]>
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 | ||
cfbf6377 | 375 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
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 | 399 | |
b15a21dd GKH |
400 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
401 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
402 | ||
403 | static struct attribute *rbd_bus_attrs[] = { | |
404 | &bus_attr_add.attr, | |
405 | &bus_attr_remove.attr, | |
406 | NULL, | |
f0f8cef5 | 407 | }; |
b15a21dd | 408 | ATTRIBUTE_GROUPS(rbd_bus); |
f0f8cef5 AE |
409 | |
410 | static struct bus_type rbd_bus_type = { | |
411 | .name = "rbd", | |
b15a21dd | 412 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
413 | }; |
414 | ||
415 | static void rbd_root_dev_release(struct device *dev) | |
416 | { | |
417 | } | |
418 | ||
419 | static struct device rbd_root_dev = { | |
420 | .init_name = "rbd", | |
421 | .release = rbd_root_dev_release, | |
422 | }; | |
423 | ||
06ecc6cb AE |
424 | static __printf(2, 3) |
425 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
426 | { | |
427 | struct va_format vaf; | |
428 | va_list args; | |
429 | ||
430 | va_start(args, fmt); | |
431 | vaf.fmt = fmt; | |
432 | vaf.va = &args; | |
433 | ||
434 | if (!rbd_dev) | |
435 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
436 | else if (rbd_dev->disk) | |
437 | printk(KERN_WARNING "%s: %s: %pV\n", | |
438 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
439 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
440 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
441 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
442 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
443 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
444 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
445 | else /* punt */ | |
446 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
447 | RBD_DRV_NAME, rbd_dev, &vaf); | |
448 | va_end(args); | |
449 | } | |
450 | ||
aafb230e AE |
451 | #ifdef RBD_DEBUG |
452 | #define rbd_assert(expr) \ | |
453 | if (unlikely(!(expr))) { \ | |
454 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
455 | "at line %d:\n\n" \ | |
456 | "\trbd_assert(%s);\n\n", \ | |
457 | __func__, __LINE__, #expr); \ | |
458 | BUG(); \ | |
459 | } | |
460 | #else /* !RBD_DEBUG */ | |
461 | # define rbd_assert(expr) ((void) 0) | |
462 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 463 | |
b454e36d | 464 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
465 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
466 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 467 | |
cc4a38bd | 468 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 AE |
469 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
470 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev); | |
54cac61f AE |
471 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
472 | u64 snap_id); | |
2ad3d716 AE |
473 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
474 | u8 *order, u64 *snap_size); | |
475 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
476 | u64 *snap_features); | |
477 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 478 | |
602adf40 YS |
479 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
480 | { | |
f0f8cef5 | 481 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 482 | bool removing = false; |
602adf40 | 483 | |
f84344f3 | 484 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
485 | return -EROFS; |
486 | ||
a14ea269 | 487 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
488 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
489 | removing = true; | |
490 | else | |
491 | rbd_dev->open_count++; | |
a14ea269 | 492 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
493 | if (removing) |
494 | return -ENOENT; | |
495 | ||
c3e946ce | 496 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 497 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
340c7a2b | 498 | |
602adf40 YS |
499 | return 0; |
500 | } | |
501 | ||
db2a144b | 502 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
503 | { |
504 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
505 | unsigned long open_count_before; |
506 | ||
a14ea269 | 507 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 508 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 509 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 510 | rbd_assert(open_count_before > 0); |
dfc5606d | 511 | |
c3e946ce | 512 | put_device(&rbd_dev->dev); |
dfc5606d YS |
513 | } |
514 | ||
602adf40 YS |
515 | static const struct block_device_operations rbd_bd_ops = { |
516 | .owner = THIS_MODULE, | |
517 | .open = rbd_open, | |
dfc5606d | 518 | .release = rbd_release, |
602adf40 YS |
519 | }; |
520 | ||
521 | /* | |
7262cfca | 522 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 523 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 524 | */ |
f8c38929 | 525 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
526 | { |
527 | struct rbd_client *rbdc; | |
528 | int ret = -ENOMEM; | |
529 | ||
37206ee5 | 530 | dout("%s:\n", __func__); |
602adf40 YS |
531 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
532 | if (!rbdc) | |
533 | goto out_opt; | |
534 | ||
535 | kref_init(&rbdc->kref); | |
536 | INIT_LIST_HEAD(&rbdc->node); | |
537 | ||
43ae4701 | 538 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 539 | if (IS_ERR(rbdc->client)) |
08f75463 | 540 | goto out_rbdc; |
43ae4701 | 541 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
542 | |
543 | ret = ceph_open_session(rbdc->client); | |
544 | if (ret < 0) | |
08f75463 | 545 | goto out_client; |
602adf40 | 546 | |
432b8587 | 547 | spin_lock(&rbd_client_list_lock); |
602adf40 | 548 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 549 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 550 | |
37206ee5 | 551 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 552 | |
602adf40 | 553 | return rbdc; |
08f75463 | 554 | out_client: |
602adf40 | 555 | ceph_destroy_client(rbdc->client); |
08f75463 | 556 | out_rbdc: |
602adf40 YS |
557 | kfree(rbdc); |
558 | out_opt: | |
43ae4701 AE |
559 | if (ceph_opts) |
560 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
561 | dout("%s: error %d\n", __func__, ret); |
562 | ||
28f259b7 | 563 | return ERR_PTR(ret); |
602adf40 YS |
564 | } |
565 | ||
2f82ee54 AE |
566 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
567 | { | |
568 | kref_get(&rbdc->kref); | |
569 | ||
570 | return rbdc; | |
571 | } | |
572 | ||
602adf40 | 573 | /* |
1f7ba331 AE |
574 | * Find a ceph client with specific addr and configuration. If |
575 | * found, bump its reference count. | |
602adf40 | 576 | */ |
1f7ba331 | 577 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
578 | { |
579 | struct rbd_client *client_node; | |
1f7ba331 | 580 | bool found = false; |
602adf40 | 581 | |
43ae4701 | 582 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
583 | return NULL; |
584 | ||
1f7ba331 AE |
585 | spin_lock(&rbd_client_list_lock); |
586 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
587 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
588 | __rbd_get_client(client_node); |
589 | ||
1f7ba331 AE |
590 | found = true; |
591 | break; | |
592 | } | |
593 | } | |
594 | spin_unlock(&rbd_client_list_lock); | |
595 | ||
596 | return found ? client_node : NULL; | |
602adf40 YS |
597 | } |
598 | ||
59c2be1e YS |
599 | /* |
600 | * mount options | |
601 | */ | |
602 | enum { | |
59c2be1e YS |
603 | Opt_last_int, |
604 | /* int args above */ | |
605 | Opt_last_string, | |
606 | /* string args above */ | |
cc0538b6 AE |
607 | Opt_read_only, |
608 | Opt_read_write, | |
609 | /* Boolean args above */ | |
610 | Opt_last_bool, | |
59c2be1e YS |
611 | }; |
612 | ||
43ae4701 | 613 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
614 | /* int args above */ |
615 | /* string args above */ | |
be466c1c | 616 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
617 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
618 | {Opt_read_write, "read_write"}, | |
619 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
620 | /* Boolean args above */ | |
59c2be1e YS |
621 | {-1, NULL} |
622 | }; | |
623 | ||
98571b5a AE |
624 | struct rbd_options { |
625 | bool read_only; | |
626 | }; | |
627 | ||
628 | #define RBD_READ_ONLY_DEFAULT false | |
629 | ||
59c2be1e YS |
630 | static int parse_rbd_opts_token(char *c, void *private) |
631 | { | |
43ae4701 | 632 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
633 | substring_t argstr[MAX_OPT_ARGS]; |
634 | int token, intval, ret; | |
635 | ||
43ae4701 | 636 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
637 | if (token < 0) |
638 | return -EINVAL; | |
639 | ||
640 | if (token < Opt_last_int) { | |
641 | ret = match_int(&argstr[0], &intval); | |
642 | if (ret < 0) { | |
643 | pr_err("bad mount option arg (not int) " | |
644 | "at '%s'\n", c); | |
645 | return ret; | |
646 | } | |
647 | dout("got int token %d val %d\n", token, intval); | |
648 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
649 | dout("got string token %d val %s\n", token, | |
650 | argstr[0].from); | |
cc0538b6 AE |
651 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
652 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
653 | } else { |
654 | dout("got token %d\n", token); | |
655 | } | |
656 | ||
657 | switch (token) { | |
cc0538b6 AE |
658 | case Opt_read_only: |
659 | rbd_opts->read_only = true; | |
660 | break; | |
661 | case Opt_read_write: | |
662 | rbd_opts->read_only = false; | |
663 | break; | |
59c2be1e | 664 | default: |
aafb230e AE |
665 | rbd_assert(false); |
666 | break; | |
59c2be1e YS |
667 | } |
668 | return 0; | |
669 | } | |
670 | ||
602adf40 YS |
671 | /* |
672 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
673 | * not exist create it. Either way, ceph_opts is consumed by this |
674 | * function. | |
602adf40 | 675 | */ |
9d3997fd | 676 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 677 | { |
f8c38929 | 678 | struct rbd_client *rbdc; |
59c2be1e | 679 | |
cfbf6377 | 680 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 681 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 682 | if (rbdc) /* using an existing client */ |
43ae4701 | 683 | ceph_destroy_options(ceph_opts); |
9d3997fd | 684 | else |
f8c38929 | 685 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 686 | mutex_unlock(&client_mutex); |
602adf40 | 687 | |
9d3997fd | 688 | return rbdc; |
602adf40 YS |
689 | } |
690 | ||
691 | /* | |
692 | * Destroy ceph client | |
d23a4b3f | 693 | * |
432b8587 | 694 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
695 | */ |
696 | static void rbd_client_release(struct kref *kref) | |
697 | { | |
698 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
699 | ||
37206ee5 | 700 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 701 | spin_lock(&rbd_client_list_lock); |
602adf40 | 702 | list_del(&rbdc->node); |
cd9d9f5d | 703 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
704 | |
705 | ceph_destroy_client(rbdc->client); | |
706 | kfree(rbdc); | |
707 | } | |
708 | ||
709 | /* | |
710 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
711 | * it. | |
712 | */ | |
9d3997fd | 713 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 714 | { |
c53d5893 AE |
715 | if (rbdc) |
716 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
717 | } |
718 | ||
a30b71b9 AE |
719 | static bool rbd_image_format_valid(u32 image_format) |
720 | { | |
721 | return image_format == 1 || image_format == 2; | |
722 | } | |
723 | ||
8e94af8e AE |
724 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
725 | { | |
103a150f AE |
726 | size_t size; |
727 | u32 snap_count; | |
728 | ||
729 | /* The header has to start with the magic rbd header text */ | |
730 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
731 | return false; | |
732 | ||
db2388b6 AE |
733 | /* The bio layer requires at least sector-sized I/O */ |
734 | ||
735 | if (ondisk->options.order < SECTOR_SHIFT) | |
736 | return false; | |
737 | ||
738 | /* If we use u64 in a few spots we may be able to loosen this */ | |
739 | ||
740 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
741 | return false; | |
742 | ||
103a150f AE |
743 | /* |
744 | * The size of a snapshot header has to fit in a size_t, and | |
745 | * that limits the number of snapshots. | |
746 | */ | |
747 | snap_count = le32_to_cpu(ondisk->snap_count); | |
748 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
749 | if (snap_count > size / sizeof (__le64)) | |
750 | return false; | |
751 | ||
752 | /* | |
753 | * Not only that, but the size of the entire the snapshot | |
754 | * header must also be representable in a size_t. | |
755 | */ | |
756 | size -= snap_count * sizeof (__le64); | |
757 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
758 | return false; | |
759 | ||
760 | return true; | |
8e94af8e AE |
761 | } |
762 | ||
602adf40 | 763 | /* |
bb23e37a AE |
764 | * Fill an rbd image header with information from the given format 1 |
765 | * on-disk header. | |
602adf40 | 766 | */ |
662518b1 | 767 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 768 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 769 | { |
662518b1 | 770 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
771 | bool first_time = header->object_prefix == NULL; |
772 | struct ceph_snap_context *snapc; | |
773 | char *object_prefix = NULL; | |
774 | char *snap_names = NULL; | |
775 | u64 *snap_sizes = NULL; | |
ccece235 | 776 | u32 snap_count; |
d2bb24e5 | 777 | size_t size; |
bb23e37a | 778 | int ret = -ENOMEM; |
621901d6 | 779 | u32 i; |
602adf40 | 780 | |
bb23e37a | 781 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 782 | |
bb23e37a AE |
783 | if (first_time) { |
784 | size_t len; | |
103a150f | 785 | |
bb23e37a AE |
786 | len = strnlen(ondisk->object_prefix, |
787 | sizeof (ondisk->object_prefix)); | |
788 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
789 | if (!object_prefix) | |
790 | return -ENOMEM; | |
791 | memcpy(object_prefix, ondisk->object_prefix, len); | |
792 | object_prefix[len] = '\0'; | |
793 | } | |
00f1f36f | 794 | |
bb23e37a | 795 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 796 | |
bb23e37a AE |
797 | snap_count = le32_to_cpu(ondisk->snap_count); |
798 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
799 | if (!snapc) | |
800 | goto out_err; | |
801 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 802 | if (snap_count) { |
bb23e37a | 803 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
804 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
805 | ||
bb23e37a | 806 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 807 | |
bb23e37a AE |
808 | if (snap_names_len > (u64)SIZE_MAX) |
809 | goto out_2big; | |
810 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
811 | if (!snap_names) | |
6a52325f AE |
812 | goto out_err; |
813 | ||
bb23e37a | 814 | /* ...as well as the array of their sizes. */ |
621901d6 | 815 | |
d2bb24e5 | 816 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
817 | snap_sizes = kmalloc(size, GFP_KERNEL); |
818 | if (!snap_sizes) | |
6a52325f | 819 | goto out_err; |
bb23e37a | 820 | |
f785cc1d | 821 | /* |
bb23e37a AE |
822 | * Copy the names, and fill in each snapshot's id |
823 | * and size. | |
824 | * | |
99a41ebc | 825 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 826 | * ondisk buffer we're working with has |
f785cc1d AE |
827 | * snap_names_len bytes beyond the end of the |
828 | * snapshot id array, this memcpy() is safe. | |
829 | */ | |
bb23e37a AE |
830 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
831 | snaps = ondisk->snaps; | |
832 | for (i = 0; i < snap_count; i++) { | |
833 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
834 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
835 | } | |
602adf40 | 836 | } |
6a52325f | 837 | |
bb23e37a | 838 | /* We won't fail any more, fill in the header */ |
621901d6 | 839 | |
bb23e37a AE |
840 | if (first_time) { |
841 | header->object_prefix = object_prefix; | |
842 | header->obj_order = ondisk->options.order; | |
843 | header->crypt_type = ondisk->options.crypt_type; | |
844 | header->comp_type = ondisk->options.comp_type; | |
845 | /* The rest aren't used for format 1 images */ | |
846 | header->stripe_unit = 0; | |
847 | header->stripe_count = 0; | |
848 | header->features = 0; | |
602adf40 | 849 | } else { |
662518b1 AE |
850 | ceph_put_snap_context(header->snapc); |
851 | kfree(header->snap_names); | |
852 | kfree(header->snap_sizes); | |
602adf40 | 853 | } |
849b4260 | 854 | |
bb23e37a | 855 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 856 | |
f84344f3 | 857 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
858 | header->snapc = snapc; |
859 | header->snap_names = snap_names; | |
860 | header->snap_sizes = snap_sizes; | |
468521c1 | 861 | |
662518b1 | 862 | /* Make sure mapping size is consistent with header info */ |
602adf40 | 863 | |
662518b1 AE |
864 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time) |
865 | if (rbd_dev->mapping.size != header->image_size) | |
866 | rbd_dev->mapping.size = header->image_size; | |
867 | ||
602adf40 | 868 | return 0; |
bb23e37a AE |
869 | out_2big: |
870 | ret = -EIO; | |
6a52325f | 871 | out_err: |
bb23e37a AE |
872 | kfree(snap_sizes); |
873 | kfree(snap_names); | |
874 | ceph_put_snap_context(snapc); | |
875 | kfree(object_prefix); | |
ccece235 | 876 | |
bb23e37a | 877 | return ret; |
602adf40 YS |
878 | } |
879 | ||
9682fc6d AE |
880 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
881 | { | |
882 | const char *snap_name; | |
883 | ||
884 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
885 | ||
886 | /* Skip over names until we find the one we are looking for */ | |
887 | ||
888 | snap_name = rbd_dev->header.snap_names; | |
889 | while (which--) | |
890 | snap_name += strlen(snap_name) + 1; | |
891 | ||
892 | return kstrdup(snap_name, GFP_KERNEL); | |
893 | } | |
894 | ||
30d1cff8 AE |
895 | /* |
896 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
897 | * Note that result is for snapshots in *descending* order. | |
898 | */ | |
899 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
900 | { | |
901 | u64 snap_id1 = *(u64 *)s1; | |
902 | u64 snap_id2 = *(u64 *)s2; | |
903 | ||
904 | if (snap_id1 < snap_id2) | |
905 | return 1; | |
906 | return snap_id1 == snap_id2 ? 0 : -1; | |
907 | } | |
908 | ||
909 | /* | |
910 | * Search a snapshot context to see if the given snapshot id is | |
911 | * present. | |
912 | * | |
913 | * Returns the position of the snapshot id in the array if it's found, | |
914 | * or BAD_SNAP_INDEX otherwise. | |
915 | * | |
916 | * Note: The snapshot array is in kept sorted (by the osd) in | |
917 | * reverse order, highest snapshot id first. | |
918 | */ | |
9682fc6d AE |
919 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
920 | { | |
921 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 922 | u64 *found; |
9682fc6d | 923 | |
30d1cff8 AE |
924 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
925 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 926 | |
30d1cff8 | 927 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
928 | } |
929 | ||
2ad3d716 AE |
930 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
931 | u64 snap_id) | |
9e15b77d | 932 | { |
54cac61f | 933 | u32 which; |
da6a6b63 | 934 | const char *snap_name; |
9e15b77d | 935 | |
54cac61f AE |
936 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
937 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 938 | return ERR_PTR(-ENOENT); |
54cac61f | 939 | |
da6a6b63 JD |
940 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
941 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
942 | } |
943 | ||
944 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
945 | { | |
9e15b77d AE |
946 | if (snap_id == CEPH_NOSNAP) |
947 | return RBD_SNAP_HEAD_NAME; | |
948 | ||
54cac61f AE |
949 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
950 | if (rbd_dev->image_format == 1) | |
951 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 952 | |
54cac61f | 953 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
954 | } |
955 | ||
2ad3d716 AE |
956 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
957 | u64 *snap_size) | |
602adf40 | 958 | { |
2ad3d716 AE |
959 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
960 | if (snap_id == CEPH_NOSNAP) { | |
961 | *snap_size = rbd_dev->header.image_size; | |
962 | } else if (rbd_dev->image_format == 1) { | |
963 | u32 which; | |
602adf40 | 964 | |
2ad3d716 AE |
965 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
966 | if (which == BAD_SNAP_INDEX) | |
967 | return -ENOENT; | |
e86924a8 | 968 | |
2ad3d716 AE |
969 | *snap_size = rbd_dev->header.snap_sizes[which]; |
970 | } else { | |
971 | u64 size = 0; | |
972 | int ret; | |
973 | ||
974 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
975 | if (ret) | |
976 | return ret; | |
977 | ||
978 | *snap_size = size; | |
979 | } | |
980 | return 0; | |
602adf40 YS |
981 | } |
982 | ||
2ad3d716 AE |
983 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
984 | u64 *snap_features) | |
602adf40 | 985 | { |
2ad3d716 AE |
986 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
987 | if (snap_id == CEPH_NOSNAP) { | |
988 | *snap_features = rbd_dev->header.features; | |
989 | } else if (rbd_dev->image_format == 1) { | |
990 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 991 | } else { |
2ad3d716 AE |
992 | u64 features = 0; |
993 | int ret; | |
8b0241f8 | 994 | |
2ad3d716 AE |
995 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
996 | if (ret) | |
997 | return ret; | |
998 | ||
999 | *snap_features = features; | |
1000 | } | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1005 | { | |
8f4b7d98 | 1006 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1007 | u64 size = 0; |
1008 | u64 features = 0; | |
1009 | int ret; | |
1010 | ||
2ad3d716 AE |
1011 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1012 | if (ret) | |
1013 | return ret; | |
1014 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1015 | if (ret) | |
1016 | return ret; | |
1017 | ||
1018 | rbd_dev->mapping.size = size; | |
1019 | rbd_dev->mapping.features = features; | |
1020 | ||
8b0241f8 | 1021 | return 0; |
602adf40 YS |
1022 | } |
1023 | ||
d1cf5788 AE |
1024 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1025 | { | |
1026 | rbd_dev->mapping.size = 0; | |
1027 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1028 | } |
1029 | ||
98571b5a | 1030 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1031 | { |
65ccfe21 AE |
1032 | char *name; |
1033 | u64 segment; | |
1034 | int ret; | |
3a96d5cd | 1035 | char *name_format; |
602adf40 | 1036 | |
78c2a44a | 1037 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1038 | if (!name) |
1039 | return NULL; | |
1040 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1041 | name_format = "%s.%012llx"; |
1042 | if (rbd_dev->image_format == 2) | |
1043 | name_format = "%s.%016llx"; | |
1044 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, name_format, | |
65ccfe21 | 1045 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 1046 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
1047 | pr_err("error formatting segment name for #%llu (%d)\n", |
1048 | segment, ret); | |
1049 | kfree(name); | |
1050 | name = NULL; | |
1051 | } | |
602adf40 | 1052 | |
65ccfe21 AE |
1053 | return name; |
1054 | } | |
602adf40 | 1055 | |
78c2a44a AE |
1056 | static void rbd_segment_name_free(const char *name) |
1057 | { | |
1058 | /* The explicit cast here is needed to drop the const qualifier */ | |
1059 | ||
1060 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1061 | } | |
1062 | ||
65ccfe21 AE |
1063 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1064 | { | |
1065 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1066 | |
65ccfe21 AE |
1067 | return offset & (segment_size - 1); |
1068 | } | |
1069 | ||
1070 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1071 | u64 offset, u64 length) | |
1072 | { | |
1073 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1074 | ||
1075 | offset &= segment_size - 1; | |
1076 | ||
aafb230e | 1077 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1078 | if (offset + length > segment_size) |
1079 | length = segment_size - offset; | |
1080 | ||
1081 | return length; | |
602adf40 YS |
1082 | } |
1083 | ||
029bcbd8 JD |
1084 | /* |
1085 | * returns the size of an object in the image | |
1086 | */ | |
1087 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1088 | { | |
1089 | return 1 << header->obj_order; | |
1090 | } | |
1091 | ||
602adf40 YS |
1092 | /* |
1093 | * bio helpers | |
1094 | */ | |
1095 | ||
1096 | static void bio_chain_put(struct bio *chain) | |
1097 | { | |
1098 | struct bio *tmp; | |
1099 | ||
1100 | while (chain) { | |
1101 | tmp = chain; | |
1102 | chain = chain->bi_next; | |
1103 | bio_put(tmp); | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * zeros a bio chain, starting at specific offset | |
1109 | */ | |
1110 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1111 | { | |
1112 | struct bio_vec *bv; | |
1113 | unsigned long flags; | |
1114 | void *buf; | |
1115 | int i; | |
1116 | int pos = 0; | |
1117 | ||
1118 | while (chain) { | |
1119 | bio_for_each_segment(bv, chain, i) { | |
1120 | if (pos + bv->bv_len > start_ofs) { | |
1121 | int remainder = max(start_ofs - pos, 0); | |
1122 | buf = bvec_kmap_irq(bv, &flags); | |
1123 | memset(buf + remainder, 0, | |
1124 | bv->bv_len - remainder); | |
e2156054 | 1125 | flush_dcache_page(bv->bv_page); |
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 | ||
491205a8 GU |
1153 | page_offset = offset & ~PAGE_MASK; |
1154 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1155 | local_irq_save(flags); |
1156 | kaddr = kmap_atomic(*page); | |
1157 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1158 | flush_dcache_page(*page); |
b9434c5b AE |
1159 | kunmap_atomic(kaddr); |
1160 | local_irq_restore(flags); | |
1161 | ||
1162 | offset += length; | |
1163 | page++; | |
1164 | } | |
1165 | } | |
1166 | ||
602adf40 | 1167 | /* |
f7760dad AE |
1168 | * Clone a portion of a bio, starting at the given byte offset |
1169 | * and continuing for the number of bytes indicated. | |
602adf40 | 1170 | */ |
f7760dad AE |
1171 | static struct bio *bio_clone_range(struct bio *bio_src, |
1172 | unsigned int offset, | |
1173 | unsigned int len, | |
1174 | gfp_t gfpmask) | |
602adf40 | 1175 | { |
f7760dad AE |
1176 | struct bio_vec *bv; |
1177 | unsigned int resid; | |
1178 | unsigned short idx; | |
1179 | unsigned int voff; | |
1180 | unsigned short end_idx; | |
1181 | unsigned short vcnt; | |
1182 | struct bio *bio; | |
1183 | ||
1184 | /* Handle the easy case for the caller */ | |
1185 | ||
4f024f37 | 1186 | if (!offset && len == bio_src->bi_iter.bi_size) |
f7760dad AE |
1187 | return bio_clone(bio_src, gfpmask); |
1188 | ||
1189 | if (WARN_ON_ONCE(!len)) | |
1190 | return NULL; | |
4f024f37 | 1191 | if (WARN_ON_ONCE(len > bio_src->bi_iter.bi_size)) |
f7760dad | 1192 | return NULL; |
4f024f37 | 1193 | if (WARN_ON_ONCE(offset > bio_src->bi_iter.bi_size - len)) |
f7760dad AE |
1194 | return NULL; |
1195 | ||
1196 | /* Find first affected segment... */ | |
1197 | ||
1198 | resid = offset; | |
d74c6d51 | 1199 | bio_for_each_segment(bv, bio_src, idx) { |
f7760dad AE |
1200 | if (resid < bv->bv_len) |
1201 | break; | |
1202 | resid -= bv->bv_len; | |
602adf40 | 1203 | } |
f7760dad | 1204 | voff = resid; |
602adf40 | 1205 | |
f7760dad | 1206 | /* ...and the last affected segment */ |
602adf40 | 1207 | |
f7760dad AE |
1208 | resid += len; |
1209 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
1210 | if (resid <= bv->bv_len) | |
1211 | break; | |
1212 | resid -= bv->bv_len; | |
1213 | } | |
1214 | vcnt = end_idx - idx + 1; | |
1215 | ||
1216 | /* Build the clone */ | |
1217 | ||
1218 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
1219 | if (!bio) | |
1220 | return NULL; /* ENOMEM */ | |
602adf40 | 1221 | |
f7760dad | 1222 | bio->bi_bdev = bio_src->bi_bdev; |
4f024f37 KO |
1223 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector + |
1224 | (offset >> SECTOR_SHIFT); | |
f7760dad AE |
1225 | bio->bi_rw = bio_src->bi_rw; |
1226 | bio->bi_flags |= 1 << BIO_CLONED; | |
1227 | ||
1228 | /* | |
1229 | * Copy over our part of the bio_vec, then update the first | |
1230 | * and last (or only) entries. | |
1231 | */ | |
1232 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
1233 | vcnt * sizeof (struct bio_vec)); | |
1234 | bio->bi_io_vec[0].bv_offset += voff; | |
1235 | if (vcnt > 1) { | |
1236 | bio->bi_io_vec[0].bv_len -= voff; | |
1237 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
1238 | } else { | |
1239 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
1240 | } |
1241 | ||
f7760dad | 1242 | bio->bi_vcnt = vcnt; |
4f024f37 | 1243 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1244 | |
1245 | return bio; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Clone a portion of a bio chain, starting at the given byte offset | |
1250 | * into the first bio in the source chain and continuing for the | |
1251 | * number of bytes indicated. The result is another bio chain of | |
1252 | * exactly the given length, or a null pointer on error. | |
1253 | * | |
1254 | * The bio_src and offset parameters are both in-out. On entry they | |
1255 | * refer to the first source bio and the offset into that bio where | |
1256 | * the start of data to be cloned is located. | |
1257 | * | |
1258 | * On return, bio_src is updated to refer to the bio in the source | |
1259 | * chain that contains first un-cloned byte, and *offset will | |
1260 | * contain the offset of that byte within that bio. | |
1261 | */ | |
1262 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1263 | unsigned int *offset, | |
1264 | unsigned int len, | |
1265 | gfp_t gfpmask) | |
1266 | { | |
1267 | struct bio *bi = *bio_src; | |
1268 | unsigned int off = *offset; | |
1269 | struct bio *chain = NULL; | |
1270 | struct bio **end; | |
1271 | ||
1272 | /* Build up a chain of clone bios up to the limit */ | |
1273 | ||
4f024f37 | 1274 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1275 | return NULL; /* Nothing to clone */ |
602adf40 | 1276 | |
f7760dad AE |
1277 | end = &chain; |
1278 | while (len) { | |
1279 | unsigned int bi_size; | |
1280 | struct bio *bio; | |
1281 | ||
f5400b7a AE |
1282 | if (!bi) { |
1283 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1284 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1285 | } |
4f024f37 | 1286 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1287 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1288 | if (!bio) | |
1289 | goto out_err; /* ENOMEM */ | |
1290 | ||
1291 | *end = bio; | |
1292 | end = &bio->bi_next; | |
602adf40 | 1293 | |
f7760dad | 1294 | off += bi_size; |
4f024f37 | 1295 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1296 | bi = bi->bi_next; |
1297 | off = 0; | |
1298 | } | |
1299 | len -= bi_size; | |
1300 | } | |
1301 | *bio_src = bi; | |
1302 | *offset = off; | |
1303 | ||
1304 | return chain; | |
1305 | out_err: | |
1306 | bio_chain_put(chain); | |
602adf40 | 1307 | |
602adf40 YS |
1308 | return NULL; |
1309 | } | |
1310 | ||
926f9b3f AE |
1311 | /* |
1312 | * The default/initial value for all object request flags is 0. For | |
1313 | * each flag, once its value is set to 1 it is never reset to 0 | |
1314 | * again. | |
1315 | */ | |
57acbaa7 | 1316 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1317 | { |
57acbaa7 | 1318 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1319 | struct rbd_device *rbd_dev; |
1320 | ||
57acbaa7 AE |
1321 | rbd_dev = obj_request->img_request->rbd_dev; |
1322 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1323 | obj_request); |
1324 | } | |
1325 | } | |
1326 | ||
57acbaa7 | 1327 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1328 | { |
1329 | smp_mb(); | |
57acbaa7 | 1330 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1331 | } |
1332 | ||
57acbaa7 | 1333 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1334 | { |
57acbaa7 AE |
1335 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1336 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1337 | |
57acbaa7 AE |
1338 | if (obj_request_img_data_test(obj_request)) |
1339 | rbd_dev = obj_request->img_request->rbd_dev; | |
1340 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1341 | obj_request); |
1342 | } | |
1343 | } | |
1344 | ||
57acbaa7 | 1345 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1346 | { |
1347 | smp_mb(); | |
57acbaa7 | 1348 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1349 | } |
1350 | ||
5679c59f AE |
1351 | /* |
1352 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1353 | * flag. The latter is set based on the "exists" value provided. | |
1354 | * | |
1355 | * Note that for our purposes once an object exists it never goes | |
1356 | * away again. It's possible that the response from two existence | |
1357 | * checks are separated by the creation of the target object, and | |
1358 | * the first ("doesn't exist") response arrives *after* the second | |
1359 | * ("does exist"). In that case we ignore the second one. | |
1360 | */ | |
1361 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1362 | bool exists) | |
1363 | { | |
1364 | if (exists) | |
1365 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1366 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1367 | smp_mb(); | |
1368 | } | |
1369 | ||
1370 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1371 | { | |
1372 | smp_mb(); | |
1373 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1374 | } | |
1375 | ||
1376 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1377 | { | |
1378 | smp_mb(); | |
1379 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1380 | } | |
1381 | ||
bf0d5f50 AE |
1382 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1383 | { | |
37206ee5 AE |
1384 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1385 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1386 | kref_get(&obj_request->kref); |
1387 | } | |
1388 | ||
1389 | static void rbd_obj_request_destroy(struct kref *kref); | |
1390 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1391 | { | |
1392 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1393 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1394 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1395 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1396 | } | |
1397 | ||
e93f3152 AE |
1398 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1399 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1400 | static void rbd_img_request_destroy(struct kref *kref); |
1401 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1402 | { | |
1403 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1404 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1405 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1406 | if (img_request_child_test(img_request)) |
1407 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1408 | else | |
1409 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1410 | } |
1411 | ||
1412 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1413 | struct rbd_obj_request *obj_request) | |
1414 | { | |
25dcf954 AE |
1415 | rbd_assert(obj_request->img_request == NULL); |
1416 | ||
b155e86c | 1417 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1418 | obj_request->img_request = img_request; |
25dcf954 | 1419 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1420 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1421 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1422 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1423 | img_request->obj_request_count++; |
1424 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1425 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1426 | obj_request->which); | |
bf0d5f50 AE |
1427 | } |
1428 | ||
1429 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1430 | struct rbd_obj_request *obj_request) | |
1431 | { | |
1432 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1433 | |
37206ee5 AE |
1434 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1435 | obj_request->which); | |
bf0d5f50 | 1436 | list_del(&obj_request->links); |
25dcf954 AE |
1437 | rbd_assert(img_request->obj_request_count > 0); |
1438 | img_request->obj_request_count--; | |
1439 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1440 | obj_request->which = BAD_WHICH; | |
6365d33a | 1441 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1442 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1443 | obj_request->img_request = NULL; |
25dcf954 | 1444 | obj_request->callback = NULL; |
bf0d5f50 AE |
1445 | rbd_obj_request_put(obj_request); |
1446 | } | |
1447 | ||
1448 | static bool obj_request_type_valid(enum obj_request_type type) | |
1449 | { | |
1450 | switch (type) { | |
9969ebc5 | 1451 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1452 | case OBJ_REQUEST_BIO: |
788e2df3 | 1453 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1454 | return true; |
1455 | default: | |
1456 | return false; | |
1457 | } | |
1458 | } | |
1459 | ||
bf0d5f50 AE |
1460 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1461 | struct rbd_obj_request *obj_request) | |
1462 | { | |
37206ee5 AE |
1463 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1464 | ||
bf0d5f50 AE |
1465 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1466 | } | |
1467 | ||
1468 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1469 | { | |
55f27e09 | 1470 | |
37206ee5 | 1471 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1472 | |
1473 | /* | |
1474 | * If no error occurred, compute the aggregate transfer | |
1475 | * count for the image request. We could instead use | |
1476 | * atomic64_cmpxchg() to update it as each object request | |
1477 | * completes; not clear which way is better off hand. | |
1478 | */ | |
1479 | if (!img_request->result) { | |
1480 | struct rbd_obj_request *obj_request; | |
1481 | u64 xferred = 0; | |
1482 | ||
1483 | for_each_obj_request(img_request, obj_request) | |
1484 | xferred += obj_request->xferred; | |
1485 | img_request->xferred = xferred; | |
1486 | } | |
1487 | ||
bf0d5f50 AE |
1488 | if (img_request->callback) |
1489 | img_request->callback(img_request); | |
1490 | else | |
1491 | rbd_img_request_put(img_request); | |
1492 | } | |
1493 | ||
788e2df3 AE |
1494 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1495 | ||
1496 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1497 | { | |
37206ee5 AE |
1498 | dout("%s: obj %p\n", __func__, obj_request); |
1499 | ||
788e2df3 AE |
1500 | return wait_for_completion_interruptible(&obj_request->completion); |
1501 | } | |
1502 | ||
0c425248 AE |
1503 | /* |
1504 | * The default/initial value for all image request flags is 0. Each | |
1505 | * is conditionally set to 1 at image request initialization time | |
1506 | * and currently never change thereafter. | |
1507 | */ | |
1508 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1509 | { | |
1510 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1511 | smp_mb(); | |
1512 | } | |
1513 | ||
1514 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1515 | { | |
1516 | smp_mb(); | |
1517 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1518 | } | |
1519 | ||
9849e986 AE |
1520 | static void img_request_child_set(struct rbd_img_request *img_request) |
1521 | { | |
1522 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1523 | smp_mb(); | |
1524 | } | |
1525 | ||
e93f3152 AE |
1526 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1527 | { | |
1528 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1529 | smp_mb(); | |
1530 | } | |
1531 | ||
9849e986 AE |
1532 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1533 | { | |
1534 | smp_mb(); | |
1535 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1536 | } | |
1537 | ||
d0b2e944 AE |
1538 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1539 | { | |
1540 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1541 | smp_mb(); | |
1542 | } | |
1543 | ||
a2acd00e AE |
1544 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1545 | { | |
1546 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1547 | smp_mb(); | |
1548 | } | |
1549 | ||
d0b2e944 AE |
1550 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1551 | { | |
1552 | smp_mb(); | |
1553 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1554 | } | |
1555 | ||
6e2a4505 AE |
1556 | static void |
1557 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1558 | { | |
b9434c5b AE |
1559 | u64 xferred = obj_request->xferred; |
1560 | u64 length = obj_request->length; | |
1561 | ||
6e2a4505 AE |
1562 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1563 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1564 | xferred, length); |
6e2a4505 | 1565 | /* |
17c1cc1d JD |
1566 | * ENOENT means a hole in the image. We zero-fill the entire |
1567 | * length of the request. A short read also implies zero-fill | |
1568 | * to the end of the request. An error requires the whole | |
1569 | * length of the request to be reported finished with an error | |
1570 | * to the block layer. In each case we update the xferred | |
1571 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1572 | */ |
b9434c5b | 1573 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1574 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1575 | if (obj_request->type == OBJ_REQUEST_BIO) |
1576 | zero_bio_chain(obj_request->bio_list, 0); | |
1577 | else | |
1578 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1579 | obj_request->result = 0; |
b9434c5b AE |
1580 | } else if (xferred < length && !obj_request->result) { |
1581 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1582 | zero_bio_chain(obj_request->bio_list, xferred); | |
1583 | else | |
1584 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1585 | } |
17c1cc1d | 1586 | obj_request->xferred = length; |
6e2a4505 AE |
1587 | obj_request_done_set(obj_request); |
1588 | } | |
1589 | ||
bf0d5f50 AE |
1590 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1591 | { | |
37206ee5 AE |
1592 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1593 | obj_request->callback); | |
bf0d5f50 AE |
1594 | if (obj_request->callback) |
1595 | obj_request->callback(obj_request); | |
788e2df3 AE |
1596 | else |
1597 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1598 | } |
1599 | ||
c47f9371 | 1600 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1601 | { |
1602 | dout("%s: obj %p\n", __func__, obj_request); | |
1603 | obj_request_done_set(obj_request); | |
1604 | } | |
1605 | ||
c47f9371 | 1606 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1607 | { |
57acbaa7 | 1608 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1609 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1610 | bool layered = false; |
1611 | ||
1612 | if (obj_request_img_data_test(obj_request)) { | |
1613 | img_request = obj_request->img_request; | |
1614 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1615 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1616 | } |
8b3e1a56 AE |
1617 | |
1618 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1619 | obj_request, img_request, obj_request->result, | |
1620 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1621 | if (layered && obj_request->result == -ENOENT && |
1622 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1623 | rbd_img_parent_read(obj_request); |
1624 | else if (img_request) | |
6e2a4505 AE |
1625 | rbd_img_obj_request_read_callback(obj_request); |
1626 | else | |
1627 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1628 | } |
1629 | ||
c47f9371 | 1630 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1631 | { |
1b83bef2 SW |
1632 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1633 | obj_request->result, obj_request->length); | |
1634 | /* | |
8b3e1a56 AE |
1635 | * There is no such thing as a successful short write. Set |
1636 | * it to our originally-requested length. | |
1b83bef2 SW |
1637 | */ |
1638 | obj_request->xferred = obj_request->length; | |
07741308 | 1639 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1640 | } |
1641 | ||
fbfab539 AE |
1642 | /* |
1643 | * For a simple stat call there's nothing to do. We'll do more if | |
1644 | * this is part of a write sequence for a layered image. | |
1645 | */ | |
c47f9371 | 1646 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1647 | { |
37206ee5 | 1648 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1649 | obj_request_done_set(obj_request); |
1650 | } | |
1651 | ||
bf0d5f50 AE |
1652 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1653 | struct ceph_msg *msg) | |
1654 | { | |
1655 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1656 | u16 opcode; |
1657 | ||
37206ee5 | 1658 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1659 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1660 | if (obj_request_img_data_test(obj_request)) { |
1661 | rbd_assert(obj_request->img_request); | |
1662 | rbd_assert(obj_request->which != BAD_WHICH); | |
1663 | } else { | |
1664 | rbd_assert(obj_request->which == BAD_WHICH); | |
1665 | } | |
bf0d5f50 | 1666 | |
1b83bef2 SW |
1667 | if (osd_req->r_result < 0) |
1668 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1669 | |
0eefd470 | 1670 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1671 | |
c47f9371 AE |
1672 | /* |
1673 | * We support a 64-bit length, but ultimately it has to be | |
1674 | * passed to blk_end_request(), which takes an unsigned int. | |
1675 | */ | |
1b83bef2 | 1676 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1677 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1678 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1679 | switch (opcode) { |
1680 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1681 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1682 | break; |
1683 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1684 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1685 | break; |
fbfab539 | 1686 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1687 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1688 | break; |
36be9a76 | 1689 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1690 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1691 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1692 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1693 | break; |
bf0d5f50 AE |
1694 | default: |
1695 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1696 | obj_request->object_name, (unsigned short) opcode); | |
1697 | break; | |
1698 | } | |
1699 | ||
07741308 | 1700 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1701 | rbd_obj_request_complete(obj_request); |
1702 | } | |
1703 | ||
9d4df01f | 1704 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1705 | { |
1706 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1707 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1708 | u64 snap_id; |
430c28c3 | 1709 | |
8c042b0d | 1710 | rbd_assert(osd_req != NULL); |
430c28c3 | 1711 | |
9d4df01f | 1712 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1713 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1714 | NULL, snap_id, NULL); |
1715 | } | |
1716 | ||
1717 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1718 | { | |
1719 | struct rbd_img_request *img_request = obj_request->img_request; | |
1720 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1721 | struct ceph_snap_context *snapc; | |
1722 | struct timespec mtime = CURRENT_TIME; | |
1723 | ||
1724 | rbd_assert(osd_req != NULL); | |
1725 | ||
1726 | snapc = img_request ? img_request->snapc : NULL; | |
1727 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1728 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1729 | } |
1730 | ||
bf0d5f50 AE |
1731 | static struct ceph_osd_request *rbd_osd_req_create( |
1732 | struct rbd_device *rbd_dev, | |
1733 | bool write_request, | |
430c28c3 | 1734 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1735 | { |
bf0d5f50 AE |
1736 | struct ceph_snap_context *snapc = NULL; |
1737 | struct ceph_osd_client *osdc; | |
1738 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1739 | |
6365d33a AE |
1740 | if (obj_request_img_data_test(obj_request)) { |
1741 | struct rbd_img_request *img_request = obj_request->img_request; | |
1742 | ||
0c425248 AE |
1743 | rbd_assert(write_request == |
1744 | img_request_write_test(img_request)); | |
1745 | if (write_request) | |
bf0d5f50 | 1746 | snapc = img_request->snapc; |
bf0d5f50 AE |
1747 | } |
1748 | ||
1749 | /* Allocate and initialize the request, for the single op */ | |
1750 | ||
1751 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1752 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1753 | if (!osd_req) | |
1754 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1755 | |
430c28c3 | 1756 | if (write_request) |
bf0d5f50 | 1757 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1758 | else |
bf0d5f50 | 1759 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1760 | |
1761 | osd_req->r_callback = rbd_osd_req_callback; | |
1762 | osd_req->r_priv = obj_request; | |
1763 | ||
1764 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1765 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1766 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1767 | ||
1768 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1769 | ||
bf0d5f50 AE |
1770 | return osd_req; |
1771 | } | |
1772 | ||
0eefd470 AE |
1773 | /* |
1774 | * Create a copyup osd request based on the information in the | |
1775 | * object request supplied. A copyup request has two osd ops, | |
1776 | * a copyup method call, and a "normal" write request. | |
1777 | */ | |
1778 | static struct ceph_osd_request * | |
1779 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1780 | { | |
1781 | struct rbd_img_request *img_request; | |
1782 | struct ceph_snap_context *snapc; | |
1783 | struct rbd_device *rbd_dev; | |
1784 | struct ceph_osd_client *osdc; | |
1785 | struct ceph_osd_request *osd_req; | |
1786 | ||
1787 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1788 | img_request = obj_request->img_request; | |
1789 | rbd_assert(img_request); | |
1790 | rbd_assert(img_request_write_test(img_request)); | |
1791 | ||
1792 | /* Allocate and initialize the request, for the two ops */ | |
1793 | ||
1794 | snapc = img_request->snapc; | |
1795 | rbd_dev = img_request->rbd_dev; | |
1796 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1797 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1798 | if (!osd_req) | |
1799 | return NULL; /* ENOMEM */ | |
1800 | ||
1801 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1802 | osd_req->r_callback = rbd_osd_req_callback; | |
1803 | osd_req->r_priv = obj_request; | |
1804 | ||
1805 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1806 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1807 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1808 | ||
1809 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1810 | ||
1811 | return osd_req; | |
1812 | } | |
1813 | ||
1814 | ||
bf0d5f50 AE |
1815 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1816 | { | |
1817 | ceph_osdc_put_request(osd_req); | |
1818 | } | |
1819 | ||
1820 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1821 | ||
1822 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1823 | u64 offset, u64 length, | |
1824 | enum obj_request_type type) | |
1825 | { | |
1826 | struct rbd_obj_request *obj_request; | |
1827 | size_t size; | |
1828 | char *name; | |
1829 | ||
1830 | rbd_assert(obj_request_type_valid(type)); | |
1831 | ||
1832 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1833 | name = kmalloc(size, GFP_KERNEL); |
1834 | if (!name) | |
bf0d5f50 AE |
1835 | return NULL; |
1836 | ||
868311b1 | 1837 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1838 | if (!obj_request) { |
1839 | kfree(name); | |
1840 | return NULL; | |
1841 | } | |
1842 | ||
bf0d5f50 AE |
1843 | obj_request->object_name = memcpy(name, object_name, size); |
1844 | obj_request->offset = offset; | |
1845 | obj_request->length = length; | |
926f9b3f | 1846 | obj_request->flags = 0; |
bf0d5f50 AE |
1847 | obj_request->which = BAD_WHICH; |
1848 | obj_request->type = type; | |
1849 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1850 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1851 | kref_init(&obj_request->kref); |
1852 | ||
37206ee5 AE |
1853 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1854 | offset, length, (int)type, obj_request); | |
1855 | ||
bf0d5f50 AE |
1856 | return obj_request; |
1857 | } | |
1858 | ||
1859 | static void rbd_obj_request_destroy(struct kref *kref) | |
1860 | { | |
1861 | struct rbd_obj_request *obj_request; | |
1862 | ||
1863 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1864 | ||
37206ee5 AE |
1865 | dout("%s: obj %p\n", __func__, obj_request); |
1866 | ||
bf0d5f50 AE |
1867 | rbd_assert(obj_request->img_request == NULL); |
1868 | rbd_assert(obj_request->which == BAD_WHICH); | |
1869 | ||
1870 | if (obj_request->osd_req) | |
1871 | rbd_osd_req_destroy(obj_request->osd_req); | |
1872 | ||
1873 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1874 | switch (obj_request->type) { | |
9969ebc5 AE |
1875 | case OBJ_REQUEST_NODATA: |
1876 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1877 | case OBJ_REQUEST_BIO: |
1878 | if (obj_request->bio_list) | |
1879 | bio_chain_put(obj_request->bio_list); | |
1880 | break; | |
788e2df3 AE |
1881 | case OBJ_REQUEST_PAGES: |
1882 | if (obj_request->pages) | |
1883 | ceph_release_page_vector(obj_request->pages, | |
1884 | obj_request->page_count); | |
1885 | break; | |
bf0d5f50 AE |
1886 | } |
1887 | ||
f907ad55 | 1888 | kfree(obj_request->object_name); |
868311b1 AE |
1889 | obj_request->object_name = NULL; |
1890 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
1891 | } |
1892 | ||
fb65d228 AE |
1893 | /* It's OK to call this for a device with no parent */ |
1894 | ||
1895 | static void rbd_spec_put(struct rbd_spec *spec); | |
1896 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
1897 | { | |
1898 | rbd_dev_remove_parent(rbd_dev); | |
1899 | rbd_spec_put(rbd_dev->parent_spec); | |
1900 | rbd_dev->parent_spec = NULL; | |
1901 | rbd_dev->parent_overlap = 0; | |
1902 | } | |
1903 | ||
a2acd00e AE |
1904 | /* |
1905 | * Parent image reference counting is used to determine when an | |
1906 | * image's parent fields can be safely torn down--after there are no | |
1907 | * more in-flight requests to the parent image. When the last | |
1908 | * reference is dropped, cleaning them up is safe. | |
1909 | */ | |
1910 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
1911 | { | |
1912 | int counter; | |
1913 | ||
1914 | if (!rbd_dev->parent_spec) | |
1915 | return; | |
1916 | ||
1917 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
1918 | if (counter > 0) | |
1919 | return; | |
1920 | ||
1921 | /* Last reference; clean up parent data structures */ | |
1922 | ||
1923 | if (!counter) | |
1924 | rbd_dev_unparent(rbd_dev); | |
1925 | else | |
1926 | rbd_warn(rbd_dev, "parent reference underflow\n"); | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * If an image has a non-zero parent overlap, get a reference to its | |
1931 | * parent. | |
1932 | * | |
392a9dad AE |
1933 | * We must get the reference before checking for the overlap to |
1934 | * coordinate properly with zeroing the parent overlap in | |
1935 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
1936 | * drop it again if there is no overlap. | |
1937 | * | |
a2acd00e AE |
1938 | * Returns true if the rbd device has a parent with a non-zero |
1939 | * overlap and a reference for it was successfully taken, or | |
1940 | * false otherwise. | |
1941 | */ | |
1942 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
1943 | { | |
1944 | int counter; | |
1945 | ||
1946 | if (!rbd_dev->parent_spec) | |
1947 | return false; | |
1948 | ||
1949 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
1950 | if (counter > 0 && rbd_dev->parent_overlap) | |
1951 | return true; | |
1952 | ||
1953 | /* Image was flattened, but parent is not yet torn down */ | |
1954 | ||
1955 | if (counter < 0) | |
1956 | rbd_warn(rbd_dev, "parent reference overflow\n"); | |
1957 | ||
1958 | return false; | |
1959 | } | |
1960 | ||
bf0d5f50 AE |
1961 | /* |
1962 | * Caller is responsible for filling in the list of object requests | |
1963 | * that comprises the image request, and the Linux request pointer | |
1964 | * (if there is one). | |
1965 | */ | |
cc344fa1 AE |
1966 | static struct rbd_img_request *rbd_img_request_create( |
1967 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1968 | u64 offset, u64 length, |
e93f3152 | 1969 | bool write_request) |
bf0d5f50 AE |
1970 | { |
1971 | struct rbd_img_request *img_request; | |
bf0d5f50 | 1972 | |
1c2a9dfe | 1973 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC); |
bf0d5f50 AE |
1974 | if (!img_request) |
1975 | return NULL; | |
1976 | ||
1977 | if (write_request) { | |
1978 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 1979 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 1980 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1981 | } |
1982 | ||
1983 | img_request->rq = NULL; | |
1984 | img_request->rbd_dev = rbd_dev; | |
1985 | img_request->offset = offset; | |
1986 | img_request->length = length; | |
0c425248 AE |
1987 | img_request->flags = 0; |
1988 | if (write_request) { | |
1989 | img_request_write_set(img_request); | |
468521c1 | 1990 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1991 | } else { |
bf0d5f50 | 1992 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1993 | } |
a2acd00e | 1994 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 1995 | img_request_layered_set(img_request); |
bf0d5f50 AE |
1996 | spin_lock_init(&img_request->completion_lock); |
1997 | img_request->next_completion = 0; | |
1998 | img_request->callback = NULL; | |
a5a337d4 | 1999 | img_request->result = 0; |
bf0d5f50 AE |
2000 | img_request->obj_request_count = 0; |
2001 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2002 | kref_init(&img_request->kref); | |
2003 | ||
37206ee5 AE |
2004 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
2005 | write_request ? "write" : "read", offset, length, | |
2006 | img_request); | |
2007 | ||
bf0d5f50 AE |
2008 | return img_request; |
2009 | } | |
2010 | ||
2011 | static void rbd_img_request_destroy(struct kref *kref) | |
2012 | { | |
2013 | struct rbd_img_request *img_request; | |
2014 | struct rbd_obj_request *obj_request; | |
2015 | struct rbd_obj_request *next_obj_request; | |
2016 | ||
2017 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2018 | ||
37206ee5 AE |
2019 | dout("%s: img %p\n", __func__, img_request); |
2020 | ||
bf0d5f50 AE |
2021 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2022 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2023 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2024 | |
a2acd00e AE |
2025 | if (img_request_layered_test(img_request)) { |
2026 | img_request_layered_clear(img_request); | |
2027 | rbd_dev_parent_put(img_request->rbd_dev); | |
2028 | } | |
2029 | ||
0c425248 | 2030 | if (img_request_write_test(img_request)) |
812164f8 | 2031 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2032 | |
1c2a9dfe | 2033 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2034 | } |
2035 | ||
e93f3152 AE |
2036 | static struct rbd_img_request *rbd_parent_request_create( |
2037 | struct rbd_obj_request *obj_request, | |
2038 | u64 img_offset, u64 length) | |
2039 | { | |
2040 | struct rbd_img_request *parent_request; | |
2041 | struct rbd_device *rbd_dev; | |
2042 | ||
2043 | rbd_assert(obj_request->img_request); | |
2044 | rbd_dev = obj_request->img_request->rbd_dev; | |
2045 | ||
2046 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2047 | img_offset, length, false); | |
2048 | if (!parent_request) | |
2049 | return NULL; | |
2050 | ||
2051 | img_request_child_set(parent_request); | |
2052 | rbd_obj_request_get(obj_request); | |
2053 | parent_request->obj_request = obj_request; | |
2054 | ||
2055 | return parent_request; | |
2056 | } | |
2057 | ||
2058 | static void rbd_parent_request_destroy(struct kref *kref) | |
2059 | { | |
2060 | struct rbd_img_request *parent_request; | |
2061 | struct rbd_obj_request *orig_request; | |
2062 | ||
2063 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2064 | orig_request = parent_request->obj_request; | |
2065 | ||
2066 | parent_request->obj_request = NULL; | |
2067 | rbd_obj_request_put(orig_request); | |
2068 | img_request_child_clear(parent_request); | |
2069 | ||
2070 | rbd_img_request_destroy(kref); | |
2071 | } | |
2072 | ||
1217857f AE |
2073 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2074 | { | |
6365d33a | 2075 | struct rbd_img_request *img_request; |
1217857f AE |
2076 | unsigned int xferred; |
2077 | int result; | |
8b3e1a56 | 2078 | bool more; |
1217857f | 2079 | |
6365d33a AE |
2080 | rbd_assert(obj_request_img_data_test(obj_request)); |
2081 | img_request = obj_request->img_request; | |
2082 | ||
1217857f AE |
2083 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2084 | xferred = (unsigned int)obj_request->xferred; | |
2085 | result = obj_request->result; | |
2086 | if (result) { | |
2087 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2088 | ||
2089 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
2090 | img_request_write_test(img_request) ? "write" : "read", | |
2091 | obj_request->length, obj_request->img_offset, | |
2092 | obj_request->offset); | |
2093 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
2094 | result, xferred); | |
2095 | if (!img_request->result) | |
2096 | img_request->result = result; | |
2097 | } | |
2098 | ||
f1a4739f AE |
2099 | /* Image object requests don't own their page array */ |
2100 | ||
2101 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2102 | obj_request->pages = NULL; | |
2103 | obj_request->page_count = 0; | |
2104 | } | |
2105 | ||
8b3e1a56 AE |
2106 | if (img_request_child_test(img_request)) { |
2107 | rbd_assert(img_request->obj_request != NULL); | |
2108 | more = obj_request->which < img_request->obj_request_count - 1; | |
2109 | } else { | |
2110 | rbd_assert(img_request->rq != NULL); | |
2111 | more = blk_end_request(img_request->rq, result, xferred); | |
2112 | } | |
2113 | ||
2114 | return more; | |
1217857f AE |
2115 | } |
2116 | ||
2169238d AE |
2117 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2118 | { | |
2119 | struct rbd_img_request *img_request; | |
2120 | u32 which = obj_request->which; | |
2121 | bool more = true; | |
2122 | ||
6365d33a | 2123 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2124 | img_request = obj_request->img_request; |
2125 | ||
2126 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2127 | rbd_assert(img_request != NULL); | |
2169238d AE |
2128 | rbd_assert(img_request->obj_request_count > 0); |
2129 | rbd_assert(which != BAD_WHICH); | |
2130 | rbd_assert(which < img_request->obj_request_count); | |
2131 | rbd_assert(which >= img_request->next_completion); | |
2132 | ||
2133 | spin_lock_irq(&img_request->completion_lock); | |
2134 | if (which != img_request->next_completion) | |
2135 | goto out; | |
2136 | ||
2137 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2138 | rbd_assert(more); |
2139 | rbd_assert(which < img_request->obj_request_count); | |
2140 | ||
2141 | if (!obj_request_done_test(obj_request)) | |
2142 | break; | |
1217857f | 2143 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2144 | which++; |
2145 | } | |
2146 | ||
2147 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2148 | img_request->next_completion = which; | |
2149 | out: | |
2150 | spin_unlock_irq(&img_request->completion_lock); | |
2151 | ||
2152 | if (!more) | |
2153 | rbd_img_request_complete(img_request); | |
2154 | } | |
2155 | ||
f1a4739f AE |
2156 | /* |
2157 | * Split up an image request into one or more object requests, each | |
2158 | * to a different object. The "type" parameter indicates whether | |
2159 | * "data_desc" is the pointer to the head of a list of bio | |
2160 | * structures, or the base of a page array. In either case this | |
2161 | * function assumes data_desc describes memory sufficient to hold | |
2162 | * all data described by the image request. | |
2163 | */ | |
2164 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2165 | enum obj_request_type type, | |
2166 | void *data_desc) | |
bf0d5f50 AE |
2167 | { |
2168 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2169 | struct rbd_obj_request *obj_request = NULL; | |
2170 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 2171 | bool write_request = img_request_write_test(img_request); |
a158073c | 2172 | struct bio *bio_list = NULL; |
f1a4739f | 2173 | unsigned int bio_offset = 0; |
a158073c | 2174 | struct page **pages = NULL; |
7da22d29 | 2175 | u64 img_offset; |
bf0d5f50 AE |
2176 | u64 resid; |
2177 | u16 opcode; | |
2178 | ||
f1a4739f AE |
2179 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2180 | (int)type, data_desc); | |
37206ee5 | 2181 | |
430c28c3 | 2182 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 2183 | img_offset = img_request->offset; |
bf0d5f50 | 2184 | resid = img_request->length; |
4dda41d3 | 2185 | rbd_assert(resid > 0); |
f1a4739f AE |
2186 | |
2187 | if (type == OBJ_REQUEST_BIO) { | |
2188 | bio_list = data_desc; | |
4f024f37 KO |
2189 | rbd_assert(img_offset == |
2190 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
f1a4739f AE |
2191 | } else { |
2192 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
2193 | pages = data_desc; | |
2194 | } | |
2195 | ||
bf0d5f50 | 2196 | while (resid) { |
2fa12320 | 2197 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2198 | const char *object_name; |
bf0d5f50 AE |
2199 | u64 offset; |
2200 | u64 length; | |
2201 | ||
7da22d29 | 2202 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2203 | if (!object_name) |
2204 | goto out_unwind; | |
7da22d29 AE |
2205 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2206 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2207 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2208 | offset, length, type); |
78c2a44a AE |
2209 | /* object request has its own copy of the object name */ |
2210 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2211 | if (!obj_request) |
2212 | goto out_unwind; | |
03507db6 JD |
2213 | /* |
2214 | * set obj_request->img_request before creating the | |
2215 | * osd_request so that it gets the right snapc | |
2216 | */ | |
2217 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2218 | |
f1a4739f AE |
2219 | if (type == OBJ_REQUEST_BIO) { |
2220 | unsigned int clone_size; | |
2221 | ||
2222 | rbd_assert(length <= (u64)UINT_MAX); | |
2223 | clone_size = (unsigned int)length; | |
2224 | obj_request->bio_list = | |
2225 | bio_chain_clone_range(&bio_list, | |
2226 | &bio_offset, | |
2227 | clone_size, | |
2228 | GFP_ATOMIC); | |
2229 | if (!obj_request->bio_list) | |
2230 | goto out_partial; | |
2231 | } else { | |
2232 | unsigned int page_count; | |
2233 | ||
2234 | obj_request->pages = pages; | |
2235 | page_count = (u32)calc_pages_for(offset, length); | |
2236 | obj_request->page_count = page_count; | |
2237 | if ((offset + length) & ~PAGE_MASK) | |
2238 | page_count--; /* more on last page */ | |
2239 | pages += page_count; | |
2240 | } | |
bf0d5f50 | 2241 | |
2fa12320 AE |
2242 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
2243 | obj_request); | |
2244 | if (!osd_req) | |
bf0d5f50 | 2245 | goto out_partial; |
2fa12320 | 2246 | obj_request->osd_req = osd_req; |
2169238d | 2247 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 2248 | |
2fa12320 AE |
2249 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
2250 | 0, 0); | |
f1a4739f AE |
2251 | if (type == OBJ_REQUEST_BIO) |
2252 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
2253 | obj_request->bio_list, length); | |
2254 | else | |
2255 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
2256 | obj_request->pages, length, | |
2257 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
2258 | |
2259 | if (write_request) | |
2260 | rbd_osd_req_format_write(obj_request); | |
2261 | else | |
2262 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2263 | |
7da22d29 | 2264 | obj_request->img_offset = img_offset; |
bf0d5f50 | 2265 | |
7da22d29 | 2266 | img_offset += length; |
bf0d5f50 AE |
2267 | resid -= length; |
2268 | } | |
2269 | ||
2270 | return 0; | |
2271 | ||
2272 | out_partial: | |
2273 | rbd_obj_request_put(obj_request); | |
2274 | out_unwind: | |
2275 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
2276 | rbd_obj_request_put(obj_request); | |
2277 | ||
2278 | return -ENOMEM; | |
2279 | } | |
2280 | ||
0eefd470 AE |
2281 | static void |
2282 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2283 | { | |
2284 | struct rbd_img_request *img_request; | |
2285 | struct rbd_device *rbd_dev; | |
ebda6408 | 2286 | struct page **pages; |
0eefd470 AE |
2287 | u32 page_count; |
2288 | ||
2289 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2290 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2291 | img_request = obj_request->img_request; | |
2292 | rbd_assert(img_request); | |
2293 | ||
2294 | rbd_dev = img_request->rbd_dev; | |
2295 | rbd_assert(rbd_dev); | |
0eefd470 | 2296 | |
ebda6408 AE |
2297 | pages = obj_request->copyup_pages; |
2298 | rbd_assert(pages != NULL); | |
0eefd470 | 2299 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2300 | page_count = obj_request->copyup_page_count; |
2301 | rbd_assert(page_count); | |
2302 | obj_request->copyup_page_count = 0; | |
2303 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2304 | |
2305 | /* | |
2306 | * We want the transfer count to reflect the size of the | |
2307 | * original write request. There is no such thing as a | |
2308 | * successful short write, so if the request was successful | |
2309 | * we can just set it to the originally-requested length. | |
2310 | */ | |
2311 | if (!obj_request->result) | |
2312 | obj_request->xferred = obj_request->length; | |
2313 | ||
2314 | /* Finish up with the normal image object callback */ | |
2315 | ||
2316 | rbd_img_obj_callback(obj_request); | |
2317 | } | |
2318 | ||
3d7efd18 AE |
2319 | static void |
2320 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2321 | { | |
2322 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2323 | struct ceph_osd_request *osd_req; |
2324 | struct ceph_osd_client *osdc; | |
2325 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2326 | struct page **pages; |
ebda6408 | 2327 | u32 page_count; |
bbea1c1a | 2328 | int img_result; |
ebda6408 | 2329 | u64 parent_length; |
b91f09f1 AE |
2330 | u64 offset; |
2331 | u64 length; | |
3d7efd18 AE |
2332 | |
2333 | rbd_assert(img_request_child_test(img_request)); | |
2334 | ||
2335 | /* First get what we need from the image request */ | |
2336 | ||
2337 | pages = img_request->copyup_pages; | |
2338 | rbd_assert(pages != NULL); | |
2339 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2340 | page_count = img_request->copyup_page_count; |
2341 | rbd_assert(page_count); | |
2342 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2343 | |
2344 | orig_request = img_request->obj_request; | |
2345 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2346 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2347 | img_result = img_request->result; |
ebda6408 AE |
2348 | parent_length = img_request->length; |
2349 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2350 | rbd_img_request_put(img_request); |
3d7efd18 | 2351 | |
91c6febb AE |
2352 | rbd_assert(orig_request->img_request); |
2353 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2354 | rbd_assert(rbd_dev); |
0eefd470 | 2355 | |
bbea1c1a AE |
2356 | /* |
2357 | * If the overlap has become 0 (most likely because the | |
2358 | * image has been flattened) we need to free the pages | |
2359 | * and re-submit the original write request. | |
2360 | */ | |
2361 | if (!rbd_dev->parent_overlap) { | |
2362 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2363 | |
bbea1c1a AE |
2364 | ceph_release_page_vector(pages, page_count); |
2365 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2366 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2367 | if (!img_result) | |
2368 | return; | |
2369 | } | |
0eefd470 | 2370 | |
bbea1c1a | 2371 | if (img_result) |
0eefd470 | 2372 | goto out_err; |
0eefd470 | 2373 | |
8785b1d4 AE |
2374 | /* |
2375 | * The original osd request is of no use to use any more. | |
2376 | * We need a new one that can hold the two ops in a copyup | |
2377 | * request. Allocate the new copyup osd request for the | |
2378 | * original request, and release the old one. | |
2379 | */ | |
bbea1c1a | 2380 | img_result = -ENOMEM; |
0eefd470 AE |
2381 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2382 | if (!osd_req) | |
2383 | goto out_err; | |
8785b1d4 | 2384 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2385 | orig_request->osd_req = osd_req; |
2386 | orig_request->copyup_pages = pages; | |
ebda6408 | 2387 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2388 | |
0eefd470 | 2389 | /* Initialize the copyup op */ |
3d7efd18 | 2390 | |
0eefd470 | 2391 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2392 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2393 | false, false); |
3d7efd18 | 2394 | |
0eefd470 AE |
2395 | /* Then the original write request op */ |
2396 | ||
b91f09f1 AE |
2397 | offset = orig_request->offset; |
2398 | length = orig_request->length; | |
0eefd470 | 2399 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, |
b91f09f1 AE |
2400 | offset, length, 0, 0); |
2401 | if (orig_request->type == OBJ_REQUEST_BIO) | |
2402 | osd_req_op_extent_osd_data_bio(osd_req, 1, | |
2403 | orig_request->bio_list, length); | |
2404 | else | |
2405 | osd_req_op_extent_osd_data_pages(osd_req, 1, | |
2406 | orig_request->pages, length, | |
2407 | offset & ~PAGE_MASK, false, false); | |
0eefd470 AE |
2408 | |
2409 | rbd_osd_req_format_write(orig_request); | |
2410 | ||
2411 | /* All set, send it off. */ | |
2412 | ||
2413 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2414 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2415 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2416 | if (!img_result) | |
0eefd470 AE |
2417 | return; |
2418 | out_err: | |
2419 | /* Record the error code and complete the request */ | |
2420 | ||
bbea1c1a | 2421 | orig_request->result = img_result; |
0eefd470 AE |
2422 | orig_request->xferred = 0; |
2423 | obj_request_done_set(orig_request); | |
2424 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2425 | } |
2426 | ||
2427 | /* | |
2428 | * Read from the parent image the range of data that covers the | |
2429 | * entire target of the given object request. This is used for | |
2430 | * satisfying a layered image write request when the target of an | |
2431 | * object request from the image request does not exist. | |
2432 | * | |
2433 | * A page array big enough to hold the returned data is allocated | |
2434 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2435 | * When the read completes, this page array will be transferred to | |
2436 | * the original object request for the copyup operation. | |
2437 | * | |
2438 | * If an error occurs, record it as the result of the original | |
2439 | * object request and mark it done so it gets completed. | |
2440 | */ | |
2441 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2442 | { | |
2443 | struct rbd_img_request *img_request = NULL; | |
2444 | struct rbd_img_request *parent_request = NULL; | |
2445 | struct rbd_device *rbd_dev; | |
2446 | u64 img_offset; | |
2447 | u64 length; | |
2448 | struct page **pages = NULL; | |
2449 | u32 page_count; | |
2450 | int result; | |
2451 | ||
2452 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2453 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2454 | |
2455 | img_request = obj_request->img_request; | |
2456 | rbd_assert(img_request != NULL); | |
2457 | rbd_dev = img_request->rbd_dev; | |
2458 | rbd_assert(rbd_dev->parent != NULL); | |
2459 | ||
2460 | /* | |
2461 | * Determine the byte range covered by the object in the | |
2462 | * child image to which the original request was to be sent. | |
2463 | */ | |
2464 | img_offset = obj_request->img_offset - obj_request->offset; | |
2465 | length = (u64)1 << rbd_dev->header.obj_order; | |
2466 | ||
a9e8ba2c AE |
2467 | /* |
2468 | * There is no defined parent data beyond the parent | |
2469 | * overlap, so limit what we read at that boundary if | |
2470 | * necessary. | |
2471 | */ | |
2472 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2473 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2474 | length = rbd_dev->parent_overlap - img_offset; | |
2475 | } | |
2476 | ||
3d7efd18 AE |
2477 | /* |
2478 | * Allocate a page array big enough to receive the data read | |
2479 | * from the parent. | |
2480 | */ | |
2481 | page_count = (u32)calc_pages_for(0, length); | |
2482 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2483 | if (IS_ERR(pages)) { | |
2484 | result = PTR_ERR(pages); | |
2485 | pages = NULL; | |
2486 | goto out_err; | |
2487 | } | |
2488 | ||
2489 | result = -ENOMEM; | |
e93f3152 AE |
2490 | parent_request = rbd_parent_request_create(obj_request, |
2491 | img_offset, length); | |
3d7efd18 AE |
2492 | if (!parent_request) |
2493 | goto out_err; | |
3d7efd18 AE |
2494 | |
2495 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2496 | if (result) | |
2497 | goto out_err; | |
2498 | parent_request->copyup_pages = pages; | |
ebda6408 | 2499 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2500 | |
2501 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2502 | result = rbd_img_request_submit(parent_request); | |
2503 | if (!result) | |
2504 | return 0; | |
2505 | ||
2506 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2507 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2508 | parent_request->obj_request = NULL; |
2509 | rbd_obj_request_put(obj_request); | |
2510 | out_err: | |
2511 | if (pages) | |
2512 | ceph_release_page_vector(pages, page_count); | |
2513 | if (parent_request) | |
2514 | rbd_img_request_put(parent_request); | |
2515 | obj_request->result = result; | |
2516 | obj_request->xferred = 0; | |
2517 | obj_request_done_set(obj_request); | |
2518 | ||
2519 | return result; | |
2520 | } | |
2521 | ||
c5b5ef6c AE |
2522 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2523 | { | |
c5b5ef6c | 2524 | struct rbd_obj_request *orig_request; |
638f5abe | 2525 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2526 | int result; |
2527 | ||
2528 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2529 | ||
2530 | /* | |
2531 | * All we need from the object request is the original | |
2532 | * request and the result of the STAT op. Grab those, then | |
2533 | * we're done with the request. | |
2534 | */ | |
2535 | orig_request = obj_request->obj_request; | |
2536 | obj_request->obj_request = NULL; | |
912c317d | 2537 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2538 | rbd_assert(orig_request); |
2539 | rbd_assert(orig_request->img_request); | |
2540 | ||
2541 | result = obj_request->result; | |
2542 | obj_request->result = 0; | |
2543 | ||
2544 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2545 | obj_request, orig_request, result, | |
2546 | obj_request->xferred, obj_request->length); | |
2547 | rbd_obj_request_put(obj_request); | |
2548 | ||
638f5abe AE |
2549 | /* |
2550 | * If the overlap has become 0 (most likely because the | |
2551 | * image has been flattened) we need to free the pages | |
2552 | * and re-submit the original write request. | |
2553 | */ | |
2554 | rbd_dev = orig_request->img_request->rbd_dev; | |
2555 | if (!rbd_dev->parent_overlap) { | |
2556 | struct ceph_osd_client *osdc; | |
2557 | ||
638f5abe AE |
2558 | osdc = &rbd_dev->rbd_client->client->osdc; |
2559 | result = rbd_obj_request_submit(osdc, orig_request); | |
2560 | if (!result) | |
2561 | return; | |
2562 | } | |
c5b5ef6c AE |
2563 | |
2564 | /* | |
2565 | * Our only purpose here is to determine whether the object | |
2566 | * exists, and we don't want to treat the non-existence as | |
2567 | * an error. If something else comes back, transfer the | |
2568 | * error to the original request and complete it now. | |
2569 | */ | |
2570 | if (!result) { | |
2571 | obj_request_existence_set(orig_request, true); | |
2572 | } else if (result == -ENOENT) { | |
2573 | obj_request_existence_set(orig_request, false); | |
2574 | } else if (result) { | |
2575 | orig_request->result = result; | |
3d7efd18 | 2576 | goto out; |
c5b5ef6c AE |
2577 | } |
2578 | ||
2579 | /* | |
2580 | * Resubmit the original request now that we have recorded | |
2581 | * whether the target object exists. | |
2582 | */ | |
b454e36d | 2583 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2584 | out: |
c5b5ef6c AE |
2585 | if (orig_request->result) |
2586 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2587 | } |
2588 | ||
2589 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2590 | { | |
2591 | struct rbd_obj_request *stat_request; | |
2592 | struct rbd_device *rbd_dev; | |
2593 | struct ceph_osd_client *osdc; | |
2594 | struct page **pages = NULL; | |
2595 | u32 page_count; | |
2596 | size_t size; | |
2597 | int ret; | |
2598 | ||
2599 | /* | |
2600 | * The response data for a STAT call consists of: | |
2601 | * le64 length; | |
2602 | * struct { | |
2603 | * le32 tv_sec; | |
2604 | * le32 tv_nsec; | |
2605 | * } mtime; | |
2606 | */ | |
2607 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2608 | page_count = (u32)calc_pages_for(0, size); | |
2609 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2610 | if (IS_ERR(pages)) | |
2611 | return PTR_ERR(pages); | |
2612 | ||
2613 | ret = -ENOMEM; | |
2614 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2615 | OBJ_REQUEST_PAGES); | |
2616 | if (!stat_request) | |
2617 | goto out; | |
2618 | ||
2619 | rbd_obj_request_get(obj_request); | |
2620 | stat_request->obj_request = obj_request; | |
2621 | stat_request->pages = pages; | |
2622 | stat_request->page_count = page_count; | |
2623 | ||
2624 | rbd_assert(obj_request->img_request); | |
2625 | rbd_dev = obj_request->img_request->rbd_dev; | |
2626 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2627 | stat_request); | |
2628 | if (!stat_request->osd_req) | |
2629 | goto out; | |
2630 | stat_request->callback = rbd_img_obj_exists_callback; | |
2631 | ||
2632 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2633 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2634 | false, false); | |
9d4df01f | 2635 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2636 | |
2637 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2638 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2639 | out: | |
2640 | if (ret) | |
2641 | rbd_obj_request_put(obj_request); | |
2642 | ||
2643 | return ret; | |
2644 | } | |
2645 | ||
b454e36d AE |
2646 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2647 | { | |
2648 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2649 | struct rbd_device *rbd_dev; |
3d7efd18 | 2650 | bool known; |
b454e36d AE |
2651 | |
2652 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2653 | ||
2654 | img_request = obj_request->img_request; | |
2655 | rbd_assert(img_request); | |
a9e8ba2c | 2656 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2657 | |
b454e36d | 2658 | /* |
a9e8ba2c AE |
2659 | * Only writes to layered images need special handling. |
2660 | * Reads and non-layered writes are simple object requests. | |
2661 | * Layered writes that start beyond the end of the overlap | |
2662 | * with the parent have no parent data, so they too are | |
2663 | * simple object requests. Finally, if the target object is | |
2664 | * known to already exist, its parent data has already been | |
2665 | * copied, so a write to the object can also be handled as a | |
2666 | * simple object request. | |
b454e36d AE |
2667 | */ |
2668 | if (!img_request_write_test(img_request) || | |
2669 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2670 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2671 | ((known = obj_request_known_test(obj_request)) && |
2672 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2673 | |
2674 | struct rbd_device *rbd_dev; | |
2675 | struct ceph_osd_client *osdc; | |
2676 | ||
2677 | rbd_dev = obj_request->img_request->rbd_dev; | |
2678 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2679 | ||
2680 | return rbd_obj_request_submit(osdc, obj_request); | |
2681 | } | |
2682 | ||
2683 | /* | |
3d7efd18 AE |
2684 | * It's a layered write. The target object might exist but |
2685 | * we may not know that yet. If we know it doesn't exist, | |
2686 | * start by reading the data for the full target object from | |
2687 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2688 | */ |
3d7efd18 AE |
2689 | if (known) |
2690 | return rbd_img_obj_parent_read_full(obj_request); | |
2691 | ||
2692 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2693 | |
2694 | return rbd_img_obj_exists_submit(obj_request); | |
2695 | } | |
2696 | ||
bf0d5f50 AE |
2697 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2698 | { | |
bf0d5f50 | 2699 | struct rbd_obj_request *obj_request; |
46faeed4 | 2700 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2701 | |
37206ee5 | 2702 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2703 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2704 | int ret; |
2705 | ||
b454e36d | 2706 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2707 | if (ret) |
2708 | return ret; | |
bf0d5f50 AE |
2709 | } |
2710 | ||
2711 | return 0; | |
2712 | } | |
8b3e1a56 AE |
2713 | |
2714 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2715 | { | |
2716 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2717 | struct rbd_device *rbd_dev; |
2718 | u64 obj_end; | |
02c74fba AE |
2719 | u64 img_xferred; |
2720 | int img_result; | |
8b3e1a56 AE |
2721 | |
2722 | rbd_assert(img_request_child_test(img_request)); | |
2723 | ||
02c74fba AE |
2724 | /* First get what we need from the image request and release it */ |
2725 | ||
8b3e1a56 | 2726 | obj_request = img_request->obj_request; |
02c74fba AE |
2727 | img_xferred = img_request->xferred; |
2728 | img_result = img_request->result; | |
2729 | rbd_img_request_put(img_request); | |
2730 | ||
2731 | /* | |
2732 | * If the overlap has become 0 (most likely because the | |
2733 | * image has been flattened) we need to re-submit the | |
2734 | * original request. | |
2735 | */ | |
a9e8ba2c AE |
2736 | rbd_assert(obj_request); |
2737 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2738 | rbd_dev = obj_request->img_request->rbd_dev; |
2739 | if (!rbd_dev->parent_overlap) { | |
2740 | struct ceph_osd_client *osdc; | |
2741 | ||
2742 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2743 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2744 | if (!img_result) | |
2745 | return; | |
2746 | } | |
a9e8ba2c | 2747 | |
02c74fba | 2748 | obj_request->result = img_result; |
a9e8ba2c AE |
2749 | if (obj_request->result) |
2750 | goto out; | |
2751 | ||
2752 | /* | |
2753 | * We need to zero anything beyond the parent overlap | |
2754 | * boundary. Since rbd_img_obj_request_read_callback() | |
2755 | * will zero anything beyond the end of a short read, an | |
2756 | * easy way to do this is to pretend the data from the | |
2757 | * parent came up short--ending at the overlap boundary. | |
2758 | */ | |
2759 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2760 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2761 | if (obj_end > rbd_dev->parent_overlap) { |
2762 | u64 xferred = 0; | |
2763 | ||
2764 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2765 | xferred = rbd_dev->parent_overlap - | |
2766 | obj_request->img_offset; | |
8b3e1a56 | 2767 | |
02c74fba | 2768 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2769 | } else { |
02c74fba | 2770 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2771 | } |
2772 | out: | |
8b3e1a56 AE |
2773 | rbd_img_obj_request_read_callback(obj_request); |
2774 | rbd_obj_request_complete(obj_request); | |
2775 | } | |
2776 | ||
2777 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2778 | { | |
8b3e1a56 AE |
2779 | struct rbd_img_request *img_request; |
2780 | int result; | |
2781 | ||
2782 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2783 | rbd_assert(obj_request->img_request != NULL); | |
2784 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 2785 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 2786 | |
8b3e1a56 | 2787 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 2788 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 2789 | obj_request->img_offset, |
e93f3152 | 2790 | obj_request->length); |
8b3e1a56 AE |
2791 | result = -ENOMEM; |
2792 | if (!img_request) | |
2793 | goto out_err; | |
2794 | ||
5b2ab72d AE |
2795 | if (obj_request->type == OBJ_REQUEST_BIO) |
2796 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
2797 | obj_request->bio_list); | |
2798 | else | |
2799 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
2800 | obj_request->pages); | |
8b3e1a56 AE |
2801 | if (result) |
2802 | goto out_err; | |
2803 | ||
2804 | img_request->callback = rbd_img_parent_read_callback; | |
2805 | result = rbd_img_request_submit(img_request); | |
2806 | if (result) | |
2807 | goto out_err; | |
2808 | ||
2809 | return; | |
2810 | out_err: | |
2811 | if (img_request) | |
2812 | rbd_img_request_put(img_request); | |
2813 | obj_request->result = result; | |
2814 | obj_request->xferred = 0; | |
2815 | obj_request_done_set(obj_request); | |
2816 | } | |
bf0d5f50 | 2817 | |
20e0af67 | 2818 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2819 | { |
2820 | struct rbd_obj_request *obj_request; | |
2169238d | 2821 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2822 | int ret; |
2823 | ||
2824 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2825 | OBJ_REQUEST_NODATA); | |
2826 | if (!obj_request) | |
2827 | return -ENOMEM; | |
2828 | ||
2829 | ret = -ENOMEM; | |
430c28c3 | 2830 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2831 | if (!obj_request->osd_req) |
2832 | goto out; | |
2833 | ||
c99d2d4a | 2834 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2835 | notify_id, 0, 0); |
9d4df01f | 2836 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2837 | |
b8d70035 | 2838 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 2839 | if (ret) |
20e0af67 JD |
2840 | goto out; |
2841 | ret = rbd_obj_request_wait(obj_request); | |
2842 | out: | |
2843 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2844 | |
2845 | return ret; | |
2846 | } | |
2847 | ||
2848 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2849 | { | |
2850 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 2851 | int ret; |
b8d70035 AE |
2852 | |
2853 | if (!rbd_dev) | |
2854 | return; | |
2855 | ||
37206ee5 | 2856 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2857 | rbd_dev->header_name, (unsigned long long)notify_id, |
2858 | (unsigned int)opcode); | |
e627db08 AE |
2859 | ret = rbd_dev_refresh(rbd_dev); |
2860 | if (ret) | |
3b5cf2a2 | 2861 | rbd_warn(rbd_dev, "header refresh error (%d)\n", ret); |
b8d70035 | 2862 | |
20e0af67 | 2863 | rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
b8d70035 AE |
2864 | } |
2865 | ||
9969ebc5 AE |
2866 | /* |
2867 | * Request sync osd watch/unwatch. The value of "start" determines | |
2868 | * whether a watch request is being initiated or torn down. | |
2869 | */ | |
1f3ef788 | 2870 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start) |
9969ebc5 AE |
2871 | { |
2872 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2873 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2874 | int ret; |
2875 | ||
2876 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2877 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2878 | ||
2879 | if (start) { | |
3c663bbd | 2880 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2881 | &rbd_dev->watch_event); |
2882 | if (ret < 0) | |
2883 | return ret; | |
8eb87565 | 2884 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2885 | } |
2886 | ||
2887 | ret = -ENOMEM; | |
2888 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2889 | OBJ_REQUEST_NODATA); | |
2890 | if (!obj_request) | |
2891 | goto out_cancel; | |
2892 | ||
430c28c3 AE |
2893 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2894 | if (!obj_request->osd_req) | |
2895 | goto out_cancel; | |
2896 | ||
8eb87565 | 2897 | if (start) |
975241af | 2898 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2899 | else |
6977c3f9 | 2900 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2901 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2902 | |
2903 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
1f3ef788 | 2904 | rbd_dev->watch_event->cookie, 0, start ? 1 : 0); |
9d4df01f | 2905 | rbd_osd_req_format_write(obj_request); |
2169238d | 2906 | |
9969ebc5 AE |
2907 | ret = rbd_obj_request_submit(osdc, obj_request); |
2908 | if (ret) | |
2909 | goto out_cancel; | |
2910 | ret = rbd_obj_request_wait(obj_request); | |
2911 | if (ret) | |
2912 | goto out_cancel; | |
9969ebc5 AE |
2913 | ret = obj_request->result; |
2914 | if (ret) | |
2915 | goto out_cancel; | |
2916 | ||
8eb87565 AE |
2917 | /* |
2918 | * A watch request is set to linger, so the underlying osd | |
2919 | * request won't go away until we unregister it. We retain | |
2920 | * a pointer to the object request during that time (in | |
2921 | * rbd_dev->watch_request), so we'll keep a reference to | |
2922 | * it. We'll drop that reference (below) after we've | |
2923 | * unregistered it. | |
2924 | */ | |
2925 | if (start) { | |
2926 | rbd_dev->watch_request = obj_request; | |
2927 | ||
2928 | return 0; | |
2929 | } | |
2930 | ||
2931 | /* We have successfully torn down the watch request */ | |
2932 | ||
2933 | rbd_obj_request_put(rbd_dev->watch_request); | |
2934 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2935 | out_cancel: |
2936 | /* Cancel the event if we're tearing down, or on error */ | |
2937 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2938 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2939 | if (obj_request) |
2940 | rbd_obj_request_put(obj_request); | |
2941 | ||
2942 | return ret; | |
2943 | } | |
2944 | ||
36be9a76 | 2945 | /* |
f40eb349 AE |
2946 | * Synchronous osd object method call. Returns the number of bytes |
2947 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2948 | */ |
2949 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2950 | const char *object_name, | |
2951 | const char *class_name, | |
2952 | const char *method_name, | |
4157976b | 2953 | const void *outbound, |
36be9a76 | 2954 | size_t outbound_size, |
4157976b | 2955 | void *inbound, |
e2a58ee5 | 2956 | size_t inbound_size) |
36be9a76 | 2957 | { |
2169238d | 2958 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2959 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2960 | struct page **pages; |
2961 | u32 page_count; | |
2962 | int ret; | |
2963 | ||
2964 | /* | |
6010a451 AE |
2965 | * Method calls are ultimately read operations. The result |
2966 | * should placed into the inbound buffer provided. They | |
2967 | * also supply outbound data--parameters for the object | |
2968 | * method. Currently if this is present it will be a | |
2969 | * snapshot id. | |
36be9a76 | 2970 | */ |
57385b51 | 2971 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2972 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2973 | if (IS_ERR(pages)) | |
2974 | return PTR_ERR(pages); | |
2975 | ||
2976 | ret = -ENOMEM; | |
6010a451 | 2977 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2978 | OBJ_REQUEST_PAGES); |
2979 | if (!obj_request) | |
2980 | goto out; | |
2981 | ||
2982 | obj_request->pages = pages; | |
2983 | obj_request->page_count = page_count; | |
2984 | ||
430c28c3 | 2985 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2986 | if (!obj_request->osd_req) |
2987 | goto out; | |
2988 | ||
c99d2d4a | 2989 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2990 | class_name, method_name); |
2991 | if (outbound_size) { | |
2992 | struct ceph_pagelist *pagelist; | |
2993 | ||
2994 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2995 | if (!pagelist) | |
2996 | goto out; | |
2997 | ||
2998 | ceph_pagelist_init(pagelist); | |
2999 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3000 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3001 | pagelist); | |
3002 | } | |
a4ce40a9 AE |
3003 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3004 | obj_request->pages, inbound_size, | |
44cd188d | 3005 | 0, false, false); |
9d4df01f | 3006 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3007 | |
36be9a76 AE |
3008 | ret = rbd_obj_request_submit(osdc, obj_request); |
3009 | if (ret) | |
3010 | goto out; | |
3011 | ret = rbd_obj_request_wait(obj_request); | |
3012 | if (ret) | |
3013 | goto out; | |
3014 | ||
3015 | ret = obj_request->result; | |
3016 | if (ret < 0) | |
3017 | goto out; | |
57385b51 AE |
3018 | |
3019 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3020 | ret = (int)obj_request->xferred; | |
903bb32e | 3021 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3022 | out: |
3023 | if (obj_request) | |
3024 | rbd_obj_request_put(obj_request); | |
3025 | else | |
3026 | ceph_release_page_vector(pages, page_count); | |
3027 | ||
3028 | return ret; | |
3029 | } | |
3030 | ||
bf0d5f50 | 3031 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 3032 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
3033 | { |
3034 | struct rbd_device *rbd_dev = q->queuedata; | |
3035 | bool read_only = rbd_dev->mapping.read_only; | |
3036 | struct request *rq; | |
3037 | int result; | |
3038 | ||
3039 | while ((rq = blk_fetch_request(q))) { | |
3040 | bool write_request = rq_data_dir(rq) == WRITE; | |
3041 | struct rbd_img_request *img_request; | |
3042 | u64 offset; | |
3043 | u64 length; | |
3044 | ||
3045 | /* Ignore any non-FS requests that filter through. */ | |
3046 | ||
3047 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
3048 | dout("%s: non-fs request type %d\n", __func__, |
3049 | (int) rq->cmd_type); | |
3050 | __blk_end_request_all(rq, 0); | |
3051 | continue; | |
3052 | } | |
3053 | ||
3054 | /* Ignore/skip any zero-length requests */ | |
3055 | ||
3056 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
3057 | length = (u64) blk_rq_bytes(rq); | |
3058 | ||
3059 | if (!length) { | |
3060 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
3061 | __blk_end_request_all(rq, 0); |
3062 | continue; | |
3063 | } | |
3064 | ||
3065 | spin_unlock_irq(q->queue_lock); | |
3066 | ||
3067 | /* Disallow writes to a read-only device */ | |
3068 | ||
3069 | if (write_request) { | |
3070 | result = -EROFS; | |
3071 | if (read_only) | |
3072 | goto end_request; | |
3073 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
3074 | } | |
3075 | ||
6d292906 AE |
3076 | /* |
3077 | * Quit early if the mapped snapshot no longer | |
3078 | * exists. It's still possible the snapshot will | |
3079 | * have disappeared by the time our request arrives | |
3080 | * at the osd, but there's no sense in sending it if | |
3081 | * we already know. | |
3082 | */ | |
3083 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
3084 | dout("request for non-existent snapshot"); |
3085 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3086 | result = -ENXIO; | |
3087 | goto end_request; | |
3088 | } | |
3089 | ||
bf0d5f50 | 3090 | result = -EINVAL; |
c0cd10db AE |
3091 | if (offset && length > U64_MAX - offset + 1) { |
3092 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
3093 | offset, length); | |
bf0d5f50 | 3094 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 3095 | } |
bf0d5f50 | 3096 | |
00a653e2 AE |
3097 | result = -EIO; |
3098 | if (offset + length > rbd_dev->mapping.size) { | |
3099 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n", | |
3100 | offset, length, rbd_dev->mapping.size); | |
3101 | goto end_request; | |
3102 | } | |
3103 | ||
bf0d5f50 AE |
3104 | result = -ENOMEM; |
3105 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
e93f3152 | 3106 | write_request); |
bf0d5f50 AE |
3107 | if (!img_request) |
3108 | goto end_request; | |
3109 | ||
3110 | img_request->rq = rq; | |
3111 | ||
f1a4739f AE |
3112 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
3113 | rq->bio); | |
bf0d5f50 AE |
3114 | if (!result) |
3115 | result = rbd_img_request_submit(img_request); | |
3116 | if (result) | |
3117 | rbd_img_request_put(img_request); | |
3118 | end_request: | |
3119 | spin_lock_irq(q->queue_lock); | |
3120 | if (result < 0) { | |
7da22d29 AE |
3121 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
3122 | write_request ? "write" : "read", | |
3123 | length, offset, result); | |
3124 | ||
bf0d5f50 AE |
3125 | __blk_end_request_all(rq, result); |
3126 | } | |
3127 | } | |
3128 | } | |
3129 | ||
602adf40 YS |
3130 | /* |
3131 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3132 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3133 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3134 | */ |
3135 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3136 | struct bio_vec *bvec) | |
3137 | { | |
3138 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3139 | sector_t sector_offset; |
3140 | sector_t sectors_per_obj; | |
3141 | sector_t obj_sector_offset; | |
3142 | int ret; | |
3143 | ||
3144 | /* | |
3145 | * Find how far into its rbd object the partition-relative | |
3146 | * bio start sector is to offset relative to the enclosing | |
3147 | * device. | |
3148 | */ | |
3149 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3150 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3151 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3152 | ||
3153 | /* | |
3154 | * Compute the number of bytes from that offset to the end | |
3155 | * of the object. Account for what's already used by the bio. | |
3156 | */ | |
3157 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3158 | if (ret > bmd->bi_size) | |
3159 | ret -= bmd->bi_size; | |
3160 | else | |
3161 | ret = 0; | |
3162 | ||
3163 | /* | |
3164 | * Don't send back more than was asked for. And if the bio | |
3165 | * was empty, let the whole thing through because: "Note | |
3166 | * that a block device *must* allow a single page to be | |
3167 | * added to an empty bio." | |
3168 | */ | |
3169 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3170 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3171 | ret = (int) bvec->bv_len; | |
3172 | ||
3173 | return ret; | |
602adf40 YS |
3174 | } |
3175 | ||
3176 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3177 | { | |
3178 | struct gendisk *disk = rbd_dev->disk; | |
3179 | ||
3180 | if (!disk) | |
3181 | return; | |
3182 | ||
a0cab924 AE |
3183 | rbd_dev->disk = NULL; |
3184 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3185 | del_gendisk(disk); |
a0cab924 AE |
3186 | if (disk->queue) |
3187 | blk_cleanup_queue(disk->queue); | |
3188 | } | |
602adf40 YS |
3189 | put_disk(disk); |
3190 | } | |
3191 | ||
788e2df3 AE |
3192 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3193 | const char *object_name, | |
7097f8df | 3194 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3195 | |
3196 | { | |
2169238d | 3197 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3198 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3199 | struct page **pages = NULL; |
3200 | u32 page_count; | |
1ceae7ef | 3201 | size_t size; |
788e2df3 AE |
3202 | int ret; |
3203 | ||
3204 | page_count = (u32) calc_pages_for(offset, length); | |
3205 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3206 | if (IS_ERR(pages)) | |
3207 | ret = PTR_ERR(pages); | |
3208 | ||
3209 | ret = -ENOMEM; | |
3210 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3211 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3212 | if (!obj_request) |
3213 | goto out; | |
3214 | ||
3215 | obj_request->pages = pages; | |
3216 | obj_request->page_count = page_count; | |
3217 | ||
430c28c3 | 3218 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
3219 | if (!obj_request->osd_req) |
3220 | goto out; | |
3221 | ||
c99d2d4a AE |
3222 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3223 | offset, length, 0, 0); | |
406e2c9f | 3224 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3225 | obj_request->pages, |
44cd188d AE |
3226 | obj_request->length, |
3227 | obj_request->offset & ~PAGE_MASK, | |
3228 | false, false); | |
9d4df01f | 3229 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3230 | |
788e2df3 AE |
3231 | ret = rbd_obj_request_submit(osdc, obj_request); |
3232 | if (ret) | |
3233 | goto out; | |
3234 | ret = rbd_obj_request_wait(obj_request); | |
3235 | if (ret) | |
3236 | goto out; | |
3237 | ||
3238 | ret = obj_request->result; | |
3239 | if (ret < 0) | |
3240 | goto out; | |
1ceae7ef AE |
3241 | |
3242 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3243 | size = (size_t) obj_request->xferred; | |
903bb32e | 3244 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3245 | rbd_assert(size <= (size_t)INT_MAX); |
3246 | ret = (int)size; | |
788e2df3 AE |
3247 | out: |
3248 | if (obj_request) | |
3249 | rbd_obj_request_put(obj_request); | |
3250 | else | |
3251 | ceph_release_page_vector(pages, page_count); | |
3252 | ||
3253 | return ret; | |
3254 | } | |
3255 | ||
602adf40 | 3256 | /* |
662518b1 AE |
3257 | * Read the complete header for the given rbd device. On successful |
3258 | * return, the rbd_dev->header field will contain up-to-date | |
3259 | * information about the image. | |
602adf40 | 3260 | */ |
99a41ebc | 3261 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3262 | { |
4156d998 | 3263 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3264 | u32 snap_count = 0; |
4156d998 AE |
3265 | u64 names_size = 0; |
3266 | u32 want_count; | |
3267 | int ret; | |
602adf40 | 3268 | |
00f1f36f | 3269 | /* |
4156d998 AE |
3270 | * The complete header will include an array of its 64-bit |
3271 | * snapshot ids, followed by the names of those snapshots as | |
3272 | * a contiguous block of NUL-terminated strings. Note that | |
3273 | * the number of snapshots could change by the time we read | |
3274 | * it in, in which case we re-read it. | |
00f1f36f | 3275 | */ |
4156d998 AE |
3276 | do { |
3277 | size_t size; | |
3278 | ||
3279 | kfree(ondisk); | |
3280 | ||
3281 | size = sizeof (*ondisk); | |
3282 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3283 | size += names_size; | |
3284 | ondisk = kmalloc(size, GFP_KERNEL); | |
3285 | if (!ondisk) | |
662518b1 | 3286 | return -ENOMEM; |
4156d998 | 3287 | |
788e2df3 | 3288 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3289 | 0, size, ondisk); |
4156d998 | 3290 | if (ret < 0) |
662518b1 | 3291 | goto out; |
c0cd10db | 3292 | if ((size_t)ret < size) { |
4156d998 | 3293 | ret = -ENXIO; |
06ecc6cb AE |
3294 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3295 | size, ret); | |
662518b1 | 3296 | goto out; |
4156d998 AE |
3297 | } |
3298 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3299 | ret = -ENXIO; | |
06ecc6cb | 3300 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3301 | goto out; |
81e759fb | 3302 | } |
602adf40 | 3303 | |
4156d998 AE |
3304 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3305 | want_count = snap_count; | |
3306 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3307 | } while (snap_count != want_count); | |
00f1f36f | 3308 | |
662518b1 AE |
3309 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3310 | out: | |
4156d998 AE |
3311 | kfree(ondisk); |
3312 | ||
3313 | return ret; | |
602adf40 YS |
3314 | } |
3315 | ||
15228ede AE |
3316 | /* |
3317 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3318 | * has disappeared from the (just updated) snapshot context. | |
3319 | */ | |
3320 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3321 | { | |
3322 | u64 snap_id; | |
3323 | ||
3324 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3325 | return; | |
3326 | ||
3327 | snap_id = rbd_dev->spec->snap_id; | |
3328 | if (snap_id == CEPH_NOSNAP) | |
3329 | return; | |
3330 | ||
3331 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3332 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3333 | } | |
3334 | ||
9875201e JD |
3335 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3336 | { | |
3337 | sector_t size; | |
3338 | bool removing; | |
3339 | ||
3340 | /* | |
3341 | * Don't hold the lock while doing disk operations, | |
3342 | * or lock ordering will conflict with the bdev mutex via: | |
3343 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3344 | */ | |
3345 | spin_lock_irq(&rbd_dev->lock); | |
3346 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3347 | spin_unlock_irq(&rbd_dev->lock); | |
3348 | /* | |
3349 | * If the device is being removed, rbd_dev->disk has | |
3350 | * been destroyed, so don't try to update its size | |
3351 | */ | |
3352 | if (!removing) { | |
3353 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3354 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3355 | set_capacity(rbd_dev->disk, size); | |
3356 | revalidate_disk(rbd_dev->disk); | |
3357 | } | |
3358 | } | |
3359 | ||
cc4a38bd | 3360 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3361 | { |
e627db08 | 3362 | u64 mapping_size; |
1fe5e993 AE |
3363 | int ret; |
3364 | ||
117973fb | 3365 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
cfbf6377 | 3366 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3367 | mapping_size = rbd_dev->mapping.size; |
117973fb | 3368 | if (rbd_dev->image_format == 1) |
99a41ebc | 3369 | ret = rbd_dev_v1_header_info(rbd_dev); |
117973fb | 3370 | else |
2df3fac7 | 3371 | ret = rbd_dev_v2_header_info(rbd_dev); |
15228ede AE |
3372 | |
3373 | /* If it's a mapped snapshot, validate its EXISTS flag */ | |
3374 | ||
3375 | rbd_exists_validate(rbd_dev); | |
cfbf6377 AE |
3376 | up_write(&rbd_dev->header_rwsem); |
3377 | ||
00a653e2 | 3378 | if (mapping_size != rbd_dev->mapping.size) { |
9875201e | 3379 | rbd_dev_update_size(rbd_dev); |
00a653e2 | 3380 | } |
1fe5e993 AE |
3381 | |
3382 | return ret; | |
3383 | } | |
3384 | ||
602adf40 YS |
3385 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3386 | { | |
3387 | struct gendisk *disk; | |
3388 | struct request_queue *q; | |
593a9e7b | 3389 | u64 segment_size; |
602adf40 | 3390 | |
602adf40 | 3391 | /* create gendisk info */ |
602adf40 YS |
3392 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
3393 | if (!disk) | |
1fcdb8aa | 3394 | return -ENOMEM; |
602adf40 | 3395 | |
f0f8cef5 | 3396 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3397 | rbd_dev->dev_id); |
602adf40 YS |
3398 | disk->major = rbd_dev->major; |
3399 | disk->first_minor = 0; | |
3400 | disk->fops = &rbd_bd_ops; | |
3401 | disk->private_data = rbd_dev; | |
3402 | ||
bf0d5f50 | 3403 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3404 | if (!q) |
3405 | goto out_disk; | |
029bcbd8 | 3406 | |
593a9e7b AE |
3407 | /* We use the default size, but let's be explicit about it. */ |
3408 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3409 | ||
029bcbd8 | 3410 | /* set io sizes to object size */ |
593a9e7b AE |
3411 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3412 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3413 | blk_queue_max_segment_size(q, segment_size); | |
3414 | blk_queue_io_min(q, segment_size); | |
3415 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3416 | |
602adf40 YS |
3417 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3418 | disk->queue = q; | |
3419 | ||
3420 | q->queuedata = rbd_dev; | |
3421 | ||
3422 | rbd_dev->disk = disk; | |
602adf40 | 3423 | |
602adf40 | 3424 | return 0; |
602adf40 YS |
3425 | out_disk: |
3426 | put_disk(disk); | |
1fcdb8aa AE |
3427 | |
3428 | return -ENOMEM; | |
602adf40 YS |
3429 | } |
3430 | ||
dfc5606d YS |
3431 | /* |
3432 | sysfs | |
3433 | */ | |
3434 | ||
593a9e7b AE |
3435 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3436 | { | |
3437 | return container_of(dev, struct rbd_device, dev); | |
3438 | } | |
3439 | ||
dfc5606d YS |
3440 | static ssize_t rbd_size_show(struct device *dev, |
3441 | struct device_attribute *attr, char *buf) | |
3442 | { | |
593a9e7b | 3443 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3444 | |
fc71d833 AE |
3445 | return sprintf(buf, "%llu\n", |
3446 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3447 | } |
3448 | ||
34b13184 AE |
3449 | /* |
3450 | * Note this shows the features for whatever's mapped, which is not | |
3451 | * necessarily the base image. | |
3452 | */ | |
3453 | static ssize_t rbd_features_show(struct device *dev, | |
3454 | struct device_attribute *attr, char *buf) | |
3455 | { | |
3456 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3457 | ||
3458 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3459 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3460 | } |
3461 | ||
dfc5606d YS |
3462 | static ssize_t rbd_major_show(struct device *dev, |
3463 | struct device_attribute *attr, char *buf) | |
3464 | { | |
593a9e7b | 3465 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3466 | |
fc71d833 AE |
3467 | if (rbd_dev->major) |
3468 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3469 | ||
3470 | return sprintf(buf, "(none)\n"); | |
3471 | ||
dfc5606d YS |
3472 | } |
3473 | ||
3474 | static ssize_t rbd_client_id_show(struct device *dev, | |
3475 | struct device_attribute *attr, char *buf) | |
602adf40 | 3476 | { |
593a9e7b | 3477 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3478 | |
1dbb4399 AE |
3479 | return sprintf(buf, "client%lld\n", |
3480 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3481 | } |
3482 | ||
dfc5606d YS |
3483 | static ssize_t rbd_pool_show(struct device *dev, |
3484 | struct device_attribute *attr, char *buf) | |
602adf40 | 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->pool_name); |
dfc5606d YS |
3489 | } |
3490 | ||
9bb2f334 AE |
3491 | static ssize_t rbd_pool_id_show(struct device *dev, |
3492 | struct device_attribute *attr, char *buf) | |
3493 | { | |
3494 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3495 | ||
0d7dbfce | 3496 | return sprintf(buf, "%llu\n", |
fc71d833 | 3497 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3498 | } |
3499 | ||
dfc5606d YS |
3500 | static ssize_t rbd_name_show(struct device *dev, |
3501 | struct device_attribute *attr, char *buf) | |
3502 | { | |
593a9e7b | 3503 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3504 | |
a92ffdf8 AE |
3505 | if (rbd_dev->spec->image_name) |
3506 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3507 | ||
3508 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3509 | } |
3510 | ||
589d30e0 AE |
3511 | static ssize_t rbd_image_id_show(struct device *dev, |
3512 | struct device_attribute *attr, char *buf) | |
3513 | { | |
3514 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3515 | ||
0d7dbfce | 3516 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3517 | } |
3518 | ||
34b13184 AE |
3519 | /* |
3520 | * Shows the name of the currently-mapped snapshot (or | |
3521 | * RBD_SNAP_HEAD_NAME for the base image). | |
3522 | */ | |
dfc5606d YS |
3523 | static ssize_t rbd_snap_show(struct device *dev, |
3524 | struct device_attribute *attr, | |
3525 | char *buf) | |
3526 | { | |
593a9e7b | 3527 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3528 | |
0d7dbfce | 3529 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3530 | } |
3531 | ||
86b00e0d AE |
3532 | /* |
3533 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3534 | * for the parent image. If there is no parent, simply shows | |
3535 | * "(no parent image)". | |
3536 | */ | |
3537 | static ssize_t rbd_parent_show(struct device *dev, | |
3538 | struct device_attribute *attr, | |
3539 | char *buf) | |
3540 | { | |
3541 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3542 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3543 | int count; | |
3544 | char *bufp = buf; | |
3545 | ||
3546 | if (!spec) | |
3547 | return sprintf(buf, "(no parent image)\n"); | |
3548 | ||
3549 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3550 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3551 | if (count < 0) | |
3552 | return count; | |
3553 | bufp += count; | |
3554 | ||
3555 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3556 | spec->image_name ? spec->image_name : "(unknown)"); | |
3557 | if (count < 0) | |
3558 | return count; | |
3559 | bufp += count; | |
3560 | ||
3561 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3562 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3563 | if (count < 0) | |
3564 | return count; | |
3565 | bufp += count; | |
3566 | ||
3567 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3568 | if (count < 0) | |
3569 | return count; | |
3570 | bufp += count; | |
3571 | ||
3572 | return (ssize_t) (bufp - buf); | |
3573 | } | |
3574 | ||
dfc5606d YS |
3575 | static ssize_t rbd_image_refresh(struct device *dev, |
3576 | struct device_attribute *attr, | |
3577 | const char *buf, | |
3578 | size_t size) | |
3579 | { | |
593a9e7b | 3580 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3581 | int ret; |
602adf40 | 3582 | |
cc4a38bd | 3583 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 AE |
3584 | if (ret) |
3585 | rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret); | |
b813623a AE |
3586 | |
3587 | return ret < 0 ? ret : size; | |
dfc5606d | 3588 | } |
602adf40 | 3589 | |
dfc5606d | 3590 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3591 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
3592 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
3593 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
3594 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3595 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3596 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3597 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3598 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3599 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3600 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3601 | |
3602 | static struct attribute *rbd_attrs[] = { | |
3603 | &dev_attr_size.attr, | |
34b13184 | 3604 | &dev_attr_features.attr, |
dfc5606d YS |
3605 | &dev_attr_major.attr, |
3606 | &dev_attr_client_id.attr, | |
3607 | &dev_attr_pool.attr, | |
9bb2f334 | 3608 | &dev_attr_pool_id.attr, |
dfc5606d | 3609 | &dev_attr_name.attr, |
589d30e0 | 3610 | &dev_attr_image_id.attr, |
dfc5606d | 3611 | &dev_attr_current_snap.attr, |
86b00e0d | 3612 | &dev_attr_parent.attr, |
dfc5606d | 3613 | &dev_attr_refresh.attr, |
dfc5606d YS |
3614 | NULL |
3615 | }; | |
3616 | ||
3617 | static struct attribute_group rbd_attr_group = { | |
3618 | .attrs = rbd_attrs, | |
3619 | }; | |
3620 | ||
3621 | static const struct attribute_group *rbd_attr_groups[] = { | |
3622 | &rbd_attr_group, | |
3623 | NULL | |
3624 | }; | |
3625 | ||
3626 | static void rbd_sysfs_dev_release(struct device *dev) | |
3627 | { | |
3628 | } | |
3629 | ||
3630 | static struct device_type rbd_device_type = { | |
3631 | .name = "rbd", | |
3632 | .groups = rbd_attr_groups, | |
3633 | .release = rbd_sysfs_dev_release, | |
3634 | }; | |
3635 | ||
8b8fb99c AE |
3636 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3637 | { | |
3638 | kref_get(&spec->kref); | |
3639 | ||
3640 | return spec; | |
3641 | } | |
3642 | ||
3643 | static void rbd_spec_free(struct kref *kref); | |
3644 | static void rbd_spec_put(struct rbd_spec *spec) | |
3645 | { | |
3646 | if (spec) | |
3647 | kref_put(&spec->kref, rbd_spec_free); | |
3648 | } | |
3649 | ||
3650 | static struct rbd_spec *rbd_spec_alloc(void) | |
3651 | { | |
3652 | struct rbd_spec *spec; | |
3653 | ||
3654 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3655 | if (!spec) | |
3656 | return NULL; | |
3657 | kref_init(&spec->kref); | |
3658 | ||
8b8fb99c AE |
3659 | return spec; |
3660 | } | |
3661 | ||
3662 | static void rbd_spec_free(struct kref *kref) | |
3663 | { | |
3664 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3665 | ||
3666 | kfree(spec->pool_name); | |
3667 | kfree(spec->image_id); | |
3668 | kfree(spec->image_name); | |
3669 | kfree(spec->snap_name); | |
3670 | kfree(spec); | |
3671 | } | |
3672 | ||
cc344fa1 | 3673 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3674 | struct rbd_spec *spec) |
3675 | { | |
3676 | struct rbd_device *rbd_dev; | |
3677 | ||
3678 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3679 | if (!rbd_dev) | |
3680 | return NULL; | |
3681 | ||
3682 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3683 | rbd_dev->flags = 0; |
a2acd00e | 3684 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 3685 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
3686 | init_rwsem(&rbd_dev->header_rwsem); |
3687 | ||
3688 | rbd_dev->spec = spec; | |
3689 | rbd_dev->rbd_client = rbdc; | |
3690 | ||
0903e875 AE |
3691 | /* Initialize the layout used for all rbd requests */ |
3692 | ||
3693 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3694 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3695 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3696 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3697 | ||
c53d5893 AE |
3698 | return rbd_dev; |
3699 | } | |
3700 | ||
3701 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3702 | { | |
c53d5893 AE |
3703 | rbd_put_client(rbd_dev->rbd_client); |
3704 | rbd_spec_put(rbd_dev->spec); | |
3705 | kfree(rbd_dev); | |
3706 | } | |
3707 | ||
9d475de5 AE |
3708 | /* |
3709 | * Get the size and object order for an image snapshot, or if | |
3710 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3711 | * image. | |
3712 | */ | |
3713 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3714 | u8 *order, u64 *snap_size) | |
3715 | { | |
3716 | __le64 snapid = cpu_to_le64(snap_id); | |
3717 | int ret; | |
3718 | struct { | |
3719 | u8 order; | |
3720 | __le64 size; | |
3721 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3722 | ||
36be9a76 | 3723 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3724 | "rbd", "get_size", |
4157976b | 3725 | &snapid, sizeof (snapid), |
e2a58ee5 | 3726 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3727 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3728 | if (ret < 0) |
3729 | return ret; | |
57385b51 AE |
3730 | if (ret < sizeof (size_buf)) |
3731 | return -ERANGE; | |
9d475de5 | 3732 | |
c3545579 | 3733 | if (order) { |
c86f86e9 | 3734 | *order = size_buf.order; |
c3545579 JD |
3735 | dout(" order %u", (unsigned int)*order); |
3736 | } | |
9d475de5 AE |
3737 | *snap_size = le64_to_cpu(size_buf.size); |
3738 | ||
c3545579 JD |
3739 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
3740 | (unsigned long long)snap_id, | |
57385b51 | 3741 | (unsigned long long)*snap_size); |
9d475de5 AE |
3742 | |
3743 | return 0; | |
3744 | } | |
3745 | ||
3746 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3747 | { | |
3748 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3749 | &rbd_dev->header.obj_order, | |
3750 | &rbd_dev->header.image_size); | |
3751 | } | |
3752 | ||
1e130199 AE |
3753 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3754 | { | |
3755 | void *reply_buf; | |
3756 | int ret; | |
3757 | void *p; | |
3758 | ||
3759 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3760 | if (!reply_buf) | |
3761 | return -ENOMEM; | |
3762 | ||
36be9a76 | 3763 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3764 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3765 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3766 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3767 | if (ret < 0) |
3768 | goto out; | |
3769 | ||
3770 | p = reply_buf; | |
3771 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3772 | p + ret, NULL, GFP_NOIO); |
3773 | ret = 0; | |
1e130199 AE |
3774 | |
3775 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3776 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3777 | rbd_dev->header.object_prefix = NULL; | |
3778 | } else { | |
3779 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3780 | } | |
1e130199 AE |
3781 | out: |
3782 | kfree(reply_buf); | |
3783 | ||
3784 | return ret; | |
3785 | } | |
3786 | ||
b1b5402a AE |
3787 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3788 | u64 *snap_features) | |
3789 | { | |
3790 | __le64 snapid = cpu_to_le64(snap_id); | |
3791 | struct { | |
3792 | __le64 features; | |
3793 | __le64 incompat; | |
4157976b | 3794 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3795 | u64 incompat; |
b1b5402a AE |
3796 | int ret; |
3797 | ||
36be9a76 | 3798 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3799 | "rbd", "get_features", |
4157976b | 3800 | &snapid, sizeof (snapid), |
e2a58ee5 | 3801 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3802 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3803 | if (ret < 0) |
3804 | return ret; | |
57385b51 AE |
3805 | if (ret < sizeof (features_buf)) |
3806 | return -ERANGE; | |
d889140c AE |
3807 | |
3808 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3809 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3810 | return -ENXIO; |
d889140c | 3811 | |
b1b5402a AE |
3812 | *snap_features = le64_to_cpu(features_buf.features); |
3813 | ||
3814 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3815 | (unsigned long long)snap_id, |
3816 | (unsigned long long)*snap_features, | |
3817 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3818 | |
3819 | return 0; | |
3820 | } | |
3821 | ||
3822 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3823 | { | |
3824 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3825 | &rbd_dev->header.features); | |
3826 | } | |
3827 | ||
86b00e0d AE |
3828 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3829 | { | |
3830 | struct rbd_spec *parent_spec; | |
3831 | size_t size; | |
3832 | void *reply_buf = NULL; | |
3833 | __le64 snapid; | |
3834 | void *p; | |
3835 | void *end; | |
642a2537 | 3836 | u64 pool_id; |
86b00e0d | 3837 | char *image_id; |
3b5cf2a2 | 3838 | u64 snap_id; |
86b00e0d | 3839 | u64 overlap; |
86b00e0d AE |
3840 | int ret; |
3841 | ||
3842 | parent_spec = rbd_spec_alloc(); | |
3843 | if (!parent_spec) | |
3844 | return -ENOMEM; | |
3845 | ||
3846 | size = sizeof (__le64) + /* pool_id */ | |
3847 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3848 | sizeof (__le64) + /* snap_id */ | |
3849 | sizeof (__le64); /* overlap */ | |
3850 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3851 | if (!reply_buf) { | |
3852 | ret = -ENOMEM; | |
3853 | goto out_err; | |
3854 | } | |
3855 | ||
3856 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3857 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3858 | "rbd", "get_parent", |
4157976b | 3859 | &snapid, sizeof (snapid), |
e2a58ee5 | 3860 | reply_buf, size); |
36be9a76 | 3861 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3862 | if (ret < 0) |
3863 | goto out_err; | |
3864 | ||
86b00e0d | 3865 | p = reply_buf; |
57385b51 AE |
3866 | end = reply_buf + ret; |
3867 | ret = -ERANGE; | |
642a2537 | 3868 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
3869 | if (pool_id == CEPH_NOPOOL) { |
3870 | /* | |
3871 | * Either the parent never existed, or we have | |
3872 | * record of it but the image got flattened so it no | |
3873 | * longer has a parent. When the parent of a | |
3874 | * layered image disappears we immediately set the | |
3875 | * overlap to 0. The effect of this is that all new | |
3876 | * requests will be treated as if the image had no | |
3877 | * parent. | |
3878 | */ | |
3879 | if (rbd_dev->parent_overlap) { | |
3880 | rbd_dev->parent_overlap = 0; | |
3881 | smp_mb(); | |
3882 | rbd_dev_parent_put(rbd_dev); | |
3883 | pr_info("%s: clone image has been flattened\n", | |
3884 | rbd_dev->disk->disk_name); | |
3885 | } | |
3886 | ||
86b00e0d | 3887 | goto out; /* No parent? No problem. */ |
392a9dad | 3888 | } |
86b00e0d | 3889 | |
0903e875 AE |
3890 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3891 | ||
3892 | ret = -EIO; | |
642a2537 | 3893 | if (pool_id > (u64)U32_MAX) { |
c0cd10db | 3894 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", |
642a2537 | 3895 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 3896 | goto out_err; |
c0cd10db | 3897 | } |
0903e875 | 3898 | |
979ed480 | 3899 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3900 | if (IS_ERR(image_id)) { |
3901 | ret = PTR_ERR(image_id); | |
3902 | goto out_err; | |
3903 | } | |
3b5cf2a2 | 3904 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
3905 | ceph_decode_64_safe(&p, end, overlap, out_err); |
3906 | ||
3b5cf2a2 AE |
3907 | /* |
3908 | * The parent won't change (except when the clone is | |
3909 | * flattened, already handled that). So we only need to | |
3910 | * record the parent spec we have not already done so. | |
3911 | */ | |
3912 | if (!rbd_dev->parent_spec) { | |
3913 | parent_spec->pool_id = pool_id; | |
3914 | parent_spec->image_id = image_id; | |
3915 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
3916 | rbd_dev->parent_spec = parent_spec; |
3917 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3b5cf2a2 AE |
3918 | } |
3919 | ||
3920 | /* | |
3921 | * We always update the parent overlap. If it's zero we | |
3922 | * treat it specially. | |
3923 | */ | |
3924 | rbd_dev->parent_overlap = overlap; | |
3925 | smp_mb(); | |
3926 | if (!overlap) { | |
3927 | ||
3928 | /* A null parent_spec indicates it's the initial probe */ | |
3929 | ||
3930 | if (parent_spec) { | |
3931 | /* | |
3932 | * The overlap has become zero, so the clone | |
3933 | * must have been resized down to 0 at some | |
3934 | * point. Treat this the same as a flatten. | |
3935 | */ | |
3936 | rbd_dev_parent_put(rbd_dev); | |
3937 | pr_info("%s: clone image now standalone\n", | |
3938 | rbd_dev->disk->disk_name); | |
3939 | } else { | |
3940 | /* | |
3941 | * For the initial probe, if we find the | |
3942 | * overlap is zero we just pretend there was | |
3943 | * no parent image. | |
3944 | */ | |
3945 | rbd_warn(rbd_dev, "ignoring parent of " | |
3946 | "clone with overlap 0\n"); | |
3947 | } | |
70cf49cf | 3948 | } |
86b00e0d AE |
3949 | out: |
3950 | ret = 0; | |
3951 | out_err: | |
3952 | kfree(reply_buf); | |
3953 | rbd_spec_put(parent_spec); | |
3954 | ||
3955 | return ret; | |
3956 | } | |
3957 | ||
cc070d59 AE |
3958 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3959 | { | |
3960 | struct { | |
3961 | __le64 stripe_unit; | |
3962 | __le64 stripe_count; | |
3963 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3964 | size_t size = sizeof (striping_info_buf); | |
3965 | void *p; | |
3966 | u64 obj_size; | |
3967 | u64 stripe_unit; | |
3968 | u64 stripe_count; | |
3969 | int ret; | |
3970 | ||
3971 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3972 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 3973 | (char *)&striping_info_buf, size); |
cc070d59 AE |
3974 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
3975 | if (ret < 0) | |
3976 | return ret; | |
3977 | if (ret < size) | |
3978 | return -ERANGE; | |
3979 | ||
3980 | /* | |
3981 | * We don't actually support the "fancy striping" feature | |
3982 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3983 | * defaults the behavior is the same as before. So find | |
3984 | * out, and only fail if the image has non-default values. | |
3985 | */ | |
3986 | ret = -EINVAL; | |
3987 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3988 | p = &striping_info_buf; | |
3989 | stripe_unit = ceph_decode_64(&p); | |
3990 | if (stripe_unit != obj_size) { | |
3991 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
3992 | "(got %llu want %llu)", | |
3993 | stripe_unit, obj_size); | |
3994 | return -EINVAL; | |
3995 | } | |
3996 | stripe_count = ceph_decode_64(&p); | |
3997 | if (stripe_count != 1) { | |
3998 | rbd_warn(rbd_dev, "unsupported stripe count " | |
3999 | "(got %llu want 1)", stripe_count); | |
4000 | return -EINVAL; | |
4001 | } | |
500d0c0f AE |
4002 | rbd_dev->header.stripe_unit = stripe_unit; |
4003 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4004 | |
4005 | return 0; | |
4006 | } | |
4007 | ||
9e15b77d AE |
4008 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4009 | { | |
4010 | size_t image_id_size; | |
4011 | char *image_id; | |
4012 | void *p; | |
4013 | void *end; | |
4014 | size_t size; | |
4015 | void *reply_buf = NULL; | |
4016 | size_t len = 0; | |
4017 | char *image_name = NULL; | |
4018 | int ret; | |
4019 | ||
4020 | rbd_assert(!rbd_dev->spec->image_name); | |
4021 | ||
69e7a02f AE |
4022 | len = strlen(rbd_dev->spec->image_id); |
4023 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4024 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4025 | if (!image_id) | |
4026 | return NULL; | |
4027 | ||
4028 | p = image_id; | |
4157976b | 4029 | end = image_id + image_id_size; |
57385b51 | 4030 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4031 | |
4032 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4033 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4034 | if (!reply_buf) | |
4035 | goto out; | |
4036 | ||
36be9a76 | 4037 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4038 | "rbd", "dir_get_name", |
4039 | image_id, image_id_size, | |
e2a58ee5 | 4040 | reply_buf, size); |
9e15b77d AE |
4041 | if (ret < 0) |
4042 | goto out; | |
4043 | p = reply_buf; | |
f40eb349 AE |
4044 | end = reply_buf + ret; |
4045 | ||
9e15b77d AE |
4046 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4047 | if (IS_ERR(image_name)) | |
4048 | image_name = NULL; | |
4049 | else | |
4050 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4051 | out: | |
4052 | kfree(reply_buf); | |
4053 | kfree(image_id); | |
4054 | ||
4055 | return image_name; | |
4056 | } | |
4057 | ||
2ad3d716 AE |
4058 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4059 | { | |
4060 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4061 | const char *snap_name; | |
4062 | u32 which = 0; | |
4063 | ||
4064 | /* Skip over names until we find the one we are looking for */ | |
4065 | ||
4066 | snap_name = rbd_dev->header.snap_names; | |
4067 | while (which < snapc->num_snaps) { | |
4068 | if (!strcmp(name, snap_name)) | |
4069 | return snapc->snaps[which]; | |
4070 | snap_name += strlen(snap_name) + 1; | |
4071 | which++; | |
4072 | } | |
4073 | return CEPH_NOSNAP; | |
4074 | } | |
4075 | ||
4076 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4077 | { | |
4078 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4079 | u32 which; | |
4080 | bool found = false; | |
4081 | u64 snap_id; | |
4082 | ||
4083 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4084 | const char *snap_name; | |
4085 | ||
4086 | snap_id = snapc->snaps[which]; | |
4087 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4088 | if (IS_ERR(snap_name)) { |
4089 | /* ignore no-longer existing snapshots */ | |
4090 | if (PTR_ERR(snap_name) == -ENOENT) | |
4091 | continue; | |
4092 | else | |
4093 | break; | |
4094 | } | |
2ad3d716 AE |
4095 | found = !strcmp(name, snap_name); |
4096 | kfree(snap_name); | |
4097 | } | |
4098 | return found ? snap_id : CEPH_NOSNAP; | |
4099 | } | |
4100 | ||
4101 | /* | |
4102 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4103 | * no snapshot by that name is found, or if an error occurs. | |
4104 | */ | |
4105 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4106 | { | |
4107 | if (rbd_dev->image_format == 1) | |
4108 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4109 | ||
4110 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4111 | } | |
4112 | ||
9e15b77d | 4113 | /* |
2e9f7f1c AE |
4114 | * When an rbd image has a parent image, it is identified by the |
4115 | * pool, image, and snapshot ids (not names). This function fills | |
4116 | * in the names for those ids. (It's OK if we can't figure out the | |
4117 | * name for an image id, but the pool and snapshot ids should always | |
4118 | * exist and have names.) All names in an rbd spec are dynamically | |
4119 | * allocated. | |
e1d4213f AE |
4120 | * |
4121 | * When an image being mapped (not a parent) is probed, we have the | |
4122 | * pool name and pool id, image name and image id, and the snapshot | |
4123 | * name. The only thing we're missing is the snapshot id. | |
9e15b77d | 4124 | */ |
2e9f7f1c | 4125 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 4126 | { |
2e9f7f1c AE |
4127 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4128 | struct rbd_spec *spec = rbd_dev->spec; | |
4129 | const char *pool_name; | |
4130 | const char *image_name; | |
4131 | const char *snap_name; | |
9e15b77d AE |
4132 | int ret; |
4133 | ||
e1d4213f AE |
4134 | /* |
4135 | * An image being mapped will have the pool name (etc.), but | |
4136 | * we need to look up the snapshot id. | |
4137 | */ | |
2e9f7f1c AE |
4138 | if (spec->pool_name) { |
4139 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
2ad3d716 | 4140 | u64 snap_id; |
e1d4213f | 4141 | |
2ad3d716 AE |
4142 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); |
4143 | if (snap_id == CEPH_NOSNAP) | |
e1d4213f | 4144 | return -ENOENT; |
2ad3d716 | 4145 | spec->snap_id = snap_id; |
e1d4213f | 4146 | } else { |
2e9f7f1c | 4147 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
4148 | } |
4149 | ||
4150 | return 0; | |
4151 | } | |
9e15b77d | 4152 | |
2e9f7f1c | 4153 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4154 | |
2e9f7f1c AE |
4155 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4156 | if (!pool_name) { | |
4157 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4158 | return -EIO; |
4159 | } | |
2e9f7f1c AE |
4160 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4161 | if (!pool_name) | |
9e15b77d AE |
4162 | return -ENOMEM; |
4163 | ||
4164 | /* Fetch the image name; tolerate failure here */ | |
4165 | ||
2e9f7f1c AE |
4166 | image_name = rbd_dev_image_name(rbd_dev); |
4167 | if (!image_name) | |
06ecc6cb | 4168 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4169 | |
2e9f7f1c | 4170 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 4171 | |
2e9f7f1c | 4172 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4173 | if (IS_ERR(snap_name)) { |
4174 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4175 | goto out_err; |
2e9f7f1c AE |
4176 | } |
4177 | ||
4178 | spec->pool_name = pool_name; | |
4179 | spec->image_name = image_name; | |
4180 | spec->snap_name = snap_name; | |
9e15b77d AE |
4181 | |
4182 | return 0; | |
4183 | out_err: | |
2e9f7f1c AE |
4184 | kfree(image_name); |
4185 | kfree(pool_name); | |
9e15b77d AE |
4186 | |
4187 | return ret; | |
4188 | } | |
4189 | ||
cc4a38bd | 4190 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4191 | { |
4192 | size_t size; | |
4193 | int ret; | |
4194 | void *reply_buf; | |
4195 | void *p; | |
4196 | void *end; | |
4197 | u64 seq; | |
4198 | u32 snap_count; | |
4199 | struct ceph_snap_context *snapc; | |
4200 | u32 i; | |
4201 | ||
4202 | /* | |
4203 | * We'll need room for the seq value (maximum snapshot id), | |
4204 | * snapshot count, and array of that many snapshot ids. | |
4205 | * For now we have a fixed upper limit on the number we're | |
4206 | * prepared to receive. | |
4207 | */ | |
4208 | size = sizeof (__le64) + sizeof (__le32) + | |
4209 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4210 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4211 | if (!reply_buf) | |
4212 | return -ENOMEM; | |
4213 | ||
36be9a76 | 4214 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4215 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4216 | reply_buf, size); |
36be9a76 | 4217 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4218 | if (ret < 0) |
4219 | goto out; | |
4220 | ||
35d489f9 | 4221 | p = reply_buf; |
57385b51 AE |
4222 | end = reply_buf + ret; |
4223 | ret = -ERANGE; | |
35d489f9 AE |
4224 | ceph_decode_64_safe(&p, end, seq, out); |
4225 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4226 | ||
4227 | /* | |
4228 | * Make sure the reported number of snapshot ids wouldn't go | |
4229 | * beyond the end of our buffer. But before checking that, | |
4230 | * make sure the computed size of the snapshot context we | |
4231 | * allocate is representable in a size_t. | |
4232 | */ | |
4233 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4234 | / sizeof (u64)) { | |
4235 | ret = -EINVAL; | |
4236 | goto out; | |
4237 | } | |
4238 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4239 | goto out; | |
468521c1 | 4240 | ret = 0; |
35d489f9 | 4241 | |
812164f8 | 4242 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4243 | if (!snapc) { |
4244 | ret = -ENOMEM; | |
4245 | goto out; | |
4246 | } | |
35d489f9 | 4247 | snapc->seq = seq; |
35d489f9 AE |
4248 | for (i = 0; i < snap_count; i++) |
4249 | snapc->snaps[i] = ceph_decode_64(&p); | |
4250 | ||
49ece554 | 4251 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4252 | rbd_dev->header.snapc = snapc; |
4253 | ||
4254 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4255 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4256 | out: |
4257 | kfree(reply_buf); | |
4258 | ||
57385b51 | 4259 | return ret; |
35d489f9 AE |
4260 | } |
4261 | ||
54cac61f AE |
4262 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4263 | u64 snap_id) | |
b8b1e2db AE |
4264 | { |
4265 | size_t size; | |
4266 | void *reply_buf; | |
54cac61f | 4267 | __le64 snapid; |
b8b1e2db AE |
4268 | int ret; |
4269 | void *p; | |
4270 | void *end; | |
b8b1e2db AE |
4271 | char *snap_name; |
4272 | ||
4273 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4274 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4275 | if (!reply_buf) | |
4276 | return ERR_PTR(-ENOMEM); | |
4277 | ||
54cac61f | 4278 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4279 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4280 | "rbd", "get_snapshot_name", |
54cac61f | 4281 | &snapid, sizeof (snapid), |
e2a58ee5 | 4282 | reply_buf, size); |
36be9a76 | 4283 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4284 | if (ret < 0) { |
4285 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4286 | goto out; |
f40eb349 | 4287 | } |
b8b1e2db AE |
4288 | |
4289 | p = reply_buf; | |
f40eb349 | 4290 | end = reply_buf + ret; |
e5c35534 | 4291 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4292 | if (IS_ERR(snap_name)) |
b8b1e2db | 4293 | goto out; |
b8b1e2db | 4294 | |
f40eb349 | 4295 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4296 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4297 | out: |
4298 | kfree(reply_buf); | |
4299 | ||
f40eb349 | 4300 | return snap_name; |
b8b1e2db AE |
4301 | } |
4302 | ||
2df3fac7 | 4303 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4304 | { |
2df3fac7 | 4305 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4306 | int ret; |
117973fb | 4307 | |
1617e40c JD |
4308 | ret = rbd_dev_v2_image_size(rbd_dev); |
4309 | if (ret) | |
cfbf6377 | 4310 | return ret; |
1617e40c | 4311 | |
2df3fac7 AE |
4312 | if (first_time) { |
4313 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4314 | if (ret) | |
cfbf6377 | 4315 | return ret; |
2df3fac7 AE |
4316 | } |
4317 | ||
642a2537 AE |
4318 | /* |
4319 | * If the image supports layering, get the parent info. We | |
4320 | * need to probe the first time regardless. Thereafter we | |
4321 | * only need to if there's a parent, to see if it has | |
4322 | * disappeared due to the mapped image getting flattened. | |
4323 | */ | |
4324 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING && | |
4325 | (first_time || rbd_dev->parent_spec)) { | |
4326 | bool warn; | |
4327 | ||
4328 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4329 | if (ret) | |
cfbf6377 | 4330 | return ret; |
642a2537 AE |
4331 | |
4332 | /* | |
4333 | * Print a warning if this is the initial probe and | |
4334 | * the image has a parent. Don't print it if the | |
4335 | * image now being probed is itself a parent. We | |
4336 | * can tell at this point because we won't know its | |
4337 | * pool name yet (just its pool id). | |
4338 | */ | |
4339 | warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name; | |
4340 | if (first_time && warn) | |
4341 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4342 | "is EXPERIMENTAL!"); | |
4343 | } | |
4344 | ||
29334ba4 AE |
4345 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) |
4346 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
4347 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
117973fb | 4348 | |
cc4a38bd | 4349 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4350 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4351 | |
4352 | return ret; | |
4353 | } | |
4354 | ||
dfc5606d YS |
4355 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4356 | { | |
dfc5606d | 4357 | struct device *dev; |
cd789ab9 | 4358 | int ret; |
dfc5606d | 4359 | |
cd789ab9 | 4360 | dev = &rbd_dev->dev; |
dfc5606d YS |
4361 | dev->bus = &rbd_bus_type; |
4362 | dev->type = &rbd_device_type; | |
4363 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4364 | dev->release = rbd_dev_device_release; |
de71a297 | 4365 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4366 | ret = device_register(dev); |
dfc5606d | 4367 | |
dfc5606d | 4368 | return ret; |
602adf40 YS |
4369 | } |
4370 | ||
dfc5606d YS |
4371 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4372 | { | |
4373 | device_unregister(&rbd_dev->dev); | |
4374 | } | |
4375 | ||
e2839308 | 4376 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
4377 | |
4378 | /* | |
499afd5b AE |
4379 | * Get a unique rbd identifier for the given new rbd_dev, and add |
4380 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 4381 | */ |
e2839308 | 4382 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4383 | { |
e2839308 | 4384 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
4385 | |
4386 | spin_lock(&rbd_dev_list_lock); | |
4387 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4388 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
4389 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
4390 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 4391 | } |
b7f23c36 | 4392 | |
1ddbe94e | 4393 | /* |
499afd5b AE |
4394 | * Remove an rbd_dev from the global list, and record that its |
4395 | * identifier is no longer in use. | |
1ddbe94e | 4396 | */ |
e2839308 | 4397 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4398 | { |
d184f6bf | 4399 | struct list_head *tmp; |
de71a297 | 4400 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
4401 | int max_id; |
4402 | ||
aafb230e | 4403 | rbd_assert(rbd_id > 0); |
499afd5b | 4404 | |
e2839308 AE |
4405 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
4406 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
4407 | spin_lock(&rbd_dev_list_lock); |
4408 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
4409 | |
4410 | /* | |
4411 | * If the id being "put" is not the current maximum, there | |
4412 | * is nothing special we need to do. | |
4413 | */ | |
e2839308 | 4414 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
4415 | spin_unlock(&rbd_dev_list_lock); |
4416 | return; | |
4417 | } | |
4418 | ||
4419 | /* | |
4420 | * We need to update the current maximum id. Search the | |
4421 | * list to find out what it is. We're more likely to find | |
4422 | * the maximum at the end, so search the list backward. | |
4423 | */ | |
4424 | max_id = 0; | |
4425 | list_for_each_prev(tmp, &rbd_dev_list) { | |
4426 | struct rbd_device *rbd_dev; | |
4427 | ||
4428 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
4429 | if (rbd_dev->dev_id > max_id) |
4430 | max_id = rbd_dev->dev_id; | |
d184f6bf | 4431 | } |
499afd5b | 4432 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 4433 | |
1ddbe94e | 4434 | /* |
e2839308 | 4435 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
4436 | * which case it now accurately reflects the new maximum. |
4437 | * Be careful not to overwrite the maximum value in that | |
4438 | * case. | |
1ddbe94e | 4439 | */ |
e2839308 AE |
4440 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
4441 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
4442 | } |
4443 | ||
e28fff26 AE |
4444 | /* |
4445 | * Skips over white space at *buf, and updates *buf to point to the | |
4446 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4447 | * the token (string of non-white space characters) found. Note |
4448 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4449 | */ |
4450 | static inline size_t next_token(const char **buf) | |
4451 | { | |
4452 | /* | |
4453 | * These are the characters that produce nonzero for | |
4454 | * isspace() in the "C" and "POSIX" locales. | |
4455 | */ | |
4456 | const char *spaces = " \f\n\r\t\v"; | |
4457 | ||
4458 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4459 | ||
4460 | return strcspn(*buf, spaces); /* Return token length */ | |
4461 | } | |
4462 | ||
4463 | /* | |
4464 | * Finds the next token in *buf, and if the provided token buffer is | |
4465 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4466 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4467 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4468 | * |
4469 | * Returns the length of the token found (not including the '\0'). | |
4470 | * Return value will be 0 if no token is found, and it will be >= | |
4471 | * token_size if the token would not fit. | |
4472 | * | |
593a9e7b | 4473 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4474 | * found token. Note that this occurs even if the token buffer is |
4475 | * too small to hold it. | |
4476 | */ | |
4477 | static inline size_t copy_token(const char **buf, | |
4478 | char *token, | |
4479 | size_t token_size) | |
4480 | { | |
4481 | size_t len; | |
4482 | ||
4483 | len = next_token(buf); | |
4484 | if (len < token_size) { | |
4485 | memcpy(token, *buf, len); | |
4486 | *(token + len) = '\0'; | |
4487 | } | |
4488 | *buf += len; | |
4489 | ||
4490 | return len; | |
4491 | } | |
4492 | ||
ea3352f4 AE |
4493 | /* |
4494 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4495 | * enough to hold a copy of it, and copies the token into the new | |
4496 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4497 | * that a duplicate buffer is created even for a zero-length token. | |
4498 | * | |
4499 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4500 | * pointer if memory for the duplicate was not available. If | |
4501 | * the lenp argument is a non-null pointer, the length of the token | |
4502 | * (not including the '\0') is returned in *lenp. | |
4503 | * | |
4504 | * If successful, the *buf pointer will be updated to point beyond | |
4505 | * the end of the found token. | |
4506 | * | |
4507 | * Note: uses GFP_KERNEL for allocation. | |
4508 | */ | |
4509 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4510 | { | |
4511 | char *dup; | |
4512 | size_t len; | |
4513 | ||
4514 | len = next_token(buf); | |
4caf35f9 | 4515 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4516 | if (!dup) |
4517 | return NULL; | |
ea3352f4 AE |
4518 | *(dup + len) = '\0'; |
4519 | *buf += len; | |
4520 | ||
4521 | if (lenp) | |
4522 | *lenp = len; | |
4523 | ||
4524 | return dup; | |
4525 | } | |
4526 | ||
a725f65e | 4527 | /* |
859c31df AE |
4528 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4529 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4530 | * and the data written is passed here via a NUL-terminated buffer. | |
4531 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4532 | * |
859c31df AE |
4533 | * The information extracted from these options is recorded in |
4534 | * the other parameters which return dynamically-allocated | |
4535 | * structures: | |
4536 | * ceph_opts | |
4537 | * The address of a pointer that will refer to a ceph options | |
4538 | * structure. Caller must release the returned pointer using | |
4539 | * ceph_destroy_options() when it is no longer needed. | |
4540 | * rbd_opts | |
4541 | * Address of an rbd options pointer. Fully initialized by | |
4542 | * this function; caller must release with kfree(). | |
4543 | * spec | |
4544 | * Address of an rbd image specification pointer. Fully | |
4545 | * initialized by this function based on parsed options. | |
4546 | * Caller must release with rbd_spec_put(). | |
4547 | * | |
4548 | * The options passed take this form: | |
4549 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4550 | * where: | |
4551 | * <mon_addrs> | |
4552 | * A comma-separated list of one or more monitor addresses. | |
4553 | * A monitor address is an ip address, optionally followed | |
4554 | * by a port number (separated by a colon). | |
4555 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4556 | * <options> | |
4557 | * A comma-separated list of ceph and/or rbd options. | |
4558 | * <pool_name> | |
4559 | * The name of the rados pool containing the rbd image. | |
4560 | * <image_name> | |
4561 | * The name of the image in that pool to map. | |
4562 | * <snap_id> | |
4563 | * An optional snapshot id. If provided, the mapping will | |
4564 | * present data from the image at the time that snapshot was | |
4565 | * created. The image head is used if no snapshot id is | |
4566 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4567 | */ |
859c31df | 4568 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4569 | struct ceph_options **ceph_opts, |
859c31df AE |
4570 | struct rbd_options **opts, |
4571 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4572 | { |
d22f76e7 | 4573 | size_t len; |
859c31df | 4574 | char *options; |
0ddebc0c | 4575 | const char *mon_addrs; |
ecb4dc22 | 4576 | char *snap_name; |
0ddebc0c | 4577 | size_t mon_addrs_size; |
859c31df | 4578 | struct rbd_spec *spec = NULL; |
4e9afeba | 4579 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4580 | struct ceph_options *copts; |
dc79b113 | 4581 | int ret; |
e28fff26 AE |
4582 | |
4583 | /* The first four tokens are required */ | |
4584 | ||
7ef3214a | 4585 | len = next_token(&buf); |
4fb5d671 AE |
4586 | if (!len) { |
4587 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4588 | return -EINVAL; | |
4589 | } | |
0ddebc0c | 4590 | mon_addrs = buf; |
f28e565a | 4591 | mon_addrs_size = len + 1; |
7ef3214a | 4592 | buf += len; |
a725f65e | 4593 | |
dc79b113 | 4594 | ret = -EINVAL; |
f28e565a AE |
4595 | options = dup_token(&buf, NULL); |
4596 | if (!options) | |
dc79b113 | 4597 | return -ENOMEM; |
4fb5d671 AE |
4598 | if (!*options) { |
4599 | rbd_warn(NULL, "no options provided"); | |
4600 | goto out_err; | |
4601 | } | |
e28fff26 | 4602 | |
859c31df AE |
4603 | spec = rbd_spec_alloc(); |
4604 | if (!spec) | |
f28e565a | 4605 | goto out_mem; |
859c31df AE |
4606 | |
4607 | spec->pool_name = dup_token(&buf, NULL); | |
4608 | if (!spec->pool_name) | |
4609 | goto out_mem; | |
4fb5d671 AE |
4610 | if (!*spec->pool_name) { |
4611 | rbd_warn(NULL, "no pool name provided"); | |
4612 | goto out_err; | |
4613 | } | |
e28fff26 | 4614 | |
69e7a02f | 4615 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4616 | if (!spec->image_name) |
f28e565a | 4617 | goto out_mem; |
4fb5d671 AE |
4618 | if (!*spec->image_name) { |
4619 | rbd_warn(NULL, "no image name provided"); | |
4620 | goto out_err; | |
4621 | } | |
d4b125e9 | 4622 | |
f28e565a AE |
4623 | /* |
4624 | * Snapshot name is optional; default is to use "-" | |
4625 | * (indicating the head/no snapshot). | |
4626 | */ | |
3feeb894 | 4627 | len = next_token(&buf); |
820a5f3e | 4628 | if (!len) { |
3feeb894 AE |
4629 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4630 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4631 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4632 | ret = -ENAMETOOLONG; |
f28e565a | 4633 | goto out_err; |
849b4260 | 4634 | } |
ecb4dc22 AE |
4635 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4636 | if (!snap_name) | |
f28e565a | 4637 | goto out_mem; |
ecb4dc22 AE |
4638 | *(snap_name + len) = '\0'; |
4639 | spec->snap_name = snap_name; | |
e5c35534 | 4640 | |
0ddebc0c | 4641 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4642 | |
4e9afeba AE |
4643 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4644 | if (!rbd_opts) | |
4645 | goto out_mem; | |
4646 | ||
4647 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4648 | |
859c31df | 4649 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4650 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4651 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4652 | if (IS_ERR(copts)) { |
4653 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4654 | goto out_err; |
4655 | } | |
859c31df AE |
4656 | kfree(options); |
4657 | ||
4658 | *ceph_opts = copts; | |
4e9afeba | 4659 | *opts = rbd_opts; |
859c31df | 4660 | *rbd_spec = spec; |
0ddebc0c | 4661 | |
dc79b113 | 4662 | return 0; |
f28e565a | 4663 | out_mem: |
dc79b113 | 4664 | ret = -ENOMEM; |
d22f76e7 | 4665 | out_err: |
859c31df AE |
4666 | kfree(rbd_opts); |
4667 | rbd_spec_put(spec); | |
f28e565a | 4668 | kfree(options); |
d22f76e7 | 4669 | |
dc79b113 | 4670 | return ret; |
a725f65e AE |
4671 | } |
4672 | ||
589d30e0 AE |
4673 | /* |
4674 | * An rbd format 2 image has a unique identifier, distinct from the | |
4675 | * name given to it by the user. Internally, that identifier is | |
4676 | * what's used to specify the names of objects related to the image. | |
4677 | * | |
4678 | * A special "rbd id" object is used to map an rbd image name to its | |
4679 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4680 | * with the supplied name. | |
4681 | * | |
4682 | * This function will record the given rbd_dev's image_id field if | |
4683 | * it can be determined, and in that case will return 0. If any | |
4684 | * errors occur a negative errno will be returned and the rbd_dev's | |
4685 | * image_id field will be unchanged (and should be NULL). | |
4686 | */ | |
4687 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4688 | { | |
4689 | int ret; | |
4690 | size_t size; | |
4691 | char *object_name; | |
4692 | void *response; | |
c0fba368 | 4693 | char *image_id; |
2f82ee54 | 4694 | |
2c0d0a10 AE |
4695 | /* |
4696 | * When probing a parent image, the image id is already | |
4697 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4698 | * need to fetch the image id again in this case. We |
4699 | * do still need to set the image format though. | |
2c0d0a10 | 4700 | */ |
c0fba368 AE |
4701 | if (rbd_dev->spec->image_id) { |
4702 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4703 | ||
2c0d0a10 | 4704 | return 0; |
c0fba368 | 4705 | } |
2c0d0a10 | 4706 | |
589d30e0 AE |
4707 | /* |
4708 | * First, see if the format 2 image id file exists, and if | |
4709 | * so, get the image's persistent id from it. | |
4710 | */ | |
69e7a02f | 4711 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4712 | object_name = kmalloc(size, GFP_NOIO); |
4713 | if (!object_name) | |
4714 | return -ENOMEM; | |
0d7dbfce | 4715 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4716 | dout("rbd id object name is %s\n", object_name); |
4717 | ||
4718 | /* Response will be an encoded string, which includes a length */ | |
4719 | ||
4720 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4721 | response = kzalloc(size, GFP_NOIO); | |
4722 | if (!response) { | |
4723 | ret = -ENOMEM; | |
4724 | goto out; | |
4725 | } | |
4726 | ||
c0fba368 AE |
4727 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4728 | ||
36be9a76 | 4729 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4730 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4731 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4732 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4733 | if (ret == -ENOENT) { |
4734 | image_id = kstrdup("", GFP_KERNEL); | |
4735 | ret = image_id ? 0 : -ENOMEM; | |
4736 | if (!ret) | |
4737 | rbd_dev->image_format = 1; | |
4738 | } else if (ret > sizeof (__le32)) { | |
4739 | void *p = response; | |
4740 | ||
4741 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4742 | NULL, GFP_NOIO); |
c0fba368 AE |
4743 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4744 | if (!ret) | |
4745 | rbd_dev->image_format = 2; | |
589d30e0 | 4746 | } else { |
c0fba368 AE |
4747 | ret = -EINVAL; |
4748 | } | |
4749 | ||
4750 | if (!ret) { | |
4751 | rbd_dev->spec->image_id = image_id; | |
4752 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4753 | } |
4754 | out: | |
4755 | kfree(response); | |
4756 | kfree(object_name); | |
4757 | ||
4758 | return ret; | |
4759 | } | |
4760 | ||
3abef3b3 AE |
4761 | /* |
4762 | * Undo whatever state changes are made by v1 or v2 header info | |
4763 | * call. | |
4764 | */ | |
6fd48b3b AE |
4765 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
4766 | { | |
4767 | struct rbd_image_header *header; | |
4768 | ||
392a9dad AE |
4769 | /* Drop parent reference unless it's already been done (or none) */ |
4770 | ||
4771 | if (rbd_dev->parent_overlap) | |
4772 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
4773 | |
4774 | /* Free dynamic fields from the header, then zero it out */ | |
4775 | ||
4776 | header = &rbd_dev->header; | |
812164f8 | 4777 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4778 | kfree(header->snap_sizes); |
4779 | kfree(header->snap_names); | |
4780 | kfree(header->object_prefix); | |
4781 | memset(header, 0, sizeof (*header)); | |
4782 | } | |
4783 | ||
2df3fac7 | 4784 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4785 | { |
4786 | int ret; | |
a30b71b9 | 4787 | |
1e130199 | 4788 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 4789 | if (ret) |
b1b5402a AE |
4790 | goto out_err; |
4791 | ||
2df3fac7 AE |
4792 | /* |
4793 | * Get the and check features for the image. Currently the | |
4794 | * features are assumed to never change. | |
4795 | */ | |
b1b5402a | 4796 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 4797 | if (ret) |
9d475de5 | 4798 | goto out_err; |
35d489f9 | 4799 | |
cc070d59 AE |
4800 | /* If the image supports fancy striping, get its parameters */ |
4801 | ||
4802 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4803 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4804 | if (ret < 0) | |
4805 | goto out_err; | |
4806 | } | |
2df3fac7 | 4807 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 4808 | |
35152979 | 4809 | return 0; |
9d475de5 | 4810 | out_err: |
642a2537 | 4811 | rbd_dev->header.features = 0; |
1e130199 AE |
4812 | kfree(rbd_dev->header.object_prefix); |
4813 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4814 | |
4815 | return ret; | |
a30b71b9 AE |
4816 | } |
4817 | ||
124afba2 | 4818 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4819 | { |
2f82ee54 | 4820 | struct rbd_device *parent = NULL; |
124afba2 AE |
4821 | struct rbd_spec *parent_spec; |
4822 | struct rbd_client *rbdc; | |
4823 | int ret; | |
4824 | ||
4825 | if (!rbd_dev->parent_spec) | |
4826 | return 0; | |
4827 | /* | |
4828 | * We need to pass a reference to the client and the parent | |
4829 | * spec when creating the parent rbd_dev. Images related by | |
4830 | * parent/child relationships always share both. | |
4831 | */ | |
4832 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4833 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4834 | ||
4835 | ret = -ENOMEM; | |
4836 | parent = rbd_dev_create(rbdc, parent_spec); | |
4837 | if (!parent) | |
4838 | goto out_err; | |
4839 | ||
1f3ef788 | 4840 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
4841 | if (ret < 0) |
4842 | goto out_err; | |
4843 | rbd_dev->parent = parent; | |
a2acd00e | 4844 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
4845 | |
4846 | return 0; | |
4847 | out_err: | |
4848 | if (parent) { | |
fb65d228 | 4849 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
4850 | kfree(rbd_dev->header_name); |
4851 | rbd_dev_destroy(parent); | |
4852 | } else { | |
4853 | rbd_put_client(rbdc); | |
4854 | rbd_spec_put(parent_spec); | |
4855 | } | |
4856 | ||
4857 | return ret; | |
4858 | } | |
4859 | ||
200a6a8b | 4860 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4861 | { |
83a06263 | 4862 | int ret; |
d1cf5788 | 4863 | |
83a06263 AE |
4864 | /* generate unique id: find highest unique id, add one */ |
4865 | rbd_dev_id_get(rbd_dev); | |
4866 | ||
4867 | /* Fill in the device name, now that we have its id. */ | |
4868 | BUILD_BUG_ON(DEV_NAME_LEN | |
4869 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4870 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4871 | ||
4872 | /* Get our block major device number. */ | |
4873 | ||
4874 | ret = register_blkdev(0, rbd_dev->name); | |
4875 | if (ret < 0) | |
4876 | goto err_out_id; | |
4877 | rbd_dev->major = ret; | |
4878 | ||
4879 | /* Set up the blkdev mapping. */ | |
4880 | ||
4881 | ret = rbd_init_disk(rbd_dev); | |
4882 | if (ret) | |
4883 | goto err_out_blkdev; | |
4884 | ||
f35a4dee | 4885 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
4886 | if (ret) |
4887 | goto err_out_disk; | |
f35a4dee AE |
4888 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
4889 | ||
4890 | ret = rbd_bus_add_dev(rbd_dev); | |
4891 | if (ret) | |
4892 | goto err_out_mapping; | |
83a06263 | 4893 | |
83a06263 AE |
4894 | /* Everything's ready. Announce the disk to the world. */ |
4895 | ||
129b79d4 | 4896 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4897 | add_disk(rbd_dev->disk); |
4898 | ||
4899 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4900 | (unsigned long long) rbd_dev->mapping.size); | |
4901 | ||
4902 | return ret; | |
2f82ee54 | 4903 | |
f35a4dee AE |
4904 | err_out_mapping: |
4905 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
4906 | err_out_disk: |
4907 | rbd_free_disk(rbd_dev); | |
4908 | err_out_blkdev: | |
4909 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4910 | err_out_id: | |
4911 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4912 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4913 | |
4914 | return ret; | |
4915 | } | |
4916 | ||
332bb12d AE |
4917 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4918 | { | |
4919 | struct rbd_spec *spec = rbd_dev->spec; | |
4920 | size_t size; | |
4921 | ||
4922 | /* Record the header object name for this rbd image. */ | |
4923 | ||
4924 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4925 | ||
4926 | if (rbd_dev->image_format == 1) | |
4927 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4928 | else | |
4929 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4930 | ||
4931 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4932 | if (!rbd_dev->header_name) | |
4933 | return -ENOMEM; | |
4934 | ||
4935 | if (rbd_dev->image_format == 1) | |
4936 | sprintf(rbd_dev->header_name, "%s%s", | |
4937 | spec->image_name, RBD_SUFFIX); | |
4938 | else | |
4939 | sprintf(rbd_dev->header_name, "%s%s", | |
4940 | RBD_HEADER_PREFIX, spec->image_id); | |
4941 | return 0; | |
4942 | } | |
4943 | ||
200a6a8b AE |
4944 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4945 | { | |
6fd48b3b | 4946 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 4947 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4948 | rbd_dev->header_name = NULL; |
4949 | rbd_dev->image_format = 0; | |
4950 | kfree(rbd_dev->spec->image_id); | |
4951 | rbd_dev->spec->image_id = NULL; | |
4952 | ||
200a6a8b AE |
4953 | rbd_dev_destroy(rbd_dev); |
4954 | } | |
4955 | ||
a30b71b9 AE |
4956 | /* |
4957 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
4958 | * device. If this image is the one being mapped (i.e., not a |
4959 | * parent), initiate a watch on its header object before using that | |
4960 | * object to get detailed information about the rbd image. | |
a30b71b9 | 4961 | */ |
1f3ef788 | 4962 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
4963 | { |
4964 | int ret; | |
b644de2b | 4965 | int tmp; |
a30b71b9 AE |
4966 | |
4967 | /* | |
3abef3b3 AE |
4968 | * Get the id from the image id object. Unless there's an |
4969 | * error, rbd_dev->spec->image_id will be filled in with | |
4970 | * a dynamically-allocated string, and rbd_dev->image_format | |
4971 | * will be set to either 1 or 2. | |
a30b71b9 AE |
4972 | */ |
4973 | ret = rbd_dev_image_id(rbd_dev); | |
4974 | if (ret) | |
c0fba368 AE |
4975 | return ret; |
4976 | rbd_assert(rbd_dev->spec->image_id); | |
4977 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4978 | ||
332bb12d AE |
4979 | ret = rbd_dev_header_name(rbd_dev); |
4980 | if (ret) | |
4981 | goto err_out_format; | |
4982 | ||
1f3ef788 AE |
4983 | if (mapping) { |
4984 | ret = rbd_dev_header_watch_sync(rbd_dev, true); | |
4985 | if (ret) | |
4986 | goto out_header_name; | |
4987 | } | |
b644de2b | 4988 | |
c0fba368 | 4989 | if (rbd_dev->image_format == 1) |
99a41ebc | 4990 | ret = rbd_dev_v1_header_info(rbd_dev); |
a30b71b9 | 4991 | else |
2df3fac7 | 4992 | ret = rbd_dev_v2_header_info(rbd_dev); |
5655c4d9 | 4993 | if (ret) |
b644de2b | 4994 | goto err_out_watch; |
83a06263 | 4995 | |
9bb81c9b AE |
4996 | ret = rbd_dev_spec_update(rbd_dev); |
4997 | if (ret) | |
33dca39f | 4998 | goto err_out_probe; |
9bb81c9b AE |
4999 | |
5000 | ret = rbd_dev_probe_parent(rbd_dev); | |
30d60ba2 AE |
5001 | if (ret) |
5002 | goto err_out_probe; | |
5003 | ||
5004 | dout("discovered format %u image, header name is %s\n", | |
5005 | rbd_dev->image_format, rbd_dev->header_name); | |
83a06263 | 5006 | |
30d60ba2 | 5007 | return 0; |
6fd48b3b AE |
5008 | err_out_probe: |
5009 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5010 | err_out_watch: |
1f3ef788 AE |
5011 | if (mapping) { |
5012 | tmp = rbd_dev_header_watch_sync(rbd_dev, false); | |
5013 | if (tmp) | |
5014 | rbd_warn(rbd_dev, "unable to tear down " | |
5015 | "watch request (%d)\n", tmp); | |
5016 | } | |
332bb12d AE |
5017 | out_header_name: |
5018 | kfree(rbd_dev->header_name); | |
5019 | rbd_dev->header_name = NULL; | |
5020 | err_out_format: | |
5021 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5022 | kfree(rbd_dev->spec->image_id); |
5023 | rbd_dev->spec->image_id = NULL; | |
5024 | ||
5025 | dout("probe failed, returning %d\n", ret); | |
5026 | ||
a30b71b9 AE |
5027 | return ret; |
5028 | } | |
5029 | ||
59c2be1e YS |
5030 | static ssize_t rbd_add(struct bus_type *bus, |
5031 | const char *buf, | |
5032 | size_t count) | |
602adf40 | 5033 | { |
cb8627c7 | 5034 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5035 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5036 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5037 | struct rbd_spec *spec = NULL; |
9d3997fd | 5038 | struct rbd_client *rbdc; |
27cc2594 | 5039 | struct ceph_osd_client *osdc; |
51344a38 | 5040 | bool read_only; |
27cc2594 | 5041 | int rc = -ENOMEM; |
602adf40 YS |
5042 | |
5043 | if (!try_module_get(THIS_MODULE)) | |
5044 | return -ENODEV; | |
5045 | ||
602adf40 | 5046 | /* parse add command */ |
859c31df | 5047 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5048 | if (rc < 0) |
bd4ba655 | 5049 | goto err_out_module; |
51344a38 AE |
5050 | read_only = rbd_opts->read_only; |
5051 | kfree(rbd_opts); | |
5052 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5053 | |
9d3997fd AE |
5054 | rbdc = rbd_get_client(ceph_opts); |
5055 | if (IS_ERR(rbdc)) { | |
5056 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5057 | goto err_out_args; |
9d3997fd | 5058 | } |
602adf40 | 5059 | |
602adf40 | 5060 | /* pick the pool */ |
9d3997fd | 5061 | osdc = &rbdc->client->osdc; |
859c31df | 5062 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
5063 | if (rc < 0) |
5064 | goto err_out_client; | |
c0cd10db | 5065 | spec->pool_id = (u64)rc; |
859c31df | 5066 | |
0903e875 AE |
5067 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5068 | ||
c0cd10db AE |
5069 | if (spec->pool_id > (u64)U32_MAX) { |
5070 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
5071 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
5072 | rc = -EIO; |
5073 | goto err_out_client; | |
5074 | } | |
5075 | ||
c53d5893 | 5076 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5077 | if (!rbd_dev) |
5078 | goto err_out_client; | |
c53d5893 AE |
5079 | rbdc = NULL; /* rbd_dev now owns this */ |
5080 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5081 | |
1f3ef788 | 5082 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5083 | if (rc < 0) |
c53d5893 | 5084 | goto err_out_rbd_dev; |
05fd6f6f | 5085 | |
7ce4eef7 AE |
5086 | /* If we are mapping a snapshot it must be marked read-only */ |
5087 | ||
5088 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5089 | read_only = true; | |
5090 | rbd_dev->mapping.read_only = read_only; | |
5091 | ||
b536f69a | 5092 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 AE |
5093 | if (rc) { |
5094 | rbd_dev_image_release(rbd_dev); | |
5095 | goto err_out_module; | |
5096 | } | |
5097 | ||
5098 | return count; | |
b536f69a | 5099 | |
c53d5893 AE |
5100 | err_out_rbd_dev: |
5101 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5102 | err_out_client: |
9d3997fd | 5103 | rbd_put_client(rbdc); |
0ddebc0c | 5104 | err_out_args: |
859c31df | 5105 | rbd_spec_put(spec); |
bd4ba655 AE |
5106 | err_out_module: |
5107 | module_put(THIS_MODULE); | |
27cc2594 | 5108 | |
602adf40 | 5109 | dout("Error adding device %s\n", buf); |
27cc2594 | 5110 | |
c0cd10db | 5111 | return (ssize_t)rc; |
602adf40 YS |
5112 | } |
5113 | ||
200a6a8b | 5114 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5115 | { |
593a9e7b | 5116 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5117 | |
602adf40 | 5118 | rbd_free_disk(rbd_dev); |
200a6a8b | 5119 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5120 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 | 5121 | unregister_blkdev(rbd_dev->major, rbd_dev->name); |
200a6a8b | 5122 | rbd_dev->major = 0; |
e2839308 | 5123 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5124 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5125 | } |
5126 | ||
05a46afd AE |
5127 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5128 | { | |
ad945fc1 | 5129 | while (rbd_dev->parent) { |
05a46afd AE |
5130 | struct rbd_device *first = rbd_dev; |
5131 | struct rbd_device *second = first->parent; | |
5132 | struct rbd_device *third; | |
5133 | ||
5134 | /* | |
5135 | * Follow to the parent with no grandparent and | |
5136 | * remove it. | |
5137 | */ | |
5138 | while (second && (third = second->parent)) { | |
5139 | first = second; | |
5140 | second = third; | |
5141 | } | |
ad945fc1 | 5142 | rbd_assert(second); |
8ad42cd0 | 5143 | rbd_dev_image_release(second); |
ad945fc1 AE |
5144 | first->parent = NULL; |
5145 | first->parent_overlap = 0; | |
5146 | ||
5147 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5148 | rbd_spec_put(first->parent_spec); |
5149 | first->parent_spec = NULL; | |
05a46afd AE |
5150 | } |
5151 | } | |
5152 | ||
dfc5606d YS |
5153 | static ssize_t rbd_remove(struct bus_type *bus, |
5154 | const char *buf, | |
5155 | size_t count) | |
602adf40 YS |
5156 | { |
5157 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5158 | struct list_head *tmp; |
5159 | int dev_id; | |
602adf40 | 5160 | unsigned long ul; |
82a442d2 | 5161 | bool already = false; |
0d8189e1 | 5162 | int ret; |
602adf40 | 5163 | |
bb8e0e84 | 5164 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5165 | if (ret) |
5166 | return ret; | |
602adf40 YS |
5167 | |
5168 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5169 | dev_id = (int)ul; |
5170 | if (dev_id != ul) | |
602adf40 YS |
5171 | return -EINVAL; |
5172 | ||
751cc0e3 AE |
5173 | ret = -ENOENT; |
5174 | spin_lock(&rbd_dev_list_lock); | |
5175 | list_for_each(tmp, &rbd_dev_list) { | |
5176 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5177 | if (rbd_dev->dev_id == dev_id) { | |
5178 | ret = 0; | |
5179 | break; | |
5180 | } | |
42382b70 | 5181 | } |
751cc0e3 AE |
5182 | if (!ret) { |
5183 | spin_lock_irq(&rbd_dev->lock); | |
5184 | if (rbd_dev->open_count) | |
5185 | ret = -EBUSY; | |
5186 | else | |
82a442d2 AE |
5187 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5188 | &rbd_dev->flags); | |
751cc0e3 AE |
5189 | spin_unlock_irq(&rbd_dev->lock); |
5190 | } | |
5191 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5192 | if (ret < 0 || already) |
1ba0f1e7 | 5193 | return ret; |
751cc0e3 | 5194 | |
1f3ef788 AE |
5195 | ret = rbd_dev_header_watch_sync(rbd_dev, false); |
5196 | if (ret) | |
5197 | rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret); | |
9abc5990 JD |
5198 | |
5199 | /* | |
5200 | * flush remaining watch callbacks - these must be complete | |
5201 | * before the osd_client is shutdown | |
5202 | */ | |
5203 | dout("%s: flushing notifies", __func__); | |
5204 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
9875201e JD |
5205 | /* |
5206 | * Don't free anything from rbd_dev->disk until after all | |
5207 | * notifies are completely processed. Otherwise | |
5208 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5209 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5210 | */ | |
5211 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5212 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5213 | module_put(THIS_MODULE); |
aafb230e | 5214 | |
1ba0f1e7 | 5215 | return count; |
602adf40 YS |
5216 | } |
5217 | ||
602adf40 YS |
5218 | /* |
5219 | * create control files in sysfs | |
dfc5606d | 5220 | * /sys/bus/rbd/... |
602adf40 YS |
5221 | */ |
5222 | static int rbd_sysfs_init(void) | |
5223 | { | |
dfc5606d | 5224 | int ret; |
602adf40 | 5225 | |
fed4c143 | 5226 | ret = device_register(&rbd_root_dev); |
21079786 | 5227 | if (ret < 0) |
dfc5606d | 5228 | return ret; |
602adf40 | 5229 | |
fed4c143 AE |
5230 | ret = bus_register(&rbd_bus_type); |
5231 | if (ret < 0) | |
5232 | device_unregister(&rbd_root_dev); | |
602adf40 | 5233 | |
602adf40 YS |
5234 | return ret; |
5235 | } | |
5236 | ||
5237 | static void rbd_sysfs_cleanup(void) | |
5238 | { | |
dfc5606d | 5239 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5240 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5241 | } |
5242 | ||
1c2a9dfe AE |
5243 | static int rbd_slab_init(void) |
5244 | { | |
5245 | rbd_assert(!rbd_img_request_cache); | |
5246 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5247 | sizeof (struct rbd_img_request), | |
5248 | __alignof__(struct rbd_img_request), | |
5249 | 0, NULL); | |
868311b1 AE |
5250 | if (!rbd_img_request_cache) |
5251 | return -ENOMEM; | |
5252 | ||
5253 | rbd_assert(!rbd_obj_request_cache); | |
5254 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5255 | sizeof (struct rbd_obj_request), | |
5256 | __alignof__(struct rbd_obj_request), | |
5257 | 0, NULL); | |
78c2a44a AE |
5258 | if (!rbd_obj_request_cache) |
5259 | goto out_err; | |
5260 | ||
5261 | rbd_assert(!rbd_segment_name_cache); | |
5262 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
5263 | MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL); | |
5264 | if (rbd_segment_name_cache) | |
1c2a9dfe | 5265 | return 0; |
78c2a44a AE |
5266 | out_err: |
5267 | if (rbd_obj_request_cache) { | |
5268 | kmem_cache_destroy(rbd_obj_request_cache); | |
5269 | rbd_obj_request_cache = NULL; | |
5270 | } | |
1c2a9dfe | 5271 | |
868311b1 AE |
5272 | kmem_cache_destroy(rbd_img_request_cache); |
5273 | rbd_img_request_cache = NULL; | |
5274 | ||
1c2a9dfe AE |
5275 | return -ENOMEM; |
5276 | } | |
5277 | ||
5278 | static void rbd_slab_exit(void) | |
5279 | { | |
78c2a44a AE |
5280 | rbd_assert(rbd_segment_name_cache); |
5281 | kmem_cache_destroy(rbd_segment_name_cache); | |
5282 | rbd_segment_name_cache = NULL; | |
5283 | ||
868311b1 AE |
5284 | rbd_assert(rbd_obj_request_cache); |
5285 | kmem_cache_destroy(rbd_obj_request_cache); | |
5286 | rbd_obj_request_cache = NULL; | |
5287 | ||
1c2a9dfe AE |
5288 | rbd_assert(rbd_img_request_cache); |
5289 | kmem_cache_destroy(rbd_img_request_cache); | |
5290 | rbd_img_request_cache = NULL; | |
5291 | } | |
5292 | ||
cc344fa1 | 5293 | static int __init rbd_init(void) |
602adf40 YS |
5294 | { |
5295 | int rc; | |
5296 | ||
1e32d34c AE |
5297 | if (!libceph_compatible(NULL)) { |
5298 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
5299 | ||
5300 | return -EINVAL; | |
5301 | } | |
1c2a9dfe | 5302 | rc = rbd_slab_init(); |
602adf40 YS |
5303 | if (rc) |
5304 | return rc; | |
1c2a9dfe AE |
5305 | rc = rbd_sysfs_init(); |
5306 | if (rc) | |
5307 | rbd_slab_exit(); | |
5308 | else | |
5309 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); | |
5310 | ||
5311 | return rc; | |
602adf40 YS |
5312 | } |
5313 | ||
cc344fa1 | 5314 | static void __exit rbd_exit(void) |
602adf40 YS |
5315 | { |
5316 | rbd_sysfs_cleanup(); | |
1c2a9dfe | 5317 | rbd_slab_exit(); |
602adf40 YS |
5318 | } |
5319 | ||
5320 | module_init(rbd_init); | |
5321 | module_exit(rbd_exit); | |
5322 | ||
d552c619 | 5323 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5324 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5325 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
5326 | MODULE_DESCRIPTION("rados block device"); | |
5327 | ||
5328 | /* following authorship retained from original osdblk.c */ | |
5329 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5330 | ||
5331 | MODULE_LICENSE("GPL"); |