |
662 | } | |
663 | ||
664 | switch (token) { | |
cc0538b6 AE |
665 | case Opt_read_only: |
666 | rbd_opts->read_only = true; | |
667 | break; | |
668 | case Opt_read_write: | |
669 | rbd_opts->read_only = false; | |
670 | break; | |
59c2be1e | 671 | default: |
aafb230e AE |
672 | rbd_assert(false); |
673 | break; | |
59c2be1e YS |
674 | } |
675 | return 0; | |
676 | } | |
677 | ||
602adf40 YS |
678 | /* |
679 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
680 | * not exist create it. Either way, ceph_opts is consumed by this |
681 | * function. | |
602adf40 | 682 | */ |
9d3997fd | 683 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 684 | { |
f8c38929 | 685 | struct rbd_client *rbdc; |
59c2be1e | 686 | |
1f7ba331 | 687 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 688 | if (rbdc) /* using an existing client */ |
43ae4701 | 689 | ceph_destroy_options(ceph_opts); |
9d3997fd | 690 | else |
f8c38929 | 691 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 692 | |
9d3997fd | 693 | return rbdc; |
602adf40 YS |
694 | } |
695 | ||
696 | /* | |
697 | * Destroy ceph client | |
d23a4b3f | 698 | * |
432b8587 | 699 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
700 | */ |
701 | static void rbd_client_release(struct kref *kref) | |
702 | { | |
703 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
704 | ||
37206ee5 | 705 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 706 | spin_lock(&rbd_client_list_lock); |
602adf40 | 707 | list_del(&rbdc->node); |
cd9d9f5d | 708 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
709 | |
710 | ceph_destroy_client(rbdc->client); | |
711 | kfree(rbdc); | |
712 | } | |
713 | ||
714 | /* | |
715 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
716 | * it. | |
717 | */ | |
9d3997fd | 718 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 719 | { |
c53d5893 AE |
720 | if (rbdc) |
721 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
722 | } |
723 | ||
a30b71b9 AE |
724 | static bool rbd_image_format_valid(u32 image_format) |
725 | { | |
726 | return image_format == 1 || image_format == 2; | |
727 | } | |
728 | ||
8e94af8e AE |
729 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
730 | { | |
103a150f AE |
731 | size_t size; |
732 | u32 snap_count; | |
733 | ||
734 | /* The header has to start with the magic rbd header text */ | |
735 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
736 | return false; | |
737 | ||
db2388b6 AE |
738 | /* The bio layer requires at least sector-sized I/O */ |
739 | ||
740 | if (ondisk->options.order < SECTOR_SHIFT) | |
741 | return false; | |
742 | ||
743 | /* If we use u64 in a few spots we may be able to loosen this */ | |
744 | ||
745 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
746 | return false; | |
747 | ||
103a150f AE |
748 | /* |
749 | * The size of a snapshot header has to fit in a size_t, and | |
750 | * that limits the number of snapshots. | |
751 | */ | |
752 | snap_count = le32_to_cpu(ondisk->snap_count); | |
753 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
754 | if (snap_count > size / sizeof (__le64)) | |
755 | return false; | |
756 | ||
757 | /* | |
758 | * Not only that, but the size of the entire the snapshot | |
759 | * header must also be representable in a size_t. | |
760 | */ | |
761 | size -= snap_count * sizeof (__le64); | |
762 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
763 | return false; | |
764 | ||
765 | return true; | |
8e94af8e AE |
766 | } |
767 | ||
602adf40 | 768 | /* |
bb23e37a AE |
769 | * Fill an rbd image header with information from the given format 1 |
770 | * on-disk header. | |
602adf40 | 771 | */ |
662518b1 | 772 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 773 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 774 | { |
662518b1 | 775 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
776 | bool first_time = header->object_prefix == NULL; |
777 | struct ceph_snap_context *snapc; | |
778 | char *object_prefix = NULL; | |
779 | char *snap_names = NULL; | |
780 | u64 *snap_sizes = NULL; | |
ccece235 | 781 | u32 snap_count; |
d2bb24e5 | 782 | size_t size; |
bb23e37a | 783 | int ret = -ENOMEM; |
621901d6 | 784 | u32 i; |
602adf40 | 785 | |
bb23e37a AE |
786 | /* Allocate this now to avoid having to handle failure below */ |
787 | ||
788 | if (first_time) { | |
789 | size_t len; | |
790 | ||
791 | len = strnlen(ondisk->object_prefix, | |
792 | sizeof (ondisk->object_prefix)); | |
793 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
794 | if (!object_prefix) | |
795 | return -ENOMEM; | |
796 | memcpy(object_prefix, ondisk->object_prefix, len); | |
797 | object_prefix[len] = '\0'; | |
798 | } | |
103a150f | 799 | |
bb23e37a | 800 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 801 | |
bb23e37a AE |
802 | snap_count = le32_to_cpu(ondisk->snap_count); |
803 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
804 | if (!snapc) | |
805 | goto out_err; | |
806 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 807 | if (snap_count) { |
bb23e37a | 808 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
809 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
810 | ||
bb23e37a | 811 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 812 | |
bb23e37a AE |
813 | if (snap_names_len > (u64)SIZE_MAX) |
814 | goto out_2big; | |
815 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
816 | if (!snap_names) | |
6a52325f AE |
817 | goto out_err; |
818 | ||
bb23e37a | 819 | /* ...as well as the array of their sizes. */ |
621901d6 | 820 | |
d2bb24e5 | 821 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
822 | snap_sizes = kmalloc(size, GFP_KERNEL); |
823 | if (!snap_sizes) | |
6a52325f | 824 | goto out_err; |
bb23e37a AE |
825 | |
826 | /* | |
827 | * Copy the names, and fill in each snapshot's id | |
828 | * and size. | |
829 | * | |
99a41ebc | 830 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a AE |
831 | * ondisk buffer we're working with has |
832 | * snap_names_len bytes beyond the end of the | |
833 | * snapshot id array, this memcpy() is safe. | |
834 | */ | |
835 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); | |
836 | snaps = ondisk->snaps; | |
837 | for (i = 0; i < snap_count; i++) { | |
838 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
839 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
840 | } | |
602adf40 | 841 | } |
849b4260 | 842 | |
bb23e37a AE |
843 | /* We won't fail any more, fill in the header */ |
844 | ||
662518b1 | 845 | down_write(&rbd_dev->header_rwsem); |
bb23e37a AE |
846 | if (first_time) { |
847 | header->object_prefix = object_prefix; | |
848 | header->obj_order = ondisk->options.order; | |
849 | header->crypt_type = ondisk->options.crypt_type; | |
850 | header->comp_type = ondisk->options.comp_type; | |
851 | /* The rest aren't used for format 1 images */ | |
852 | header->stripe_unit = 0; | |
853 | header->stripe_count = 0; | |
854 | header->features = 0; | |
662518b1 AE |
855 | } else { |
856 | ceph_put_snap_context(header->snapc); | |
857 | kfree(header->snap_names); | |
858 | kfree(header->snap_sizes); | |
bb23e37a | 859 | } |
6a52325f | 860 | |
bb23e37a | 861 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 862 | |
f84344f3 | 863 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
864 | header->snapc = snapc; |
865 | header->snap_names = snap_names; | |
866 | header->snap_sizes = snap_sizes; | |
602adf40 | 867 | |
662518b1 AE |
868 | /* Make sure mapping size is consistent with header info */ |
869 | ||
870 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time) | |
871 | if (rbd_dev->mapping.size != header->image_size) | |
872 | rbd_dev->mapping.size = header->image_size; | |
873 | ||
874 | up_write(&rbd_dev->header_rwsem); | |
875 | ||
602adf40 | 876 | return 0; |
bb23e37a AE |
877 | out_2big: |
878 | ret = -EIO; | |
6a52325f | 879 | out_err: |
bb23e37a AE |
880 | kfree(snap_sizes); |
881 | kfree(snap_names); | |
882 | ceph_put_snap_context(snapc); | |
883 | kfree(object_prefix); | |
884 | ||
885 | return ret; | |
602adf40 YS |
886 | } |
887 | ||
9682fc6d AE |
888 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
889 | { | |
890 | const char *snap_name; | |
891 | ||
892 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
893 | ||
894 | /* Skip over names until we find the one we are looking for */ | |
895 | ||
896 | snap_name = rbd_dev->header.snap_names; | |
897 | while (which--) | |
898 | snap_name += strlen(snap_name) + 1; | |
899 | ||
900 | return kstrdup(snap_name, GFP_KERNEL); | |
901 | } | |
902 | ||
30d1cff8 AE |
903 | /* |
904 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
905 | * Note that result is for snapshots in *descending* order. | |
906 | */ | |
907 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
908 | { | |
909 | u64 snap_id1 = *(u64 *)s1; | |
910 | u64 snap_id2 = *(u64 *)s2; | |
911 | ||
912 | if (snap_id1 < snap_id2) | |
913 | return 1; | |
914 | return snap_id1 == snap_id2 ? 0 : -1; | |
915 | } | |
916 | ||
917 | /* | |
918 | * Search a snapshot context to see if the given snapshot id is | |
919 | * present. | |
920 | * | |
921 | * Returns the position of the snapshot id in the array if it's found, | |
922 | * or BAD_SNAP_INDEX otherwise. | |
923 | * | |
924 | * Note: The snapshot array is in kept sorted (by the osd) in | |
925 | * reverse order, highest snapshot id first. | |
926 | */ | |
9682fc6d AE |
927 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
928 | { | |
929 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 930 | u64 *found; |
9682fc6d | 931 | |
30d1cff8 AE |
932 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
933 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 934 | |
30d1cff8 | 935 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
936 | } |
937 | ||
2ad3d716 AE |
938 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
939 | u64 snap_id) | |
9e15b77d | 940 | { |
54cac61f | 941 | u32 which; |
9e15b77d | 942 | |
54cac61f AE |
943 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
944 | if (which == BAD_SNAP_INDEX) | |
945 | return NULL; | |
946 | ||
947 | return _rbd_dev_v1_snap_name(rbd_dev, which); | |
948 | } | |
949 | ||
950 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
951 | { | |
9e15b77d AE |
952 | if (snap_id == CEPH_NOSNAP) |
953 | return RBD_SNAP_HEAD_NAME; | |
954 | ||
54cac61f AE |
955 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
956 | if (rbd_dev->image_format == 1) | |
957 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 958 | |
54cac61f | 959 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
960 | } |
961 | ||
2ad3d716 AE |
962 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
963 | u64 *snap_size) | |
602adf40 | 964 | { |
2ad3d716 AE |
965 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
966 | if (snap_id == CEPH_NOSNAP) { | |
967 | *snap_size = rbd_dev->header.image_size; | |
968 | } else if (rbd_dev->image_format == 1) { | |
969 | u32 which; | |
602adf40 | 970 | |
2ad3d716 AE |
971 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
972 | if (which == BAD_SNAP_INDEX) | |
973 | return -ENOENT; | |
e86924a8 | 974 | |
2ad3d716 AE |
975 | *snap_size = rbd_dev->header.snap_sizes[which]; |
976 | } else { | |
977 | u64 size = 0; | |
978 | int ret; | |
979 | ||
980 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
981 | if (ret) | |
982 | return ret; | |
983 | ||
984 | *snap_size = size; | |
985 | } | |
986 | return 0; | |
602adf40 YS |
987 | } |
988 | ||
2ad3d716 AE |
989 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
990 | u64 *snap_features) | |
602adf40 | 991 | { |
2ad3d716 AE |
992 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
993 | if (snap_id == CEPH_NOSNAP) { | |
994 | *snap_features = rbd_dev->header.features; | |
995 | } else if (rbd_dev->image_format == 1) { | |
996 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 997 | } else { |
2ad3d716 AE |
998 | u64 features = 0; |
999 | int ret; | |
8b0241f8 | 1000 | |
2ad3d716 AE |
1001 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1002 | if (ret) | |
1003 | return ret; | |
1004 | ||
1005 | *snap_features = features; | |
1006 | } | |
1007 | return 0; | |
1008 | } | |
1009 | ||
1010 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1011 | { | |
8f4b7d98 | 1012 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1013 | u64 size = 0; |
1014 | u64 features = 0; | |
1015 | int ret; | |
1016 | ||
2ad3d716 AE |
1017 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1018 | if (ret) | |
1019 | return ret; | |
1020 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1021 | if (ret) | |
1022 | return ret; | |
1023 | ||
1024 | rbd_dev->mapping.size = size; | |
1025 | rbd_dev->mapping.features = features; | |
1026 | ||
8b0241f8 | 1027 | return 0; |
602adf40 YS |
1028 | } |
1029 | ||
d1cf5788 AE |
1030 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1031 | { | |
1032 | rbd_dev->mapping.size = 0; | |
1033 | rbd_dev->mapping.features = 0; | |
d1cf5788 AE |
1034 | } |
1035 | ||
98571b5a | 1036 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1037 | { |
65ccfe21 AE |
1038 | char *name; |
1039 | u64 segment; | |
1040 | int ret; | |
602adf40 | 1041 | |
78c2a44a | 1042 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1043 | if (!name) |
1044 | return NULL; | |
1045 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 1046 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 1047 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 1048 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
1049 | pr_err("error formatting segment name for #%llu (%d)\n", |
1050 | segment, ret); | |
1051 | kfree(name); | |
1052 | name = NULL; | |
1053 | } | |
602adf40 | 1054 | |
65ccfe21 AE |
1055 | return name; |
1056 | } | |
602adf40 | 1057 | |
78c2a44a AE |
1058 | static void rbd_segment_name_free(const char *name) |
1059 | { | |
1060 | /* The explicit cast here is needed to drop the const qualifier */ | |
1061 | ||
1062 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1063 | } | |
1064 | ||
65ccfe21 AE |
1065 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1066 | { | |
1067 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1068 | |
65ccfe21 AE |
1069 | return offset & (segment_size - 1); |
1070 | } | |
1071 | ||
1072 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1073 | u64 offset, u64 length) | |
1074 | { | |
1075 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1076 | ||
1077 | offset &= segment_size - 1; | |
1078 | ||
aafb230e | 1079 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1080 | if (offset + length > segment_size) |
1081 | length = segment_size - offset; | |
1082 | ||
1083 | return length; | |
602adf40 YS |
1084 | } |
1085 | ||
029bcbd8 JD |
1086 | /* |
1087 | * returns the size of an object in the image | |
1088 | */ | |
1089 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1090 | { | |
1091 | return 1 << header->obj_order; | |
1092 | } | |
1093 | ||
602adf40 YS |
1094 | /* |
1095 | * bio helpers | |
1096 | */ | |
1097 | ||
1098 | static void bio_chain_put(struct bio *chain) | |
1099 | { | |
1100 | struct bio *tmp; | |
1101 | ||
1102 | while (chain) { | |
1103 | tmp = chain; | |
1104 | chain = chain->bi_next; | |
1105 | bio_put(tmp); | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * zeros a bio chain, starting at specific offset | |
1111 | */ | |
1112 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1113 | { | |
1114 | struct bio_vec *bv; | |
1115 | unsigned long flags; | |
1116 | void *buf; | |
1117 | int i; | |
1118 | int pos = 0; | |
1119 | ||
1120 | while (chain) { | |
1121 | bio_for_each_segment(bv, chain, i) { | |
1122 | if (pos + bv->bv_len > start_ofs) { | |
1123 | int remainder = max(start_ofs - pos, 0); | |
1124 | buf = bvec_kmap_irq(bv, &flags); | |
1125 | memset(buf + remainder, 0, | |
1126 | bv->bv_len - remainder); | |
85b5aaa6 | 1127 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
1128 | } |
1129 | pos += bv->bv_len; | |
1130 | } | |
1131 | ||
1132 | chain = chain->bi_next; | |
1133 | } | |
1134 | } | |
1135 | ||
b9434c5b AE |
1136 | /* |
1137 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1138 | * starting at the given byte offset from the start of the array and | |
1139 | * continuing up to the given end offset. The pages array is | |
1140 | * assumed to be big enough to hold all bytes up to the end. | |
1141 | */ | |
1142 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1143 | { | |
1144 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1145 | ||
1146 | rbd_assert(end > offset); | |
1147 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1148 | while (offset < end) { | |
1149 | size_t page_offset; | |
1150 | size_t length; | |
1151 | unsigned long flags; | |
1152 | void *kaddr; | |
1153 | ||
1154 | page_offset = (size_t)(offset & ~PAGE_MASK); | |
1155 | length = min(PAGE_SIZE - page_offset, (size_t)(end - offset)); | |
1156 | local_irq_save(flags); | |
1157 | kaddr = kmap_atomic(*page); | |
1158 | memset(kaddr + page_offset, 0, length); | |
1159 | kunmap_atomic(kaddr); | |
1160 | local_irq_restore(flags); | |
1161 | ||
1162 | offset += length; | |
1163 | page++; | |
1164 | } | |
1165 | } | |
1166 | ||
602adf40 | 1167 | /* |
f7760dad AE |
1168 | * Clone a portion of a bio, starting at the given byte offset |
1169 | * and continuing for the number of bytes indicated. | |
602adf40 | 1170 | */ |
f7760dad AE |
1171 | static struct bio *bio_clone_range(struct bio *bio_src, |
1172 | unsigned int offset, | |
1173 | unsigned int len, | |
1174 | gfp_t gfpmask) | |
602adf40 | 1175 | { |
f7760dad AE |
1176 | struct bio_vec *bv; |
1177 | unsigned int resid; | |
1178 | unsigned short idx; | |
1179 | unsigned int voff; | |
1180 | unsigned short end_idx; | |
1181 | unsigned short vcnt; | |
1182 | struct bio *bio; | |
1183 | ||
1184 | /* Handle the easy case for the caller */ | |
1185 | ||
1186 | if (!offset && len == bio_src->bi_size) | |
1187 | return bio_clone(bio_src, gfpmask); | |
1188 | ||
1189 | if (WARN_ON_ONCE(!len)) | |
1190 | return NULL; | |
1191 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
1192 | return NULL; | |
1193 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
1194 | return NULL; | |
1195 | ||
1196 | /* Find first affected segment... */ | |
1197 | ||
1198 | resid = offset; | |
1199 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
1200 | if (resid < bv->bv_len) | |
1201 | break; | |
1202 | resid -= bv->bv_len; | |
602adf40 | 1203 | } |
f7760dad | 1204 | voff = resid; |
602adf40 | 1205 | |
f7760dad | 1206 | /* ...and the last affected segment */ |
602adf40 | 1207 | |
f7760dad AE |
1208 | resid += len; |
1209 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
1210 | if (resid <= bv->bv_len) | |
1211 | break; | |
1212 | resid -= bv->bv_len; | |
1213 | } | |
1214 | vcnt = end_idx - idx + 1; | |
1215 | ||
1216 | /* Build the clone */ | |
1217 | ||
1218 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
1219 | if (!bio) | |
1220 | return NULL; /* ENOMEM */ | |
602adf40 | 1221 | |
f7760dad AE |
1222 | bio->bi_bdev = bio_src->bi_bdev; |
1223 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
1224 | bio->bi_rw = bio_src->bi_rw; | |
1225 | bio->bi_flags |= 1 << BIO_CLONED; | |
1226 | ||
1227 | /* | |
1228 | * Copy over our part of the bio_vec, then update the first | |
1229 | * and last (or only) entries. | |
1230 | */ | |
1231 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
1232 | vcnt * sizeof (struct bio_vec)); | |
1233 | bio->bi_io_vec[0].bv_offset += voff; | |
1234 | if (vcnt > 1) { | |
1235 | bio->bi_io_vec[0].bv_len -= voff; | |
1236 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
1237 | } else { | |
1238 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
1239 | } |
1240 | ||
f7760dad AE |
1241 | bio->bi_vcnt = vcnt; |
1242 | bio->bi_size = len; | |
1243 | bio->bi_idx = 0; | |
1244 | ||
1245 | return bio; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Clone a portion of a bio chain, starting at the given byte offset | |
1250 | * into the first bio in the source chain and continuing for the | |
1251 | * number of bytes indicated. The result is another bio chain of | |
1252 | * exactly the given length, or a null pointer on error. | |
1253 | * | |
1254 | * The bio_src and offset parameters are both in-out. On entry they | |
1255 | * refer to the first source bio and the offset into that bio where | |
1256 | * the start of data to be cloned is located. | |
1257 | * | |
1258 | * On return, bio_src is updated to refer to the bio in the source | |
1259 | * chain that contains first un-cloned byte, and *offset will | |
1260 | * contain the offset of that byte within that bio. | |
1261 | */ | |
1262 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1263 | unsigned int *offset, | |
1264 | unsigned int len, | |
1265 | gfp_t gfpmask) | |
1266 | { | |
1267 | struct bio *bi = *bio_src; | |
1268 | unsigned int off = *offset; | |
1269 | struct bio *chain = NULL; | |
1270 | struct bio **end; | |
1271 | ||
1272 | /* Build up a chain of clone bios up to the limit */ | |
1273 | ||
1274 | if (!bi || off >= bi->bi_size || !len) | |
1275 | return NULL; /* Nothing to clone */ | |
602adf40 | 1276 | |
f7760dad AE |
1277 | end = &chain; |
1278 | while (len) { | |
1279 | unsigned int bi_size; | |
1280 | struct bio *bio; | |
1281 | ||
f5400b7a AE |
1282 | if (!bi) { |
1283 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1284 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1285 | } |
f7760dad AE |
1286 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1287 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1288 | if (!bio) | |
1289 | goto out_err; /* ENOMEM */ | |
1290 | ||
1291 | *end = bio; | |
1292 | end = &bio->bi_next; | |
602adf40 | 1293 | |
f7760dad AE |
1294 | off += bi_size; |
1295 | if (off == bi->bi_size) { | |
1296 | bi = bi->bi_next; | |
1297 | off = 0; | |
1298 | } | |
1299 | len -= bi_size; | |
1300 | } | |
1301 | *bio_src = bi; | |
1302 | *offset = off; | |
1303 | ||
1304 | return chain; | |
1305 | out_err: | |
1306 | bio_chain_put(chain); | |
602adf40 | 1307 | |
602adf40 YS |
1308 | return NULL; |
1309 | } | |
1310 | ||
926f9b3f AE |
1311 | /* |
1312 | * The default/initial value for all object request flags is 0. For | |
1313 | * each flag, once its value is set to 1 it is never reset to 0 | |
1314 | * again. | |
1315 | */ | |
57acbaa7 | 1316 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1317 | { |
57acbaa7 | 1318 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1319 | struct rbd_device *rbd_dev; |
1320 | ||
57acbaa7 AE |
1321 | rbd_dev = obj_request->img_request->rbd_dev; |
1322 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1323 | obj_request); |
1324 | } | |
1325 | } | |
1326 | ||
57acbaa7 | 1327 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1328 | { |
1329 | smp_mb(); | |
57acbaa7 | 1330 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1331 | } |
1332 | ||
57acbaa7 | 1333 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1334 | { |
57acbaa7 AE |
1335 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1336 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1337 | |
57acbaa7 AE |
1338 | if (obj_request_img_data_test(obj_request)) |
1339 | rbd_dev = obj_request->img_request->rbd_dev; | |
1340 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1341 | obj_request); |
1342 | } | |
1343 | } | |
1344 | ||
57acbaa7 | 1345 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1346 | { |
1347 | smp_mb(); | |
57acbaa7 | 1348 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1349 | } |
1350 | ||
5679c59f AE |
1351 | /* |
1352 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1353 | * flag. The latter is set based on the "exists" value provided. | |
1354 | * | |
1355 | * Note that for our purposes once an object exists it never goes | |
1356 | * away again. It's possible that the response from two existence | |
1357 | * checks are separated by the creation of the target object, and | |
1358 | * the first ("doesn't exist") response arrives *after* the second | |
1359 | * ("does exist"). In that case we ignore the second one. | |
1360 | */ | |
1361 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1362 | bool exists) | |
1363 | { | |
1364 | if (exists) | |
1365 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1366 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1367 | smp_mb(); | |
1368 | } | |
1369 | ||
1370 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1371 | { | |
1372 | smp_mb(); | |
1373 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1374 | } | |
1375 | ||
1376 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1377 | { | |
1378 | smp_mb(); | |
1379 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1380 | } | |
1381 | ||
bf0d5f50 AE |
1382 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1383 | { | |
37206ee5 AE |
1384 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1385 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1386 | kref_get(&obj_request->kref); |
1387 | } | |
1388 | ||
1389 | static void rbd_obj_request_destroy(struct kref *kref); | |
1390 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1391 | { | |
1392 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1393 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1394 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1395 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1396 | } | |
1397 | ||
e93f3152 AE |
1398 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1399 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1400 | static void rbd_img_request_destroy(struct kref *kref); |
1401 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1402 | { | |
1403 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1404 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1405 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1406 | if (img_request_child_test(img_request)) |
1407 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1408 | else | |
1409 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1410 | } |
1411 | ||
1412 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1413 | struct rbd_obj_request *obj_request) | |
1414 | { | |
25dcf954 AE |
1415 | rbd_assert(obj_request->img_request == NULL); |
1416 | ||
b155e86c | 1417 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1418 | obj_request->img_request = img_request; |
25dcf954 | 1419 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1420 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1421 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1422 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1423 | img_request->obj_request_count++; |
1424 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1425 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1426 | obj_request->which); | |
bf0d5f50 AE |
1427 | } |
1428 | ||
1429 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1430 | struct rbd_obj_request *obj_request) | |
1431 | { | |
1432 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1433 | |
37206ee5 AE |
1434 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1435 | obj_request->which); | |
bf0d5f50 | 1436 | list_del(&obj_request->links); |
25dcf954 AE |
1437 | rbd_assert(img_request->obj_request_count > 0); |
1438 | img_request->obj_request_count--; | |
1439 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1440 | obj_request->which = BAD_WHICH; | |
6365d33a | 1441 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1442 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1443 | obj_request->img_request = NULL; |
25dcf954 | 1444 | obj_request->callback = NULL; |
bf0d5f50 AE |
1445 | rbd_obj_request_put(obj_request); |
1446 | } | |
1447 | ||
1448 | static bool obj_request_type_valid(enum obj_request_type type) | |
1449 | { | |
1450 | switch (type) { | |
9969ebc5 | 1451 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1452 | case OBJ_REQUEST_BIO: |
788e2df3 | 1453 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1454 | return true; |
1455 | default: | |
1456 | return false; | |
1457 | } | |
1458 | } | |
1459 | ||
bf0d5f50 AE |
1460 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1461 | struct rbd_obj_request *obj_request) | |
1462 | { | |
37206ee5 AE |
1463 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1464 | ||
bf0d5f50 AE |
1465 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1466 | } | |
1467 | ||
1468 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1469 | { | |
55f27e09 | 1470 | |
37206ee5 | 1471 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1472 | |
1473 | /* | |
1474 | * If no error occurred, compute the aggregate transfer | |
1475 | * count for the image request. We could instead use | |
1476 | * atomic64_cmpxchg() to update it as each object request | |
1477 | * completes; not clear which way is better off hand. | |
1478 | */ | |
1479 | if (!img_request->result) { | |
1480 | struct rbd_obj_request *obj_request; | |
1481 | u64 xferred = 0; | |
1482 | ||
1483 | for_each_obj_request(img_request, obj_request) | |
1484 | xferred += obj_request->xferred; | |
1485 | img_request->xferred = xferred; | |
1486 | } | |
1487 | ||
bf0d5f50 AE |
1488 | if (img_request->callback) |
1489 | img_request->callback(img_request); | |
1490 | else | |
1491 | rbd_img_request_put(img_request); | |
1492 | } | |
1493 | ||
788e2df3 AE |
1494 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1495 | ||
1496 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1497 | { | |
37206ee5 AE |
1498 | dout("%s: obj %p\n", __func__, obj_request); |
1499 | ||
788e2df3 AE |
1500 | return wait_for_completion_interruptible(&obj_request->completion); |
1501 | } | |
1502 | ||
0c425248 AE |
1503 | /* |
1504 | * The default/initial value for all image request flags is 0. Each | |
1505 | * is conditionally set to 1 at image request initialization time | |
1506 | * and currently never change thereafter. | |
1507 | */ | |
1508 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1509 | { | |
1510 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1511 | smp_mb(); | |
1512 | } | |
1513 | ||
1514 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1515 | { | |
1516 | smp_mb(); | |
1517 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1518 | } | |
1519 | ||
9849e986 AE |
1520 | static void img_request_child_set(struct rbd_img_request *img_request) |
1521 | { | |
1522 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1523 | smp_mb(); | |
1524 | } | |
1525 | ||
e93f3152 AE |
1526 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1527 | { | |
1528 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1529 | smp_mb(); | |
1530 | } | |
1531 | ||
9849e986 AE |
1532 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1533 | { | |
1534 | smp_mb(); | |
1535 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1536 | } | |
1537 | ||
d0b2e944 AE |
1538 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1539 | { | |
1540 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1541 | smp_mb(); | |
1542 | } | |
1543 | ||
a2acd00e AE |
1544 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1545 | { | |
1546 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1547 | smp_mb(); | |
1548 | } | |
1549 | ||
d0b2e944 AE |
1550 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1551 | { | |
1552 | smp_mb(); | |
1553 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1554 | } | |
1555 | ||
6e2a4505 AE |
1556 | static void |
1557 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1558 | { | |
b9434c5b AE |
1559 | u64 xferred = obj_request->xferred; |
1560 | u64 length = obj_request->length; | |
1561 | ||
6e2a4505 AE |
1562 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1563 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1564 | xferred, length); |
6e2a4505 AE |
1565 | /* |
1566 | * ENOENT means a hole in the image. We zero-fill the | |
1567 | * entire length of the request. A short read also implies | |
1568 | * zero-fill to the end of the request. Either way we | |
1569 | * update the xferred count to indicate the whole request | |
1570 | * was satisfied. | |
1571 | */ | |
b9434c5b | 1572 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1573 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1574 | if (obj_request->type == OBJ_REQUEST_BIO) |
1575 | zero_bio_chain(obj_request->bio_list, 0); | |
1576 | else | |
1577 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1578 | obj_request->result = 0; |
b9434c5b AE |
1579 | obj_request->xferred = length; |
1580 | } else if (xferred < length && !obj_request->result) { | |
1581 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1582 | zero_bio_chain(obj_request->bio_list, xferred); | |
1583 | else | |
1584 | zero_pages(obj_request->pages, xferred, length); | |
1585 | obj_request->xferred = length; | |
6e2a4505 AE |
1586 | } |
1587 | obj_request_done_set(obj_request); | |
1588 | } | |
1589 | ||
bf0d5f50 AE |
1590 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1591 | { | |
37206ee5 AE |
1592 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1593 | obj_request->callback); | |
bf0d5f50 AE |
1594 | if (obj_request->callback) |
1595 | obj_request->callback(obj_request); | |
788e2df3 AE |
1596 | else |
1597 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1598 | } |
1599 | ||
c47f9371 | 1600 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1601 | { |
1602 | dout("%s: obj %p\n", __func__, obj_request); | |
1603 | obj_request_done_set(obj_request); | |
1604 | } | |
1605 | ||
c47f9371 | 1606 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1607 | { |
57acbaa7 | 1608 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1609 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1610 | bool layered = false; |
1611 | ||
1612 | if (obj_request_img_data_test(obj_request)) { | |
1613 | img_request = obj_request->img_request; | |
1614 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1615 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1616 | } |
8b3e1a56 AE |
1617 | |
1618 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1619 | obj_request, img_request, obj_request->result, | |
1620 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1621 | if (layered && obj_request->result == -ENOENT && |
1622 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1623 | rbd_img_parent_read(obj_request); |
1624 | else if (img_request) | |
6e2a4505 AE |
1625 | rbd_img_obj_request_read_callback(obj_request); |
1626 | else | |
1627 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1628 | } |
1629 | ||
c47f9371 | 1630 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1631 | { |
1b83bef2 SW |
1632 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1633 | obj_request->result, obj_request->length); | |
1634 | /* | |
8b3e1a56 AE |
1635 | * There is no such thing as a successful short write. Set |
1636 | * it to our originally-requested length. | |
1b83bef2 SW |
1637 | */ |
1638 | obj_request->xferred = obj_request->length; | |
07741308 | 1639 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1640 | } |
1641 | ||
fbfab539 AE |
1642 | /* |
1643 | * For a simple stat call there's nothing to do. We'll do more if | |
1644 | * this is part of a write sequence for a layered image. | |
1645 | */ | |
c47f9371 | 1646 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1647 | { |
37206ee5 | 1648 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1649 | obj_request_done_set(obj_request); |
1650 | } | |
1651 | ||
bf0d5f50 AE |
1652 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1653 | struct ceph_msg *msg) | |
1654 | { | |
1655 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1656 | u16 opcode; |
1657 | ||
37206ee5 | 1658 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1659 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1660 | if (obj_request_img_data_test(obj_request)) { |
1661 | rbd_assert(obj_request->img_request); | |
1662 | rbd_assert(obj_request->which != BAD_WHICH); | |
1663 | } else { | |
1664 | rbd_assert(obj_request->which == BAD_WHICH); | |
1665 | } | |
bf0d5f50 | 1666 | |
1b83bef2 SW |
1667 | if (osd_req->r_result < 0) |
1668 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1669 | |
0eefd470 | 1670 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1671 | |
c47f9371 AE |
1672 | /* |
1673 | * We support a 64-bit length, but ultimately it has to be | |
1674 | * passed to blk_end_request(), which takes an unsigned int. | |
1675 | */ | |
1b83bef2 | 1676 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1677 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1678 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1679 | switch (opcode) { |
1680 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1681 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1682 | break; |
1683 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1684 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1685 | break; |
fbfab539 | 1686 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1687 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1688 | break; |
36be9a76 | 1689 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1690 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1691 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1692 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1693 | break; |
bf0d5f50 AE |
1694 | default: |
1695 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1696 | obj_request->object_name, (unsigned short) opcode); | |
1697 | break; | |
1698 | } | |
1699 | ||
07741308 | 1700 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1701 | rbd_obj_request_complete(obj_request); |
1702 | } | |
1703 | ||
9d4df01f | 1704 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1705 | { |
1706 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1707 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1708 | u64 snap_id; |
430c28c3 | 1709 | |
8c042b0d | 1710 | rbd_assert(osd_req != NULL); |
430c28c3 | 1711 | |
9d4df01f | 1712 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1713 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1714 | NULL, snap_id, NULL); |
1715 | } | |
1716 | ||
1717 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1718 | { | |
1719 | struct rbd_img_request *img_request = obj_request->img_request; | |
1720 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1721 | struct ceph_snap_context *snapc; | |
1722 | struct timespec mtime = CURRENT_TIME; | |
1723 | ||
1724 | rbd_assert(osd_req != NULL); | |
1725 | ||
1726 | snapc = img_request ? img_request->snapc : NULL; | |
1727 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1728 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1729 | } |
1730 | ||
bf0d5f50 AE |
1731 | static struct ceph_osd_request *rbd_osd_req_create( |
1732 | struct rbd_device *rbd_dev, | |
1733 | bool write_request, | |
430c28c3 | 1734 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1735 | { |
bf0d5f50 AE |
1736 | struct ceph_snap_context *snapc = NULL; |
1737 | struct ceph_osd_client *osdc; | |
1738 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1739 | |
6365d33a AE |
1740 | if (obj_request_img_data_test(obj_request)) { |
1741 | struct rbd_img_request *img_request = obj_request->img_request; | |
1742 | ||
0c425248 AE |
1743 | rbd_assert(write_request == |
1744 | img_request_write_test(img_request)); | |
1745 | if (write_request) | |
bf0d5f50 | 1746 | snapc = img_request->snapc; |
bf0d5f50 AE |
1747 | } |
1748 | ||
1749 | /* Allocate and initialize the request, for the single op */ | |
1750 | ||
1751 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1752 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1753 | if (!osd_req) | |
1754 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1755 | |
430c28c3 | 1756 | if (write_request) |
bf0d5f50 | 1757 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1758 | else |
bf0d5f50 | 1759 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1760 | |
1761 | osd_req->r_callback = rbd_osd_req_callback; | |
1762 | osd_req->r_priv = obj_request; | |
1763 | ||
1764 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1765 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1766 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1767 | ||
1768 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1769 | ||
bf0d5f50 AE |
1770 | return osd_req; |
1771 | } | |
1772 | ||
0eefd470 AE |
1773 | /* |
1774 | * Create a copyup osd request based on the information in the | |
1775 | * object request supplied. A copyup request has two osd ops, | |
1776 | * a copyup method call, and a "normal" write request. | |
1777 | */ | |
1778 | static struct ceph_osd_request * | |
1779 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1780 | { | |
1781 | struct rbd_img_request *img_request; | |
1782 | struct ceph_snap_context *snapc; | |
1783 | struct rbd_device *rbd_dev; | |
1784 | struct ceph_osd_client *osdc; | |
1785 | struct ceph_osd_request *osd_req; | |
1786 | ||
1787 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1788 | img_request = obj_request->img_request; | |
1789 | rbd_assert(img_request); | |
1790 | rbd_assert(img_request_write_test(img_request)); | |
1791 | ||
1792 | /* Allocate and initialize the request, for the two ops */ | |
1793 | ||
1794 | snapc = img_request->snapc; | |
1795 | rbd_dev = img_request->rbd_dev; | |
1796 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1797 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1798 | if (!osd_req) | |
1799 | return NULL; /* ENOMEM */ | |
1800 | ||
1801 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1802 | osd_req->r_callback = rbd_osd_req_callback; | |
1803 | osd_req->r_priv = obj_request; | |
1804 | ||
1805 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1806 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1807 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1808 | ||
1809 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1810 | ||
1811 | return osd_req; | |
1812 | } | |
1813 | ||
1814 | ||
bf0d5f50 AE |
1815 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1816 | { | |
1817 | ceph_osdc_put_request(osd_req); | |
1818 | } | |
1819 | ||
1820 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1821 | ||
1822 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1823 | u64 offset, u64 length, | |
1824 | enum obj_request_type type) | |
1825 | { | |
1826 | struct rbd_obj_request *obj_request; | |
1827 | size_t size; | |
1828 | char *name; | |
1829 | ||
1830 | rbd_assert(obj_request_type_valid(type)); | |
1831 | ||
1832 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1833 | name = kmalloc(size, GFP_KERNEL); |
1834 | if (!name) | |
bf0d5f50 AE |
1835 | return NULL; |
1836 | ||
868311b1 | 1837 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1838 | if (!obj_request) { |
1839 | kfree(name); | |
1840 | return NULL; | |
1841 | } | |
1842 | ||
bf0d5f50 AE |
1843 | obj_request->object_name = memcpy(name, object_name, size); |
1844 | obj_request->offset = offset; | |
1845 | obj_request->length = length; | |
926f9b3f | 1846 | obj_request->flags = 0; |
bf0d5f50 AE |
1847 | obj_request->which = BAD_WHICH; |
1848 | obj_request->type = type; | |
1849 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1850 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1851 | kref_init(&obj_request->kref); |
1852 | ||
37206ee5 AE |
1853 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1854 | offset, length, (int)type, obj_request); | |
1855 | ||
bf0d5f50 AE |
1856 | return obj_request; |
1857 | } | |
1858 | ||
1859 | static void rbd_obj_request_destroy(struct kref *kref) | |
1860 | { | |
1861 | struct rbd_obj_request *obj_request; | |
1862 | ||
1863 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1864 | ||
37206ee5 AE |
1865 | dout("%s: obj %p\n", __func__, obj_request); |
1866 | ||
bf0d5f50 AE |
1867 | rbd_assert(obj_request->img_request == NULL); |
1868 | rbd_assert(obj_request->which == BAD_WHICH); | |
1869 | ||
1870 | if (obj_request->osd_req) | |
1871 | rbd_osd_req_destroy(obj_request->osd_req); | |
1872 | ||
1873 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1874 | switch (obj_request->type) { | |
9969ebc5 AE |
1875 | case OBJ_REQUEST_NODATA: |
1876 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1877 | case OBJ_REQUEST_BIO: |
1878 | if (obj_request->bio_list) | |
1879 | bio_chain_put(obj_request->bio_list); | |
1880 | break; | |
788e2df3 AE |
1881 | case OBJ_REQUEST_PAGES: |
1882 | if (obj_request->pages) | |
1883 | ceph_release_page_vector(obj_request->pages, | |
1884 | obj_request->page_count); | |
1885 | break; | |
bf0d5f50 AE |
1886 | } |
1887 | ||
f907ad55 | 1888 | kfree(obj_request->object_name); |
868311b1 AE |
1889 | obj_request->object_name = NULL; |
1890 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
1891 | } |
1892 | ||
fb65d228 AE |
1893 | /* It's OK to call this for a device with no parent */ |
1894 | ||
1895 | static void rbd_spec_put(struct rbd_spec *spec); | |
1896 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
1897 | { | |
1898 | rbd_dev_remove_parent(rbd_dev); | |
1899 | rbd_spec_put(rbd_dev->parent_spec); | |
1900 | rbd_dev->parent_spec = NULL; | |
1901 | rbd_dev->parent_overlap = 0; | |
1902 | } | |
1903 | ||
a2acd00e AE |
1904 | /* |
1905 | * Parent image reference counting is used to determine when an | |
1906 | * image's parent fields can be safely torn down--after there are no | |
1907 | * more in-flight requests to the parent image. When the last | |
1908 | * reference is dropped, cleaning them up is safe. | |
1909 | */ | |
1910 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
1911 | { | |
1912 | int counter; | |
1913 | ||
1914 | if (!rbd_dev->parent_spec) | |
1915 | return; | |
1916 | ||
1917 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
1918 | if (counter > 0) | |
1919 | return; | |
1920 | ||
1921 | /* Last reference; clean up parent data structures */ | |
1922 | ||
1923 | if (!counter) | |
1924 | rbd_dev_unparent(rbd_dev); | |
1925 | else | |
1926 | rbd_warn(rbd_dev, "parent reference underflow\n"); | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * If an image has a non-zero parent overlap, get a reference to its | |
1931 | * parent. | |
1932 | * | |
392a9dad AE |
1933 | * We must get the reference before checking for the overlap to |
1934 | * coordinate properly with zeroing the parent overlap in | |
1935 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
1936 | * drop it again if there is no overlap. | |
1937 | * | |
a2acd00e AE |
1938 | * Returns true if the rbd device has a parent with a non-zero |
1939 | * overlap and a reference for it was successfully taken, or | |
1940 | * false otherwise. | |
1941 | */ | |
1942 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
1943 | { | |
1944 | int counter; | |
1945 | ||
1946 | if (!rbd_dev->parent_spec) | |
1947 | return false; | |
1948 | ||
1949 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
1950 | if (counter > 0 && rbd_dev->parent_overlap) | |
1951 | return true; | |
1952 | ||
1953 | /* Image was flattened, but parent is not yet torn down */ | |
1954 | ||
1955 | if (counter < 0) | |
1956 | rbd_warn(rbd_dev, "parent reference overflow\n"); | |
1957 | ||
1958 | return false; | |
1959 | } | |
1960 | ||
bf0d5f50 AE |
1961 | /* |
1962 | * Caller is responsible for filling in the list of object requests | |
1963 | * that comprises the image request, and the Linux request pointer | |
1964 | * (if there is one). | |
1965 | */ | |
cc344fa1 AE |
1966 | static struct rbd_img_request *rbd_img_request_create( |
1967 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1968 | u64 offset, u64 length, |
e93f3152 | 1969 | bool write_request) |
bf0d5f50 AE |
1970 | { |
1971 | struct rbd_img_request *img_request; | |
bf0d5f50 | 1972 | |
1c2a9dfe | 1973 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC); |
bf0d5f50 AE |
1974 | if (!img_request) |
1975 | return NULL; | |
1976 | ||
1977 | if (write_request) { | |
1978 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 1979 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 1980 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1981 | } |
1982 | ||
1983 | img_request->rq = NULL; | |
1984 | img_request->rbd_dev = rbd_dev; | |
1985 | img_request->offset = offset; | |
1986 | img_request->length = length; | |
0c425248 AE |
1987 | img_request->flags = 0; |
1988 | if (write_request) { | |
1989 | img_request_write_set(img_request); | |
468521c1 | 1990 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1991 | } else { |
bf0d5f50 | 1992 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1993 | } |
a2acd00e | 1994 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 1995 | img_request_layered_set(img_request); |
bf0d5f50 AE |
1996 | spin_lock_init(&img_request->completion_lock); |
1997 | img_request->next_completion = 0; | |
1998 | img_request->callback = NULL; | |
a5a337d4 | 1999 | img_request->result = 0; |
bf0d5f50 AE |
2000 | img_request->obj_request_count = 0; |
2001 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2002 | kref_init(&img_request->kref); | |
2003 | ||
37206ee5 AE |
2004 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
2005 | write_request ? "write" : "read", offset, length, | |
2006 | img_request); | |
2007 | ||
bf0d5f50 AE |
2008 | return img_request; |
2009 | } | |
2010 | ||
2011 | static void rbd_img_request_destroy(struct kref *kref) | |
2012 | { | |
2013 | struct rbd_img_request *img_request; | |
2014 | struct rbd_obj_request *obj_request; | |
2015 | struct rbd_obj_request *next_obj_request; | |
2016 | ||
2017 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2018 | ||
37206ee5 AE |
2019 | dout("%s: img %p\n", __func__, img_request); |
2020 | ||
bf0d5f50 AE |
2021 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2022 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2023 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2024 | |
a2acd00e AE |
2025 | if (img_request_layered_test(img_request)) { |
2026 | img_request_layered_clear(img_request); | |
2027 | rbd_dev_parent_put(img_request->rbd_dev); | |
2028 | } | |
2029 | ||
0c425248 | 2030 | if (img_request_write_test(img_request)) |
812164f8 | 2031 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2032 | |
1c2a9dfe | 2033 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2034 | } |
2035 | ||
e93f3152 AE |
2036 | static struct rbd_img_request *rbd_parent_request_create( |
2037 | struct rbd_obj_request *obj_request, | |
2038 | u64 img_offset, u64 length) | |
2039 | { | |
2040 | struct rbd_img_request *parent_request; | |
2041 | struct rbd_device *rbd_dev; | |
2042 | ||
2043 | rbd_assert(obj_request->img_request); | |
2044 | rbd_dev = obj_request->img_request->rbd_dev; | |
2045 | ||
2046 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2047 | img_offset, length, false); | |
2048 | if (!parent_request) | |
2049 | return NULL; | |
2050 | ||
2051 | img_request_child_set(parent_request); | |
2052 | rbd_obj_request_get(obj_request); | |
2053 | parent_request->obj_request = obj_request; | |
2054 | ||
2055 | return parent_request; | |
2056 | } | |
2057 | ||
2058 | static void rbd_parent_request_destroy(struct kref *kref) | |
2059 | { | |
2060 | struct rbd_img_request *parent_request; | |
2061 | struct rbd_obj_request *orig_request; | |
2062 | ||
2063 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2064 | orig_request = parent_request->obj_request; | |
2065 | ||
2066 | parent_request->obj_request = NULL; | |
2067 | rbd_obj_request_put(orig_request); | |
2068 | img_request_child_clear(parent_request); | |
2069 | ||
2070 | rbd_img_request_destroy(kref); | |
2071 | } | |
2072 | ||
1217857f AE |
2073 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2074 | { | |
6365d33a | 2075 | struct rbd_img_request *img_request; |
1217857f AE |
2076 | unsigned int xferred; |
2077 | int result; | |
8b3e1a56 | 2078 | bool more; |
1217857f | 2079 | |
6365d33a AE |
2080 | rbd_assert(obj_request_img_data_test(obj_request)); |
2081 | img_request = obj_request->img_request; | |
2082 | ||
1217857f AE |
2083 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2084 | xferred = (unsigned int)obj_request->xferred; | |
2085 | result = obj_request->result; | |
2086 | if (result) { | |
2087 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2088 | ||
2089 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
2090 | img_request_write_test(img_request) ? "write" : "read", | |
2091 | obj_request->length, obj_request->img_offset, | |
2092 | obj_request->offset); | |
2093 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
2094 | result, xferred); | |
2095 | if (!img_request->result) | |
2096 | img_request->result = result; | |
2097 | } | |
2098 | ||
f1a4739f AE |
2099 | /* Image object requests don't own their page array */ |
2100 | ||
2101 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2102 | obj_request->pages = NULL; | |
2103 | obj_request->page_count = 0; | |
2104 | } | |
2105 | ||
8b3e1a56 AE |
2106 | if (img_request_child_test(img_request)) { |
2107 | rbd_assert(img_request->obj_request != NULL); | |
2108 | more = obj_request->which < img_request->obj_request_count - 1; | |
2109 | } else { | |
2110 | rbd_assert(img_request->rq != NULL); | |
2111 | more = blk_end_request(img_request->rq, result, xferred); | |
2112 | } | |
2113 | ||
2114 | return more; | |
1217857f AE |
2115 | } |
2116 | ||
2169238d AE |
2117 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2118 | { | |
2119 | struct rbd_img_request *img_request; | |
2120 | u32 which = obj_request->which; | |
2121 | bool more = true; | |
2122 | ||
6365d33a | 2123 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2124 | img_request = obj_request->img_request; |
2125 | ||
2126 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2127 | rbd_assert(img_request != NULL); | |
2169238d AE |
2128 | rbd_assert(img_request->obj_request_count > 0); |
2129 | rbd_assert(which != BAD_WHICH); | |
2130 | rbd_assert(which < img_request->obj_request_count); | |
2131 | rbd_assert(which >= img_request->next_completion); | |
2132 | ||
2133 | spin_lock_irq(&img_request->completion_lock); | |
2134 | if (which != img_request->next_completion) | |
2135 | goto out; | |
2136 | ||
2137 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2138 | rbd_assert(more); |
2139 | rbd_assert(which < img_request->obj_request_count); | |
2140 | ||
2141 | if (!obj_request_done_test(obj_request)) | |
2142 | break; | |
1217857f | 2143 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2144 | which++; |
2145 | } | |
2146 | ||
2147 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2148 | img_request->next_completion = which; | |
2149 | out: | |
2150 | spin_unlock_irq(&img_request->completion_lock); | |
2151 | ||
2152 | if (!more) | |
2153 | rbd_img_request_complete(img_request); | |
2154 | } | |
2155 | ||
f1a4739f AE |
2156 | /* |
2157 | * Split up an image request into one or more object requests, each | |
2158 | * to a different object. The "type" parameter indicates whether | |
2159 | * "data_desc" is the pointer to the head of a list of bio | |
2160 | * structures, or the base of a page array. In either case this | |
2161 | * function assumes data_desc describes memory sufficient to hold | |
2162 | * all data described by the image request. | |
2163 | */ | |
2164 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2165 | enum obj_request_type type, | |
2166 | void *data_desc) | |
bf0d5f50 AE |
2167 | { |
2168 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2169 | struct rbd_obj_request *obj_request = NULL; | |
2170 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 2171 | bool write_request = img_request_write_test(img_request); |
f1a4739f AE |
2172 | struct bio *bio_list; |
2173 | unsigned int bio_offset = 0; | |
2174 | struct page **pages; | |
7da22d29 | 2175 | u64 img_offset; |
bf0d5f50 AE |
2176 | u64 resid; |
2177 | u16 opcode; | |
2178 | ||
f1a4739f AE |
2179 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2180 | (int)type, data_desc); | |
37206ee5 | 2181 | |
430c28c3 | 2182 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 2183 | img_offset = img_request->offset; |
bf0d5f50 | 2184 | resid = img_request->length; |
4dda41d3 | 2185 | rbd_assert(resid > 0); |
f1a4739f AE |
2186 | |
2187 | if (type == OBJ_REQUEST_BIO) { | |
2188 | bio_list = data_desc; | |
2189 | rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
2190 | } else { | |
2191 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
2192 | pages = data_desc; | |
2193 | } | |
2194 | ||
bf0d5f50 | 2195 | while (resid) { |
2fa12320 | 2196 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2197 | const char *object_name; |
bf0d5f50 AE |
2198 | u64 offset; |
2199 | u64 length; | |
2200 | ||
7da22d29 | 2201 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2202 | if (!object_name) |
2203 | goto out_unwind; | |
7da22d29 AE |
2204 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2205 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2206 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2207 | offset, length, type); |
78c2a44a AE |
2208 | /* object request has its own copy of the object name */ |
2209 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2210 | if (!obj_request) |
2211 | goto out_unwind; | |
2212 | ||
f1a4739f AE |
2213 | if (type == OBJ_REQUEST_BIO) { |
2214 | unsigned int clone_size; | |
2215 | ||
2216 | rbd_assert(length <= (u64)UINT_MAX); | |
2217 | clone_size = (unsigned int)length; | |
2218 | obj_request->bio_list = | |
2219 | bio_chain_clone_range(&bio_list, | |
2220 | &bio_offset, | |
2221 | clone_size, | |
2222 | GFP_ATOMIC); | |
2223 | if (!obj_request->bio_list) | |
2224 | goto out_partial; | |
2225 | } else { | |
2226 | unsigned int page_count; | |
2227 | ||
2228 | obj_request->pages = pages; | |
2229 | page_count = (u32)calc_pages_for(offset, length); | |
2230 | obj_request->page_count = page_count; | |
2231 | if ((offset + length) & ~PAGE_MASK) | |
2232 | page_count--; /* more on last page */ | |
2233 | pages += page_count; | |
2234 | } | |
bf0d5f50 | 2235 | |
2fa12320 AE |
2236 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
2237 | obj_request); | |
2238 | if (!osd_req) | |
bf0d5f50 | 2239 | goto out_partial; |
2fa12320 | 2240 | obj_request->osd_req = osd_req; |
2169238d | 2241 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 2242 | |
2fa12320 AE |
2243 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
2244 | 0, 0); | |
f1a4739f AE |
2245 | if (type == OBJ_REQUEST_BIO) |
2246 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
2247 | obj_request->bio_list, length); | |
2248 | else | |
2249 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
2250 | obj_request->pages, length, | |
2251 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
2252 | |
2253 | if (write_request) | |
2254 | rbd_osd_req_format_write(obj_request); | |
2255 | else | |
2256 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2257 | |
7da22d29 | 2258 | obj_request->img_offset = img_offset; |
bf0d5f50 AE |
2259 | rbd_img_obj_request_add(img_request, obj_request); |
2260 | ||
7da22d29 | 2261 | img_offset += length; |
bf0d5f50 AE |
2262 | resid -= length; |
2263 | } | |
2264 | ||
2265 | return 0; | |
2266 | ||
2267 | out_partial: | |
2268 | rbd_obj_request_put(obj_request); | |
2269 | out_unwind: | |
2270 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
2271 | rbd_obj_request_put(obj_request); | |
2272 | ||
2273 | return -ENOMEM; | |
2274 | } | |
2275 | ||
0eefd470 AE |
2276 | static void |
2277 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2278 | { | |
2279 | struct rbd_img_request *img_request; | |
2280 | struct rbd_device *rbd_dev; | |
ebda6408 | 2281 | struct page **pages; |
0eefd470 AE |
2282 | u32 page_count; |
2283 | ||
2284 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2285 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2286 | img_request = obj_request->img_request; | |
2287 | rbd_assert(img_request); | |
2288 | ||
2289 | rbd_dev = img_request->rbd_dev; | |
2290 | rbd_assert(rbd_dev); | |
0eefd470 | 2291 | |
ebda6408 AE |
2292 | pages = obj_request->copyup_pages; |
2293 | rbd_assert(pages != NULL); | |
0eefd470 | 2294 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2295 | page_count = obj_request->copyup_page_count; |
2296 | rbd_assert(page_count); | |
2297 | obj_request->copyup_page_count = 0; | |
2298 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2299 | |
2300 | /* | |
2301 | * We want the transfer count to reflect the size of the | |
2302 | * original write request. There is no such thing as a | |
2303 | * successful short write, so if the request was successful | |
2304 | * we can just set it to the originally-requested length. | |
2305 | */ | |
2306 | if (!obj_request->result) | |
2307 | obj_request->xferred = obj_request->length; | |
2308 | ||
2309 | /* Finish up with the normal image object callback */ | |
2310 | ||
2311 | rbd_img_obj_callback(obj_request); | |
2312 | } | |
2313 | ||
3d7efd18 AE |
2314 | static void |
2315 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2316 | { | |
2317 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2318 | struct ceph_osd_request *osd_req; |
2319 | struct ceph_osd_client *osdc; | |
2320 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2321 | struct page **pages; |
ebda6408 | 2322 | u32 page_count; |
bbea1c1a | 2323 | int img_result; |
ebda6408 | 2324 | u64 parent_length; |
b91f09f1 AE |
2325 | u64 offset; |
2326 | u64 length; | |
3d7efd18 AE |
2327 | |
2328 | rbd_assert(img_request_child_test(img_request)); | |
2329 | ||
2330 | /* First get what we need from the image request */ | |
2331 | ||
2332 | pages = img_request->copyup_pages; | |
2333 | rbd_assert(pages != NULL); | |
2334 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2335 | page_count = img_request->copyup_page_count; |
2336 | rbd_assert(page_count); | |
2337 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2338 | |
2339 | orig_request = img_request->obj_request; | |
2340 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2341 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2342 | img_result = img_request->result; |
ebda6408 AE |
2343 | parent_length = img_request->length; |
2344 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2345 | rbd_img_request_put(img_request); |
3d7efd18 | 2346 | |
91c6febb AE |
2347 | rbd_assert(orig_request->img_request); |
2348 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2349 | rbd_assert(rbd_dev); |
0eefd470 | 2350 | |
bbea1c1a AE |
2351 | /* |
2352 | * If the overlap has become 0 (most likely because the | |
2353 | * image has been flattened) we need to free the pages | |
2354 | * and re-submit the original write request. | |
2355 | */ | |
2356 | if (!rbd_dev->parent_overlap) { | |
2357 | struct ceph_osd_client *osdc; | |
2358 | ||
2359 | ceph_release_page_vector(pages, page_count); | |
2360 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2361 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2362 | if (!img_result) | |
2363 | return; | |
2364 | } | |
2365 | ||
2366 | if (img_result) | |
0eefd470 AE |
2367 | goto out_err; |
2368 | ||
8785b1d4 AE |
2369 | /* |
2370 | * The original osd request is of no use to use any more. | |
2371 | * We need a new one that can hold the two ops in a copyup | |
2372 | * request. Allocate the new copyup osd request for the | |
2373 | * original request, and release the old one. | |
2374 | */ | |
bbea1c1a | 2375 | img_result = -ENOMEM; |
0eefd470 AE |
2376 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2377 | if (!osd_req) | |
2378 | goto out_err; | |
8785b1d4 | 2379 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2380 | orig_request->osd_req = osd_req; |
2381 | orig_request->copyup_pages = pages; | |
ebda6408 | 2382 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2383 | |
0eefd470 | 2384 | /* Initialize the copyup op */ |
3d7efd18 | 2385 | |
0eefd470 | 2386 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2387 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2388 | false, false); |
3d7efd18 | 2389 | |
0eefd470 AE |
2390 | /* Then the original write request op */ |
2391 | ||
b91f09f1 AE |
2392 | offset = orig_request->offset; |
2393 | length = orig_request->length; | |
0eefd470 | 2394 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, |
b91f09f1 AE |
2395 | offset, length, 0, 0); |
2396 | if (orig_request->type == OBJ_REQUEST_BIO) | |
2397 | osd_req_op_extent_osd_data_bio(osd_req, 1, | |
2398 | orig_request->bio_list, length); | |
2399 | else | |
2400 | osd_req_op_extent_osd_data_pages(osd_req, 1, | |
2401 | orig_request->pages, length, | |
2402 | offset & ~PAGE_MASK, false, false); | |
0eefd470 AE |
2403 | |
2404 | rbd_osd_req_format_write(orig_request); | |
2405 | ||
2406 | /* All set, send it off. */ | |
2407 | ||
2408 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2409 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2410 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2411 | if (!img_result) | |
0eefd470 AE |
2412 | return; |
2413 | out_err: | |
2414 | /* Record the error code and complete the request */ | |
2415 | ||
bbea1c1a | 2416 | orig_request->result = img_result; |
0eefd470 AE |
2417 | orig_request->xferred = 0; |
2418 | obj_request_done_set(orig_request); | |
2419 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2420 | } |
2421 | ||
2422 | /* | |
2423 | * Read from the parent image the range of data that covers the | |
2424 | * entire target of the given object request. This is used for | |
2425 | * satisfying a layered image write request when the target of an | |
2426 | * object request from the image request does not exist. | |
2427 | * | |
2428 | * A page array big enough to hold the returned data is allocated | |
2429 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2430 | * When the read completes, this page array will be transferred to | |
2431 | * the original object request for the copyup operation. | |
2432 | * | |
2433 | * If an error occurs, record it as the result of the original | |
2434 | * object request and mark it done so it gets completed. | |
2435 | */ | |
2436 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2437 | { | |
2438 | struct rbd_img_request *img_request = NULL; | |
2439 | struct rbd_img_request *parent_request = NULL; | |
2440 | struct rbd_device *rbd_dev; | |
2441 | u64 img_offset; | |
2442 | u64 length; | |
2443 | struct page **pages = NULL; | |
2444 | u32 page_count; | |
2445 | int result; | |
2446 | ||
2447 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2448 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2449 | |
2450 | img_request = obj_request->img_request; | |
2451 | rbd_assert(img_request != NULL); | |
2452 | rbd_dev = img_request->rbd_dev; | |
2453 | rbd_assert(rbd_dev->parent != NULL); | |
2454 | ||
2455 | /* | |
2456 | * Determine the byte range covered by the object in the | |
2457 | * child image to which the original request was to be sent. | |
2458 | */ | |
2459 | img_offset = obj_request->img_offset - obj_request->offset; | |
2460 | length = (u64)1 << rbd_dev->header.obj_order; | |
2461 | ||
a9e8ba2c AE |
2462 | /* |
2463 | * There is no defined parent data beyond the parent | |
2464 | * overlap, so limit what we read at that boundary if | |
2465 | * necessary. | |
2466 | */ | |
2467 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2468 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2469 | length = rbd_dev->parent_overlap - img_offset; | |
2470 | } | |
2471 | ||
3d7efd18 AE |
2472 | /* |
2473 | * Allocate a page array big enough to receive the data read | |
2474 | * from the parent. | |
2475 | */ | |
2476 | page_count = (u32)calc_pages_for(0, length); | |
2477 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2478 | if (IS_ERR(pages)) { | |
2479 | result = PTR_ERR(pages); | |
2480 | pages = NULL; | |
2481 | goto out_err; | |
2482 | } | |
2483 | ||
2484 | result = -ENOMEM; | |
e93f3152 AE |
2485 | parent_request = rbd_parent_request_create(obj_request, |
2486 | img_offset, length); | |
3d7efd18 AE |
2487 | if (!parent_request) |
2488 | goto out_err; | |
3d7efd18 AE |
2489 | |
2490 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2491 | if (result) | |
2492 | goto out_err; | |
2493 | parent_request->copyup_pages = pages; | |
ebda6408 | 2494 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2495 | |
2496 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2497 | result = rbd_img_request_submit(parent_request); | |
2498 | if (!result) | |
2499 | return 0; | |
2500 | ||
2501 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2502 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2503 | parent_request->obj_request = NULL; |
2504 | rbd_obj_request_put(obj_request); | |
2505 | out_err: | |
2506 | if (pages) | |
2507 | ceph_release_page_vector(pages, page_count); | |
2508 | if (parent_request) | |
2509 | rbd_img_request_put(parent_request); | |
2510 | obj_request->result = result; | |
2511 | obj_request->xferred = 0; | |
2512 | obj_request_done_set(obj_request); | |
2513 | ||
2514 | return result; | |
2515 | } | |
2516 | ||
c5b5ef6c AE |
2517 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2518 | { | |
c5b5ef6c | 2519 | struct rbd_obj_request *orig_request; |
638f5abe | 2520 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2521 | int result; |
2522 | ||
2523 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2524 | ||
2525 | /* | |
2526 | * All we need from the object request is the original | |
2527 | * request and the result of the STAT op. Grab those, then | |
2528 | * we're done with the request. | |
2529 | */ | |
2530 | orig_request = obj_request->obj_request; | |
2531 | obj_request->obj_request = NULL; | |
2532 | rbd_assert(orig_request); | |
2533 | rbd_assert(orig_request->img_request); | |
2534 | ||
2535 | result = obj_request->result; | |
2536 | obj_request->result = 0; | |
2537 | ||
2538 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2539 | obj_request, orig_request, result, | |
2540 | obj_request->xferred, obj_request->length); | |
2541 | rbd_obj_request_put(obj_request); | |
2542 | ||
638f5abe AE |
2543 | /* |
2544 | * If the overlap has become 0 (most likely because the | |
2545 | * image has been flattened) we need to free the pages | |
2546 | * and re-submit the original write request. | |
2547 | */ | |
2548 | rbd_dev = orig_request->img_request->rbd_dev; | |
2549 | if (!rbd_dev->parent_overlap) { | |
2550 | struct ceph_osd_client *osdc; | |
2551 | ||
2552 | rbd_obj_request_put(orig_request); | |
2553 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2554 | result = rbd_obj_request_submit(osdc, orig_request); | |
2555 | if (!result) | |
2556 | return; | |
2557 | } | |
c5b5ef6c AE |
2558 | |
2559 | /* | |
2560 | * Our only purpose here is to determine whether the object | |
2561 | * exists, and we don't want to treat the non-existence as | |
2562 | * an error. If something else comes back, transfer the | |
2563 | * error to the original request and complete it now. | |
2564 | */ | |
2565 | if (!result) { | |
2566 | obj_request_existence_set(orig_request, true); | |
2567 | } else if (result == -ENOENT) { | |
2568 | obj_request_existence_set(orig_request, false); | |
2569 | } else if (result) { | |
2570 | orig_request->result = result; | |
3d7efd18 | 2571 | goto out; |
c5b5ef6c AE |
2572 | } |
2573 | ||
2574 | /* | |
2575 | * Resubmit the original request now that we have recorded | |
2576 | * whether the target object exists. | |
2577 | */ | |
b454e36d | 2578 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2579 | out: |
c5b5ef6c AE |
2580 | if (orig_request->result) |
2581 | rbd_obj_request_complete(orig_request); | |
2582 | rbd_obj_request_put(orig_request); | |
2583 | } | |
2584 | ||
2585 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2586 | { | |
2587 | struct rbd_obj_request *stat_request; | |
2588 | struct rbd_device *rbd_dev; | |
2589 | struct ceph_osd_client *osdc; | |
2590 | struct page **pages = NULL; | |
2591 | u32 page_count; | |
2592 | size_t size; | |
2593 | int ret; | |
2594 | ||
2595 | /* | |
2596 | * The response data for a STAT call consists of: | |
2597 | * le64 length; | |
2598 | * struct { | |
2599 | * le32 tv_sec; | |
2600 | * le32 tv_nsec; | |
2601 | * } mtime; | |
2602 | */ | |
2603 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2604 | page_count = (u32)calc_pages_for(0, size); | |
2605 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2606 | if (IS_ERR(pages)) | |
2607 | return PTR_ERR(pages); | |
2608 | ||
2609 | ret = -ENOMEM; | |
2610 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2611 | OBJ_REQUEST_PAGES); | |
2612 | if (!stat_request) | |
2613 | goto out; | |
2614 | ||
2615 | rbd_obj_request_get(obj_request); | |
2616 | stat_request->obj_request = obj_request; | |
2617 | stat_request->pages = pages; | |
2618 | stat_request->page_count = page_count; | |
2619 | ||
2620 | rbd_assert(obj_request->img_request); | |
2621 | rbd_dev = obj_request->img_request->rbd_dev; | |
2622 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2623 | stat_request); | |
2624 | if (!stat_request->osd_req) | |
2625 | goto out; | |
2626 | stat_request->callback = rbd_img_obj_exists_callback; | |
2627 | ||
2628 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2629 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2630 | false, false); | |
9d4df01f | 2631 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2632 | |
2633 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2634 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2635 | out: | |
2636 | if (ret) | |
2637 | rbd_obj_request_put(obj_request); | |
2638 | ||
2639 | return ret; | |
2640 | } | |
2641 | ||
b454e36d AE |
2642 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2643 | { | |
2644 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2645 | struct rbd_device *rbd_dev; |
3d7efd18 | 2646 | bool known; |
b454e36d AE |
2647 | |
2648 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2649 | ||
2650 | img_request = obj_request->img_request; | |
2651 | rbd_assert(img_request); | |
a9e8ba2c | 2652 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2653 | |
b454e36d | 2654 | /* |
a9e8ba2c AE |
2655 | * Only writes to layered images need special handling. |
2656 | * Reads and non-layered writes are simple object requests. | |
2657 | * Layered writes that start beyond the end of the overlap | |
2658 | * with the parent have no parent data, so they too are | |
2659 | * simple object requests. Finally, if the target object is | |
2660 | * known to already exist, its parent data has already been | |
2661 | * copied, so a write to the object can also be handled as a | |
2662 | * simple object request. | |
b454e36d AE |
2663 | */ |
2664 | if (!img_request_write_test(img_request) || | |
2665 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2666 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2667 | ((known = obj_request_known_test(obj_request)) && |
2668 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2669 | |
2670 | struct rbd_device *rbd_dev; | |
2671 | struct ceph_osd_client *osdc; | |
2672 | ||
2673 | rbd_dev = obj_request->img_request->rbd_dev; | |
2674 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2675 | ||
2676 | return rbd_obj_request_submit(osdc, obj_request); | |
2677 | } | |
2678 | ||
2679 | /* | |
3d7efd18 AE |
2680 | * It's a layered write. The target object might exist but |
2681 | * we may not know that yet. If we know it doesn't exist, | |
2682 | * start by reading the data for the full target object from | |
2683 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2684 | */ |
3d7efd18 AE |
2685 | if (known) |
2686 | return rbd_img_obj_parent_read_full(obj_request); | |
2687 | ||
2688 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2689 | |
2690 | return rbd_img_obj_exists_submit(obj_request); | |
2691 | } | |
2692 | ||
bf0d5f50 AE |
2693 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2694 | { | |
bf0d5f50 | 2695 | struct rbd_obj_request *obj_request; |
46faeed4 | 2696 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2697 | |
37206ee5 | 2698 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2699 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2700 | int ret; |
2701 | ||
b454e36d | 2702 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2703 | if (ret) |
2704 | return ret; | |
bf0d5f50 AE |
2705 | } |
2706 | ||
2707 | return 0; | |
2708 | } | |
8b3e1a56 AE |
2709 | |
2710 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2711 | { | |
2712 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2713 | struct rbd_device *rbd_dev; |
2714 | u64 obj_end; | |
02c74fba AE |
2715 | u64 img_xferred; |
2716 | int img_result; | |
8b3e1a56 AE |
2717 | |
2718 | rbd_assert(img_request_child_test(img_request)); | |
2719 | ||
02c74fba AE |
2720 | /* First get what we need from the image request and release it */ |
2721 | ||
8b3e1a56 | 2722 | obj_request = img_request->obj_request; |
02c74fba AE |
2723 | img_xferred = img_request->xferred; |
2724 | img_result = img_request->result; | |
2725 | rbd_img_request_put(img_request); | |
2726 | ||
2727 | /* | |
2728 | * If the overlap has become 0 (most likely because the | |
2729 | * image has been flattened) we need to re-submit the | |
2730 | * original request. | |
2731 | */ | |
a9e8ba2c AE |
2732 | rbd_assert(obj_request); |
2733 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2734 | rbd_dev = obj_request->img_request->rbd_dev; |
2735 | if (!rbd_dev->parent_overlap) { | |
2736 | struct ceph_osd_client *osdc; | |
2737 | ||
2738 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2739 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2740 | if (!img_result) | |
2741 | return; | |
2742 | } | |
a9e8ba2c | 2743 | |
02c74fba | 2744 | obj_request->result = img_result; |
a9e8ba2c AE |
2745 | if (obj_request->result) |
2746 | goto out; | |
2747 | ||
2748 | /* | |
2749 | * We need to zero anything beyond the parent overlap | |
2750 | * boundary. Since rbd_img_obj_request_read_callback() | |
2751 | * will zero anything beyond the end of a short read, an | |
2752 | * easy way to do this is to pretend the data from the | |
2753 | * parent came up short--ending at the overlap boundary. | |
2754 | */ | |
2755 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2756 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2757 | if (obj_end > rbd_dev->parent_overlap) { |
2758 | u64 xferred = 0; | |
2759 | ||
2760 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2761 | xferred = rbd_dev->parent_overlap - | |
2762 | obj_request->img_offset; | |
8b3e1a56 | 2763 | |
02c74fba | 2764 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2765 | } else { |
02c74fba | 2766 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2767 | } |
2768 | out: | |
8b3e1a56 AE |
2769 | rbd_img_obj_request_read_callback(obj_request); |
2770 | rbd_obj_request_complete(obj_request); | |
2771 | } | |
2772 | ||
2773 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2774 | { | |
8b3e1a56 AE |
2775 | struct rbd_img_request *img_request; |
2776 | int result; | |
2777 | ||
2778 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2779 | rbd_assert(obj_request->img_request != NULL); | |
2780 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 2781 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 2782 | |
8b3e1a56 | 2783 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 2784 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 2785 | obj_request->img_offset, |
e93f3152 | 2786 | obj_request->length); |
8b3e1a56 AE |
2787 | result = -ENOMEM; |
2788 | if (!img_request) | |
2789 | goto out_err; | |
2790 | ||
5b2ab72d AE |
2791 | if (obj_request->type == OBJ_REQUEST_BIO) |
2792 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
2793 | obj_request->bio_list); | |
2794 | else | |
2795 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
2796 | obj_request->pages); | |
8b3e1a56 AE |
2797 | if (result) |
2798 | goto out_err; | |
2799 | ||
2800 | img_request->callback = rbd_img_parent_read_callback; | |
2801 | result = rbd_img_request_submit(img_request); | |
2802 | if (result) | |
2803 | goto out_err; | |
2804 | ||
2805 | return; | |
2806 | out_err: | |
2807 | if (img_request) | |
2808 | rbd_img_request_put(img_request); | |
2809 | obj_request->result = result; | |
2810 | obj_request->xferred = 0; | |
2811 | obj_request_done_set(obj_request); | |
2812 | } | |
bf0d5f50 | 2813 | |
cc4a38bd | 2814 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2815 | { |
2816 | struct rbd_obj_request *obj_request; | |
2169238d | 2817 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2818 | int ret; |
2819 | ||
2820 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2821 | OBJ_REQUEST_NODATA); | |
2822 | if (!obj_request) | |
2823 | return -ENOMEM; | |
2824 | ||
2825 | ret = -ENOMEM; | |
430c28c3 | 2826 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2827 | if (!obj_request->osd_req) |
2828 | goto out; | |
2169238d | 2829 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 2830 | |
c99d2d4a | 2831 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2832 | notify_id, 0, 0); |
9d4df01f | 2833 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2834 | |
b8d70035 | 2835 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 2836 | out: |
cf81b60e AE |
2837 | if (ret) |
2838 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2839 | |
2840 | return ret; | |
2841 | } | |
2842 | ||
2843 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2844 | { | |
2845 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 2846 | int ret; |
b8d70035 AE |
2847 | |
2848 | if (!rbd_dev) | |
2849 | return; | |
2850 | ||
37206ee5 | 2851 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2852 | rbd_dev->header_name, (unsigned long long)notify_id, |
2853 | (unsigned int)opcode); | |
e627db08 AE |
2854 | ret = rbd_dev_refresh(rbd_dev); |
2855 | if (ret) | |
2856 | rbd_warn(rbd_dev, ": header refresh error (%d)\n", ret); | |
b8d70035 | 2857 | |
cc4a38bd | 2858 | rbd_obj_notify_ack(rbd_dev, notify_id); |
b8d70035 AE |
2859 | } |
2860 | ||
9969ebc5 AE |
2861 | /* |
2862 | * Request sync osd watch/unwatch. The value of "start" determines | |
2863 | * whether a watch request is being initiated or torn down. | |
2864 | */ | |
1f3ef788 | 2865 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start) |
9969ebc5 AE |
2866 | { |
2867 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2868 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2869 | int ret; |
2870 | ||
2871 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2872 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2873 | ||
2874 | if (start) { | |
3c663bbd | 2875 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2876 | &rbd_dev->watch_event); |
2877 | if (ret < 0) | |
2878 | return ret; | |
8eb87565 | 2879 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2880 | } |
2881 | ||
2882 | ret = -ENOMEM; | |
2883 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2884 | OBJ_REQUEST_NODATA); | |
2885 | if (!obj_request) | |
2886 | goto out_cancel; | |
2887 | ||
430c28c3 AE |
2888 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2889 | if (!obj_request->osd_req) | |
2890 | goto out_cancel; | |
2891 | ||
8eb87565 | 2892 | if (start) |
975241af | 2893 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2894 | else |
6977c3f9 | 2895 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2896 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2897 | |
2898 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
1f3ef788 | 2899 | rbd_dev->watch_event->cookie, 0, start ? 1 : 0); |
9d4df01f | 2900 | rbd_osd_req_format_write(obj_request); |
2169238d | 2901 | |
9969ebc5 AE |
2902 | ret = rbd_obj_request_submit(osdc, obj_request); |
2903 | if (ret) | |
2904 | goto out_cancel; | |
2905 | ret = rbd_obj_request_wait(obj_request); | |
2906 | if (ret) | |
2907 | goto out_cancel; | |
9969ebc5 AE |
2908 | ret = obj_request->result; |
2909 | if (ret) | |
2910 | goto out_cancel; | |
2911 | ||
8eb87565 AE |
2912 | /* |
2913 | * A watch request is set to linger, so the underlying osd | |
2914 | * request won't go away until we unregister it. We retain | |
2915 | * a pointer to the object request during that time (in | |
2916 | * rbd_dev->watch_request), so we'll keep a reference to | |
2917 | * it. We'll drop that reference (below) after we've | |
2918 | * unregistered it. | |
2919 | */ | |
2920 | if (start) { | |
2921 | rbd_dev->watch_request = obj_request; | |
2922 | ||
2923 | return 0; | |
2924 | } | |
2925 | ||
2926 | /* We have successfully torn down the watch request */ | |
2927 | ||
2928 | rbd_obj_request_put(rbd_dev->watch_request); | |
2929 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2930 | out_cancel: |
2931 | /* Cancel the event if we're tearing down, or on error */ | |
2932 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2933 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2934 | if (obj_request) |
2935 | rbd_obj_request_put(obj_request); | |
2936 | ||
2937 | return ret; | |
2938 | } | |
2939 | ||
36be9a76 | 2940 | /* |
f40eb349 AE |
2941 | * Synchronous osd object method call. Returns the number of bytes |
2942 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2943 | */ |
2944 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2945 | const char *object_name, | |
2946 | const char *class_name, | |
2947 | const char *method_name, | |
4157976b | 2948 | const void *outbound, |
36be9a76 | 2949 | size_t outbound_size, |
4157976b | 2950 | void *inbound, |
e2a58ee5 | 2951 | size_t inbound_size) |
36be9a76 | 2952 | { |
2169238d | 2953 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2954 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2955 | struct page **pages; |
2956 | u32 page_count; | |
2957 | int ret; | |
2958 | ||
2959 | /* | |
6010a451 AE |
2960 | * Method calls are ultimately read operations. The result |
2961 | * should placed into the inbound buffer provided. They | |
2962 | * also supply outbound data--parameters for the object | |
2963 | * method. Currently if this is present it will be a | |
2964 | * snapshot id. | |
36be9a76 | 2965 | */ |
57385b51 | 2966 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2967 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2968 | if (IS_ERR(pages)) | |
2969 | return PTR_ERR(pages); | |
2970 | ||
2971 | ret = -ENOMEM; | |
6010a451 | 2972 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2973 | OBJ_REQUEST_PAGES); |
2974 | if (!obj_request) | |
2975 | goto out; | |
2976 | ||
2977 | obj_request->pages = pages; | |
2978 | obj_request->page_count = page_count; | |
2979 | ||
430c28c3 | 2980 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2981 | if (!obj_request->osd_req) |
2982 | goto out; | |
2983 | ||
c99d2d4a | 2984 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2985 | class_name, method_name); |
2986 | if (outbound_size) { | |
2987 | struct ceph_pagelist *pagelist; | |
2988 | ||
2989 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2990 | if (!pagelist) | |
2991 | goto out; | |
2992 | ||
2993 | ceph_pagelist_init(pagelist); | |
2994 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
2995 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
2996 | pagelist); | |
2997 | } | |
a4ce40a9 AE |
2998 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
2999 | obj_request->pages, inbound_size, | |
44cd188d | 3000 | 0, false, false); |
9d4df01f | 3001 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3002 | |
36be9a76 AE |
3003 | ret = rbd_obj_request_submit(osdc, obj_request); |
3004 | if (ret) | |
3005 | goto out; | |
3006 | ret = rbd_obj_request_wait(obj_request); | |
3007 | if (ret) | |
3008 | goto out; | |
3009 | ||
3010 | ret = obj_request->result; | |
3011 | if (ret < 0) | |
3012 | goto out; | |
57385b51 AE |
3013 | |
3014 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3015 | ret = (int)obj_request->xferred; | |
903bb32e | 3016 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3017 | out: |
3018 | if (obj_request) | |
3019 | rbd_obj_request_put(obj_request); | |
3020 | else | |
3021 | ceph_release_page_vector(pages, page_count); | |
3022 | ||
3023 | return ret; | |
3024 | } | |
3025 | ||
bf0d5f50 | 3026 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 3027 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
3028 | { |
3029 | struct rbd_device *rbd_dev = q->queuedata; | |
3030 | bool read_only = rbd_dev->mapping.read_only; | |
3031 | struct request *rq; | |
3032 | int result; | |
3033 | ||
3034 | while ((rq = blk_fetch_request(q))) { | |
3035 | bool write_request = rq_data_dir(rq) == WRITE; | |
3036 | struct rbd_img_request *img_request; | |
3037 | u64 offset; | |
3038 | u64 length; | |
3039 | ||
3040 | /* Ignore any non-FS requests that filter through. */ | |
3041 | ||
3042 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
3043 | dout("%s: non-fs request type %d\n", __func__, |
3044 | (int) rq->cmd_type); | |
3045 | __blk_end_request_all(rq, 0); | |
3046 | continue; | |
3047 | } | |
3048 | ||
3049 | /* Ignore/skip any zero-length requests */ | |
3050 | ||
3051 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
3052 | length = (u64) blk_rq_bytes(rq); | |
3053 | ||
3054 | if (!length) { | |
3055 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
3056 | __blk_end_request_all(rq, 0); |
3057 | continue; | |
3058 | } | |
3059 | ||
3060 | spin_unlock_irq(q->queue_lock); | |
3061 | ||
3062 | /* Disallow writes to a read-only device */ | |
3063 | ||
3064 | if (write_request) { | |
3065 | result = -EROFS; | |
3066 | if (read_only) | |
3067 | goto end_request; | |
3068 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
3069 | } | |
3070 | ||
6d292906 AE |
3071 | /* |
3072 | * Quit early if the mapped snapshot no longer | |
3073 | * exists. It's still possible the snapshot will | |
3074 | * have disappeared by the time our request arrives | |
3075 | * at the osd, but there's no sense in sending it if | |
3076 | * we already know. | |
3077 | */ | |
3078 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
3079 | dout("request for non-existent snapshot"); |
3080 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3081 | result = -ENXIO; | |
3082 | goto end_request; | |
3083 | } | |
3084 | ||
bf0d5f50 | 3085 | result = -EINVAL; |
c0cd10db AE |
3086 | if (offset && length > U64_MAX - offset + 1) { |
3087 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
3088 | offset, length); | |
bf0d5f50 | 3089 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 3090 | } |
bf0d5f50 | 3091 | |
00a653e2 AE |
3092 | result = -EIO; |
3093 | if (offset + length > rbd_dev->mapping.size) { | |
3094 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n", | |
3095 | offset, length, rbd_dev->mapping.size); | |
3096 | goto end_request; | |
3097 | } | |
3098 | ||
bf0d5f50 AE |
3099 | result = -ENOMEM; |
3100 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
e93f3152 | 3101 | write_request); |
bf0d5f50 AE |
3102 | if (!img_request) |
3103 | goto end_request; | |
3104 | ||
3105 | img_request->rq = rq; | |
3106 | ||
f1a4739f AE |
3107 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
3108 | rq->bio); | |
bf0d5f50 AE |
3109 | if (!result) |
3110 | result = rbd_img_request_submit(img_request); | |
3111 | if (result) | |
3112 | rbd_img_request_put(img_request); | |
3113 | end_request: | |
3114 | spin_lock_irq(q->queue_lock); | |
3115 | if (result < 0) { | |
7da22d29 AE |
3116 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
3117 | write_request ? "write" : "read", | |
3118 | length, offset, result); | |
3119 | ||
bf0d5f50 AE |
3120 | __blk_end_request_all(rq, result); |
3121 | } | |
3122 | } | |
3123 | } | |
3124 | ||
602adf40 YS |
3125 | /* |
3126 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3127 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3128 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3129 | */ |
3130 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3131 | struct bio_vec *bvec) | |
3132 | { | |
3133 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3134 | sector_t sector_offset; |
3135 | sector_t sectors_per_obj; | |
3136 | sector_t obj_sector_offset; | |
3137 | int ret; | |
3138 | ||
3139 | /* | |
3140 | * Find how far into its rbd object the partition-relative | |
3141 | * bio start sector is to offset relative to the enclosing | |
3142 | * device. | |
3143 | */ | |
3144 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3145 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3146 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3147 | ||
3148 | /* | |
3149 | * Compute the number of bytes from that offset to the end | |
3150 | * of the object. Account for what's already used by the bio. | |
3151 | */ | |
3152 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3153 | if (ret > bmd->bi_size) | |
3154 | ret -= bmd->bi_size; | |
3155 | else | |
3156 | ret = 0; | |
3157 | ||
3158 | /* | |
3159 | * Don't send back more than was asked for. And if the bio | |
3160 | * was empty, let the whole thing through because: "Note | |
3161 | * that a block device *must* allow a single page to be | |
3162 | * added to an empty bio." | |
3163 | */ | |
3164 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3165 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3166 | ret = (int) bvec->bv_len; | |
3167 | ||
3168 | return ret; | |
602adf40 YS |
3169 | } |
3170 | ||
3171 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3172 | { | |
3173 | struct gendisk *disk = rbd_dev->disk; | |
3174 | ||
3175 | if (!disk) | |
3176 | return; | |
3177 | ||
a0cab924 AE |
3178 | rbd_dev->disk = NULL; |
3179 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3180 | del_gendisk(disk); |
a0cab924 AE |
3181 | if (disk->queue) |
3182 | blk_cleanup_queue(disk->queue); | |
3183 | } | |
602adf40 YS |
3184 | put_disk(disk); |
3185 | } | |
3186 | ||
788e2df3 AE |
3187 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3188 | const char *object_name, | |
7097f8df | 3189 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3190 | |
3191 | { | |
2169238d | 3192 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3193 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3194 | struct page **pages = NULL; |
3195 | u32 page_count; | |
1ceae7ef | 3196 | size_t size; |
788e2df3 AE |
3197 | int ret; |
3198 | ||
3199 | page_count = (u32) calc_pages_for(offset, length); | |
3200 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3201 | if (IS_ERR(pages)) | |
3202 | ret = PTR_ERR(pages); | |
3203 | ||
3204 | ret = -ENOMEM; | |
3205 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3206 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3207 | if (!obj_request) |
3208 | goto out; | |
3209 | ||
3210 | obj_request->pages = pages; | |
3211 | obj_request->page_count = page_count; | |
3212 | ||
430c28c3 | 3213 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
3214 | if (!obj_request->osd_req) |
3215 | goto out; | |
3216 | ||
c99d2d4a AE |
3217 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3218 | offset, length, 0, 0); | |
406e2c9f | 3219 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3220 | obj_request->pages, |
44cd188d AE |
3221 | obj_request->length, |
3222 | obj_request->offset & ~PAGE_MASK, | |
3223 | false, false); | |
9d4df01f | 3224 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3225 | |
788e2df3 AE |
3226 | ret = rbd_obj_request_submit(osdc, obj_request); |
3227 | if (ret) | |
3228 | goto out; | |
3229 | ret = rbd_obj_request_wait(obj_request); | |
3230 | if (ret) | |
3231 | goto out; | |
3232 | ||
3233 | ret = obj_request->result; | |
3234 | if (ret < 0) | |
3235 | goto out; | |
1ceae7ef AE |
3236 | |
3237 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3238 | size = (size_t) obj_request->xferred; | |
903bb32e | 3239 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3240 | rbd_assert(size <= (size_t)INT_MAX); |
3241 | ret = (int)size; | |
788e2df3 AE |
3242 | out: |
3243 | if (obj_request) | |
3244 | rbd_obj_request_put(obj_request); | |
3245 | else | |
3246 | ceph_release_page_vector(pages, page_count); | |
3247 | ||
3248 | return ret; | |
3249 | } | |
3250 | ||
602adf40 | 3251 | /* |
662518b1 AE |
3252 | * Read the complete header for the given rbd device. On successful |
3253 | * return, the rbd_dev->header field will contain up-to-date | |
3254 | * information about the image. | |
602adf40 | 3255 | */ |
99a41ebc | 3256 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3257 | { |
4156d998 | 3258 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3259 | u32 snap_count = 0; |
4156d998 AE |
3260 | u64 names_size = 0; |
3261 | u32 want_count; | |
3262 | int ret; | |
602adf40 | 3263 | |
00f1f36f | 3264 | /* |
4156d998 AE |
3265 | * The complete header will include an array of its 64-bit |
3266 | * snapshot ids, followed by the names of those snapshots as | |
3267 | * a contiguous block of NUL-terminated strings. Note that | |
3268 | * the number of snapshots could change by the time we read | |
3269 | * it in, in which case we re-read it. | |
00f1f36f | 3270 | */ |
4156d998 AE |
3271 | do { |
3272 | size_t size; | |
3273 | ||
3274 | kfree(ondisk); | |
3275 | ||
3276 | size = sizeof (*ondisk); | |
3277 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3278 | size += names_size; | |
3279 | ondisk = kmalloc(size, GFP_KERNEL); | |
3280 | if (!ondisk) | |
662518b1 | 3281 | return -ENOMEM; |
4156d998 | 3282 | |
788e2df3 | 3283 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3284 | 0, size, ondisk); |
4156d998 | 3285 | if (ret < 0) |
662518b1 | 3286 | goto out; |
c0cd10db | 3287 | if ((size_t)ret < size) { |
4156d998 | 3288 | ret = -ENXIO; |
06ecc6cb AE |
3289 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3290 | size, ret); | |
662518b1 | 3291 | goto out; |
4156d998 AE |
3292 | } |
3293 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3294 | ret = -ENXIO; | |
06ecc6cb | 3295 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3296 | goto out; |
81e759fb | 3297 | } |
602adf40 | 3298 | |
4156d998 AE |
3299 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3300 | want_count = snap_count; | |
3301 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3302 | } while (snap_count != want_count); | |
00f1f36f | 3303 | |
662518b1 AE |
3304 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3305 | out: | |
4156d998 AE |
3306 | kfree(ondisk); |
3307 | ||
3308 | return ret; | |
602adf40 YS |
3309 | } |
3310 | ||
15228ede AE |
3311 | /* |
3312 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3313 | * has disappeared from the (just updated) snapshot context. | |
3314 | */ | |
3315 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3316 | { | |
3317 | u64 snap_id; | |
3318 | ||
3319 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3320 | return; | |
3321 | ||
3322 | snap_id = rbd_dev->spec->snap_id; | |
3323 | if (snap_id == CEPH_NOSNAP) | |
3324 | return; | |
3325 | ||
3326 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3327 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3328 | } | |
3329 | ||
cc4a38bd | 3330 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3331 | { |
e627db08 | 3332 | u64 mapping_size; |
1fe5e993 AE |
3333 | int ret; |
3334 | ||
117973fb | 3335 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
e627db08 | 3336 | mapping_size = rbd_dev->mapping.size; |
1fe5e993 | 3337 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb | 3338 | if (rbd_dev->image_format == 1) |
99a41ebc | 3339 | ret = rbd_dev_v1_header_info(rbd_dev); |
117973fb | 3340 | else |
2df3fac7 | 3341 | ret = rbd_dev_v2_header_info(rbd_dev); |
15228ede AE |
3342 | |
3343 | /* If it's a mapped snapshot, validate its EXISTS flag */ | |
3344 | ||
3345 | rbd_exists_validate(rbd_dev); | |
1fe5e993 | 3346 | mutex_unlock(&ctl_mutex); |
00a653e2 AE |
3347 | if (mapping_size != rbd_dev->mapping.size) { |
3348 | sector_t size; | |
3349 | ||
3350 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3351 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3352 | set_capacity(rbd_dev->disk, size); | |
a3fbe5d4 | 3353 | revalidate_disk(rbd_dev->disk); |
00a653e2 | 3354 | } |
1fe5e993 AE |
3355 | |
3356 | return ret; | |
3357 | } | |
3358 | ||
602adf40 YS |
3359 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3360 | { | |
3361 | struct gendisk *disk; | |
3362 | struct request_queue *q; | |
593a9e7b | 3363 | u64 segment_size; |
602adf40 | 3364 | |
602adf40 | 3365 | /* create gendisk info */ |
602adf40 YS |
3366 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
3367 | if (!disk) | |
1fcdb8aa | 3368 | return -ENOMEM; |
602adf40 | 3369 | |
f0f8cef5 | 3370 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3371 | rbd_dev->dev_id); |
602adf40 YS |
3372 | disk->major = rbd_dev->major; |
3373 | disk->first_minor = 0; | |
3374 | disk->fops = &rbd_bd_ops; | |
3375 | disk->private_data = rbd_dev; | |
3376 | ||
bf0d5f50 | 3377 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3378 | if (!q) |
3379 | goto out_disk; | |
029bcbd8 | 3380 | |
593a9e7b AE |
3381 | /* We use the default size, but let's be explicit about it. */ |
3382 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3383 | ||
029bcbd8 | 3384 | /* set io sizes to object size */ |
593a9e7b AE |
3385 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3386 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3387 | blk_queue_max_segment_size(q, segment_size); | |
3388 | blk_queue_io_min(q, segment_size); | |
3389 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3390 | |
602adf40 YS |
3391 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3392 | disk->queue = q; | |
3393 | ||
3394 | q->queuedata = rbd_dev; | |
3395 | ||
3396 | rbd_dev->disk = disk; | |
602adf40 | 3397 | |
602adf40 | 3398 | return 0; |
602adf40 YS |
3399 | out_disk: |
3400 | put_disk(disk); | |
1fcdb8aa AE |
3401 | |
3402 | return -ENOMEM; | |
602adf40 YS |
3403 | } |
3404 | ||
dfc5606d YS |
3405 | /* |
3406 | sysfs | |
3407 | */ | |
3408 | ||
593a9e7b AE |
3409 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3410 | { | |
3411 | return container_of(dev, struct rbd_device, dev); | |
3412 | } | |
3413 | ||
dfc5606d YS |
3414 | static ssize_t rbd_size_show(struct device *dev, |
3415 | struct device_attribute *attr, char *buf) | |
3416 | { | |
593a9e7b | 3417 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3418 | |
fc71d833 AE |
3419 | return sprintf(buf, "%llu\n", |
3420 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3421 | } |
3422 | ||
34b13184 AE |
3423 | /* |
3424 | * Note this shows the features for whatever's mapped, which is not | |
3425 | * necessarily the base image. | |
3426 | */ | |
3427 | static ssize_t rbd_features_show(struct device *dev, | |
3428 | struct device_attribute *attr, char *buf) | |
3429 | { | |
3430 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3431 | ||
3432 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3433 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3434 | } |
3435 | ||
dfc5606d YS |
3436 | static ssize_t rbd_major_show(struct device *dev, |
3437 | struct device_attribute *attr, char *buf) | |
3438 | { | |
593a9e7b | 3439 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3440 | |
fc71d833 AE |
3441 | if (rbd_dev->major) |
3442 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3443 | ||
3444 | return sprintf(buf, "(none)\n"); | |
3445 | ||
dfc5606d YS |
3446 | } |
3447 | ||
3448 | static ssize_t rbd_client_id_show(struct device *dev, | |
3449 | struct device_attribute *attr, char *buf) | |
602adf40 | 3450 | { |
593a9e7b | 3451 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3452 | |
1dbb4399 AE |
3453 | return sprintf(buf, "client%lld\n", |
3454 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3455 | } |
3456 | ||
dfc5606d YS |
3457 | static ssize_t rbd_pool_show(struct device *dev, |
3458 | struct device_attribute *attr, char *buf) | |
602adf40 | 3459 | { |
593a9e7b | 3460 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3461 | |
0d7dbfce | 3462 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3463 | } |
3464 | ||
9bb2f334 AE |
3465 | static ssize_t rbd_pool_id_show(struct device *dev, |
3466 | struct device_attribute *attr, char *buf) | |
3467 | { | |
3468 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3469 | ||
0d7dbfce | 3470 | return sprintf(buf, "%llu\n", |
fc71d833 | 3471 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3472 | } |
3473 | ||
dfc5606d YS |
3474 | static ssize_t rbd_name_show(struct device *dev, |
3475 | struct device_attribute *attr, char *buf) | |
3476 | { | |
593a9e7b | 3477 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3478 | |
a92ffdf8 AE |
3479 | if (rbd_dev->spec->image_name) |
3480 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3481 | ||
3482 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3483 | } |
3484 | ||
589d30e0 AE |
3485 | static ssize_t rbd_image_id_show(struct device *dev, |
3486 | struct device_attribute *attr, char *buf) | |
3487 | { | |
3488 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3489 | ||
0d7dbfce | 3490 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3491 | } |
3492 | ||
34b13184 AE |
3493 | /* |
3494 | * Shows the name of the currently-mapped snapshot (or | |
3495 | * RBD_SNAP_HEAD_NAME for the base image). | |
3496 | */ | |
dfc5606d YS |
3497 | static ssize_t rbd_snap_show(struct device *dev, |
3498 | struct device_attribute *attr, | |
3499 | char *buf) | |
3500 | { | |
593a9e7b | 3501 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3502 | |
0d7dbfce | 3503 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3504 | } |
3505 | ||
86b00e0d AE |
3506 | /* |
3507 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3508 | * for the parent image. If there is no parent, simply shows | |
3509 | * "(no parent image)". | |
3510 | */ | |
3511 | static ssize_t rbd_parent_show(struct device *dev, | |
3512 | struct device_attribute *attr, | |
3513 | char *buf) | |
3514 | { | |
3515 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3516 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3517 | int count; | |
3518 | char *bufp = buf; | |
3519 | ||
3520 | if (!spec) | |
3521 | return sprintf(buf, "(no parent image)\n"); | |
3522 | ||
3523 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3524 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3525 | if (count < 0) | |
3526 | return count; | |
3527 | bufp += count; | |
3528 | ||
3529 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3530 | spec->image_name ? spec->image_name : "(unknown)"); | |
3531 | if (count < 0) | |
3532 | return count; | |
3533 | bufp += count; | |
3534 | ||
3535 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3536 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3537 | if (count < 0) | |
3538 | return count; | |
3539 | bufp += count; | |
3540 | ||
3541 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3542 | if (count < 0) | |
3543 | return count; | |
3544 | bufp += count; | |
3545 | ||
3546 | return (ssize_t) (bufp - buf); | |
3547 | } | |
3548 | ||
dfc5606d YS |
3549 | static ssize_t rbd_image_refresh(struct device *dev, |
3550 | struct device_attribute *attr, | |
3551 | const char *buf, | |
3552 | size_t size) | |
3553 | { | |
593a9e7b | 3554 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3555 | int ret; |
602adf40 | 3556 | |
cc4a38bd | 3557 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 AE |
3558 | if (ret) |
3559 | rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret); | |
b813623a AE |
3560 | |
3561 | return ret < 0 ? ret : size; | |
dfc5606d | 3562 | } |
602adf40 | 3563 | |
dfc5606d | 3564 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3565 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
3566 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
3567 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
3568 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3569 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3570 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3571 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3572 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3573 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3574 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3575 | |
3576 | static struct attribute *rbd_attrs[] = { | |
3577 | &dev_attr_size.attr, | |
34b13184 | 3578 | &dev_attr_features.attr, |
dfc5606d YS |
3579 | &dev_attr_major.attr, |
3580 | &dev_attr_client_id.attr, | |
3581 | &dev_attr_pool.attr, | |
9bb2f334 | 3582 | &dev_attr_pool_id.attr, |
dfc5606d | 3583 | &dev_attr_name.attr, |
589d30e0 | 3584 | &dev_attr_image_id.attr, |
dfc5606d | 3585 | &dev_attr_current_snap.attr, |
86b00e0d | 3586 | &dev_attr_parent.attr, |
dfc5606d | 3587 | &dev_attr_refresh.attr, |
dfc5606d YS |
3588 | NULL |
3589 | }; | |
3590 | ||
3591 | static struct attribute_group rbd_attr_group = { | |
3592 | .attrs = rbd_attrs, | |
3593 | }; | |
3594 | ||
3595 | static const struct attribute_group *rbd_attr_groups[] = { | |
3596 | &rbd_attr_group, | |
3597 | NULL | |
3598 | }; | |
3599 | ||
3600 | static void rbd_sysfs_dev_release(struct device *dev) | |
3601 | { | |
3602 | } | |
3603 | ||
3604 | static struct device_type rbd_device_type = { | |
3605 | .name = "rbd", | |
3606 | .groups = rbd_attr_groups, | |
3607 | .release = rbd_sysfs_dev_release, | |
3608 | }; | |
3609 | ||
8b8fb99c AE |
3610 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3611 | { | |
3612 | kref_get(&spec->kref); | |
3613 | ||
3614 | return spec; | |
3615 | } | |
3616 | ||
3617 | static void rbd_spec_free(struct kref *kref); | |
3618 | static void rbd_spec_put(struct rbd_spec *spec) | |
3619 | { | |
3620 | if (spec) | |
3621 | kref_put(&spec->kref, rbd_spec_free); | |
3622 | } | |
3623 | ||
3624 | static struct rbd_spec *rbd_spec_alloc(void) | |
3625 | { | |
3626 | struct rbd_spec *spec; | |
3627 | ||
3628 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3629 | if (!spec) | |
3630 | return NULL; | |
3631 | kref_init(&spec->kref); | |
3632 | ||
8b8fb99c AE |
3633 | return spec; |
3634 | } | |
3635 | ||
3636 | static void rbd_spec_free(struct kref *kref) | |
3637 | { | |
3638 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3639 | ||
3640 | kfree(spec->pool_name); | |
3641 | kfree(spec->image_id); | |
3642 | kfree(spec->image_name); | |
3643 | kfree(spec->snap_name); | |
3644 | kfree(spec); | |
3645 | } | |
3646 | ||
cc344fa1 | 3647 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3648 | struct rbd_spec *spec) |
3649 | { | |
3650 | struct rbd_device *rbd_dev; | |
3651 | ||
3652 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3653 | if (!rbd_dev) | |
3654 | return NULL; | |
3655 | ||
3656 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3657 | rbd_dev->flags = 0; |
a2acd00e | 3658 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 3659 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
3660 | init_rwsem(&rbd_dev->header_rwsem); |
3661 | ||
3662 | rbd_dev->spec = spec; | |
3663 | rbd_dev->rbd_client = rbdc; | |
3664 | ||
0903e875 AE |
3665 | /* Initialize the layout used for all rbd requests */ |
3666 | ||
3667 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3668 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3669 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3670 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3671 | ||
c53d5893 AE |
3672 | return rbd_dev; |
3673 | } | |
3674 | ||
3675 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3676 | { | |
c53d5893 AE |
3677 | rbd_put_client(rbd_dev->rbd_client); |
3678 | rbd_spec_put(rbd_dev->spec); | |
3679 | kfree(rbd_dev); | |
3680 | } | |
3681 | ||
9d475de5 AE |
3682 | /* |
3683 | * Get the size and object order for an image snapshot, or if | |
3684 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3685 | * image. | |
3686 | */ | |
3687 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3688 | u8 *order, u64 *snap_size) | |
3689 | { | |
3690 | __le64 snapid = cpu_to_le64(snap_id); | |
3691 | int ret; | |
3692 | struct { | |
3693 | u8 order; | |
3694 | __le64 size; | |
3695 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3696 | ||
36be9a76 | 3697 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3698 | "rbd", "get_size", |
4157976b | 3699 | &snapid, sizeof (snapid), |
e2a58ee5 | 3700 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3701 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3702 | if (ret < 0) |
3703 | return ret; | |
57385b51 AE |
3704 | if (ret < sizeof (size_buf)) |
3705 | return -ERANGE; | |
9d475de5 | 3706 | |
c86f86e9 AE |
3707 | if (order) |
3708 | *order = size_buf.order; | |
9d475de5 AE |
3709 | *snap_size = le64_to_cpu(size_buf.size); |
3710 | ||
3711 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
57385b51 AE |
3712 | (unsigned long long)snap_id, (unsigned int)*order, |
3713 | (unsigned long long)*snap_size); | |
9d475de5 AE |
3714 | |
3715 | return 0; | |
3716 | } | |
3717 | ||
3718 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3719 | { | |
3720 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3721 | &rbd_dev->header.obj_order, | |
3722 | &rbd_dev->header.image_size); | |
3723 | } | |
3724 | ||
1e130199 AE |
3725 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3726 | { | |
3727 | void *reply_buf; | |
3728 | int ret; | |
3729 | void *p; | |
3730 | ||
3731 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3732 | if (!reply_buf) | |
3733 | return -ENOMEM; | |
3734 | ||
36be9a76 | 3735 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3736 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3737 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3738 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3739 | if (ret < 0) |
3740 | goto out; | |
3741 | ||
3742 | p = reply_buf; | |
3743 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3744 | p + ret, NULL, GFP_NOIO); |
3745 | ret = 0; | |
1e130199 AE |
3746 | |
3747 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3748 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3749 | rbd_dev->header.object_prefix = NULL; | |
3750 | } else { | |
3751 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3752 | } | |
1e130199 AE |
3753 | out: |
3754 | kfree(reply_buf); | |
3755 | ||
3756 | return ret; | |
3757 | } | |
3758 | ||
b1b5402a AE |
3759 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3760 | u64 *snap_features) | |
3761 | { | |
3762 | __le64 snapid = cpu_to_le64(snap_id); | |
3763 | struct { | |
3764 | __le64 features; | |
3765 | __le64 incompat; | |
4157976b | 3766 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3767 | u64 incompat; |
b1b5402a AE |
3768 | int ret; |
3769 | ||
36be9a76 | 3770 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3771 | "rbd", "get_features", |
4157976b | 3772 | &snapid, sizeof (snapid), |
e2a58ee5 | 3773 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3774 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3775 | if (ret < 0) |
3776 | return ret; | |
57385b51 AE |
3777 | if (ret < sizeof (features_buf)) |
3778 | return -ERANGE; | |
d889140c AE |
3779 | |
3780 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3781 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3782 | return -ENXIO; |
d889140c | 3783 | |
b1b5402a AE |
3784 | *snap_features = le64_to_cpu(features_buf.features); |
3785 | ||
3786 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3787 | (unsigned long long)snap_id, |
3788 | (unsigned long long)*snap_features, | |
3789 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3790 | |
3791 | return 0; | |
3792 | } | |
3793 | ||
3794 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3795 | { | |
3796 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3797 | &rbd_dev->header.features); | |
3798 | } | |
3799 | ||
86b00e0d AE |
3800 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3801 | { | |
3802 | struct rbd_spec *parent_spec; | |
3803 | size_t size; | |
3804 | void *reply_buf = NULL; | |
3805 | __le64 snapid; | |
3806 | void *p; | |
3807 | void *end; | |
642a2537 | 3808 | u64 pool_id; |
86b00e0d AE |
3809 | char *image_id; |
3810 | u64 overlap; | |
86b00e0d AE |
3811 | int ret; |
3812 | ||
3813 | parent_spec = rbd_spec_alloc(); | |
3814 | if (!parent_spec) | |
3815 | return -ENOMEM; | |
3816 | ||
3817 | size = sizeof (__le64) + /* pool_id */ | |
3818 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3819 | sizeof (__le64) + /* snap_id */ | |
3820 | sizeof (__le64); /* overlap */ | |
3821 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3822 | if (!reply_buf) { | |
3823 | ret = -ENOMEM; | |
3824 | goto out_err; | |
3825 | } | |
3826 | ||
3827 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3828 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3829 | "rbd", "get_parent", |
4157976b | 3830 | &snapid, sizeof (snapid), |
e2a58ee5 | 3831 | reply_buf, size); |
36be9a76 | 3832 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3833 | if (ret < 0) |
3834 | goto out_err; | |
3835 | ||
86b00e0d | 3836 | p = reply_buf; |
57385b51 AE |
3837 | end = reply_buf + ret; |
3838 | ret = -ERANGE; | |
642a2537 | 3839 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
3840 | if (pool_id == CEPH_NOPOOL) { |
3841 | /* | |
3842 | * Either the parent never existed, or we have | |
3843 | * record of it but the image got flattened so it no | |
3844 | * longer has a parent. When the parent of a | |
3845 | * layered image disappears we immediately set the | |
3846 | * overlap to 0. The effect of this is that all new | |
3847 | * requests will be treated as if the image had no | |
3848 | * parent. | |
3849 | */ | |
3850 | if (rbd_dev->parent_overlap) { | |
3851 | rbd_dev->parent_overlap = 0; | |
3852 | smp_mb(); | |
3853 | rbd_dev_parent_put(rbd_dev); | |
3854 | pr_info("%s: clone image has been flattened\n", | |
3855 | rbd_dev->disk->disk_name); | |
3856 | } | |
3857 | ||
86b00e0d | 3858 | goto out; /* No parent? No problem. */ |
392a9dad | 3859 | } |
86b00e0d | 3860 | |
0903e875 AE |
3861 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3862 | ||
3863 | ret = -EIO; | |
642a2537 | 3864 | if (pool_id > (u64)U32_MAX) { |
c0cd10db | 3865 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", |
642a2537 | 3866 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 3867 | goto out_err; |
c0cd10db | 3868 | } |
642a2537 | 3869 | parent_spec->pool_id = pool_id; |
0903e875 | 3870 | |
979ed480 | 3871 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3872 | if (IS_ERR(image_id)) { |
3873 | ret = PTR_ERR(image_id); | |
3874 | goto out_err; | |
3875 | } | |
3876 | parent_spec->image_id = image_id; | |
3877 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
3878 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
3879 | ||
70cf49cf | 3880 | if (overlap) { |
642a2537 | 3881 | rbd_spec_put(rbd_dev->parent_spec); |
70cf49cf AE |
3882 | rbd_dev->parent_spec = parent_spec; |
3883 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3884 | rbd_dev->parent_overlap = overlap; | |
3885 | } else { | |
3886 | rbd_warn(rbd_dev, "ignoring parent of clone with overlap 0\n"); | |
3887 | } | |
86b00e0d AE |
3888 | out: |
3889 | ret = 0; | |
3890 | out_err: | |
3891 | kfree(reply_buf); | |
3892 | rbd_spec_put(parent_spec); | |
3893 | ||
3894 | return ret; | |
3895 | } | |
3896 | ||
cc070d59 AE |
3897 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3898 | { | |
3899 | struct { | |
3900 | __le64 stripe_unit; | |
3901 | __le64 stripe_count; | |
3902 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3903 | size_t size = sizeof (striping_info_buf); | |
3904 | void *p; | |
3905 | u64 obj_size; | |
3906 | u64 stripe_unit; | |
3907 | u64 stripe_count; | |
3908 | int ret; | |
3909 | ||
3910 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3911 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 3912 | (char *)&striping_info_buf, size); |
cc070d59 AE |
3913 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
3914 | if (ret < 0) | |
3915 | return ret; | |
3916 | if (ret < size) | |
3917 | return -ERANGE; | |
3918 | ||
3919 | /* | |
3920 | * We don't actually support the "fancy striping" feature | |
3921 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3922 | * defaults the behavior is the same as before. So find | |
3923 | * out, and only fail if the image has non-default values. | |
3924 | */ | |
3925 | ret = -EINVAL; | |
3926 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3927 | p = &striping_info_buf; | |
3928 | stripe_unit = ceph_decode_64(&p); | |
3929 | if (stripe_unit != obj_size) { | |
3930 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
3931 | "(got %llu want %llu)", | |
3932 | stripe_unit, obj_size); | |
3933 | return -EINVAL; | |
3934 | } | |
3935 | stripe_count = ceph_decode_64(&p); | |
3936 | if (stripe_count != 1) { | |
3937 | rbd_warn(rbd_dev, "unsupported stripe count " | |
3938 | "(got %llu want 1)", stripe_count); | |
3939 | return -EINVAL; | |
3940 | } | |
500d0c0f AE |
3941 | rbd_dev->header.stripe_unit = stripe_unit; |
3942 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
3943 | |
3944 | return 0; | |
3945 | } | |
3946 | ||
9e15b77d AE |
3947 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3948 | { | |
3949 | size_t image_id_size; | |
3950 | char *image_id; | |
3951 | void *p; | |
3952 | void *end; | |
3953 | size_t size; | |
3954 | void *reply_buf = NULL; | |
3955 | size_t len = 0; | |
3956 | char *image_name = NULL; | |
3957 | int ret; | |
3958 | ||
3959 | rbd_assert(!rbd_dev->spec->image_name); | |
3960 | ||
69e7a02f AE |
3961 | len = strlen(rbd_dev->spec->image_id); |
3962 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3963 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3964 | if (!image_id) | |
3965 | return NULL; | |
3966 | ||
3967 | p = image_id; | |
4157976b | 3968 | end = image_id + image_id_size; |
57385b51 | 3969 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
3970 | |
3971 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3972 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3973 | if (!reply_buf) | |
3974 | goto out; | |
3975 | ||
36be9a76 | 3976 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3977 | "rbd", "dir_get_name", |
3978 | image_id, image_id_size, | |
e2a58ee5 | 3979 | reply_buf, size); |
9e15b77d AE |
3980 | if (ret < 0) |
3981 | goto out; | |
3982 | p = reply_buf; | |
f40eb349 AE |
3983 | end = reply_buf + ret; |
3984 | ||
9e15b77d AE |
3985 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
3986 | if (IS_ERR(image_name)) | |
3987 | image_name = NULL; | |
3988 | else | |
3989 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3990 | out: | |
3991 | kfree(reply_buf); | |
3992 | kfree(image_id); | |
3993 | ||
3994 | return image_name; | |
3995 | } | |
3996 | ||
2ad3d716 AE |
3997 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
3998 | { | |
3999 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4000 | const char *snap_name; | |
4001 | u32 which = 0; | |
4002 | ||
4003 | /* Skip over names until we find the one we are looking for */ | |
4004 | ||
4005 | snap_name = rbd_dev->header.snap_names; | |
4006 | while (which < snapc->num_snaps) { | |
4007 | if (!strcmp(name, snap_name)) | |
4008 | return snapc->snaps[which]; | |
4009 | snap_name += strlen(snap_name) + 1; | |
4010 | which++; | |
4011 | } | |
4012 | return CEPH_NOSNAP; | |
4013 | } | |
4014 | ||
4015 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4016 | { | |
4017 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4018 | u32 which; | |
4019 | bool found = false; | |
4020 | u64 snap_id; | |
4021 | ||
4022 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4023 | const char *snap_name; | |
4024 | ||
4025 | snap_id = snapc->snaps[which]; | |
4026 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
4027 | if (IS_ERR(snap_name)) | |
4028 | break; | |
4029 | found = !strcmp(name, snap_name); | |
4030 | kfree(snap_name); | |
4031 | } | |
4032 | return found ? snap_id : CEPH_NOSNAP; | |
4033 | } | |
4034 | ||
4035 | /* | |
4036 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4037 | * no snapshot by that name is found, or if an error occurs. | |
4038 | */ | |
4039 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4040 | { | |
4041 | if (rbd_dev->image_format == 1) | |
4042 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4043 | ||
4044 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4045 | } | |
4046 | ||
9e15b77d | 4047 | /* |
2e9f7f1c AE |
4048 | * When an rbd image has a parent image, it is identified by the |
4049 | * pool, image, and snapshot ids (not names). This function fills | |
4050 | * in the names for those ids. (It's OK if we can't figure out the | |
4051 | * name for an image id, but the pool and snapshot ids should always | |
4052 | * exist and have names.) All names in an rbd spec are dynamically | |
4053 | * allocated. | |
e1d4213f AE |
4054 | * |
4055 | * When an image being mapped (not a parent) is probed, we have the | |
4056 | * pool name and pool id, image name and image id, and the snapshot | |
4057 | * name. The only thing we're missing is the snapshot id. | |
9e15b77d | 4058 | */ |
2e9f7f1c | 4059 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 4060 | { |
2e9f7f1c AE |
4061 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4062 | struct rbd_spec *spec = rbd_dev->spec; | |
4063 | const char *pool_name; | |
4064 | const char *image_name; | |
4065 | const char *snap_name; | |
9e15b77d AE |
4066 | int ret; |
4067 | ||
e1d4213f AE |
4068 | /* |
4069 | * An image being mapped will have the pool name (etc.), but | |
4070 | * we need to look up the snapshot id. | |
4071 | */ | |
2e9f7f1c AE |
4072 | if (spec->pool_name) { |
4073 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
2ad3d716 | 4074 | u64 snap_id; |
e1d4213f | 4075 | |
2ad3d716 AE |
4076 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); |
4077 | if (snap_id == CEPH_NOSNAP) | |
e1d4213f | 4078 | return -ENOENT; |
2ad3d716 | 4079 | spec->snap_id = snap_id; |
e1d4213f | 4080 | } else { |
2e9f7f1c | 4081 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
4082 | } |
4083 | ||
4084 | return 0; | |
4085 | } | |
9e15b77d | 4086 | |
2e9f7f1c | 4087 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4088 | |
2e9f7f1c AE |
4089 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4090 | if (!pool_name) { | |
4091 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4092 | return -EIO; |
4093 | } | |
2e9f7f1c AE |
4094 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4095 | if (!pool_name) | |
9e15b77d AE |
4096 | return -ENOMEM; |
4097 | ||
4098 | /* Fetch the image name; tolerate failure here */ | |
4099 | ||
2e9f7f1c AE |
4100 | image_name = rbd_dev_image_name(rbd_dev); |
4101 | if (!image_name) | |
06ecc6cb | 4102 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4103 | |
2e9f7f1c | 4104 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 4105 | |
2e9f7f1c | 4106 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
2e9f7f1c AE |
4107 | if (!snap_name) { |
4108 | ret = -ENOMEM; | |
9e15b77d | 4109 | goto out_err; |
2e9f7f1c AE |
4110 | } |
4111 | ||
4112 | spec->pool_name = pool_name; | |
4113 | spec->image_name = image_name; | |
4114 | spec->snap_name = snap_name; | |
9e15b77d AE |
4115 | |
4116 | return 0; | |
4117 | out_err: | |
2e9f7f1c AE |
4118 | kfree(image_name); |
4119 | kfree(pool_name); | |
9e15b77d AE |
4120 | |
4121 | return ret; | |
4122 | } | |
4123 | ||
cc4a38bd | 4124 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4125 | { |
4126 | size_t size; | |
4127 | int ret; | |
4128 | void *reply_buf; | |
4129 | void *p; | |
4130 | void *end; | |
4131 | u64 seq; | |
4132 | u32 snap_count; | |
4133 | struct ceph_snap_context *snapc; | |
4134 | u32 i; | |
4135 | ||
4136 | /* | |
4137 | * We'll need room for the seq value (maximum snapshot id), | |
4138 | * snapshot count, and array of that many snapshot ids. | |
4139 | * For now we have a fixed upper limit on the number we're | |
4140 | * prepared to receive. | |
4141 | */ | |
4142 | size = sizeof (__le64) + sizeof (__le32) + | |
4143 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4144 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4145 | if (!reply_buf) | |
4146 | return -ENOMEM; | |
4147 | ||
36be9a76 | 4148 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4149 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4150 | reply_buf, size); |
36be9a76 | 4151 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4152 | if (ret < 0) |
4153 | goto out; | |
4154 | ||
35d489f9 | 4155 | p = reply_buf; |
57385b51 AE |
4156 | end = reply_buf + ret; |
4157 | ret = -ERANGE; | |
35d489f9 AE |
4158 | ceph_decode_64_safe(&p, end, seq, out); |
4159 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4160 | ||
4161 | /* | |
4162 | * Make sure the reported number of snapshot ids wouldn't go | |
4163 | * beyond the end of our buffer. But before checking that, | |
4164 | * make sure the computed size of the snapshot context we | |
4165 | * allocate is representable in a size_t. | |
4166 | */ | |
4167 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4168 | / sizeof (u64)) { | |
4169 | ret = -EINVAL; | |
4170 | goto out; | |
4171 | } | |
4172 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4173 | goto out; | |
468521c1 | 4174 | ret = 0; |
35d489f9 | 4175 | |
812164f8 | 4176 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4177 | if (!snapc) { |
4178 | ret = -ENOMEM; | |
4179 | goto out; | |
4180 | } | |
35d489f9 | 4181 | snapc->seq = seq; |
35d489f9 AE |
4182 | for (i = 0; i < snap_count; i++) |
4183 | snapc->snaps[i] = ceph_decode_64(&p); | |
4184 | ||
49ece554 | 4185 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4186 | rbd_dev->header.snapc = snapc; |
4187 | ||
4188 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4189 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4190 | out: |
4191 | kfree(reply_buf); | |
4192 | ||
57385b51 | 4193 | return ret; |
35d489f9 AE |
4194 | } |
4195 | ||
54cac61f AE |
4196 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4197 | u64 snap_id) | |
b8b1e2db AE |
4198 | { |
4199 | size_t size; | |
4200 | void *reply_buf; | |
54cac61f | 4201 | __le64 snapid; |
b8b1e2db AE |
4202 | int ret; |
4203 | void *p; | |
4204 | void *end; | |
b8b1e2db AE |
4205 | char *snap_name; |
4206 | ||
4207 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4208 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4209 | if (!reply_buf) | |
4210 | return ERR_PTR(-ENOMEM); | |
4211 | ||
54cac61f | 4212 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4213 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4214 | "rbd", "get_snapshot_name", |
54cac61f | 4215 | &snapid, sizeof (snapid), |
e2a58ee5 | 4216 | reply_buf, size); |
36be9a76 | 4217 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4218 | if (ret < 0) { |
4219 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4220 | goto out; |
f40eb349 | 4221 | } |
b8b1e2db AE |
4222 | |
4223 | p = reply_buf; | |
f40eb349 | 4224 | end = reply_buf + ret; |
e5c35534 | 4225 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4226 | if (IS_ERR(snap_name)) |
b8b1e2db | 4227 | goto out; |
b8b1e2db | 4228 | |
f40eb349 | 4229 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4230 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4231 | out: |
4232 | kfree(reply_buf); | |
4233 | ||
f40eb349 | 4234 | return snap_name; |
b8b1e2db AE |
4235 | } |
4236 | ||
2df3fac7 | 4237 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4238 | { |
2df3fac7 | 4239 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4240 | int ret; |
117973fb AE |
4241 | |
4242 | down_write(&rbd_dev->header_rwsem); | |
4243 | ||
2df3fac7 AE |
4244 | if (first_time) { |
4245 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4246 | if (ret) | |
4247 | goto out; | |
4248 | } | |
4249 | ||
642a2537 AE |
4250 | /* |
4251 | * If the image supports layering, get the parent info. We | |
4252 | * need to probe the first time regardless. Thereafter we | |
4253 | * only need to if there's a parent, to see if it has | |
4254 | * disappeared due to the mapped image getting flattened. | |
4255 | */ | |
4256 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING && | |
4257 | (first_time || rbd_dev->parent_spec)) { | |
4258 | bool warn; | |
4259 | ||
4260 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4261 | if (ret) | |
4262 | goto out; | |
4263 | ||
4264 | /* | |
4265 | * Print a warning if this is the initial probe and | |
4266 | * the image has a parent. Don't print it if the | |
4267 | * image now being probed is itself a parent. We | |
4268 | * can tell at this point because we won't know its | |
4269 | * pool name yet (just its pool id). | |
4270 | */ | |
4271 | warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name; | |
4272 | if (first_time && warn) | |
4273 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4274 | "is EXPERIMENTAL!"); | |
4275 | } | |
4276 | ||
117973fb AE |
4277 | ret = rbd_dev_v2_image_size(rbd_dev); |
4278 | if (ret) | |
4279 | goto out; | |
642a2537 | 4280 | |
29334ba4 AE |
4281 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) |
4282 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
4283 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
117973fb | 4284 | |
cc4a38bd | 4285 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4286 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4287 | out: |
4288 | up_write(&rbd_dev->header_rwsem); | |
4289 | ||
4290 | return ret; | |
4291 | } | |
4292 | ||
dfc5606d YS |
4293 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4294 | { | |
dfc5606d | 4295 | struct device *dev; |
cd789ab9 | 4296 | int ret; |
dfc5606d YS |
4297 | |
4298 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 4299 | |
cd789ab9 | 4300 | dev = &rbd_dev->dev; |
dfc5606d YS |
4301 | dev->bus = &rbd_bus_type; |
4302 | dev->type = &rbd_device_type; | |
4303 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4304 | dev->release = rbd_dev_device_release; |
de71a297 | 4305 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4306 | ret = device_register(dev); |
dfc5606d | 4307 | |
dfc5606d | 4308 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 4309 | |
dfc5606d | 4310 | return ret; |
602adf40 YS |
4311 | } |
4312 | ||
dfc5606d YS |
4313 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4314 | { | |
4315 | device_unregister(&rbd_dev->dev); | |
4316 | } | |
4317 | ||
e2839308 | 4318 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
4319 | |
4320 | /* | |
499afd5b AE |
4321 | * Get a unique rbd identifier for the given new rbd_dev, and add |
4322 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 4323 | */ |
e2839308 | 4324 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4325 | { |
e2839308 | 4326 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
4327 | |
4328 | spin_lock(&rbd_dev_list_lock); | |
4329 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4330 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
4331 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
4332 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 4333 | } |
b7f23c36 | 4334 | |
1ddbe94e | 4335 | /* |
499afd5b AE |
4336 | * Remove an rbd_dev from the global list, and record that its |
4337 | * identifier is no longer in use. | |
1ddbe94e | 4338 | */ |
e2839308 | 4339 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4340 | { |
d184f6bf | 4341 | struct list_head *tmp; |
de71a297 | 4342 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
4343 | int max_id; |
4344 | ||
aafb230e | 4345 | rbd_assert(rbd_id > 0); |
499afd5b | 4346 | |
e2839308 AE |
4347 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
4348 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
4349 | spin_lock(&rbd_dev_list_lock); |
4350 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
4351 | |
4352 | /* | |
4353 | * If the id being "put" is not the current maximum, there | |
4354 | * is nothing special we need to do. | |
4355 | */ | |
e2839308 | 4356 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
4357 | spin_unlock(&rbd_dev_list_lock); |
4358 | return; | |
4359 | } | |
4360 | ||
4361 | /* | |
4362 | * We need to update the current maximum id. Search the | |
4363 | * list to find out what it is. We're more likely to find | |
4364 | * the maximum at the end, so search the list backward. | |
4365 | */ | |
4366 | max_id = 0; | |
4367 | list_for_each_prev(tmp, &rbd_dev_list) { | |
4368 | struct rbd_device *rbd_dev; | |
4369 | ||
4370 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
4371 | if (rbd_dev->dev_id > max_id) |
4372 | max_id = rbd_dev->dev_id; | |
d184f6bf | 4373 | } |
499afd5b | 4374 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 4375 | |
1ddbe94e | 4376 | /* |
e2839308 | 4377 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
4378 | * which case it now accurately reflects the new maximum. |
4379 | * Be careful not to overwrite the maximum value in that | |
4380 | * case. | |
1ddbe94e | 4381 | */ |
e2839308 AE |
4382 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
4383 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
4384 | } |
4385 | ||
e28fff26 AE |
4386 | /* |
4387 | * Skips over white space at *buf, and updates *buf to point to the | |
4388 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4389 | * the token (string of non-white space characters) found. Note |
4390 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4391 | */ |
4392 | static inline size_t next_token(const char **buf) | |
4393 | { | |
4394 | /* | |
4395 | * These are the characters that produce nonzero for | |
4396 | * isspace() in the "C" and "POSIX" locales. | |
4397 | */ | |
4398 | const char *spaces = " \f\n\r\t\v"; | |
4399 | ||
4400 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4401 | ||
4402 | return strcspn(*buf, spaces); /* Return token length */ | |
4403 | } | |
4404 | ||
4405 | /* | |
4406 | * Finds the next token in *buf, and if the provided token buffer is | |
4407 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4408 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4409 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4410 | * |
4411 | * Returns the length of the token found (not including the '\0'). | |
4412 | * Return value will be 0 if no token is found, and it will be >= | |
4413 | * token_size if the token would not fit. | |
4414 | * | |
593a9e7b | 4415 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4416 | * found token. Note that this occurs even if the token buffer is |
4417 | * too small to hold it. | |
4418 | */ | |
4419 | static inline size_t copy_token(const char **buf, | |
4420 | char *token, | |
4421 | size_t token_size) | |
4422 | { | |
4423 | size_t len; | |
4424 | ||
4425 | len = next_token(buf); | |
4426 | if (len < token_size) { | |
4427 | memcpy(token, *buf, len); | |
4428 | *(token + len) = '\0'; | |
4429 | } | |
4430 | *buf += len; | |
4431 | ||
4432 | return len; | |
4433 | } | |
4434 | ||
ea3352f4 AE |
4435 | /* |
4436 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4437 | * enough to hold a copy of it, and copies the token into the new | |
4438 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4439 | * that a duplicate buffer is created even for a zero-length token. | |
4440 | * | |
4441 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4442 | * pointer if memory for the duplicate was not available. If | |
4443 | * the lenp argument is a non-null pointer, the length of the token | |
4444 | * (not including the '\0') is returned in *lenp. | |
4445 | * | |
4446 | * If successful, the *buf pointer will be updated to point beyond | |
4447 | * the end of the found token. | |
4448 | * | |
4449 | * Note: uses GFP_KERNEL for allocation. | |
4450 | */ | |
4451 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4452 | { | |
4453 | char *dup; | |
4454 | size_t len; | |
4455 | ||
4456 | len = next_token(buf); | |
4caf35f9 | 4457 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4458 | if (!dup) |
4459 | return NULL; | |
ea3352f4 AE |
4460 | *(dup + len) = '\0'; |
4461 | *buf += len; | |
4462 | ||
4463 | if (lenp) | |
4464 | *lenp = len; | |
4465 | ||
4466 | return dup; | |
4467 | } | |
4468 | ||
a725f65e | 4469 | /* |
859c31df AE |
4470 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4471 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4472 | * and the data written is passed here via a NUL-terminated buffer. | |
4473 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4474 | * |
859c31df AE |
4475 | * The information extracted from these options is recorded in |
4476 | * the other parameters which return dynamically-allocated | |
4477 | * structures: | |
4478 | * ceph_opts | |
4479 | * The address of a pointer that will refer to a ceph options | |
4480 | * structure. Caller must release the returned pointer using | |
4481 | * ceph_destroy_options() when it is no longer needed. | |
4482 | * rbd_opts | |
4483 | * Address of an rbd options pointer. Fully initialized by | |
4484 | * this function; caller must release with kfree(). | |
4485 | * spec | |
4486 | * Address of an rbd image specification pointer. Fully | |
4487 | * initialized by this function based on parsed options. | |
4488 | * Caller must release with rbd_spec_put(). | |
4489 | * | |
4490 | * The options passed take this form: | |
4491 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4492 | * where: | |
4493 | * <mon_addrs> | |
4494 | * A comma-separated list of one or more monitor addresses. | |
4495 | * A monitor address is an ip address, optionally followed | |
4496 | * by a port number (separated by a colon). | |
4497 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4498 | * <options> | |
4499 | * A comma-separated list of ceph and/or rbd options. | |
4500 | * <pool_name> | |
4501 | * The name of the rados pool containing the rbd image. | |
4502 | * <image_name> | |
4503 | * The name of the image in that pool to map. | |
4504 | * <snap_id> | |
4505 | * An optional snapshot id. If provided, the mapping will | |
4506 | * present data from the image at the time that snapshot was | |
4507 | * created. The image head is used if no snapshot id is | |
4508 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4509 | */ |
859c31df | 4510 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4511 | struct ceph_options **ceph_opts, |
859c31df AE |
4512 | struct rbd_options **opts, |
4513 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4514 | { |
d22f76e7 | 4515 | size_t len; |
859c31df | 4516 | char *options; |
0ddebc0c | 4517 | const char *mon_addrs; |
ecb4dc22 | 4518 | char *snap_name; |
0ddebc0c | 4519 | size_t mon_addrs_size; |
859c31df | 4520 | struct rbd_spec *spec = NULL; |
4e9afeba | 4521 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4522 | struct ceph_options *copts; |
dc79b113 | 4523 | int ret; |
e28fff26 AE |
4524 | |
4525 | /* The first four tokens are required */ | |
4526 | ||
7ef3214a | 4527 | len = next_token(&buf); |
4fb5d671 AE |
4528 | if (!len) { |
4529 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4530 | return -EINVAL; | |
4531 | } | |
0ddebc0c | 4532 | mon_addrs = buf; |
f28e565a | 4533 | mon_addrs_size = len + 1; |
7ef3214a | 4534 | buf += len; |
a725f65e | 4535 | |
dc79b113 | 4536 | ret = -EINVAL; |
f28e565a AE |
4537 | options = dup_token(&buf, NULL); |
4538 | if (!options) | |
dc79b113 | 4539 | return -ENOMEM; |
4fb5d671 AE |
4540 | if (!*options) { |
4541 | rbd_warn(NULL, "no options provided"); | |
4542 | goto out_err; | |
4543 | } | |
e28fff26 | 4544 | |
859c31df AE |
4545 | spec = rbd_spec_alloc(); |
4546 | if (!spec) | |
f28e565a | 4547 | goto out_mem; |
859c31df AE |
4548 | |
4549 | spec->pool_name = dup_token(&buf, NULL); | |
4550 | if (!spec->pool_name) | |
4551 | goto out_mem; | |
4fb5d671 AE |
4552 | if (!*spec->pool_name) { |
4553 | rbd_warn(NULL, "no pool name provided"); | |
4554 | goto out_err; | |
4555 | } | |
e28fff26 | 4556 | |
69e7a02f | 4557 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4558 | if (!spec->image_name) |
f28e565a | 4559 | goto out_mem; |
4fb5d671 AE |
4560 | if (!*spec->image_name) { |
4561 | rbd_warn(NULL, "no image name provided"); | |
4562 | goto out_err; | |
4563 | } | |
d4b125e9 | 4564 | |
f28e565a AE |
4565 | /* |
4566 | * Snapshot name is optional; default is to use "-" | |
4567 | * (indicating the head/no snapshot). | |
4568 | */ | |
3feeb894 | 4569 | len = next_token(&buf); |
820a5f3e | 4570 | if (!len) { |
3feeb894 AE |
4571 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4572 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4573 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4574 | ret = -ENAMETOOLONG; |
f28e565a | 4575 | goto out_err; |
849b4260 | 4576 | } |
ecb4dc22 AE |
4577 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4578 | if (!snap_name) | |
f28e565a | 4579 | goto out_mem; |
ecb4dc22 AE |
4580 | *(snap_name + len) = '\0'; |
4581 | spec->snap_name = snap_name; | |
e5c35534 | 4582 | |
0ddebc0c | 4583 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4584 | |
4e9afeba AE |
4585 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4586 | if (!rbd_opts) | |
4587 | goto out_mem; | |
4588 | ||
4589 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4590 | |
859c31df | 4591 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4592 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4593 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4594 | if (IS_ERR(copts)) { |
4595 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4596 | goto out_err; |
4597 | } | |
859c31df AE |
4598 | kfree(options); |
4599 | ||
4600 | *ceph_opts = copts; | |
4e9afeba | 4601 | *opts = rbd_opts; |
859c31df | 4602 | *rbd_spec = spec; |
0ddebc0c | 4603 | |
dc79b113 | 4604 | return 0; |
f28e565a | 4605 | out_mem: |
dc79b113 | 4606 | ret = -ENOMEM; |
d22f76e7 | 4607 | out_err: |
859c31df AE |
4608 | kfree(rbd_opts); |
4609 | rbd_spec_put(spec); | |
f28e565a | 4610 | kfree(options); |
d22f76e7 | 4611 | |
dc79b113 | 4612 | return ret; |
a725f65e AE |
4613 | } |
4614 | ||
589d30e0 AE |
4615 | /* |
4616 | * An rbd format 2 image has a unique identifier, distinct from the | |
4617 | * name given to it by the user. Internally, that identifier is | |
4618 | * what's used to specify the names of objects related to the image. | |
4619 | * | |
4620 | * A special "rbd id" object is used to map an rbd image name to its | |
4621 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4622 | * with the supplied name. | |
4623 | * | |
4624 | * This function will record the given rbd_dev's image_id field if | |
4625 | * it can be determined, and in that case will return 0. If any | |
4626 | * errors occur a negative errno will be returned and the rbd_dev's | |
4627 | * image_id field will be unchanged (and should be NULL). | |
4628 | */ | |
4629 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4630 | { | |
4631 | int ret; | |
4632 | size_t size; | |
4633 | char *object_name; | |
4634 | void *response; | |
c0fba368 | 4635 | char *image_id; |
2f82ee54 | 4636 | |
2c0d0a10 AE |
4637 | /* |
4638 | * When probing a parent image, the image id is already | |
4639 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4640 | * need to fetch the image id again in this case. We |
4641 | * do still need to set the image format though. | |
2c0d0a10 | 4642 | */ |
c0fba368 AE |
4643 | if (rbd_dev->spec->image_id) { |
4644 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4645 | ||
2c0d0a10 | 4646 | return 0; |
c0fba368 | 4647 | } |
2c0d0a10 | 4648 | |
589d30e0 AE |
4649 | /* |
4650 | * First, see if the format 2 image id file exists, and if | |
4651 | * so, get the image's persistent id from it. | |
4652 | */ | |
69e7a02f | 4653 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4654 | object_name = kmalloc(size, GFP_NOIO); |
4655 | if (!object_name) | |
4656 | return -ENOMEM; | |
0d7dbfce | 4657 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4658 | dout("rbd id object name is %s\n", object_name); |
4659 | ||
4660 | /* Response will be an encoded string, which includes a length */ | |
4661 | ||
4662 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4663 | response = kzalloc(size, GFP_NOIO); | |
4664 | if (!response) { | |
4665 | ret = -ENOMEM; | |
4666 | goto out; | |
4667 | } | |
4668 | ||
c0fba368 AE |
4669 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4670 | ||
36be9a76 | 4671 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4672 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4673 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4674 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4675 | if (ret == -ENOENT) { |
4676 | image_id = kstrdup("", GFP_KERNEL); | |
4677 | ret = image_id ? 0 : -ENOMEM; | |
4678 | if (!ret) | |
4679 | rbd_dev->image_format = 1; | |
4680 | } else if (ret > sizeof (__le32)) { | |
4681 | void *p = response; | |
4682 | ||
4683 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4684 | NULL, GFP_NOIO); |
c0fba368 AE |
4685 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4686 | if (!ret) | |
4687 | rbd_dev->image_format = 2; | |
589d30e0 | 4688 | } else { |
c0fba368 AE |
4689 | ret = -EINVAL; |
4690 | } | |
4691 | ||
4692 | if (!ret) { | |
4693 | rbd_dev->spec->image_id = image_id; | |
4694 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4695 | } |
4696 | out: | |
4697 | kfree(response); | |
4698 | kfree(object_name); | |
4699 | ||
4700 | return ret; | |
4701 | } | |
4702 | ||
6fd48b3b AE |
4703 | /* Undo whatever state changes are made by v1 or v2 image probe */ |
4704 | ||
4705 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) | |
4706 | { | |
4707 | struct rbd_image_header *header; | |
4708 | ||
392a9dad AE |
4709 | /* Drop parent reference unless it's already been done (or none) */ |
4710 | ||
4711 | if (rbd_dev->parent_overlap) | |
4712 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
4713 | |
4714 | /* Free dynamic fields from the header, then zero it out */ | |
4715 | ||
4716 | header = &rbd_dev->header; | |
812164f8 | 4717 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4718 | kfree(header->snap_sizes); |
4719 | kfree(header->snap_names); | |
4720 | kfree(header->object_prefix); | |
4721 | memset(header, 0, sizeof (*header)); | |
4722 | } | |
4723 | ||
2df3fac7 | 4724 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 | 4725 | { |
9d475de5 | 4726 | int ret; |
a30b71b9 | 4727 | |
1e130199 | 4728 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 4729 | if (ret) |
b1b5402a AE |
4730 | goto out_err; |
4731 | ||
2df3fac7 AE |
4732 | /* |
4733 | * Get the and check features for the image. Currently the | |
4734 | * features are assumed to never change. | |
4735 | */ | |
b1b5402a | 4736 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 4737 | if (ret) |
9d475de5 | 4738 | goto out_err; |
35d489f9 | 4739 | |
cc070d59 AE |
4740 | /* If the image supports fancy striping, get its parameters */ |
4741 | ||
4742 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4743 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4744 | if (ret < 0) | |
4745 | goto out_err; | |
4746 | } | |
2df3fac7 | 4747 | /* No support for crypto and compression type format 2 images */ |
6e14b1a6 | 4748 | |
35152979 | 4749 | return 0; |
9d475de5 | 4750 | out_err: |
642a2537 | 4751 | rbd_dev->header.features = 0; |
1e130199 AE |
4752 | kfree(rbd_dev->header.object_prefix); |
4753 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4754 | |
4755 | return ret; | |
a30b71b9 AE |
4756 | } |
4757 | ||
124afba2 | 4758 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4759 | { |
2f82ee54 | 4760 | struct rbd_device *parent = NULL; |
124afba2 AE |
4761 | struct rbd_spec *parent_spec; |
4762 | struct rbd_client *rbdc; | |
4763 | int ret; | |
4764 | ||
4765 | if (!rbd_dev->parent_spec) | |
4766 | return 0; | |
4767 | /* | |
4768 | * We need to pass a reference to the client and the parent | |
4769 | * spec when creating the parent rbd_dev. Images related by | |
4770 | * parent/child relationships always share both. | |
4771 | */ | |
4772 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4773 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4774 | ||
4775 | ret = -ENOMEM; | |
4776 | parent = rbd_dev_create(rbdc, parent_spec); | |
4777 | if (!parent) | |
4778 | goto out_err; | |
4779 | ||
1f3ef788 | 4780 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
4781 | if (ret < 0) |
4782 | goto out_err; | |
4783 | rbd_dev->parent = parent; | |
a2acd00e | 4784 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
4785 | |
4786 | return 0; | |
4787 | out_err: | |
4788 | if (parent) { | |
fb65d228 | 4789 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
4790 | kfree(rbd_dev->header_name); |
4791 | rbd_dev_destroy(parent); | |
4792 | } else { | |
4793 | rbd_put_client(rbdc); | |
4794 | rbd_spec_put(parent_spec); | |
4795 | } | |
4796 | ||
4797 | return ret; | |
4798 | } | |
4799 | ||
200a6a8b | 4800 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4801 | { |
83a06263 | 4802 | int ret; |
d1cf5788 | 4803 | |
83a06263 AE |
4804 | /* generate unique id: find highest unique id, add one */ |
4805 | rbd_dev_id_get(rbd_dev); | |
4806 | ||
4807 | /* Fill in the device name, now that we have its id. */ | |
4808 | BUILD_BUG_ON(DEV_NAME_LEN | |
4809 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4810 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4811 | ||
4812 | /* Get our block major device number. */ | |
4813 | ||
4814 | ret = register_blkdev(0, rbd_dev->name); | |
4815 | if (ret < 0) | |
4816 | goto err_out_id; | |
4817 | rbd_dev->major = ret; | |
4818 | ||
4819 | /* Set up the blkdev mapping. */ | |
4820 | ||
4821 | ret = rbd_init_disk(rbd_dev); | |
4822 | if (ret) | |
4823 | goto err_out_blkdev; | |
4824 | ||
f35a4dee | 4825 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
4826 | if (ret) |
4827 | goto err_out_disk; | |
f35a4dee AE |
4828 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
4829 | ||
4830 | ret = rbd_bus_add_dev(rbd_dev); | |
4831 | if (ret) | |
4832 | goto err_out_mapping; | |
83a06263 | 4833 | |
83a06263 AE |
4834 | /* Everything's ready. Announce the disk to the world. */ |
4835 | ||
129b79d4 | 4836 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4837 | add_disk(rbd_dev->disk); |
4838 | ||
4839 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4840 | (unsigned long long) rbd_dev->mapping.size); | |
4841 | ||
4842 | return ret; | |
2f82ee54 | 4843 | |
f35a4dee AE |
4844 | err_out_mapping: |
4845 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
4846 | err_out_disk: |
4847 | rbd_free_disk(rbd_dev); | |
4848 | err_out_blkdev: | |
4849 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4850 | err_out_id: | |
4851 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4852 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4853 | |
4854 | return ret; | |
4855 | } | |
4856 | ||
332bb12d AE |
4857 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4858 | { | |
4859 | struct rbd_spec *spec = rbd_dev->spec; | |
4860 | size_t size; | |
4861 | ||
4862 | /* Record the header object name for this rbd image. */ | |
4863 | ||
4864 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4865 | ||
4866 | if (rbd_dev->image_format == 1) | |
4867 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4868 | else | |
4869 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4870 | ||
4871 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4872 | if (!rbd_dev->header_name) | |
4873 | return -ENOMEM; | |
4874 | ||
4875 | if (rbd_dev->image_format == 1) | |
4876 | sprintf(rbd_dev->header_name, "%s%s", | |
4877 | spec->image_name, RBD_SUFFIX); | |
4878 | else | |
4879 | sprintf(rbd_dev->header_name, "%s%s", | |
4880 | RBD_HEADER_PREFIX, spec->image_id); | |
4881 | return 0; | |
4882 | } | |
4883 | ||
200a6a8b AE |
4884 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4885 | { | |
6fd48b3b | 4886 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 4887 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4888 | rbd_dev->header_name = NULL; |
4889 | rbd_dev->image_format = 0; | |
4890 | kfree(rbd_dev->spec->image_id); | |
4891 | rbd_dev->spec->image_id = NULL; | |
4892 | ||
200a6a8b AE |
4893 | rbd_dev_destroy(rbd_dev); |
4894 | } | |
4895 | ||
a30b71b9 AE |
4896 | /* |
4897 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
4898 | * device. If this image is the one being mapped (i.e., not a |
4899 | * parent), initiate a watch on its header object before using that | |
4900 | * object to get detailed information about the rbd image. | |
a30b71b9 | 4901 | */ |
1f3ef788 | 4902 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
4903 | { |
4904 | int ret; | |
b644de2b | 4905 | int tmp; |
a30b71b9 AE |
4906 | |
4907 | /* | |
4908 | * Get the id from the image id object. If it's not a | |
4909 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4910 | * it's a format 1 image. | |
4911 | */ | |
4912 | ret = rbd_dev_image_id(rbd_dev); | |
4913 | if (ret) | |
c0fba368 AE |
4914 | return ret; |
4915 | rbd_assert(rbd_dev->spec->image_id); | |
4916 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4917 | ||
332bb12d AE |
4918 | ret = rbd_dev_header_name(rbd_dev); |
4919 | if (ret) | |
4920 | goto err_out_format; | |
4921 | ||
1f3ef788 AE |
4922 | if (mapping) { |
4923 | ret = rbd_dev_header_watch_sync(rbd_dev, true); | |
4924 | if (ret) | |
4925 | goto out_header_name; | |
4926 | } | |
b644de2b | 4927 | |
c0fba368 | 4928 | if (rbd_dev->image_format == 1) |
99a41ebc | 4929 | ret = rbd_dev_v1_header_info(rbd_dev); |
a30b71b9 | 4930 | else |
2df3fac7 | 4931 | ret = rbd_dev_v2_header_info(rbd_dev); |
5655c4d9 | 4932 | if (ret) |
b644de2b | 4933 | goto err_out_watch; |
83a06263 | 4934 | |
9bb81c9b AE |
4935 | ret = rbd_dev_spec_update(rbd_dev); |
4936 | if (ret) | |
33dca39f | 4937 | goto err_out_probe; |
9bb81c9b AE |
4938 | |
4939 | ret = rbd_dev_probe_parent(rbd_dev); | |
30d60ba2 AE |
4940 | if (ret) |
4941 | goto err_out_probe; | |
4942 | ||
4943 | dout("discovered format %u image, header name is %s\n", | |
4944 | rbd_dev->image_format, rbd_dev->header_name); | |
83a06263 | 4945 | |
30d60ba2 | 4946 | return 0; |
6fd48b3b AE |
4947 | err_out_probe: |
4948 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 4949 | err_out_watch: |
1f3ef788 AE |
4950 | if (mapping) { |
4951 | tmp = rbd_dev_header_watch_sync(rbd_dev, false); | |
4952 | if (tmp) | |
4953 | rbd_warn(rbd_dev, "unable to tear down " | |
4954 | "watch request (%d)\n", tmp); | |
4955 | } | |
332bb12d AE |
4956 | out_header_name: |
4957 | kfree(rbd_dev->header_name); | |
4958 | rbd_dev->header_name = NULL; | |
4959 | err_out_format: | |
4960 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
4961 | kfree(rbd_dev->spec->image_id); |
4962 | rbd_dev->spec->image_id = NULL; | |
4963 | ||
4964 | dout("probe failed, returning %d\n", ret); | |
4965 | ||
a30b71b9 AE |
4966 | return ret; |
4967 | } | |
4968 | ||
59c2be1e YS |
4969 | static ssize_t rbd_add(struct bus_type *bus, |
4970 | const char *buf, | |
4971 | size_t count) | |
602adf40 | 4972 | { |
cb8627c7 | 4973 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4974 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4975 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4976 | struct rbd_spec *spec = NULL; |
9d3997fd | 4977 | struct rbd_client *rbdc; |
27cc2594 | 4978 | struct ceph_osd_client *osdc; |
51344a38 | 4979 | bool read_only; |
27cc2594 | 4980 | int rc = -ENOMEM; |
602adf40 YS |
4981 | |
4982 | if (!try_module_get(THIS_MODULE)) | |
4983 | return -ENODEV; | |
4984 | ||
602adf40 | 4985 | /* parse add command */ |
859c31df | 4986 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4987 | if (rc < 0) |
bd4ba655 | 4988 | goto err_out_module; |
51344a38 AE |
4989 | read_only = rbd_opts->read_only; |
4990 | kfree(rbd_opts); | |
4991 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 4992 | |
9d3997fd AE |
4993 | rbdc = rbd_get_client(ceph_opts); |
4994 | if (IS_ERR(rbdc)) { | |
4995 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4996 | goto err_out_args; |
9d3997fd | 4997 | } |
602adf40 | 4998 | |
602adf40 | 4999 | /* pick the pool */ |
9d3997fd | 5000 | osdc = &rbdc->client->osdc; |
859c31df | 5001 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
5002 | if (rc < 0) |
5003 | goto err_out_client; | |
c0cd10db | 5004 | spec->pool_id = (u64)rc; |
859c31df | 5005 | |
0903e875 AE |
5006 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5007 | ||
c0cd10db AE |
5008 | if (spec->pool_id > (u64)U32_MAX) { |
5009 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
5010 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
5011 | rc = -EIO; |
5012 | goto err_out_client; | |
5013 | } | |
5014 | ||
c53d5893 | 5015 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5016 | if (!rbd_dev) |
5017 | goto err_out_client; | |
c53d5893 AE |
5018 | rbdc = NULL; /* rbd_dev now owns this */ |
5019 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5020 | |
1f3ef788 | 5021 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5022 | if (rc < 0) |
c53d5893 | 5023 | goto err_out_rbd_dev; |
05fd6f6f | 5024 | |
7ce4eef7 AE |
5025 | /* If we are mapping a snapshot it must be marked read-only */ |
5026 | ||
5027 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5028 | read_only = true; | |
5029 | rbd_dev->mapping.read_only = read_only; | |
5030 | ||
b536f69a AE |
5031 | rc = rbd_dev_device_setup(rbd_dev); |
5032 | if (!rc) | |
5033 | return count; | |
5034 | ||
5035 | rbd_dev_image_release(rbd_dev); | |
c53d5893 AE |
5036 | err_out_rbd_dev: |
5037 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5038 | err_out_client: |
9d3997fd | 5039 | rbd_put_client(rbdc); |
0ddebc0c | 5040 | err_out_args: |
4e9afeba | 5041 | kfree(rbd_opts); |
859c31df | 5042 | rbd_spec_put(spec); |
bd4ba655 AE |
5043 | err_out_module: |
5044 | module_put(THIS_MODULE); | |
27cc2594 | 5045 | |
602adf40 | 5046 | dout("Error adding device %s\n", buf); |
27cc2594 | 5047 | |
c0cd10db | 5048 | return (ssize_t)rc; |
602adf40 YS |
5049 | } |
5050 | ||
de71a297 | 5051 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
5052 | { |
5053 | struct list_head *tmp; | |
5054 | struct rbd_device *rbd_dev; | |
5055 | ||
e124a82f | 5056 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
5057 | list_for_each(tmp, &rbd_dev_list) { |
5058 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 5059 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 5060 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 5061 | return rbd_dev; |
e124a82f | 5062 | } |
602adf40 | 5063 | } |
e124a82f | 5064 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
5065 | return NULL; |
5066 | } | |
5067 | ||
200a6a8b | 5068 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5069 | { |
593a9e7b | 5070 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5071 | |
602adf40 | 5072 | rbd_free_disk(rbd_dev); |
200a6a8b | 5073 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5074 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 | 5075 | unregister_blkdev(rbd_dev->major, rbd_dev->name); |
200a6a8b | 5076 | rbd_dev->major = 0; |
e2839308 | 5077 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5078 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5079 | } |
5080 | ||
05a46afd AE |
5081 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5082 | { | |
ad945fc1 | 5083 | while (rbd_dev->parent) { |
05a46afd AE |
5084 | struct rbd_device *first = rbd_dev; |
5085 | struct rbd_device *second = first->parent; | |
5086 | struct rbd_device *third; | |
5087 | ||
5088 | /* | |
5089 | * Follow to the parent with no grandparent and | |
5090 | * remove it. | |
5091 | */ | |
5092 | while (second && (third = second->parent)) { | |
5093 | first = second; | |
5094 | second = third; | |
5095 | } | |
ad945fc1 | 5096 | rbd_assert(second); |
8ad42cd0 | 5097 | rbd_dev_image_release(second); |
ad945fc1 AE |
5098 | first->parent = NULL; |
5099 | first->parent_overlap = 0; | |
5100 | ||
5101 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5102 | rbd_spec_put(first->parent_spec); |
5103 | first->parent_spec = NULL; | |
05a46afd AE |
5104 | } |
5105 | } | |
5106 | ||
dfc5606d YS |
5107 | static ssize_t rbd_remove(struct bus_type *bus, |
5108 | const char *buf, | |
5109 | size_t count) | |
602adf40 YS |
5110 | { |
5111 | struct rbd_device *rbd_dev = NULL; | |
0d8189e1 | 5112 | int target_id; |
602adf40 | 5113 | unsigned long ul; |
0d8189e1 | 5114 | int ret; |
602adf40 | 5115 | |
0d8189e1 AE |
5116 | ret = strict_strtoul(buf, 10, &ul); |
5117 | if (ret) | |
5118 | return ret; | |
602adf40 YS |
5119 | |
5120 | /* convert to int; abort if we lost anything in the conversion */ | |
5121 | target_id = (int) ul; | |
5122 | if (target_id != ul) | |
5123 | return -EINVAL; | |
5124 | ||
5125 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
5126 | ||
5127 | rbd_dev = __rbd_get_dev(target_id); | |
5128 | if (!rbd_dev) { | |
5129 | ret = -ENOENT; | |
5130 | goto done; | |
42382b70 AE |
5131 | } |
5132 | ||
a14ea269 | 5133 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 5134 | if (rbd_dev->open_count) |
42382b70 | 5135 | ret = -EBUSY; |
b82d167b AE |
5136 | else |
5137 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 5138 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 5139 | if (ret < 0) |
42382b70 | 5140 | goto done; |
b480815a | 5141 | rbd_bus_del_dev(rbd_dev); |
1f3ef788 AE |
5142 | ret = rbd_dev_header_watch_sync(rbd_dev, false); |
5143 | if (ret) | |
5144 | rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret); | |
8ad42cd0 | 5145 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5146 | module_put(THIS_MODULE); |
1f3ef788 | 5147 | ret = count; |
602adf40 YS |
5148 | done: |
5149 | mutex_unlock(&ctl_mutex); | |
aafb230e | 5150 | |
602adf40 YS |
5151 | return ret; |
5152 | } | |
5153 | ||
602adf40 YS |
5154 | /* |
5155 | * create control files in sysfs | |
dfc5606d | 5156 | * /sys/bus/rbd/... |
602adf40 YS |
5157 | */ |
5158 | static int rbd_sysfs_init(void) | |
5159 | { | |
dfc5606d | 5160 | int ret; |
602adf40 | 5161 | |
fed4c143 | 5162 | ret = device_register(&rbd_root_dev); |
21079786 | 5163 | if (ret < 0) |
dfc5606d | 5164 | return ret; |
602adf40 | 5165 | |
fed4c143 AE |
5166 | ret = bus_register(&rbd_bus_type); |
5167 | if (ret < 0) | |
5168 | device_unregister(&rbd_root_dev); | |
602adf40 | 5169 | |
602adf40 YS |
5170 | return ret; |
5171 | } | |
5172 | ||
5173 | static void rbd_sysfs_cleanup(void) | |
5174 | { | |
dfc5606d | 5175 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5176 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5177 | } |
5178 | ||
1c2a9dfe AE |
5179 | static int rbd_slab_init(void) |
5180 | { | |
5181 | rbd_assert(!rbd_img_request_cache); | |
5182 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5183 | sizeof (struct rbd_img_request), | |
5184 | __alignof__(struct rbd_img_request), | |
5185 | 0, NULL); | |
868311b1 AE |
5186 | if (!rbd_img_request_cache) |
5187 | return -ENOMEM; | |
5188 | ||
5189 | rbd_assert(!rbd_obj_request_cache); | |
5190 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5191 | sizeof (struct rbd_obj_request), | |
5192 | __alignof__(struct rbd_obj_request), | |
5193 | 0, NULL); | |
78c2a44a AE |
5194 | if (!rbd_obj_request_cache) |
5195 | goto out_err; | |
5196 | ||
5197 | rbd_assert(!rbd_segment_name_cache); | |
5198 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
5199 | MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL); | |
5200 | if (rbd_segment_name_cache) | |
1c2a9dfe | 5201 | return 0; |
78c2a44a AE |
5202 | out_err: |
5203 | if (rbd_obj_request_cache) { | |
5204 | kmem_cache_destroy(rbd_obj_request_cache); | |
5205 | rbd_obj_request_cache = NULL; | |
5206 | } | |
1c2a9dfe | 5207 | |
868311b1 AE |
5208 | kmem_cache_destroy(rbd_img_request_cache); |
5209 | rbd_img_request_cache = NULL; | |
5210 | ||
1c2a9dfe AE |
5211 | return -ENOMEM; |
5212 | } | |
5213 | ||
5214 | static void rbd_slab_exit(void) | |
5215 | { | |
78c2a44a AE |
5216 | rbd_assert(rbd_segment_name_cache); |
5217 | kmem_cache_destroy(rbd_segment_name_cache); | |
5218 | rbd_segment_name_cache = NULL; | |
5219 | ||
868311b1 AE |
5220 | rbd_assert(rbd_obj_request_cache); |
5221 | kmem_cache_destroy(rbd_obj_request_cache); | |
5222 | rbd_obj_request_cache = NULL; | |
5223 | ||
1c2a9dfe AE |
5224 | rbd_assert(rbd_img_request_cache); |
5225 | kmem_cache_destroy(rbd_img_request_cache); | |
5226 | rbd_img_request_cache = NULL; | |
5227 | } | |
5228 | ||
cc344fa1 | 5229 | static int __init rbd_init(void) |
602adf40 YS |
5230 | { |
5231 | int rc; | |
5232 | ||
1e32d34c AE |
5233 | if (!libceph_compatible(NULL)) { |
5234 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
5235 | ||
5236 | return -EINVAL; | |
5237 | } | |
1c2a9dfe | 5238 | rc = rbd_slab_init(); |
602adf40 YS |
5239 | if (rc) |
5240 | return rc; | |
1c2a9dfe AE |
5241 | rc = rbd_sysfs_init(); |
5242 | if (rc) | |
5243 | rbd_slab_exit(); | |
5244 | else | |
5245 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); | |
5246 | ||
5247 | return rc; | |
602adf40 YS |
5248 | } |
5249 | ||
cc344fa1 | 5250 | static void __exit rbd_exit(void) |
602adf40 YS |
5251 | { |
5252 | rbd_sysfs_cleanup(); | |
1c2a9dfe | 5253 | rbd_slab_exit(); |
602adf40 YS |
5254 | } |
5255 | ||
5256 | module_init(rbd_init); | |
5257 | module_exit(rbd_exit); | |
5258 | ||
5259 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
5260 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
5261 | MODULE_DESCRIPTION("rados block device"); | |
5262 | ||
5263 | /* following authorship retained from original osdblk.c */ | |
5264 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5265 | ||
5266 | MODULE_LICENSE("GPL"); |