]>
Commit | Line | Data |
---|---|---|
602adf40 YS |
1 | /* |
2 | rbd.c -- Export ceph rados objects as a Linux block device | |
3 | ||
4 | ||
5 | based on drivers/block/osdblk.c: | |
6 | ||
7 | Copyright 2009 Red Hat, Inc. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; see the file COPYING. If not, write to | |
20 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | ||
22 | ||
23 | ||
dfc5606d | 24 | For usage instructions, please refer to: |
602adf40 | 25 | |
dfc5606d | 26 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
27 | |
28 | */ | |
29 | ||
30 | #include <linux/ceph/libceph.h> | |
31 | #include <linux/ceph/osd_client.h> | |
32 | #include <linux/ceph/mon_client.h> | |
33 | #include <linux/ceph/decode.h> | |
59c2be1e | 34 | #include <linux/parser.h> |
602adf40 YS |
35 | |
36 | #include <linux/kernel.h> | |
37 | #include <linux/device.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/fs.h> | |
40 | #include <linux/blkdev.h> | |
41 | ||
42 | #include "rbd_types.h" | |
43 | ||
aafb230e AE |
44 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
45 | ||
593a9e7b AE |
46 | /* |
47 | * The basic unit of block I/O is a sector. It is interpreted in a | |
48 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
49 | * universally 512 bytes. These symbols are just slightly more | |
50 | * meaningful than the bare numbers they represent. | |
51 | */ | |
52 | #define SECTOR_SHIFT 9 | |
53 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
54 | ||
f0f8cef5 AE |
55 | #define RBD_DRV_NAME "rbd" |
56 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
57 | |
58 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
59 | ||
d4b125e9 AE |
60 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
61 | #define RBD_MAX_SNAP_NAME_LEN \ | |
62 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
63 | ||
35d489f9 | 64 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
65 | |
66 | #define RBD_SNAP_HEAD_NAME "-" | |
67 | ||
9e15b77d AE |
68 | /* This allows a single page to hold an image name sent by OSD */ |
69 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 70 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 71 | |
1e130199 | 72 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 73 | |
d889140c AE |
74 | /* Feature bits */ |
75 | ||
5cbf6f12 AE |
76 | #define RBD_FEATURE_LAYERING (1<<0) |
77 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
78 | #define RBD_FEATURES_ALL \ | |
79 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
80 | |
81 | /* Features supported by this (client software) implementation. */ | |
82 | ||
770eba6e | 83 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 84 | |
81a89793 AE |
85 | /* |
86 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
87 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
88 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
89 | * enough to hold all possible device names. | |
90 | */ | |
602adf40 | 91 | #define DEV_NAME_LEN 32 |
81a89793 | 92 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
93 | |
94 | /* | |
95 | * block device image metadata (in-memory version) | |
96 | */ | |
97 | struct rbd_image_header { | |
f84344f3 | 98 | /* These four fields never change for a given rbd image */ |
849b4260 | 99 | char *object_prefix; |
34b13184 | 100 | u64 features; |
602adf40 YS |
101 | __u8 obj_order; |
102 | __u8 crypt_type; | |
103 | __u8 comp_type; | |
602adf40 | 104 | |
f84344f3 AE |
105 | /* The remaining fields need to be updated occasionally */ |
106 | u64 image_size; | |
107 | struct ceph_snap_context *snapc; | |
602adf40 YS |
108 | char *snap_names; |
109 | u64 *snap_sizes; | |
59c2be1e | 110 | |
500d0c0f AE |
111 | u64 stripe_unit; |
112 | u64 stripe_count; | |
113 | ||
59c2be1e YS |
114 | u64 obj_version; |
115 | }; | |
116 | ||
0d7dbfce AE |
117 | /* |
118 | * An rbd image specification. | |
119 | * | |
120 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
121 | * identify an image. Each rbd_dev structure includes a pointer to |
122 | * an rbd_spec structure that encapsulates this identity. | |
123 | * | |
124 | * Each of the id's in an rbd_spec has an associated name. For a | |
125 | * user-mapped image, the names are supplied and the id's associated | |
126 | * with them are looked up. For a layered image, a parent image is | |
127 | * defined by the tuple, and the names are looked up. | |
128 | * | |
129 | * An rbd_dev structure contains a parent_spec pointer which is | |
130 | * non-null if the image it represents is a child in a layered | |
131 | * image. This pointer will refer to the rbd_spec structure used | |
132 | * by the parent rbd_dev for its own identity (i.e., the structure | |
133 | * is shared between the parent and child). | |
134 | * | |
135 | * Since these structures are populated once, during the discovery | |
136 | * phase of image construction, they are effectively immutable so | |
137 | * we make no effort to synchronize access to them. | |
138 | * | |
139 | * Note that code herein does not assume the image name is known (it | |
140 | * could be a null pointer). | |
0d7dbfce AE |
141 | */ |
142 | struct rbd_spec { | |
143 | u64 pool_id; | |
ecb4dc22 | 144 | const char *pool_name; |
0d7dbfce | 145 | |
ecb4dc22 AE |
146 | const char *image_id; |
147 | const char *image_name; | |
0d7dbfce AE |
148 | |
149 | u64 snap_id; | |
ecb4dc22 | 150 | const char *snap_name; |
0d7dbfce AE |
151 | |
152 | struct kref kref; | |
153 | }; | |
154 | ||
602adf40 | 155 | /* |
f0f8cef5 | 156 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
157 | */ |
158 | struct rbd_client { | |
159 | struct ceph_client *client; | |
160 | struct kref kref; | |
161 | struct list_head node; | |
162 | }; | |
163 | ||
bf0d5f50 AE |
164 | struct rbd_img_request; |
165 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
166 | ||
167 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
168 | ||
169 | struct rbd_obj_request; | |
170 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
171 | ||
9969ebc5 AE |
172 | enum obj_request_type { |
173 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
174 | }; | |
bf0d5f50 | 175 | |
926f9b3f AE |
176 | enum obj_req_flags { |
177 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 178 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
179 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
180 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
181 | }; |
182 | ||
bf0d5f50 AE |
183 | struct rbd_obj_request { |
184 | const char *object_name; | |
185 | u64 offset; /* object start byte */ | |
186 | u64 length; /* bytes from offset */ | |
926f9b3f | 187 | unsigned long flags; |
bf0d5f50 | 188 | |
c5b5ef6c AE |
189 | /* |
190 | * An object request associated with an image will have its | |
191 | * img_data flag set; a standalone object request will not. | |
192 | * | |
193 | * A standalone object request will have which == BAD_WHICH | |
194 | * and a null obj_request pointer. | |
195 | * | |
196 | * An object request initiated in support of a layered image | |
197 | * object (to check for its existence before a write) will | |
198 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
199 | * | |
200 | * Finally, an object request for rbd image data will have | |
201 | * which != BAD_WHICH, and will have a non-null img_request | |
202 | * pointer. The value of which will be in the range | |
203 | * 0..(img_request->obj_request_count-1). | |
204 | */ | |
205 | union { | |
206 | struct rbd_obj_request *obj_request; /* STAT op */ | |
207 | struct { | |
208 | struct rbd_img_request *img_request; | |
209 | u64 img_offset; | |
210 | /* links for img_request->obj_requests list */ | |
211 | struct list_head links; | |
212 | }; | |
213 | }; | |
bf0d5f50 AE |
214 | u32 which; /* posn image request list */ |
215 | ||
216 | enum obj_request_type type; | |
788e2df3 AE |
217 | union { |
218 | struct bio *bio_list; | |
219 | struct { | |
220 | struct page **pages; | |
221 | u32 page_count; | |
222 | }; | |
223 | }; | |
0eefd470 | 224 | struct page **copyup_pages; |
bf0d5f50 AE |
225 | |
226 | struct ceph_osd_request *osd_req; | |
227 | ||
228 | u64 xferred; /* bytes transferred */ | |
229 | u64 version; | |
1b83bef2 | 230 | int result; |
bf0d5f50 AE |
231 | |
232 | rbd_obj_callback_t callback; | |
788e2df3 | 233 | struct completion completion; |
bf0d5f50 AE |
234 | |
235 | struct kref kref; | |
236 | }; | |
237 | ||
0c425248 | 238 | enum img_req_flags { |
9849e986 AE |
239 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
240 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 241 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
242 | }; |
243 | ||
bf0d5f50 | 244 | struct rbd_img_request { |
bf0d5f50 AE |
245 | struct rbd_device *rbd_dev; |
246 | u64 offset; /* starting image byte offset */ | |
247 | u64 length; /* byte count from offset */ | |
0c425248 | 248 | unsigned long flags; |
bf0d5f50 | 249 | union { |
9849e986 | 250 | u64 snap_id; /* for reads */ |
bf0d5f50 | 251 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
252 | }; |
253 | union { | |
254 | struct request *rq; /* block request */ | |
255 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 256 | }; |
3d7efd18 | 257 | struct page **copyup_pages; |
bf0d5f50 AE |
258 | spinlock_t completion_lock;/* protects next_completion */ |
259 | u32 next_completion; | |
260 | rbd_img_callback_t callback; | |
55f27e09 | 261 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 262 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
263 | |
264 | u32 obj_request_count; | |
265 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
266 | ||
267 | struct kref kref; | |
268 | }; | |
269 | ||
270 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 271 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 272 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 273 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 274 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 275 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 276 | |
dfc5606d | 277 | struct rbd_snap { |
dfc5606d | 278 | const char *name; |
3591538f | 279 | u64 size; |
dfc5606d YS |
280 | struct list_head node; |
281 | u64 id; | |
34b13184 | 282 | u64 features; |
dfc5606d YS |
283 | }; |
284 | ||
f84344f3 | 285 | struct rbd_mapping { |
99c1f08f | 286 | u64 size; |
34b13184 | 287 | u64 features; |
f84344f3 AE |
288 | bool read_only; |
289 | }; | |
290 | ||
602adf40 YS |
291 | /* |
292 | * a single device | |
293 | */ | |
294 | struct rbd_device { | |
de71a297 | 295 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
296 | |
297 | int major; /* blkdev assigned major */ | |
298 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 299 | |
a30b71b9 | 300 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
301 | struct rbd_client *rbd_client; |
302 | ||
303 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
304 | ||
b82d167b | 305 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
306 | |
307 | struct rbd_image_header header; | |
b82d167b | 308 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 309 | struct rbd_spec *spec; |
602adf40 | 310 | |
0d7dbfce | 311 | char *header_name; |
971f839a | 312 | |
0903e875 AE |
313 | struct ceph_file_layout layout; |
314 | ||
59c2be1e | 315 | struct ceph_osd_event *watch_event; |
975241af | 316 | struct rbd_obj_request *watch_request; |
59c2be1e | 317 | |
86b00e0d AE |
318 | struct rbd_spec *parent_spec; |
319 | u64 parent_overlap; | |
2f82ee54 | 320 | struct rbd_device *parent; |
86b00e0d | 321 | |
c666601a JD |
322 | /* protects updating the header */ |
323 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
324 | |
325 | struct rbd_mapping mapping; | |
602adf40 YS |
326 | |
327 | struct list_head node; | |
dfc5606d YS |
328 | |
329 | /* list of snapshots */ | |
330 | struct list_head snaps; | |
331 | ||
332 | /* sysfs related */ | |
333 | struct device dev; | |
b82d167b | 334 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
335 | }; |
336 | ||
b82d167b AE |
337 | /* |
338 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
339 | * rbd_dev->lock is used to protect access. | |
340 | * | |
341 | * Currently, only the "removing" flag (which is coupled with the | |
342 | * "open_count" field) requires atomic access. | |
343 | */ | |
6d292906 AE |
344 | enum rbd_dev_flags { |
345 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 346 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
347 | }; |
348 | ||
602adf40 | 349 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 350 | |
602adf40 | 351 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
352 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
353 | ||
432b8587 AE |
354 | static LIST_HEAD(rbd_client_list); /* clients */ |
355 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 356 | |
3d7efd18 AE |
357 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
358 | ||
304f6808 | 359 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev); |
304f6808 | 360 | |
200a6a8b | 361 | static void rbd_dev_device_release(struct device *dev); |
6087b51b | 362 | static void rbd_snap_destroy(struct rbd_snap *snap); |
dfc5606d | 363 | |
f0f8cef5 AE |
364 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
365 | size_t count); | |
366 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
367 | size_t count); | |
71f293e2 | 368 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev); |
f0f8cef5 AE |
369 | |
370 | static struct bus_attribute rbd_bus_attrs[] = { | |
371 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
372 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
373 | __ATTR_NULL | |
374 | }; | |
375 | ||
376 | static struct bus_type rbd_bus_type = { | |
377 | .name = "rbd", | |
378 | .bus_attrs = rbd_bus_attrs, | |
379 | }; | |
380 | ||
381 | static void rbd_root_dev_release(struct device *dev) | |
382 | { | |
383 | } | |
384 | ||
385 | static struct device rbd_root_dev = { | |
386 | .init_name = "rbd", | |
387 | .release = rbd_root_dev_release, | |
388 | }; | |
389 | ||
06ecc6cb AE |
390 | static __printf(2, 3) |
391 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
392 | { | |
393 | struct va_format vaf; | |
394 | va_list args; | |
395 | ||
396 | va_start(args, fmt); | |
397 | vaf.fmt = fmt; | |
398 | vaf.va = &args; | |
399 | ||
400 | if (!rbd_dev) | |
401 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
402 | else if (rbd_dev->disk) | |
403 | printk(KERN_WARNING "%s: %s: %pV\n", | |
404 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
405 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
406 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
407 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
408 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
409 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
410 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
411 | else /* punt */ | |
412 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
413 | RBD_DRV_NAME, rbd_dev, &vaf); | |
414 | va_end(args); | |
415 | } | |
416 | ||
aafb230e AE |
417 | #ifdef RBD_DEBUG |
418 | #define rbd_assert(expr) \ | |
419 | if (unlikely(!(expr))) { \ | |
420 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
421 | "at line %d:\n\n" \ | |
422 | "\trbd_assert(%s);\n\n", \ | |
423 | __func__, __LINE__, #expr); \ | |
424 | BUG(); \ | |
425 | } | |
426 | #else /* !RBD_DEBUG */ | |
427 | # define rbd_assert(expr) ((void) 0) | |
428 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 429 | |
b454e36d | 430 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
431 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
432 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 433 | |
117973fb AE |
434 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver); |
435 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver); | |
59c2be1e | 436 | |
602adf40 YS |
437 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
438 | { | |
f0f8cef5 | 439 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 440 | bool removing = false; |
602adf40 | 441 | |
f84344f3 | 442 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
443 | return -EROFS; |
444 | ||
a14ea269 | 445 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
446 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
447 | removing = true; | |
448 | else | |
449 | rbd_dev->open_count++; | |
a14ea269 | 450 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
451 | if (removing) |
452 | return -ENOENT; | |
453 | ||
42382b70 | 454 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 455 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 456 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 457 | mutex_unlock(&ctl_mutex); |
340c7a2b | 458 | |
602adf40 YS |
459 | return 0; |
460 | } | |
461 | ||
dfc5606d YS |
462 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
463 | { | |
464 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
465 | unsigned long open_count_before; |
466 | ||
a14ea269 | 467 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 468 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 469 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 470 | rbd_assert(open_count_before > 0); |
dfc5606d | 471 | |
42382b70 | 472 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 473 | put_device(&rbd_dev->dev); |
42382b70 | 474 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
475 | |
476 | return 0; | |
477 | } | |
478 | ||
602adf40 YS |
479 | static const struct block_device_operations rbd_bd_ops = { |
480 | .owner = THIS_MODULE, | |
481 | .open = rbd_open, | |
dfc5606d | 482 | .release = rbd_release, |
602adf40 YS |
483 | }; |
484 | ||
485 | /* | |
486 | * Initialize an rbd client instance. | |
43ae4701 | 487 | * We own *ceph_opts. |
602adf40 | 488 | */ |
f8c38929 | 489 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
490 | { |
491 | struct rbd_client *rbdc; | |
492 | int ret = -ENOMEM; | |
493 | ||
37206ee5 | 494 | dout("%s:\n", __func__); |
602adf40 YS |
495 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
496 | if (!rbdc) | |
497 | goto out_opt; | |
498 | ||
499 | kref_init(&rbdc->kref); | |
500 | INIT_LIST_HEAD(&rbdc->node); | |
501 | ||
bc534d86 AE |
502 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
503 | ||
43ae4701 | 504 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 505 | if (IS_ERR(rbdc->client)) |
bc534d86 | 506 | goto out_mutex; |
43ae4701 | 507 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
508 | |
509 | ret = ceph_open_session(rbdc->client); | |
510 | if (ret < 0) | |
511 | goto out_err; | |
512 | ||
432b8587 | 513 | spin_lock(&rbd_client_list_lock); |
602adf40 | 514 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 515 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 516 | |
bc534d86 | 517 | mutex_unlock(&ctl_mutex); |
37206ee5 | 518 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 519 | |
602adf40 YS |
520 | return rbdc; |
521 | ||
522 | out_err: | |
523 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
524 | out_mutex: |
525 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
526 | kfree(rbdc); |
527 | out_opt: | |
43ae4701 AE |
528 | if (ceph_opts) |
529 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
530 | dout("%s: error %d\n", __func__, ret); |
531 | ||
28f259b7 | 532 | return ERR_PTR(ret); |
602adf40 YS |
533 | } |
534 | ||
2f82ee54 AE |
535 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
536 | { | |
537 | kref_get(&rbdc->kref); | |
538 | ||
539 | return rbdc; | |
540 | } | |
541 | ||
602adf40 | 542 | /* |
1f7ba331 AE |
543 | * Find a ceph client with specific addr and configuration. If |
544 | * found, bump its reference count. | |
602adf40 | 545 | */ |
1f7ba331 | 546 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
547 | { |
548 | struct rbd_client *client_node; | |
1f7ba331 | 549 | bool found = false; |
602adf40 | 550 | |
43ae4701 | 551 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
552 | return NULL; |
553 | ||
1f7ba331 AE |
554 | spin_lock(&rbd_client_list_lock); |
555 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
556 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
557 | __rbd_get_client(client_node); |
558 | ||
1f7ba331 AE |
559 | found = true; |
560 | break; | |
561 | } | |
562 | } | |
563 | spin_unlock(&rbd_client_list_lock); | |
564 | ||
565 | return found ? client_node : NULL; | |
602adf40 YS |
566 | } |
567 | ||
59c2be1e YS |
568 | /* |
569 | * mount options | |
570 | */ | |
571 | enum { | |
59c2be1e YS |
572 | Opt_last_int, |
573 | /* int args above */ | |
574 | Opt_last_string, | |
575 | /* string args above */ | |
cc0538b6 AE |
576 | Opt_read_only, |
577 | Opt_read_write, | |
578 | /* Boolean args above */ | |
579 | Opt_last_bool, | |
59c2be1e YS |
580 | }; |
581 | ||
43ae4701 | 582 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
583 | /* int args above */ |
584 | /* string args above */ | |
be466c1c | 585 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
586 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
587 | {Opt_read_write, "read_write"}, | |
588 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
589 | /* Boolean args above */ | |
59c2be1e YS |
590 | {-1, NULL} |
591 | }; | |
592 | ||
98571b5a AE |
593 | struct rbd_options { |
594 | bool read_only; | |
595 | }; | |
596 | ||
597 | #define RBD_READ_ONLY_DEFAULT false | |
598 | ||
59c2be1e YS |
599 | static int parse_rbd_opts_token(char *c, void *private) |
600 | { | |
43ae4701 | 601 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
602 | substring_t argstr[MAX_OPT_ARGS]; |
603 | int token, intval, ret; | |
604 | ||
43ae4701 | 605 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
606 | if (token < 0) |
607 | return -EINVAL; | |
608 | ||
609 | if (token < Opt_last_int) { | |
610 | ret = match_int(&argstr[0], &intval); | |
611 | if (ret < 0) { | |
612 | pr_err("bad mount option arg (not int) " | |
613 | "at '%s'\n", c); | |
614 | return ret; | |
615 | } | |
616 | dout("got int token %d val %d\n", token, intval); | |
617 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
618 | dout("got string token %d val %s\n", token, | |
619 | argstr[0].from); | |
cc0538b6 AE |
620 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
621 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
622 | } else { |
623 | dout("got token %d\n", token); | |
624 | } | |
625 | ||
626 | switch (token) { | |
cc0538b6 AE |
627 | case Opt_read_only: |
628 | rbd_opts->read_only = true; | |
629 | break; | |
630 | case Opt_read_write: | |
631 | rbd_opts->read_only = false; | |
632 | break; | |
59c2be1e | 633 | default: |
aafb230e AE |
634 | rbd_assert(false); |
635 | break; | |
59c2be1e YS |
636 | } |
637 | return 0; | |
638 | } | |
639 | ||
602adf40 YS |
640 | /* |
641 | * Get a ceph client with specific addr and configuration, if one does | |
642 | * not exist create it. | |
643 | */ | |
9d3997fd | 644 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 645 | { |
f8c38929 | 646 | struct rbd_client *rbdc; |
59c2be1e | 647 | |
1f7ba331 | 648 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 649 | if (rbdc) /* using an existing client */ |
43ae4701 | 650 | ceph_destroy_options(ceph_opts); |
9d3997fd | 651 | else |
f8c38929 | 652 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 653 | |
9d3997fd | 654 | return rbdc; |
602adf40 YS |
655 | } |
656 | ||
657 | /* | |
658 | * Destroy ceph client | |
d23a4b3f | 659 | * |
432b8587 | 660 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
661 | */ |
662 | static void rbd_client_release(struct kref *kref) | |
663 | { | |
664 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
665 | ||
37206ee5 | 666 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 667 | spin_lock(&rbd_client_list_lock); |
602adf40 | 668 | list_del(&rbdc->node); |
cd9d9f5d | 669 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
670 | |
671 | ceph_destroy_client(rbdc->client); | |
672 | kfree(rbdc); | |
673 | } | |
674 | ||
468521c1 AE |
675 | /* Caller has to fill in snapc->seq and snapc->snaps[0..snap_count-1] */ |
676 | ||
677 | static struct ceph_snap_context *rbd_snap_context_create(u32 snap_count) | |
678 | { | |
679 | struct ceph_snap_context *snapc; | |
680 | size_t size; | |
681 | ||
682 | size = sizeof (struct ceph_snap_context); | |
683 | size += snap_count * sizeof (snapc->snaps[0]); | |
684 | snapc = kzalloc(size, GFP_KERNEL); | |
685 | if (!snapc) | |
686 | return NULL; | |
687 | ||
688 | atomic_set(&snapc->nref, 1); | |
689 | snapc->num_snaps = snap_count; | |
690 | ||
691 | return snapc; | |
692 | } | |
693 | ||
694 | static inline void rbd_snap_context_get(struct ceph_snap_context *snapc) | |
695 | { | |
696 | (void)ceph_get_snap_context(snapc); | |
697 | } | |
698 | ||
699 | static inline void rbd_snap_context_put(struct ceph_snap_context *snapc) | |
700 | { | |
701 | ceph_put_snap_context(snapc); | |
702 | } | |
703 | ||
602adf40 YS |
704 | /* |
705 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
706 | * it. | |
707 | */ | |
9d3997fd | 708 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 709 | { |
c53d5893 AE |
710 | if (rbdc) |
711 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
712 | } |
713 | ||
a30b71b9 AE |
714 | static bool rbd_image_format_valid(u32 image_format) |
715 | { | |
716 | return image_format == 1 || image_format == 2; | |
717 | } | |
718 | ||
8e94af8e AE |
719 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
720 | { | |
103a150f AE |
721 | size_t size; |
722 | u32 snap_count; | |
723 | ||
724 | /* The header has to start with the magic rbd header text */ | |
725 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
726 | return false; | |
727 | ||
db2388b6 AE |
728 | /* The bio layer requires at least sector-sized I/O */ |
729 | ||
730 | if (ondisk->options.order < SECTOR_SHIFT) | |
731 | return false; | |
732 | ||
733 | /* If we use u64 in a few spots we may be able to loosen this */ | |
734 | ||
735 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
736 | return false; | |
737 | ||
103a150f AE |
738 | /* |
739 | * The size of a snapshot header has to fit in a size_t, and | |
740 | * that limits the number of snapshots. | |
741 | */ | |
742 | snap_count = le32_to_cpu(ondisk->snap_count); | |
743 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
744 | if (snap_count > size / sizeof (__le64)) | |
745 | return false; | |
746 | ||
747 | /* | |
748 | * Not only that, but the size of the entire the snapshot | |
749 | * header must also be representable in a size_t. | |
750 | */ | |
751 | size -= snap_count * sizeof (__le64); | |
752 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
753 | return false; | |
754 | ||
755 | return true; | |
8e94af8e AE |
756 | } |
757 | ||
602adf40 YS |
758 | /* |
759 | * Create a new header structure, translate header format from the on-disk | |
760 | * header. | |
761 | */ | |
762 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 763 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 764 | { |
ccece235 | 765 | u32 snap_count; |
58c17b0e | 766 | size_t len; |
d2bb24e5 | 767 | size_t size; |
621901d6 | 768 | u32 i; |
602adf40 | 769 | |
6a52325f AE |
770 | memset(header, 0, sizeof (*header)); |
771 | ||
103a150f AE |
772 | snap_count = le32_to_cpu(ondisk->snap_count); |
773 | ||
58c17b0e AE |
774 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
775 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 776 | if (!header->object_prefix) |
602adf40 | 777 | return -ENOMEM; |
58c17b0e AE |
778 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
779 | header->object_prefix[len] = '\0'; | |
00f1f36f | 780 | |
602adf40 | 781 | if (snap_count) { |
f785cc1d AE |
782 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
783 | ||
621901d6 AE |
784 | /* Save a copy of the snapshot names */ |
785 | ||
f785cc1d AE |
786 | if (snap_names_len > (u64) SIZE_MAX) |
787 | return -EIO; | |
788 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 789 | if (!header->snap_names) |
6a52325f | 790 | goto out_err; |
f785cc1d AE |
791 | /* |
792 | * Note that rbd_dev_v1_header_read() guarantees | |
793 | * the ondisk buffer we're working with has | |
794 | * snap_names_len bytes beyond the end of the | |
795 | * snapshot id array, this memcpy() is safe. | |
796 | */ | |
797 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
798 | snap_names_len); | |
6a52325f | 799 | |
621901d6 AE |
800 | /* Record each snapshot's size */ |
801 | ||
d2bb24e5 AE |
802 | size = snap_count * sizeof (*header->snap_sizes); |
803 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 804 | if (!header->snap_sizes) |
6a52325f | 805 | goto out_err; |
621901d6 AE |
806 | for (i = 0; i < snap_count; i++) |
807 | header->snap_sizes[i] = | |
808 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 YS |
809 | } else { |
810 | header->snap_names = NULL; | |
811 | header->snap_sizes = NULL; | |
812 | } | |
849b4260 | 813 | |
34b13184 | 814 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
815 | header->obj_order = ondisk->options.order; |
816 | header->crypt_type = ondisk->options.crypt_type; | |
817 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 818 | |
621901d6 AE |
819 | /* Allocate and fill in the snapshot context */ |
820 | ||
f84344f3 | 821 | header->image_size = le64_to_cpu(ondisk->image_size); |
468521c1 AE |
822 | |
823 | header->snapc = rbd_snap_context_create(snap_count); | |
6a52325f AE |
824 | if (!header->snapc) |
825 | goto out_err; | |
505cbb9b | 826 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
621901d6 | 827 | for (i = 0; i < snap_count; i++) |
468521c1 | 828 | header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id); |
602adf40 YS |
829 | |
830 | return 0; | |
831 | ||
6a52325f | 832 | out_err: |
849b4260 | 833 | kfree(header->snap_sizes); |
ccece235 | 834 | header->snap_sizes = NULL; |
602adf40 | 835 | kfree(header->snap_names); |
ccece235 | 836 | header->snap_names = NULL; |
6a52325f AE |
837 | kfree(header->object_prefix); |
838 | header->object_prefix = NULL; | |
ccece235 | 839 | |
00f1f36f | 840 | return -ENOMEM; |
602adf40 YS |
841 | } |
842 | ||
9e15b77d AE |
843 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) |
844 | { | |
845 | struct rbd_snap *snap; | |
846 | ||
847 | if (snap_id == CEPH_NOSNAP) | |
848 | return RBD_SNAP_HEAD_NAME; | |
849 | ||
850 | list_for_each_entry(snap, &rbd_dev->snaps, node) | |
851 | if (snap_id == snap->id) | |
852 | return snap->name; | |
853 | ||
854 | return NULL; | |
855 | } | |
856 | ||
8b0241f8 AE |
857 | static struct rbd_snap *snap_by_name(struct rbd_device *rbd_dev, |
858 | const char *snap_name) | |
602adf40 | 859 | { |
e86924a8 | 860 | struct rbd_snap *snap; |
602adf40 | 861 | |
8b0241f8 AE |
862 | list_for_each_entry(snap, &rbd_dev->snaps, node) |
863 | if (!strcmp(snap_name, snap->name)) | |
864 | return snap; | |
e86924a8 | 865 | |
8b0241f8 | 866 | return NULL; |
602adf40 YS |
867 | } |
868 | ||
d1cf5788 | 869 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) |
602adf40 | 870 | { |
0d7dbfce | 871 | if (!memcmp(rbd_dev->spec->snap_name, RBD_SNAP_HEAD_NAME, |
cc9d734c | 872 | sizeof (RBD_SNAP_HEAD_NAME))) { |
99c1f08f | 873 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
34b13184 | 874 | rbd_dev->mapping.features = rbd_dev->header.features; |
602adf40 | 875 | } else { |
8b0241f8 AE |
876 | struct rbd_snap *snap; |
877 | ||
878 | snap = snap_by_name(rbd_dev, rbd_dev->spec->snap_name); | |
879 | if (!snap) | |
880 | return -ENOENT; | |
8b0241f8 AE |
881 | rbd_dev->mapping.size = snap->size; |
882 | rbd_dev->mapping.features = snap->features; | |
f84344f3 | 883 | rbd_dev->mapping.read_only = true; |
602adf40 | 884 | } |
6d292906 | 885 | |
8b0241f8 | 886 | return 0; |
602adf40 YS |
887 | } |
888 | ||
d1cf5788 AE |
889 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
890 | { | |
891 | rbd_dev->mapping.size = 0; | |
892 | rbd_dev->mapping.features = 0; | |
893 | rbd_dev->mapping.read_only = true; | |
894 | } | |
895 | ||
200a6a8b AE |
896 | static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev) |
897 | { | |
898 | rbd_dev->mapping.size = 0; | |
899 | rbd_dev->mapping.features = 0; | |
900 | rbd_dev->mapping.read_only = true; | |
901 | } | |
902 | ||
98571b5a | 903 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 904 | { |
65ccfe21 AE |
905 | char *name; |
906 | u64 segment; | |
907 | int ret; | |
602adf40 | 908 | |
2fd82b9e | 909 | name = kmalloc(MAX_OBJ_NAME_SIZE + 1, GFP_NOIO); |
65ccfe21 AE |
910 | if (!name) |
911 | return NULL; | |
912 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 913 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 914 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 915 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
916 | pr_err("error formatting segment name for #%llu (%d)\n", |
917 | segment, ret); | |
918 | kfree(name); | |
919 | name = NULL; | |
920 | } | |
602adf40 | 921 | |
65ccfe21 AE |
922 | return name; |
923 | } | |
602adf40 | 924 | |
65ccfe21 AE |
925 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
926 | { | |
927 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 928 | |
65ccfe21 AE |
929 | return offset & (segment_size - 1); |
930 | } | |
931 | ||
932 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
933 | u64 offset, u64 length) | |
934 | { | |
935 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
936 | ||
937 | offset &= segment_size - 1; | |
938 | ||
aafb230e | 939 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
940 | if (offset + length > segment_size) |
941 | length = segment_size - offset; | |
942 | ||
943 | return length; | |
602adf40 YS |
944 | } |
945 | ||
029bcbd8 JD |
946 | /* |
947 | * returns the size of an object in the image | |
948 | */ | |
949 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
950 | { | |
951 | return 1 << header->obj_order; | |
952 | } | |
953 | ||
602adf40 YS |
954 | /* |
955 | * bio helpers | |
956 | */ | |
957 | ||
958 | static void bio_chain_put(struct bio *chain) | |
959 | { | |
960 | struct bio *tmp; | |
961 | ||
962 | while (chain) { | |
963 | tmp = chain; | |
964 | chain = chain->bi_next; | |
965 | bio_put(tmp); | |
966 | } | |
967 | } | |
968 | ||
969 | /* | |
970 | * zeros a bio chain, starting at specific offset | |
971 | */ | |
972 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
973 | { | |
974 | struct bio_vec *bv; | |
975 | unsigned long flags; | |
976 | void *buf; | |
977 | int i; | |
978 | int pos = 0; | |
979 | ||
980 | while (chain) { | |
981 | bio_for_each_segment(bv, chain, i) { | |
982 | if (pos + bv->bv_len > start_ofs) { | |
983 | int remainder = max(start_ofs - pos, 0); | |
984 | buf = bvec_kmap_irq(bv, &flags); | |
985 | memset(buf + remainder, 0, | |
986 | bv->bv_len - remainder); | |
85b5aaa6 | 987 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
988 | } |
989 | pos += bv->bv_len; | |
990 | } | |
991 | ||
992 | chain = chain->bi_next; | |
993 | } | |
994 | } | |
995 | ||
b9434c5b AE |
996 | /* |
997 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
998 | * starting at the given byte offset from the start of the array and | |
999 | * continuing up to the given end offset. The pages array is | |
1000 | * assumed to be big enough to hold all bytes up to the end. | |
1001 | */ | |
1002 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1003 | { | |
1004 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1005 | ||
1006 | rbd_assert(end > offset); | |
1007 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1008 | while (offset < end) { | |
1009 | size_t page_offset; | |
1010 | size_t length; | |
1011 | unsigned long flags; | |
1012 | void *kaddr; | |
1013 | ||
1014 | page_offset = (size_t)(offset & ~PAGE_MASK); | |
1015 | length = min(PAGE_SIZE - page_offset, (size_t)(end - offset)); | |
1016 | local_irq_save(flags); | |
1017 | kaddr = kmap_atomic(*page); | |
1018 | memset(kaddr + page_offset, 0, length); | |
1019 | kunmap_atomic(kaddr); | |
1020 | local_irq_restore(flags); | |
1021 | ||
1022 | offset += length; | |
1023 | page++; | |
1024 | } | |
1025 | } | |
1026 | ||
602adf40 | 1027 | /* |
f7760dad AE |
1028 | * Clone a portion of a bio, starting at the given byte offset |
1029 | * and continuing for the number of bytes indicated. | |
602adf40 | 1030 | */ |
f7760dad AE |
1031 | static struct bio *bio_clone_range(struct bio *bio_src, |
1032 | unsigned int offset, | |
1033 | unsigned int len, | |
1034 | gfp_t gfpmask) | |
602adf40 | 1035 | { |
f7760dad AE |
1036 | struct bio_vec *bv; |
1037 | unsigned int resid; | |
1038 | unsigned short idx; | |
1039 | unsigned int voff; | |
1040 | unsigned short end_idx; | |
1041 | unsigned short vcnt; | |
1042 | struct bio *bio; | |
1043 | ||
1044 | /* Handle the easy case for the caller */ | |
1045 | ||
1046 | if (!offset && len == bio_src->bi_size) | |
1047 | return bio_clone(bio_src, gfpmask); | |
1048 | ||
1049 | if (WARN_ON_ONCE(!len)) | |
1050 | return NULL; | |
1051 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
1052 | return NULL; | |
1053 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
1054 | return NULL; | |
1055 | ||
1056 | /* Find first affected segment... */ | |
1057 | ||
1058 | resid = offset; | |
1059 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
1060 | if (resid < bv->bv_len) | |
1061 | break; | |
1062 | resid -= bv->bv_len; | |
602adf40 | 1063 | } |
f7760dad | 1064 | voff = resid; |
602adf40 | 1065 | |
f7760dad | 1066 | /* ...and the last affected segment */ |
602adf40 | 1067 | |
f7760dad AE |
1068 | resid += len; |
1069 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
1070 | if (resid <= bv->bv_len) | |
1071 | break; | |
1072 | resid -= bv->bv_len; | |
1073 | } | |
1074 | vcnt = end_idx - idx + 1; | |
1075 | ||
1076 | /* Build the clone */ | |
1077 | ||
1078 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
1079 | if (!bio) | |
1080 | return NULL; /* ENOMEM */ | |
602adf40 | 1081 | |
f7760dad AE |
1082 | bio->bi_bdev = bio_src->bi_bdev; |
1083 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
1084 | bio->bi_rw = bio_src->bi_rw; | |
1085 | bio->bi_flags |= 1 << BIO_CLONED; | |
1086 | ||
1087 | /* | |
1088 | * Copy over our part of the bio_vec, then update the first | |
1089 | * and last (or only) entries. | |
1090 | */ | |
1091 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
1092 | vcnt * sizeof (struct bio_vec)); | |
1093 | bio->bi_io_vec[0].bv_offset += voff; | |
1094 | if (vcnt > 1) { | |
1095 | bio->bi_io_vec[0].bv_len -= voff; | |
1096 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
1097 | } else { | |
1098 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
1099 | } |
1100 | ||
f7760dad AE |
1101 | bio->bi_vcnt = vcnt; |
1102 | bio->bi_size = len; | |
1103 | bio->bi_idx = 0; | |
1104 | ||
1105 | return bio; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Clone a portion of a bio chain, starting at the given byte offset | |
1110 | * into the first bio in the source chain and continuing for the | |
1111 | * number of bytes indicated. The result is another bio chain of | |
1112 | * exactly the given length, or a null pointer on error. | |
1113 | * | |
1114 | * The bio_src and offset parameters are both in-out. On entry they | |
1115 | * refer to the first source bio and the offset into that bio where | |
1116 | * the start of data to be cloned is located. | |
1117 | * | |
1118 | * On return, bio_src is updated to refer to the bio in the source | |
1119 | * chain that contains first un-cloned byte, and *offset will | |
1120 | * contain the offset of that byte within that bio. | |
1121 | */ | |
1122 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1123 | unsigned int *offset, | |
1124 | unsigned int len, | |
1125 | gfp_t gfpmask) | |
1126 | { | |
1127 | struct bio *bi = *bio_src; | |
1128 | unsigned int off = *offset; | |
1129 | struct bio *chain = NULL; | |
1130 | struct bio **end; | |
1131 | ||
1132 | /* Build up a chain of clone bios up to the limit */ | |
1133 | ||
1134 | if (!bi || off >= bi->bi_size || !len) | |
1135 | return NULL; /* Nothing to clone */ | |
602adf40 | 1136 | |
f7760dad AE |
1137 | end = &chain; |
1138 | while (len) { | |
1139 | unsigned int bi_size; | |
1140 | struct bio *bio; | |
1141 | ||
f5400b7a AE |
1142 | if (!bi) { |
1143 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1144 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1145 | } |
f7760dad AE |
1146 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1147 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1148 | if (!bio) | |
1149 | goto out_err; /* ENOMEM */ | |
1150 | ||
1151 | *end = bio; | |
1152 | end = &bio->bi_next; | |
602adf40 | 1153 | |
f7760dad AE |
1154 | off += bi_size; |
1155 | if (off == bi->bi_size) { | |
1156 | bi = bi->bi_next; | |
1157 | off = 0; | |
1158 | } | |
1159 | len -= bi_size; | |
1160 | } | |
1161 | *bio_src = bi; | |
1162 | *offset = off; | |
1163 | ||
1164 | return chain; | |
1165 | out_err: | |
1166 | bio_chain_put(chain); | |
602adf40 | 1167 | |
602adf40 YS |
1168 | return NULL; |
1169 | } | |
1170 | ||
926f9b3f AE |
1171 | /* |
1172 | * The default/initial value for all object request flags is 0. For | |
1173 | * each flag, once its value is set to 1 it is never reset to 0 | |
1174 | * again. | |
1175 | */ | |
57acbaa7 | 1176 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1177 | { |
57acbaa7 | 1178 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1179 | struct rbd_device *rbd_dev; |
1180 | ||
57acbaa7 AE |
1181 | rbd_dev = obj_request->img_request->rbd_dev; |
1182 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1183 | obj_request); |
1184 | } | |
1185 | } | |
1186 | ||
57acbaa7 | 1187 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1188 | { |
1189 | smp_mb(); | |
57acbaa7 | 1190 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1191 | } |
1192 | ||
57acbaa7 | 1193 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1194 | { |
57acbaa7 AE |
1195 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1196 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1197 | |
57acbaa7 AE |
1198 | if (obj_request_img_data_test(obj_request)) |
1199 | rbd_dev = obj_request->img_request->rbd_dev; | |
1200 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1201 | obj_request); |
1202 | } | |
1203 | } | |
1204 | ||
57acbaa7 | 1205 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1206 | { |
1207 | smp_mb(); | |
57acbaa7 | 1208 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1209 | } |
1210 | ||
5679c59f AE |
1211 | /* |
1212 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1213 | * flag. The latter is set based on the "exists" value provided. | |
1214 | * | |
1215 | * Note that for our purposes once an object exists it never goes | |
1216 | * away again. It's possible that the response from two existence | |
1217 | * checks are separated by the creation of the target object, and | |
1218 | * the first ("doesn't exist") response arrives *after* the second | |
1219 | * ("does exist"). In that case we ignore the second one. | |
1220 | */ | |
1221 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1222 | bool exists) | |
1223 | { | |
1224 | if (exists) | |
1225 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1226 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1227 | smp_mb(); | |
1228 | } | |
1229 | ||
1230 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1231 | { | |
1232 | smp_mb(); | |
1233 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1234 | } | |
1235 | ||
1236 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1237 | { | |
1238 | smp_mb(); | |
1239 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1240 | } | |
1241 | ||
bf0d5f50 AE |
1242 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1243 | { | |
37206ee5 AE |
1244 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1245 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1246 | kref_get(&obj_request->kref); |
1247 | } | |
1248 | ||
1249 | static void rbd_obj_request_destroy(struct kref *kref); | |
1250 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1251 | { | |
1252 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1253 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1254 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1255 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1256 | } | |
1257 | ||
1258 | static void rbd_img_request_get(struct rbd_img_request *img_request) | |
1259 | { | |
37206ee5 AE |
1260 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1261 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1262 | kref_get(&img_request->kref); |
1263 | } | |
1264 | ||
1265 | static void rbd_img_request_destroy(struct kref *kref); | |
1266 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1267 | { | |
1268 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1269 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1270 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1271 | kref_put(&img_request->kref, rbd_img_request_destroy); |
1272 | } | |
1273 | ||
1274 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1275 | struct rbd_obj_request *obj_request) | |
1276 | { | |
25dcf954 AE |
1277 | rbd_assert(obj_request->img_request == NULL); |
1278 | ||
b155e86c | 1279 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1280 | obj_request->img_request = img_request; |
25dcf954 | 1281 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1282 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1283 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1284 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1285 | img_request->obj_request_count++; |
1286 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1287 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1288 | obj_request->which); | |
bf0d5f50 AE |
1289 | } |
1290 | ||
1291 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1292 | struct rbd_obj_request *obj_request) | |
1293 | { | |
1294 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1295 | |
37206ee5 AE |
1296 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1297 | obj_request->which); | |
bf0d5f50 | 1298 | list_del(&obj_request->links); |
25dcf954 AE |
1299 | rbd_assert(img_request->obj_request_count > 0); |
1300 | img_request->obj_request_count--; | |
1301 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1302 | obj_request->which = BAD_WHICH; | |
6365d33a | 1303 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1304 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1305 | obj_request->img_request = NULL; |
25dcf954 | 1306 | obj_request->callback = NULL; |
bf0d5f50 AE |
1307 | rbd_obj_request_put(obj_request); |
1308 | } | |
1309 | ||
1310 | static bool obj_request_type_valid(enum obj_request_type type) | |
1311 | { | |
1312 | switch (type) { | |
9969ebc5 | 1313 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1314 | case OBJ_REQUEST_BIO: |
788e2df3 | 1315 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1316 | return true; |
1317 | default: | |
1318 | return false; | |
1319 | } | |
1320 | } | |
1321 | ||
bf0d5f50 AE |
1322 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1323 | struct rbd_obj_request *obj_request) | |
1324 | { | |
37206ee5 AE |
1325 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1326 | ||
bf0d5f50 AE |
1327 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1328 | } | |
1329 | ||
1330 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1331 | { | |
55f27e09 | 1332 | |
37206ee5 | 1333 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1334 | |
1335 | /* | |
1336 | * If no error occurred, compute the aggregate transfer | |
1337 | * count for the image request. We could instead use | |
1338 | * atomic64_cmpxchg() to update it as each object request | |
1339 | * completes; not clear which way is better off hand. | |
1340 | */ | |
1341 | if (!img_request->result) { | |
1342 | struct rbd_obj_request *obj_request; | |
1343 | u64 xferred = 0; | |
1344 | ||
1345 | for_each_obj_request(img_request, obj_request) | |
1346 | xferred += obj_request->xferred; | |
1347 | img_request->xferred = xferred; | |
1348 | } | |
1349 | ||
bf0d5f50 AE |
1350 | if (img_request->callback) |
1351 | img_request->callback(img_request); | |
1352 | else | |
1353 | rbd_img_request_put(img_request); | |
1354 | } | |
1355 | ||
788e2df3 AE |
1356 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1357 | ||
1358 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1359 | { | |
37206ee5 AE |
1360 | dout("%s: obj %p\n", __func__, obj_request); |
1361 | ||
788e2df3 AE |
1362 | return wait_for_completion_interruptible(&obj_request->completion); |
1363 | } | |
1364 | ||
0c425248 AE |
1365 | /* |
1366 | * The default/initial value for all image request flags is 0. Each | |
1367 | * is conditionally set to 1 at image request initialization time | |
1368 | * and currently never change thereafter. | |
1369 | */ | |
1370 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1371 | { | |
1372 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1373 | smp_mb(); | |
1374 | } | |
1375 | ||
1376 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1377 | { | |
1378 | smp_mb(); | |
1379 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1380 | } | |
1381 | ||
9849e986 AE |
1382 | static void img_request_child_set(struct rbd_img_request *img_request) |
1383 | { | |
1384 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1385 | smp_mb(); | |
1386 | } | |
1387 | ||
1388 | static bool img_request_child_test(struct rbd_img_request *img_request) | |
1389 | { | |
1390 | smp_mb(); | |
1391 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1392 | } | |
1393 | ||
d0b2e944 AE |
1394 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1395 | { | |
1396 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1397 | smp_mb(); | |
1398 | } | |
1399 | ||
1400 | static bool img_request_layered_test(struct rbd_img_request *img_request) | |
1401 | { | |
1402 | smp_mb(); | |
1403 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1404 | } | |
1405 | ||
6e2a4505 AE |
1406 | static void |
1407 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1408 | { | |
b9434c5b AE |
1409 | u64 xferred = obj_request->xferred; |
1410 | u64 length = obj_request->length; | |
1411 | ||
6e2a4505 AE |
1412 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1413 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1414 | xferred, length); |
6e2a4505 AE |
1415 | /* |
1416 | * ENOENT means a hole in the image. We zero-fill the | |
1417 | * entire length of the request. A short read also implies | |
1418 | * zero-fill to the end of the request. Either way we | |
1419 | * update the xferred count to indicate the whole request | |
1420 | * was satisfied. | |
1421 | */ | |
b9434c5b | 1422 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1423 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1424 | if (obj_request->type == OBJ_REQUEST_BIO) |
1425 | zero_bio_chain(obj_request->bio_list, 0); | |
1426 | else | |
1427 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1428 | obj_request->result = 0; |
b9434c5b AE |
1429 | obj_request->xferred = length; |
1430 | } else if (xferred < length && !obj_request->result) { | |
1431 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1432 | zero_bio_chain(obj_request->bio_list, xferred); | |
1433 | else | |
1434 | zero_pages(obj_request->pages, xferred, length); | |
1435 | obj_request->xferred = length; | |
6e2a4505 AE |
1436 | } |
1437 | obj_request_done_set(obj_request); | |
1438 | } | |
1439 | ||
bf0d5f50 AE |
1440 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1441 | { | |
37206ee5 AE |
1442 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1443 | obj_request->callback); | |
bf0d5f50 AE |
1444 | if (obj_request->callback) |
1445 | obj_request->callback(obj_request); | |
788e2df3 AE |
1446 | else |
1447 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1448 | } |
1449 | ||
c47f9371 | 1450 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1451 | { |
1452 | dout("%s: obj %p\n", __func__, obj_request); | |
1453 | obj_request_done_set(obj_request); | |
1454 | } | |
1455 | ||
c47f9371 | 1456 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1457 | { |
57acbaa7 | 1458 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1459 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1460 | bool layered = false; |
1461 | ||
1462 | if (obj_request_img_data_test(obj_request)) { | |
1463 | img_request = obj_request->img_request; | |
1464 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1465 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1466 | } |
8b3e1a56 AE |
1467 | |
1468 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1469 | obj_request, img_request, obj_request->result, | |
1470 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1471 | if (layered && obj_request->result == -ENOENT && |
1472 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1473 | rbd_img_parent_read(obj_request); |
1474 | else if (img_request) | |
6e2a4505 AE |
1475 | rbd_img_obj_request_read_callback(obj_request); |
1476 | else | |
1477 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1478 | } |
1479 | ||
c47f9371 | 1480 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1481 | { |
1b83bef2 SW |
1482 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1483 | obj_request->result, obj_request->length); | |
1484 | /* | |
8b3e1a56 AE |
1485 | * There is no such thing as a successful short write. Set |
1486 | * it to our originally-requested length. | |
1b83bef2 SW |
1487 | */ |
1488 | obj_request->xferred = obj_request->length; | |
07741308 | 1489 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1490 | } |
1491 | ||
fbfab539 AE |
1492 | /* |
1493 | * For a simple stat call there's nothing to do. We'll do more if | |
1494 | * this is part of a write sequence for a layered image. | |
1495 | */ | |
c47f9371 | 1496 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1497 | { |
37206ee5 | 1498 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1499 | obj_request_done_set(obj_request); |
1500 | } | |
1501 | ||
bf0d5f50 AE |
1502 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1503 | struct ceph_msg *msg) | |
1504 | { | |
1505 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1506 | u16 opcode; |
1507 | ||
37206ee5 | 1508 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1509 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1510 | if (obj_request_img_data_test(obj_request)) { |
1511 | rbd_assert(obj_request->img_request); | |
1512 | rbd_assert(obj_request->which != BAD_WHICH); | |
1513 | } else { | |
1514 | rbd_assert(obj_request->which == BAD_WHICH); | |
1515 | } | |
bf0d5f50 | 1516 | |
1b83bef2 SW |
1517 | if (osd_req->r_result < 0) |
1518 | obj_request->result = osd_req->r_result; | |
bf0d5f50 AE |
1519 | obj_request->version = le64_to_cpu(osd_req->r_reassert_version.version); |
1520 | ||
0eefd470 | 1521 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1522 | |
c47f9371 AE |
1523 | /* |
1524 | * We support a 64-bit length, but ultimately it has to be | |
1525 | * passed to blk_end_request(), which takes an unsigned int. | |
1526 | */ | |
1b83bef2 | 1527 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1528 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1529 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1530 | switch (opcode) { |
1531 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1532 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1533 | break; |
1534 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1535 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1536 | break; |
fbfab539 | 1537 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1538 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1539 | break; |
36be9a76 | 1540 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1541 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1542 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1543 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1544 | break; |
bf0d5f50 AE |
1545 | default: |
1546 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1547 | obj_request->object_name, (unsigned short) opcode); | |
1548 | break; | |
1549 | } | |
1550 | ||
07741308 | 1551 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1552 | rbd_obj_request_complete(obj_request); |
1553 | } | |
1554 | ||
9d4df01f | 1555 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1556 | { |
1557 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1558 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1559 | u64 snap_id; |
430c28c3 | 1560 | |
8c042b0d | 1561 | rbd_assert(osd_req != NULL); |
430c28c3 | 1562 | |
9d4df01f | 1563 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1564 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1565 | NULL, snap_id, NULL); |
1566 | } | |
1567 | ||
1568 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1569 | { | |
1570 | struct rbd_img_request *img_request = obj_request->img_request; | |
1571 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1572 | struct ceph_snap_context *snapc; | |
1573 | struct timespec mtime = CURRENT_TIME; | |
1574 | ||
1575 | rbd_assert(osd_req != NULL); | |
1576 | ||
1577 | snapc = img_request ? img_request->snapc : NULL; | |
1578 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1579 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1580 | } |
1581 | ||
bf0d5f50 AE |
1582 | static struct ceph_osd_request *rbd_osd_req_create( |
1583 | struct rbd_device *rbd_dev, | |
1584 | bool write_request, | |
430c28c3 | 1585 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1586 | { |
bf0d5f50 AE |
1587 | struct ceph_snap_context *snapc = NULL; |
1588 | struct ceph_osd_client *osdc; | |
1589 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1590 | |
6365d33a AE |
1591 | if (obj_request_img_data_test(obj_request)) { |
1592 | struct rbd_img_request *img_request = obj_request->img_request; | |
1593 | ||
0c425248 AE |
1594 | rbd_assert(write_request == |
1595 | img_request_write_test(img_request)); | |
1596 | if (write_request) | |
bf0d5f50 | 1597 | snapc = img_request->snapc; |
bf0d5f50 AE |
1598 | } |
1599 | ||
1600 | /* Allocate and initialize the request, for the single op */ | |
1601 | ||
1602 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1603 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1604 | if (!osd_req) | |
1605 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1606 | |
430c28c3 | 1607 | if (write_request) |
bf0d5f50 | 1608 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1609 | else |
bf0d5f50 | 1610 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1611 | |
1612 | osd_req->r_callback = rbd_osd_req_callback; | |
1613 | osd_req->r_priv = obj_request; | |
1614 | ||
1615 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1616 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1617 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1618 | ||
1619 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1620 | ||
bf0d5f50 AE |
1621 | return osd_req; |
1622 | } | |
1623 | ||
0eefd470 AE |
1624 | /* |
1625 | * Create a copyup osd request based on the information in the | |
1626 | * object request supplied. A copyup request has two osd ops, | |
1627 | * a copyup method call, and a "normal" write request. | |
1628 | */ | |
1629 | static struct ceph_osd_request * | |
1630 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1631 | { | |
1632 | struct rbd_img_request *img_request; | |
1633 | struct ceph_snap_context *snapc; | |
1634 | struct rbd_device *rbd_dev; | |
1635 | struct ceph_osd_client *osdc; | |
1636 | struct ceph_osd_request *osd_req; | |
1637 | ||
1638 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1639 | img_request = obj_request->img_request; | |
1640 | rbd_assert(img_request); | |
1641 | rbd_assert(img_request_write_test(img_request)); | |
1642 | ||
1643 | /* Allocate and initialize the request, for the two ops */ | |
1644 | ||
1645 | snapc = img_request->snapc; | |
1646 | rbd_dev = img_request->rbd_dev; | |
1647 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1648 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1649 | if (!osd_req) | |
1650 | return NULL; /* ENOMEM */ | |
1651 | ||
1652 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1653 | osd_req->r_callback = rbd_osd_req_callback; | |
1654 | osd_req->r_priv = obj_request; | |
1655 | ||
1656 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1657 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1658 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1659 | ||
1660 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1661 | ||
1662 | return osd_req; | |
1663 | } | |
1664 | ||
1665 | ||
bf0d5f50 AE |
1666 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1667 | { | |
1668 | ceph_osdc_put_request(osd_req); | |
1669 | } | |
1670 | ||
1671 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1672 | ||
1673 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1674 | u64 offset, u64 length, | |
1675 | enum obj_request_type type) | |
1676 | { | |
1677 | struct rbd_obj_request *obj_request; | |
1678 | size_t size; | |
1679 | char *name; | |
1680 | ||
1681 | rbd_assert(obj_request_type_valid(type)); | |
1682 | ||
1683 | size = strlen(object_name) + 1; | |
1684 | obj_request = kzalloc(sizeof (*obj_request) + size, GFP_KERNEL); | |
1685 | if (!obj_request) | |
1686 | return NULL; | |
1687 | ||
1688 | name = (char *)(obj_request + 1); | |
1689 | obj_request->object_name = memcpy(name, object_name, size); | |
1690 | obj_request->offset = offset; | |
1691 | obj_request->length = length; | |
926f9b3f | 1692 | obj_request->flags = 0; |
bf0d5f50 AE |
1693 | obj_request->which = BAD_WHICH; |
1694 | obj_request->type = type; | |
1695 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1696 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1697 | kref_init(&obj_request->kref); |
1698 | ||
37206ee5 AE |
1699 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1700 | offset, length, (int)type, obj_request); | |
1701 | ||
bf0d5f50 AE |
1702 | return obj_request; |
1703 | } | |
1704 | ||
1705 | static void rbd_obj_request_destroy(struct kref *kref) | |
1706 | { | |
1707 | struct rbd_obj_request *obj_request; | |
1708 | ||
1709 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1710 | ||
37206ee5 AE |
1711 | dout("%s: obj %p\n", __func__, obj_request); |
1712 | ||
bf0d5f50 AE |
1713 | rbd_assert(obj_request->img_request == NULL); |
1714 | rbd_assert(obj_request->which == BAD_WHICH); | |
1715 | ||
1716 | if (obj_request->osd_req) | |
1717 | rbd_osd_req_destroy(obj_request->osd_req); | |
1718 | ||
1719 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1720 | switch (obj_request->type) { | |
9969ebc5 AE |
1721 | case OBJ_REQUEST_NODATA: |
1722 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1723 | case OBJ_REQUEST_BIO: |
1724 | if (obj_request->bio_list) | |
1725 | bio_chain_put(obj_request->bio_list); | |
1726 | break; | |
788e2df3 AE |
1727 | case OBJ_REQUEST_PAGES: |
1728 | if (obj_request->pages) | |
1729 | ceph_release_page_vector(obj_request->pages, | |
1730 | obj_request->page_count); | |
1731 | break; | |
bf0d5f50 AE |
1732 | } |
1733 | ||
1734 | kfree(obj_request); | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * Caller is responsible for filling in the list of object requests | |
1739 | * that comprises the image request, and the Linux request pointer | |
1740 | * (if there is one). | |
1741 | */ | |
cc344fa1 AE |
1742 | static struct rbd_img_request *rbd_img_request_create( |
1743 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1744 | u64 offset, u64 length, |
9849e986 AE |
1745 | bool write_request, |
1746 | bool child_request) | |
bf0d5f50 AE |
1747 | { |
1748 | struct rbd_img_request *img_request; | |
bf0d5f50 AE |
1749 | |
1750 | img_request = kmalloc(sizeof (*img_request), GFP_ATOMIC); | |
1751 | if (!img_request) | |
1752 | return NULL; | |
1753 | ||
1754 | if (write_request) { | |
1755 | down_read(&rbd_dev->header_rwsem); | |
468521c1 | 1756 | rbd_snap_context_get(rbd_dev->header.snapc); |
bf0d5f50 | 1757 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1758 | } |
1759 | ||
1760 | img_request->rq = NULL; | |
1761 | img_request->rbd_dev = rbd_dev; | |
1762 | img_request->offset = offset; | |
1763 | img_request->length = length; | |
0c425248 AE |
1764 | img_request->flags = 0; |
1765 | if (write_request) { | |
1766 | img_request_write_set(img_request); | |
468521c1 | 1767 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1768 | } else { |
bf0d5f50 | 1769 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1770 | } |
9849e986 AE |
1771 | if (child_request) |
1772 | img_request_child_set(img_request); | |
d0b2e944 AE |
1773 | if (rbd_dev->parent_spec) |
1774 | img_request_layered_set(img_request); | |
bf0d5f50 AE |
1775 | spin_lock_init(&img_request->completion_lock); |
1776 | img_request->next_completion = 0; | |
1777 | img_request->callback = NULL; | |
a5a337d4 | 1778 | img_request->result = 0; |
bf0d5f50 AE |
1779 | img_request->obj_request_count = 0; |
1780 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1781 | kref_init(&img_request->kref); | |
1782 | ||
1783 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1784 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1785 | ||
37206ee5 AE |
1786 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1787 | write_request ? "write" : "read", offset, length, | |
1788 | img_request); | |
1789 | ||
bf0d5f50 AE |
1790 | return img_request; |
1791 | } | |
1792 | ||
1793 | static void rbd_img_request_destroy(struct kref *kref) | |
1794 | { | |
1795 | struct rbd_img_request *img_request; | |
1796 | struct rbd_obj_request *obj_request; | |
1797 | struct rbd_obj_request *next_obj_request; | |
1798 | ||
1799 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1800 | ||
37206ee5 AE |
1801 | dout("%s: img %p\n", __func__, img_request); |
1802 | ||
bf0d5f50 AE |
1803 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1804 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1805 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 1806 | |
0c425248 | 1807 | if (img_request_write_test(img_request)) |
468521c1 | 1808 | rbd_snap_context_put(img_request->snapc); |
bf0d5f50 | 1809 | |
8b3e1a56 AE |
1810 | if (img_request_child_test(img_request)) |
1811 | rbd_obj_request_put(img_request->obj_request); | |
1812 | ||
bf0d5f50 AE |
1813 | kfree(img_request); |
1814 | } | |
1815 | ||
1217857f AE |
1816 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
1817 | { | |
6365d33a | 1818 | struct rbd_img_request *img_request; |
1217857f AE |
1819 | unsigned int xferred; |
1820 | int result; | |
8b3e1a56 | 1821 | bool more; |
1217857f | 1822 | |
6365d33a AE |
1823 | rbd_assert(obj_request_img_data_test(obj_request)); |
1824 | img_request = obj_request->img_request; | |
1825 | ||
1217857f AE |
1826 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
1827 | xferred = (unsigned int)obj_request->xferred; | |
1828 | result = obj_request->result; | |
1829 | if (result) { | |
1830 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1831 | ||
1832 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
1833 | img_request_write_test(img_request) ? "write" : "read", | |
1834 | obj_request->length, obj_request->img_offset, | |
1835 | obj_request->offset); | |
1836 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
1837 | result, xferred); | |
1838 | if (!img_request->result) | |
1839 | img_request->result = result; | |
1840 | } | |
1841 | ||
f1a4739f AE |
1842 | /* Image object requests don't own their page array */ |
1843 | ||
1844 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
1845 | obj_request->pages = NULL; | |
1846 | obj_request->page_count = 0; | |
1847 | } | |
1848 | ||
8b3e1a56 AE |
1849 | if (img_request_child_test(img_request)) { |
1850 | rbd_assert(img_request->obj_request != NULL); | |
1851 | more = obj_request->which < img_request->obj_request_count - 1; | |
1852 | } else { | |
1853 | rbd_assert(img_request->rq != NULL); | |
1854 | more = blk_end_request(img_request->rq, result, xferred); | |
1855 | } | |
1856 | ||
1857 | return more; | |
1217857f AE |
1858 | } |
1859 | ||
2169238d AE |
1860 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
1861 | { | |
1862 | struct rbd_img_request *img_request; | |
1863 | u32 which = obj_request->which; | |
1864 | bool more = true; | |
1865 | ||
6365d33a | 1866 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
1867 | img_request = obj_request->img_request; |
1868 | ||
1869 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
1870 | rbd_assert(img_request != NULL); | |
2169238d AE |
1871 | rbd_assert(img_request->obj_request_count > 0); |
1872 | rbd_assert(which != BAD_WHICH); | |
1873 | rbd_assert(which < img_request->obj_request_count); | |
1874 | rbd_assert(which >= img_request->next_completion); | |
1875 | ||
1876 | spin_lock_irq(&img_request->completion_lock); | |
1877 | if (which != img_request->next_completion) | |
1878 | goto out; | |
1879 | ||
1880 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
1881 | rbd_assert(more); |
1882 | rbd_assert(which < img_request->obj_request_count); | |
1883 | ||
1884 | if (!obj_request_done_test(obj_request)) | |
1885 | break; | |
1217857f | 1886 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
1887 | which++; |
1888 | } | |
1889 | ||
1890 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
1891 | img_request->next_completion = which; | |
1892 | out: | |
1893 | spin_unlock_irq(&img_request->completion_lock); | |
1894 | ||
1895 | if (!more) | |
1896 | rbd_img_request_complete(img_request); | |
1897 | } | |
1898 | ||
f1a4739f AE |
1899 | /* |
1900 | * Split up an image request into one or more object requests, each | |
1901 | * to a different object. The "type" parameter indicates whether | |
1902 | * "data_desc" is the pointer to the head of a list of bio | |
1903 | * structures, or the base of a page array. In either case this | |
1904 | * function assumes data_desc describes memory sufficient to hold | |
1905 | * all data described by the image request. | |
1906 | */ | |
1907 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
1908 | enum obj_request_type type, | |
1909 | void *data_desc) | |
bf0d5f50 AE |
1910 | { |
1911 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1912 | struct rbd_obj_request *obj_request = NULL; | |
1913 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 1914 | bool write_request = img_request_write_test(img_request); |
f1a4739f AE |
1915 | struct bio *bio_list; |
1916 | unsigned int bio_offset = 0; | |
1917 | struct page **pages; | |
7da22d29 | 1918 | u64 img_offset; |
bf0d5f50 AE |
1919 | u64 resid; |
1920 | u16 opcode; | |
1921 | ||
f1a4739f AE |
1922 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
1923 | (int)type, data_desc); | |
37206ee5 | 1924 | |
430c28c3 | 1925 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 1926 | img_offset = img_request->offset; |
bf0d5f50 | 1927 | resid = img_request->length; |
4dda41d3 | 1928 | rbd_assert(resid > 0); |
f1a4739f AE |
1929 | |
1930 | if (type == OBJ_REQUEST_BIO) { | |
1931 | bio_list = data_desc; | |
1932 | rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
1933 | } else { | |
1934 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
1935 | pages = data_desc; | |
1936 | } | |
1937 | ||
bf0d5f50 | 1938 | while (resid) { |
2fa12320 | 1939 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 1940 | const char *object_name; |
bf0d5f50 AE |
1941 | u64 offset; |
1942 | u64 length; | |
1943 | ||
7da22d29 | 1944 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
1945 | if (!object_name) |
1946 | goto out_unwind; | |
7da22d29 AE |
1947 | offset = rbd_segment_offset(rbd_dev, img_offset); |
1948 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 1949 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 1950 | offset, length, type); |
bf0d5f50 AE |
1951 | kfree(object_name); /* object request has its own copy */ |
1952 | if (!obj_request) | |
1953 | goto out_unwind; | |
1954 | ||
f1a4739f AE |
1955 | if (type == OBJ_REQUEST_BIO) { |
1956 | unsigned int clone_size; | |
1957 | ||
1958 | rbd_assert(length <= (u64)UINT_MAX); | |
1959 | clone_size = (unsigned int)length; | |
1960 | obj_request->bio_list = | |
1961 | bio_chain_clone_range(&bio_list, | |
1962 | &bio_offset, | |
1963 | clone_size, | |
1964 | GFP_ATOMIC); | |
1965 | if (!obj_request->bio_list) | |
1966 | goto out_partial; | |
1967 | } else { | |
1968 | unsigned int page_count; | |
1969 | ||
1970 | obj_request->pages = pages; | |
1971 | page_count = (u32)calc_pages_for(offset, length); | |
1972 | obj_request->page_count = page_count; | |
1973 | if ((offset + length) & ~PAGE_MASK) | |
1974 | page_count--; /* more on last page */ | |
1975 | pages += page_count; | |
1976 | } | |
bf0d5f50 | 1977 | |
2fa12320 AE |
1978 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
1979 | obj_request); | |
1980 | if (!osd_req) | |
bf0d5f50 | 1981 | goto out_partial; |
2fa12320 | 1982 | obj_request->osd_req = osd_req; |
2169238d | 1983 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 1984 | |
2fa12320 AE |
1985 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
1986 | 0, 0); | |
f1a4739f AE |
1987 | if (type == OBJ_REQUEST_BIO) |
1988 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
1989 | obj_request->bio_list, length); | |
1990 | else | |
1991 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
1992 | obj_request->pages, length, | |
1993 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
1994 | |
1995 | if (write_request) | |
1996 | rbd_osd_req_format_write(obj_request); | |
1997 | else | |
1998 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 1999 | |
7da22d29 | 2000 | obj_request->img_offset = img_offset; |
bf0d5f50 AE |
2001 | rbd_img_obj_request_add(img_request, obj_request); |
2002 | ||
7da22d29 | 2003 | img_offset += length; |
bf0d5f50 AE |
2004 | resid -= length; |
2005 | } | |
2006 | ||
2007 | return 0; | |
2008 | ||
2009 | out_partial: | |
2010 | rbd_obj_request_put(obj_request); | |
2011 | out_unwind: | |
2012 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
2013 | rbd_obj_request_put(obj_request); | |
2014 | ||
2015 | return -ENOMEM; | |
2016 | } | |
2017 | ||
0eefd470 AE |
2018 | static void |
2019 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2020 | { | |
2021 | struct rbd_img_request *img_request; | |
2022 | struct rbd_device *rbd_dev; | |
2023 | u64 length; | |
2024 | u32 page_count; | |
2025 | ||
2026 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2027 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2028 | img_request = obj_request->img_request; | |
2029 | rbd_assert(img_request); | |
2030 | ||
2031 | rbd_dev = img_request->rbd_dev; | |
2032 | rbd_assert(rbd_dev); | |
2033 | length = (u64)1 << rbd_dev->header.obj_order; | |
2034 | page_count = (u32)calc_pages_for(0, length); | |
2035 | ||
2036 | rbd_assert(obj_request->copyup_pages); | |
2037 | ceph_release_page_vector(obj_request->copyup_pages, page_count); | |
2038 | obj_request->copyup_pages = NULL; | |
2039 | ||
2040 | /* | |
2041 | * We want the transfer count to reflect the size of the | |
2042 | * original write request. There is no such thing as a | |
2043 | * successful short write, so if the request was successful | |
2044 | * we can just set it to the originally-requested length. | |
2045 | */ | |
2046 | if (!obj_request->result) | |
2047 | obj_request->xferred = obj_request->length; | |
2048 | ||
2049 | /* Finish up with the normal image object callback */ | |
2050 | ||
2051 | rbd_img_obj_callback(obj_request); | |
2052 | } | |
2053 | ||
3d7efd18 AE |
2054 | static void |
2055 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2056 | { | |
2057 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2058 | struct ceph_osd_request *osd_req; |
2059 | struct ceph_osd_client *osdc; | |
2060 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2061 | struct page **pages; |
3d7efd18 AE |
2062 | int result; |
2063 | u64 obj_size; | |
2064 | u64 xferred; | |
2065 | ||
2066 | rbd_assert(img_request_child_test(img_request)); | |
2067 | ||
2068 | /* First get what we need from the image request */ | |
2069 | ||
2070 | pages = img_request->copyup_pages; | |
2071 | rbd_assert(pages != NULL); | |
2072 | img_request->copyup_pages = NULL; | |
2073 | ||
2074 | orig_request = img_request->obj_request; | |
2075 | rbd_assert(orig_request != NULL); | |
0eefd470 | 2076 | rbd_assert(orig_request->type == OBJ_REQUEST_BIO); |
3d7efd18 AE |
2077 | result = img_request->result; |
2078 | obj_size = img_request->length; | |
2079 | xferred = img_request->xferred; | |
2080 | ||
0eefd470 AE |
2081 | rbd_dev = img_request->rbd_dev; |
2082 | rbd_assert(rbd_dev); | |
2083 | rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order); | |
2084 | ||
3d7efd18 AE |
2085 | rbd_img_request_put(img_request); |
2086 | ||
0eefd470 AE |
2087 | if (result) |
2088 | goto out_err; | |
2089 | ||
2090 | /* Allocate the new copyup osd request for the original request */ | |
2091 | ||
2092 | result = -ENOMEM; | |
2093 | rbd_assert(!orig_request->osd_req); | |
2094 | osd_req = rbd_osd_req_create_copyup(orig_request); | |
2095 | if (!osd_req) | |
2096 | goto out_err; | |
2097 | orig_request->osd_req = osd_req; | |
2098 | orig_request->copyup_pages = pages; | |
3d7efd18 | 2099 | |
0eefd470 | 2100 | /* Initialize the copyup op */ |
3d7efd18 | 2101 | |
0eefd470 AE |
2102 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
2103 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0, | |
2104 | false, false); | |
3d7efd18 | 2105 | |
0eefd470 AE |
2106 | /* Then the original write request op */ |
2107 | ||
2108 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, | |
2109 | orig_request->offset, | |
2110 | orig_request->length, 0, 0); | |
2111 | osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list, | |
2112 | orig_request->length); | |
2113 | ||
2114 | rbd_osd_req_format_write(orig_request); | |
2115 | ||
2116 | /* All set, send it off. */ | |
2117 | ||
2118 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2119 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2120 | result = rbd_obj_request_submit(osdc, orig_request); | |
2121 | if (!result) | |
2122 | return; | |
2123 | out_err: | |
2124 | /* Record the error code and complete the request */ | |
2125 | ||
2126 | orig_request->result = result; | |
2127 | orig_request->xferred = 0; | |
2128 | obj_request_done_set(orig_request); | |
2129 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2130 | } |
2131 | ||
2132 | /* | |
2133 | * Read from the parent image the range of data that covers the | |
2134 | * entire target of the given object request. This is used for | |
2135 | * satisfying a layered image write request when the target of an | |
2136 | * object request from the image request does not exist. | |
2137 | * | |
2138 | * A page array big enough to hold the returned data is allocated | |
2139 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2140 | * When the read completes, this page array will be transferred to | |
2141 | * the original object request for the copyup operation. | |
2142 | * | |
2143 | * If an error occurs, record it as the result of the original | |
2144 | * object request and mark it done so it gets completed. | |
2145 | */ | |
2146 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2147 | { | |
2148 | struct rbd_img_request *img_request = NULL; | |
2149 | struct rbd_img_request *parent_request = NULL; | |
2150 | struct rbd_device *rbd_dev; | |
2151 | u64 img_offset; | |
2152 | u64 length; | |
2153 | struct page **pages = NULL; | |
2154 | u32 page_count; | |
2155 | int result; | |
2156 | ||
2157 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2158 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2159 | ||
2160 | img_request = obj_request->img_request; | |
2161 | rbd_assert(img_request != NULL); | |
2162 | rbd_dev = img_request->rbd_dev; | |
2163 | rbd_assert(rbd_dev->parent != NULL); | |
2164 | ||
0eefd470 AE |
2165 | /* |
2166 | * First things first. The original osd request is of no | |
2167 | * use to use any more, we'll need a new one that can hold | |
2168 | * the two ops in a copyup request. We'll get that later, | |
2169 | * but for now we can release the old one. | |
2170 | */ | |
2171 | rbd_osd_req_destroy(obj_request->osd_req); | |
2172 | obj_request->osd_req = NULL; | |
2173 | ||
3d7efd18 AE |
2174 | /* |
2175 | * Determine the byte range covered by the object in the | |
2176 | * child image to which the original request was to be sent. | |
2177 | */ | |
2178 | img_offset = obj_request->img_offset - obj_request->offset; | |
2179 | length = (u64)1 << rbd_dev->header.obj_order; | |
2180 | ||
a9e8ba2c AE |
2181 | /* |
2182 | * There is no defined parent data beyond the parent | |
2183 | * overlap, so limit what we read at that boundary if | |
2184 | * necessary. | |
2185 | */ | |
2186 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2187 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2188 | length = rbd_dev->parent_overlap - img_offset; | |
2189 | } | |
2190 | ||
3d7efd18 AE |
2191 | /* |
2192 | * Allocate a page array big enough to receive the data read | |
2193 | * from the parent. | |
2194 | */ | |
2195 | page_count = (u32)calc_pages_for(0, length); | |
2196 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2197 | if (IS_ERR(pages)) { | |
2198 | result = PTR_ERR(pages); | |
2199 | pages = NULL; | |
2200 | goto out_err; | |
2201 | } | |
2202 | ||
2203 | result = -ENOMEM; | |
2204 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2205 | img_offset, length, | |
2206 | false, true); | |
2207 | if (!parent_request) | |
2208 | goto out_err; | |
2209 | rbd_obj_request_get(obj_request); | |
2210 | parent_request->obj_request = obj_request; | |
2211 | ||
2212 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2213 | if (result) | |
2214 | goto out_err; | |
2215 | parent_request->copyup_pages = pages; | |
2216 | ||
2217 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2218 | result = rbd_img_request_submit(parent_request); | |
2219 | if (!result) | |
2220 | return 0; | |
2221 | ||
2222 | parent_request->copyup_pages = NULL; | |
2223 | parent_request->obj_request = NULL; | |
2224 | rbd_obj_request_put(obj_request); | |
2225 | out_err: | |
2226 | if (pages) | |
2227 | ceph_release_page_vector(pages, page_count); | |
2228 | if (parent_request) | |
2229 | rbd_img_request_put(parent_request); | |
2230 | obj_request->result = result; | |
2231 | obj_request->xferred = 0; | |
2232 | obj_request_done_set(obj_request); | |
2233 | ||
2234 | return result; | |
2235 | } | |
2236 | ||
c5b5ef6c AE |
2237 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2238 | { | |
c5b5ef6c AE |
2239 | struct rbd_obj_request *orig_request; |
2240 | int result; | |
2241 | ||
2242 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2243 | ||
2244 | /* | |
2245 | * All we need from the object request is the original | |
2246 | * request and the result of the STAT op. Grab those, then | |
2247 | * we're done with the request. | |
2248 | */ | |
2249 | orig_request = obj_request->obj_request; | |
2250 | obj_request->obj_request = NULL; | |
2251 | rbd_assert(orig_request); | |
2252 | rbd_assert(orig_request->img_request); | |
2253 | ||
2254 | result = obj_request->result; | |
2255 | obj_request->result = 0; | |
2256 | ||
2257 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2258 | obj_request, orig_request, result, | |
2259 | obj_request->xferred, obj_request->length); | |
2260 | rbd_obj_request_put(obj_request); | |
2261 | ||
2262 | rbd_assert(orig_request); | |
2263 | rbd_assert(orig_request->img_request); | |
c5b5ef6c AE |
2264 | |
2265 | /* | |
2266 | * Our only purpose here is to determine whether the object | |
2267 | * exists, and we don't want to treat the non-existence as | |
2268 | * an error. If something else comes back, transfer the | |
2269 | * error to the original request and complete it now. | |
2270 | */ | |
2271 | if (!result) { | |
2272 | obj_request_existence_set(orig_request, true); | |
2273 | } else if (result == -ENOENT) { | |
2274 | obj_request_existence_set(orig_request, false); | |
2275 | } else if (result) { | |
2276 | orig_request->result = result; | |
3d7efd18 | 2277 | goto out; |
c5b5ef6c AE |
2278 | } |
2279 | ||
2280 | /* | |
2281 | * Resubmit the original request now that we have recorded | |
2282 | * whether the target object exists. | |
2283 | */ | |
b454e36d | 2284 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2285 | out: |
c5b5ef6c AE |
2286 | if (orig_request->result) |
2287 | rbd_obj_request_complete(orig_request); | |
2288 | rbd_obj_request_put(orig_request); | |
2289 | } | |
2290 | ||
2291 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2292 | { | |
2293 | struct rbd_obj_request *stat_request; | |
2294 | struct rbd_device *rbd_dev; | |
2295 | struct ceph_osd_client *osdc; | |
2296 | struct page **pages = NULL; | |
2297 | u32 page_count; | |
2298 | size_t size; | |
2299 | int ret; | |
2300 | ||
2301 | /* | |
2302 | * The response data for a STAT call consists of: | |
2303 | * le64 length; | |
2304 | * struct { | |
2305 | * le32 tv_sec; | |
2306 | * le32 tv_nsec; | |
2307 | * } mtime; | |
2308 | */ | |
2309 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2310 | page_count = (u32)calc_pages_for(0, size); | |
2311 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2312 | if (IS_ERR(pages)) | |
2313 | return PTR_ERR(pages); | |
2314 | ||
2315 | ret = -ENOMEM; | |
2316 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2317 | OBJ_REQUEST_PAGES); | |
2318 | if (!stat_request) | |
2319 | goto out; | |
2320 | ||
2321 | rbd_obj_request_get(obj_request); | |
2322 | stat_request->obj_request = obj_request; | |
2323 | stat_request->pages = pages; | |
2324 | stat_request->page_count = page_count; | |
2325 | ||
2326 | rbd_assert(obj_request->img_request); | |
2327 | rbd_dev = obj_request->img_request->rbd_dev; | |
2328 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2329 | stat_request); | |
2330 | if (!stat_request->osd_req) | |
2331 | goto out; | |
2332 | stat_request->callback = rbd_img_obj_exists_callback; | |
2333 | ||
2334 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2335 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2336 | false, false); | |
9d4df01f | 2337 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2338 | |
2339 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2340 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2341 | out: | |
2342 | if (ret) | |
2343 | rbd_obj_request_put(obj_request); | |
2344 | ||
2345 | return ret; | |
2346 | } | |
2347 | ||
b454e36d AE |
2348 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2349 | { | |
2350 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2351 | struct rbd_device *rbd_dev; |
3d7efd18 | 2352 | bool known; |
b454e36d AE |
2353 | |
2354 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2355 | ||
2356 | img_request = obj_request->img_request; | |
2357 | rbd_assert(img_request); | |
a9e8ba2c | 2358 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2359 | |
b454e36d | 2360 | /* |
a9e8ba2c AE |
2361 | * Only writes to layered images need special handling. |
2362 | * Reads and non-layered writes are simple object requests. | |
2363 | * Layered writes that start beyond the end of the overlap | |
2364 | * with the parent have no parent data, so they too are | |
2365 | * simple object requests. Finally, if the target object is | |
2366 | * known to already exist, its parent data has already been | |
2367 | * copied, so a write to the object can also be handled as a | |
2368 | * simple object request. | |
b454e36d AE |
2369 | */ |
2370 | if (!img_request_write_test(img_request) || | |
2371 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2372 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2373 | ((known = obj_request_known_test(obj_request)) && |
2374 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2375 | |
2376 | struct rbd_device *rbd_dev; | |
2377 | struct ceph_osd_client *osdc; | |
2378 | ||
2379 | rbd_dev = obj_request->img_request->rbd_dev; | |
2380 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2381 | ||
2382 | return rbd_obj_request_submit(osdc, obj_request); | |
2383 | } | |
2384 | ||
2385 | /* | |
3d7efd18 AE |
2386 | * It's a layered write. The target object might exist but |
2387 | * we may not know that yet. If we know it doesn't exist, | |
2388 | * start by reading the data for the full target object from | |
2389 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2390 | */ |
3d7efd18 AE |
2391 | if (known) |
2392 | return rbd_img_obj_parent_read_full(obj_request); | |
2393 | ||
2394 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2395 | |
2396 | return rbd_img_obj_exists_submit(obj_request); | |
2397 | } | |
2398 | ||
bf0d5f50 AE |
2399 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2400 | { | |
bf0d5f50 | 2401 | struct rbd_obj_request *obj_request; |
46faeed4 | 2402 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2403 | |
37206ee5 | 2404 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2405 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2406 | int ret; |
2407 | ||
b454e36d | 2408 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2409 | if (ret) |
2410 | return ret; | |
bf0d5f50 AE |
2411 | } |
2412 | ||
2413 | return 0; | |
2414 | } | |
8b3e1a56 AE |
2415 | |
2416 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2417 | { | |
2418 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2419 | struct rbd_device *rbd_dev; |
2420 | u64 obj_end; | |
8b3e1a56 AE |
2421 | |
2422 | rbd_assert(img_request_child_test(img_request)); | |
2423 | ||
2424 | obj_request = img_request->obj_request; | |
a9e8ba2c AE |
2425 | rbd_assert(obj_request); |
2426 | rbd_assert(obj_request->img_request); | |
2427 | ||
8b3e1a56 | 2428 | obj_request->result = img_request->result; |
a9e8ba2c AE |
2429 | if (obj_request->result) |
2430 | goto out; | |
2431 | ||
2432 | /* | |
2433 | * We need to zero anything beyond the parent overlap | |
2434 | * boundary. Since rbd_img_obj_request_read_callback() | |
2435 | * will zero anything beyond the end of a short read, an | |
2436 | * easy way to do this is to pretend the data from the | |
2437 | * parent came up short--ending at the overlap boundary. | |
2438 | */ | |
2439 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2440 | obj_end = obj_request->img_offset + obj_request->length; | |
2441 | rbd_dev = obj_request->img_request->rbd_dev; | |
2442 | if (obj_end > rbd_dev->parent_overlap) { | |
2443 | u64 xferred = 0; | |
2444 | ||
2445 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2446 | xferred = rbd_dev->parent_overlap - | |
2447 | obj_request->img_offset; | |
8b3e1a56 | 2448 | |
a9e8ba2c AE |
2449 | obj_request->xferred = min(img_request->xferred, xferred); |
2450 | } else { | |
2451 | obj_request->xferred = img_request->xferred; | |
2452 | } | |
2453 | out: | |
8b3e1a56 AE |
2454 | rbd_img_obj_request_read_callback(obj_request); |
2455 | rbd_obj_request_complete(obj_request); | |
2456 | } | |
2457 | ||
2458 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2459 | { | |
2460 | struct rbd_device *rbd_dev; | |
2461 | struct rbd_img_request *img_request; | |
2462 | int result; | |
2463 | ||
2464 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2465 | rbd_assert(obj_request->img_request != NULL); | |
2466 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
2467 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2468 | ||
2469 | rbd_dev = obj_request->img_request->rbd_dev; | |
2470 | rbd_assert(rbd_dev->parent != NULL); | |
2471 | /* rbd_read_finish(obj_request, obj_request->length); */ | |
2472 | img_request = rbd_img_request_create(rbd_dev->parent, | |
2473 | obj_request->img_offset, | |
2474 | obj_request->length, | |
2475 | false, true); | |
2476 | result = -ENOMEM; | |
2477 | if (!img_request) | |
2478 | goto out_err; | |
2479 | ||
2480 | rbd_obj_request_get(obj_request); | |
2481 | img_request->obj_request = obj_request; | |
2482 | ||
f1a4739f AE |
2483 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2484 | obj_request->bio_list); | |
8b3e1a56 AE |
2485 | if (result) |
2486 | goto out_err; | |
2487 | ||
2488 | img_request->callback = rbd_img_parent_read_callback; | |
2489 | result = rbd_img_request_submit(img_request); | |
2490 | if (result) | |
2491 | goto out_err; | |
2492 | ||
2493 | return; | |
2494 | out_err: | |
2495 | if (img_request) | |
2496 | rbd_img_request_put(img_request); | |
2497 | obj_request->result = result; | |
2498 | obj_request->xferred = 0; | |
2499 | obj_request_done_set(obj_request); | |
2500 | } | |
bf0d5f50 | 2501 | |
cf81b60e | 2502 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, |
b8d70035 AE |
2503 | u64 ver, u64 notify_id) |
2504 | { | |
2505 | struct rbd_obj_request *obj_request; | |
2169238d | 2506 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2507 | int ret; |
2508 | ||
2509 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2510 | OBJ_REQUEST_NODATA); | |
2511 | if (!obj_request) | |
2512 | return -ENOMEM; | |
2513 | ||
2514 | ret = -ENOMEM; | |
430c28c3 | 2515 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2516 | if (!obj_request->osd_req) |
2517 | goto out; | |
2169238d | 2518 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 2519 | |
c99d2d4a AE |
2520 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
2521 | notify_id, ver, 0); | |
9d4df01f | 2522 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2523 | |
b8d70035 | 2524 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 2525 | out: |
cf81b60e AE |
2526 | if (ret) |
2527 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2528 | |
2529 | return ret; | |
2530 | } | |
2531 | ||
2532 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2533 | { | |
2534 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
2535 | u64 hver; | |
b8d70035 AE |
2536 | |
2537 | if (!rbd_dev) | |
2538 | return; | |
2539 | ||
37206ee5 | 2540 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
b8d70035 AE |
2541 | rbd_dev->header_name, (unsigned long long) notify_id, |
2542 | (unsigned int) opcode); | |
522a0cc0 | 2543 | (void)rbd_dev_refresh(rbd_dev, &hver); |
b8d70035 | 2544 | |
cf81b60e | 2545 | rbd_obj_notify_ack(rbd_dev, hver, notify_id); |
b8d70035 AE |
2546 | } |
2547 | ||
9969ebc5 AE |
2548 | /* |
2549 | * Request sync osd watch/unwatch. The value of "start" determines | |
2550 | * whether a watch request is being initiated or torn down. | |
2551 | */ | |
2552 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
2553 | { | |
2554 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2555 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2556 | int ret; |
2557 | ||
2558 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2559 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2560 | ||
2561 | if (start) { | |
3c663bbd | 2562 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2563 | &rbd_dev->watch_event); |
2564 | if (ret < 0) | |
2565 | return ret; | |
8eb87565 | 2566 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2567 | } |
2568 | ||
2569 | ret = -ENOMEM; | |
2570 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2571 | OBJ_REQUEST_NODATA); | |
2572 | if (!obj_request) | |
2573 | goto out_cancel; | |
2574 | ||
430c28c3 AE |
2575 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2576 | if (!obj_request->osd_req) | |
2577 | goto out_cancel; | |
2578 | ||
8eb87565 | 2579 | if (start) |
975241af | 2580 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2581 | else |
6977c3f9 | 2582 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2583 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2584 | |
2585 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
2586 | rbd_dev->watch_event->cookie, | |
2587 | rbd_dev->header.obj_version, start); | |
9d4df01f | 2588 | rbd_osd_req_format_write(obj_request); |
2169238d | 2589 | |
9969ebc5 AE |
2590 | ret = rbd_obj_request_submit(osdc, obj_request); |
2591 | if (ret) | |
2592 | goto out_cancel; | |
2593 | ret = rbd_obj_request_wait(obj_request); | |
2594 | if (ret) | |
2595 | goto out_cancel; | |
9969ebc5 AE |
2596 | ret = obj_request->result; |
2597 | if (ret) | |
2598 | goto out_cancel; | |
2599 | ||
8eb87565 AE |
2600 | /* |
2601 | * A watch request is set to linger, so the underlying osd | |
2602 | * request won't go away until we unregister it. We retain | |
2603 | * a pointer to the object request during that time (in | |
2604 | * rbd_dev->watch_request), so we'll keep a reference to | |
2605 | * it. We'll drop that reference (below) after we've | |
2606 | * unregistered it. | |
2607 | */ | |
2608 | if (start) { | |
2609 | rbd_dev->watch_request = obj_request; | |
2610 | ||
2611 | return 0; | |
2612 | } | |
2613 | ||
2614 | /* We have successfully torn down the watch request */ | |
2615 | ||
2616 | rbd_obj_request_put(rbd_dev->watch_request); | |
2617 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2618 | out_cancel: |
2619 | /* Cancel the event if we're tearing down, or on error */ | |
2620 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2621 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2622 | if (obj_request) |
2623 | rbd_obj_request_put(obj_request); | |
2624 | ||
2625 | return ret; | |
2626 | } | |
2627 | ||
36be9a76 | 2628 | /* |
f40eb349 AE |
2629 | * Synchronous osd object method call. Returns the number of bytes |
2630 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2631 | */ |
2632 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2633 | const char *object_name, | |
2634 | const char *class_name, | |
2635 | const char *method_name, | |
4157976b | 2636 | const void *outbound, |
36be9a76 | 2637 | size_t outbound_size, |
4157976b | 2638 | void *inbound, |
36be9a76 AE |
2639 | size_t inbound_size, |
2640 | u64 *version) | |
2641 | { | |
2169238d | 2642 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2643 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2644 | struct page **pages; |
2645 | u32 page_count; | |
2646 | int ret; | |
2647 | ||
2648 | /* | |
6010a451 AE |
2649 | * Method calls are ultimately read operations. The result |
2650 | * should placed into the inbound buffer provided. They | |
2651 | * also supply outbound data--parameters for the object | |
2652 | * method. Currently if this is present it will be a | |
2653 | * snapshot id. | |
36be9a76 | 2654 | */ |
57385b51 | 2655 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2656 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2657 | if (IS_ERR(pages)) | |
2658 | return PTR_ERR(pages); | |
2659 | ||
2660 | ret = -ENOMEM; | |
6010a451 | 2661 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2662 | OBJ_REQUEST_PAGES); |
2663 | if (!obj_request) | |
2664 | goto out; | |
2665 | ||
2666 | obj_request->pages = pages; | |
2667 | obj_request->page_count = page_count; | |
2668 | ||
430c28c3 | 2669 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2670 | if (!obj_request->osd_req) |
2671 | goto out; | |
2672 | ||
c99d2d4a | 2673 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2674 | class_name, method_name); |
2675 | if (outbound_size) { | |
2676 | struct ceph_pagelist *pagelist; | |
2677 | ||
2678 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2679 | if (!pagelist) | |
2680 | goto out; | |
2681 | ||
2682 | ceph_pagelist_init(pagelist); | |
2683 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
2684 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
2685 | pagelist); | |
2686 | } | |
a4ce40a9 AE |
2687 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
2688 | obj_request->pages, inbound_size, | |
44cd188d | 2689 | 0, false, false); |
9d4df01f | 2690 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2691 | |
36be9a76 AE |
2692 | ret = rbd_obj_request_submit(osdc, obj_request); |
2693 | if (ret) | |
2694 | goto out; | |
2695 | ret = rbd_obj_request_wait(obj_request); | |
2696 | if (ret) | |
2697 | goto out; | |
2698 | ||
2699 | ret = obj_request->result; | |
2700 | if (ret < 0) | |
2701 | goto out; | |
57385b51 AE |
2702 | |
2703 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
2704 | ret = (int)obj_request->xferred; | |
903bb32e | 2705 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
2706 | if (version) |
2707 | *version = obj_request->version; | |
2708 | out: | |
2709 | if (obj_request) | |
2710 | rbd_obj_request_put(obj_request); | |
2711 | else | |
2712 | ceph_release_page_vector(pages, page_count); | |
2713 | ||
2714 | return ret; | |
2715 | } | |
2716 | ||
bf0d5f50 | 2717 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 2718 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
2719 | { |
2720 | struct rbd_device *rbd_dev = q->queuedata; | |
2721 | bool read_only = rbd_dev->mapping.read_only; | |
2722 | struct request *rq; | |
2723 | int result; | |
2724 | ||
2725 | while ((rq = blk_fetch_request(q))) { | |
2726 | bool write_request = rq_data_dir(rq) == WRITE; | |
2727 | struct rbd_img_request *img_request; | |
2728 | u64 offset; | |
2729 | u64 length; | |
2730 | ||
2731 | /* Ignore any non-FS requests that filter through. */ | |
2732 | ||
2733 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
2734 | dout("%s: non-fs request type %d\n", __func__, |
2735 | (int) rq->cmd_type); | |
2736 | __blk_end_request_all(rq, 0); | |
2737 | continue; | |
2738 | } | |
2739 | ||
2740 | /* Ignore/skip any zero-length requests */ | |
2741 | ||
2742 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
2743 | length = (u64) blk_rq_bytes(rq); | |
2744 | ||
2745 | if (!length) { | |
2746 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
2747 | __blk_end_request_all(rq, 0); |
2748 | continue; | |
2749 | } | |
2750 | ||
2751 | spin_unlock_irq(q->queue_lock); | |
2752 | ||
2753 | /* Disallow writes to a read-only device */ | |
2754 | ||
2755 | if (write_request) { | |
2756 | result = -EROFS; | |
2757 | if (read_only) | |
2758 | goto end_request; | |
2759 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
2760 | } | |
2761 | ||
6d292906 AE |
2762 | /* |
2763 | * Quit early if the mapped snapshot no longer | |
2764 | * exists. It's still possible the snapshot will | |
2765 | * have disappeared by the time our request arrives | |
2766 | * at the osd, but there's no sense in sending it if | |
2767 | * we already know. | |
2768 | */ | |
2769 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
2770 | dout("request for non-existent snapshot"); |
2771 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
2772 | result = -ENXIO; | |
2773 | goto end_request; | |
2774 | } | |
2775 | ||
bf0d5f50 | 2776 | result = -EINVAL; |
c0cd10db AE |
2777 | if (offset && length > U64_MAX - offset + 1) { |
2778 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
2779 | offset, length); | |
bf0d5f50 | 2780 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 2781 | } |
bf0d5f50 AE |
2782 | |
2783 | result = -ENOMEM; | |
2784 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
9849e986 | 2785 | write_request, false); |
bf0d5f50 AE |
2786 | if (!img_request) |
2787 | goto end_request; | |
2788 | ||
2789 | img_request->rq = rq; | |
2790 | ||
f1a4739f AE |
2791 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2792 | rq->bio); | |
bf0d5f50 AE |
2793 | if (!result) |
2794 | result = rbd_img_request_submit(img_request); | |
2795 | if (result) | |
2796 | rbd_img_request_put(img_request); | |
2797 | end_request: | |
2798 | spin_lock_irq(q->queue_lock); | |
2799 | if (result < 0) { | |
7da22d29 AE |
2800 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
2801 | write_request ? "write" : "read", | |
2802 | length, offset, result); | |
2803 | ||
bf0d5f50 AE |
2804 | __blk_end_request_all(rq, result); |
2805 | } | |
2806 | } | |
2807 | } | |
2808 | ||
602adf40 YS |
2809 | /* |
2810 | * a queue callback. Makes sure that we don't create a bio that spans across | |
2811 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 2812 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
2813 | */ |
2814 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
2815 | struct bio_vec *bvec) | |
2816 | { | |
2817 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
2818 | sector_t sector_offset; |
2819 | sector_t sectors_per_obj; | |
2820 | sector_t obj_sector_offset; | |
2821 | int ret; | |
2822 | ||
2823 | /* | |
2824 | * Find how far into its rbd object the partition-relative | |
2825 | * bio start sector is to offset relative to the enclosing | |
2826 | * device. | |
2827 | */ | |
2828 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
2829 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
2830 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
2831 | ||
2832 | /* | |
2833 | * Compute the number of bytes from that offset to the end | |
2834 | * of the object. Account for what's already used by the bio. | |
2835 | */ | |
2836 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2837 | if (ret > bmd->bi_size) | |
2838 | ret -= bmd->bi_size; | |
2839 | else | |
2840 | ret = 0; | |
2841 | ||
2842 | /* | |
2843 | * Don't send back more than was asked for. And if the bio | |
2844 | * was empty, let the whole thing through because: "Note | |
2845 | * that a block device *must* allow a single page to be | |
2846 | * added to an empty bio." | |
2847 | */ | |
2848 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2849 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2850 | ret = (int) bvec->bv_len; | |
2851 | ||
2852 | return ret; | |
602adf40 YS |
2853 | } |
2854 | ||
2855 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2856 | { | |
2857 | struct gendisk *disk = rbd_dev->disk; | |
2858 | ||
2859 | if (!disk) | |
2860 | return; | |
2861 | ||
a0cab924 AE |
2862 | rbd_dev->disk = NULL; |
2863 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 2864 | del_gendisk(disk); |
a0cab924 AE |
2865 | if (disk->queue) |
2866 | blk_cleanup_queue(disk->queue); | |
2867 | } | |
602adf40 YS |
2868 | put_disk(disk); |
2869 | } | |
2870 | ||
788e2df3 AE |
2871 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2872 | const char *object_name, | |
2873 | u64 offset, u64 length, | |
80ef15bf | 2874 | void *buf, u64 *version) |
788e2df3 AE |
2875 | |
2876 | { | |
2169238d | 2877 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 2878 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
2879 | struct page **pages = NULL; |
2880 | u32 page_count; | |
1ceae7ef | 2881 | size_t size; |
788e2df3 AE |
2882 | int ret; |
2883 | ||
2884 | page_count = (u32) calc_pages_for(offset, length); | |
2885 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2886 | if (IS_ERR(pages)) | |
2887 | ret = PTR_ERR(pages); | |
2888 | ||
2889 | ret = -ENOMEM; | |
2890 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2891 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2892 | if (!obj_request) |
2893 | goto out; | |
2894 | ||
2895 | obj_request->pages = pages; | |
2896 | obj_request->page_count = page_count; | |
2897 | ||
430c28c3 | 2898 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
2899 | if (!obj_request->osd_req) |
2900 | goto out; | |
2901 | ||
c99d2d4a AE |
2902 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
2903 | offset, length, 0, 0); | |
406e2c9f | 2904 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 2905 | obj_request->pages, |
44cd188d AE |
2906 | obj_request->length, |
2907 | obj_request->offset & ~PAGE_MASK, | |
2908 | false, false); | |
9d4df01f | 2909 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2910 | |
788e2df3 AE |
2911 | ret = rbd_obj_request_submit(osdc, obj_request); |
2912 | if (ret) | |
2913 | goto out; | |
2914 | ret = rbd_obj_request_wait(obj_request); | |
2915 | if (ret) | |
2916 | goto out; | |
2917 | ||
2918 | ret = obj_request->result; | |
2919 | if (ret < 0) | |
2920 | goto out; | |
1ceae7ef AE |
2921 | |
2922 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
2923 | size = (size_t) obj_request->xferred; | |
903bb32e | 2924 | ceph_copy_from_page_vector(pages, buf, 0, size); |
23ed6e13 AE |
2925 | rbd_assert(size <= (size_t) INT_MAX); |
2926 | ret = (int) size; | |
788e2df3 AE |
2927 | if (version) |
2928 | *version = obj_request->version; | |
2929 | out: | |
2930 | if (obj_request) | |
2931 | rbd_obj_request_put(obj_request); | |
2932 | else | |
2933 | ceph_release_page_vector(pages, page_count); | |
2934 | ||
2935 | return ret; | |
2936 | } | |
2937 | ||
602adf40 | 2938 | /* |
4156d998 AE |
2939 | * Read the complete header for the given rbd device. |
2940 | * | |
2941 | * Returns a pointer to a dynamically-allocated buffer containing | |
2942 | * the complete and validated header. Caller can pass the address | |
2943 | * of a variable that will be filled in with the version of the | |
2944 | * header object at the time it was read. | |
2945 | * | |
2946 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 2947 | */ |
4156d998 AE |
2948 | static struct rbd_image_header_ondisk * |
2949 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev, u64 *version) | |
602adf40 | 2950 | { |
4156d998 | 2951 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 2952 | u32 snap_count = 0; |
4156d998 AE |
2953 | u64 names_size = 0; |
2954 | u32 want_count; | |
2955 | int ret; | |
602adf40 | 2956 | |
00f1f36f | 2957 | /* |
4156d998 AE |
2958 | * The complete header will include an array of its 64-bit |
2959 | * snapshot ids, followed by the names of those snapshots as | |
2960 | * a contiguous block of NUL-terminated strings. Note that | |
2961 | * the number of snapshots could change by the time we read | |
2962 | * it in, in which case we re-read it. | |
00f1f36f | 2963 | */ |
4156d998 AE |
2964 | do { |
2965 | size_t size; | |
2966 | ||
2967 | kfree(ondisk); | |
2968 | ||
2969 | size = sizeof (*ondisk); | |
2970 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
2971 | size += names_size; | |
2972 | ondisk = kmalloc(size, GFP_KERNEL); | |
2973 | if (!ondisk) | |
2974 | return ERR_PTR(-ENOMEM); | |
2975 | ||
788e2df3 | 2976 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
80ef15bf | 2977 | 0, size, ondisk, version); |
4156d998 AE |
2978 | if (ret < 0) |
2979 | goto out_err; | |
c0cd10db | 2980 | if ((size_t)ret < size) { |
4156d998 | 2981 | ret = -ENXIO; |
06ecc6cb AE |
2982 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
2983 | size, ret); | |
4156d998 AE |
2984 | goto out_err; |
2985 | } | |
2986 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
2987 | ret = -ENXIO; | |
06ecc6cb | 2988 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 2989 | goto out_err; |
81e759fb | 2990 | } |
602adf40 | 2991 | |
4156d998 AE |
2992 | names_size = le64_to_cpu(ondisk->snap_names_len); |
2993 | want_count = snap_count; | |
2994 | snap_count = le32_to_cpu(ondisk->snap_count); | |
2995 | } while (snap_count != want_count); | |
00f1f36f | 2996 | |
4156d998 | 2997 | return ondisk; |
00f1f36f | 2998 | |
4156d998 AE |
2999 | out_err: |
3000 | kfree(ondisk); | |
3001 | ||
3002 | return ERR_PTR(ret); | |
3003 | } | |
3004 | ||
3005 | /* | |
3006 | * reload the ondisk the header | |
3007 | */ | |
3008 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
3009 | struct rbd_image_header *header) | |
3010 | { | |
3011 | struct rbd_image_header_ondisk *ondisk; | |
3012 | u64 ver = 0; | |
3013 | int ret; | |
602adf40 | 3014 | |
4156d998 AE |
3015 | ondisk = rbd_dev_v1_header_read(rbd_dev, &ver); |
3016 | if (IS_ERR(ondisk)) | |
3017 | return PTR_ERR(ondisk); | |
3018 | ret = rbd_header_from_disk(header, ondisk); | |
3019 | if (ret >= 0) | |
3020 | header->obj_version = ver; | |
3021 | kfree(ondisk); | |
3022 | ||
3023 | return ret; | |
602adf40 YS |
3024 | } |
3025 | ||
41f38c2b | 3026 | static void rbd_remove_all_snaps(struct rbd_device *rbd_dev) |
dfc5606d YS |
3027 | { |
3028 | struct rbd_snap *snap; | |
a0593290 | 3029 | struct rbd_snap *next; |
dfc5606d | 3030 | |
6087b51b AE |
3031 | list_for_each_entry_safe(snap, next, &rbd_dev->snaps, node) { |
3032 | list_del(&snap->node); | |
3033 | rbd_snap_destroy(snap); | |
3034 | } | |
dfc5606d YS |
3035 | } |
3036 | ||
9478554a AE |
3037 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
3038 | { | |
0d7dbfce | 3039 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
3040 | return; |
3041 | ||
e28626a0 AE |
3042 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) { |
3043 | sector_t size; | |
3044 | ||
3045 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
3046 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3047 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3048 | set_capacity(rbd_dev->disk, size); | |
3049 | } | |
9478554a AE |
3050 | } |
3051 | ||
602adf40 YS |
3052 | /* |
3053 | * only read the first part of the ondisk header, without the snaps info | |
3054 | */ | |
117973fb | 3055 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev, u64 *hver) |
602adf40 YS |
3056 | { |
3057 | int ret; | |
3058 | struct rbd_image_header h; | |
602adf40 YS |
3059 | |
3060 | ret = rbd_read_header(rbd_dev, &h); | |
3061 | if (ret < 0) | |
3062 | return ret; | |
3063 | ||
a51aa0c0 JD |
3064 | down_write(&rbd_dev->header_rwsem); |
3065 | ||
9478554a AE |
3066 | /* Update image size, and check for resize of mapped image */ |
3067 | rbd_dev->header.image_size = h.image_size; | |
3068 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 3069 | |
849b4260 | 3070 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 3071 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 3072 | kfree(rbd_dev->header.snap_names); |
d1d25646 | 3073 | /* osd requests may still refer to snapc */ |
468521c1 | 3074 | rbd_snap_context_put(rbd_dev->header.snapc); |
602adf40 | 3075 | |
b813623a AE |
3076 | if (hver) |
3077 | *hver = h.obj_version; | |
a71b891b | 3078 | rbd_dev->header.obj_version = h.obj_version; |
93a24e08 | 3079 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
3080 | rbd_dev->header.snapc = h.snapc; |
3081 | rbd_dev->header.snap_names = h.snap_names; | |
3082 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 | 3083 | /* Free the extra copy of the object prefix */ |
c0cd10db AE |
3084 | if (strcmp(rbd_dev->header.object_prefix, h.object_prefix)) |
3085 | rbd_warn(rbd_dev, "object prefix changed (ignoring)"); | |
849b4260 AE |
3086 | kfree(h.object_prefix); |
3087 | ||
304f6808 | 3088 | ret = rbd_dev_snaps_update(rbd_dev); |
dfc5606d | 3089 | |
c666601a | 3090 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 3091 | |
dfc5606d | 3092 | return ret; |
602adf40 YS |
3093 | } |
3094 | ||
117973fb | 3095 | static int rbd_dev_refresh(struct rbd_device *rbd_dev, u64 *hver) |
1fe5e993 AE |
3096 | { |
3097 | int ret; | |
3098 | ||
117973fb | 3099 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1fe5e993 | 3100 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb AE |
3101 | if (rbd_dev->image_format == 1) |
3102 | ret = rbd_dev_v1_refresh(rbd_dev, hver); | |
3103 | else | |
3104 | ret = rbd_dev_v2_refresh(rbd_dev, hver); | |
1fe5e993 | 3105 | mutex_unlock(&ctl_mutex); |
d98df63e | 3106 | revalidate_disk(rbd_dev->disk); |
522a0cc0 AE |
3107 | if (ret) |
3108 | rbd_warn(rbd_dev, "got notification but failed to " | |
3109 | " update snaps: %d\n", ret); | |
1fe5e993 AE |
3110 | |
3111 | return ret; | |
3112 | } | |
3113 | ||
602adf40 YS |
3114 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3115 | { | |
3116 | struct gendisk *disk; | |
3117 | struct request_queue *q; | |
593a9e7b | 3118 | u64 segment_size; |
602adf40 | 3119 | |
602adf40 | 3120 | /* create gendisk info */ |
602adf40 YS |
3121 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
3122 | if (!disk) | |
1fcdb8aa | 3123 | return -ENOMEM; |
602adf40 | 3124 | |
f0f8cef5 | 3125 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3126 | rbd_dev->dev_id); |
602adf40 YS |
3127 | disk->major = rbd_dev->major; |
3128 | disk->first_minor = 0; | |
3129 | disk->fops = &rbd_bd_ops; | |
3130 | disk->private_data = rbd_dev; | |
3131 | ||
bf0d5f50 | 3132 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3133 | if (!q) |
3134 | goto out_disk; | |
029bcbd8 | 3135 | |
593a9e7b AE |
3136 | /* We use the default size, but let's be explicit about it. */ |
3137 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3138 | ||
029bcbd8 | 3139 | /* set io sizes to object size */ |
593a9e7b AE |
3140 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3141 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3142 | blk_queue_max_segment_size(q, segment_size); | |
3143 | blk_queue_io_min(q, segment_size); | |
3144 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3145 | |
602adf40 YS |
3146 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3147 | disk->queue = q; | |
3148 | ||
3149 | q->queuedata = rbd_dev; | |
3150 | ||
3151 | rbd_dev->disk = disk; | |
602adf40 | 3152 | |
602adf40 | 3153 | return 0; |
602adf40 YS |
3154 | out_disk: |
3155 | put_disk(disk); | |
1fcdb8aa AE |
3156 | |
3157 | return -ENOMEM; | |
602adf40 YS |
3158 | } |
3159 | ||
dfc5606d YS |
3160 | /* |
3161 | sysfs | |
3162 | */ | |
3163 | ||
593a9e7b AE |
3164 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3165 | { | |
3166 | return container_of(dev, struct rbd_device, dev); | |
3167 | } | |
3168 | ||
dfc5606d YS |
3169 | static ssize_t rbd_size_show(struct device *dev, |
3170 | struct device_attribute *attr, char *buf) | |
3171 | { | |
593a9e7b | 3172 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3173 | |
fc71d833 AE |
3174 | return sprintf(buf, "%llu\n", |
3175 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3176 | } |
3177 | ||
34b13184 AE |
3178 | /* |
3179 | * Note this shows the features for whatever's mapped, which is not | |
3180 | * necessarily the base image. | |
3181 | */ | |
3182 | static ssize_t rbd_features_show(struct device *dev, | |
3183 | struct device_attribute *attr, char *buf) | |
3184 | { | |
3185 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3186 | ||
3187 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3188 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3189 | } |
3190 | ||
dfc5606d YS |
3191 | static ssize_t rbd_major_show(struct device *dev, |
3192 | struct device_attribute *attr, char *buf) | |
3193 | { | |
593a9e7b | 3194 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3195 | |
fc71d833 AE |
3196 | if (rbd_dev->major) |
3197 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3198 | ||
3199 | return sprintf(buf, "(none)\n"); | |
3200 | ||
dfc5606d YS |
3201 | } |
3202 | ||
3203 | static ssize_t rbd_client_id_show(struct device *dev, | |
3204 | struct device_attribute *attr, char *buf) | |
602adf40 | 3205 | { |
593a9e7b | 3206 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3207 | |
1dbb4399 AE |
3208 | return sprintf(buf, "client%lld\n", |
3209 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3210 | } |
3211 | ||
dfc5606d YS |
3212 | static ssize_t rbd_pool_show(struct device *dev, |
3213 | struct device_attribute *attr, char *buf) | |
602adf40 | 3214 | { |
593a9e7b | 3215 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3216 | |
0d7dbfce | 3217 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3218 | } |
3219 | ||
9bb2f334 AE |
3220 | static ssize_t rbd_pool_id_show(struct device *dev, |
3221 | struct device_attribute *attr, char *buf) | |
3222 | { | |
3223 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3224 | ||
0d7dbfce | 3225 | return sprintf(buf, "%llu\n", |
fc71d833 | 3226 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3227 | } |
3228 | ||
dfc5606d YS |
3229 | static ssize_t rbd_name_show(struct device *dev, |
3230 | struct device_attribute *attr, char *buf) | |
3231 | { | |
593a9e7b | 3232 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3233 | |
a92ffdf8 AE |
3234 | if (rbd_dev->spec->image_name) |
3235 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3236 | ||
3237 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3238 | } |
3239 | ||
589d30e0 AE |
3240 | static ssize_t rbd_image_id_show(struct device *dev, |
3241 | struct device_attribute *attr, char *buf) | |
3242 | { | |
3243 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3244 | ||
0d7dbfce | 3245 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3246 | } |
3247 | ||
34b13184 AE |
3248 | /* |
3249 | * Shows the name of the currently-mapped snapshot (or | |
3250 | * RBD_SNAP_HEAD_NAME for the base image). | |
3251 | */ | |
dfc5606d YS |
3252 | static ssize_t rbd_snap_show(struct device *dev, |
3253 | struct device_attribute *attr, | |
3254 | char *buf) | |
3255 | { | |
593a9e7b | 3256 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3257 | |
0d7dbfce | 3258 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3259 | } |
3260 | ||
86b00e0d AE |
3261 | /* |
3262 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3263 | * for the parent image. If there is no parent, simply shows | |
3264 | * "(no parent image)". | |
3265 | */ | |
3266 | static ssize_t rbd_parent_show(struct device *dev, | |
3267 | struct device_attribute *attr, | |
3268 | char *buf) | |
3269 | { | |
3270 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3271 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3272 | int count; | |
3273 | char *bufp = buf; | |
3274 | ||
3275 | if (!spec) | |
3276 | return sprintf(buf, "(no parent image)\n"); | |
3277 | ||
3278 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3279 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3280 | if (count < 0) | |
3281 | return count; | |
3282 | bufp += count; | |
3283 | ||
3284 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3285 | spec->image_name ? spec->image_name : "(unknown)"); | |
3286 | if (count < 0) | |
3287 | return count; | |
3288 | bufp += count; | |
3289 | ||
3290 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3291 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3292 | if (count < 0) | |
3293 | return count; | |
3294 | bufp += count; | |
3295 | ||
3296 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3297 | if (count < 0) | |
3298 | return count; | |
3299 | bufp += count; | |
3300 | ||
3301 | return (ssize_t) (bufp - buf); | |
3302 | } | |
3303 | ||
dfc5606d YS |
3304 | static ssize_t rbd_image_refresh(struct device *dev, |
3305 | struct device_attribute *attr, | |
3306 | const char *buf, | |
3307 | size_t size) | |
3308 | { | |
593a9e7b | 3309 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3310 | int ret; |
602adf40 | 3311 | |
117973fb | 3312 | ret = rbd_dev_refresh(rbd_dev, NULL); |
b813623a AE |
3313 | |
3314 | return ret < 0 ? ret : size; | |
dfc5606d | 3315 | } |
602adf40 | 3316 | |
dfc5606d | 3317 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3318 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
3319 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
3320 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
3321 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3322 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3323 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3324 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3325 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3326 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3327 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3328 | |
3329 | static struct attribute *rbd_attrs[] = { | |
3330 | &dev_attr_size.attr, | |
34b13184 | 3331 | &dev_attr_features.attr, |
dfc5606d YS |
3332 | &dev_attr_major.attr, |
3333 | &dev_attr_client_id.attr, | |
3334 | &dev_attr_pool.attr, | |
9bb2f334 | 3335 | &dev_attr_pool_id.attr, |
dfc5606d | 3336 | &dev_attr_name.attr, |
589d30e0 | 3337 | &dev_attr_image_id.attr, |
dfc5606d | 3338 | &dev_attr_current_snap.attr, |
86b00e0d | 3339 | &dev_attr_parent.attr, |
dfc5606d | 3340 | &dev_attr_refresh.attr, |
dfc5606d YS |
3341 | NULL |
3342 | }; | |
3343 | ||
3344 | static struct attribute_group rbd_attr_group = { | |
3345 | .attrs = rbd_attrs, | |
3346 | }; | |
3347 | ||
3348 | static const struct attribute_group *rbd_attr_groups[] = { | |
3349 | &rbd_attr_group, | |
3350 | NULL | |
3351 | }; | |
3352 | ||
3353 | static void rbd_sysfs_dev_release(struct device *dev) | |
3354 | { | |
3355 | } | |
3356 | ||
3357 | static struct device_type rbd_device_type = { | |
3358 | .name = "rbd", | |
3359 | .groups = rbd_attr_groups, | |
3360 | .release = rbd_sysfs_dev_release, | |
3361 | }; | |
3362 | ||
8b8fb99c AE |
3363 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3364 | { | |
3365 | kref_get(&spec->kref); | |
3366 | ||
3367 | return spec; | |
3368 | } | |
3369 | ||
3370 | static void rbd_spec_free(struct kref *kref); | |
3371 | static void rbd_spec_put(struct rbd_spec *spec) | |
3372 | { | |
3373 | if (spec) | |
3374 | kref_put(&spec->kref, rbd_spec_free); | |
3375 | } | |
3376 | ||
3377 | static struct rbd_spec *rbd_spec_alloc(void) | |
3378 | { | |
3379 | struct rbd_spec *spec; | |
3380 | ||
3381 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3382 | if (!spec) | |
3383 | return NULL; | |
3384 | kref_init(&spec->kref); | |
3385 | ||
8b8fb99c AE |
3386 | return spec; |
3387 | } | |
3388 | ||
3389 | static void rbd_spec_free(struct kref *kref) | |
3390 | { | |
3391 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3392 | ||
3393 | kfree(spec->pool_name); | |
3394 | kfree(spec->image_id); | |
3395 | kfree(spec->image_name); | |
3396 | kfree(spec->snap_name); | |
3397 | kfree(spec); | |
3398 | } | |
3399 | ||
cc344fa1 | 3400 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3401 | struct rbd_spec *spec) |
3402 | { | |
3403 | struct rbd_device *rbd_dev; | |
3404 | ||
3405 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3406 | if (!rbd_dev) | |
3407 | return NULL; | |
3408 | ||
3409 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3410 | rbd_dev->flags = 0; |
c53d5893 AE |
3411 | INIT_LIST_HEAD(&rbd_dev->node); |
3412 | INIT_LIST_HEAD(&rbd_dev->snaps); | |
3413 | init_rwsem(&rbd_dev->header_rwsem); | |
3414 | ||
3415 | rbd_dev->spec = spec; | |
3416 | rbd_dev->rbd_client = rbdc; | |
3417 | ||
0903e875 AE |
3418 | /* Initialize the layout used for all rbd requests */ |
3419 | ||
3420 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3421 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3422 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3423 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3424 | ||
c53d5893 AE |
3425 | return rbd_dev; |
3426 | } | |
3427 | ||
3428 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3429 | { | |
c53d5893 AE |
3430 | rbd_put_client(rbd_dev->rbd_client); |
3431 | rbd_spec_put(rbd_dev->spec); | |
3432 | kfree(rbd_dev); | |
3433 | } | |
3434 | ||
6087b51b | 3435 | static void rbd_snap_destroy(struct rbd_snap *snap) |
dfc5606d | 3436 | { |
3e83b65b AE |
3437 | kfree(snap->name); |
3438 | kfree(snap); | |
dfc5606d YS |
3439 | } |
3440 | ||
6087b51b | 3441 | static struct rbd_snap *rbd_snap_create(struct rbd_device *rbd_dev, |
c8d18425 | 3442 | const char *snap_name, |
34b13184 AE |
3443 | u64 snap_id, u64 snap_size, |
3444 | u64 snap_features) | |
dfc5606d | 3445 | { |
4e891e0a | 3446 | struct rbd_snap *snap; |
4e891e0a AE |
3447 | |
3448 | snap = kzalloc(sizeof (*snap), GFP_KERNEL); | |
dfc5606d | 3449 | if (!snap) |
4e891e0a AE |
3450 | return ERR_PTR(-ENOMEM); |
3451 | ||
6e584f52 | 3452 | snap->name = snap_name; |
c8d18425 AE |
3453 | snap->id = snap_id; |
3454 | snap->size = snap_size; | |
34b13184 | 3455 | snap->features = snap_features; |
4e891e0a AE |
3456 | |
3457 | return snap; | |
dfc5606d YS |
3458 | } |
3459 | ||
6e584f52 AE |
3460 | /* |
3461 | * Returns a dynamically-allocated snapshot name if successful, or a | |
3462 | * pointer-coded error otherwise. | |
3463 | */ | |
cd892126 AE |
3464 | static char *rbd_dev_v1_snap_info(struct rbd_device *rbd_dev, u32 which, |
3465 | u64 *snap_size, u64 *snap_features) | |
3466 | { | |
3467 | char *snap_name; | |
6e584f52 | 3468 | int i; |
cd892126 AE |
3469 | |
3470 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
3471 | ||
cd892126 AE |
3472 | /* Skip over names until we find the one we are looking for */ |
3473 | ||
3474 | snap_name = rbd_dev->header.snap_names; | |
6e584f52 | 3475 | for (i = 0; i < which; i++) |
cd892126 AE |
3476 | snap_name += strlen(snap_name) + 1; |
3477 | ||
6e584f52 AE |
3478 | snap_name = kstrdup(snap_name, GFP_KERNEL); |
3479 | if (!snap_name) | |
3480 | return ERR_PTR(-ENOMEM); | |
3481 | ||
3482 | *snap_size = rbd_dev->header.snap_sizes[which]; | |
3483 | *snap_features = 0; /* No features for v1 */ | |
3484 | ||
cd892126 AE |
3485 | return snap_name; |
3486 | } | |
3487 | ||
9d475de5 AE |
3488 | /* |
3489 | * Get the size and object order for an image snapshot, or if | |
3490 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3491 | * image. | |
3492 | */ | |
3493 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3494 | u8 *order, u64 *snap_size) | |
3495 | { | |
3496 | __le64 snapid = cpu_to_le64(snap_id); | |
3497 | int ret; | |
3498 | struct { | |
3499 | u8 order; | |
3500 | __le64 size; | |
3501 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3502 | ||
36be9a76 | 3503 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3504 | "rbd", "get_size", |
4157976b AE |
3505 | &snapid, sizeof (snapid), |
3506 | &size_buf, sizeof (size_buf), NULL); | |
36be9a76 | 3507 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3508 | if (ret < 0) |
3509 | return ret; | |
57385b51 AE |
3510 | if (ret < sizeof (size_buf)) |
3511 | return -ERANGE; | |
9d475de5 | 3512 | |
c86f86e9 AE |
3513 | if (order) |
3514 | *order = size_buf.order; | |
9d475de5 AE |
3515 | *snap_size = le64_to_cpu(size_buf.size); |
3516 | ||
3517 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
57385b51 AE |
3518 | (unsigned long long)snap_id, (unsigned int)*order, |
3519 | (unsigned long long)*snap_size); | |
9d475de5 AE |
3520 | |
3521 | return 0; | |
3522 | } | |
3523 | ||
3524 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3525 | { | |
3526 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3527 | &rbd_dev->header.obj_order, | |
3528 | &rbd_dev->header.image_size); | |
3529 | } | |
3530 | ||
1e130199 AE |
3531 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3532 | { | |
3533 | void *reply_buf; | |
3534 | int ret; | |
3535 | void *p; | |
3536 | ||
3537 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3538 | if (!reply_buf) | |
3539 | return -ENOMEM; | |
3540 | ||
36be9a76 | 3541 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3542 | "rbd", "get_object_prefix", NULL, 0, |
07b2391f | 3543 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX, NULL); |
36be9a76 | 3544 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3545 | if (ret < 0) |
3546 | goto out; | |
3547 | ||
3548 | p = reply_buf; | |
3549 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3550 | p + ret, NULL, GFP_NOIO); |
3551 | ret = 0; | |
1e130199 AE |
3552 | |
3553 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3554 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3555 | rbd_dev->header.object_prefix = NULL; | |
3556 | } else { | |
3557 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3558 | } | |
1e130199 AE |
3559 | out: |
3560 | kfree(reply_buf); | |
3561 | ||
3562 | return ret; | |
3563 | } | |
3564 | ||
b1b5402a AE |
3565 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3566 | u64 *snap_features) | |
3567 | { | |
3568 | __le64 snapid = cpu_to_le64(snap_id); | |
3569 | struct { | |
3570 | __le64 features; | |
3571 | __le64 incompat; | |
4157976b | 3572 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3573 | u64 incompat; |
b1b5402a AE |
3574 | int ret; |
3575 | ||
36be9a76 | 3576 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3577 | "rbd", "get_features", |
4157976b AE |
3578 | &snapid, sizeof (snapid), |
3579 | &features_buf, sizeof (features_buf), NULL); | |
36be9a76 | 3580 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3581 | if (ret < 0) |
3582 | return ret; | |
57385b51 AE |
3583 | if (ret < sizeof (features_buf)) |
3584 | return -ERANGE; | |
d889140c AE |
3585 | |
3586 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3587 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3588 | return -ENXIO; |
d889140c | 3589 | |
b1b5402a AE |
3590 | *snap_features = le64_to_cpu(features_buf.features); |
3591 | ||
3592 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3593 | (unsigned long long)snap_id, |
3594 | (unsigned long long)*snap_features, | |
3595 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3596 | |
3597 | return 0; | |
3598 | } | |
3599 | ||
3600 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3601 | { | |
3602 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3603 | &rbd_dev->header.features); | |
3604 | } | |
3605 | ||
86b00e0d AE |
3606 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3607 | { | |
3608 | struct rbd_spec *parent_spec; | |
3609 | size_t size; | |
3610 | void *reply_buf = NULL; | |
3611 | __le64 snapid; | |
3612 | void *p; | |
3613 | void *end; | |
3614 | char *image_id; | |
3615 | u64 overlap; | |
86b00e0d AE |
3616 | int ret; |
3617 | ||
3618 | parent_spec = rbd_spec_alloc(); | |
3619 | if (!parent_spec) | |
3620 | return -ENOMEM; | |
3621 | ||
3622 | size = sizeof (__le64) + /* pool_id */ | |
3623 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3624 | sizeof (__le64) + /* snap_id */ | |
3625 | sizeof (__le64); /* overlap */ | |
3626 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3627 | if (!reply_buf) { | |
3628 | ret = -ENOMEM; | |
3629 | goto out_err; | |
3630 | } | |
3631 | ||
3632 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3633 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3634 | "rbd", "get_parent", |
4157976b AE |
3635 | &snapid, sizeof (snapid), |
3636 | reply_buf, size, NULL); | |
36be9a76 | 3637 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3638 | if (ret < 0) |
3639 | goto out_err; | |
3640 | ||
86b00e0d | 3641 | p = reply_buf; |
57385b51 AE |
3642 | end = reply_buf + ret; |
3643 | ret = -ERANGE; | |
86b00e0d AE |
3644 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); |
3645 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
3646 | goto out; /* No parent? No problem. */ | |
3647 | ||
0903e875 AE |
3648 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3649 | ||
3650 | ret = -EIO; | |
c0cd10db AE |
3651 | if (parent_spec->pool_id > (u64)U32_MAX) { |
3652 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", | |
3653 | (unsigned long long)parent_spec->pool_id, U32_MAX); | |
57385b51 | 3654 | goto out_err; |
c0cd10db | 3655 | } |
0903e875 | 3656 | |
979ed480 | 3657 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3658 | if (IS_ERR(image_id)) { |
3659 | ret = PTR_ERR(image_id); | |
3660 | goto out_err; | |
3661 | } | |
3662 | parent_spec->image_id = image_id; | |
3663 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
3664 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
3665 | ||
3666 | rbd_dev->parent_overlap = overlap; | |
3667 | rbd_dev->parent_spec = parent_spec; | |
3668 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3669 | out: | |
3670 | ret = 0; | |
3671 | out_err: | |
3672 | kfree(reply_buf); | |
3673 | rbd_spec_put(parent_spec); | |
3674 | ||
3675 | return ret; | |
3676 | } | |
3677 | ||
cc070d59 AE |
3678 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3679 | { | |
3680 | struct { | |
3681 | __le64 stripe_unit; | |
3682 | __le64 stripe_count; | |
3683 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3684 | size_t size = sizeof (striping_info_buf); | |
3685 | void *p; | |
3686 | u64 obj_size; | |
3687 | u64 stripe_unit; | |
3688 | u64 stripe_count; | |
3689 | int ret; | |
3690 | ||
3691 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3692 | "rbd", "get_stripe_unit_count", NULL, 0, | |
3693 | (char *)&striping_info_buf, size, NULL); | |
3694 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); | |
3695 | if (ret < 0) | |
3696 | return ret; | |
3697 | if (ret < size) | |
3698 | return -ERANGE; | |
3699 | ||
3700 | /* | |
3701 | * We don't actually support the "fancy striping" feature | |
3702 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3703 | * defaults the behavior is the same as before. So find | |
3704 | * out, and only fail if the image has non-default values. | |
3705 | */ | |
3706 | ret = -EINVAL; | |
3707 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3708 | p = &striping_info_buf; | |
3709 | stripe_unit = ceph_decode_64(&p); | |
3710 | if (stripe_unit != obj_size) { | |
3711 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
3712 | "(got %llu want %llu)", | |
3713 | stripe_unit, obj_size); | |
3714 | return -EINVAL; | |
3715 | } | |
3716 | stripe_count = ceph_decode_64(&p); | |
3717 | if (stripe_count != 1) { | |
3718 | rbd_warn(rbd_dev, "unsupported stripe count " | |
3719 | "(got %llu want 1)", stripe_count); | |
3720 | return -EINVAL; | |
3721 | } | |
500d0c0f AE |
3722 | rbd_dev->header.stripe_unit = stripe_unit; |
3723 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
3724 | |
3725 | return 0; | |
3726 | } | |
3727 | ||
9e15b77d AE |
3728 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3729 | { | |
3730 | size_t image_id_size; | |
3731 | char *image_id; | |
3732 | void *p; | |
3733 | void *end; | |
3734 | size_t size; | |
3735 | void *reply_buf = NULL; | |
3736 | size_t len = 0; | |
3737 | char *image_name = NULL; | |
3738 | int ret; | |
3739 | ||
3740 | rbd_assert(!rbd_dev->spec->image_name); | |
3741 | ||
69e7a02f AE |
3742 | len = strlen(rbd_dev->spec->image_id); |
3743 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3744 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3745 | if (!image_id) | |
3746 | return NULL; | |
3747 | ||
3748 | p = image_id; | |
4157976b | 3749 | end = image_id + image_id_size; |
57385b51 | 3750 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
3751 | |
3752 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3753 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3754 | if (!reply_buf) | |
3755 | goto out; | |
3756 | ||
36be9a76 | 3757 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3758 | "rbd", "dir_get_name", |
3759 | image_id, image_id_size, | |
4157976b | 3760 | reply_buf, size, NULL); |
9e15b77d AE |
3761 | if (ret < 0) |
3762 | goto out; | |
3763 | p = reply_buf; | |
f40eb349 AE |
3764 | end = reply_buf + ret; |
3765 | ||
9e15b77d AE |
3766 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
3767 | if (IS_ERR(image_name)) | |
3768 | image_name = NULL; | |
3769 | else | |
3770 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3771 | out: | |
3772 | kfree(reply_buf); | |
3773 | kfree(image_id); | |
3774 | ||
3775 | return image_name; | |
3776 | } | |
3777 | ||
3778 | /* | |
2e9f7f1c AE |
3779 | * When an rbd image has a parent image, it is identified by the |
3780 | * pool, image, and snapshot ids (not names). This function fills | |
3781 | * in the names for those ids. (It's OK if we can't figure out the | |
3782 | * name for an image id, but the pool and snapshot ids should always | |
3783 | * exist and have names.) All names in an rbd spec are dynamically | |
3784 | * allocated. | |
e1d4213f AE |
3785 | * |
3786 | * When an image being mapped (not a parent) is probed, we have the | |
3787 | * pool name and pool id, image name and image id, and the snapshot | |
3788 | * name. The only thing we're missing is the snapshot id. | |
2e9f7f1c AE |
3789 | * |
3790 | * The set of snapshots for an image is not known until they have | |
3791 | * been read by rbd_dev_snaps_update(), so we can't completely fill | |
3792 | * in this information until after that has been called. | |
9e15b77d | 3793 | */ |
2e9f7f1c | 3794 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 3795 | { |
2e9f7f1c AE |
3796 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3797 | struct rbd_spec *spec = rbd_dev->spec; | |
3798 | const char *pool_name; | |
3799 | const char *image_name; | |
3800 | const char *snap_name; | |
9e15b77d AE |
3801 | int ret; |
3802 | ||
e1d4213f AE |
3803 | /* |
3804 | * An image being mapped will have the pool name (etc.), but | |
3805 | * we need to look up the snapshot id. | |
3806 | */ | |
2e9f7f1c AE |
3807 | if (spec->pool_name) { |
3808 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
e1d4213f AE |
3809 | struct rbd_snap *snap; |
3810 | ||
2e9f7f1c | 3811 | snap = snap_by_name(rbd_dev, spec->snap_name); |
e1d4213f AE |
3812 | if (!snap) |
3813 | return -ENOENT; | |
2e9f7f1c | 3814 | spec->snap_id = snap->id; |
e1d4213f | 3815 | } else { |
2e9f7f1c | 3816 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
3817 | } |
3818 | ||
3819 | return 0; | |
3820 | } | |
9e15b77d | 3821 | |
2e9f7f1c | 3822 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 3823 | |
2e9f7f1c AE |
3824 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
3825 | if (!pool_name) { | |
3826 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
3827 | return -EIO; |
3828 | } | |
2e9f7f1c AE |
3829 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
3830 | if (!pool_name) | |
9e15b77d AE |
3831 | return -ENOMEM; |
3832 | ||
3833 | /* Fetch the image name; tolerate failure here */ | |
3834 | ||
2e9f7f1c AE |
3835 | image_name = rbd_dev_image_name(rbd_dev); |
3836 | if (!image_name) | |
06ecc6cb | 3837 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 3838 | |
2e9f7f1c | 3839 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 3840 | |
2e9f7f1c AE |
3841 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
3842 | if (!snap_name) { | |
3843 | rbd_warn(rbd_dev, "no snapshot with id %llu", spec->snap_id); | |
9e15b77d AE |
3844 | ret = -EIO; |
3845 | goto out_err; | |
3846 | } | |
2e9f7f1c AE |
3847 | snap_name = kstrdup(snap_name, GFP_KERNEL); |
3848 | if (!snap_name) { | |
3849 | ret = -ENOMEM; | |
9e15b77d | 3850 | goto out_err; |
2e9f7f1c AE |
3851 | } |
3852 | ||
3853 | spec->pool_name = pool_name; | |
3854 | spec->image_name = image_name; | |
3855 | spec->snap_name = snap_name; | |
9e15b77d AE |
3856 | |
3857 | return 0; | |
3858 | out_err: | |
2e9f7f1c AE |
3859 | kfree(image_name); |
3860 | kfree(pool_name); | |
9e15b77d AE |
3861 | |
3862 | return ret; | |
3863 | } | |
3864 | ||
6e14b1a6 | 3865 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev, u64 *ver) |
35d489f9 AE |
3866 | { |
3867 | size_t size; | |
3868 | int ret; | |
3869 | void *reply_buf; | |
3870 | void *p; | |
3871 | void *end; | |
3872 | u64 seq; | |
3873 | u32 snap_count; | |
3874 | struct ceph_snap_context *snapc; | |
3875 | u32 i; | |
3876 | ||
3877 | /* | |
3878 | * We'll need room for the seq value (maximum snapshot id), | |
3879 | * snapshot count, and array of that many snapshot ids. | |
3880 | * For now we have a fixed upper limit on the number we're | |
3881 | * prepared to receive. | |
3882 | */ | |
3883 | size = sizeof (__le64) + sizeof (__le32) + | |
3884 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3885 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3886 | if (!reply_buf) | |
3887 | return -ENOMEM; | |
3888 | ||
36be9a76 | 3889 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3890 | "rbd", "get_snapcontext", NULL, 0, |
07b2391f | 3891 | reply_buf, size, ver); |
36be9a76 | 3892 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3893 | if (ret < 0) |
3894 | goto out; | |
3895 | ||
35d489f9 | 3896 | p = reply_buf; |
57385b51 AE |
3897 | end = reply_buf + ret; |
3898 | ret = -ERANGE; | |
35d489f9 AE |
3899 | ceph_decode_64_safe(&p, end, seq, out); |
3900 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3901 | ||
3902 | /* | |
3903 | * Make sure the reported number of snapshot ids wouldn't go | |
3904 | * beyond the end of our buffer. But before checking that, | |
3905 | * make sure the computed size of the snapshot context we | |
3906 | * allocate is representable in a size_t. | |
3907 | */ | |
3908 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3909 | / sizeof (u64)) { | |
3910 | ret = -EINVAL; | |
3911 | goto out; | |
3912 | } | |
3913 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3914 | goto out; | |
468521c1 | 3915 | ret = 0; |
35d489f9 | 3916 | |
468521c1 | 3917 | snapc = rbd_snap_context_create(snap_count); |
35d489f9 AE |
3918 | if (!snapc) { |
3919 | ret = -ENOMEM; | |
3920 | goto out; | |
3921 | } | |
35d489f9 | 3922 | snapc->seq = seq; |
35d489f9 AE |
3923 | for (i = 0; i < snap_count; i++) |
3924 | snapc->snaps[i] = ceph_decode_64(&p); | |
3925 | ||
3926 | rbd_dev->header.snapc = snapc; | |
3927 | ||
3928 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 3929 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
3930 | out: |
3931 | kfree(reply_buf); | |
3932 | ||
57385b51 | 3933 | return ret; |
35d489f9 AE |
3934 | } |
3935 | ||
b8b1e2db AE |
3936 | static char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, u32 which) |
3937 | { | |
3938 | size_t size; | |
3939 | void *reply_buf; | |
3940 | __le64 snap_id; | |
3941 | int ret; | |
3942 | void *p; | |
3943 | void *end; | |
b8b1e2db AE |
3944 | char *snap_name; |
3945 | ||
3946 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
3947 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3948 | if (!reply_buf) | |
3949 | return ERR_PTR(-ENOMEM); | |
3950 | ||
acb1b6ca | 3951 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); |
b8b1e2db | 3952 | snap_id = cpu_to_le64(rbd_dev->header.snapc->snaps[which]); |
36be9a76 | 3953 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 3954 | "rbd", "get_snapshot_name", |
4157976b | 3955 | &snap_id, sizeof (snap_id), |
07b2391f | 3956 | reply_buf, size, NULL); |
36be9a76 | 3957 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
3958 | if (ret < 0) { |
3959 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 3960 | goto out; |
f40eb349 | 3961 | } |
b8b1e2db AE |
3962 | |
3963 | p = reply_buf; | |
f40eb349 | 3964 | end = reply_buf + ret; |
e5c35534 | 3965 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 3966 | if (IS_ERR(snap_name)) |
b8b1e2db | 3967 | goto out; |
b8b1e2db | 3968 | |
f40eb349 AE |
3969 | dout(" snap_id 0x%016llx snap_name = %s\n", |
3970 | (unsigned long long)le64_to_cpu(snap_id), snap_name); | |
b8b1e2db AE |
3971 | out: |
3972 | kfree(reply_buf); | |
3973 | ||
f40eb349 | 3974 | return snap_name; |
b8b1e2db AE |
3975 | } |
3976 | ||
3977 | static char *rbd_dev_v2_snap_info(struct rbd_device *rbd_dev, u32 which, | |
3978 | u64 *snap_size, u64 *snap_features) | |
3979 | { | |
e0b49868 | 3980 | u64 snap_id; |
acb1b6ca AE |
3981 | u64 size; |
3982 | u64 features; | |
3983 | char *snap_name; | |
b8b1e2db AE |
3984 | int ret; |
3985 | ||
acb1b6ca | 3986 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); |
b8b1e2db | 3987 | snap_id = rbd_dev->header.snapc->snaps[which]; |
acb1b6ca | 3988 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); |
b8b1e2db | 3989 | if (ret) |
acb1b6ca AE |
3990 | goto out_err; |
3991 | ||
3992 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); | |
b8b1e2db | 3993 | if (ret) |
acb1b6ca AE |
3994 | goto out_err; |
3995 | ||
3996 | snap_name = rbd_dev_v2_snap_name(rbd_dev, which); | |
3997 | if (!IS_ERR(snap_name)) { | |
3998 | *snap_size = size; | |
3999 | *snap_features = features; | |
4000 | } | |
b8b1e2db | 4001 | |
acb1b6ca AE |
4002 | return snap_name; |
4003 | out_err: | |
4004 | return ERR_PTR(ret); | |
b8b1e2db AE |
4005 | } |
4006 | ||
4007 | static char *rbd_dev_snap_info(struct rbd_device *rbd_dev, u32 which, | |
4008 | u64 *snap_size, u64 *snap_features) | |
4009 | { | |
4010 | if (rbd_dev->image_format == 1) | |
4011 | return rbd_dev_v1_snap_info(rbd_dev, which, | |
4012 | snap_size, snap_features); | |
4013 | if (rbd_dev->image_format == 2) | |
4014 | return rbd_dev_v2_snap_info(rbd_dev, which, | |
4015 | snap_size, snap_features); | |
4016 | return ERR_PTR(-EINVAL); | |
4017 | } | |
4018 | ||
117973fb AE |
4019 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev, u64 *hver) |
4020 | { | |
4021 | int ret; | |
117973fb AE |
4022 | |
4023 | down_write(&rbd_dev->header_rwsem); | |
4024 | ||
117973fb AE |
4025 | ret = rbd_dev_v2_image_size(rbd_dev); |
4026 | if (ret) | |
4027 | goto out; | |
117973fb AE |
4028 | rbd_update_mapping_size(rbd_dev); |
4029 | ||
4030 | ret = rbd_dev_v2_snap_context(rbd_dev, hver); | |
4031 | dout("rbd_dev_v2_snap_context returned %d\n", ret); | |
4032 | if (ret) | |
4033 | goto out; | |
4034 | ret = rbd_dev_snaps_update(rbd_dev); | |
4035 | dout("rbd_dev_snaps_update returned %d\n", ret); | |
4036 | if (ret) | |
4037 | goto out; | |
117973fb AE |
4038 | out: |
4039 | up_write(&rbd_dev->header_rwsem); | |
4040 | ||
4041 | return ret; | |
4042 | } | |
4043 | ||
dfc5606d | 4044 | /* |
35938150 AE |
4045 | * Scan the rbd device's current snapshot list and compare it to the |
4046 | * newly-received snapshot context. Remove any existing snapshots | |
4047 | * not present in the new snapshot context. Add a new snapshot for | |
4048 | * any snaphots in the snapshot context not in the current list. | |
4049 | * And verify there are no changes to snapshots we already know | |
4050 | * about. | |
4051 | * | |
4052 | * Assumes the snapshots in the snapshot context are sorted by | |
4053 | * snapshot id, highest id first. (Snapshots in the rbd_dev's list | |
4054 | * are also maintained in that order.) | |
522a0cc0 AE |
4055 | * |
4056 | * Note that any error occurs while updating the snapshot list | |
4057 | * aborts the update, and the entire list is cleared. The snapshot | |
4058 | * list becomes inconsistent at that point anyway, so it might as | |
4059 | * well be empty. | |
dfc5606d | 4060 | */ |
304f6808 | 4061 | static int rbd_dev_snaps_update(struct rbd_device *rbd_dev) |
dfc5606d | 4062 | { |
35938150 AE |
4063 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; |
4064 | const u32 snap_count = snapc->num_snaps; | |
35938150 AE |
4065 | struct list_head *head = &rbd_dev->snaps; |
4066 | struct list_head *links = head->next; | |
4067 | u32 index = 0; | |
522a0cc0 | 4068 | int ret = 0; |
dfc5606d | 4069 | |
522a0cc0 | 4070 | dout("%s: snap count is %u\n", __func__, (unsigned int)snap_count); |
35938150 AE |
4071 | while (index < snap_count || links != head) { |
4072 | u64 snap_id; | |
4073 | struct rbd_snap *snap; | |
cd892126 AE |
4074 | char *snap_name; |
4075 | u64 snap_size = 0; | |
4076 | u64 snap_features = 0; | |
dfc5606d | 4077 | |
35938150 AE |
4078 | snap_id = index < snap_count ? snapc->snaps[index] |
4079 | : CEPH_NOSNAP; | |
4080 | snap = links != head ? list_entry(links, struct rbd_snap, node) | |
4081 | : NULL; | |
aafb230e | 4082 | rbd_assert(!snap || snap->id != CEPH_NOSNAP); |
dfc5606d | 4083 | |
35938150 AE |
4084 | if (snap_id == CEPH_NOSNAP || (snap && snap->id > snap_id)) { |
4085 | struct list_head *next = links->next; | |
dfc5606d | 4086 | |
6d292906 AE |
4087 | /* |
4088 | * A previously-existing snapshot is not in | |
4089 | * the new snap context. | |
4090 | * | |
522a0cc0 AE |
4091 | * If the now-missing snapshot is the one |
4092 | * the image represents, clear its existence | |
4093 | * flag so we can avoid sending any more | |
4094 | * requests to it. | |
6d292906 | 4095 | */ |
0d7dbfce | 4096 | if (rbd_dev->spec->snap_id == snap->id) |
6d292906 | 4097 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
3e83b65b | 4098 | dout("removing %ssnap id %llu\n", |
0d7dbfce AE |
4099 | rbd_dev->spec->snap_id == snap->id ? |
4100 | "mapped " : "", | |
522a0cc0 | 4101 | (unsigned long long)snap->id); |
6087b51b AE |
4102 | |
4103 | list_del(&snap->node); | |
4104 | rbd_snap_destroy(snap); | |
35938150 AE |
4105 | |
4106 | /* Done with this list entry; advance */ | |
4107 | ||
4108 | links = next; | |
dfc5606d YS |
4109 | continue; |
4110 | } | |
35938150 | 4111 | |
b8b1e2db AE |
4112 | snap_name = rbd_dev_snap_info(rbd_dev, index, |
4113 | &snap_size, &snap_features); | |
522a0cc0 AE |
4114 | if (IS_ERR(snap_name)) { |
4115 | ret = PTR_ERR(snap_name); | |
4116 | dout("failed to get snap info, error %d\n", ret); | |
4117 | goto out_err; | |
4118 | } | |
cd892126 | 4119 | |
522a0cc0 AE |
4120 | dout("entry %u: snap_id = %llu\n", (unsigned int)snap_count, |
4121 | (unsigned long long)snap_id); | |
35938150 AE |
4122 | if (!snap || (snap_id != CEPH_NOSNAP && snap->id < snap_id)) { |
4123 | struct rbd_snap *new_snap; | |
4124 | ||
4125 | /* We haven't seen this snapshot before */ | |
4126 | ||
6087b51b | 4127 | new_snap = rbd_snap_create(rbd_dev, snap_name, |
cd892126 | 4128 | snap_id, snap_size, snap_features); |
9fcbb800 | 4129 | if (IS_ERR(new_snap)) { |
522a0cc0 AE |
4130 | ret = PTR_ERR(new_snap); |
4131 | dout(" failed to add dev, error %d\n", ret); | |
4132 | goto out_err; | |
9fcbb800 | 4133 | } |
35938150 AE |
4134 | |
4135 | /* New goes before existing, or at end of list */ | |
4136 | ||
9fcbb800 | 4137 | dout(" added dev%s\n", snap ? "" : " at end\n"); |
35938150 AE |
4138 | if (snap) |
4139 | list_add_tail(&new_snap->node, &snap->node); | |
4140 | else | |
523f3258 | 4141 | list_add_tail(&new_snap->node, head); |
35938150 AE |
4142 | } else { |
4143 | /* Already have this one */ | |
4144 | ||
9fcbb800 AE |
4145 | dout(" already present\n"); |
4146 | ||
cd892126 | 4147 | rbd_assert(snap->size == snap_size); |
aafb230e | 4148 | rbd_assert(!strcmp(snap->name, snap_name)); |
cd892126 | 4149 | rbd_assert(snap->features == snap_features); |
35938150 AE |
4150 | |
4151 | /* Done with this list entry; advance */ | |
4152 | ||
4153 | links = links->next; | |
dfc5606d | 4154 | } |
35938150 AE |
4155 | |
4156 | /* Advance to the next entry in the snapshot context */ | |
4157 | ||
4158 | index++; | |
dfc5606d | 4159 | } |
9fcbb800 | 4160 | dout("%s: done\n", __func__); |
dfc5606d YS |
4161 | |
4162 | return 0; | |
522a0cc0 AE |
4163 | out_err: |
4164 | rbd_remove_all_snaps(rbd_dev); | |
4165 | ||
4166 | return ret; | |
dfc5606d YS |
4167 | } |
4168 | ||
dfc5606d YS |
4169 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4170 | { | |
dfc5606d | 4171 | struct device *dev; |
cd789ab9 | 4172 | int ret; |
dfc5606d YS |
4173 | |
4174 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 4175 | |
cd789ab9 | 4176 | dev = &rbd_dev->dev; |
dfc5606d YS |
4177 | dev->bus = &rbd_bus_type; |
4178 | dev->type = &rbd_device_type; | |
4179 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4180 | dev->release = rbd_dev_device_release; |
de71a297 | 4181 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4182 | ret = device_register(dev); |
dfc5606d | 4183 | |
dfc5606d | 4184 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 4185 | |
dfc5606d | 4186 | return ret; |
602adf40 YS |
4187 | } |
4188 | ||
dfc5606d YS |
4189 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4190 | { | |
4191 | device_unregister(&rbd_dev->dev); | |
4192 | } | |
4193 | ||
e2839308 | 4194 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
4195 | |
4196 | /* | |
499afd5b AE |
4197 | * Get a unique rbd identifier for the given new rbd_dev, and add |
4198 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 4199 | */ |
e2839308 | 4200 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4201 | { |
e2839308 | 4202 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
4203 | |
4204 | spin_lock(&rbd_dev_list_lock); | |
4205 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4206 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
4207 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
4208 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 4209 | } |
b7f23c36 | 4210 | |
1ddbe94e | 4211 | /* |
499afd5b AE |
4212 | * Remove an rbd_dev from the global list, and record that its |
4213 | * identifier is no longer in use. | |
1ddbe94e | 4214 | */ |
e2839308 | 4215 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4216 | { |
d184f6bf | 4217 | struct list_head *tmp; |
de71a297 | 4218 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
4219 | int max_id; |
4220 | ||
aafb230e | 4221 | rbd_assert(rbd_id > 0); |
499afd5b | 4222 | |
e2839308 AE |
4223 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
4224 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
4225 | spin_lock(&rbd_dev_list_lock); |
4226 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
4227 | |
4228 | /* | |
4229 | * If the id being "put" is not the current maximum, there | |
4230 | * is nothing special we need to do. | |
4231 | */ | |
e2839308 | 4232 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
4233 | spin_unlock(&rbd_dev_list_lock); |
4234 | return; | |
4235 | } | |
4236 | ||
4237 | /* | |
4238 | * We need to update the current maximum id. Search the | |
4239 | * list to find out what it is. We're more likely to find | |
4240 | * the maximum at the end, so search the list backward. | |
4241 | */ | |
4242 | max_id = 0; | |
4243 | list_for_each_prev(tmp, &rbd_dev_list) { | |
4244 | struct rbd_device *rbd_dev; | |
4245 | ||
4246 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
4247 | if (rbd_dev->dev_id > max_id) |
4248 | max_id = rbd_dev->dev_id; | |
d184f6bf | 4249 | } |
499afd5b | 4250 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 4251 | |
1ddbe94e | 4252 | /* |
e2839308 | 4253 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
4254 | * which case it now accurately reflects the new maximum. |
4255 | * Be careful not to overwrite the maximum value in that | |
4256 | * case. | |
1ddbe94e | 4257 | */ |
e2839308 AE |
4258 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
4259 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
4260 | } |
4261 | ||
e28fff26 AE |
4262 | /* |
4263 | * Skips over white space at *buf, and updates *buf to point to the | |
4264 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4265 | * the token (string of non-white space characters) found. Note |
4266 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4267 | */ |
4268 | static inline size_t next_token(const char **buf) | |
4269 | { | |
4270 | /* | |
4271 | * These are the characters that produce nonzero for | |
4272 | * isspace() in the "C" and "POSIX" locales. | |
4273 | */ | |
4274 | const char *spaces = " \f\n\r\t\v"; | |
4275 | ||
4276 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4277 | ||
4278 | return strcspn(*buf, spaces); /* Return token length */ | |
4279 | } | |
4280 | ||
4281 | /* | |
4282 | * Finds the next token in *buf, and if the provided token buffer is | |
4283 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4284 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4285 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4286 | * |
4287 | * Returns the length of the token found (not including the '\0'). | |
4288 | * Return value will be 0 if no token is found, and it will be >= | |
4289 | * token_size if the token would not fit. | |
4290 | * | |
593a9e7b | 4291 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4292 | * found token. Note that this occurs even if the token buffer is |
4293 | * too small to hold it. | |
4294 | */ | |
4295 | static inline size_t copy_token(const char **buf, | |
4296 | char *token, | |
4297 | size_t token_size) | |
4298 | { | |
4299 | size_t len; | |
4300 | ||
4301 | len = next_token(buf); | |
4302 | if (len < token_size) { | |
4303 | memcpy(token, *buf, len); | |
4304 | *(token + len) = '\0'; | |
4305 | } | |
4306 | *buf += len; | |
4307 | ||
4308 | return len; | |
4309 | } | |
4310 | ||
ea3352f4 AE |
4311 | /* |
4312 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4313 | * enough to hold a copy of it, and copies the token into the new | |
4314 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4315 | * that a duplicate buffer is created even for a zero-length token. | |
4316 | * | |
4317 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4318 | * pointer if memory for the duplicate was not available. If | |
4319 | * the lenp argument is a non-null pointer, the length of the token | |
4320 | * (not including the '\0') is returned in *lenp. | |
4321 | * | |
4322 | * If successful, the *buf pointer will be updated to point beyond | |
4323 | * the end of the found token. | |
4324 | * | |
4325 | * Note: uses GFP_KERNEL for allocation. | |
4326 | */ | |
4327 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4328 | { | |
4329 | char *dup; | |
4330 | size_t len; | |
4331 | ||
4332 | len = next_token(buf); | |
4caf35f9 | 4333 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4334 | if (!dup) |
4335 | return NULL; | |
ea3352f4 AE |
4336 | *(dup + len) = '\0'; |
4337 | *buf += len; | |
4338 | ||
4339 | if (lenp) | |
4340 | *lenp = len; | |
4341 | ||
4342 | return dup; | |
4343 | } | |
4344 | ||
a725f65e | 4345 | /* |
859c31df AE |
4346 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4347 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4348 | * and the data written is passed here via a NUL-terminated buffer. | |
4349 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4350 | * |
859c31df AE |
4351 | * The information extracted from these options is recorded in |
4352 | * the other parameters which return dynamically-allocated | |
4353 | * structures: | |
4354 | * ceph_opts | |
4355 | * The address of a pointer that will refer to a ceph options | |
4356 | * structure. Caller must release the returned pointer using | |
4357 | * ceph_destroy_options() when it is no longer needed. | |
4358 | * rbd_opts | |
4359 | * Address of an rbd options pointer. Fully initialized by | |
4360 | * this function; caller must release with kfree(). | |
4361 | * spec | |
4362 | * Address of an rbd image specification pointer. Fully | |
4363 | * initialized by this function based on parsed options. | |
4364 | * Caller must release with rbd_spec_put(). | |
4365 | * | |
4366 | * The options passed take this form: | |
4367 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4368 | * where: | |
4369 | * <mon_addrs> | |
4370 | * A comma-separated list of one or more monitor addresses. | |
4371 | * A monitor address is an ip address, optionally followed | |
4372 | * by a port number (separated by a colon). | |
4373 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4374 | * <options> | |
4375 | * A comma-separated list of ceph and/or rbd options. | |
4376 | * <pool_name> | |
4377 | * The name of the rados pool containing the rbd image. | |
4378 | * <image_name> | |
4379 | * The name of the image in that pool to map. | |
4380 | * <snap_id> | |
4381 | * An optional snapshot id. If provided, the mapping will | |
4382 | * present data from the image at the time that snapshot was | |
4383 | * created. The image head is used if no snapshot id is | |
4384 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4385 | */ |
859c31df | 4386 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4387 | struct ceph_options **ceph_opts, |
859c31df AE |
4388 | struct rbd_options **opts, |
4389 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4390 | { |
d22f76e7 | 4391 | size_t len; |
859c31df | 4392 | char *options; |
0ddebc0c | 4393 | const char *mon_addrs; |
ecb4dc22 | 4394 | char *snap_name; |
0ddebc0c | 4395 | size_t mon_addrs_size; |
859c31df | 4396 | struct rbd_spec *spec = NULL; |
4e9afeba | 4397 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4398 | struct ceph_options *copts; |
dc79b113 | 4399 | int ret; |
e28fff26 AE |
4400 | |
4401 | /* The first four tokens are required */ | |
4402 | ||
7ef3214a | 4403 | len = next_token(&buf); |
4fb5d671 AE |
4404 | if (!len) { |
4405 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4406 | return -EINVAL; | |
4407 | } | |
0ddebc0c | 4408 | mon_addrs = buf; |
f28e565a | 4409 | mon_addrs_size = len + 1; |
7ef3214a | 4410 | buf += len; |
a725f65e | 4411 | |
dc79b113 | 4412 | ret = -EINVAL; |
f28e565a AE |
4413 | options = dup_token(&buf, NULL); |
4414 | if (!options) | |
dc79b113 | 4415 | return -ENOMEM; |
4fb5d671 AE |
4416 | if (!*options) { |
4417 | rbd_warn(NULL, "no options provided"); | |
4418 | goto out_err; | |
4419 | } | |
e28fff26 | 4420 | |
859c31df AE |
4421 | spec = rbd_spec_alloc(); |
4422 | if (!spec) | |
f28e565a | 4423 | goto out_mem; |
859c31df AE |
4424 | |
4425 | spec->pool_name = dup_token(&buf, NULL); | |
4426 | if (!spec->pool_name) | |
4427 | goto out_mem; | |
4fb5d671 AE |
4428 | if (!*spec->pool_name) { |
4429 | rbd_warn(NULL, "no pool name provided"); | |
4430 | goto out_err; | |
4431 | } | |
e28fff26 | 4432 | |
69e7a02f | 4433 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4434 | if (!spec->image_name) |
f28e565a | 4435 | goto out_mem; |
4fb5d671 AE |
4436 | if (!*spec->image_name) { |
4437 | rbd_warn(NULL, "no image name provided"); | |
4438 | goto out_err; | |
4439 | } | |
d4b125e9 | 4440 | |
f28e565a AE |
4441 | /* |
4442 | * Snapshot name is optional; default is to use "-" | |
4443 | * (indicating the head/no snapshot). | |
4444 | */ | |
3feeb894 | 4445 | len = next_token(&buf); |
820a5f3e | 4446 | if (!len) { |
3feeb894 AE |
4447 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4448 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4449 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4450 | ret = -ENAMETOOLONG; |
f28e565a | 4451 | goto out_err; |
849b4260 | 4452 | } |
ecb4dc22 AE |
4453 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4454 | if (!snap_name) | |
f28e565a | 4455 | goto out_mem; |
ecb4dc22 AE |
4456 | *(snap_name + len) = '\0'; |
4457 | spec->snap_name = snap_name; | |
e5c35534 | 4458 | |
0ddebc0c | 4459 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4460 | |
4e9afeba AE |
4461 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4462 | if (!rbd_opts) | |
4463 | goto out_mem; | |
4464 | ||
4465 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4466 | |
859c31df | 4467 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4468 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4469 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4470 | if (IS_ERR(copts)) { |
4471 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4472 | goto out_err; |
4473 | } | |
859c31df AE |
4474 | kfree(options); |
4475 | ||
4476 | *ceph_opts = copts; | |
4e9afeba | 4477 | *opts = rbd_opts; |
859c31df | 4478 | *rbd_spec = spec; |
0ddebc0c | 4479 | |
dc79b113 | 4480 | return 0; |
f28e565a | 4481 | out_mem: |
dc79b113 | 4482 | ret = -ENOMEM; |
d22f76e7 | 4483 | out_err: |
859c31df AE |
4484 | kfree(rbd_opts); |
4485 | rbd_spec_put(spec); | |
f28e565a | 4486 | kfree(options); |
d22f76e7 | 4487 | |
dc79b113 | 4488 | return ret; |
a725f65e AE |
4489 | } |
4490 | ||
589d30e0 AE |
4491 | /* |
4492 | * An rbd format 2 image has a unique identifier, distinct from the | |
4493 | * name given to it by the user. Internally, that identifier is | |
4494 | * what's used to specify the names of objects related to the image. | |
4495 | * | |
4496 | * A special "rbd id" object is used to map an rbd image name to its | |
4497 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4498 | * with the supplied name. | |
4499 | * | |
4500 | * This function will record the given rbd_dev's image_id field if | |
4501 | * it can be determined, and in that case will return 0. If any | |
4502 | * errors occur a negative errno will be returned and the rbd_dev's | |
4503 | * image_id field will be unchanged (and should be NULL). | |
4504 | */ | |
4505 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4506 | { | |
4507 | int ret; | |
4508 | size_t size; | |
4509 | char *object_name; | |
4510 | void *response; | |
c0fba368 | 4511 | char *image_id; |
2f82ee54 | 4512 | |
2c0d0a10 AE |
4513 | /* |
4514 | * When probing a parent image, the image id is already | |
4515 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4516 | * need to fetch the image id again in this case. We |
4517 | * do still need to set the image format though. | |
2c0d0a10 | 4518 | */ |
c0fba368 AE |
4519 | if (rbd_dev->spec->image_id) { |
4520 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4521 | ||
2c0d0a10 | 4522 | return 0; |
c0fba368 | 4523 | } |
2c0d0a10 | 4524 | |
589d30e0 AE |
4525 | /* |
4526 | * First, see if the format 2 image id file exists, and if | |
4527 | * so, get the image's persistent id from it. | |
4528 | */ | |
69e7a02f | 4529 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4530 | object_name = kmalloc(size, GFP_NOIO); |
4531 | if (!object_name) | |
4532 | return -ENOMEM; | |
0d7dbfce | 4533 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4534 | dout("rbd id object name is %s\n", object_name); |
4535 | ||
4536 | /* Response will be an encoded string, which includes a length */ | |
4537 | ||
4538 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4539 | response = kzalloc(size, GFP_NOIO); | |
4540 | if (!response) { | |
4541 | ret = -ENOMEM; | |
4542 | goto out; | |
4543 | } | |
4544 | ||
c0fba368 AE |
4545 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4546 | ||
36be9a76 | 4547 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4548 | "rbd", "get_id", NULL, 0, |
07b2391f | 4549 | response, RBD_IMAGE_ID_LEN_MAX, NULL); |
36be9a76 | 4550 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4551 | if (ret == -ENOENT) { |
4552 | image_id = kstrdup("", GFP_KERNEL); | |
4553 | ret = image_id ? 0 : -ENOMEM; | |
4554 | if (!ret) | |
4555 | rbd_dev->image_format = 1; | |
4556 | } else if (ret > sizeof (__le32)) { | |
4557 | void *p = response; | |
4558 | ||
4559 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4560 | NULL, GFP_NOIO); |
c0fba368 AE |
4561 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4562 | if (!ret) | |
4563 | rbd_dev->image_format = 2; | |
589d30e0 | 4564 | } else { |
c0fba368 AE |
4565 | ret = -EINVAL; |
4566 | } | |
4567 | ||
4568 | if (!ret) { | |
4569 | rbd_dev->spec->image_id = image_id; | |
4570 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4571 | } |
4572 | out: | |
4573 | kfree(response); | |
4574 | kfree(object_name); | |
4575 | ||
4576 | return ret; | |
4577 | } | |
4578 | ||
6fd48b3b AE |
4579 | /* Undo whatever state changes are made by v1 or v2 image probe */ |
4580 | ||
4581 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) | |
4582 | { | |
4583 | struct rbd_image_header *header; | |
4584 | ||
4585 | rbd_dev_remove_parent(rbd_dev); | |
4586 | rbd_spec_put(rbd_dev->parent_spec); | |
4587 | rbd_dev->parent_spec = NULL; | |
4588 | rbd_dev->parent_overlap = 0; | |
4589 | ||
4590 | /* Free dynamic fields from the header, then zero it out */ | |
4591 | ||
4592 | header = &rbd_dev->header; | |
4593 | rbd_snap_context_put(header->snapc); | |
4594 | kfree(header->snap_sizes); | |
4595 | kfree(header->snap_names); | |
4596 | kfree(header->object_prefix); | |
4597 | memset(header, 0, sizeof (*header)); | |
4598 | } | |
4599 | ||
a30b71b9 AE |
4600 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
4601 | { | |
4602 | int ret; | |
a30b71b9 AE |
4603 | |
4604 | /* Populate rbd image metadata */ | |
4605 | ||
4606 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
4607 | if (ret < 0) | |
4608 | goto out_err; | |
86b00e0d AE |
4609 | |
4610 | /* Version 1 images have no parent (no layering) */ | |
4611 | ||
4612 | rbd_dev->parent_spec = NULL; | |
4613 | rbd_dev->parent_overlap = 0; | |
4614 | ||
a30b71b9 AE |
4615 | dout("discovered version 1 image, header name is %s\n", |
4616 | rbd_dev->header_name); | |
4617 | ||
4618 | return 0; | |
4619 | ||
4620 | out_err: | |
4621 | kfree(rbd_dev->header_name); | |
4622 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
4623 | kfree(rbd_dev->spec->image_id); |
4624 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
4625 | |
4626 | return ret; | |
4627 | } | |
4628 | ||
4629 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
4630 | { | |
9d475de5 | 4631 | int ret; |
6e14b1a6 | 4632 | u64 ver = 0; |
a30b71b9 | 4633 | |
9d475de5 | 4634 | ret = rbd_dev_v2_image_size(rbd_dev); |
57385b51 | 4635 | if (ret) |
1e130199 AE |
4636 | goto out_err; |
4637 | ||
4638 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
4639 | ||
4640 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
57385b51 | 4641 | if (ret) |
b1b5402a AE |
4642 | goto out_err; |
4643 | ||
d889140c | 4644 | /* Get the and check features for the image */ |
b1b5402a AE |
4645 | |
4646 | ret = rbd_dev_v2_features(rbd_dev); | |
57385b51 | 4647 | if (ret) |
9d475de5 | 4648 | goto out_err; |
35d489f9 | 4649 | |
86b00e0d AE |
4650 | /* If the image supports layering, get the parent info */ |
4651 | ||
4652 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
4653 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
57385b51 | 4654 | if (ret) |
86b00e0d | 4655 | goto out_err; |
770eba6e AE |
4656 | rbd_warn(rbd_dev, "WARNING: kernel support for " |
4657 | "layered rbd images is EXPERIMENTAL!"); | |
86b00e0d AE |
4658 | } |
4659 | ||
cc070d59 AE |
4660 | /* If the image supports fancy striping, get its parameters */ |
4661 | ||
4662 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4663 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4664 | if (ret < 0) | |
4665 | goto out_err; | |
4666 | } | |
4667 | ||
6e14b1a6 AE |
4668 | /* crypto and compression type aren't (yet) supported for v2 images */ |
4669 | ||
4670 | rbd_dev->header.crypt_type = 0; | |
4671 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 4672 | |
6e14b1a6 AE |
4673 | /* Get the snapshot context, plus the header version */ |
4674 | ||
4675 | ret = rbd_dev_v2_snap_context(rbd_dev, &ver); | |
35d489f9 AE |
4676 | if (ret) |
4677 | goto out_err; | |
6e14b1a6 AE |
4678 | rbd_dev->header.obj_version = ver; |
4679 | ||
a30b71b9 AE |
4680 | dout("discovered version 2 image, header name is %s\n", |
4681 | rbd_dev->header_name); | |
4682 | ||
35152979 | 4683 | return 0; |
9d475de5 | 4684 | out_err: |
86b00e0d AE |
4685 | rbd_dev->parent_overlap = 0; |
4686 | rbd_spec_put(rbd_dev->parent_spec); | |
4687 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
4688 | kfree(rbd_dev->header_name); |
4689 | rbd_dev->header_name = NULL; | |
1e130199 AE |
4690 | kfree(rbd_dev->header.object_prefix); |
4691 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4692 | |
4693 | return ret; | |
a30b71b9 AE |
4694 | } |
4695 | ||
124afba2 | 4696 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4697 | { |
2f82ee54 | 4698 | struct rbd_device *parent = NULL; |
124afba2 AE |
4699 | struct rbd_spec *parent_spec; |
4700 | struct rbd_client *rbdc; | |
4701 | int ret; | |
4702 | ||
4703 | if (!rbd_dev->parent_spec) | |
4704 | return 0; | |
4705 | /* | |
4706 | * We need to pass a reference to the client and the parent | |
4707 | * spec when creating the parent rbd_dev. Images related by | |
4708 | * parent/child relationships always share both. | |
4709 | */ | |
4710 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4711 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4712 | ||
4713 | ret = -ENOMEM; | |
4714 | parent = rbd_dev_create(rbdc, parent_spec); | |
4715 | if (!parent) | |
4716 | goto out_err; | |
4717 | ||
4718 | ret = rbd_dev_image_probe(parent); | |
4719 | if (ret < 0) | |
4720 | goto out_err; | |
4721 | rbd_dev->parent = parent; | |
4722 | ||
4723 | return 0; | |
4724 | out_err: | |
4725 | if (parent) { | |
4726 | rbd_spec_put(rbd_dev->parent_spec); | |
4727 | kfree(rbd_dev->header_name); | |
4728 | rbd_dev_destroy(parent); | |
4729 | } else { | |
4730 | rbd_put_client(rbdc); | |
4731 | rbd_spec_put(parent_spec); | |
4732 | } | |
4733 | ||
4734 | return ret; | |
4735 | } | |
4736 | ||
200a6a8b | 4737 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4738 | { |
83a06263 | 4739 | int ret; |
d1cf5788 AE |
4740 | |
4741 | ret = rbd_dev_mapping_set(rbd_dev); | |
83a06263 | 4742 | if (ret) |
9bb81c9b | 4743 | return ret; |
5de10f3b | 4744 | |
83a06263 AE |
4745 | /* generate unique id: find highest unique id, add one */ |
4746 | rbd_dev_id_get(rbd_dev); | |
4747 | ||
4748 | /* Fill in the device name, now that we have its id. */ | |
4749 | BUILD_BUG_ON(DEV_NAME_LEN | |
4750 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4751 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4752 | ||
4753 | /* Get our block major device number. */ | |
4754 | ||
4755 | ret = register_blkdev(0, rbd_dev->name); | |
4756 | if (ret < 0) | |
4757 | goto err_out_id; | |
4758 | rbd_dev->major = ret; | |
4759 | ||
4760 | /* Set up the blkdev mapping. */ | |
4761 | ||
4762 | ret = rbd_init_disk(rbd_dev); | |
4763 | if (ret) | |
4764 | goto err_out_blkdev; | |
4765 | ||
4766 | ret = rbd_bus_add_dev(rbd_dev); | |
4767 | if (ret) | |
4768 | goto err_out_disk; | |
4769 | ||
83a06263 AE |
4770 | /* Everything's ready. Announce the disk to the world. */ |
4771 | ||
b5156e76 | 4772 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
129b79d4 | 4773 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4774 | add_disk(rbd_dev->disk); |
4775 | ||
4776 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4777 | (unsigned long long) rbd_dev->mapping.size); | |
4778 | ||
4779 | return ret; | |
2f82ee54 | 4780 | |
83a06263 AE |
4781 | err_out_disk: |
4782 | rbd_free_disk(rbd_dev); | |
4783 | err_out_blkdev: | |
4784 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4785 | err_out_id: | |
4786 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4787 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4788 | |
4789 | return ret; | |
4790 | } | |
4791 | ||
332bb12d AE |
4792 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4793 | { | |
4794 | struct rbd_spec *spec = rbd_dev->spec; | |
4795 | size_t size; | |
4796 | ||
4797 | /* Record the header object name for this rbd image. */ | |
4798 | ||
4799 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4800 | ||
4801 | if (rbd_dev->image_format == 1) | |
4802 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4803 | else | |
4804 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4805 | ||
4806 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4807 | if (!rbd_dev->header_name) | |
4808 | return -ENOMEM; | |
4809 | ||
4810 | if (rbd_dev->image_format == 1) | |
4811 | sprintf(rbd_dev->header_name, "%s%s", | |
4812 | spec->image_name, RBD_SUFFIX); | |
4813 | else | |
4814 | sprintf(rbd_dev->header_name, "%s%s", | |
4815 | RBD_HEADER_PREFIX, spec->image_id); | |
4816 | return 0; | |
4817 | } | |
4818 | ||
200a6a8b AE |
4819 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4820 | { | |
6fd48b3b AE |
4821 | int ret; |
4822 | ||
4823 | rbd_remove_all_snaps(rbd_dev); | |
4824 | rbd_dev_unprobe(rbd_dev); | |
4825 | ret = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4826 | if (ret) | |
4827 | rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret); | |
200a6a8b | 4828 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4829 | rbd_dev->header_name = NULL; |
4830 | rbd_dev->image_format = 0; | |
4831 | kfree(rbd_dev->spec->image_id); | |
4832 | rbd_dev->spec->image_id = NULL; | |
4833 | ||
200a6a8b AE |
4834 | rbd_dev_destroy(rbd_dev); |
4835 | } | |
4836 | ||
a30b71b9 AE |
4837 | /* |
4838 | * Probe for the existence of the header object for the given rbd | |
4839 | * device. For format 2 images this includes determining the image | |
4840 | * id. | |
4841 | */ | |
71f293e2 | 4842 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4843 | { |
4844 | int ret; | |
b644de2b | 4845 | int tmp; |
a30b71b9 AE |
4846 | |
4847 | /* | |
4848 | * Get the id from the image id object. If it's not a | |
4849 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4850 | * it's a format 1 image. | |
4851 | */ | |
4852 | ret = rbd_dev_image_id(rbd_dev); | |
4853 | if (ret) | |
c0fba368 AE |
4854 | return ret; |
4855 | rbd_assert(rbd_dev->spec->image_id); | |
4856 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4857 | ||
332bb12d AE |
4858 | ret = rbd_dev_header_name(rbd_dev); |
4859 | if (ret) | |
4860 | goto err_out_format; | |
4861 | ||
b644de2b AE |
4862 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
4863 | if (ret) | |
4864 | goto out_header_name; | |
4865 | ||
c0fba368 | 4866 | if (rbd_dev->image_format == 1) |
a30b71b9 AE |
4867 | ret = rbd_dev_v1_probe(rbd_dev); |
4868 | else | |
4869 | ret = rbd_dev_v2_probe(rbd_dev); | |
5655c4d9 | 4870 | if (ret) |
b644de2b | 4871 | goto err_out_watch; |
83a06263 | 4872 | |
9bb81c9b AE |
4873 | ret = rbd_dev_snaps_update(rbd_dev); |
4874 | if (ret) | |
6fd48b3b | 4875 | goto err_out_probe; |
9bb81c9b AE |
4876 | |
4877 | ret = rbd_dev_spec_update(rbd_dev); | |
4878 | if (ret) | |
4879 | goto err_out_snaps; | |
4880 | ||
4881 | ret = rbd_dev_probe_parent(rbd_dev); | |
4882 | if (ret) | |
4883 | goto err_out_snaps; | |
4884 | ||
200a6a8b | 4885 | ret = rbd_dev_device_setup(rbd_dev); |
6fd48b3b AE |
4886 | if (!ret) |
4887 | return 0; | |
83a06263 | 4888 | |
9bb81c9b AE |
4889 | err_out_snaps: |
4890 | rbd_remove_all_snaps(rbd_dev); | |
6fd48b3b AE |
4891 | err_out_probe: |
4892 | rbd_dev_unprobe(rbd_dev); | |
b644de2b AE |
4893 | err_out_watch: |
4894 | tmp = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4895 | if (tmp) | |
4896 | rbd_warn(rbd_dev, "unable to tear down watch request\n"); | |
332bb12d AE |
4897 | out_header_name: |
4898 | kfree(rbd_dev->header_name); | |
4899 | rbd_dev->header_name = NULL; | |
4900 | err_out_format: | |
4901 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
4902 | kfree(rbd_dev->spec->image_id); |
4903 | rbd_dev->spec->image_id = NULL; | |
4904 | ||
4905 | dout("probe failed, returning %d\n", ret); | |
4906 | ||
a30b71b9 AE |
4907 | return ret; |
4908 | } | |
4909 | ||
59c2be1e YS |
4910 | static ssize_t rbd_add(struct bus_type *bus, |
4911 | const char *buf, | |
4912 | size_t count) | |
602adf40 | 4913 | { |
cb8627c7 | 4914 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4915 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4916 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4917 | struct rbd_spec *spec = NULL; |
9d3997fd | 4918 | struct rbd_client *rbdc; |
27cc2594 AE |
4919 | struct ceph_osd_client *osdc; |
4920 | int rc = -ENOMEM; | |
602adf40 YS |
4921 | |
4922 | if (!try_module_get(THIS_MODULE)) | |
4923 | return -ENODEV; | |
4924 | ||
602adf40 | 4925 | /* parse add command */ |
859c31df | 4926 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4927 | if (rc < 0) |
bd4ba655 | 4928 | goto err_out_module; |
78cea76e | 4929 | |
9d3997fd AE |
4930 | rbdc = rbd_get_client(ceph_opts); |
4931 | if (IS_ERR(rbdc)) { | |
4932 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4933 | goto err_out_args; |
9d3997fd | 4934 | } |
c53d5893 | 4935 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4936 | |
602adf40 | 4937 | /* pick the pool */ |
9d3997fd | 4938 | osdc = &rbdc->client->osdc; |
859c31df | 4939 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4940 | if (rc < 0) |
4941 | goto err_out_client; | |
c0cd10db | 4942 | spec->pool_id = (u64)rc; |
859c31df | 4943 | |
0903e875 AE |
4944 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4945 | ||
c0cd10db AE |
4946 | if (spec->pool_id > (u64)U32_MAX) { |
4947 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
4948 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
4949 | rc = -EIO; |
4950 | goto err_out_client; | |
4951 | } | |
4952 | ||
c53d5893 | 4953 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4954 | if (!rbd_dev) |
4955 | goto err_out_client; | |
c53d5893 AE |
4956 | rbdc = NULL; /* rbd_dev now owns this */ |
4957 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4958 | |
bd4ba655 | 4959 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4960 | kfree(rbd_opts); |
4961 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4962 | |
71f293e2 | 4963 | rc = rbd_dev_image_probe(rbd_dev); |
a30b71b9 | 4964 | if (rc < 0) |
c53d5893 | 4965 | goto err_out_rbd_dev; |
05fd6f6f | 4966 | |
602adf40 | 4967 | return count; |
c53d5893 | 4968 | err_out_rbd_dev: |
9f5dffdc | 4969 | kfree(rbd_dev->header_name); |
c53d5893 | 4970 | rbd_dev_destroy(rbd_dev); |
bd4ba655 | 4971 | err_out_client: |
9d3997fd | 4972 | rbd_put_client(rbdc); |
0ddebc0c | 4973 | err_out_args: |
78cea76e AE |
4974 | if (ceph_opts) |
4975 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4976 | kfree(rbd_opts); |
859c31df | 4977 | rbd_spec_put(spec); |
bd4ba655 AE |
4978 | err_out_module: |
4979 | module_put(THIS_MODULE); | |
27cc2594 | 4980 | |
602adf40 | 4981 | dout("Error adding device %s\n", buf); |
27cc2594 | 4982 | |
c0cd10db | 4983 | return (ssize_t)rc; |
602adf40 YS |
4984 | } |
4985 | ||
de71a297 | 4986 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4987 | { |
4988 | struct list_head *tmp; | |
4989 | struct rbd_device *rbd_dev; | |
4990 | ||
e124a82f | 4991 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4992 | list_for_each(tmp, &rbd_dev_list) { |
4993 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4994 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4995 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4996 | return rbd_dev; |
e124a82f | 4997 | } |
602adf40 | 4998 | } |
e124a82f | 4999 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
5000 | return NULL; |
5001 | } | |
5002 | ||
200a6a8b | 5003 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5004 | { |
593a9e7b | 5005 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5006 | |
602adf40 | 5007 | rbd_free_disk(rbd_dev); |
200a6a8b AE |
5008 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
5009 | rbd_dev_clear_mapping(rbd_dev); | |
602adf40 | 5010 | unregister_blkdev(rbd_dev->major, rbd_dev->name); |
200a6a8b | 5011 | rbd_dev->major = 0; |
e2839308 | 5012 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5013 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5014 | } |
5015 | ||
05a46afd AE |
5016 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5017 | { | |
ad945fc1 | 5018 | while (rbd_dev->parent) { |
05a46afd AE |
5019 | struct rbd_device *first = rbd_dev; |
5020 | struct rbd_device *second = first->parent; | |
5021 | struct rbd_device *third; | |
5022 | ||
5023 | /* | |
5024 | * Follow to the parent with no grandparent and | |
5025 | * remove it. | |
5026 | */ | |
5027 | while (second && (third = second->parent)) { | |
5028 | first = second; | |
5029 | second = third; | |
5030 | } | |
ad945fc1 | 5031 | rbd_assert(second); |
b480815a | 5032 | rbd_bus_del_dev(second); |
8ad42cd0 | 5033 | rbd_dev_image_release(second); |
ad945fc1 AE |
5034 | first->parent = NULL; |
5035 | first->parent_overlap = 0; | |
5036 | ||
5037 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5038 | rbd_spec_put(first->parent_spec); |
5039 | first->parent_spec = NULL; | |
05a46afd AE |
5040 | } |
5041 | } | |
5042 | ||
dfc5606d YS |
5043 | static ssize_t rbd_remove(struct bus_type *bus, |
5044 | const char *buf, | |
5045 | size_t count) | |
602adf40 YS |
5046 | { |
5047 | struct rbd_device *rbd_dev = NULL; | |
0d8189e1 | 5048 | int target_id; |
602adf40 | 5049 | unsigned long ul; |
0d8189e1 | 5050 | int ret; |
602adf40 | 5051 | |
0d8189e1 AE |
5052 | ret = strict_strtoul(buf, 10, &ul); |
5053 | if (ret) | |
5054 | return ret; | |
602adf40 YS |
5055 | |
5056 | /* convert to int; abort if we lost anything in the conversion */ | |
5057 | target_id = (int) ul; | |
5058 | if (target_id != ul) | |
5059 | return -EINVAL; | |
5060 | ||
5061 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
5062 | ||
5063 | rbd_dev = __rbd_get_dev(target_id); | |
5064 | if (!rbd_dev) { | |
5065 | ret = -ENOENT; | |
5066 | goto done; | |
42382b70 AE |
5067 | } |
5068 | ||
a14ea269 | 5069 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 5070 | if (rbd_dev->open_count) |
42382b70 | 5071 | ret = -EBUSY; |
b82d167b AE |
5072 | else |
5073 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 5074 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 5075 | if (ret < 0) |
42382b70 | 5076 | goto done; |
0d8189e1 | 5077 | ret = count; |
b480815a | 5078 | rbd_bus_del_dev(rbd_dev); |
8ad42cd0 | 5079 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5080 | module_put(THIS_MODULE); |
602adf40 YS |
5081 | done: |
5082 | mutex_unlock(&ctl_mutex); | |
aafb230e | 5083 | |
602adf40 YS |
5084 | return ret; |
5085 | } | |
5086 | ||
602adf40 YS |
5087 | /* |
5088 | * create control files in sysfs | |
dfc5606d | 5089 | * /sys/bus/rbd/... |
602adf40 YS |
5090 | */ |
5091 | static int rbd_sysfs_init(void) | |
5092 | { | |
dfc5606d | 5093 | int ret; |
602adf40 | 5094 | |
fed4c143 | 5095 | ret = device_register(&rbd_root_dev); |
21079786 | 5096 | if (ret < 0) |
dfc5606d | 5097 | return ret; |
602adf40 | 5098 | |
fed4c143 AE |
5099 | ret = bus_register(&rbd_bus_type); |
5100 | if (ret < 0) | |
5101 | device_unregister(&rbd_root_dev); | |
602adf40 | 5102 | |
602adf40 YS |
5103 | return ret; |
5104 | } | |
5105 | ||
5106 | static void rbd_sysfs_cleanup(void) | |
5107 | { | |
dfc5606d | 5108 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5109 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5110 | } |
5111 | ||
cc344fa1 | 5112 | static int __init rbd_init(void) |
602adf40 YS |
5113 | { |
5114 | int rc; | |
5115 | ||
1e32d34c AE |
5116 | if (!libceph_compatible(NULL)) { |
5117 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
5118 | ||
5119 | return -EINVAL; | |
5120 | } | |
602adf40 YS |
5121 | rc = rbd_sysfs_init(); |
5122 | if (rc) | |
5123 | return rc; | |
f0f8cef5 | 5124 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); |
602adf40 YS |
5125 | return 0; |
5126 | } | |
5127 | ||
cc344fa1 | 5128 | static void __exit rbd_exit(void) |
602adf40 YS |
5129 | { |
5130 | rbd_sysfs_cleanup(); | |
5131 | } | |
5132 | ||
5133 | module_init(rbd_init); | |
5134 | module_exit(rbd_exit); | |
5135 | ||
5136 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
5137 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
5138 | MODULE_DESCRIPTION("rados block device"); | |
5139 | ||
5140 | /* following authorship retained from original osdblk.c */ | |
5141 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5142 | ||
5143 | MODULE_LICENSE("GPL"); |