]>
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> |
602adf40 YS |
36 | |
37 | #include <linux/kernel.h> | |
38 | #include <linux/device.h> | |
39 | #include <linux/module.h> | |
40 | #include <linux/fs.h> | |
41 | #include <linux/blkdev.h> | |
42 | ||
43 | #include "rbd_types.h" | |
44 | ||
aafb230e AE |
45 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
46 | ||
593a9e7b AE |
47 | /* |
48 | * The basic unit of block I/O is a sector. It is interpreted in a | |
49 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
50 | * universally 512 bytes. These symbols are just slightly more | |
51 | * meaningful than the bare numbers they represent. | |
52 | */ | |
53 | #define SECTOR_SHIFT 9 | |
54 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
55 | ||
f0f8cef5 AE |
56 | #define RBD_DRV_NAME "rbd" |
57 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
58 | |
59 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
60 | ||
d4b125e9 AE |
61 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
62 | #define RBD_MAX_SNAP_NAME_LEN \ | |
63 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
64 | ||
35d489f9 | 65 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
66 | |
67 | #define RBD_SNAP_HEAD_NAME "-" | |
68 | ||
9e15b77d AE |
69 | /* This allows a single page to hold an image name sent by OSD */ |
70 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 71 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 72 | |
1e130199 | 73 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 74 | |
d889140c AE |
75 | /* Feature bits */ |
76 | ||
5cbf6f12 AE |
77 | #define RBD_FEATURE_LAYERING (1<<0) |
78 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
79 | #define RBD_FEATURES_ALL \ | |
80 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
81 | |
82 | /* Features supported by this (client software) implementation. */ | |
83 | ||
770eba6e | 84 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 85 | |
81a89793 AE |
86 | /* |
87 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
88 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
89 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
90 | * enough to hold all possible device names. | |
91 | */ | |
602adf40 | 92 | #define DEV_NAME_LEN 32 |
81a89793 | 93 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
94 | |
95 | /* | |
96 | * block device image metadata (in-memory version) | |
97 | */ | |
98 | struct rbd_image_header { | |
f84344f3 | 99 | /* These four fields never change for a given rbd image */ |
849b4260 | 100 | char *object_prefix; |
34b13184 | 101 | u64 features; |
602adf40 YS |
102 | __u8 obj_order; |
103 | __u8 crypt_type; | |
104 | __u8 comp_type; | |
602adf40 | 105 | |
f84344f3 AE |
106 | /* The remaining fields need to be updated occasionally */ |
107 | u64 image_size; | |
108 | struct ceph_snap_context *snapc; | |
602adf40 YS |
109 | char *snap_names; |
110 | u64 *snap_sizes; | |
59c2be1e | 111 | |
500d0c0f AE |
112 | u64 stripe_unit; |
113 | u64 stripe_count; | |
59c2be1e YS |
114 | }; |
115 | ||
0d7dbfce AE |
116 | /* |
117 | * An rbd image specification. | |
118 | * | |
119 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
120 | * identify an image. Each rbd_dev structure includes a pointer to |
121 | * an rbd_spec structure that encapsulates this identity. | |
122 | * | |
123 | * Each of the id's in an rbd_spec has an associated name. For a | |
124 | * user-mapped image, the names are supplied and the id's associated | |
125 | * with them are looked up. For a layered image, a parent image is | |
126 | * defined by the tuple, and the names are looked up. | |
127 | * | |
128 | * An rbd_dev structure contains a parent_spec pointer which is | |
129 | * non-null if the image it represents is a child in a layered | |
130 | * image. This pointer will refer to the rbd_spec structure used | |
131 | * by the parent rbd_dev for its own identity (i.e., the structure | |
132 | * is shared between the parent and child). | |
133 | * | |
134 | * Since these structures are populated once, during the discovery | |
135 | * phase of image construction, they are effectively immutable so | |
136 | * we make no effort to synchronize access to them. | |
137 | * | |
138 | * Note that code herein does not assume the image name is known (it | |
139 | * could be a null pointer). | |
0d7dbfce AE |
140 | */ |
141 | struct rbd_spec { | |
142 | u64 pool_id; | |
ecb4dc22 | 143 | const char *pool_name; |
0d7dbfce | 144 | |
ecb4dc22 AE |
145 | const char *image_id; |
146 | const char *image_name; | |
0d7dbfce AE |
147 | |
148 | u64 snap_id; | |
ecb4dc22 | 149 | const char *snap_name; |
0d7dbfce AE |
150 | |
151 | struct kref kref; | |
152 | }; | |
153 | ||
602adf40 | 154 | /* |
f0f8cef5 | 155 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
156 | */ |
157 | struct rbd_client { | |
158 | struct ceph_client *client; | |
159 | struct kref kref; | |
160 | struct list_head node; | |
161 | }; | |
162 | ||
bf0d5f50 AE |
163 | struct rbd_img_request; |
164 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
165 | ||
166 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
167 | ||
168 | struct rbd_obj_request; | |
169 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
170 | ||
9969ebc5 AE |
171 | enum obj_request_type { |
172 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
173 | }; | |
bf0d5f50 | 174 | |
926f9b3f AE |
175 | enum obj_req_flags { |
176 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 177 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
178 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
179 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
180 | }; |
181 | ||
bf0d5f50 AE |
182 | struct rbd_obj_request { |
183 | const char *object_name; | |
184 | u64 offset; /* object start byte */ | |
185 | u64 length; /* bytes from offset */ | |
926f9b3f | 186 | unsigned long flags; |
bf0d5f50 | 187 | |
c5b5ef6c AE |
188 | /* |
189 | * An object request associated with an image will have its | |
190 | * img_data flag set; a standalone object request will not. | |
191 | * | |
192 | * A standalone object request will have which == BAD_WHICH | |
193 | * and a null obj_request pointer. | |
194 | * | |
195 | * An object request initiated in support of a layered image | |
196 | * object (to check for its existence before a write) will | |
197 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
198 | * | |
199 | * Finally, an object request for rbd image data will have | |
200 | * which != BAD_WHICH, and will have a non-null img_request | |
201 | * pointer. The value of which will be in the range | |
202 | * 0..(img_request->obj_request_count-1). | |
203 | */ | |
204 | union { | |
205 | struct rbd_obj_request *obj_request; /* STAT op */ | |
206 | struct { | |
207 | struct rbd_img_request *img_request; | |
208 | u64 img_offset; | |
209 | /* links for img_request->obj_requests list */ | |
210 | struct list_head links; | |
211 | }; | |
212 | }; | |
bf0d5f50 AE |
213 | u32 which; /* posn image request list */ |
214 | ||
215 | enum obj_request_type type; | |
788e2df3 AE |
216 | union { |
217 | struct bio *bio_list; | |
218 | struct { | |
219 | struct page **pages; | |
220 | u32 page_count; | |
221 | }; | |
222 | }; | |
0eefd470 | 223 | struct page **copyup_pages; |
bf0d5f50 AE |
224 | |
225 | struct ceph_osd_request *osd_req; | |
226 | ||
227 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 228 | int result; |
bf0d5f50 AE |
229 | |
230 | rbd_obj_callback_t callback; | |
788e2df3 | 231 | struct completion completion; |
bf0d5f50 AE |
232 | |
233 | struct kref kref; | |
234 | }; | |
235 | ||
0c425248 | 236 | enum img_req_flags { |
9849e986 AE |
237 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
238 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 239 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
240 | }; |
241 | ||
bf0d5f50 | 242 | struct rbd_img_request { |
bf0d5f50 AE |
243 | struct rbd_device *rbd_dev; |
244 | u64 offset; /* starting image byte offset */ | |
245 | u64 length; /* byte count from offset */ | |
0c425248 | 246 | unsigned long flags; |
bf0d5f50 | 247 | union { |
9849e986 | 248 | u64 snap_id; /* for reads */ |
bf0d5f50 | 249 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
250 | }; |
251 | union { | |
252 | struct request *rq; /* block request */ | |
253 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 254 | }; |
3d7efd18 | 255 | struct page **copyup_pages; |
bf0d5f50 AE |
256 | spinlock_t completion_lock;/* protects next_completion */ |
257 | u32 next_completion; | |
258 | rbd_img_callback_t callback; | |
55f27e09 | 259 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 260 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
261 | |
262 | u32 obj_request_count; | |
263 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
264 | ||
265 | struct kref kref; | |
266 | }; | |
267 | ||
268 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 269 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 270 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 271 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 272 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 273 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 274 | |
dfc5606d | 275 | struct rbd_snap { |
dfc5606d | 276 | const char *name; |
3591538f | 277 | u64 size; |
dfc5606d YS |
278 | struct list_head node; |
279 | u64 id; | |
34b13184 | 280 | u64 features; |
dfc5606d YS |
281 | }; |
282 | ||
f84344f3 | 283 | struct rbd_mapping { |
99c1f08f | 284 | u64 size; |
34b13184 | 285 | u64 features; |
f84344f3 AE |
286 | bool read_only; |
287 | }; | |
288 | ||
602adf40 YS |
289 | /* |
290 | * a single device | |
291 | */ | |
292 | struct rbd_device { | |
de71a297 | 293 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
294 | |
295 | int major; /* blkdev assigned major */ | |
296 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 297 | |
a30b71b9 | 298 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
299 | struct rbd_client *rbd_client; |
300 | ||
301 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
302 | ||
b82d167b | 303 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
304 | |
305 | struct rbd_image_header header; | |
b82d167b | 306 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 307 | struct rbd_spec *spec; |
602adf40 | 308 | |
0d7dbfce | 309 | char *header_name; |
971f839a | 310 | |
0903e875 AE |
311 | struct ceph_file_layout layout; |
312 | ||
59c2be1e | 313 | struct ceph_osd_event *watch_event; |
975241af | 314 | struct rbd_obj_request *watch_request; |
59c2be1e | 315 | |
86b00e0d AE |
316 | struct rbd_spec *parent_spec; |
317 | u64 parent_overlap; | |
2f82ee54 | 318 | struct rbd_device *parent; |
86b00e0d | 319 | |
c666601a JD |
320 | /* protects updating the header */ |
321 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
322 | |
323 | struct rbd_mapping mapping; | |
602adf40 YS |
324 | |
325 | struct list_head node; | |
dfc5606d YS |
326 | |
327 | /* list of snapshots */ | |
328 | struct list_head snaps; | |
329 | ||
330 | /* sysfs related */ | |
331 | struct device dev; | |
b82d167b | 332 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
333 | }; |
334 | ||
b82d167b AE |
335 | /* |
336 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
337 | * rbd_dev->lock is used to protect access. | |
338 | * | |
339 | * Currently, only the "removing" flag (which is coupled with the | |
340 | * "open_count" field) requires atomic access. | |
341 | */ | |
6d292906 AE |
342 | enum rbd_dev_flags { |
343 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 344 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
345 | }; |
346 | ||
602adf40 | 347 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 348 | |
602adf40 | 349 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
350 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
351 | ||
432b8587 AE |
352 | static LIST_HEAD(rbd_client_list); /* clients */ |
353 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 354 | |
3d7efd18 AE |
355 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
356 | ||
304f6808 | 357 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
304f6808 | 358 | |
200a6a8b | 359 | static void rbd_dev_device_release(struct device *dev); |
6087b51b | 360 | static void rbd_snap_destroy(struct rbd_snap *snap); |
dfc5606d | 361 | |
f0f8cef5 AE |
362 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
363 | size_t count); | |
364 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
365 | size_t count); | |
71f293e2 | 366 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev); |
f0f8cef5 AE |
367 | |
368 | static struct bus_attribute rbd_bus_attrs[] = { | |
369 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
370 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
371 | __ATTR_NULL | |
372 | }; | |
373 | ||
374 | static struct bus_type rbd_bus_type = { | |
375 | .name = "rbd", | |
376 | .bus_attrs = rbd_bus_attrs, | |
377 | }; | |
378 | ||
379 | static void rbd_root_dev_release(struct device *dev) | |
380 | { | |
381 | } | |
382 | ||
383 | static struct device rbd_root_dev = { | |
384 | .init_name = "rbd", | |
385 | .release = rbd_root_dev_release, | |
386 | }; | |
387 | ||
06ecc6cb AE |
388 | static __printf(2, 3) |
389 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
390 | { | |
391 | struct va_format vaf; | |
392 | va_list args; | |
393 | ||
394 | va_start(args, fmt); | |
395 | vaf.fmt = fmt; | |
396 | vaf.va = &args; | |
397 | ||
398 | if (!rbd_dev) | |
399 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
400 | else if (rbd_dev->disk) | |
401 | printk(KERN_WARNING "%s: %s: %pV\n", | |
402 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
403 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
404 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
405 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
406 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
407 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
408 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
409 | else /* punt */ | |
410 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
411 | RBD_DRV_NAME, rbd_dev, &vaf); | |
412 | va_end(args); | |
413 | } | |
414 | ||
aafb230e AE |
415 | #ifdef RBD_DEBUG |
416 | #define rbd_assert(expr) \ | |
417 | if (unlikely(!(expr))) { \ | |
418 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
419 | "at line %d:\n\n" \ | |
420 | "\trbd_assert(%s);\n\n", \ | |
421 | __func__, __LINE__, #expr); \ | |
422 | BUG(); \ | |
423 | } | |
424 | #else /* !RBD_DEBUG */ | |
425 | # define rbd_assert(expr) ((void) 0) | |
426 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 427 | |
b454e36d | 428 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
429 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
430 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 431 | |
cc4a38bd AE |
432 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
433 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev); | |
59c2be1e | 434 | |
602adf40 YS |
435 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
436 | { | |
f0f8cef5 | 437 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 438 | bool removing = false; |
602adf40 | 439 | |
f84344f3 | 440 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
441 | return -EROFS; |
442 | ||
a14ea269 | 443 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
444 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
445 | removing = true; | |
446 | else | |
447 | rbd_dev->open_count++; | |
a14ea269 | 448 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
449 | if (removing) |
450 | return -ENOENT; | |
451 | ||
42382b70 | 452 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 453 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 454 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 455 | mutex_unlock(&ctl_mutex); |
340c7a2b | 456 | |
602adf40 YS |
457 | return 0; |
458 | } | |
459 | ||
dfc5606d YS |
460 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
461 | { | |
462 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
463 | unsigned long open_count_before; |
464 | ||
a14ea269 | 465 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 466 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 467 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 468 | rbd_assert(open_count_before > 0); |
dfc5606d | 469 | |
42382b70 | 470 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 471 | put_device(&rbd_dev->dev); |
42382b70 | 472 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
473 | |
474 | return 0; | |
475 | } | |
476 | ||
602adf40 YS |
477 | static const struct block_device_operations rbd_bd_ops = { |
478 | .owner = THIS_MODULE, | |
479 | .open = rbd_open, | |
dfc5606d | 480 | .release = rbd_release, |
602adf40 YS |
481 | }; |
482 | ||
483 | /* | |
484 | * Initialize an rbd client instance. | |
43ae4701 | 485 | * We own *ceph_opts. |
602adf40 | 486 | */ |
f8c38929 | 487 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
488 | { |
489 | struct rbd_client *rbdc; | |
490 | int ret = -ENOMEM; | |
491 | ||
37206ee5 | 492 | dout("%s:\n", __func__); |
602adf40 YS |
493 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
494 | if (!rbdc) | |
495 | goto out_opt; | |
496 | ||
497 | kref_init(&rbdc->kref); | |
498 | INIT_LIST_HEAD(&rbdc->node); | |
499 | ||
bc534d86 AE |
500 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
501 | ||
43ae4701 | 502 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 503 | if (IS_ERR(rbdc->client)) |
bc534d86 | 504 | goto out_mutex; |
43ae4701 | 505 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
506 | |
507 | ret = ceph_open_session(rbdc->client); | |
508 | if (ret < 0) | |
509 | goto out_err; | |
510 | ||
432b8587 | 511 | spin_lock(&rbd_client_list_lock); |
602adf40 | 512 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 513 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 514 | |
bc534d86 | 515 | mutex_unlock(&ctl_mutex); |
37206ee5 | 516 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 517 | |
602adf40 YS |
518 | return rbdc; |
519 | ||
520 | out_err: | |
521 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
522 | out_mutex: |
523 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
524 | kfree(rbdc); |
525 | out_opt: | |
43ae4701 AE |
526 | if (ceph_opts) |
527 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
528 | dout("%s: error %d\n", __func__, ret); |
529 | ||
28f259b7 | 530 | return ERR_PTR(ret); |
602adf40 YS |
531 | } |
532 | ||
2f82ee54 AE |
533 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
534 | { | |
535 | kref_get(&rbdc->kref); | |
536 | ||
537 | return rbdc; | |
538 | } | |
539 | ||
602adf40 | 540 | /* |
1f7ba331 AE |
541 | * Find a ceph client with specific addr and configuration. If |
542 | * found, bump its reference count. | |
602adf40 | 543 | */ |
1f7ba331 | 544 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
545 | { |
546 | struct rbd_client *client_node; | |
1f7ba331 | 547 | bool found = false; |
602adf40 | 548 | |
43ae4701 | 549 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
550 | return NULL; |
551 | ||
1f7ba331 AE |
552 | spin_lock(&rbd_client_list_lock); |
553 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
554 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
555 | __rbd_get_client(client_node); |
556 | ||
1f7ba331 AE |
557 | found = true; |
558 | break; | |
559 | } | |
560 | } | |
561 | spin_unlock(&rbd_client_list_lock); | |
562 | ||
563 | return found ? client_node : NULL; | |
602adf40 YS |
564 | } |
565 | ||
59c2be1e YS |
566 | /* |
567 | * mount options | |
568 | */ | |
569 | enum { | |
59c2be1e YS |
570 | Opt_last_int, |
571 | /* int args above */ | |
572 | Opt_last_string, | |
573 | /* string args above */ | |
cc0538b6 AE |
574 | Opt_read_only, |
575 | Opt_read_write, | |
576 | /* Boolean args above */ | |
577 | Opt_last_bool, | |
59c2be1e YS |
578 | }; |
579 | ||
43ae4701 | 580 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
581 | /* int args above */ |
582 | /* string args above */ | |
be466c1c | 583 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
584 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
585 | {Opt_read_write, "read_write"}, | |
586 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
587 | /* Boolean args above */ | |
59c2be1e YS |
588 | {-1, NULL} |
589 | }; | |
590 | ||
98571b5a AE |
591 | struct rbd_options { |
592 | bool read_only; | |
593 | }; | |
594 | ||
595 | #define RBD_READ_ONLY_DEFAULT false | |
596 | ||
59c2be1e YS |
597 | static int parse_rbd_opts_token(char *c, void *private) |
598 | { | |
43ae4701 | 599 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
600 | substring_t argstr[MAX_OPT_ARGS]; |
601 | int token, intval, ret; | |
602 | ||
43ae4701 | 603 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
604 | if (token < 0) |
605 | return -EINVAL; | |
606 | ||
607 | if (token < Opt_last_int) { | |
608 | ret = match_int(&argstr[0], &intval); | |
609 | if (ret < 0) { | |
610 | pr_err("bad mount option arg (not int) " | |
611 | "at '%s'\n", c); | |
612 | return ret; | |
613 | } | |
614 | dout("got int token %d val %d\n", token, intval); | |
615 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
616 | dout("got string token %d val %s\n", token, | |
617 | argstr[0].from); | |
cc0538b6 AE |
618 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
619 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
620 | } else { |
621 | dout("got token %d\n", token); | |
622 | } | |
623 | ||
624 | switch (token) { | |
cc0538b6 AE |
625 | case Opt_read_only: |
626 | rbd_opts->read_only = true; | |
627 | break; | |
628 | case Opt_read_write: | |
629 | rbd_opts->read_only = false; | |
630 | break; | |
59c2be1e | 631 | default: |
aafb230e AE |
632 | rbd_assert(false); |
633 | break; | |
59c2be1e YS |
634 | } |
635 | return 0; | |
636 | } | |
637 | ||
602adf40 YS |
638 | /* |
639 | * Get a ceph client with specific addr and configuration, if one does | |
640 | * not exist create it. | |
641 | */ | |
9d3997fd | 642 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 643 | { |
f8c38929 | 644 | struct rbd_client *rbdc; |
59c2be1e | 645 | |
1f7ba331 | 646 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 647 | if (rbdc) /* using an existing client */ |
43ae4701 | 648 | ceph_destroy_options(ceph_opts); |
9d3997fd | 649 | else |
f8c38929 | 650 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 651 | |
9d3997fd | 652 | return rbdc; |
602adf40 YS |
653 | } |
654 | ||
655 | /* | |
656 | * Destroy ceph client | |
d23a4b3f | 657 | * |
432b8587 | 658 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
659 | */ |
660 | static void rbd_client_release(struct kref *kref) | |
661 | { | |
662 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
663 | ||
37206ee5 | 664 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 665 | spin_lock(&rbd_client_list_lock); |
602adf40 | 666 | list_del(&rbdc->node); |
cd9d9f5d | 667 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
668 | |
669 | ceph_destroy_client(rbdc->client); | |
670 | kfree(rbdc); | |
671 | } | |
672 | ||
673 | /* | |
674 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
675 | * it. | |
676 | */ | |
9d3997fd | 677 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 678 | { |
c53d5893 AE |
679 | if (rbdc) |
680 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
681 | } |
682 | ||
a30b71b9 AE |
683 | static bool rbd_image_format_valid(u32 image_format) |
684 | { | |
685 | return image_format == 1 || image_format == 2; | |
686 | } | |
687 | ||
8e94af8e AE |
688 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
689 | { | |
103a150f AE |
690 | size_t size; |
691 | u32 snap_count; | |
692 | ||
693 | /* The header has to start with the magic rbd header text */ | |
694 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
695 | return false; | |
696 | ||
db2388b6 AE |
697 | /* The bio layer requires at least sector-sized I/O */ |
698 | ||
699 | if (ondisk->options.order < SECTOR_SHIFT) | |
700 | return false; | |
701 | ||
702 | /* If we use u64 in a few spots we may be able to loosen this */ | |
703 | ||
704 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
705 | return false; | |
706 | ||
103a150f AE |
707 | /* |
708 | * The size of a snapshot header has to fit in a size_t, and | |
709 | * that limits the number of snapshots. | |
710 | */ | |
711 | snap_count = le32_to_cpu(ondisk->snap_count); | |
712 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
713 | if (snap_count > size / sizeof (__le64)) | |
714 | return false; | |
715 | ||
716 | /* | |
717 | * Not only that, but the size of the entire the snapshot | |
718 | * header must also be representable in a size_t. | |
719 | */ | |
720 | size -= snap_count * sizeof (__le64); | |
721 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
722 | return false; | |
723 | ||
724 | return true; | |
8e94af8e AE |
725 | } |
726 | ||
602adf40 YS |
727 | /* |
728 | * Create a new header structure, translate header format from the on-disk | |
729 | * header. | |
730 | */ | |
731 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 732 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 733 | { |
ccece235 | 734 | u32 snap_count; |
58c17b0e | 735 | size_t len; |
d2bb24e5 | 736 | size_t size; |
621901d6 | 737 | u32 i; |
602adf40 | 738 | |
6a52325f AE |
739 | memset(header, 0, sizeof (*header)); |
740 | ||
103a150f AE |
741 | snap_count = le32_to_cpu(ondisk->snap_count); |
742 | ||
58c17b0e AE |
743 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
744 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 745 | if (!header->object_prefix) |
602adf40 | 746 | return -ENOMEM; |
58c17b0e AE |
747 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
748 | header->object_prefix[len] = '\0'; | |
00f1f36f | 749 | |
602adf40 | 750 | if (snap_count) { |
f785cc1d AE |
751 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
752 | ||
621901d6 AE |
753 | /* Save a copy of the snapshot names */ |
754 | ||
f785cc1d AE |
755 | if (snap_names_len > (u64) SIZE_MAX) |
756 | return -EIO; | |
757 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 758 | if (!header->snap_names) |
6a52325f | 759 | goto out_err; |
f785cc1d AE |
760 | /* |
761 | * Note that rbd_dev_v1_header_read() guarantees | |
762 | * the ondisk buffer we're working with has | |
763 | * snap_names_len bytes beyond the end of the | |
764 | * snapshot id array, this memcpy() is safe. | |
765 | */ | |
766 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
767 | snap_names_len); | |
6a52325f | 768 | |
621901d6 AE |
769 | /* Record each snapshot's size */ |
770 | ||
d2bb24e5 AE |
771 | size = snap_count * sizeof (*header->snap_sizes); |
772 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 773 | if (!header->snap_sizes) |
6a52325f | 774 | goto out_err; |
621901d6 AE |
775 | for (i = 0; i < snap_count; i++) |
776 | header->snap_sizes[i] = | |
777 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 YS |
778 | } else { |
779 | header->snap_names = NULL; | |
780 | header->snap_sizes = NULL; | |
781 | } | |
849b4260 | 782 | |
34b13184 | 783 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
784 | header->obj_order = ondisk->options.order; |
785 | header->crypt_type = ondisk->options.crypt_type; | |
786 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 787 | |
621901d6 AE |
788 | /* Allocate and fill in the snapshot context */ |
789 | ||
f84344f3 | 790 | header->image_size = le64_to_cpu(ondisk->image_size); |
468521c1 | 791 | |
812164f8 | 792 | header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
6a52325f AE |
793 | if (!header->snapc) |
794 | goto out_err; | |
505cbb9b | 795 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
621901d6 | 796 | for (i = 0; i < snap_count; i++) |
468521c1 | 797 | header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id); |
602adf40 YS |
798 | |
799 | return 0; | |
800 | ||
6a52325f | 801 | out_err: |
849b4260 | 802 | kfree(header->snap_sizes); |
ccece235 | 803 | header->snap_sizes = NULL; |
602adf40 | 804 | kfree(header->snap_names); |
ccece235 | 805 | header->snap_names = NULL; |
6a52325f AE |
806 | kfree(header->object_prefix); |
807 | header->object_prefix = NULL; | |
ccece235 | 808 | |
00f1f36f | 809 | return -ENOMEM; |
602adf40 YS |
810 | } |
811 | ||
9e15b77d AE |
812 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
813 | { | |
814 | struct rbd_snap *snap; | |
815 | ||
816 | if (snap_id == CEPH_NOSNAP) | |
817 | return RBD_SNAP_HEAD_NAME; | |
818 | ||
819 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
820 | if (snap_id == snap->id) | |
821 | return snap->name; | |
822 | ||
823 | return NULL; | |
824 | } | |
825 | ||
8b0241f8 AE |
826 | static struct rbd_snap *snap_by_name(struct rbd_device *rbd_dev, |
827 | const char *snap_name) | |
602adf40 | 828 | { |
e86924a8 | 829 | struct rbd_snap *snap; |
602adf40 | 830 | |
8b0241f8 AE |
831 | list_for_each_entry(snap, &rbd_dev->snaps, node) |
832 | if (!strcmp(snap_name, snap->name)) | |
833 | return snap; | |
e86924a8 | 834 | |
8b0241f8 | 835 | return NULL; |
602adf40 YS |
836 | } |
837 | ||
d1cf5788 | 838 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) |
602adf40 | 839 | { |
0d7dbfce | 840 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 841 | sizeof (RBD_SNAP_HEAD_NAME))) { |
99c1f08f | 842 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 843 | rbd_dev->mapping.features = rbd_dev->header.features; |
602adf40 | 844 | } else { |
8b0241f8 AE |
845 | struct rbd_snap *snap; |
846 | ||
847 | snap = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); | |
848 | if (!snap) | |
849 | return -ENOENT; | |
8b0241f8 AE |
850 | rbd_dev->mapping.size = snap->size; |
851 | rbd_dev->mapping.features = snap->features; | |
f84344f3 | 852 | rbd_dev->mapping.read_only = true; |
602adf40 | 853 | } |
6d292906 | 854 | |
8b0241f8 | 855 | return 0; |
602adf40 YS |
856 | } |
857 | ||
d1cf5788 AE |
858 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
859 | { | |
860 | rbd_dev->mapping.size = 0; | |
861 | rbd_dev->mapping.features = 0; | |
862 | rbd_dev->mapping.read_only = true; | |
863 | } | |
864 | ||
200a6a8b AE |
865 | static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev) |
866 | { | |
867 | rbd_dev->mapping.size = 0; | |
868 | rbd_dev->mapping.features = 0; | |
869 | rbd_dev->mapping.read_only = true; | |
870 | } | |
871 | ||
98571b5a | 872 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 873 | { |
65ccfe21 AE |
874 | char *name; |
875 | u64 segment; | |
876 | int ret; | |
602adf40 | 877 | |
2fd82b9e | 878 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
879 | if (!name) |
880 | return NULL; | |
881 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 882 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 883 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 884 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
885 | pr_err("error formatting segment name for #%llu (%d)\n", |
886 | segment, ret); | |
887 | kfree(name); | |
888 | name = NULL; | |
889 | } | |
602adf40 | 890 | |
65ccfe21 AE |
891 | return name; |
892 | } | |
602adf40 | 893 | |
65ccfe21 AE |
894 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
895 | { | |
896 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 897 | |
65ccfe21 AE |
898 | return offset & (segment_size - 1); |
899 | } | |
900 | ||
901 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
902 | u64 offset, u64 length) | |
903 | { | |
904 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
905 | ||
906 | offset &= segment_size - 1; | |
907 | ||
aafb230e | 908 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
909 | if (offset + length > segment_size) |
910 | length = segment_size - offset; | |
911 | ||
912 | return length; | |
602adf40 YS |
913 | } |
914 | ||
029bcbd8 JD |
915 | /* |
916 | * returns the size of an object in the image | |
917 | */ | |
918 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
919 | { | |
920 | return 1 << header->obj_order; | |
921 | } | |
922 | ||
602adf40 YS |
923 | /* |
924 | * bio helpers | |
925 | */ | |
926 | ||
927 | static void bio_chain_put(struct bio *chain) | |
928 | { | |
929 | struct bio *tmp; | |
930 | ||
931 | while (chain) { | |
932 | tmp = chain; | |
933 | chain = chain->bi_next; | |
934 | bio_put(tmp); | |
935 | } | |
936 | } | |
937 | ||
938 | /* | |
939 | * zeros a bio chain, starting at specific offset | |
940 | */ | |
941 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
942 | { | |
943 | struct bio_vec *bv; | |
944 | unsigned long flags; | |
945 | void *buf; | |
946 | int i; | |
947 | int pos = 0; | |
948 | ||
949 | while (chain) { | |
950 | bio_for_each_segment(bv, chain, i) { | |
951 | if (pos + bv->bv_len > start_ofs) { | |
952 | int remainder = max(start_ofs - pos, 0); | |
953 | buf = bvec_kmap_irq(bv, &flags); | |
954 | memset(buf + remainder, 0, | |
955 | bv->bv_len - remainder); | |
85b5aaa6 | 956 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
957 | } |
958 | pos += bv->bv_len; | |
959 | } | |
960 | ||
961 | chain = chain->bi_next; | |
962 | } | |
963 | } | |
964 | ||
b9434c5b AE |
965 | /* |
966 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
967 | * starting at the given byte offset from the start of the array and | |
968 | * continuing up to the given end offset. The pages array is | |
969 | * assumed to be big enough to hold all bytes up to the end. | |
970 | */ | |
971 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
972 | { | |
973 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
974 | ||
975 | rbd_assert(end > offset); | |
976 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
977 | while (offset < end) { | |
978 | size_t page_offset; | |
979 | size_t length; | |
980 | unsigned long flags; | |
981 | void *kaddr; | |
982 | ||
983 | page_offset = (size_t)(offset & ~PAGE_MASK); | |
984 | length = min(PAGE_SIZE - page_offset, (size_t)(end - offset)); | |
985 | local_irq_save(flags); | |
986 | kaddr = kmap_atomic(*page); | |
987 | memset(kaddr + page_offset, 0, length); | |
988 | kunmap_atomic(kaddr); | |
989 | local_irq_restore(flags); | |
990 | ||
991 | offset += length; | |
992 | page++; | |
993 | } | |
994 | } | |
995 | ||
602adf40 | 996 | /* |
f7760dad AE |
997 | * Clone a portion of a bio, starting at the given byte offset |
998 | * and continuing for the number of bytes indicated. | |
602adf40 | 999 | */ |
f7760dad AE |
1000 | static struct bio *bio_clone_range(struct bio *bio_src, |
1001 | unsigned int offset, | |
1002 | unsigned int len, | |
1003 | gfp_t gfpmask) | |
602adf40 | 1004 | { |
f7760dad AE |
1005 | struct bio_vec *bv; |
1006 | unsigned int resid; | |
1007 | unsigned short idx; | |
1008 | unsigned int voff; | |
1009 | unsigned short end_idx; | |
1010 | unsigned short vcnt; | |
1011 | struct bio *bio; | |
1012 | ||
1013 | /* Handle the easy case for the caller */ | |
1014 | ||
1015 | if (!offset && len == bio_src->bi_size) | |
1016 | return bio_clone(bio_src, gfpmask); | |
1017 | ||
1018 | if (WARN_ON_ONCE(!len)) | |
1019 | return NULL; | |
1020 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
1021 | return NULL; | |
1022 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
1023 | return NULL; | |
1024 | ||
1025 | /* Find first affected segment... */ | |
1026 | ||
1027 | resid = offset; | |
1028 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
1029 | if (resid < bv->bv_len) | |
1030 | break; | |
1031 | resid -= bv->bv_len; | |
602adf40 | 1032 | } |
f7760dad | 1033 | voff = resid; |
602adf40 | 1034 | |
f7760dad | 1035 | /* ...and the last affected segment */ |
602adf40 | 1036 | |
f7760dad AE |
1037 | resid += len; |
1038 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
1039 | if (resid <= bv->bv_len) | |
1040 | break; | |
1041 | resid -= bv->bv_len; | |
1042 | } | |
1043 | vcnt = end_idx - idx + 1; | |
1044 | ||
1045 | /* Build the clone */ | |
1046 | ||
1047 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
1048 | if (!bio) | |
1049 | return NULL; /* ENOMEM */ | |
602adf40 | 1050 | |
f7760dad AE |
1051 | bio->bi_bdev = bio_src->bi_bdev; |
1052 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
1053 | bio->bi_rw = bio_src->bi_rw; | |
1054 | bio->bi_flags |= 1 << BIO_CLONED; | |
1055 | ||
1056 | /* | |
1057 | * Copy over our part of the bio_vec, then update the first | |
1058 | * and last (or only) entries. | |
1059 | */ | |
1060 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
1061 | vcnt * sizeof (struct bio_vec)); | |
1062 | bio->bi_io_vec[0].bv_offset += voff; | |
1063 | if (vcnt > 1) { | |
1064 | bio->bi_io_vec[0].bv_len -= voff; | |
1065 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
1066 | } else { | |
1067 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
1068 | } |
1069 | ||
f7760dad AE |
1070 | bio->bi_vcnt = vcnt; |
1071 | bio->bi_size = len; | |
1072 | bio->bi_idx = 0; | |
1073 | ||
1074 | return bio; | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * Clone a portion of a bio chain, starting at the given byte offset | |
1079 | * into the first bio in the source chain and continuing for the | |
1080 | * number of bytes indicated. The result is another bio chain of | |
1081 | * exactly the given length, or a null pointer on error. | |
1082 | * | |
1083 | * The bio_src and offset parameters are both in-out. On entry they | |
1084 | * refer to the first source bio and the offset into that bio where | |
1085 | * the start of data to be cloned is located. | |
1086 | * | |
1087 | * On return, bio_src is updated to refer to the bio in the source | |
1088 | * chain that contains first un-cloned byte, and *offset will | |
1089 | * contain the offset of that byte within that bio. | |
1090 | */ | |
1091 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1092 | unsigned int *offset, | |
1093 | unsigned int len, | |
1094 | gfp_t gfpmask) | |
1095 | { | |
1096 | struct bio *bi = *bio_src; | |
1097 | unsigned int off = *offset; | |
1098 | struct bio *chain = NULL; | |
1099 | struct bio **end; | |
1100 | ||
1101 | /* Build up a chain of clone bios up to the limit */ | |
1102 | ||
1103 | if (!bi || off >= bi->bi_size || !len) | |
1104 | return NULL; /* Nothing to clone */ | |
602adf40 | 1105 | |
f7760dad AE |
1106 | end = &chain; |
1107 | while (len) { | |
1108 | unsigned int bi_size; | |
1109 | struct bio *bio; | |
1110 | ||
f5400b7a AE |
1111 | if (!bi) { |
1112 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1113 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1114 | } |
f7760dad AE |
1115 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1116 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1117 | if (!bio) | |
1118 | goto out_err; /* ENOMEM */ | |
1119 | ||
1120 | *end = bio; | |
1121 | end = &bio->bi_next; | |
602adf40 | 1122 | |
f7760dad AE |
1123 | off += bi_size; |
1124 | if (off == bi->bi_size) { | |
1125 | bi = bi->bi_next; | |
1126 | off = 0; | |
1127 | } | |
1128 | len -= bi_size; | |
1129 | } | |
1130 | *bio_src = bi; | |
1131 | *offset = off; | |
1132 | ||
1133 | return chain; | |
1134 | out_err: | |
1135 | bio_chain_put(chain); | |
602adf40 | 1136 | |
602adf40 YS |
1137 | return NULL; |
1138 | } | |
1139 | ||
926f9b3f AE |
1140 | /* |
1141 | * The default/initial value for all object request flags is 0. For | |
1142 | * each flag, once its value is set to 1 it is never reset to 0 | |
1143 | * again. | |
1144 | */ | |
57acbaa7 | 1145 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1146 | { |
57acbaa7 | 1147 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1148 | struct rbd_device *rbd_dev; |
1149 | ||
57acbaa7 AE |
1150 | rbd_dev = obj_request->img_request->rbd_dev; |
1151 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1152 | obj_request); |
1153 | } | |
1154 | } | |
1155 | ||
57acbaa7 | 1156 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1157 | { |
1158 | smp_mb(); | |
57acbaa7 | 1159 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1160 | } |
1161 | ||
57acbaa7 | 1162 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1163 | { |
57acbaa7 AE |
1164 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1165 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1166 | |
57acbaa7 AE |
1167 | if (obj_request_img_data_test(obj_request)) |
1168 | rbd_dev = obj_request->img_request->rbd_dev; | |
1169 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1170 | obj_request); |
1171 | } | |
1172 | } | |
1173 | ||
57acbaa7 | 1174 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1175 | { |
1176 | smp_mb(); | |
57acbaa7 | 1177 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1178 | } |
1179 | ||
5679c59f AE |
1180 | /* |
1181 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1182 | * flag. The latter is set based on the "exists" value provided. | |
1183 | * | |
1184 | * Note that for our purposes once an object exists it never goes | |
1185 | * away again. It's possible that the response from two existence | |
1186 | * checks are separated by the creation of the target object, and | |
1187 | * the first ("doesn't exist") response arrives *after* the second | |
1188 | * ("does exist"). In that case we ignore the second one. | |
1189 | */ | |
1190 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1191 | bool exists) | |
1192 | { | |
1193 | if (exists) | |
1194 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1195 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1196 | smp_mb(); | |
1197 | } | |
1198 | ||
1199 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1200 | { | |
1201 | smp_mb(); | |
1202 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1203 | } | |
1204 | ||
1205 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1206 | { | |
1207 | smp_mb(); | |
1208 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1209 | } | |
1210 | ||
bf0d5f50 AE |
1211 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1212 | { | |
37206ee5 AE |
1213 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1214 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1215 | kref_get(&obj_request->kref); |
1216 | } | |
1217 | ||
1218 | static void rbd_obj_request_destroy(struct kref *kref); | |
1219 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1220 | { | |
1221 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1222 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1223 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1224 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1225 | } | |
1226 | ||
1227 | static void rbd_img_request_get(struct rbd_img_request *img_request) | |
1228 | { | |
37206ee5 AE |
1229 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1230 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1231 | kref_get(&img_request->kref); |
1232 | } | |
1233 | ||
1234 | static void rbd_img_request_destroy(struct kref *kref); | |
1235 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1236 | { | |
1237 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1238 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1239 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1240 | kref_put(&img_request->kref, rbd_img_request_destroy); |
1241 | } | |
1242 | ||
1243 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1244 | struct rbd_obj_request *obj_request) | |
1245 | { | |
25dcf954 AE |
1246 | rbd_assert(obj_request->img_request == NULL); |
1247 | ||
b155e86c | 1248 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1249 | obj_request->img_request = img_request; |
25dcf954 | 1250 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1251 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1252 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1253 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1254 | img_request->obj_request_count++; |
1255 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1256 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1257 | obj_request->which); | |
bf0d5f50 AE |
1258 | } |
1259 | ||
1260 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1261 | struct rbd_obj_request *obj_request) | |
1262 | { | |
1263 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1264 | |
37206ee5 AE |
1265 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1266 | obj_request->which); | |
bf0d5f50 | 1267 | list_del(&obj_request->links); |
25dcf954 AE |
1268 | rbd_assert(img_request->obj_request_count > 0); |
1269 | img_request->obj_request_count--; | |
1270 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1271 | obj_request->which = BAD_WHICH; | |
6365d33a | 1272 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1273 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1274 | obj_request->img_request = NULL; |
25dcf954 | 1275 | obj_request->callback = NULL; |
bf0d5f50 AE |
1276 | rbd_obj_request_put(obj_request); |
1277 | } | |
1278 | ||
1279 | static bool obj_request_type_valid(enum obj_request_type type) | |
1280 | { | |
1281 | switch (type) { | |
9969ebc5 | 1282 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1283 | case OBJ_REQUEST_BIO: |
788e2df3 | 1284 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1285 | return true; |
1286 | default: | |
1287 | return false; | |
1288 | } | |
1289 | } | |
1290 | ||
bf0d5f50 AE |
1291 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1292 | struct rbd_obj_request *obj_request) | |
1293 | { | |
37206ee5 AE |
1294 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1295 | ||
bf0d5f50 AE |
1296 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1297 | } | |
1298 | ||
1299 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1300 | { | |
55f27e09 | 1301 | |
37206ee5 | 1302 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1303 | |
1304 | /* | |
1305 | * If no error occurred, compute the aggregate transfer | |
1306 | * count for the image request. We could instead use | |
1307 | * atomic64_cmpxchg() to update it as each object request | |
1308 | * completes; not clear which way is better off hand. | |
1309 | */ | |
1310 | if (!img_request->result) { | |
1311 | struct rbd_obj_request *obj_request; | |
1312 | u64 xferred = 0; | |
1313 | ||
1314 | for_each_obj_request(img_request, obj_request) | |
1315 | xferred += obj_request->xferred; | |
1316 | img_request->xferred = xferred; | |
1317 | } | |
1318 | ||
bf0d5f50 AE |
1319 | if (img_request->callback) |
1320 | img_request->callback(img_request); | |
1321 | else | |
1322 | rbd_img_request_put(img_request); | |
1323 | } | |
1324 | ||
788e2df3 AE |
1325 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1326 | ||
1327 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1328 | { | |
37206ee5 AE |
1329 | dout("%s: obj %p\n", __func__, obj_request); |
1330 | ||
788e2df3 AE |
1331 | return wait_for_completion_interruptible(&obj_request->completion); |
1332 | } | |
1333 | ||
0c425248 AE |
1334 | /* |
1335 | * The default/initial value for all image request flags is 0. Each | |
1336 | * is conditionally set to 1 at image request initialization time | |
1337 | * and currently never change thereafter. | |
1338 | */ | |
1339 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1340 | { | |
1341 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1342 | smp_mb(); | |
1343 | } | |
1344 | ||
1345 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1346 | { | |
1347 | smp_mb(); | |
1348 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1349 | } | |
1350 | ||
9849e986 AE |
1351 | static void img_request_child_set(struct rbd_img_request *img_request) |
1352 | { | |
1353 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1354 | smp_mb(); | |
1355 | } | |
1356 | ||
1357 | static bool img_request_child_test(struct rbd_img_request *img_request) | |
1358 | { | |
1359 | smp_mb(); | |
1360 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1361 | } | |
1362 | ||
d0b2e944 AE |
1363 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1364 | { | |
1365 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1366 | smp_mb(); | |
1367 | } | |
1368 | ||
1369 | static bool img_request_layered_test(struct rbd_img_request *img_request) | |
1370 | { | |
1371 | smp_mb(); | |
1372 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1373 | } | |
1374 | ||
6e2a4505 AE |
1375 | static void |
1376 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1377 | { | |
b9434c5b AE |
1378 | u64 xferred = obj_request->xferred; |
1379 | u64 length = obj_request->length; | |
1380 | ||
6e2a4505 AE |
1381 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1382 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1383 | xferred, length); |
6e2a4505 AE |
1384 | /* |
1385 | * ENOENT means a hole in the image. We zero-fill the | |
1386 | * entire length of the request. A short read also implies | |
1387 | * zero-fill to the end of the request. Either way we | |
1388 | * update the xferred count to indicate the whole request | |
1389 | * was satisfied. | |
1390 | */ | |
b9434c5b | 1391 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1392 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1393 | if (obj_request->type == OBJ_REQUEST_BIO) |
1394 | zero_bio_chain(obj_request->bio_list, 0); | |
1395 | else | |
1396 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1397 | obj_request->result = 0; |
b9434c5b AE |
1398 | obj_request->xferred = length; |
1399 | } else if (xferred < length && !obj_request->result) { | |
1400 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1401 | zero_bio_chain(obj_request->bio_list, xferred); | |
1402 | else | |
1403 | zero_pages(obj_request->pages, xferred, length); | |
1404 | obj_request->xferred = length; | |
6e2a4505 AE |
1405 | } |
1406 | obj_request_done_set(obj_request); | |
1407 | } | |
1408 | ||
bf0d5f50 AE |
1409 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1410 | { | |
37206ee5 AE |
1411 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1412 | obj_request->callback); | |
bf0d5f50 AE |
1413 | if (obj_request->callback) |
1414 | obj_request->callback(obj_request); | |
788e2df3 AE |
1415 | else |
1416 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1417 | } |
1418 | ||
c47f9371 | 1419 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1420 | { |
1421 | dout("%s: obj %p\n", __func__, obj_request); | |
1422 | obj_request_done_set(obj_request); | |
1423 | } | |
1424 | ||
c47f9371 | 1425 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1426 | { |
57acbaa7 | 1427 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1428 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1429 | bool layered = false; |
1430 | ||
1431 | if (obj_request_img_data_test(obj_request)) { | |
1432 | img_request = obj_request->img_request; | |
1433 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1434 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1435 | } |
8b3e1a56 AE |
1436 | |
1437 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1438 | obj_request, img_request, obj_request->result, | |
1439 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1440 | if (layered && obj_request->result == -ENOENT && |
1441 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1442 | rbd_img_parent_read(obj_request); |
1443 | else if (img_request) | |
6e2a4505 AE |
1444 | rbd_img_obj_request_read_callback(obj_request); |
1445 | else | |
1446 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1447 | } |
1448 | ||
c47f9371 | 1449 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1450 | { |
1b83bef2 SW |
1451 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1452 | obj_request->result, obj_request->length); | |
1453 | /* | |
8b3e1a56 AE |
1454 | * There is no such thing as a successful short write. Set |
1455 | * it to our originally-requested length. | |
1b83bef2 SW |
1456 | */ |
1457 | obj_request->xferred = obj_request->length; | |
07741308 | 1458 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1459 | } |
1460 | ||
fbfab539 AE |
1461 | /* |
1462 | * For a simple stat call there's nothing to do. We'll do more if | |
1463 | * this is part of a write sequence for a layered image. | |
1464 | */ | |
c47f9371 | 1465 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1466 | { |
37206ee5 | 1467 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1468 | obj_request_done_set(obj_request); |
1469 | } | |
1470 | ||
bf0d5f50 AE |
1471 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1472 | struct ceph_msg *msg) | |
1473 | { | |
1474 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1475 | u16 opcode; |
1476 | ||
37206ee5 | 1477 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1478 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1479 | if (obj_request_img_data_test(obj_request)) { |
1480 | rbd_assert(obj_request->img_request); | |
1481 | rbd_assert(obj_request->which != BAD_WHICH); | |
1482 | } else { | |
1483 | rbd_assert(obj_request->which == BAD_WHICH); | |
1484 | } | |
bf0d5f50 | 1485 | |
1b83bef2 SW |
1486 | if (osd_req->r_result < 0) |
1487 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1488 | |
0eefd470 | 1489 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1490 | |
c47f9371 AE |
1491 | /* |
1492 | * We support a 64-bit length, but ultimately it has to be | |
1493 | * passed to blk_end_request(), which takes an unsigned int. | |
1494 | */ | |
1b83bef2 | 1495 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1496 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1497 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1498 | switch (opcode) { |
1499 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1500 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1501 | break; |
1502 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1503 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1504 | break; |
fbfab539 | 1505 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1506 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1507 | break; |
36be9a76 | 1508 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1509 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1510 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1511 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1512 | break; |
bf0d5f50 AE |
1513 | default: |
1514 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1515 | obj_request->object_name, (unsigned short) opcode); | |
1516 | break; | |
1517 | } | |
1518 | ||
07741308 | 1519 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1520 | rbd_obj_request_complete(obj_request); |
1521 | } | |
1522 | ||
9d4df01f | 1523 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1524 | { |
1525 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1526 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1527 | u64 snap_id; |
430c28c3 | 1528 | |
8c042b0d | 1529 | rbd_assert(osd_req != NULL); |
430c28c3 | 1530 | |
9d4df01f | 1531 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1532 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1533 | NULL, snap_id, NULL); |
1534 | } | |
1535 | ||
1536 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1537 | { | |
1538 | struct rbd_img_request *img_request = obj_request->img_request; | |
1539 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1540 | struct ceph_snap_context *snapc; | |
1541 | struct timespec mtime = CURRENT_TIME; | |
1542 | ||
1543 | rbd_assert(osd_req != NULL); | |
1544 | ||
1545 | snapc = img_request ? img_request->snapc : NULL; | |
1546 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1547 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1548 | } |
1549 | ||
bf0d5f50 AE |
1550 | static struct ceph_osd_request *rbd_osd_req_create( |
1551 | struct rbd_device *rbd_dev, | |
1552 | bool write_request, | |
430c28c3 | 1553 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1554 | { |
bf0d5f50 AE |
1555 | struct ceph_snap_context *snapc = NULL; |
1556 | struct ceph_osd_client *osdc; | |
1557 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1558 | |
6365d33a AE |
1559 | if (obj_request_img_data_test(obj_request)) { |
1560 | struct rbd_img_request *img_request = obj_request->img_request; | |
1561 | ||
0c425248 AE |
1562 | rbd_assert(write_request == |
1563 | img_request_write_test(img_request)); | |
1564 | if (write_request) | |
bf0d5f50 | 1565 | snapc = img_request->snapc; |
bf0d5f50 AE |
1566 | } |
1567 | ||
1568 | /* Allocate and initialize the request, for the single op */ | |
1569 | ||
1570 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1571 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1572 | if (!osd_req) | |
1573 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1574 | |
430c28c3 | 1575 | if (write_request) |
bf0d5f50 | 1576 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1577 | else |
bf0d5f50 | 1578 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1579 | |
1580 | osd_req->r_callback = rbd_osd_req_callback; | |
1581 | osd_req->r_priv = obj_request; | |
1582 | ||
1583 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1584 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1585 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1586 | ||
1587 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1588 | ||
bf0d5f50 AE |
1589 | return osd_req; |
1590 | } | |
1591 | ||
0eefd470 AE |
1592 | /* |
1593 | * Create a copyup osd request based on the information in the | |
1594 | * object request supplied. A copyup request has two osd ops, | |
1595 | * a copyup method call, and a "normal" write request. | |
1596 | */ | |
1597 | static struct ceph_osd_request * | |
1598 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1599 | { | |
1600 | struct rbd_img_request *img_request; | |
1601 | struct ceph_snap_context *snapc; | |
1602 | struct rbd_device *rbd_dev; | |
1603 | struct ceph_osd_client *osdc; | |
1604 | struct ceph_osd_request *osd_req; | |
1605 | ||
1606 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1607 | img_request = obj_request->img_request; | |
1608 | rbd_assert(img_request); | |
1609 | rbd_assert(img_request_write_test(img_request)); | |
1610 | ||
1611 | /* Allocate and initialize the request, for the two ops */ | |
1612 | ||
1613 | snapc = img_request->snapc; | |
1614 | rbd_dev = img_request->rbd_dev; | |
1615 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1616 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1617 | if (!osd_req) | |
1618 | return NULL; /* ENOMEM */ | |
1619 | ||
1620 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1621 | osd_req->r_callback = rbd_osd_req_callback; | |
1622 | osd_req->r_priv = obj_request; | |
1623 | ||
1624 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1625 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1626 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1627 | ||
1628 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1629 | ||
1630 | return osd_req; | |
1631 | } | |
1632 | ||
1633 | ||
bf0d5f50 AE |
1634 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1635 | { | |
1636 | ceph_osdc_put_request(osd_req); | |
1637 | } | |
1638 | ||
1639 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1640 | ||
1641 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1642 | u64 offset, u64 length, | |
1643 | enum obj_request_type type) | |
1644 | { | |
1645 | struct rbd_obj_request *obj_request; | |
1646 | size_t size; | |
1647 | char *name; | |
1648 | ||
1649 | rbd_assert(obj_request_type_valid(type)); | |
1650 | ||
1651 | size = strlen(object_name) + 1; | |
1652 | obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL); | |
1653 | if (!obj_request) | |
1654 | return NULL; | |
1655 | ||
1656 | name = (char *)(obj_request + 1); | |
1657 | obj_request->object_name = memcpy(name, object_name, size); | |
1658 | obj_request->offset = offset; | |
1659 | obj_request->length = length; | |
926f9b3f | 1660 | obj_request->flags = 0; |
bf0d5f50 AE |
1661 | obj_request->which = BAD_WHICH; |
1662 | obj_request->type = type; | |
1663 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1664 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1665 | kref_init(&obj_request->kref); |
1666 | ||
37206ee5 AE |
1667 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1668 | offset, length, (int)type, obj_request); | |
1669 | ||
bf0d5f50 AE |
1670 | return obj_request; |
1671 | } | |
1672 | ||
1673 | static void rbd_obj_request_destroy(struct kref *kref) | |
1674 | { | |
1675 | struct rbd_obj_request *obj_request; | |
1676 | ||
1677 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1678 | ||
37206ee5 AE |
1679 | dout("%s: obj %p\n", __func__, obj_request); |
1680 | ||
bf0d5f50 AE |
1681 | rbd_assert(obj_request->img_request == NULL); |
1682 | rbd_assert(obj_request->which == BAD_WHICH); | |
1683 | ||
1684 | if (obj_request->osd_req) | |
1685 | rbd_osd_req_destroy(obj_request->osd_req); | |
1686 | ||
1687 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1688 | switch (obj_request->type) { | |
9969ebc5 AE |
1689 | case OBJ_REQUEST_NODATA: |
1690 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1691 | case OBJ_REQUEST_BIO: |
1692 | if (obj_request->bio_list) | |
1693 | bio_chain_put(obj_request->bio_list); | |
1694 | break; | |
788e2df3 AE |
1695 | case OBJ_REQUEST_PAGES: |
1696 | if (obj_request->pages) | |
1697 | ceph_release_page_vector(obj_request->pages, | |
1698 | obj_request->page_count); | |
1699 | break; | |
bf0d5f50 AE |
1700 | } |
1701 | ||
1702 | kfree(obj_request); | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * Caller is responsible for filling in the list of object requests | |
1707 | * that comprises the image request, and the Linux request pointer | |
1708 | * (if there is one). | |
1709 | */ | |
cc344fa1 AE |
1710 | static struct rbd_img_request *rbd_img_request_create( |
1711 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1712 | u64 offset, u64 length, |
9849e986 AE |
1713 | bool write_request, |
1714 | bool child_request) | |
bf0d5f50 AE |
1715 | { |
1716 | struct rbd_img_request *img_request; | |
bf0d5f50 AE |
1717 | |
1718 | img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC); | |
1719 | if (!img_request) | |
1720 | return NULL; | |
1721 | ||
1722 | if (write_request) { | |
1723 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 1724 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 1725 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1726 | } |
1727 | ||
1728 | img_request->rq = NULL; | |
1729 | img_request->rbd_dev = rbd_dev; | |
1730 | img_request->offset = offset; | |
1731 | img_request->length = length; | |
0c425248 AE |
1732 | img_request->flags = 0; |
1733 | if (write_request) { | |
1734 | img_request_write_set(img_request); | |
468521c1 | 1735 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1736 | } else { |
bf0d5f50 | 1737 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1738 | } |
9849e986 AE |
1739 | if (child_request) |
1740 | img_request_child_set(img_request); | |
d0b2e944 AE |
1741 | if (rbd_dev->parent_spec) |
1742 | img_request_layered_set(img_request); | |
bf0d5f50 AE |
1743 | spin_lock_init(&img_request->completion_lock); |
1744 | img_request->next_completion = 0; | |
1745 | img_request->callback = NULL; | |
a5a337d4 | 1746 | img_request->result = 0; |
bf0d5f50 AE |
1747 | img_request->obj_request_count = 0; |
1748 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1749 | kref_init(&img_request->kref); | |
1750 | ||
1751 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1752 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1753 | ||
37206ee5 AE |
1754 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1755 | write_request ? "write" : "read", offset, length, | |
1756 | img_request); | |
1757 | ||
bf0d5f50 AE |
1758 | return img_request; |
1759 | } | |
1760 | ||
1761 | static void rbd_img_request_destroy(struct kref *kref) | |
1762 | { | |
1763 | struct rbd_img_request *img_request; | |
1764 | struct rbd_obj_request *obj_request; | |
1765 | struct rbd_obj_request *next_obj_request; | |
1766 | ||
1767 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1768 | ||
37206ee5 AE |
1769 | dout("%s: img %p\n", __func__, img_request); |
1770 | ||
bf0d5f50 AE |
1771 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1772 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1773 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 1774 | |
0c425248 | 1775 | if (img_request_write_test(img_request)) |
812164f8 | 1776 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 1777 | |
8b3e1a56 AE |
1778 | if (img_request_child_test(img_request)) |
1779 | rbd_obj_request_put(img_request->obj_request); | |
1780 | ||
bf0d5f50 AE |
1781 | kfree(img_request); |
1782 | } | |
1783 | ||
1217857f AE |
1784 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
1785 | { | |
6365d33a | 1786 | struct rbd_img_request *img_request; |
1217857f AE |
1787 | unsigned int xferred; |
1788 | int result; | |
8b3e1a56 | 1789 | bool more; |
1217857f | 1790 | |
6365d33a AE |
1791 | rbd_assert(obj_request_img_data_test(obj_request)); |
1792 | img_request = obj_request->img_request; | |
1793 | ||
1217857f AE |
1794 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
1795 | xferred = (unsigned int)obj_request->xferred; | |
1796 | result = obj_request->result; | |
1797 | if (result) { | |
1798 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1799 | ||
1800 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
1801 | img_request_write_test(img_request) ? "write" : "read", | |
1802 | obj_request->length, obj_request->img_offset, | |
1803 | obj_request->offset); | |
1804 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
1805 | result, xferred); | |
1806 | if (!img_request->result) | |
1807 | img_request->result = result; | |
1808 | } | |
1809 | ||
f1a4739f AE |
1810 | /* Image object requests don't own their page array */ |
1811 | ||
1812 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
1813 | obj_request->pages = NULL; | |
1814 | obj_request->page_count = 0; | |
1815 | } | |
1816 | ||
8b3e1a56 AE |
1817 | if (img_request_child_test(img_request)) { |
1818 | rbd_assert(img_request->obj_request != NULL); | |
1819 | more = obj_request->which < img_request->obj_request_count - 1; | |
1820 | } else { | |
1821 | rbd_assert(img_request->rq != NULL); | |
1822 | more = blk_end_request(img_request->rq, result, xferred); | |
1823 | } | |
1824 | ||
1825 | return more; | |
1217857f AE |
1826 | } |
1827 | ||
2169238d AE |
1828 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
1829 | { | |
1830 | struct rbd_img_request *img_request; | |
1831 | u32 which = obj_request->which; | |
1832 | bool more = true; | |
1833 | ||
6365d33a | 1834 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
1835 | img_request = obj_request->img_request; |
1836 | ||
1837 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
1838 | rbd_assert(img_request != NULL); | |
2169238d AE |
1839 | rbd_assert(img_request->obj_request_count > 0); |
1840 | rbd_assert(which != BAD_WHICH); | |
1841 | rbd_assert(which < img_request->obj_request_count); | |
1842 | rbd_assert(which >= img_request->next_completion); | |
1843 | ||
1844 | spin_lock_irq(&img_request->completion_lock); | |
1845 | if (which != img_request->next_completion) | |
1846 | goto out; | |
1847 | ||
1848 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
1849 | rbd_assert(more); |
1850 | rbd_assert(which < img_request->obj_request_count); | |
1851 | ||
1852 | if (!obj_request_done_test(obj_request)) | |
1853 | break; | |
1217857f | 1854 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
1855 | which++; |
1856 | } | |
1857 | ||
1858 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
1859 | img_request->next_completion = which; | |
1860 | out: | |
1861 | spin_unlock_irq(&img_request->completion_lock); | |
1862 | ||
1863 | if (!more) | |
1864 | rbd_img_request_complete(img_request); | |
1865 | } | |
1866 | ||
f1a4739f AE |
1867 | /* |
1868 | * Split up an image request into one or more object requests, each | |
1869 | * to a different object. The "type" parameter indicates whether | |
1870 | * "data_desc" is the pointer to the head of a list of bio | |
1871 | * structures, or the base of a page array. In either case this | |
1872 | * function assumes data_desc describes memory sufficient to hold | |
1873 | * all data described by the image request. | |
1874 | */ | |
1875 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
1876 | enum obj_request_type type, | |
1877 | void *data_desc) | |
bf0d5f50 AE |
1878 | { |
1879 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1880 | struct rbd_obj_request *obj_request = NULL; | |
1881 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 1882 | bool write_request = img_request_write_test(img_request); |
f1a4739f AE |
1883 | struct bio *bio_list; |
1884 | unsigned int bio_offset = 0; | |
1885 | struct page **pages; | |
7da22d29 | 1886 | u64 img_offset; |
bf0d5f50 AE |
1887 | u64 resid; |
1888 | u16 opcode; | |
1889 | ||
f1a4739f AE |
1890 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
1891 | (int)type, data_desc); | |
37206ee5 | 1892 | |
430c28c3 | 1893 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 1894 | img_offset = img_request->offset; |
bf0d5f50 | 1895 | resid = img_request->length; |
4dda41d3 | 1896 | rbd_assert(resid > 0); |
f1a4739f AE |
1897 | |
1898 | if (type == OBJ_REQUEST_BIO) { | |
1899 | bio_list = data_desc; | |
1900 | rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
1901 | } else { | |
1902 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
1903 | pages = data_desc; | |
1904 | } | |
1905 | ||
bf0d5f50 | 1906 | while (resid) { |
2fa12320 | 1907 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 1908 | const char *object_name; |
bf0d5f50 AE |
1909 | u64 offset; |
1910 | u64 length; | |
1911 | ||
7da22d29 | 1912 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
1913 | if (!object_name) |
1914 | goto out_unwind; | |
7da22d29 AE |
1915 | offset = rbd_segment_offset(rbd_dev, img_offset); |
1916 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 1917 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 1918 | offset, length, type); |
bf0d5f50 AE |
1919 | kfree(object_name); /* object request has its own copy */ |
1920 | if (!obj_request) | |
1921 | goto out_unwind; | |
1922 | ||
f1a4739f AE |
1923 | if (type == OBJ_REQUEST_BIO) { |
1924 | unsigned int clone_size; | |
1925 | ||
1926 | rbd_assert(length <= (u64)UINT_MAX); | |
1927 | clone_size = (unsigned int)length; | |
1928 | obj_request->bio_list = | |
1929 | bio_chain_clone_range(&bio_list, | |
1930 | &bio_offset, | |
1931 | clone_size, | |
1932 | GFP_ATOMIC); | |
1933 | if (!obj_request->bio_list) | |
1934 | goto out_partial; | |
1935 | } else { | |
1936 | unsigned int page_count; | |
1937 | ||
1938 | obj_request->pages = pages; | |
1939 | page_count = (u32)calc_pages_for(offset, length); | |
1940 | obj_request->page_count = page_count; | |
1941 | if ((offset + length) & ~PAGE_MASK) | |
1942 | page_count--; /* more on last page */ | |
1943 | pages += page_count; | |
1944 | } | |
bf0d5f50 | 1945 | |
2fa12320 AE |
1946 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
1947 | obj_request); | |
1948 | if (!osd_req) | |
bf0d5f50 | 1949 | goto out_partial; |
2fa12320 | 1950 | obj_request->osd_req = osd_req; |
2169238d | 1951 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 1952 | |
2fa12320 AE |
1953 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
1954 | 0, 0); | |
f1a4739f AE |
1955 | if (type == OBJ_REQUEST_BIO) |
1956 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
1957 | obj_request->bio_list, length); | |
1958 | else | |
1959 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
1960 | obj_request->pages, length, | |
1961 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
1962 | |
1963 | if (write_request) | |
1964 | rbd_osd_req_format_write(obj_request); | |
1965 | else | |
1966 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 1967 | |
7da22d29 | 1968 | obj_request->img_offset = img_offset; |
bf0d5f50 AE |
1969 | rbd_img_obj_request_add(img_request, obj_request); |
1970 | ||
7da22d29 | 1971 | img_offset += length; |
bf0d5f50 AE |
1972 | resid -= length; |
1973 | } | |
1974 | ||
1975 | return 0; | |
1976 | ||
1977 | out_partial: | |
1978 | rbd_obj_request_put(obj_request); | |
1979 | out_unwind: | |
1980 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
1981 | rbd_obj_request_put(obj_request); | |
1982 | ||
1983 | return -ENOMEM; | |
1984 | } | |
1985 | ||
0eefd470 AE |
1986 | static void |
1987 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
1988 | { | |
1989 | struct rbd_img_request *img_request; | |
1990 | struct rbd_device *rbd_dev; | |
1991 | u64 length; | |
1992 | u32 page_count; | |
1993 | ||
1994 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
1995 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1996 | img_request = obj_request->img_request; | |
1997 | rbd_assert(img_request); | |
1998 | ||
1999 | rbd_dev = img_request->rbd_dev; | |
2000 | rbd_assert(rbd_dev); | |
2001 | length = (u64)1 << rbd_dev->header.obj_order; | |
2002 | page_count = (u32)calc_pages_for(0, length); | |
2003 | ||
2004 | rbd_assert(obj_request->copyup_pages); | |
2005 | ceph_release_page_vector(obj_request->copyup_pages, page_count); | |
2006 | obj_request->copyup_pages = NULL; | |
2007 | ||
2008 | /* | |
2009 | * We want the transfer count to reflect the size of the | |
2010 | * original write request. There is no such thing as a | |
2011 | * successful short write, so if the request was successful | |
2012 | * we can just set it to the originally-requested length. | |
2013 | */ | |
2014 | if (!obj_request->result) | |
2015 | obj_request->xferred = obj_request->length; | |
2016 | ||
2017 | /* Finish up with the normal image object callback */ | |
2018 | ||
2019 | rbd_img_obj_callback(obj_request); | |
2020 | } | |
2021 | ||
3d7efd18 AE |
2022 | static void |
2023 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2024 | { | |
2025 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2026 | struct ceph_osd_request *osd_req; |
2027 | struct ceph_osd_client *osdc; | |
2028 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2029 | struct page **pages; |
3d7efd18 AE |
2030 | int result; |
2031 | u64 obj_size; | |
2032 | u64 xferred; | |
2033 | ||
2034 | rbd_assert(img_request_child_test(img_request)); | |
2035 | ||
2036 | /* First get what we need from the image request */ | |
2037 | ||
2038 | pages = img_request->copyup_pages; | |
2039 | rbd_assert(pages != NULL); | |
2040 | img_request->copyup_pages = NULL; | |
2041 | ||
2042 | orig_request = img_request->obj_request; | |
2043 | rbd_assert(orig_request != NULL); | |
0eefd470 | 2044 | rbd_assert(orig_request->type == OBJ_REQUEST_BIO); |
3d7efd18 AE |
2045 | result = img_request->result; |
2046 | obj_size = img_request->length; | |
2047 | xferred = img_request->xferred; | |
2048 | ||
0eefd470 AE |
2049 | rbd_dev = img_request->rbd_dev; |
2050 | rbd_assert(rbd_dev); | |
2051 | rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order); | |
2052 | ||
3d7efd18 AE |
2053 | rbd_img_request_put(img_request); |
2054 | ||
0eefd470 AE |
2055 | if (result) |
2056 | goto out_err; | |
2057 | ||
2058 | /* Allocate the new copyup osd request for the original request */ | |
2059 | ||
2060 | result = -ENOMEM; | |
2061 | rbd_assert(!orig_request->osd_req); | |
2062 | osd_req = rbd_osd_req_create_copyup(orig_request); | |
2063 | if (!osd_req) | |
2064 | goto out_err; | |
2065 | orig_request->osd_req = osd_req; | |
2066 | orig_request->copyup_pages = pages; | |
3d7efd18 | 2067 | |
0eefd470 | 2068 | /* Initialize the copyup op */ |
3d7efd18 | 2069 | |
0eefd470 AE |
2070 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
2071 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0, | |
2072 | false, false); | |
3d7efd18 | 2073 | |
0eefd470 AE |
2074 | /* Then the original write request op */ |
2075 | ||
2076 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, | |
2077 | orig_request->offset, | |
2078 | orig_request->length, 0, 0); | |
2079 | osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list, | |
2080 | orig_request->length); | |
2081 | ||
2082 | rbd_osd_req_format_write(orig_request); | |
2083 | ||
2084 | /* All set, send it off. */ | |
2085 | ||
2086 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2087 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2088 | result = rbd_obj_request_submit(osdc, orig_request); | |
2089 | if (!result) | |
2090 | return; | |
2091 | out_err: | |
2092 | /* Record the error code and complete the request */ | |
2093 | ||
2094 | orig_request->result = result; | |
2095 | orig_request->xferred = 0; | |
2096 | obj_request_done_set(orig_request); | |
2097 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2098 | } |
2099 | ||
2100 | /* | |
2101 | * Read from the parent image the range of data that covers the | |
2102 | * entire target of the given object request. This is used for | |
2103 | * satisfying a layered image write request when the target of an | |
2104 | * object request from the image request does not exist. | |
2105 | * | |
2106 | * A page array big enough to hold the returned data is allocated | |
2107 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2108 | * When the read completes, this page array will be transferred to | |
2109 | * the original object request for the copyup operation. | |
2110 | * | |
2111 | * If an error occurs, record it as the result of the original | |
2112 | * object request and mark it done so it gets completed. | |
2113 | */ | |
2114 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2115 | { | |
2116 | struct rbd_img_request *img_request = NULL; | |
2117 | struct rbd_img_request *parent_request = NULL; | |
2118 | struct rbd_device *rbd_dev; | |
2119 | u64 img_offset; | |
2120 | u64 length; | |
2121 | struct page **pages = NULL; | |
2122 | u32 page_count; | |
2123 | int result; | |
2124 | ||
2125 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2126 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2127 | ||
2128 | img_request = obj_request->img_request; | |
2129 | rbd_assert(img_request != NULL); | |
2130 | rbd_dev = img_request->rbd_dev; | |
2131 | rbd_assert(rbd_dev->parent != NULL); | |
2132 | ||
0eefd470 AE |
2133 | /* |
2134 | * First things first. The original osd request is of no | |
2135 | * use to use any more, we'll need a new one that can hold | |
2136 | * the two ops in a copyup request. We'll get that later, | |
2137 | * but for now we can release the old one. | |
2138 | */ | |
2139 | rbd_osd_req_destroy(obj_request->osd_req); | |
2140 | obj_request->osd_req = NULL; | |
2141 | ||
3d7efd18 AE |
2142 | /* |
2143 | * Determine the byte range covered by the object in the | |
2144 | * child image to which the original request was to be sent. | |
2145 | */ | |
2146 | img_offset = obj_request->img_offset - obj_request->offset; | |
2147 | length = (u64)1 << rbd_dev->header.obj_order; | |
2148 | ||
a9e8ba2c AE |
2149 | /* |
2150 | * There is no defined parent data beyond the parent | |
2151 | * overlap, so limit what we read at that boundary if | |
2152 | * necessary. | |
2153 | */ | |
2154 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2155 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2156 | length = rbd_dev->parent_overlap - img_offset; | |
2157 | } | |
2158 | ||
3d7efd18 AE |
2159 | /* |
2160 | * Allocate a page array big enough to receive the data read | |
2161 | * from the parent. | |
2162 | */ | |
2163 | page_count = (u32)calc_pages_for(0, length); | |
2164 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2165 | if (IS_ERR(pages)) { | |
2166 | result = PTR_ERR(pages); | |
2167 | pages = NULL; | |
2168 | goto out_err; | |
2169 | } | |
2170 | ||
2171 | result = -ENOMEM; | |
2172 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2173 | img_offset, length, | |
2174 | false, true); | |
2175 | if (!parent_request) | |
2176 | goto out_err; | |
2177 | rbd_obj_request_get(obj_request); | |
2178 | parent_request->obj_request = obj_request; | |
2179 | ||
2180 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2181 | if (result) | |
2182 | goto out_err; | |
2183 | parent_request->copyup_pages = pages; | |
2184 | ||
2185 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2186 | result = rbd_img_request_submit(parent_request); | |
2187 | if (!result) | |
2188 | return 0; | |
2189 | ||
2190 | parent_request->copyup_pages = NULL; | |
2191 | parent_request->obj_request = NULL; | |
2192 | rbd_obj_request_put(obj_request); | |
2193 | out_err: | |
2194 | if (pages) | |
2195 | ceph_release_page_vector(pages, page_count); | |
2196 | if (parent_request) | |
2197 | rbd_img_request_put(parent_request); | |
2198 | obj_request->result = result; | |
2199 | obj_request->xferred = 0; | |
2200 | obj_request_done_set(obj_request); | |
2201 | ||
2202 | return result; | |
2203 | } | |
2204 | ||
c5b5ef6c AE |
2205 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2206 | { | |
c5b5ef6c AE |
2207 | struct rbd_obj_request *orig_request; |
2208 | int result; | |
2209 | ||
2210 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2211 | ||
2212 | /* | |
2213 | * All we need from the object request is the original | |
2214 | * request and the result of the STAT op. Grab those, then | |
2215 | * we're done with the request. | |
2216 | */ | |
2217 | orig_request = obj_request->obj_request; | |
2218 | obj_request->obj_request = NULL; | |
2219 | rbd_assert(orig_request); | |
2220 | rbd_assert(orig_request->img_request); | |
2221 | ||
2222 | result = obj_request->result; | |
2223 | obj_request->result = 0; | |
2224 | ||
2225 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2226 | obj_request, orig_request, result, | |
2227 | obj_request->xferred, obj_request->length); | |
2228 | rbd_obj_request_put(obj_request); | |
2229 | ||
2230 | rbd_assert(orig_request); | |
2231 | rbd_assert(orig_request->img_request); | |
c5b5ef6c AE |
2232 | |
2233 | /* | |
2234 | * Our only purpose here is to determine whether the object | |
2235 | * exists, and we don't want to treat the non-existence as | |
2236 | * an error. If something else comes back, transfer the | |
2237 | * error to the original request and complete it now. | |
2238 | */ | |
2239 | if (!result) { | |
2240 | obj_request_existence_set(orig_request, true); | |
2241 | } else if (result == -ENOENT) { | |
2242 | obj_request_existence_set(orig_request, false); | |
2243 | } else if (result) { | |
2244 | orig_request->result = result; | |
3d7efd18 | 2245 | goto out; |
c5b5ef6c AE |
2246 | } |
2247 | ||
2248 | /* | |
2249 | * Resubmit the original request now that we have recorded | |
2250 | * whether the target object exists. | |
2251 | */ | |
b454e36d | 2252 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2253 | out: |
c5b5ef6c AE |
2254 | if (orig_request->result) |
2255 | rbd_obj_request_complete(orig_request); | |
2256 | rbd_obj_request_put(orig_request); | |
2257 | } | |
2258 | ||
2259 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2260 | { | |
2261 | struct rbd_obj_request *stat_request; | |
2262 | struct rbd_device *rbd_dev; | |
2263 | struct ceph_osd_client *osdc; | |
2264 | struct page **pages = NULL; | |
2265 | u32 page_count; | |
2266 | size_t size; | |
2267 | int ret; | |
2268 | ||
2269 | /* | |
2270 | * The response data for a STAT call consists of: | |
2271 | * le64 length; | |
2272 | * struct { | |
2273 | * le32 tv_sec; | |
2274 | * le32 tv_nsec; | |
2275 | * } mtime; | |
2276 | */ | |
2277 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2278 | page_count = (u32)calc_pages_for(0, size); | |
2279 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2280 | if (IS_ERR(pages)) | |
2281 | return PTR_ERR(pages); | |
2282 | ||
2283 | ret = -ENOMEM; | |
2284 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2285 | OBJ_REQUEST_PAGES); | |
2286 | if (!stat_request) | |
2287 | goto out; | |
2288 | ||
2289 | rbd_obj_request_get(obj_request); | |
2290 | stat_request->obj_request = obj_request; | |
2291 | stat_request->pages = pages; | |
2292 | stat_request->page_count = page_count; | |
2293 | ||
2294 | rbd_assert(obj_request->img_request); | |
2295 | rbd_dev = obj_request->img_request->rbd_dev; | |
2296 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2297 | stat_request); | |
2298 | if (!stat_request->osd_req) | |
2299 | goto out; | |
2300 | stat_request->callback = rbd_img_obj_exists_callback; | |
2301 | ||
2302 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2303 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2304 | false, false); | |
9d4df01f | 2305 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2306 | |
2307 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2308 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2309 | out: | |
2310 | if (ret) | |
2311 | rbd_obj_request_put(obj_request); | |
2312 | ||
2313 | return ret; | |
2314 | } | |
2315 | ||
b454e36d AE |
2316 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2317 | { | |
2318 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2319 | struct rbd_device *rbd_dev; |
3d7efd18 | 2320 | bool known; |
b454e36d AE |
2321 | |
2322 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2323 | ||
2324 | img_request = obj_request->img_request; | |
2325 | rbd_assert(img_request); | |
a9e8ba2c | 2326 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2327 | |
b454e36d | 2328 | /* |
a9e8ba2c AE |
2329 | * Only writes to layered images need special handling. |
2330 | * Reads and non-layered writes are simple object requests. | |
2331 | * Layered writes that start beyond the end of the overlap | |
2332 | * with the parent have no parent data, so they too are | |
2333 | * simple object requests. Finally, if the target object is | |
2334 | * known to already exist, its parent data has already been | |
2335 | * copied, so a write to the object can also be handled as a | |
2336 | * simple object request. | |
b454e36d AE |
2337 | */ |
2338 | if (!img_request_write_test(img_request) || | |
2339 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2340 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2341 | ((known = obj_request_known_test(obj_request)) && |
2342 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2343 | |
2344 | struct rbd_device *rbd_dev; | |
2345 | struct ceph_osd_client *osdc; | |
2346 | ||
2347 | rbd_dev = obj_request->img_request->rbd_dev; | |
2348 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2349 | ||
2350 | return rbd_obj_request_submit(osdc, obj_request); | |
2351 | } | |
2352 | ||
2353 | /* | |
3d7efd18 AE |
2354 | * It's a layered write. The target object might exist but |
2355 | * we may not know that yet. If we know it doesn't exist, | |
2356 | * start by reading the data for the full target object from | |
2357 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2358 | */ |
3d7efd18 AE |
2359 | if (known) |
2360 | return rbd_img_obj_parent_read_full(obj_request); | |
2361 | ||
2362 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2363 | |
2364 | return rbd_img_obj_exists_submit(obj_request); | |
2365 | } | |
2366 | ||
bf0d5f50 AE |
2367 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2368 | { | |
bf0d5f50 | 2369 | struct rbd_obj_request *obj_request; |
46faeed4 | 2370 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2371 | |
37206ee5 | 2372 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2373 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2374 | int ret; |
2375 | ||
b454e36d | 2376 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2377 | if (ret) |
2378 | return ret; | |
bf0d5f50 AE |
2379 | } |
2380 | ||
2381 | return 0; | |
2382 | } | |
8b3e1a56 AE |
2383 | |
2384 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2385 | { | |
2386 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2387 | struct rbd_device *rbd_dev; |
2388 | u64 obj_end; | |
8b3e1a56 AE |
2389 | |
2390 | rbd_assert(img_request_child_test(img_request)); | |
2391 | ||
2392 | obj_request = img_request->obj_request; | |
a9e8ba2c AE |
2393 | rbd_assert(obj_request); |
2394 | rbd_assert(obj_request->img_request); | |
2395 | ||
8b3e1a56 | 2396 | obj_request->result = img_request->result; |
a9e8ba2c AE |
2397 | if (obj_request->result) |
2398 | goto out; | |
2399 | ||
2400 | /* | |
2401 | * We need to zero anything beyond the parent overlap | |
2402 | * boundary. Since rbd_img_obj_request_read_callback() | |
2403 | * will zero anything beyond the end of a short read, an | |
2404 | * easy way to do this is to pretend the data from the | |
2405 | * parent came up short--ending at the overlap boundary. | |
2406 | */ | |
2407 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2408 | obj_end = obj_request->img_offset + obj_request->length; | |
2409 | rbd_dev = obj_request->img_request->rbd_dev; | |
2410 | if (obj_end > rbd_dev->parent_overlap) { | |
2411 | u64 xferred = 0; | |
2412 | ||
2413 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2414 | xferred = rbd_dev->parent_overlap - | |
2415 | obj_request->img_offset; | |
8b3e1a56 | 2416 | |
a9e8ba2c AE |
2417 | obj_request->xferred = min(img_request->xferred, xferred); |
2418 | } else { | |
2419 | obj_request->xferred = img_request->xferred; | |
2420 | } | |
2421 | out: | |
8b3e1a56 AE |
2422 | rbd_img_obj_request_read_callback(obj_request); |
2423 | rbd_obj_request_complete(obj_request); | |
2424 | } | |
2425 | ||
2426 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2427 | { | |
2428 | struct rbd_device *rbd_dev; | |
2429 | struct rbd_img_request *img_request; | |
2430 | int result; | |
2431 | ||
2432 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2433 | rbd_assert(obj_request->img_request != NULL); | |
2434 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
2435 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2436 | ||
2437 | rbd_dev = obj_request->img_request->rbd_dev; | |
2438 | rbd_assert(rbd_dev->parent != NULL); | |
2439 | /* rbd_read_finish(obj_request, obj_request->length); */ | |
2440 | img_request = rbd_img_request_create(rbd_dev->parent, | |
2441 | obj_request->img_offset, | |
2442 | obj_request->length, | |
2443 | false, true); | |
2444 | result = -ENOMEM; | |
2445 | if (!img_request) | |
2446 | goto out_err; | |
2447 | ||
2448 | rbd_obj_request_get(obj_request); | |
2449 | img_request->obj_request = obj_request; | |
2450 | ||
f1a4739f AE |
2451 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2452 | obj_request->bio_list); | |
8b3e1a56 AE |
2453 | if (result) |
2454 | goto out_err; | |
2455 | ||
2456 | img_request->callback = rbd_img_parent_read_callback; | |
2457 | result = rbd_img_request_submit(img_request); | |
2458 | if (result) | |
2459 | goto out_err; | |
2460 | ||
2461 | return; | |
2462 | out_err: | |
2463 | if (img_request) | |
2464 | rbd_img_request_put(img_request); | |
2465 | obj_request->result = result; | |
2466 | obj_request->xferred = 0; | |
2467 | obj_request_done_set(obj_request); | |
2468 | } | |
bf0d5f50 | 2469 | |
cc4a38bd | 2470 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2471 | { |
2472 | struct rbd_obj_request *obj_request; | |
2169238d | 2473 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2474 | int ret; |
2475 | ||
2476 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2477 | OBJ_REQUEST_NODATA); | |
2478 | if (!obj_request) | |
2479 | return -ENOMEM; | |
2480 | ||
2481 | ret = -ENOMEM; | |
430c28c3 | 2482 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2483 | if (!obj_request->osd_req) |
2484 | goto out; | |
2169238d | 2485 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 2486 | |
c99d2d4a | 2487 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2488 | notify_id, 0, 0); |
9d4df01f | 2489 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2490 | |
b8d70035 | 2491 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 2492 | out: |
cf81b60e AE |
2493 | if (ret) |
2494 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2495 | |
2496 | return ret; | |
2497 | } | |
2498 | ||
2499 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2500 | { | |
2501 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
b8d70035 AE |
2502 | |
2503 | if (!rbd_dev) | |
2504 | return; | |
2505 | ||
37206ee5 | 2506 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2507 | rbd_dev->header_name, (unsigned long long)notify_id, |
2508 | (unsigned int)opcode); | |
2509 | (void)rbd_dev_refresh(rbd_dev); | |
b8d70035 | 2510 | |
cc4a38bd | 2511 | rbd_obj_notify_ack(rbd_dev, notify_id); |
b8d70035 AE |
2512 | } |
2513 | ||
9969ebc5 AE |
2514 | /* |
2515 | * Request sync osd watch/unwatch. The value of "start" determines | |
2516 | * whether a watch request is being initiated or torn down. | |
2517 | */ | |
2518 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
2519 | { | |
2520 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2521 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2522 | int ret; |
2523 | ||
2524 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2525 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2526 | ||
2527 | if (start) { | |
3c663bbd | 2528 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2529 | &rbd_dev->watch_event); |
2530 | if (ret < 0) | |
2531 | return ret; | |
8eb87565 | 2532 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2533 | } |
2534 | ||
2535 | ret = -ENOMEM; | |
2536 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2537 | OBJ_REQUEST_NODATA); | |
2538 | if (!obj_request) | |
2539 | goto out_cancel; | |
2540 | ||
430c28c3 AE |
2541 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2542 | if (!obj_request->osd_req) | |
2543 | goto out_cancel; | |
2544 | ||
8eb87565 | 2545 | if (start) |
975241af | 2546 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2547 | else |
6977c3f9 | 2548 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2549 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2550 | |
2551 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
b21ebddd | 2552 | rbd_dev->watch_event->cookie, 0, start); |
9d4df01f | 2553 | rbd_osd_req_format_write(obj_request); |
2169238d | 2554 | |
9969ebc5 AE |
2555 | ret = rbd_obj_request_submit(osdc, obj_request); |
2556 | if (ret) | |
2557 | goto out_cancel; | |
2558 | ret = rbd_obj_request_wait(obj_request); | |
2559 | if (ret) | |
2560 | goto out_cancel; | |
9969ebc5 AE |
2561 | ret = obj_request->result; |
2562 | if (ret) | |
2563 | goto out_cancel; | |
2564 | ||
8eb87565 AE |
2565 | /* |
2566 | * A watch request is set to linger, so the underlying osd | |
2567 | * request won't go away until we unregister it. We retain | |
2568 | * a pointer to the object request during that time (in | |
2569 | * rbd_dev->watch_request), so we'll keep a reference to | |
2570 | * it. We'll drop that reference (below) after we've | |
2571 | * unregistered it. | |
2572 | */ | |
2573 | if (start) { | |
2574 | rbd_dev->watch_request = obj_request; | |
2575 | ||
2576 | return 0; | |
2577 | } | |
2578 | ||
2579 | /* We have successfully torn down the watch request */ | |
2580 | ||
2581 | rbd_obj_request_put(rbd_dev->watch_request); | |
2582 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2583 | out_cancel: |
2584 | /* Cancel the event if we're tearing down, or on error */ | |
2585 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2586 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2587 | if (obj_request) |
2588 | rbd_obj_request_put(obj_request); | |
2589 | ||
2590 | return ret; | |
2591 | } | |
2592 | ||
36be9a76 | 2593 | /* |
f40eb349 AE |
2594 | * Synchronous osd object method call. Returns the number of bytes |
2595 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2596 | */ |
2597 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2598 | const char *object_name, | |
2599 | const char *class_name, | |
2600 | const char *method_name, | |
4157976b | 2601 | const void *outbound, |
36be9a76 | 2602 | size_t outbound_size, |
4157976b | 2603 | void *inbound, |
e2a58ee5 | 2604 | size_t inbound_size) |
36be9a76 | 2605 | { |
2169238d | 2606 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2607 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2608 | struct page **pages; |
2609 | u32 page_count; | |
2610 | int ret; | |
2611 | ||
2612 | /* | |
6010a451 AE |
2613 | * Method calls are ultimately read operations. The result |
2614 | * should placed into the inbound buffer provided. They | |
2615 | * also supply outbound data--parameters for the object | |
2616 | * method. Currently if this is present it will be a | |
2617 | * snapshot id. | |
36be9a76 | 2618 | */ |
57385b51 | 2619 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2620 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2621 | if (IS_ERR(pages)) | |
2622 | return PTR_ERR(pages); | |
2623 | ||
2624 | ret = -ENOMEM; | |
6010a451 | 2625 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2626 | OBJ_REQUEST_PAGES); |
2627 | if (!obj_request) | |
2628 | goto out; | |
2629 | ||
2630 | obj_request->pages = pages; | |
2631 | obj_request->page_count = page_count; | |
2632 | ||
430c28c3 | 2633 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2634 | if (!obj_request->osd_req) |
2635 | goto out; | |
2636 | ||
c99d2d4a | 2637 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2638 | class_name, method_name); |
2639 | if (outbound_size) { | |
2640 | struct ceph_pagelist *pagelist; | |
2641 | ||
2642 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2643 | if (!pagelist) | |
2644 | goto out; | |
2645 | ||
2646 | ceph_pagelist_init(pagelist); | |
2647 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
2648 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
2649 | pagelist); | |
2650 | } | |
a4ce40a9 AE |
2651 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
2652 | obj_request->pages, inbound_size, | |
44cd188d | 2653 | 0, false, false); |
9d4df01f | 2654 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2655 | |
36be9a76 AE |
2656 | ret = rbd_obj_request_submit(osdc, obj_request); |
2657 | if (ret) | |
2658 | goto out; | |
2659 | ret = rbd_obj_request_wait(obj_request); | |
2660 | if (ret) | |
2661 | goto out; | |
2662 | ||
2663 | ret = obj_request->result; | |
2664 | if (ret < 0) | |
2665 | goto out; | |
57385b51 AE |
2666 | |
2667 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
2668 | ret = (int)obj_request->xferred; | |
903bb32e | 2669 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
2670 | out: |
2671 | if (obj_request) | |
2672 | rbd_obj_request_put(obj_request); | |
2673 | else | |
2674 | ceph_release_page_vector(pages, page_count); | |
2675 | ||
2676 | return ret; | |
2677 | } | |
2678 | ||
bf0d5f50 | 2679 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 2680 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
2681 | { |
2682 | struct rbd_device *rbd_dev = q->queuedata; | |
2683 | bool read_only = rbd_dev->mapping.read_only; | |
2684 | struct request *rq; | |
2685 | int result; | |
2686 | ||
2687 | while ((rq = blk_fetch_request(q))) { | |
2688 | bool write_request = rq_data_dir(rq) == WRITE; | |
2689 | struct rbd_img_request *img_request; | |
2690 | u64 offset; | |
2691 | u64 length; | |
2692 | ||
2693 | /* Ignore any non-FS requests that filter through. */ | |
2694 | ||
2695 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
2696 | dout("%s: non-fs request type %d\n", __func__, |
2697 | (int) rq->cmd_type); | |
2698 | __blk_end_request_all(rq, 0); | |
2699 | continue; | |
2700 | } | |
2701 | ||
2702 | /* Ignore/skip any zero-length requests */ | |
2703 | ||
2704 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
2705 | length = (u64) blk_rq_bytes(rq); | |
2706 | ||
2707 | if (!length) { | |
2708 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
2709 | __blk_end_request_all(rq, 0); |
2710 | continue; | |
2711 | } | |
2712 | ||
2713 | spin_unlock_irq(q->queue_lock); | |
2714 | ||
2715 | /* Disallow writes to a read-only device */ | |
2716 | ||
2717 | if (write_request) { | |
2718 | result = -EROFS; | |
2719 | if (read_only) | |
2720 | goto end_request; | |
2721 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
2722 | } | |
2723 | ||
6d292906 AE |
2724 | /* |
2725 | * Quit early if the mapped snapshot no longer | |
2726 | * exists. It's still possible the snapshot will | |
2727 | * have disappeared by the time our request arrives | |
2728 | * at the osd, but there's no sense in sending it if | |
2729 | * we already know. | |
2730 | */ | |
2731 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
2732 | dout("request for non-existent snapshot"); |
2733 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
2734 | result = -ENXIO; | |
2735 | goto end_request; | |
2736 | } | |
2737 | ||
bf0d5f50 | 2738 | result = -EINVAL; |
c0cd10db AE |
2739 | if (offset && length > U64_MAX - offset + 1) { |
2740 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
2741 | offset, length); | |
bf0d5f50 | 2742 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 2743 | } |
bf0d5f50 AE |
2744 | |
2745 | result = -ENOMEM; | |
2746 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
9849e986 | 2747 | write_request, false); |
bf0d5f50 AE |
2748 | if (!img_request) |
2749 | goto end_request; | |
2750 | ||
2751 | img_request->rq = rq; | |
2752 | ||
f1a4739f AE |
2753 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2754 | rq->bio); | |
bf0d5f50 AE |
2755 | if (!result) |
2756 | result = rbd_img_request_submit(img_request); | |
2757 | if (result) | |
2758 | rbd_img_request_put(img_request); | |
2759 | end_request: | |
2760 | spin_lock_irq(q->queue_lock); | |
2761 | if (result < 0) { | |
7da22d29 AE |
2762 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
2763 | write_request ? "write" : "read", | |
2764 | length, offset, result); | |
2765 | ||
bf0d5f50 AE |
2766 | __blk_end_request_all(rq, result); |
2767 | } | |
2768 | } | |
2769 | } | |
2770 | ||
602adf40 YS |
2771 | /* |
2772 | * a queue callback. Makes sure that we don't create a bio that spans across | |
2773 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 2774 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
2775 | */ |
2776 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
2777 | struct bio_vec *bvec) | |
2778 | { | |
2779 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
2780 | sector_t sector_offset; |
2781 | sector_t sectors_per_obj; | |
2782 | sector_t obj_sector_offset; | |
2783 | int ret; | |
2784 | ||
2785 | /* | |
2786 | * Find how far into its rbd object the partition-relative | |
2787 | * bio start sector is to offset relative to the enclosing | |
2788 | * device. | |
2789 | */ | |
2790 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
2791 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
2792 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
2793 | ||
2794 | /* | |
2795 | * Compute the number of bytes from that offset to the end | |
2796 | * of the object. Account for what's already used by the bio. | |
2797 | */ | |
2798 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2799 | if (ret > bmd->bi_size) | |
2800 | ret -= bmd->bi_size; | |
2801 | else | |
2802 | ret = 0; | |
2803 | ||
2804 | /* | |
2805 | * Don't send back more than was asked for. And if the bio | |
2806 | * was empty, let the whole thing through because: "Note | |
2807 | * that a block device *must* allow a single page to be | |
2808 | * added to an empty bio." | |
2809 | */ | |
2810 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2811 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2812 | ret = (int) bvec->bv_len; | |
2813 | ||
2814 | return ret; | |
602adf40 YS |
2815 | } |
2816 | ||
2817 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2818 | { | |
2819 | struct gendisk *disk = rbd_dev->disk; | |
2820 | ||
2821 | if (!disk) | |
2822 | return; | |
2823 | ||
a0cab924 AE |
2824 | rbd_dev->disk = NULL; |
2825 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 2826 | del_gendisk(disk); |
a0cab924 AE |
2827 | if (disk->queue) |
2828 | blk_cleanup_queue(disk->queue); | |
2829 | } | |
602adf40 YS |
2830 | put_disk(disk); |
2831 | } | |
2832 | ||
788e2df3 AE |
2833 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2834 | const char *object_name, | |
7097f8df | 2835 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
2836 | |
2837 | { | |
2169238d | 2838 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 2839 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
2840 | struct page **pages = NULL; |
2841 | u32 page_count; | |
1ceae7ef | 2842 | size_t size; |
788e2df3 AE |
2843 | int ret; |
2844 | ||
2845 | page_count = (u32) calc_pages_for(offset, length); | |
2846 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2847 | if (IS_ERR(pages)) | |
2848 | ret = PTR_ERR(pages); | |
2849 | ||
2850 | ret = -ENOMEM; | |
2851 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2852 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2853 | if (!obj_request) |
2854 | goto out; | |
2855 | ||
2856 | obj_request->pages = pages; | |
2857 | obj_request->page_count = page_count; | |
2858 | ||
430c28c3 | 2859 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
2860 | if (!obj_request->osd_req) |
2861 | goto out; | |
2862 | ||
c99d2d4a AE |
2863 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
2864 | offset, length, 0, 0); | |
406e2c9f | 2865 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 2866 | obj_request->pages, |
44cd188d AE |
2867 | obj_request->length, |
2868 | obj_request->offset & ~PAGE_MASK, | |
2869 | false, false); | |
9d4df01f | 2870 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2871 | |
788e2df3 AE |
2872 | ret = rbd_obj_request_submit(osdc, obj_request); |
2873 | if (ret) | |
2874 | goto out; | |
2875 | ret = rbd_obj_request_wait(obj_request); | |
2876 | if (ret) | |
2877 | goto out; | |
2878 | ||
2879 | ret = obj_request->result; | |
2880 | if (ret < 0) | |
2881 | goto out; | |
1ceae7ef AE |
2882 | |
2883 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
2884 | size = (size_t) obj_request->xferred; | |
903bb32e | 2885 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
2886 | rbd_assert(size <= (size_t)INT_MAX); |
2887 | ret = (int)size; | |
788e2df3 AE |
2888 | out: |
2889 | if (obj_request) | |
2890 | rbd_obj_request_put(obj_request); | |
2891 | else | |
2892 | ceph_release_page_vector(pages, page_count); | |
2893 | ||
2894 | return ret; | |
2895 | } | |
2896 | ||
602adf40 | 2897 | /* |
4156d998 AE |
2898 | * Read the complete header for the given rbd device. |
2899 | * | |
2900 | * Returns a pointer to a dynamically-allocated buffer containing | |
2901 | * the complete and validated header. Caller can pass the address | |
2902 | * of a variable that will be filled in with the version of the | |
2903 | * header object at the time it was read. | |
2904 | * | |
2905 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 2906 | */ |
4156d998 | 2907 | static struct rbd_image_header_ondisk * |
7097f8df | 2908 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev) |
602adf40 | 2909 | { |
4156d998 | 2910 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 2911 | u32 snap_count = 0; |
4156d998 AE |
2912 | u64 names_size = 0; |
2913 | u32 want_count; | |
2914 | int ret; | |
602adf40 | 2915 | |
00f1f36f | 2916 | /* |
4156d998 AE |
2917 | * The complete header will include an array of its 64-bit |
2918 | * snapshot ids, followed by the names of those snapshots as | |
2919 | * a contiguous block of NUL-terminated strings. Note that | |
2920 | * the number of snapshots could change by the time we read | |
2921 | * it in, in which case we re-read it. | |
00f1f36f | 2922 | */ |
4156d998 AE |
2923 | do { |
2924 | size_t size; | |
2925 | ||
2926 | kfree(ondisk); | |
2927 | ||
2928 | size = sizeof (*ondisk); | |
2929 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
2930 | size += names_size; | |
2931 | ondisk = kmalloc(size, GFP_KERNEL); | |
2932 | if (!ondisk) | |
2933 | return ERR_PTR(-ENOMEM); | |
2934 | ||
788e2df3 | 2935 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 2936 | 0, size, ondisk); |
4156d998 AE |
2937 | if (ret < 0) |
2938 | goto out_err; | |
c0cd10db | 2939 | if ((size_t)ret < size) { |
4156d998 | 2940 | ret = -ENXIO; |
06ecc6cb AE |
2941 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
2942 | size, ret); | |
4156d998 AE |
2943 | goto out_err; |
2944 | } | |
2945 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
2946 | ret = -ENXIO; | |
06ecc6cb | 2947 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 2948 | goto out_err; |
81e759fb | 2949 | } |
602adf40 | 2950 | |
4156d998 AE |
2951 | names_size = le64_to_cpu(ondisk->snap_names_len); |
2952 | want_count = snap_count; | |
2953 | snap_count = le32_to_cpu(ondisk->snap_count); | |
2954 | } while (snap_count != want_count); | |
00f1f36f | 2955 | |
4156d998 | 2956 | return ondisk; |
00f1f36f | 2957 | |
4156d998 AE |
2958 | out_err: |
2959 | kfree(ondisk); | |
2960 | ||
2961 | return ERR_PTR(ret); | |
2962 | } | |
2963 | ||
2964 | /* | |
2965 | * reload the ondisk the header | |
2966 | */ | |
2967 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
2968 | struct rbd_image_header *header) | |
2969 | { | |
2970 | struct rbd_image_header_ondisk *ondisk; | |
4156d998 | 2971 | int ret; |
602adf40 | 2972 | |
7097f8df | 2973 | ondisk = rbd_dev_v1_header_read(rbd_dev); |
4156d998 AE |
2974 | if (IS_ERR(ondisk)) |
2975 | return PTR_ERR(ondisk); | |
2976 | ret = rbd_header_from_disk(header, ondisk); | |
4156d998 AE |
2977 | kfree(ondisk); |
2978 | ||
2979 | return ret; | |
602adf40 YS |
2980 | } |
2981 | ||
41f38c2b | 2982 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
2983 | { |
2984 | struct rbd_snap *snap; | |
a0593290 | 2985 | struct rbd_snap *next; |
dfc5606d | 2986 | |
6087b51b AE |
2987 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) { |
2988 | list_del(&snap->node); | |
2989 | rbd_snap_destroy(snap); | |
2990 | } | |
dfc5606d YS |
2991 | } |
2992 | ||
9478554a AE |
2993 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
2994 | { | |
0d7dbfce | 2995 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
2996 | return; |
2997 | ||
e28626a0 AE |
2998 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) { |
2999 | sector_t size; | |
3000 | ||
3001 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
3002 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3003 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3004 | set_capacity(rbd_dev->disk, size); | |
3005 | } | |
9478554a AE |
3006 | } |
3007 | ||
602adf40 YS |
3008 | /* |
3009 | * only read the first part of the ondisk header, without the snaps info | |
3010 | */ | |
cc4a38bd | 3011 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev) |
602adf40 YS |
3012 | { |
3013 | int ret; | |
3014 | struct rbd_image_header h; | |
602adf40 YS |
3015 | |
3016 | ret = rbd_read_header(rbd_dev, &h); | |
3017 | if (ret < 0) | |
3018 | return ret; | |
3019 | ||
a51aa0c0 JD |
3020 | down_write(&rbd_dev->header_rwsem); |
3021 | ||
9478554a AE |
3022 | /* Update image size, and check for resize of mapped image */ |
3023 | rbd_dev->header.image_size = h.image_size; | |
3024 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 3025 | |
849b4260 | 3026 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 3027 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 3028 | kfree(rbd_dev->header.snap_names); |
d1d25646 | 3029 | /* osd requests may still refer to snapc */ |
812164f8 | 3030 | ceph_put_snap_context(rbd_dev->header.snapc); |
602adf40 | 3031 | |
93a24e08 | 3032 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
3033 | rbd_dev->header.snapc = h.snapc; |
3034 | rbd_dev->header.snap_names = h.snap_names; | |
3035 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 | 3036 | /* Free the extra copy of the object prefix */ |
c0cd10db AE |
3037 | if (strcmp(rbd_dev->header.object_prefix, h.object_prefix)) |
3038 | rbd_warn(rbd_dev, "object prefix changed (ignoring)"); | |
849b4260 AE |
3039 | kfree(h.object_prefix); |
3040 | ||
304f6808 | 3041 | ret = rbd_dev_snaps_update(rbd_dev); |
dfc5606d | 3042 | |
c666601a | 3043 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 3044 | |
dfc5606d | 3045 | return ret; |
602adf40 YS |
3046 | } |
3047 | ||
cc4a38bd | 3048 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3049 | { |
a3fbe5d4 | 3050 | u64 image_size; |
1fe5e993 AE |
3051 | int ret; |
3052 | ||
117973fb | 3053 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
a3fbe5d4 | 3054 | image_size = rbd_dev->header.image_size; |
1fe5e993 | 3055 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb | 3056 | if (rbd_dev->image_format == 1) |
cc4a38bd | 3057 | ret = rbd_dev_v1_refresh(rbd_dev); |
117973fb | 3058 | else |
cc4a38bd | 3059 | ret = rbd_dev_v2_refresh(rbd_dev); |
1fe5e993 | 3060 | mutex_unlock(&ctl_mutex); |
522a0cc0 AE |
3061 | if (ret) |
3062 | rbd_warn(rbd_dev, "got notification but failed to " | |
3063 | " update snaps: %d\n", ret); | |
a3fbe5d4 AE |
3064 | if (image_size != rbd_dev->header.image_size) |
3065 | revalidate_disk(rbd_dev->disk); | |
1fe5e993 AE |
3066 | |
3067 | return ret; | |
3068 | } | |
3069 | ||
602adf40 YS |
3070 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3071 | { | |
3072 | struct gendisk *disk; | |
3073 | struct request_queue *q; | |
593a9e7b | 3074 | u64 segment_size; |
602adf40 | 3075 | |
602adf40 | 3076 | /* create gendisk info */ |
602adf40 YS |
3077 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
3078 | if (!disk) | |
1fcdb8aa | 3079 | return -ENOMEM; |
602adf40 | 3080 | |
f0f8cef5 | 3081 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3082 | rbd_dev->dev_id); |
602adf40 YS |
3083 | disk->major = rbd_dev->major; |
3084 | disk->first_minor = 0; | |
3085 | disk->fops = &rbd_bd_ops; | |
3086 | disk->private_data = rbd_dev; | |
3087 | ||
bf0d5f50 | 3088 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3089 | if (!q) |
3090 | goto out_disk; | |
029bcbd8 | 3091 | |
593a9e7b AE |
3092 | /* We use the default size, but let's be explicit about it. */ |
3093 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3094 | ||
029bcbd8 | 3095 | /* set io sizes to object size */ |
593a9e7b AE |
3096 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3097 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3098 | blk_queue_max_segment_size(q, segment_size); | |
3099 | blk_queue_io_min(q, segment_size); | |
3100 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3101 | |
602adf40 YS |
3102 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3103 | disk->queue = q; | |
3104 | ||
3105 | q->queuedata = rbd_dev; | |
3106 | ||
3107 | rbd_dev->disk = disk; | |
602adf40 | 3108 | |
602adf40 | 3109 | return 0; |
602adf40 YS |
3110 | out_disk: |
3111 | put_disk(disk); | |
1fcdb8aa AE |
3112 | |
3113 | return -ENOMEM; | |
602adf40 YS |
3114 | } |
3115 | ||
dfc5606d YS |
3116 | /* |
3117 | sysfs | |
3118 | */ | |
3119 | ||
593a9e7b AE |
3120 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3121 | { | |
3122 | return container_of(dev, struct rbd_device, dev); | |
3123 | } | |
3124 | ||
dfc5606d YS |
3125 | static ssize_t rbd_size_show(struct device *dev, |
3126 | struct device_attribute *attr, char *buf) | |
3127 | { | |
593a9e7b | 3128 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3129 | |
fc71d833 AE |
3130 | return sprintf(buf, "%llu\n", |
3131 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3132 | } |
3133 | ||
34b13184 AE |
3134 | /* |
3135 | * Note this shows the features for whatever's mapped, which is not | |
3136 | * necessarily the base image. | |
3137 | */ | |
3138 | static ssize_t rbd_features_show(struct device *dev, | |
3139 | struct device_attribute *attr, char *buf) | |
3140 | { | |
3141 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3142 | ||
3143 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3144 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3145 | } |
3146 | ||
dfc5606d YS |
3147 | static ssize_t rbd_major_show(struct device *dev, |
3148 | struct device_attribute *attr, char *buf) | |
3149 | { | |
593a9e7b | 3150 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3151 | |
fc71d833 AE |
3152 | if (rbd_dev->major) |
3153 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3154 | ||
3155 | return sprintf(buf, "(none)\n"); | |
3156 | ||
dfc5606d YS |
3157 | } |
3158 | ||
3159 | static ssize_t rbd_client_id_show(struct device *dev, | |
3160 | struct device_attribute *attr, char *buf) | |
602adf40 | 3161 | { |
593a9e7b | 3162 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3163 | |
1dbb4399 AE |
3164 | return sprintf(buf, "client%lld\n", |
3165 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3166 | } |
3167 | ||
dfc5606d YS |
3168 | static ssize_t rbd_pool_show(struct device *dev, |
3169 | struct device_attribute *attr, char *buf) | |
602adf40 | 3170 | { |
593a9e7b | 3171 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3172 | |
0d7dbfce | 3173 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3174 | } |
3175 | ||
9bb2f334 AE |
3176 | static ssize_t rbd_pool_id_show(struct device *dev, |
3177 | struct device_attribute *attr, char *buf) | |
3178 | { | |
3179 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3180 | ||
0d7dbfce | 3181 | return sprintf(buf, "%llu\n", |
fc71d833 | 3182 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3183 | } |
3184 | ||
dfc5606d YS |
3185 | static ssize_t rbd_name_show(struct device *dev, |
3186 | struct device_attribute *attr, char *buf) | |
3187 | { | |
593a9e7b | 3188 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3189 | |
a92ffdf8 AE |
3190 | if (rbd_dev->spec->image_name) |
3191 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3192 | ||
3193 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3194 | } |
3195 | ||
589d30e0 AE |
3196 | static ssize_t rbd_image_id_show(struct device *dev, |
3197 | struct device_attribute *attr, char *buf) | |
3198 | { | |
3199 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3200 | ||
0d7dbfce | 3201 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3202 | } |
3203 | ||
34b13184 AE |
3204 | /* |
3205 | * Shows the name of the currently-mapped snapshot (or | |
3206 | * RBD_SNAP_HEAD_NAME for the base image). | |
3207 | */ | |
dfc5606d YS |
3208 | static ssize_t rbd_snap_show(struct device *dev, |
3209 | struct device_attribute *attr, | |
3210 | char *buf) | |
3211 | { | |
593a9e7b | 3212 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3213 | |
0d7dbfce | 3214 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3215 | } |
3216 | ||
86b00e0d AE |
3217 | /* |
3218 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3219 | * for the parent image. If there is no parent, simply shows | |
3220 | * "(no parent image)". | |
3221 | */ | |
3222 | static ssize_t rbd_parent_show(struct device *dev, | |
3223 | struct device_attribute *attr, | |
3224 | char *buf) | |
3225 | { | |
3226 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3227 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3228 | int count; | |
3229 | char *bufp = buf; | |
3230 | ||
3231 | if (!spec) | |
3232 | return sprintf(buf, "(no parent image)\n"); | |
3233 | ||
3234 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3235 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3236 | if (count < 0) | |
3237 | return count; | |
3238 | bufp += count; | |
3239 | ||
3240 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3241 | spec->image_name ? spec->image_name : "(unknown)"); | |
3242 | if (count < 0) | |
3243 | return count; | |
3244 | bufp += count; | |
3245 | ||
3246 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3247 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3248 | if (count < 0) | |
3249 | return count; | |
3250 | bufp += count; | |
3251 | ||
3252 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3253 | if (count < 0) | |
3254 | return count; | |
3255 | bufp += count; | |
3256 | ||
3257 | return (ssize_t) (bufp - buf); | |
3258 | } | |
3259 | ||
dfc5606d YS |
3260 | static ssize_t rbd_image_refresh(struct device *dev, |
3261 | struct device_attribute *attr, | |
3262 | const char *buf, | |
3263 | size_t size) | |
3264 | { | |
593a9e7b | 3265 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3266 | int ret; |
602adf40 | 3267 | |
cc4a38bd | 3268 | ret = rbd_dev_refresh(rbd_dev); |
b813623a AE |
3269 | |
3270 | return ret < 0 ? ret : size; | |
dfc5606d | 3271 | } |
602adf40 | 3272 | |
dfc5606d | 3273 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3274 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
3275 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
3276 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
3277 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3278 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3279 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3280 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3281 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3282 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3283 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3284 | |
3285 | static struct attribute *rbd_attrs[] = { | |
3286 | &dev_attr_size.attr, | |
34b13184 | 3287 | &dev_attr_features.attr, |
dfc5606d YS |
3288 | &dev_attr_major.attr, |
3289 | &dev_attr_client_id.attr, | |
3290 | &dev_attr_pool.attr, | |
9bb2f334 | 3291 | &dev_attr_pool_id.attr, |
dfc5606d | 3292 | &dev_attr_name.attr, |
589d30e0 | 3293 | &dev_attr_image_id.attr, |
dfc5606d | 3294 | &dev_attr_current_snap.attr, |
86b00e0d | 3295 | &dev_attr_parent.attr, |
dfc5606d | 3296 | &dev_attr_refresh.attr, |
dfc5606d YS |
3297 | NULL |
3298 | }; | |
3299 | ||
3300 | static struct attribute_group rbd_attr_group = { | |
3301 | .attrs = rbd_attrs, | |
3302 | }; | |
3303 | ||
3304 | static const struct attribute_group *rbd_attr_groups[] = { | |
3305 | &rbd_attr_group, | |
3306 | NULL | |
3307 | }; | |
3308 | ||
3309 | static void rbd_sysfs_dev_release(struct device *dev) | |
3310 | { | |
3311 | } | |
3312 | ||
3313 | static struct device_type rbd_device_type = { | |
3314 | .name = "rbd", | |
3315 | .groups = rbd_attr_groups, | |
3316 | .release = rbd_sysfs_dev_release, | |
3317 | }; | |
3318 | ||
8b8fb99c AE |
3319 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3320 | { | |
3321 | kref_get(&spec->kref); | |
3322 | ||
3323 | return spec; | |
3324 | } | |
3325 | ||
3326 | static void rbd_spec_free(struct kref *kref); | |
3327 | static void rbd_spec_put(struct rbd_spec *spec) | |
3328 | { | |
3329 | if (spec) | |
3330 | kref_put(&spec->kref, rbd_spec_free); | |
3331 | } | |
3332 | ||
3333 | static struct rbd_spec *rbd_spec_alloc(void) | |
3334 | { | |
3335 | struct rbd_spec *spec; | |
3336 | ||
3337 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3338 | if (!spec) | |
3339 | return NULL; | |
3340 | kref_init(&spec->kref); | |
3341 | ||
8b8fb99c AE |
3342 | return spec; |
3343 | } | |
3344 | ||
3345 | static void rbd_spec_free(struct kref *kref) | |
3346 | { | |
3347 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3348 | ||
3349 | kfree(spec->pool_name); | |
3350 | kfree(spec->image_id); | |
3351 | kfree(spec->image_name); | |
3352 | kfree(spec->snap_name); | |
3353 | kfree(spec); | |
3354 | } | |
3355 | ||
cc344fa1 | 3356 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3357 | struct rbd_spec *spec) |
3358 | { | |
3359 | struct rbd_device *rbd_dev; | |
3360 | ||
3361 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3362 | if (!rbd_dev) | |
3363 | return NULL; | |
3364 | ||
3365 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3366 | rbd_dev->flags = 0; |
c53d5893 AE |
3367 | INIT_LIST_HEAD(&rbd_dev->node); |
3368 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
3369 | init_rwsem(&rbd_dev->header_rwsem); | |
3370 | ||
3371 | rbd_dev->spec = spec; | |
3372 | rbd_dev->rbd_client = rbdc; | |
3373 | ||
0903e875 AE |
3374 | /* Initialize the layout used for all rbd requests */ |
3375 | ||
3376 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3377 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3378 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3379 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3380 | ||
c53d5893 AE |
3381 | return rbd_dev; |
3382 | } | |
3383 | ||
3384 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3385 | { | |
c53d5893 AE |
3386 | rbd_put_client(rbd_dev->rbd_client); |
3387 | rbd_spec_put(rbd_dev->spec); | |
3388 | kfree(rbd_dev); | |
3389 | } | |
3390 | ||
6087b51b | 3391 | static void rbd_snap_destroy(struct rbd_snap *snap) |
dfc5606d | 3392 | { |
3e83b65b AE |
3393 | kfree(snap->name); |
3394 | kfree(snap); | |
dfc5606d YS |
3395 | } |
3396 | ||
6087b51b | 3397 | static struct rbd_snap *rbd_snap_create(struct rbd_device *rbd_dev, |
c8d18425 | 3398 | const char *snap_name, |
34b13184 AE |
3399 | u64 snap_id, u64 snap_size, |
3400 | u64 snap_features) | |
dfc5606d | 3401 | { |
4e891e0a | 3402 | struct rbd_snap *snap; |
4e891e0a AE |
3403 | |
3404 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 3405 | if (!snap) |
4e891e0a AE |
3406 | return ERR_PTR(-ENOMEM); |
3407 | ||
6e584f52 | 3408 | snap->name = snap_name; |
c8d18425 AE |
3409 | snap->id = snap_id; |
3410 | snap->size = snap_size; | |
34b13184 | 3411 | snap->features = snap_features; |
4e891e0a AE |
3412 | |
3413 | return snap; | |
dfc5606d YS |
3414 | } |
3415 | ||
6e584f52 AE |
3416 | /* |
3417 | * Returns a dynamically-allocated snapshot name if successful, or a | |
3418 | * pointer-coded error otherwise. | |
3419 | */ | |
cb75223d | 3420 | static const char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
cd892126 AE |
3421 | u64 *snap_size, u64 *snap_features) |
3422 | { | |
cb75223d | 3423 | const char *snap_name; |
6e584f52 | 3424 | int i; |
cd892126 AE |
3425 | |
3426 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
3427 | ||
cd892126 AE |
3428 | /* Skip over names until we find the one we are looking for */ |
3429 | ||
3430 | snap_name = rbd_dev->header.snap_names; | |
6e584f52 | 3431 | for (i = 0; i < which; i++) |
cd892126 AE |
3432 | snap_name += strlen(snap_name) + 1; |
3433 | ||
6e584f52 AE |
3434 | snap_name = kstrdup(snap_name, GFP_KERNEL); |
3435 | if (!snap_name) | |
3436 | return ERR_PTR(-ENOMEM); | |
3437 | ||
3438 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
3439 | *snap_features = 0; /* No features for v1 */ | |
3440 | ||
cd892126 AE |
3441 | return snap_name; |
3442 | } | |
3443 | ||
9d475de5 AE |
3444 | /* |
3445 | * Get the size and object order for an image snapshot, or if | |
3446 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3447 | * image. | |
3448 | */ | |
3449 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3450 | u8 *order, u64 *snap_size) | |
3451 | { | |
3452 | __le64 snapid = cpu_to_le64(snap_id); | |
3453 | int ret; | |
3454 | struct { | |
3455 | u8 order; | |
3456 | __le64 size; | |
3457 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3458 | ||
36be9a76 | 3459 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3460 | "rbd", "get_size", |
4157976b | 3461 | &snapid, sizeof (snapid), |
e2a58ee5 | 3462 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3463 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3464 | if (ret < 0) |
3465 | return ret; | |
57385b51 AE |
3466 | if (ret < sizeof (size_buf)) |
3467 | return -ERANGE; | |
9d475de5 | 3468 | |
c86f86e9 AE |
3469 | if (order) |
3470 | *order = size_buf.order; | |
9d475de5 AE |
3471 | *snap_size = le64_to_cpu(size_buf.size); |
3472 | ||
3473 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
57385b51 AE |
3474 | (unsigned long long)snap_id, (unsigned int)*order, |
3475 | (unsigned long long)*snap_size); | |
9d475de5 AE |
3476 | |
3477 | return 0; | |
3478 | } | |
3479 | ||
3480 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3481 | { | |
3482 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3483 | &rbd_dev->header.obj_order, | |
3484 | &rbd_dev->header.image_size); | |
3485 | } | |
3486 | ||
1e130199 AE |
3487 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3488 | { | |
3489 | void *reply_buf; | |
3490 | int ret; | |
3491 | void *p; | |
3492 | ||
3493 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3494 | if (!reply_buf) | |
3495 | return -ENOMEM; | |
3496 | ||
36be9a76 | 3497 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3498 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3499 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3500 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3501 | if (ret < 0) |
3502 | goto out; | |
3503 | ||
3504 | p = reply_buf; | |
3505 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3506 | p + ret, NULL, GFP_NOIO); |
3507 | ret = 0; | |
1e130199 AE |
3508 | |
3509 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3510 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3511 | rbd_dev->header.object_prefix = NULL; | |
3512 | } else { | |
3513 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3514 | } | |
1e130199 AE |
3515 | out: |
3516 | kfree(reply_buf); | |
3517 | ||
3518 | return ret; | |
3519 | } | |
3520 | ||
b1b5402a AE |
3521 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3522 | u64 *snap_features) | |
3523 | { | |
3524 | __le64 snapid = cpu_to_le64(snap_id); | |
3525 | struct { | |
3526 | __le64 features; | |
3527 | __le64 incompat; | |
4157976b | 3528 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3529 | u64 incompat; |
b1b5402a AE |
3530 | int ret; |
3531 | ||
36be9a76 | 3532 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3533 | "rbd", "get_features", |
4157976b | 3534 | &snapid, sizeof (snapid), |
e2a58ee5 | 3535 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3536 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3537 | if (ret < 0) |
3538 | return ret; | |
57385b51 AE |
3539 | if (ret < sizeof (features_buf)) |
3540 | return -ERANGE; | |
d889140c AE |
3541 | |
3542 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3543 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3544 | return -ENXIO; |
d889140c | 3545 | |
b1b5402a AE |
3546 | *snap_features = le64_to_cpu(features_buf.features); |
3547 | ||
3548 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3549 | (unsigned long long)snap_id, |
3550 | (unsigned long long)*snap_features, | |
3551 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3552 | |
3553 | return 0; | |
3554 | } | |
3555 | ||
3556 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3557 | { | |
3558 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3559 | &rbd_dev->header.features); | |
3560 | } | |
3561 | ||
86b00e0d AE |
3562 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3563 | { | |
3564 | struct rbd_spec *parent_spec; | |
3565 | size_t size; | |
3566 | void *reply_buf = NULL; | |
3567 | __le64 snapid; | |
3568 | void *p; | |
3569 | void *end; | |
3570 | char *image_id; | |
3571 | u64 overlap; | |
86b00e0d AE |
3572 | int ret; |
3573 | ||
3574 | parent_spec = rbd_spec_alloc(); | |
3575 | if (!parent_spec) | |
3576 | return -ENOMEM; | |
3577 | ||
3578 | size = sizeof (__le64) + /* pool_id */ | |
3579 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3580 | sizeof (__le64) + /* snap_id */ | |
3581 | sizeof (__le64); /* overlap */ | |
3582 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3583 | if (!reply_buf) { | |
3584 | ret = -ENOMEM; | |
3585 | goto out_err; | |
3586 | } | |
3587 | ||
3588 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3589 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3590 | "rbd", "get_parent", |
4157976b | 3591 | &snapid, sizeof (snapid), |
e2a58ee5 | 3592 | reply_buf, size); |
36be9a76 | 3593 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3594 | if (ret < 0) |
3595 | goto out_err; | |
3596 | ||
86b00e0d | 3597 | p = reply_buf; |
57385b51 AE |
3598 | end = reply_buf + ret; |
3599 | ret = -ERANGE; | |
86b00e0d AE |
3600 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); |
3601 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
3602 | goto out; /* No parent? No problem. */ | |
3603 | ||
0903e875 AE |
3604 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3605 | ||
3606 | ret = -EIO; | |
c0cd10db AE |
3607 | if (parent_spec->pool_id > (u64)U32_MAX) { |
3608 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", | |
3609 | (unsigned long long)parent_spec->pool_id, U32_MAX); | |
57385b51 | 3610 | goto out_err; |
c0cd10db | 3611 | } |
0903e875 | 3612 | |
979ed480 | 3613 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3614 | if (IS_ERR(image_id)) { |
3615 | ret = PTR_ERR(image_id); | |
3616 | goto out_err; | |
3617 | } | |
3618 | parent_spec->image_id = image_id; | |
3619 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
3620 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
3621 | ||
3622 | rbd_dev->parent_overlap = overlap; | |
3623 | rbd_dev->parent_spec = parent_spec; | |
3624 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3625 | out: | |
3626 | ret = 0; | |
3627 | out_err: | |
3628 | kfree(reply_buf); | |
3629 | rbd_spec_put(parent_spec); | |
3630 | ||
3631 | return ret; | |
3632 | } | |
3633 | ||
cc070d59 AE |
3634 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3635 | { | |
3636 | struct { | |
3637 | __le64 stripe_unit; | |
3638 | __le64 stripe_count; | |
3639 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3640 | size_t size = sizeof (striping_info_buf); | |
3641 | void *p; | |
3642 | u64 obj_size; | |
3643 | u64 stripe_unit; | |
3644 | u64 stripe_count; | |
3645 | int ret; | |
3646 | ||
3647 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3648 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 3649 | (char *)&striping_info_buf, size); |
cc070d59 AE |
3650 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
3651 | if (ret < 0) | |
3652 | return ret; | |
3653 | if (ret < size) | |
3654 | return -ERANGE; | |
3655 | ||
3656 | /* | |
3657 | * We don't actually support the "fancy striping" feature | |
3658 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3659 | * defaults the behavior is the same as before. So find | |
3660 | * out, and only fail if the image has non-default values. | |
3661 | */ | |
3662 | ret = -EINVAL; | |
3663 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3664 | p = &striping_info_buf; | |
3665 | stripe_unit = ceph_decode_64(&p); | |
3666 | if (stripe_unit != obj_size) { | |
3667 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
3668 | "(got %llu want %llu)", | |
3669 | stripe_unit, obj_size); | |
3670 | return -EINVAL; | |
3671 | } | |
3672 | stripe_count = ceph_decode_64(&p); | |
3673 | if (stripe_count != 1) { | |
3674 | rbd_warn(rbd_dev, "unsupported stripe count " | |
3675 | "(got %llu want 1)", stripe_count); | |
3676 | return -EINVAL; | |
3677 | } | |
500d0c0f AE |
3678 | rbd_dev->header.stripe_unit = stripe_unit; |
3679 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
3680 | |
3681 | return 0; | |
3682 | } | |
3683 | ||
9e15b77d AE |
3684 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3685 | { | |
3686 | size_t image_id_size; | |
3687 | char *image_id; | |
3688 | void *p; | |
3689 | void *end; | |
3690 | size_t size; | |
3691 | void *reply_buf = NULL; | |
3692 | size_t len = 0; | |
3693 | char *image_name = NULL; | |
3694 | int ret; | |
3695 | ||
3696 | rbd_assert(!rbd_dev->spec->image_name); | |
3697 | ||
69e7a02f AE |
3698 | len = strlen(rbd_dev->spec->image_id); |
3699 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3700 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3701 | if (!image_id) | |
3702 | return NULL; | |
3703 | ||
3704 | p = image_id; | |
4157976b | 3705 | end = image_id + image_id_size; |
57385b51 | 3706 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
3707 | |
3708 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3709 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3710 | if (!reply_buf) | |
3711 | goto out; | |
3712 | ||
36be9a76 | 3713 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3714 | "rbd", "dir_get_name", |
3715 | image_id, image_id_size, | |
e2a58ee5 | 3716 | reply_buf, size); |
9e15b77d AE |
3717 | if (ret < 0) |
3718 | goto out; | |
3719 | p = reply_buf; | |
f40eb349 AE |
3720 | end = reply_buf + ret; |
3721 | ||
9e15b77d AE |
3722 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
3723 | if (IS_ERR(image_name)) | |
3724 | image_name = NULL; | |
3725 | else | |
3726 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3727 | out: | |
3728 | kfree(reply_buf); | |
3729 | kfree(image_id); | |
3730 | ||
3731 | return image_name; | |
3732 | } | |
3733 | ||
3734 | /* | |
2e9f7f1c AE |
3735 | * When an rbd image has a parent image, it is identified by the |
3736 | * pool, image, and snapshot ids (not names). This function fills | |
3737 | * in the names for those ids. (It's OK if we can't figure out the | |
3738 | * name for an image id, but the pool and snapshot ids should always | |
3739 | * exist and have names.) All names in an rbd spec are dynamically | |
3740 | * allocated. | |
e1d4213f AE |
3741 | * |
3742 | * When an image being mapped (not a parent) is probed, we have the | |
3743 | * pool name and pool id, image name and image id, and the snapshot | |
3744 | * name. The only thing we're missing is the snapshot id. | |
2e9f7f1c AE |
3745 | * |
3746 | * The set of snapshots for an image is not known until they have | |
3747 | * been read by rbd_dev_snaps_update(), so we can't completely fill | |
3748 | * in this information until after that has been called. | |
9e15b77d | 3749 | */ |
2e9f7f1c | 3750 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 3751 | { |
2e9f7f1c AE |
3752 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3753 | struct rbd_spec *spec = rbd_dev->spec; | |
3754 | const char *pool_name; | |
3755 | const char *image_name; | |
3756 | const char *snap_name; | |
9e15b77d AE |
3757 | int ret; |
3758 | ||
e1d4213f AE |
3759 | /* |
3760 | * An image being mapped will have the pool name (etc.), but | |
3761 | * we need to look up the snapshot id. | |
3762 | */ | |
2e9f7f1c AE |
3763 | if (spec->pool_name) { |
3764 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
e1d4213f AE |
3765 | struct rbd_snap *snap; |
3766 | ||
2e9f7f1c | 3767 | snap = snap_by_name(rbd_dev, spec->snap_name); |
e1d4213f AE |
3768 | if (!snap) |
3769 | return -ENOENT; | |
2e9f7f1c | 3770 | spec->snap_id = snap->id; |
e1d4213f | 3771 | } else { |
2e9f7f1c | 3772 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
3773 | } |
3774 | ||
3775 | return 0; | |
3776 | } | |
9e15b77d | 3777 | |
2e9f7f1c | 3778 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 3779 | |
2e9f7f1c AE |
3780 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
3781 | if (!pool_name) { | |
3782 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
3783 | return -EIO; |
3784 | } | |
2e9f7f1c AE |
3785 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
3786 | if (!pool_name) | |
9e15b77d AE |
3787 | return -ENOMEM; |
3788 | ||
3789 | /* Fetch the image name; tolerate failure here */ | |
3790 | ||
2e9f7f1c AE |
3791 | image_name = rbd_dev_image_name(rbd_dev); |
3792 | if (!image_name) | |
06ecc6cb | 3793 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 3794 | |
2e9f7f1c | 3795 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 3796 | |
2e9f7f1c AE |
3797 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
3798 | if (!snap_name) { | |
3799 | rbd_warn(rbd_dev, "no snapshot with id %llu", spec->snap_id); | |
9e15b77d AE |
3800 | ret = -EIO; |
3801 | goto out_err; | |
3802 | } | |
2e9f7f1c AE |
3803 | snap_name = kstrdup(snap_name, GFP_KERNEL); |
3804 | if (!snap_name) { | |
3805 | ret = -ENOMEM; | |
9e15b77d | 3806 | goto out_err; |
2e9f7f1c AE |
3807 | } |
3808 | ||
3809 | spec->pool_name = pool_name; | |
3810 | spec->image_name = image_name; | |
3811 | spec->snap_name = snap_name; | |
9e15b77d AE |
3812 | |
3813 | return 0; | |
3814 | out_err: | |
2e9f7f1c AE |
3815 | kfree(image_name); |
3816 | kfree(pool_name); | |
9e15b77d AE |
3817 | |
3818 | return ret; | |
3819 | } | |
3820 | ||
cc4a38bd | 3821 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
3822 | { |
3823 | size_t size; | |
3824 | int ret; | |
3825 | void *reply_buf; | |
3826 | void *p; | |
3827 | void *end; | |
3828 | u64 seq; | |
3829 | u32 snap_count; | |
3830 | struct ceph_snap_context *snapc; | |
3831 | u32 i; | |
3832 | ||
3833 | /* | |
3834 | * We'll need room for the seq value (maximum snapshot id), | |
3835 | * snapshot count, and array of that many snapshot ids. | |
3836 | * For now we have a fixed upper limit on the number we're | |
3837 | * prepared to receive. | |
3838 | */ | |
3839 | size = sizeof (__le64) + sizeof (__le32) + | |
3840 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3841 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3842 | if (!reply_buf) | |
3843 | return -ENOMEM; | |
3844 | ||
36be9a76 | 3845 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3846 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 3847 | reply_buf, size); |
36be9a76 | 3848 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3849 | if (ret < 0) |
3850 | goto out; | |
3851 | ||
35d489f9 | 3852 | p = reply_buf; |
57385b51 AE |
3853 | end = reply_buf + ret; |
3854 | ret = -ERANGE; | |
35d489f9 AE |
3855 | ceph_decode_64_safe(&p, end, seq, out); |
3856 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3857 | ||
3858 | /* | |
3859 | * Make sure the reported number of snapshot ids wouldn't go | |
3860 | * beyond the end of our buffer. But before checking that, | |
3861 | * make sure the computed size of the snapshot context we | |
3862 | * allocate is representable in a size_t. | |
3863 | */ | |
3864 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3865 | / sizeof (u64)) { | |
3866 | ret = -EINVAL; | |
3867 | goto out; | |
3868 | } | |
3869 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3870 | goto out; | |
468521c1 | 3871 | ret = 0; |
35d489f9 | 3872 | |
812164f8 | 3873 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
3874 | if (!snapc) { |
3875 | ret = -ENOMEM; | |
3876 | goto out; | |
3877 | } | |
35d489f9 | 3878 | snapc->seq = seq; |
35d489f9 AE |
3879 | for (i = 0; i < snap_count; i++) |
3880 | snapc->snaps[i] = ceph_decode_64(&p); | |
3881 | ||
3882 | rbd_dev->header.snapc = snapc; | |
3883 | ||
3884 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 3885 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
3886 | out: |
3887 | kfree(reply_buf); | |
3888 | ||
57385b51 | 3889 | return ret; |
35d489f9 AE |
3890 | } |
3891 | ||
cb75223d | 3892 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
b8b1e2db AE |
3893 | { |
3894 | size_t size; | |
3895 | void *reply_buf; | |
3896 | __le64 snap_id; | |
3897 | int ret; | |
3898 | void *p; | |
3899 | void *end; | |
b8b1e2db AE |
3900 | char *snap_name; |
3901 | ||
3902 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
3903 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3904 | if (!reply_buf) | |
3905 | return ERR_PTR(-ENOMEM); | |
3906 | ||
acb1b6ca | 3907 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); |
b8b1e2db | 3908 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); |
36be9a76 | 3909 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 3910 | "rbd", "get_snapshot_name", |
4157976b | 3911 | &snap_id, sizeof (snap_id), |
e2a58ee5 | 3912 | reply_buf, size); |
36be9a76 | 3913 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
3914 | if (ret < 0) { |
3915 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 3916 | goto out; |
f40eb349 | 3917 | } |
b8b1e2db AE |
3918 | |
3919 | p = reply_buf; | |
f40eb349 | 3920 | end = reply_buf + ret; |
e5c35534 | 3921 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 3922 | if (IS_ERR(snap_name)) |
b8b1e2db | 3923 | goto out; |
b8b1e2db | 3924 | |
f40eb349 AE |
3925 | dout(" snap_id 0x%016llx snap_name = %s\n", |
3926 | (unsigned long long)le64_to_cpu(snap_id), snap_name); | |
b8b1e2db AE |
3927 | out: |
3928 | kfree(reply_buf); | |
3929 | ||
f40eb349 | 3930 | return snap_name; |
b8b1e2db AE |
3931 | } |
3932 | ||
cb75223d | 3933 | static const char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, |
b8b1e2db AE |
3934 | u64 *snap_size, u64 *snap_features) |
3935 | { | |
e0b49868 | 3936 | u64 snap_id; |
acb1b6ca AE |
3937 | u64 size; |
3938 | u64 features; | |
cb75223d | 3939 | const char *snap_name; |
b8b1e2db AE |
3940 | int ret; |
3941 | ||
acb1b6ca | 3942 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); |
b8b1e2db | 3943 | snap_id = rbd_dev->header.snapc->snaps[which]; |
acb1b6ca | 3944 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); |
b8b1e2db | 3945 | if (ret) |
acb1b6ca AE |
3946 | goto out_err; |
3947 | ||
3948 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); | |
b8b1e2db | 3949 | if (ret) |
acb1b6ca AE |
3950 | goto out_err; |
3951 | ||
3952 | snap_name = rbd_dev_v2_snap_name(rbd_dev, which); | |
3953 | if (!IS_ERR(snap_name)) { | |
3954 | *snap_size = size; | |
3955 | *snap_features = features; | |
3956 | } | |
b8b1e2db | 3957 | |
acb1b6ca AE |
3958 | return snap_name; |
3959 | out_err: | |
3960 | return ERR_PTR(ret); | |
b8b1e2db AE |
3961 | } |
3962 | ||
cb75223d | 3963 | static const char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, |
b8b1e2db AE |
3964 | u64 *snap_size, u64 *snap_features) |
3965 | { | |
3966 | if (rbd_dev->image_format == 1) | |
3967 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
3968 | snap_size, snap_features); | |
3969 | if (rbd_dev->image_format == 2) | |
3970 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
3971 | snap_size, snap_features); | |
3972 | return ERR_PTR(-EINVAL); | |
3973 | } | |
3974 | ||
cc4a38bd | 3975 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev) |
117973fb AE |
3976 | { |
3977 | int ret; | |
117973fb AE |
3978 | |
3979 | down_write(&rbd_dev->header_rwsem); | |
3980 | ||
117973fb AE |
3981 | ret = rbd_dev_v2_image_size(rbd_dev); |
3982 | if (ret) | |
3983 | goto out; | |
117973fb AE |
3984 | rbd_update_mapping_size(rbd_dev); |
3985 | ||
cc4a38bd | 3986 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb AE |
3987 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
3988 | if (ret) | |
3989 | goto out; | |
3990 | ret = rbd_dev_snaps_update(rbd_dev); | |
3991 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
3992 | if (ret) | |
3993 | goto out; | |
117973fb AE |
3994 | out: |
3995 | up_write(&rbd_dev->header_rwsem); | |
3996 | ||
3997 | return ret; | |
3998 | } | |
3999 | ||
dfc5606d | 4000 | /* |
35938150 AE |
4001 | * Scan the rbd device's current snapshot list and compare it to the |
4002 | * newly-received snapshot context. Remove any existing snapshots | |
4003 | * not present in the new snapshot context. Add a new snapshot for | |
4004 | * any snaphots in the snapshot context not in the current list. | |
4005 | * And verify there are no changes to snapshots we already know | |
4006 | * about. | |
4007 | * | |
4008 | * Assumes the snapshots in the snapshot context are sorted by | |
4009 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
4010 | * are also maintained in that order.) | |
522a0cc0 AE |
4011 | * |
4012 | * Note that any error occurs while updating the snapshot list | |
4013 | * aborts the update, and the entire list is cleared. The snapshot | |
4014 | * list becomes inconsistent at that point anyway, so it might as | |
4015 | * well be empty. | |
dfc5606d | 4016 | */ |
304f6808 | 4017 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 4018 | { |
35938150 AE |
4019 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
4020 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
4021 | struct list_head *head = &rbd_dev->snaps; |
4022 | struct list_head *links = head->next; | |
4023 | u32 index = 0; | |
522a0cc0 | 4024 | int ret = 0; |
dfc5606d | 4025 | |
522a0cc0 | 4026 | dout("%s: snap count is %u\n", __func__, (unsigned int)snap_count); |
35938150 AE |
4027 | while (index < snap_count || links != head) { |
4028 | u64 snap_id; | |
4029 | struct rbd_snap *snap; | |
cb75223d | 4030 | const char *snap_name; |
cd892126 AE |
4031 | u64 snap_size = 0; |
4032 | u64 snap_features = 0; | |
dfc5606d | 4033 | |
35938150 AE |
4034 | snap_id = index < snap_count ? snapc->snaps[index] |
4035 | : CEPH_NOSNAP; | |
4036 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
4037 | : NULL; | |
aafb230e | 4038 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 4039 | |
35938150 AE |
4040 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
4041 | struct list_head *next = links->next; | |
dfc5606d | 4042 | |
6d292906 AE |
4043 | /* |
4044 | * A previously-existing snapshot is not in | |
4045 | * the new snap context. | |
4046 | * | |
522a0cc0 AE |
4047 | * If the now-missing snapshot is the one |
4048 | * the image represents, clear its existence | |
4049 | * flag so we can avoid sending any more | |
4050 | * requests to it. | |
6d292906 | 4051 | */ |
0d7dbfce | 4052 | if (rbd_dev->spec->snap_id == snap->id) |
6d292906 | 4053 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
3e83b65b | 4054 | dout("removing %ssnap id %llu\n", |
0d7dbfce AE |
4055 | rbd_dev->spec->snap_id == snap->id ? |
4056 | "mapped " : "", | |
522a0cc0 | 4057 | (unsigned long long)snap->id); |
6087b51b AE |
4058 | |
4059 | list_del(&snap->node); | |
4060 | rbd_snap_destroy(snap); | |
35938150 AE |
4061 | |
4062 | /* Done with this list entry; advance */ | |
4063 | ||
4064 | links = next; | |
dfc5606d YS |
4065 | continue; |
4066 | } | |
35938150 | 4067 | |
b8b1e2db AE |
4068 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
4069 | &snap_size, &snap_features); | |
522a0cc0 AE |
4070 | if (IS_ERR(snap_name)) { |
4071 | ret = PTR_ERR(snap_name); | |
4072 | dout("failed to get snap info, error %d\n", ret); | |
4073 | goto out_err; | |
4074 | } | |
cd892126 | 4075 | |
522a0cc0 AE |
4076 | dout("entry %u: snap_id = %llu\n", (unsigned int)snap_count, |
4077 | (unsigned long long)snap_id); | |
35938150 AE |
4078 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
4079 | struct rbd_snap *new_snap; | |
4080 | ||
4081 | /* We haven't seen this snapshot before */ | |
4082 | ||
6087b51b | 4083 | new_snap = rbd_snap_create(rbd_dev, snap_name, |
cd892126 | 4084 | snap_id, snap_size, snap_features); |
9fcbb800 | 4085 | if (IS_ERR(new_snap)) { |
522a0cc0 AE |
4086 | ret = PTR_ERR(new_snap); |
4087 | dout(" failed to add dev, error %d\n", ret); | |
4088 | goto out_err; | |
9fcbb800 | 4089 | } |
35938150 AE |
4090 | |
4091 | /* New goes before existing, or at end of list */ | |
4092 | ||
9fcbb800 | 4093 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
4094 | if (snap) |
4095 | list_add_tail(&new_snap->node, &snap->node); | |
4096 | else | |
523f3258 | 4097 | list_add_tail(&new_snap->node, head); |
35938150 AE |
4098 | } else { |
4099 | /* Already have this one */ | |
4100 | ||
9fcbb800 AE |
4101 | dout(" already present\n"); |
4102 | ||
cd892126 | 4103 | rbd_assert(snap->size == snap_size); |
aafb230e | 4104 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 4105 | rbd_assert(snap->features == snap_features); |
35938150 AE |
4106 | |
4107 | /* Done with this list entry; advance */ | |
4108 | ||
4109 | links = links->next; | |
dfc5606d | 4110 | } |
35938150 AE |
4111 | |
4112 | /* Advance to the next entry in the snapshot context */ | |
4113 | ||
4114 | index++; | |
dfc5606d | 4115 | } |
9fcbb800 | 4116 | dout("%s: done\n", __func__); |
dfc5606d YS |
4117 | |
4118 | return 0; | |
522a0cc0 AE |
4119 | out_err: |
4120 | rbd_remove_all_snaps(rbd_dev); | |
4121 | ||
4122 | return ret; | |
dfc5606d YS |
4123 | } |
4124 | ||
dfc5606d YS |
4125 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4126 | { | |
dfc5606d | 4127 | struct device *dev; |
cd789ab9 | 4128 | int ret; |
dfc5606d YS |
4129 | |
4130 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 4131 | |
cd789ab9 | 4132 | dev = &rbd_dev->dev; |
dfc5606d YS |
4133 | dev->bus = &rbd_bus_type; |
4134 | dev->type = &rbd_device_type; | |
4135 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4136 | dev->release = rbd_dev_device_release; |
de71a297 | 4137 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4138 | ret = device_register(dev); |
dfc5606d | 4139 | |
dfc5606d | 4140 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 4141 | |
dfc5606d | 4142 | return ret; |
602adf40 YS |
4143 | } |
4144 | ||
dfc5606d YS |
4145 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4146 | { | |
4147 | device_unregister(&rbd_dev->dev); | |
4148 | } | |
4149 | ||
e2839308 | 4150 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
4151 | |
4152 | /* | |
499afd5b AE |
4153 | * Get a unique rbd identifier for the given new rbd_dev, and add |
4154 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 4155 | */ |
e2839308 | 4156 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4157 | { |
e2839308 | 4158 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
4159 | |
4160 | spin_lock(&rbd_dev_list_lock); | |
4161 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4162 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
4163 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
4164 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 4165 | } |
b7f23c36 | 4166 | |
1ddbe94e | 4167 | /* |
499afd5b AE |
4168 | * Remove an rbd_dev from the global list, and record that its |
4169 | * identifier is no longer in use. | |
1ddbe94e | 4170 | */ |
e2839308 | 4171 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4172 | { |
d184f6bf | 4173 | struct list_head *tmp; |
de71a297 | 4174 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
4175 | int max_id; |
4176 | ||
aafb230e | 4177 | rbd_assert(rbd_id > 0); |
499afd5b | 4178 | |
e2839308 AE |
4179 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
4180 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
4181 | spin_lock(&rbd_dev_list_lock); |
4182 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
4183 | |
4184 | /* | |
4185 | * If the id being "put" is not the current maximum, there | |
4186 | * is nothing special we need to do. | |
4187 | */ | |
e2839308 | 4188 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
4189 | spin_unlock(&rbd_dev_list_lock); |
4190 | return; | |
4191 | } | |
4192 | ||
4193 | /* | |
4194 | * We need to update the current maximum id. Search the | |
4195 | * list to find out what it is. We're more likely to find | |
4196 | * the maximum at the end, so search the list backward. | |
4197 | */ | |
4198 | max_id = 0; | |
4199 | list_for_each_prev(tmp, &rbd_dev_list) { | |
4200 | struct rbd_device *rbd_dev; | |
4201 | ||
4202 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
4203 | if (rbd_dev->dev_id > max_id) |
4204 | max_id = rbd_dev->dev_id; | |
d184f6bf | 4205 | } |
499afd5b | 4206 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 4207 | |
1ddbe94e | 4208 | /* |
e2839308 | 4209 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
4210 | * which case it now accurately reflects the new maximum. |
4211 | * Be careful not to overwrite the maximum value in that | |
4212 | * case. | |
1ddbe94e | 4213 | */ |
e2839308 AE |
4214 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
4215 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
4216 | } |
4217 | ||
e28fff26 AE |
4218 | /* |
4219 | * Skips over white space at *buf, and updates *buf to point to the | |
4220 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4221 | * the token (string of non-white space characters) found. Note |
4222 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4223 | */ |
4224 | static inline size_t next_token(const char **buf) | |
4225 | { | |
4226 | /* | |
4227 | * These are the characters that produce nonzero for | |
4228 | * isspace() in the "C" and "POSIX" locales. | |
4229 | */ | |
4230 | const char *spaces = " \f\n\r\t\v"; | |
4231 | ||
4232 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4233 | ||
4234 | return strcspn(*buf, spaces); /* Return token length */ | |
4235 | } | |
4236 | ||
4237 | /* | |
4238 | * Finds the next token in *buf, and if the provided token buffer is | |
4239 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4240 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4241 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4242 | * |
4243 | * Returns the length of the token found (not including the '\0'). | |
4244 | * Return value will be 0 if no token is found, and it will be >= | |
4245 | * token_size if the token would not fit. | |
4246 | * | |
593a9e7b | 4247 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4248 | * found token. Note that this occurs even if the token buffer is |
4249 | * too small to hold it. | |
4250 | */ | |
4251 | static inline size_t copy_token(const char **buf, | |
4252 | char *token, | |
4253 | size_t token_size) | |
4254 | { | |
4255 | size_t len; | |
4256 | ||
4257 | len = next_token(buf); | |
4258 | if (len < token_size) { | |
4259 | memcpy(token, *buf, len); | |
4260 | *(token + len) = '\0'; | |
4261 | } | |
4262 | *buf += len; | |
4263 | ||
4264 | return len; | |
4265 | } | |
4266 | ||
ea3352f4 AE |
4267 | /* |
4268 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4269 | * enough to hold a copy of it, and copies the token into the new | |
4270 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4271 | * that a duplicate buffer is created even for a zero-length token. | |
4272 | * | |
4273 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4274 | * pointer if memory for the duplicate was not available. If | |
4275 | * the lenp argument is a non-null pointer, the length of the token | |
4276 | * (not including the '\0') is returned in *lenp. | |
4277 | * | |
4278 | * If successful, the *buf pointer will be updated to point beyond | |
4279 | * the end of the found token. | |
4280 | * | |
4281 | * Note: uses GFP_KERNEL for allocation. | |
4282 | */ | |
4283 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4284 | { | |
4285 | char *dup; | |
4286 | size_t len; | |
4287 | ||
4288 | len = next_token(buf); | |
4caf35f9 | 4289 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4290 | if (!dup) |
4291 | return NULL; | |
ea3352f4 AE |
4292 | *(dup + len) = '\0'; |
4293 | *buf += len; | |
4294 | ||
4295 | if (lenp) | |
4296 | *lenp = len; | |
4297 | ||
4298 | return dup; | |
4299 | } | |
4300 | ||
a725f65e | 4301 | /* |
859c31df AE |
4302 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4303 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4304 | * and the data written is passed here via a NUL-terminated buffer. | |
4305 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4306 | * |
859c31df AE |
4307 | * The information extracted from these options is recorded in |
4308 | * the other parameters which return dynamically-allocated | |
4309 | * structures: | |
4310 | * ceph_opts | |
4311 | * The address of a pointer that will refer to a ceph options | |
4312 | * structure. Caller must release the returned pointer using | |
4313 | * ceph_destroy_options() when it is no longer needed. | |
4314 | * rbd_opts | |
4315 | * Address of an rbd options pointer. Fully initialized by | |
4316 | * this function; caller must release with kfree(). | |
4317 | * spec | |
4318 | * Address of an rbd image specification pointer. Fully | |
4319 | * initialized by this function based on parsed options. | |
4320 | * Caller must release with rbd_spec_put(). | |
4321 | * | |
4322 | * The options passed take this form: | |
4323 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4324 | * where: | |
4325 | * <mon_addrs> | |
4326 | * A comma-separated list of one or more monitor addresses. | |
4327 | * A monitor address is an ip address, optionally followed | |
4328 | * by a port number (separated by a colon). | |
4329 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4330 | * <options> | |
4331 | * A comma-separated list of ceph and/or rbd options. | |
4332 | * <pool_name> | |
4333 | * The name of the rados pool containing the rbd image. | |
4334 | * <image_name> | |
4335 | * The name of the image in that pool to map. | |
4336 | * <snap_id> | |
4337 | * An optional snapshot id. If provided, the mapping will | |
4338 | * present data from the image at the time that snapshot was | |
4339 | * created. The image head is used if no snapshot id is | |
4340 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4341 | */ |
859c31df | 4342 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4343 | struct ceph_options **ceph_opts, |
859c31df AE |
4344 | struct rbd_options **opts, |
4345 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4346 | { |
d22f76e7 | 4347 | size_t len; |
859c31df | 4348 | char *options; |
0ddebc0c | 4349 | const char *mon_addrs; |
ecb4dc22 | 4350 | char *snap_name; |
0ddebc0c | 4351 | size_t mon_addrs_size; |
859c31df | 4352 | struct rbd_spec *spec = NULL; |
4e9afeba | 4353 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4354 | struct ceph_options *copts; |
dc79b113 | 4355 | int ret; |
e28fff26 AE |
4356 | |
4357 | /* The first four tokens are required */ | |
4358 | ||
7ef3214a | 4359 | len = next_token(&buf); |
4fb5d671 AE |
4360 | if (!len) { |
4361 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4362 | return -EINVAL; | |
4363 | } | |
0ddebc0c | 4364 | mon_addrs = buf; |
f28e565a | 4365 | mon_addrs_size = len + 1; |
7ef3214a | 4366 | buf += len; |
a725f65e | 4367 | |
dc79b113 | 4368 | ret = -EINVAL; |
f28e565a AE |
4369 | options = dup_token(&buf, NULL); |
4370 | if (!options) | |
dc79b113 | 4371 | return -ENOMEM; |
4fb5d671 AE |
4372 | if (!*options) { |
4373 | rbd_warn(NULL, "no options provided"); | |
4374 | goto out_err; | |
4375 | } | |
e28fff26 | 4376 | |
859c31df AE |
4377 | spec = rbd_spec_alloc(); |
4378 | if (!spec) | |
f28e565a | 4379 | goto out_mem; |
859c31df AE |
4380 | |
4381 | spec->pool_name = dup_token(&buf, NULL); | |
4382 | if (!spec->pool_name) | |
4383 | goto out_mem; | |
4fb5d671 AE |
4384 | if (!*spec->pool_name) { |
4385 | rbd_warn(NULL, "no pool name provided"); | |
4386 | goto out_err; | |
4387 | } | |
e28fff26 | 4388 | |
69e7a02f | 4389 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4390 | if (!spec->image_name) |
f28e565a | 4391 | goto out_mem; |
4fb5d671 AE |
4392 | if (!*spec->image_name) { |
4393 | rbd_warn(NULL, "no image name provided"); | |
4394 | goto out_err; | |
4395 | } | |
d4b125e9 | 4396 | |
f28e565a AE |
4397 | /* |
4398 | * Snapshot name is optional; default is to use "-" | |
4399 | * (indicating the head/no snapshot). | |
4400 | */ | |
3feeb894 | 4401 | len = next_token(&buf); |
820a5f3e | 4402 | if (!len) { |
3feeb894 AE |
4403 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4404 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4405 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4406 | ret = -ENAMETOOLONG; |
f28e565a | 4407 | goto out_err; |
849b4260 | 4408 | } |
ecb4dc22 AE |
4409 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4410 | if (!snap_name) | |
f28e565a | 4411 | goto out_mem; |
ecb4dc22 AE |
4412 | *(snap_name + len) = '\0'; |
4413 | spec->snap_name = snap_name; | |
e5c35534 | 4414 | |
0ddebc0c | 4415 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4416 | |
4e9afeba AE |
4417 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4418 | if (!rbd_opts) | |
4419 | goto out_mem; | |
4420 | ||
4421 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4422 | |
859c31df | 4423 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4424 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4425 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4426 | if (IS_ERR(copts)) { |
4427 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4428 | goto out_err; |
4429 | } | |
859c31df AE |
4430 | kfree(options); |
4431 | ||
4432 | *ceph_opts = copts; | |
4e9afeba | 4433 | *opts = rbd_opts; |
859c31df | 4434 | *rbd_spec = spec; |
0ddebc0c | 4435 | |
dc79b113 | 4436 | return 0; |
f28e565a | 4437 | out_mem: |
dc79b113 | 4438 | ret = -ENOMEM; |
d22f76e7 | 4439 | out_err: |
859c31df AE |
4440 | kfree(rbd_opts); |
4441 | rbd_spec_put(spec); | |
f28e565a | 4442 | kfree(options); |
d22f76e7 | 4443 | |
dc79b113 | 4444 | return ret; |
a725f65e AE |
4445 | } |
4446 | ||
589d30e0 AE |
4447 | /* |
4448 | * An rbd format 2 image has a unique identifier, distinct from the | |
4449 | * name given to it by the user. Internally, that identifier is | |
4450 | * what's used to specify the names of objects related to the image. | |
4451 | * | |
4452 | * A special "rbd id" object is used to map an rbd image name to its | |
4453 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4454 | * with the supplied name. | |
4455 | * | |
4456 | * This function will record the given rbd_dev's image_id field if | |
4457 | * it can be determined, and in that case will return 0. If any | |
4458 | * errors occur a negative errno will be returned and the rbd_dev's | |
4459 | * image_id field will be unchanged (and should be NULL). | |
4460 | */ | |
4461 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4462 | { | |
4463 | int ret; | |
4464 | size_t size; | |
4465 | char *object_name; | |
4466 | void *response; | |
c0fba368 | 4467 | char *image_id; |
2f82ee54 | 4468 | |
2c0d0a10 AE |
4469 | /* |
4470 | * When probing a parent image, the image id is already | |
4471 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4472 | * need to fetch the image id again in this case. We |
4473 | * do still need to set the image format though. | |
2c0d0a10 | 4474 | */ |
c0fba368 AE |
4475 | if (rbd_dev->spec->image_id) { |
4476 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4477 | ||
2c0d0a10 | 4478 | return 0; |
c0fba368 | 4479 | } |
2c0d0a10 | 4480 | |
589d30e0 AE |
4481 | /* |
4482 | * First, see if the format 2 image id file exists, and if | |
4483 | * so, get the image's persistent id from it. | |
4484 | */ | |
69e7a02f | 4485 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4486 | object_name = kmalloc(size, GFP_NOIO); |
4487 | if (!object_name) | |
4488 | return -ENOMEM; | |
0d7dbfce | 4489 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4490 | dout("rbd id object name is %s\n", object_name); |
4491 | ||
4492 | /* Response will be an encoded string, which includes a length */ | |
4493 | ||
4494 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4495 | response = kzalloc(size, GFP_NOIO); | |
4496 | if (!response) { | |
4497 | ret = -ENOMEM; | |
4498 | goto out; | |
4499 | } | |
4500 | ||
c0fba368 AE |
4501 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4502 | ||
36be9a76 | 4503 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4504 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4505 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4506 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4507 | if (ret == -ENOENT) { |
4508 | image_id = kstrdup("", GFP_KERNEL); | |
4509 | ret = image_id ? 0 : -ENOMEM; | |
4510 | if (!ret) | |
4511 | rbd_dev->image_format = 1; | |
4512 | } else if (ret > sizeof (__le32)) { | |
4513 | void *p = response; | |
4514 | ||
4515 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4516 | NULL, GFP_NOIO); |
c0fba368 AE |
4517 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4518 | if (!ret) | |
4519 | rbd_dev->image_format = 2; | |
589d30e0 | 4520 | } else { |
c0fba368 AE |
4521 | ret = -EINVAL; |
4522 | } | |
4523 | ||
4524 | if (!ret) { | |
4525 | rbd_dev->spec->image_id = image_id; | |
4526 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4527 | } |
4528 | out: | |
4529 | kfree(response); | |
4530 | kfree(object_name); | |
4531 | ||
4532 | return ret; | |
4533 | } | |
4534 | ||
6fd48b3b AE |
4535 | /* Undo whatever state changes are made by v1 or v2 image probe */ |
4536 | ||
4537 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) | |
4538 | { | |
4539 | struct rbd_image_header *header; | |
4540 | ||
4541 | rbd_dev_remove_parent(rbd_dev); | |
4542 | rbd_spec_put(rbd_dev->parent_spec); | |
4543 | rbd_dev->parent_spec = NULL; | |
4544 | rbd_dev->parent_overlap = 0; | |
4545 | ||
4546 | /* Free dynamic fields from the header, then zero it out */ | |
4547 | ||
4548 | header = &rbd_dev->header; | |
812164f8 | 4549 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4550 | kfree(header->snap_sizes); |
4551 | kfree(header->snap_names); | |
4552 | kfree(header->object_prefix); | |
4553 | memset(header, 0, sizeof (*header)); | |
4554 | } | |
4555 | ||
a30b71b9 AE |
4556 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
4557 | { | |
4558 | int ret; | |
a30b71b9 AE |
4559 | |
4560 | /* Populate rbd image metadata */ | |
4561 | ||
4562 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
4563 | if (ret < 0) | |
4564 | goto out_err; | |
86b00e0d AE |
4565 | |
4566 | /* Version 1 images have no parent (no layering) */ | |
4567 | ||
4568 | rbd_dev->parent_spec = NULL; | |
4569 | rbd_dev->parent_overlap = 0; | |
4570 | ||
a30b71b9 AE |
4571 | dout("discovered version 1 image, header name is %s\n", |
4572 | rbd_dev->header_name); | |
4573 | ||
4574 | return 0; | |
4575 | ||
4576 | out_err: | |
4577 | kfree(rbd_dev->header_name); | |
4578 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
4579 | kfree(rbd_dev->spec->image_id); |
4580 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
4581 | |
4582 | return ret; | |
4583 | } | |
4584 | ||
4585 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
4586 | { | |
9d475de5 | 4587 | int ret; |
a30b71b9 | 4588 | |
9d475de5 | 4589 | ret = rbd_dev_v2_image_size(rbd_dev); |
57385b51 | 4590 | if (ret) |
1e130199 AE |
4591 | goto out_err; |
4592 | ||
4593 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
4594 | ||
4595 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
57385b51 | 4596 | if (ret) |
b1b5402a AE |
4597 | goto out_err; |
4598 | ||
d889140c | 4599 | /* Get the and check features for the image */ |
b1b5402a AE |
4600 | |
4601 | ret = rbd_dev_v2_features(rbd_dev); | |
57385b51 | 4602 | if (ret) |
9d475de5 | 4603 | goto out_err; |
35d489f9 | 4604 | |
86b00e0d AE |
4605 | /* If the image supports layering, get the parent info */ |
4606 | ||
4607 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
4608 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
57385b51 | 4609 | if (ret) |
86b00e0d | 4610 | goto out_err; |
96882f55 AE |
4611 | |
4612 | /* | |
4613 | * Don't print a warning for parent images. We can | |
4614 | * tell this point because we won't know its pool | |
4615 | * name yet (just its pool id). | |
4616 | */ | |
4617 | if (rbd_dev->spec->pool_name) | |
4618 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4619 | "is EXPERIMENTAL!"); | |
86b00e0d AE |
4620 | } |
4621 | ||
cc070d59 AE |
4622 | /* If the image supports fancy striping, get its parameters */ |
4623 | ||
4624 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4625 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4626 | if (ret < 0) | |
4627 | goto out_err; | |
4628 | } | |
4629 | ||
6e14b1a6 AE |
4630 | /* crypto and compression type aren't (yet) supported for v2 images */ |
4631 | ||
4632 | rbd_dev->header.crypt_type = 0; | |
4633 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 4634 | |
6e14b1a6 AE |
4635 | /* Get the snapshot context, plus the header version */ |
4636 | ||
cc4a38bd | 4637 | ret = rbd_dev_v2_snap_context(rbd_dev); |
35d489f9 AE |
4638 | if (ret) |
4639 | goto out_err; | |
6e14b1a6 | 4640 | |
a30b71b9 AE |
4641 | dout("discovered version 2 image, header name is %s\n", |
4642 | rbd_dev->header_name); | |
4643 | ||
35152979 | 4644 | return 0; |
9d475de5 | 4645 | out_err: |
86b00e0d AE |
4646 | rbd_dev->parent_overlap = 0; |
4647 | rbd_spec_put(rbd_dev->parent_spec); | |
4648 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
4649 | kfree(rbd_dev->header_name); |
4650 | rbd_dev->header_name = NULL; | |
1e130199 AE |
4651 | kfree(rbd_dev->header.object_prefix); |
4652 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4653 | |
4654 | return ret; | |
a30b71b9 AE |
4655 | } |
4656 | ||
124afba2 | 4657 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4658 | { |
2f82ee54 | 4659 | struct rbd_device *parent = NULL; |
124afba2 AE |
4660 | struct rbd_spec *parent_spec; |
4661 | struct rbd_client *rbdc; | |
4662 | int ret; | |
4663 | ||
4664 | if (!rbd_dev->parent_spec) | |
4665 | return 0; | |
4666 | /* | |
4667 | * We need to pass a reference to the client and the parent | |
4668 | * spec when creating the parent rbd_dev. Images related by | |
4669 | * parent/child relationships always share both. | |
4670 | */ | |
4671 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4672 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4673 | ||
4674 | ret = -ENOMEM; | |
4675 | parent = rbd_dev_create(rbdc, parent_spec); | |
4676 | if (!parent) | |
4677 | goto out_err; | |
4678 | ||
4679 | ret = rbd_dev_image_probe(parent); | |
4680 | if (ret < 0) | |
4681 | goto out_err; | |
4682 | rbd_dev->parent = parent; | |
4683 | ||
4684 | return 0; | |
4685 | out_err: | |
4686 | if (parent) { | |
4687 | rbd_spec_put(rbd_dev->parent_spec); | |
4688 | kfree(rbd_dev->header_name); | |
4689 | rbd_dev_destroy(parent); | |
4690 | } else { | |
4691 | rbd_put_client(rbdc); | |
4692 | rbd_spec_put(parent_spec); | |
4693 | } | |
4694 | ||
4695 | return ret; | |
4696 | } | |
4697 | ||
200a6a8b | 4698 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4699 | { |
83a06263 | 4700 | int ret; |
d1cf5788 AE |
4701 | |
4702 | ret = rbd_dev_mapping_set(rbd_dev); | |
83a06263 | 4703 | if (ret) |
9bb81c9b | 4704 | return ret; |
5de10f3b | 4705 | |
83a06263 AE |
4706 | /* generate unique id: find highest unique id, add one */ |
4707 | rbd_dev_id_get(rbd_dev); | |
4708 | ||
4709 | /* Fill in the device name, now that we have its id. */ | |
4710 | BUILD_BUG_ON(DEV_NAME_LEN | |
4711 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4712 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4713 | ||
4714 | /* Get our block major device number. */ | |
4715 | ||
4716 | ret = register_blkdev(0, rbd_dev->name); | |
4717 | if (ret < 0) | |
4718 | goto err_out_id; | |
4719 | rbd_dev->major = ret; | |
4720 | ||
4721 | /* Set up the blkdev mapping. */ | |
4722 | ||
4723 | ret = rbd_init_disk(rbd_dev); | |
4724 | if (ret) | |
4725 | goto err_out_blkdev; | |
4726 | ||
4727 | ret = rbd_bus_add_dev(rbd_dev); | |
4728 | if (ret) | |
4729 | goto err_out_disk; | |
4730 | ||
83a06263 AE |
4731 | /* Everything's ready. Announce the disk to the world. */ |
4732 | ||
b5156e76 | 4733 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
129b79d4 | 4734 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4735 | add_disk(rbd_dev->disk); |
4736 | ||
4737 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4738 | (unsigned long long) rbd_dev->mapping.size); | |
4739 | ||
4740 | return ret; | |
2f82ee54 | 4741 | |
83a06263 AE |
4742 | err_out_disk: |
4743 | rbd_free_disk(rbd_dev); | |
4744 | err_out_blkdev: | |
4745 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4746 | err_out_id: | |
4747 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4748 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4749 | |
4750 | return ret; | |
4751 | } | |
4752 | ||
332bb12d AE |
4753 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4754 | { | |
4755 | struct rbd_spec *spec = rbd_dev->spec; | |
4756 | size_t size; | |
4757 | ||
4758 | /* Record the header object name for this rbd image. */ | |
4759 | ||
4760 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4761 | ||
4762 | if (rbd_dev->image_format == 1) | |
4763 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4764 | else | |
4765 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4766 | ||
4767 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4768 | if (!rbd_dev->header_name) | |
4769 | return -ENOMEM; | |
4770 | ||
4771 | if (rbd_dev->image_format == 1) | |
4772 | sprintf(rbd_dev->header_name, "%s%s", | |
4773 | spec->image_name, RBD_SUFFIX); | |
4774 | else | |
4775 | sprintf(rbd_dev->header_name, "%s%s", | |
4776 | RBD_HEADER_PREFIX, spec->image_id); | |
4777 | return 0; | |
4778 | } | |
4779 | ||
200a6a8b AE |
4780 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4781 | { | |
6fd48b3b AE |
4782 | int ret; |
4783 | ||
4784 | rbd_remove_all_snaps(rbd_dev); | |
4785 | rbd_dev_unprobe(rbd_dev); | |
4786 | ret = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4787 | if (ret) | |
4788 | rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret); | |
200a6a8b | 4789 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4790 | rbd_dev->header_name = NULL; |
4791 | rbd_dev->image_format = 0; | |
4792 | kfree(rbd_dev->spec->image_id); | |
4793 | rbd_dev->spec->image_id = NULL; | |
4794 | ||
200a6a8b AE |
4795 | rbd_dev_destroy(rbd_dev); |
4796 | } | |
4797 | ||
a30b71b9 AE |
4798 | /* |
4799 | * Probe for the existence of the header object for the given rbd | |
4800 | * device. For format 2 images this includes determining the image | |
4801 | * id. | |
4802 | */ | |
71f293e2 | 4803 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4804 | { |
4805 | int ret; | |
b644de2b | 4806 | int tmp; |
a30b71b9 AE |
4807 | |
4808 | /* | |
4809 | * Get the id from the image id object. If it's not a | |
4810 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4811 | * it's a format 1 image. | |
4812 | */ | |
4813 | ret = rbd_dev_image_id(rbd_dev); | |
4814 | if (ret) | |
c0fba368 AE |
4815 | return ret; |
4816 | rbd_assert(rbd_dev->spec->image_id); | |
4817 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4818 | ||
332bb12d AE |
4819 | ret = rbd_dev_header_name(rbd_dev); |
4820 | if (ret) | |
4821 | goto err_out_format; | |
4822 | ||
b644de2b AE |
4823 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
4824 | if (ret) | |
4825 | goto out_header_name; | |
4826 | ||
c0fba368 | 4827 | if (rbd_dev->image_format == 1) |
a30b71b9 AE |
4828 | ret = rbd_dev_v1_probe(rbd_dev); |
4829 | else | |
4830 | ret = rbd_dev_v2_probe(rbd_dev); | |
5655c4d9 | 4831 | if (ret) |
b644de2b | 4832 | goto err_out_watch; |
83a06263 | 4833 | |
9bb81c9b AE |
4834 | ret = rbd_dev_snaps_update(rbd_dev); |
4835 | if (ret) | |
6fd48b3b | 4836 | goto err_out_probe; |
9bb81c9b AE |
4837 | |
4838 | ret = rbd_dev_spec_update(rbd_dev); | |
4839 | if (ret) | |
4840 | goto err_out_snaps; | |
4841 | ||
4842 | ret = rbd_dev_probe_parent(rbd_dev); | |
6fd48b3b AE |
4843 | if (!ret) |
4844 | return 0; | |
83a06263 | 4845 | |
9bb81c9b AE |
4846 | err_out_snaps: |
4847 | rbd_remove_all_snaps(rbd_dev); | |
6fd48b3b AE |
4848 | err_out_probe: |
4849 | rbd_dev_unprobe(rbd_dev); | |
b644de2b AE |
4850 | err_out_watch: |
4851 | tmp = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4852 | if (tmp) | |
4853 | rbd_warn(rbd_dev, "unable to tear down watch request\n"); | |
332bb12d AE |
4854 | out_header_name: |
4855 | kfree(rbd_dev->header_name); | |
4856 | rbd_dev->header_name = NULL; | |
4857 | err_out_format: | |
4858 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
4859 | kfree(rbd_dev->spec->image_id); |
4860 | rbd_dev->spec->image_id = NULL; | |
4861 | ||
4862 | dout("probe failed, returning %d\n", ret); | |
4863 | ||
a30b71b9 AE |
4864 | return ret; |
4865 | } | |
4866 | ||
59c2be1e YS |
4867 | static ssize_t rbd_add(struct bus_type *bus, |
4868 | const char *buf, | |
4869 | size_t count) | |
602adf40 | 4870 | { |
cb8627c7 | 4871 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4872 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4873 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4874 | struct rbd_spec *spec = NULL; |
9d3997fd | 4875 | struct rbd_client *rbdc; |
27cc2594 AE |
4876 | struct ceph_osd_client *osdc; |
4877 | int rc = -ENOMEM; | |
602adf40 YS |
4878 | |
4879 | if (!try_module_get(THIS_MODULE)) | |
4880 | return -ENODEV; | |
4881 | ||
602adf40 | 4882 | /* parse add command */ |
859c31df | 4883 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4884 | if (rc < 0) |
bd4ba655 | 4885 | goto err_out_module; |
78cea76e | 4886 | |
9d3997fd AE |
4887 | rbdc = rbd_get_client(ceph_opts); |
4888 | if (IS_ERR(rbdc)) { | |
4889 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4890 | goto err_out_args; |
9d3997fd | 4891 | } |
c53d5893 | 4892 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4893 | |
602adf40 | 4894 | /* pick the pool */ |
9d3997fd | 4895 | osdc = &rbdc->client->osdc; |
859c31df | 4896 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4897 | if (rc < 0) |
4898 | goto err_out_client; | |
c0cd10db | 4899 | spec->pool_id = (u64)rc; |
859c31df | 4900 | |
0903e875 AE |
4901 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4902 | ||
c0cd10db AE |
4903 | if (spec->pool_id > (u64)U32_MAX) { |
4904 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
4905 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
4906 | rc = -EIO; |
4907 | goto err_out_client; | |
4908 | } | |
4909 | ||
c53d5893 | 4910 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4911 | if (!rbd_dev) |
4912 | goto err_out_client; | |
c53d5893 AE |
4913 | rbdc = NULL; /* rbd_dev now owns this */ |
4914 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4915 | |
bd4ba655 | 4916 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4917 | kfree(rbd_opts); |
4918 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4919 | |
71f293e2 | 4920 | rc = rbd_dev_image_probe(rbd_dev); |
a30b71b9 | 4921 | if (rc < 0) |
c53d5893 | 4922 | goto err_out_rbd_dev; |
05fd6f6f | 4923 | |
b536f69a AE |
4924 | rc = rbd_dev_device_setup(rbd_dev); |
4925 | if (!rc) | |
4926 | return count; | |
4927 | ||
4928 | rbd_dev_image_release(rbd_dev); | |
c53d5893 AE |
4929 | err_out_rbd_dev: |
4930 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 4931 | err_out_client: |
9d3997fd | 4932 | rbd_put_client(rbdc); |
0ddebc0c | 4933 | err_out_args: |
78cea76e AE |
4934 | if (ceph_opts) |
4935 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4936 | kfree(rbd_opts); |
859c31df | 4937 | rbd_spec_put(spec); |
bd4ba655 AE |
4938 | err_out_module: |
4939 | module_put(THIS_MODULE); | |
27cc2594 | 4940 | |
602adf40 | 4941 | dout("Error adding device %s\n", buf); |
27cc2594 | 4942 | |
c0cd10db | 4943 | return (ssize_t)rc; |
602adf40 YS |
4944 | } |
4945 | ||
de71a297 | 4946 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4947 | { |
4948 | struct list_head *tmp; | |
4949 | struct rbd_device *rbd_dev; | |
4950 | ||
e124a82f | 4951 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4952 | list_for_each(tmp, &rbd_dev_list) { |
4953 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4954 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4955 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4956 | return rbd_dev; |
e124a82f | 4957 | } |
602adf40 | 4958 | } |
e124a82f | 4959 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
4960 | return NULL; |
4961 | } | |
4962 | ||
200a6a8b | 4963 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 4964 | { |
593a9e7b | 4965 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4966 | |
602adf40 | 4967 | rbd_free_disk(rbd_dev); |
200a6a8b AE |
4968 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
4969 | rbd_dev_clear_mapping(rbd_dev); | |
602adf40 | 4970 | unregister_blkdev(rbd_dev->major, rbd_dev->name); |
200a6a8b | 4971 | rbd_dev->major = 0; |
e2839308 | 4972 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 4973 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
4974 | } |
4975 | ||
05a46afd AE |
4976 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
4977 | { | |
ad945fc1 | 4978 | while (rbd_dev->parent) { |
05a46afd AE |
4979 | struct rbd_device *first = rbd_dev; |
4980 | struct rbd_device *second = first->parent; | |
4981 | struct rbd_device *third; | |
4982 | ||
4983 | /* | |
4984 | * Follow to the parent with no grandparent and | |
4985 | * remove it. | |
4986 | */ | |
4987 | while (second && (third = second->parent)) { | |
4988 | first = second; | |
4989 | second = third; | |
4990 | } | |
ad945fc1 | 4991 | rbd_assert(second); |
8ad42cd0 | 4992 | rbd_dev_image_release(second); |
ad945fc1 AE |
4993 | first->parent = NULL; |
4994 | first->parent_overlap = 0; | |
4995 | ||
4996 | rbd_assert(first->parent_spec); | |
05a46afd AE |
4997 | rbd_spec_put(first->parent_spec); |
4998 | first->parent_spec = NULL; | |
05a46afd AE |
4999 | } |
5000 | } | |
5001 | ||
dfc5606d YS |
5002 | static ssize_t rbd_remove(struct bus_type *bus, |
5003 | const char *buf, | |
5004 | size_t count) | |
602adf40 YS |
5005 | { |
5006 | struct rbd_device *rbd_dev = NULL; | |
0d8189e1 | 5007 | int target_id; |
602adf40 | 5008 | unsigned long ul; |
0d8189e1 | 5009 | int ret; |
602adf40 | 5010 | |
0d8189e1 AE |
5011 | ret = strict_strtoul(buf, 10, &ul); |
5012 | if (ret) | |
5013 | return ret; | |
602adf40 YS |
5014 | |
5015 | /* convert to int; abort if we lost anything in the conversion */ | |
5016 | target_id = (int) ul; | |
5017 | if (target_id != ul) | |
5018 | return -EINVAL; | |
5019 | ||
5020 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
5021 | ||
5022 | rbd_dev = __rbd_get_dev(target_id); | |
5023 | if (!rbd_dev) { | |
5024 | ret = -ENOENT; | |
5025 | goto done; | |
42382b70 AE |
5026 | } |
5027 | ||
a14ea269 | 5028 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 5029 | if (rbd_dev->open_count) |
42382b70 | 5030 | ret = -EBUSY; |
b82d167b AE |
5031 | else |
5032 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 5033 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 5034 | if (ret < 0) |
42382b70 | 5035 | goto done; |
0d8189e1 | 5036 | ret = count; |
b480815a | 5037 | rbd_bus_del_dev(rbd_dev); |
8ad42cd0 | 5038 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5039 | module_put(THIS_MODULE); |
602adf40 YS |
5040 | done: |
5041 | mutex_unlock(&ctl_mutex); | |
aafb230e | 5042 | |
602adf40 YS |
5043 | return ret; |
5044 | } | |
5045 | ||
602adf40 YS |
5046 | /* |
5047 | * create control files in sysfs | |
dfc5606d | 5048 | * /sys/bus/rbd/... |
602adf40 YS |
5049 | */ |
5050 | static int rbd_sysfs_init(void) | |
5051 | { | |
dfc5606d | 5052 | int ret; |
602adf40 | 5053 | |
fed4c143 | 5054 | ret = device_register(&rbd_root_dev); |
21079786 | 5055 | if (ret < 0) |
dfc5606d | 5056 | return ret; |
602adf40 | 5057 | |
fed4c143 AE |
5058 | ret = bus_register(&rbd_bus_type); |
5059 | if (ret < 0) | |
5060 | device_unregister(&rbd_root_dev); | |
602adf40 | 5061 | |
602adf40 YS |
5062 | return ret; |
5063 | } | |
5064 | ||
5065 | static void rbd_sysfs_cleanup(void) | |
5066 | { | |
dfc5606d | 5067 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5068 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5069 | } |
5070 | ||
cc344fa1 | 5071 | static int __init rbd_init(void) |
602adf40 YS |
5072 | { |
5073 | int rc; | |
5074 | ||
1e32d34c AE |
5075 | if (!libceph_compatible(NULL)) { |
5076 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
5077 | ||
5078 | return -EINVAL; | |
5079 | } | |
602adf40 YS |
5080 | rc = rbd_sysfs_init(); |
5081 | if (rc) | |
5082 | return rc; | |
f0f8cef5 | 5083 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
5084 | return 0; |
5085 | } | |
5086 | ||
cc344fa1 | 5087 | static void __exit rbd_exit(void) |
602adf40 YS |
5088 | { |
5089 | rbd_sysfs_cleanup(); | |
5090 | } | |
5091 | ||
5092 | module_init(rbd_init); | |
5093 | module_exit(rbd_exit); | |
5094 | ||
5095 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
5096 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
5097 | MODULE_DESCRIPTION("rados block device"); | |
5098 | ||
5099 | /* following authorship retained from original osdblk.c */ | |
5100 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5101 | ||
5102 | MODULE_LICENSE("GPL"); |