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