Commit | Line | Data |
---|---|---|
e2a58ee5 | 1 | |
602adf40 YS |
2 | /* |
3 | rbd.c -- Export ceph rados objects as a Linux block device | |
4 | ||
5 | ||
6 | based on drivers/block/osdblk.c: | |
7 | ||
8 | Copyright 2009 Red Hat, Inc. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; see the file COPYING. If not, write to | |
21 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
22 | ||
23 | ||
24 | ||
dfc5606d | 25 | For usage instructions, please refer to: |
602adf40 | 26 | |
dfc5606d | 27 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
28 | |
29 | */ | |
30 | ||
31 | #include <linux/ceph/libceph.h> | |
32 | #include <linux/ceph/osd_client.h> | |
33 | #include <linux/ceph/mon_client.h> | |
34 | #include <linux/ceph/decode.h> | |
59c2be1e | 35 | #include <linux/parser.h> |
30d1cff8 | 36 | #include <linux/bsearch.h> |
602adf40 YS |
37 | |
38 | #include <linux/kernel.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/module.h> | |
41 | #include <linux/fs.h> | |
42 | #include <linux/blkdev.h> | |
1c2a9dfe | 43 | #include <linux/slab.h> |
602adf40 YS |
44 | |
45 | #include "rbd_types.h" | |
46 | ||
aafb230e AE |
47 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
48 | ||
593a9e7b AE |
49 | /* |
50 | * The basic unit of block I/O is a sector. It is interpreted in a | |
51 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
52 | * universally 512 bytes. These symbols are just slightly more | |
53 | * meaningful than the bare numbers they represent. | |
54 | */ | |
55 | #define SECTOR_SHIFT 9 | |
56 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
57 | ||
f0f8cef5 AE |
58 | #define RBD_DRV_NAME "rbd" |
59 | #define RBD_DRV_NAME_LONG "rbd (rados block device)" | |
602adf40 YS |
60 | |
61 | #define RBD_MINORS_PER_MAJOR 256 /* max minors per blkdev */ | |
62 | ||
d4b125e9 AE |
63 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
64 | #define RBD_MAX_SNAP_NAME_LEN \ | |
65 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
66 | ||
35d489f9 | 67 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
68 | |
69 | #define RBD_SNAP_HEAD_NAME "-" | |
70 | ||
9682fc6d AE |
71 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
72 | ||
9e15b77d AE |
73 | /* This allows a single page to hold an image name sent by OSD */ |
74 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 75 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 76 | |
1e130199 | 77 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 78 | |
d889140c AE |
79 | /* Feature bits */ |
80 | ||
5cbf6f12 AE |
81 | #define RBD_FEATURE_LAYERING (1<<0) |
82 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
83 | #define RBD_FEATURES_ALL \ | |
84 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
85 | |
86 | /* Features supported by this (client software) implementation. */ | |
87 | ||
770eba6e | 88 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 89 | |
81a89793 AE |
90 | /* |
91 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
92 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
93 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
94 | * enough to hold all possible device names. | |
95 | */ | |
602adf40 | 96 | #define DEV_NAME_LEN 32 |
81a89793 | 97 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
98 | |
99 | /* | |
100 | * block device image metadata (in-memory version) | |
101 | */ | |
102 | struct rbd_image_header { | |
f84344f3 | 103 | /* These four fields never change for a given rbd image */ |
849b4260 | 104 | char *object_prefix; |
34b13184 | 105 | u64 features; |
602adf40 YS |
106 | __u8 obj_order; |
107 | __u8 crypt_type; | |
108 | __u8 comp_type; | |
602adf40 | 109 | |
f84344f3 AE |
110 | /* The remaining fields need to be updated occasionally */ |
111 | u64 image_size; | |
112 | struct ceph_snap_context *snapc; | |
602adf40 YS |
113 | char *snap_names; |
114 | u64 *snap_sizes; | |
59c2be1e | 115 | |
500d0c0f AE |
116 | u64 stripe_unit; |
117 | u64 stripe_count; | |
59c2be1e YS |
118 | }; |
119 | ||
0d7dbfce AE |
120 | /* |
121 | * An rbd image specification. | |
122 | * | |
123 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
124 | * identify an image. Each rbd_dev structure includes a pointer to |
125 | * an rbd_spec structure that encapsulates this identity. | |
126 | * | |
127 | * Each of the id's in an rbd_spec has an associated name. For a | |
128 | * user-mapped image, the names are supplied and the id's associated | |
129 | * with them are looked up. For a layered image, a parent image is | |
130 | * defined by the tuple, and the names are looked up. | |
131 | * | |
132 | * An rbd_dev structure contains a parent_spec pointer which is | |
133 | * non-null if the image it represents is a child in a layered | |
134 | * image. This pointer will refer to the rbd_spec structure used | |
135 | * by the parent rbd_dev for its own identity (i.e., the structure | |
136 | * is shared between the parent and child). | |
137 | * | |
138 | * Since these structures are populated once, during the discovery | |
139 | * phase of image construction, they are effectively immutable so | |
140 | * we make no effort to synchronize access to them. | |
141 | * | |
142 | * Note that code herein does not assume the image name is known (it | |
143 | * could be a null pointer). | |
0d7dbfce AE |
144 | */ |
145 | struct rbd_spec { | |
146 | u64 pool_id; | |
ecb4dc22 | 147 | const char *pool_name; |
0d7dbfce | 148 | |
ecb4dc22 AE |
149 | const char *image_id; |
150 | const char *image_name; | |
0d7dbfce AE |
151 | |
152 | u64 snap_id; | |
ecb4dc22 | 153 | const char *snap_name; |
0d7dbfce AE |
154 | |
155 | struct kref kref; | |
156 | }; | |
157 | ||
602adf40 | 158 | /* |
f0f8cef5 | 159 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
160 | */ |
161 | struct rbd_client { | |
162 | struct ceph_client *client; | |
163 | struct kref kref; | |
164 | struct list_head node; | |
165 | }; | |
166 | ||
bf0d5f50 AE |
167 | struct rbd_img_request; |
168 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
169 | ||
170 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
171 | ||
172 | struct rbd_obj_request; | |
173 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
174 | ||
9969ebc5 AE |
175 | enum obj_request_type { |
176 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
177 | }; | |
bf0d5f50 | 178 | |
926f9b3f AE |
179 | enum obj_req_flags { |
180 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 181 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
182 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
183 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
184 | }; |
185 | ||
bf0d5f50 AE |
186 | struct rbd_obj_request { |
187 | const char *object_name; | |
188 | u64 offset; /* object start byte */ | |
189 | u64 length; /* bytes from offset */ | |
926f9b3f | 190 | unsigned long flags; |
bf0d5f50 | 191 | |
c5b5ef6c AE |
192 | /* |
193 | * An object request associated with an image will have its | |
194 | * img_data flag set; a standalone object request will not. | |
195 | * | |
196 | * A standalone object request will have which == BAD_WHICH | |
197 | * and a null obj_request pointer. | |
198 | * | |
199 | * An object request initiated in support of a layered image | |
200 | * object (to check for its existence before a write) will | |
201 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
202 | * | |
203 | * Finally, an object request for rbd image data will have | |
204 | * which != BAD_WHICH, and will have a non-null img_request | |
205 | * pointer. The value of which will be in the range | |
206 | * 0..(img_request->obj_request_count-1). | |
207 | */ | |
208 | union { | |
209 | struct rbd_obj_request *obj_request; /* STAT op */ | |
210 | struct { | |
211 | struct rbd_img_request *img_request; | |
212 | u64 img_offset; | |
213 | /* links for img_request->obj_requests list */ | |
214 | struct list_head links; | |
215 | }; | |
216 | }; | |
bf0d5f50 AE |
217 | u32 which; /* posn image request list */ |
218 | ||
219 | enum obj_request_type type; | |
788e2df3 AE |
220 | union { |
221 | struct bio *bio_list; | |
222 | struct { | |
223 | struct page **pages; | |
224 | u32 page_count; | |
225 | }; | |
226 | }; | |
0eefd470 | 227 | struct page **copyup_pages; |
bf0d5f50 AE |
228 | |
229 | struct ceph_osd_request *osd_req; | |
230 | ||
231 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 232 | int result; |
bf0d5f50 AE |
233 | |
234 | rbd_obj_callback_t callback; | |
788e2df3 | 235 | struct completion completion; |
bf0d5f50 AE |
236 | |
237 | struct kref kref; | |
238 | }; | |
239 | ||
0c425248 | 240 | enum img_req_flags { |
9849e986 AE |
241 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
242 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 243 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
244 | }; |
245 | ||
bf0d5f50 | 246 | struct rbd_img_request { |
bf0d5f50 AE |
247 | struct rbd_device *rbd_dev; |
248 | u64 offset; /* starting image byte offset */ | |
249 | u64 length; /* byte count from offset */ | |
0c425248 | 250 | unsigned long flags; |
bf0d5f50 | 251 | union { |
9849e986 | 252 | u64 snap_id; /* for reads */ |
bf0d5f50 | 253 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
254 | }; |
255 | union { | |
256 | struct request *rq; /* block request */ | |
257 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 258 | }; |
3d7efd18 | 259 | struct page **copyup_pages; |
bf0d5f50 AE |
260 | spinlock_t completion_lock;/* protects next_completion */ |
261 | u32 next_completion; | |
262 | rbd_img_callback_t callback; | |
55f27e09 | 263 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 264 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
265 | |
266 | u32 obj_request_count; | |
267 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
268 | ||
269 | struct kref kref; | |
270 | }; | |
271 | ||
272 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 273 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 274 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 275 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 276 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 277 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 278 | |
f84344f3 | 279 | struct rbd_mapping { |
99c1f08f | 280 | u64 size; |
34b13184 | 281 | u64 features; |
f84344f3 AE |
282 | bool read_only; |
283 | }; | |
284 | ||
602adf40 YS |
285 | /* |
286 | * a single device | |
287 | */ | |
288 | struct rbd_device { | |
de71a297 | 289 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
290 | |
291 | int major; /* blkdev assigned major */ | |
292 | struct gendisk *disk; /* blkdev's gendisk and rq */ | |
602adf40 | 293 | |
a30b71b9 | 294 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
295 | struct rbd_client *rbd_client; |
296 | ||
297 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
298 | ||
b82d167b | 299 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
300 | |
301 | struct rbd_image_header header; | |
b82d167b | 302 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 303 | struct rbd_spec *spec; |
602adf40 | 304 | |
0d7dbfce | 305 | char *header_name; |
971f839a | 306 | |
0903e875 AE |
307 | struct ceph_file_layout layout; |
308 | ||
59c2be1e | 309 | struct ceph_osd_event *watch_event; |
975241af | 310 | struct rbd_obj_request *watch_request; |
59c2be1e | 311 | |
86b00e0d AE |
312 | struct rbd_spec *parent_spec; |
313 | u64 parent_overlap; | |
2f82ee54 | 314 | struct rbd_device *parent; |
86b00e0d | 315 | |
c666601a JD |
316 | /* protects updating the header */ |
317 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
318 | |
319 | struct rbd_mapping mapping; | |
602adf40 YS |
320 | |
321 | struct list_head node; | |
dfc5606d | 322 | |
dfc5606d YS |
323 | /* sysfs related */ |
324 | struct device dev; | |
b82d167b | 325 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
326 | }; |
327 | ||
b82d167b AE |
328 | /* |
329 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
330 | * rbd_dev->lock is used to protect access. | |
331 | * | |
332 | * Currently, only the "removing" flag (which is coupled with the | |
333 | * "open_count" field) requires atomic access. | |
334 | */ | |
6d292906 AE |
335 | enum rbd_dev_flags { |
336 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 337 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
338 | }; |
339 | ||
602adf40 | 340 | static DEFINE_MUTEX(ctl_mutex); /* Serialize open/close/setup/teardown */ |
e124a82f | 341 | |
602adf40 | 342 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
343 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
344 | ||
432b8587 AE |
345 | static LIST_HEAD(rbd_client_list); /* clients */ |
346 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 347 | |
78c2a44a AE |
348 | /* Slab caches for frequently-allocated structures */ |
349 | ||
1c2a9dfe | 350 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 351 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 352 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 353 | |
3d7efd18 AE |
354 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
355 | ||
200a6a8b | 356 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 357 | |
f0f8cef5 AE |
358 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
359 | size_t count); | |
360 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
361 | size_t count); | |
71f293e2 | 362 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev); |
f0f8cef5 AE |
363 | |
364 | static struct bus_attribute rbd_bus_attrs[] = { | |
365 | __ATTR(add, S_IWUSR, NULL, rbd_add), | |
366 | __ATTR(remove, S_IWUSR, NULL, rbd_remove), | |
367 | __ATTR_NULL | |
368 | }; | |
369 | ||
370 | static struct bus_type rbd_bus_type = { | |
371 | .name = "rbd", | |
372 | .bus_attrs = rbd_bus_attrs, | |
373 | }; | |
374 | ||
375 | static void rbd_root_dev_release(struct device *dev) | |
376 | { | |
377 | } | |
378 | ||
379 | static struct device rbd_root_dev = { | |
380 | .init_name = "rbd", | |
381 | .release = rbd_root_dev_release, | |
382 | }; | |
383 | ||
06ecc6cb AE |
384 | static __printf(2, 3) |
385 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
386 | { | |
387 | struct va_format vaf; | |
388 | va_list args; | |
389 | ||
390 | va_start(args, fmt); | |
391 | vaf.fmt = fmt; | |
392 | vaf.va = &args; | |
393 | ||
394 | if (!rbd_dev) | |
395 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
396 | else if (rbd_dev->disk) | |
397 | printk(KERN_WARNING "%s: %s: %pV\n", | |
398 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
399 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
400 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
401 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
402 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
403 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
404 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
405 | else /* punt */ | |
406 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
407 | RBD_DRV_NAME, rbd_dev, &vaf); | |
408 | va_end(args); | |
409 | } | |
410 | ||
aafb230e AE |
411 | #ifdef RBD_DEBUG |
412 | #define rbd_assert(expr) \ | |
413 | if (unlikely(!(expr))) { \ | |
414 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
415 | "at line %d:\n\n" \ | |
416 | "\trbd_assert(%s);\n\n", \ | |
417 | __func__, __LINE__, #expr); \ | |
418 | BUG(); \ | |
419 | } | |
420 | #else /* !RBD_DEBUG */ | |
421 | # define rbd_assert(expr) ((void) 0) | |
422 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 423 | |
b454e36d | 424 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
425 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
426 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 427 | |
cc4a38bd AE |
428 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
429 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev); | |
54cac61f AE |
430 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
431 | u64 snap_id); | |
2ad3d716 AE |
432 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
433 | u8 *order, u64 *snap_size); | |
434 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
435 | u64 *snap_features); | |
436 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 437 | |
602adf40 YS |
438 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
439 | { | |
f0f8cef5 | 440 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 441 | bool removing = false; |
602adf40 | 442 | |
f84344f3 | 443 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
444 | return -EROFS; |
445 | ||
a14ea269 | 446 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
447 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
448 | removing = true; | |
449 | else | |
450 | rbd_dev->open_count++; | |
a14ea269 | 451 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
452 | if (removing) |
453 | return -ENOENT; | |
454 | ||
42382b70 | 455 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 456 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 457 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
42382b70 | 458 | mutex_unlock(&ctl_mutex); |
340c7a2b | 459 | |
602adf40 YS |
460 | return 0; |
461 | } | |
462 | ||
dfc5606d YS |
463 | static int rbd_release(struct gendisk *disk, fmode_t mode) |
464 | { | |
465 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
466 | unsigned long open_count_before; |
467 | ||
a14ea269 | 468 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 469 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 470 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 471 | rbd_assert(open_count_before > 0); |
dfc5606d | 472 | |
42382b70 | 473 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
c3e946ce | 474 | put_device(&rbd_dev->dev); |
42382b70 | 475 | mutex_unlock(&ctl_mutex); |
dfc5606d YS |
476 | |
477 | return 0; | |
478 | } | |
479 | ||
602adf40 YS |
480 | static const struct block_device_operations rbd_bd_ops = { |
481 | .owner = THIS_MODULE, | |
482 | .open = rbd_open, | |
dfc5606d | 483 | .release = rbd_release, |
602adf40 YS |
484 | }; |
485 | ||
486 | /* | |
487 | * Initialize an rbd client instance. | |
43ae4701 | 488 | * We own *ceph_opts. |
602adf40 | 489 | */ |
f8c38929 | 490 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
491 | { |
492 | struct rbd_client *rbdc; | |
493 | int ret = -ENOMEM; | |
494 | ||
37206ee5 | 495 | dout("%s:\n", __func__); |
602adf40 YS |
496 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
497 | if (!rbdc) | |
498 | goto out_opt; | |
499 | ||
500 | kref_init(&rbdc->kref); | |
501 | INIT_LIST_HEAD(&rbdc->node); | |
502 | ||
bc534d86 AE |
503 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
504 | ||
43ae4701 | 505 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 506 | if (IS_ERR(rbdc->client)) |
bc534d86 | 507 | goto out_mutex; |
43ae4701 | 508 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
509 | |
510 | ret = ceph_open_session(rbdc->client); | |
511 | if (ret < 0) | |
512 | goto out_err; | |
513 | ||
432b8587 | 514 | spin_lock(&rbd_client_list_lock); |
602adf40 | 515 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 516 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 517 | |
bc534d86 | 518 | mutex_unlock(&ctl_mutex); |
37206ee5 | 519 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 520 | |
602adf40 YS |
521 | return rbdc; |
522 | ||
523 | out_err: | |
524 | ceph_destroy_client(rbdc->client); | |
bc534d86 AE |
525 | out_mutex: |
526 | mutex_unlock(&ctl_mutex); | |
602adf40 YS |
527 | kfree(rbdc); |
528 | out_opt: | |
43ae4701 AE |
529 | if (ceph_opts) |
530 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
531 | dout("%s: error %d\n", __func__, ret); |
532 | ||
28f259b7 | 533 | return ERR_PTR(ret); |
602adf40 YS |
534 | } |
535 | ||
2f82ee54 AE |
536 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
537 | { | |
538 | kref_get(&rbdc->kref); | |
539 | ||
540 | return rbdc; | |
541 | } | |
542 | ||
602adf40 | 543 | /* |
1f7ba331 AE |
544 | * Find a ceph client with specific addr and configuration. If |
545 | * found, bump its reference count. | |
602adf40 | 546 | */ |
1f7ba331 | 547 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
548 | { |
549 | struct rbd_client *client_node; | |
1f7ba331 | 550 | bool found = false; |
602adf40 | 551 | |
43ae4701 | 552 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
553 | return NULL; |
554 | ||
1f7ba331 AE |
555 | spin_lock(&rbd_client_list_lock); |
556 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
557 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
558 | __rbd_get_client(client_node); |
559 | ||
1f7ba331 AE |
560 | found = true; |
561 | break; | |
562 | } | |
563 | } | |
564 | spin_unlock(&rbd_client_list_lock); | |
565 | ||
566 | return found ? client_node : NULL; | |
602adf40 YS |
567 | } |
568 | ||
59c2be1e YS |
569 | /* |
570 | * mount options | |
571 | */ | |
572 | enum { | |
59c2be1e YS |
573 | Opt_last_int, |
574 | /* int args above */ | |
575 | Opt_last_string, | |
576 | /* string args above */ | |
cc0538b6 AE |
577 | Opt_read_only, |
578 | Opt_read_write, | |
579 | /* Boolean args above */ | |
580 | Opt_last_bool, | |
59c2be1e YS |
581 | }; |
582 | ||
43ae4701 | 583 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
584 | /* int args above */ |
585 | /* string args above */ | |
be466c1c | 586 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
587 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
588 | {Opt_read_write, "read_write"}, | |
589 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
590 | /* Boolean args above */ | |
59c2be1e YS |
591 | {-1, NULL} |
592 | }; | |
593 | ||
98571b5a AE |
594 | struct rbd_options { |
595 | bool read_only; | |
596 | }; | |
597 | ||
598 | #define RBD_READ_ONLY_DEFAULT false | |
599 | ||
59c2be1e YS |
600 | static int parse_rbd_opts_token(char *c, void *private) |
601 | { | |
43ae4701 | 602 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
603 | substring_t argstr[MAX_OPT_ARGS]; |
604 | int token, intval, ret; | |
605 | ||
43ae4701 | 606 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
607 | if (token < 0) |
608 | return -EINVAL; | |
609 | ||
610 | if (token < Opt_last_int) { | |
611 | ret = match_int(&argstr[0], &intval); | |
612 | if (ret < 0) { | |
613 | pr_err("bad mount option arg (not int) " | |
614 | "at '%s'\n", c); | |
615 | return ret; | |
616 | } | |
617 | dout("got int token %d val %d\n", token, intval); | |
618 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
619 | dout("got string token %d val %s\n", token, | |
620 | argstr[0].from); | |
cc0538b6 AE |
621 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
622 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
623 | } else { |
624 | dout("got token %d\n", token); | |
625 | } | |
626 | ||
627 | switch (token) { | |
cc0538b6 AE |
628 | case Opt_read_only: |
629 | rbd_opts->read_only = true; | |
630 | break; | |
631 | case Opt_read_write: | |
632 | rbd_opts->read_only = false; | |
633 | break; | |
59c2be1e | 634 | default: |
aafb230e AE |
635 | rbd_assert(false); |
636 | break; | |
59c2be1e YS |
637 | } |
638 | return 0; | |
639 | } | |
640 | ||
602adf40 YS |
641 | /* |
642 | * Get a ceph client with specific addr and configuration, if one does | |
643 | * not exist create it. | |
644 | */ | |
9d3997fd | 645 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 646 | { |
f8c38929 | 647 | struct rbd_client *rbdc; |
59c2be1e | 648 | |
1f7ba331 | 649 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 650 | if (rbdc) /* using an existing client */ |
43ae4701 | 651 | ceph_destroy_options(ceph_opts); |
9d3997fd | 652 | else |
f8c38929 | 653 | rbdc = rbd_client_create(ceph_opts); |
602adf40 | 654 | |
9d3997fd | 655 | return rbdc; |
602adf40 YS |
656 | } |
657 | ||
658 | /* | |
659 | * Destroy ceph client | |
d23a4b3f | 660 | * |
432b8587 | 661 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
662 | */ |
663 | static void rbd_client_release(struct kref *kref) | |
664 | { | |
665 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
666 | ||
37206ee5 | 667 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 668 | spin_lock(&rbd_client_list_lock); |
602adf40 | 669 | list_del(&rbdc->node); |
cd9d9f5d | 670 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
671 | |
672 | ceph_destroy_client(rbdc->client); | |
673 | kfree(rbdc); | |
674 | } | |
675 | ||
676 | /* | |
677 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
678 | * it. | |
679 | */ | |
9d3997fd | 680 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 681 | { |
c53d5893 AE |
682 | if (rbdc) |
683 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
684 | } |
685 | ||
a30b71b9 AE |
686 | static bool rbd_image_format_valid(u32 image_format) |
687 | { | |
688 | return image_format == 1 || image_format == 2; | |
689 | } | |
690 | ||
8e94af8e AE |
691 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
692 | { | |
103a150f AE |
693 | size_t size; |
694 | u32 snap_count; | |
695 | ||
696 | /* The header has to start with the magic rbd header text */ | |
697 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
698 | return false; | |
699 | ||
db2388b6 AE |
700 | /* The bio layer requires at least sector-sized I/O */ |
701 | ||
702 | if (ondisk->options.order < SECTOR_SHIFT) | |
703 | return false; | |
704 | ||
705 | /* If we use u64 in a few spots we may be able to loosen this */ | |
706 | ||
707 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
708 | return false; | |
709 | ||
103a150f AE |
710 | /* |
711 | * The size of a snapshot header has to fit in a size_t, and | |
712 | * that limits the number of snapshots. | |
713 | */ | |
714 | snap_count = le32_to_cpu(ondisk->snap_count); | |
715 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
716 | if (snap_count > size / sizeof (__le64)) | |
717 | return false; | |
718 | ||
719 | /* | |
720 | * Not only that, but the size of the entire the snapshot | |
721 | * header must also be representable in a size_t. | |
722 | */ | |
723 | size -= snap_count * sizeof (__le64); | |
724 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
725 | return false; | |
726 | ||
727 | return true; | |
8e94af8e AE |
728 | } |
729 | ||
602adf40 YS |
730 | /* |
731 | * Create a new header structure, translate header format from the on-disk | |
732 | * header. | |
733 | */ | |
734 | static int rbd_header_from_disk(struct rbd_image_header *header, | |
4156d998 | 735 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 736 | { |
ccece235 | 737 | u32 snap_count; |
58c17b0e | 738 | size_t len; |
d2bb24e5 | 739 | size_t size; |
621901d6 | 740 | u32 i; |
602adf40 | 741 | |
6a52325f AE |
742 | memset(header, 0, sizeof (*header)); |
743 | ||
103a150f AE |
744 | snap_count = le32_to_cpu(ondisk->snap_count); |
745 | ||
58c17b0e AE |
746 | len = strnlen(ondisk->object_prefix, sizeof (ondisk->object_prefix)); |
747 | header->object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
6a52325f | 748 | if (!header->object_prefix) |
602adf40 | 749 | return -ENOMEM; |
58c17b0e AE |
750 | memcpy(header->object_prefix, ondisk->object_prefix, len); |
751 | header->object_prefix[len] = '\0'; | |
00f1f36f | 752 | |
602adf40 | 753 | if (snap_count) { |
f785cc1d AE |
754 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
755 | ||
621901d6 AE |
756 | /* Save a copy of the snapshot names */ |
757 | ||
f785cc1d AE |
758 | if (snap_names_len > (u64) SIZE_MAX) |
759 | return -EIO; | |
760 | header->snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
602adf40 | 761 | if (!header->snap_names) |
6a52325f | 762 | goto out_err; |
f785cc1d AE |
763 | /* |
764 | * Note that rbd_dev_v1_header_read() guarantees | |
765 | * the ondisk buffer we're working with has | |
766 | * snap_names_len bytes beyond the end of the | |
767 | * snapshot id array, this memcpy() is safe. | |
768 | */ | |
769 | memcpy(header->snap_names, &ondisk->snaps[snap_count], | |
770 | snap_names_len); | |
6a52325f | 771 | |
621901d6 AE |
772 | /* Record each snapshot's size */ |
773 | ||
d2bb24e5 AE |
774 | size = snap_count * sizeof (*header->snap_sizes); |
775 | header->snap_sizes = kmalloc(size, GFP_KERNEL); | |
602adf40 | 776 | if (!header->snap_sizes) |
6a52325f | 777 | goto out_err; |
621901d6 AE |
778 | for (i = 0; i < snap_count; i++) |
779 | header->snap_sizes[i] = | |
780 | le64_to_cpu(ondisk->snaps[i].image_size); | |
602adf40 YS |
781 | } else { |
782 | header->snap_names = NULL; | |
783 | header->snap_sizes = NULL; | |
784 | } | |
849b4260 | 785 | |
34b13184 | 786 | header->features = 0; /* No features support in v1 images */ |
602adf40 YS |
787 | header->obj_order = ondisk->options.order; |
788 | header->crypt_type = ondisk->options.crypt_type; | |
789 | header->comp_type = ondisk->options.comp_type; | |
6a52325f | 790 | |
621901d6 AE |
791 | /* Allocate and fill in the snapshot context */ |
792 | ||
f84344f3 | 793 | header->image_size = le64_to_cpu(ondisk->image_size); |
468521c1 | 794 | |
812164f8 | 795 | header->snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
6a52325f AE |
796 | if (!header->snapc) |
797 | goto out_err; | |
505cbb9b | 798 | header->snapc->seq = le64_to_cpu(ondisk->snap_seq); |
621901d6 | 799 | for (i = 0; i < snap_count; i++) |
468521c1 | 800 | header->snapc->snaps[i] = le64_to_cpu(ondisk->snaps[i].id); |
602adf40 YS |
801 | |
802 | return 0; | |
803 | ||
6a52325f | 804 | out_err: |
849b4260 | 805 | kfree(header->snap_sizes); |
ccece235 | 806 | header->snap_sizes = NULL; |
602adf40 | 807 | kfree(header->snap_names); |
ccece235 | 808 | header->snap_names = NULL; |
6a52325f AE |
809 | kfree(header->object_prefix); |
810 | header->object_prefix = NULL; | |
ccece235 | 811 | |
00f1f36f | 812 | return -ENOMEM; |
602adf40 YS |
813 | } |
814 | ||
9682fc6d AE |
815 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
816 | { | |
817 | const char *snap_name; | |
818 | ||
819 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
820 | ||
821 | /* Skip over names until we find the one we are looking for */ | |
822 | ||
823 | snap_name = rbd_dev->header.snap_names; | |
824 | while (which--) | |
825 | snap_name += strlen(snap_name) + 1; | |
826 | ||
827 | return kstrdup(snap_name, GFP_KERNEL); | |
828 | } | |
829 | ||
30d1cff8 AE |
830 | /* |
831 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
832 | * Note that result is for snapshots in *descending* order. | |
833 | */ | |
834 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
835 | { | |
836 | u64 snap_id1 = *(u64 *)s1; | |
837 | u64 snap_id2 = *(u64 *)s2; | |
838 | ||
839 | if (snap_id1 < snap_id2) | |
840 | return 1; | |
841 | return snap_id1 == snap_id2 ? 0 : -1; | |
842 | } | |
843 | ||
844 | /* | |
845 | * Search a snapshot context to see if the given snapshot id is | |
846 | * present. | |
847 | * | |
848 | * Returns the position of the snapshot id in the array if it's found, | |
849 | * or BAD_SNAP_INDEX otherwise. | |
850 | * | |
851 | * Note: The snapshot array is in kept sorted (by the osd) in | |
852 | * reverse order, highest snapshot id first. | |
853 | */ | |
9682fc6d AE |
854 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
855 | { | |
856 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 857 | u64 *found; |
9682fc6d | 858 | |
30d1cff8 AE |
859 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
860 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 861 | |
30d1cff8 | 862 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
863 | } |
864 | ||
2ad3d716 AE |
865 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
866 | u64 snap_id) | |
9e15b77d | 867 | { |
54cac61f | 868 | u32 which; |
9e15b77d | 869 | |
54cac61f AE |
870 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
871 | if (which == BAD_SNAP_INDEX) | |
872 | return NULL; | |
873 | ||
874 | return _rbd_dev_v1_snap_name(rbd_dev, which); | |
875 | } | |
876 | ||
877 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
878 | { | |
9e15b77d AE |
879 | if (snap_id == CEPH_NOSNAP) |
880 | return RBD_SNAP_HEAD_NAME; | |
881 | ||
54cac61f AE |
882 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
883 | if (rbd_dev->image_format == 1) | |
884 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 885 | |
54cac61f | 886 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
887 | } |
888 | ||
2ad3d716 AE |
889 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
890 | u64 *snap_size) | |
602adf40 | 891 | { |
2ad3d716 AE |
892 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
893 | if (snap_id == CEPH_NOSNAP) { | |
894 | *snap_size = rbd_dev->header.image_size; | |
895 | } else if (rbd_dev->image_format == 1) { | |
896 | u32 which; | |
602adf40 | 897 | |
2ad3d716 AE |
898 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
899 | if (which == BAD_SNAP_INDEX) | |
900 | return -ENOENT; | |
e86924a8 | 901 | |
2ad3d716 AE |
902 | *snap_size = rbd_dev->header.snap_sizes[which]; |
903 | } else { | |
904 | u64 size = 0; | |
905 | int ret; | |
906 | ||
907 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
908 | if (ret) | |
909 | return ret; | |
910 | ||
911 | *snap_size = size; | |
912 | } | |
913 | return 0; | |
602adf40 YS |
914 | } |
915 | ||
2ad3d716 AE |
916 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
917 | u64 *snap_features) | |
602adf40 | 918 | { |
2ad3d716 AE |
919 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
920 | if (snap_id == CEPH_NOSNAP) { | |
921 | *snap_features = rbd_dev->header.features; | |
922 | } else if (rbd_dev->image_format == 1) { | |
923 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 924 | } else { |
2ad3d716 AE |
925 | u64 features = 0; |
926 | int ret; | |
8b0241f8 | 927 | |
2ad3d716 AE |
928 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
929 | if (ret) | |
930 | return ret; | |
931 | ||
932 | *snap_features = features; | |
933 | } | |
934 | return 0; | |
935 | } | |
936 | ||
937 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
938 | { | |
939 | const char *snap_name = rbd_dev->spec->snap_name; | |
940 | u64 snap_id; | |
941 | u64 size = 0; | |
942 | u64 features = 0; | |
943 | int ret; | |
944 | ||
945 | if (strcmp(snap_name, RBD_SNAP_HEAD_NAME)) { | |
946 | snap_id = rbd_snap_id_by_name(rbd_dev, snap_name); | |
947 | if (snap_id == CEPH_NOSNAP) | |
8b0241f8 | 948 | return -ENOENT; |
2ad3d716 AE |
949 | } else { |
950 | snap_id = CEPH_NOSNAP; | |
602adf40 | 951 | } |
6d292906 | 952 | |
2ad3d716 AE |
953 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
954 | if (ret) | |
955 | return ret; | |
956 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
957 | if (ret) | |
958 | return ret; | |
959 | ||
960 | rbd_dev->mapping.size = size; | |
961 | rbd_dev->mapping.features = features; | |
962 | ||
963 | /* If we are mapping a snapshot it must be marked read-only */ | |
964 | ||
965 | if (snap_id != CEPH_NOSNAP) | |
966 | rbd_dev->mapping.read_only = true; | |
967 | ||
8b0241f8 | 968 | return 0; |
602adf40 YS |
969 | } |
970 | ||
d1cf5788 AE |
971 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
972 | { | |
973 | rbd_dev->mapping.size = 0; | |
974 | rbd_dev->mapping.features = 0; | |
975 | rbd_dev->mapping.read_only = true; | |
976 | } | |
977 | ||
200a6a8b AE |
978 | static void rbd_dev_clear_mapping(struct rbd_device *rbd_dev) |
979 | { | |
980 | rbd_dev->mapping.size = 0; | |
981 | rbd_dev->mapping.features = 0; | |
982 | rbd_dev->mapping.read_only = true; | |
983 | } | |
984 | ||
98571b5a | 985 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 986 | { |
65ccfe21 AE |
987 | char *name; |
988 | u64 segment; | |
989 | int ret; | |
602adf40 | 990 | |
78c2a44a | 991 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
992 | if (!name) |
993 | return NULL; | |
994 | segment = offset >> rbd_dev->header.obj_order; | |
2fd82b9e | 995 | ret = snprintf(name, MAX_OBJ_NAME_SIZE + 1, "%s.%012llx", |
65ccfe21 | 996 | rbd_dev->header.object_prefix, segment); |
2fd82b9e | 997 | if (ret < 0 || ret > MAX_OBJ_NAME_SIZE) { |
65ccfe21 AE |
998 | pr_err("error formatting segment name for #%llu (%d)\n", |
999 | segment, ret); | |
1000 | kfree(name); | |
1001 | name = NULL; | |
1002 | } | |
602adf40 | 1003 | |
65ccfe21 AE |
1004 | return name; |
1005 | } | |
602adf40 | 1006 | |
78c2a44a AE |
1007 | static void rbd_segment_name_free(const char *name) |
1008 | { | |
1009 | /* The explicit cast here is needed to drop the const qualifier */ | |
1010 | ||
1011 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1012 | } | |
1013 | ||
65ccfe21 AE |
1014 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1015 | { | |
1016 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1017 | |
65ccfe21 AE |
1018 | return offset & (segment_size - 1); |
1019 | } | |
1020 | ||
1021 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1022 | u64 offset, u64 length) | |
1023 | { | |
1024 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1025 | ||
1026 | offset &= segment_size - 1; | |
1027 | ||
aafb230e | 1028 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1029 | if (offset + length > segment_size) |
1030 | length = segment_size - offset; | |
1031 | ||
1032 | return length; | |
602adf40 YS |
1033 | } |
1034 | ||
029bcbd8 JD |
1035 | /* |
1036 | * returns the size of an object in the image | |
1037 | */ | |
1038 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1039 | { | |
1040 | return 1 << header->obj_order; | |
1041 | } | |
1042 | ||
602adf40 YS |
1043 | /* |
1044 | * bio helpers | |
1045 | */ | |
1046 | ||
1047 | static void bio_chain_put(struct bio *chain) | |
1048 | { | |
1049 | struct bio *tmp; | |
1050 | ||
1051 | while (chain) { | |
1052 | tmp = chain; | |
1053 | chain = chain->bi_next; | |
1054 | bio_put(tmp); | |
1055 | } | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * zeros a bio chain, starting at specific offset | |
1060 | */ | |
1061 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1062 | { | |
1063 | struct bio_vec *bv; | |
1064 | unsigned long flags; | |
1065 | void *buf; | |
1066 | int i; | |
1067 | int pos = 0; | |
1068 | ||
1069 | while (chain) { | |
1070 | bio_for_each_segment(bv, chain, i) { | |
1071 | if (pos + bv->bv_len > start_ofs) { | |
1072 | int remainder = max(start_ofs - pos, 0); | |
1073 | buf = bvec_kmap_irq(bv, &flags); | |
1074 | memset(buf + remainder, 0, | |
1075 | bv->bv_len - remainder); | |
85b5aaa6 | 1076 | bvec_kunmap_irq(buf, &flags); |
602adf40 YS |
1077 | } |
1078 | pos += bv->bv_len; | |
1079 | } | |
1080 | ||
1081 | chain = chain->bi_next; | |
1082 | } | |
1083 | } | |
1084 | ||
b9434c5b AE |
1085 | /* |
1086 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1087 | * starting at the given byte offset from the start of the array and | |
1088 | * continuing up to the given end offset. The pages array is | |
1089 | * assumed to be big enough to hold all bytes up to the end. | |
1090 | */ | |
1091 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1092 | { | |
1093 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1094 | ||
1095 | rbd_assert(end > offset); | |
1096 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1097 | while (offset < end) { | |
1098 | size_t page_offset; | |
1099 | size_t length; | |
1100 | unsigned long flags; | |
1101 | void *kaddr; | |
1102 | ||
1103 | page_offset = (size_t)(offset & ~PAGE_MASK); | |
1104 | length = min(PAGE_SIZE - page_offset, (size_t)(end - offset)); | |
1105 | local_irq_save(flags); | |
1106 | kaddr = kmap_atomic(*page); | |
1107 | memset(kaddr + page_offset, 0, length); | |
1108 | kunmap_atomic(kaddr); | |
1109 | local_irq_restore(flags); | |
1110 | ||
1111 | offset += length; | |
1112 | page++; | |
1113 | } | |
1114 | } | |
1115 | ||
602adf40 | 1116 | /* |
f7760dad AE |
1117 | * Clone a portion of a bio, starting at the given byte offset |
1118 | * and continuing for the number of bytes indicated. | |
602adf40 | 1119 | */ |
f7760dad AE |
1120 | static struct bio *bio_clone_range(struct bio *bio_src, |
1121 | unsigned int offset, | |
1122 | unsigned int len, | |
1123 | gfp_t gfpmask) | |
602adf40 | 1124 | { |
f7760dad AE |
1125 | struct bio_vec *bv; |
1126 | unsigned int resid; | |
1127 | unsigned short idx; | |
1128 | unsigned int voff; | |
1129 | unsigned short end_idx; | |
1130 | unsigned short vcnt; | |
1131 | struct bio *bio; | |
1132 | ||
1133 | /* Handle the easy case for the caller */ | |
1134 | ||
1135 | if (!offset && len == bio_src->bi_size) | |
1136 | return bio_clone(bio_src, gfpmask); | |
1137 | ||
1138 | if (WARN_ON_ONCE(!len)) | |
1139 | return NULL; | |
1140 | if (WARN_ON_ONCE(len > bio_src->bi_size)) | |
1141 | return NULL; | |
1142 | if (WARN_ON_ONCE(offset > bio_src->bi_size - len)) | |
1143 | return NULL; | |
1144 | ||
1145 | /* Find first affected segment... */ | |
1146 | ||
1147 | resid = offset; | |
1148 | __bio_for_each_segment(bv, bio_src, idx, 0) { | |
1149 | if (resid < bv->bv_len) | |
1150 | break; | |
1151 | resid -= bv->bv_len; | |
602adf40 | 1152 | } |
f7760dad | 1153 | voff = resid; |
602adf40 | 1154 | |
f7760dad | 1155 | /* ...and the last affected segment */ |
602adf40 | 1156 | |
f7760dad AE |
1157 | resid += len; |
1158 | __bio_for_each_segment(bv, bio_src, end_idx, idx) { | |
1159 | if (resid <= bv->bv_len) | |
1160 | break; | |
1161 | resid -= bv->bv_len; | |
1162 | } | |
1163 | vcnt = end_idx - idx + 1; | |
1164 | ||
1165 | /* Build the clone */ | |
1166 | ||
1167 | bio = bio_alloc(gfpmask, (unsigned int) vcnt); | |
1168 | if (!bio) | |
1169 | return NULL; /* ENOMEM */ | |
602adf40 | 1170 | |
f7760dad AE |
1171 | bio->bi_bdev = bio_src->bi_bdev; |
1172 | bio->bi_sector = bio_src->bi_sector + (offset >> SECTOR_SHIFT); | |
1173 | bio->bi_rw = bio_src->bi_rw; | |
1174 | bio->bi_flags |= 1 << BIO_CLONED; | |
1175 | ||
1176 | /* | |
1177 | * Copy over our part of the bio_vec, then update the first | |
1178 | * and last (or only) entries. | |
1179 | */ | |
1180 | memcpy(&bio->bi_io_vec[0], &bio_src->bi_io_vec[idx], | |
1181 | vcnt * sizeof (struct bio_vec)); | |
1182 | bio->bi_io_vec[0].bv_offset += voff; | |
1183 | if (vcnt > 1) { | |
1184 | bio->bi_io_vec[0].bv_len -= voff; | |
1185 | bio->bi_io_vec[vcnt - 1].bv_len = resid; | |
1186 | } else { | |
1187 | bio->bi_io_vec[0].bv_len = len; | |
602adf40 YS |
1188 | } |
1189 | ||
f7760dad AE |
1190 | bio->bi_vcnt = vcnt; |
1191 | bio->bi_size = len; | |
1192 | bio->bi_idx = 0; | |
1193 | ||
1194 | return bio; | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Clone a portion of a bio chain, starting at the given byte offset | |
1199 | * into the first bio in the source chain and continuing for the | |
1200 | * number of bytes indicated. The result is another bio chain of | |
1201 | * exactly the given length, or a null pointer on error. | |
1202 | * | |
1203 | * The bio_src and offset parameters are both in-out. On entry they | |
1204 | * refer to the first source bio and the offset into that bio where | |
1205 | * the start of data to be cloned is located. | |
1206 | * | |
1207 | * On return, bio_src is updated to refer to the bio in the source | |
1208 | * chain that contains first un-cloned byte, and *offset will | |
1209 | * contain the offset of that byte within that bio. | |
1210 | */ | |
1211 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1212 | unsigned int *offset, | |
1213 | unsigned int len, | |
1214 | gfp_t gfpmask) | |
1215 | { | |
1216 | struct bio *bi = *bio_src; | |
1217 | unsigned int off = *offset; | |
1218 | struct bio *chain = NULL; | |
1219 | struct bio **end; | |
1220 | ||
1221 | /* Build up a chain of clone bios up to the limit */ | |
1222 | ||
1223 | if (!bi || off >= bi->bi_size || !len) | |
1224 | return NULL; /* Nothing to clone */ | |
602adf40 | 1225 | |
f7760dad AE |
1226 | end = &chain; |
1227 | while (len) { | |
1228 | unsigned int bi_size; | |
1229 | struct bio *bio; | |
1230 | ||
f5400b7a AE |
1231 | if (!bi) { |
1232 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1233 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1234 | } |
f7760dad AE |
1235 | bi_size = min_t(unsigned int, bi->bi_size - off, len); |
1236 | bio = bio_clone_range(bi, off, bi_size, gfpmask); | |
1237 | if (!bio) | |
1238 | goto out_err; /* ENOMEM */ | |
1239 | ||
1240 | *end = bio; | |
1241 | end = &bio->bi_next; | |
602adf40 | 1242 | |
f7760dad AE |
1243 | off += bi_size; |
1244 | if (off == bi->bi_size) { | |
1245 | bi = bi->bi_next; | |
1246 | off = 0; | |
1247 | } | |
1248 | len -= bi_size; | |
1249 | } | |
1250 | *bio_src = bi; | |
1251 | *offset = off; | |
1252 | ||
1253 | return chain; | |
1254 | out_err: | |
1255 | bio_chain_put(chain); | |
602adf40 | 1256 | |
602adf40 YS |
1257 | return NULL; |
1258 | } | |
1259 | ||
926f9b3f AE |
1260 | /* |
1261 | * The default/initial value for all object request flags is 0. For | |
1262 | * each flag, once its value is set to 1 it is never reset to 0 | |
1263 | * again. | |
1264 | */ | |
57acbaa7 | 1265 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1266 | { |
57acbaa7 | 1267 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1268 | struct rbd_device *rbd_dev; |
1269 | ||
57acbaa7 AE |
1270 | rbd_dev = obj_request->img_request->rbd_dev; |
1271 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1272 | obj_request); |
1273 | } | |
1274 | } | |
1275 | ||
57acbaa7 | 1276 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1277 | { |
1278 | smp_mb(); | |
57acbaa7 | 1279 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1280 | } |
1281 | ||
57acbaa7 | 1282 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1283 | { |
57acbaa7 AE |
1284 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1285 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1286 | |
57acbaa7 AE |
1287 | if (obj_request_img_data_test(obj_request)) |
1288 | rbd_dev = obj_request->img_request->rbd_dev; | |
1289 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1290 | obj_request); |
1291 | } | |
1292 | } | |
1293 | ||
57acbaa7 | 1294 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1295 | { |
1296 | smp_mb(); | |
57acbaa7 | 1297 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1298 | } |
1299 | ||
5679c59f AE |
1300 | /* |
1301 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1302 | * flag. The latter is set based on the "exists" value provided. | |
1303 | * | |
1304 | * Note that for our purposes once an object exists it never goes | |
1305 | * away again. It's possible that the response from two existence | |
1306 | * checks are separated by the creation of the target object, and | |
1307 | * the first ("doesn't exist") response arrives *after* the second | |
1308 | * ("does exist"). In that case we ignore the second one. | |
1309 | */ | |
1310 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1311 | bool exists) | |
1312 | { | |
1313 | if (exists) | |
1314 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1315 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1316 | smp_mb(); | |
1317 | } | |
1318 | ||
1319 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1320 | { | |
1321 | smp_mb(); | |
1322 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1323 | } | |
1324 | ||
1325 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1326 | { | |
1327 | smp_mb(); | |
1328 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1329 | } | |
1330 | ||
bf0d5f50 AE |
1331 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1332 | { | |
37206ee5 AE |
1333 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1334 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1335 | kref_get(&obj_request->kref); |
1336 | } | |
1337 | ||
1338 | static void rbd_obj_request_destroy(struct kref *kref); | |
1339 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1340 | { | |
1341 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1342 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1343 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1344 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1345 | } | |
1346 | ||
1347 | static void rbd_img_request_get(struct rbd_img_request *img_request) | |
1348 | { | |
37206ee5 AE |
1349 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1350 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1351 | kref_get(&img_request->kref); |
1352 | } | |
1353 | ||
1354 | static void rbd_img_request_destroy(struct kref *kref); | |
1355 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1356 | { | |
1357 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1358 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1359 | atomic_read(&img_request->kref.refcount)); | |
bf0d5f50 AE |
1360 | kref_put(&img_request->kref, rbd_img_request_destroy); |
1361 | } | |
1362 | ||
1363 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1364 | struct rbd_obj_request *obj_request) | |
1365 | { | |
25dcf954 AE |
1366 | rbd_assert(obj_request->img_request == NULL); |
1367 | ||
b155e86c | 1368 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1369 | obj_request->img_request = img_request; |
25dcf954 | 1370 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1371 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1372 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1373 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1374 | img_request->obj_request_count++; |
1375 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1376 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1377 | obj_request->which); | |
bf0d5f50 AE |
1378 | } |
1379 | ||
1380 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1381 | struct rbd_obj_request *obj_request) | |
1382 | { | |
1383 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1384 | |
37206ee5 AE |
1385 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1386 | obj_request->which); | |
bf0d5f50 | 1387 | list_del(&obj_request->links); |
25dcf954 AE |
1388 | rbd_assert(img_request->obj_request_count > 0); |
1389 | img_request->obj_request_count--; | |
1390 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1391 | obj_request->which = BAD_WHICH; | |
6365d33a | 1392 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1393 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1394 | obj_request->img_request = NULL; |
25dcf954 | 1395 | obj_request->callback = NULL; |
bf0d5f50 AE |
1396 | rbd_obj_request_put(obj_request); |
1397 | } | |
1398 | ||
1399 | static bool obj_request_type_valid(enum obj_request_type type) | |
1400 | { | |
1401 | switch (type) { | |
9969ebc5 | 1402 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1403 | case OBJ_REQUEST_BIO: |
788e2df3 | 1404 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1405 | return true; |
1406 | default: | |
1407 | return false; | |
1408 | } | |
1409 | } | |
1410 | ||
bf0d5f50 AE |
1411 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1412 | struct rbd_obj_request *obj_request) | |
1413 | { | |
37206ee5 AE |
1414 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1415 | ||
bf0d5f50 AE |
1416 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1417 | } | |
1418 | ||
1419 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1420 | { | |
55f27e09 | 1421 | |
37206ee5 | 1422 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1423 | |
1424 | /* | |
1425 | * If no error occurred, compute the aggregate transfer | |
1426 | * count for the image request. We could instead use | |
1427 | * atomic64_cmpxchg() to update it as each object request | |
1428 | * completes; not clear which way is better off hand. | |
1429 | */ | |
1430 | if (!img_request->result) { | |
1431 | struct rbd_obj_request *obj_request; | |
1432 | u64 xferred = 0; | |
1433 | ||
1434 | for_each_obj_request(img_request, obj_request) | |
1435 | xferred += obj_request->xferred; | |
1436 | img_request->xferred = xferred; | |
1437 | } | |
1438 | ||
bf0d5f50 AE |
1439 | if (img_request->callback) |
1440 | img_request->callback(img_request); | |
1441 | else | |
1442 | rbd_img_request_put(img_request); | |
1443 | } | |
1444 | ||
788e2df3 AE |
1445 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1446 | ||
1447 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1448 | { | |
37206ee5 AE |
1449 | dout("%s: obj %p\n", __func__, obj_request); |
1450 | ||
788e2df3 AE |
1451 | return wait_for_completion_interruptible(&obj_request->completion); |
1452 | } | |
1453 | ||
0c425248 AE |
1454 | /* |
1455 | * The default/initial value for all image request flags is 0. Each | |
1456 | * is conditionally set to 1 at image request initialization time | |
1457 | * and currently never change thereafter. | |
1458 | */ | |
1459 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1460 | { | |
1461 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1462 | smp_mb(); | |
1463 | } | |
1464 | ||
1465 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1466 | { | |
1467 | smp_mb(); | |
1468 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1469 | } | |
1470 | ||
9849e986 AE |
1471 | static void img_request_child_set(struct rbd_img_request *img_request) |
1472 | { | |
1473 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1474 | smp_mb(); | |
1475 | } | |
1476 | ||
1477 | static bool img_request_child_test(struct rbd_img_request *img_request) | |
1478 | { | |
1479 | smp_mb(); | |
1480 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1481 | } | |
1482 | ||
d0b2e944 AE |
1483 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1484 | { | |
1485 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1486 | smp_mb(); | |
1487 | } | |
1488 | ||
1489 | static bool img_request_layered_test(struct rbd_img_request *img_request) | |
1490 | { | |
1491 | smp_mb(); | |
1492 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1493 | } | |
1494 | ||
6e2a4505 AE |
1495 | static void |
1496 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1497 | { | |
b9434c5b AE |
1498 | u64 xferred = obj_request->xferred; |
1499 | u64 length = obj_request->length; | |
1500 | ||
6e2a4505 AE |
1501 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1502 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1503 | xferred, length); |
6e2a4505 AE |
1504 | /* |
1505 | * ENOENT means a hole in the image. We zero-fill the | |
1506 | * entire length of the request. A short read also implies | |
1507 | * zero-fill to the end of the request. Either way we | |
1508 | * update the xferred count to indicate the whole request | |
1509 | * was satisfied. | |
1510 | */ | |
b9434c5b | 1511 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1512 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1513 | if (obj_request->type == OBJ_REQUEST_BIO) |
1514 | zero_bio_chain(obj_request->bio_list, 0); | |
1515 | else | |
1516 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1517 | obj_request->result = 0; |
b9434c5b AE |
1518 | obj_request->xferred = length; |
1519 | } else if (xferred < length && !obj_request->result) { | |
1520 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1521 | zero_bio_chain(obj_request->bio_list, xferred); | |
1522 | else | |
1523 | zero_pages(obj_request->pages, xferred, length); | |
1524 | obj_request->xferred = length; | |
6e2a4505 AE |
1525 | } |
1526 | obj_request_done_set(obj_request); | |
1527 | } | |
1528 | ||
bf0d5f50 AE |
1529 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1530 | { | |
37206ee5 AE |
1531 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1532 | obj_request->callback); | |
bf0d5f50 AE |
1533 | if (obj_request->callback) |
1534 | obj_request->callback(obj_request); | |
788e2df3 AE |
1535 | else |
1536 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1537 | } |
1538 | ||
c47f9371 | 1539 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1540 | { |
1541 | dout("%s: obj %p\n", __func__, obj_request); | |
1542 | obj_request_done_set(obj_request); | |
1543 | } | |
1544 | ||
c47f9371 | 1545 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1546 | { |
57acbaa7 | 1547 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1548 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1549 | bool layered = false; |
1550 | ||
1551 | if (obj_request_img_data_test(obj_request)) { | |
1552 | img_request = obj_request->img_request; | |
1553 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1554 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1555 | } |
8b3e1a56 AE |
1556 | |
1557 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1558 | obj_request, img_request, obj_request->result, | |
1559 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1560 | if (layered && obj_request->result == -ENOENT && |
1561 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1562 | rbd_img_parent_read(obj_request); |
1563 | else if (img_request) | |
6e2a4505 AE |
1564 | rbd_img_obj_request_read_callback(obj_request); |
1565 | else | |
1566 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1567 | } |
1568 | ||
c47f9371 | 1569 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1570 | { |
1b83bef2 SW |
1571 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1572 | obj_request->result, obj_request->length); | |
1573 | /* | |
8b3e1a56 AE |
1574 | * There is no such thing as a successful short write. Set |
1575 | * it to our originally-requested length. | |
1b83bef2 SW |
1576 | */ |
1577 | obj_request->xferred = obj_request->length; | |
07741308 | 1578 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1579 | } |
1580 | ||
fbfab539 AE |
1581 | /* |
1582 | * For a simple stat call there's nothing to do. We'll do more if | |
1583 | * this is part of a write sequence for a layered image. | |
1584 | */ | |
c47f9371 | 1585 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1586 | { |
37206ee5 | 1587 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1588 | obj_request_done_set(obj_request); |
1589 | } | |
1590 | ||
bf0d5f50 AE |
1591 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1592 | struct ceph_msg *msg) | |
1593 | { | |
1594 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1595 | u16 opcode; |
1596 | ||
37206ee5 | 1597 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1598 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1599 | if (obj_request_img_data_test(obj_request)) { |
1600 | rbd_assert(obj_request->img_request); | |
1601 | rbd_assert(obj_request->which != BAD_WHICH); | |
1602 | } else { | |
1603 | rbd_assert(obj_request->which == BAD_WHICH); | |
1604 | } | |
bf0d5f50 | 1605 | |
1b83bef2 SW |
1606 | if (osd_req->r_result < 0) |
1607 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1608 | |
0eefd470 | 1609 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1610 | |
c47f9371 AE |
1611 | /* |
1612 | * We support a 64-bit length, but ultimately it has to be | |
1613 | * passed to blk_end_request(), which takes an unsigned int. | |
1614 | */ | |
1b83bef2 | 1615 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1616 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1617 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1618 | switch (opcode) { |
1619 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1620 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1621 | break; |
1622 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1623 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1624 | break; |
fbfab539 | 1625 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1626 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1627 | break; |
36be9a76 | 1628 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1629 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1630 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1631 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1632 | break; |
bf0d5f50 AE |
1633 | default: |
1634 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1635 | obj_request->object_name, (unsigned short) opcode); | |
1636 | break; | |
1637 | } | |
1638 | ||
07741308 | 1639 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1640 | rbd_obj_request_complete(obj_request); |
1641 | } | |
1642 | ||
9d4df01f | 1643 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1644 | { |
1645 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1646 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1647 | u64 snap_id; |
430c28c3 | 1648 | |
8c042b0d | 1649 | rbd_assert(osd_req != NULL); |
430c28c3 | 1650 | |
9d4df01f | 1651 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1652 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1653 | NULL, snap_id, NULL); |
1654 | } | |
1655 | ||
1656 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1657 | { | |
1658 | struct rbd_img_request *img_request = obj_request->img_request; | |
1659 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1660 | struct ceph_snap_context *snapc; | |
1661 | struct timespec mtime = CURRENT_TIME; | |
1662 | ||
1663 | rbd_assert(osd_req != NULL); | |
1664 | ||
1665 | snapc = img_request ? img_request->snapc : NULL; | |
1666 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1667 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1668 | } |
1669 | ||
bf0d5f50 AE |
1670 | static struct ceph_osd_request *rbd_osd_req_create( |
1671 | struct rbd_device *rbd_dev, | |
1672 | bool write_request, | |
430c28c3 | 1673 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1674 | { |
bf0d5f50 AE |
1675 | struct ceph_snap_context *snapc = NULL; |
1676 | struct ceph_osd_client *osdc; | |
1677 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1678 | |
6365d33a AE |
1679 | if (obj_request_img_data_test(obj_request)) { |
1680 | struct rbd_img_request *img_request = obj_request->img_request; | |
1681 | ||
0c425248 AE |
1682 | rbd_assert(write_request == |
1683 | img_request_write_test(img_request)); | |
1684 | if (write_request) | |
bf0d5f50 | 1685 | snapc = img_request->snapc; |
bf0d5f50 AE |
1686 | } |
1687 | ||
1688 | /* Allocate and initialize the request, for the single op */ | |
1689 | ||
1690 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1691 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1692 | if (!osd_req) | |
1693 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1694 | |
430c28c3 | 1695 | if (write_request) |
bf0d5f50 | 1696 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1697 | else |
bf0d5f50 | 1698 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1699 | |
1700 | osd_req->r_callback = rbd_osd_req_callback; | |
1701 | osd_req->r_priv = obj_request; | |
1702 | ||
1703 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1704 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1705 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1706 | ||
1707 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1708 | ||
bf0d5f50 AE |
1709 | return osd_req; |
1710 | } | |
1711 | ||
0eefd470 AE |
1712 | /* |
1713 | * Create a copyup osd request based on the information in the | |
1714 | * object request supplied. A copyup request has two osd ops, | |
1715 | * a copyup method call, and a "normal" write request. | |
1716 | */ | |
1717 | static struct ceph_osd_request * | |
1718 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1719 | { | |
1720 | struct rbd_img_request *img_request; | |
1721 | struct ceph_snap_context *snapc; | |
1722 | struct rbd_device *rbd_dev; | |
1723 | struct ceph_osd_client *osdc; | |
1724 | struct ceph_osd_request *osd_req; | |
1725 | ||
1726 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1727 | img_request = obj_request->img_request; | |
1728 | rbd_assert(img_request); | |
1729 | rbd_assert(img_request_write_test(img_request)); | |
1730 | ||
1731 | /* Allocate and initialize the request, for the two ops */ | |
1732 | ||
1733 | snapc = img_request->snapc; | |
1734 | rbd_dev = img_request->rbd_dev; | |
1735 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1736 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1737 | if (!osd_req) | |
1738 | return NULL; /* ENOMEM */ | |
1739 | ||
1740 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1741 | osd_req->r_callback = rbd_osd_req_callback; | |
1742 | osd_req->r_priv = obj_request; | |
1743 | ||
1744 | osd_req->r_oid_len = strlen(obj_request->object_name); | |
1745 | rbd_assert(osd_req->r_oid_len < sizeof (osd_req->r_oid)); | |
1746 | memcpy(osd_req->r_oid, obj_request->object_name, osd_req->r_oid_len); | |
1747 | ||
1748 | osd_req->r_file_layout = rbd_dev->layout; /* struct */ | |
1749 | ||
1750 | return osd_req; | |
1751 | } | |
1752 | ||
1753 | ||
bf0d5f50 AE |
1754 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1755 | { | |
1756 | ceph_osdc_put_request(osd_req); | |
1757 | } | |
1758 | ||
1759 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1760 | ||
1761 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1762 | u64 offset, u64 length, | |
1763 | enum obj_request_type type) | |
1764 | { | |
1765 | struct rbd_obj_request *obj_request; | |
1766 | size_t size; | |
1767 | char *name; | |
1768 | ||
1769 | rbd_assert(obj_request_type_valid(type)); | |
1770 | ||
1771 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1772 | name = kmalloc(size, GFP_KERNEL); |
1773 | if (!name) | |
bf0d5f50 AE |
1774 | return NULL; |
1775 | ||
868311b1 | 1776 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1777 | if (!obj_request) { |
1778 | kfree(name); | |
1779 | return NULL; | |
1780 | } | |
1781 | ||
bf0d5f50 AE |
1782 | obj_request->object_name = memcpy(name, object_name, size); |
1783 | obj_request->offset = offset; | |
1784 | obj_request->length = length; | |
926f9b3f | 1785 | obj_request->flags = 0; |
bf0d5f50 AE |
1786 | obj_request->which = BAD_WHICH; |
1787 | obj_request->type = type; | |
1788 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1789 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1790 | kref_init(&obj_request->kref); |
1791 | ||
37206ee5 AE |
1792 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1793 | offset, length, (int)type, obj_request); | |
1794 | ||
bf0d5f50 AE |
1795 | return obj_request; |
1796 | } | |
1797 | ||
1798 | static void rbd_obj_request_destroy(struct kref *kref) | |
1799 | { | |
1800 | struct rbd_obj_request *obj_request; | |
1801 | ||
1802 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1803 | ||
37206ee5 AE |
1804 | dout("%s: obj %p\n", __func__, obj_request); |
1805 | ||
bf0d5f50 AE |
1806 | rbd_assert(obj_request->img_request == NULL); |
1807 | rbd_assert(obj_request->which == BAD_WHICH); | |
1808 | ||
1809 | if (obj_request->osd_req) | |
1810 | rbd_osd_req_destroy(obj_request->osd_req); | |
1811 | ||
1812 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1813 | switch (obj_request->type) { | |
9969ebc5 AE |
1814 | case OBJ_REQUEST_NODATA: |
1815 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1816 | case OBJ_REQUEST_BIO: |
1817 | if (obj_request->bio_list) | |
1818 | bio_chain_put(obj_request->bio_list); | |
1819 | break; | |
788e2df3 AE |
1820 | case OBJ_REQUEST_PAGES: |
1821 | if (obj_request->pages) | |
1822 | ceph_release_page_vector(obj_request->pages, | |
1823 | obj_request->page_count); | |
1824 | break; | |
bf0d5f50 AE |
1825 | } |
1826 | ||
f907ad55 | 1827 | kfree(obj_request->object_name); |
868311b1 AE |
1828 | obj_request->object_name = NULL; |
1829 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
1830 | } |
1831 | ||
1832 | /* | |
1833 | * Caller is responsible for filling in the list of object requests | |
1834 | * that comprises the image request, and the Linux request pointer | |
1835 | * (if there is one). | |
1836 | */ | |
cc344fa1 AE |
1837 | static struct rbd_img_request *rbd_img_request_create( |
1838 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1839 | u64 offset, u64 length, |
9849e986 AE |
1840 | bool write_request, |
1841 | bool child_request) | |
bf0d5f50 AE |
1842 | { |
1843 | struct rbd_img_request *img_request; | |
bf0d5f50 | 1844 | |
1c2a9dfe | 1845 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC); |
bf0d5f50 AE |
1846 | if (!img_request) |
1847 | return NULL; | |
1848 | ||
1849 | if (write_request) { | |
1850 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 1851 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 1852 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1853 | } |
1854 | ||
1855 | img_request->rq = NULL; | |
1856 | img_request->rbd_dev = rbd_dev; | |
1857 | img_request->offset = offset; | |
1858 | img_request->length = length; | |
0c425248 AE |
1859 | img_request->flags = 0; |
1860 | if (write_request) { | |
1861 | img_request_write_set(img_request); | |
468521c1 | 1862 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1863 | } else { |
bf0d5f50 | 1864 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1865 | } |
9849e986 AE |
1866 | if (child_request) |
1867 | img_request_child_set(img_request); | |
d0b2e944 AE |
1868 | if (rbd_dev->parent_spec) |
1869 | img_request_layered_set(img_request); | |
bf0d5f50 AE |
1870 | spin_lock_init(&img_request->completion_lock); |
1871 | img_request->next_completion = 0; | |
1872 | img_request->callback = NULL; | |
a5a337d4 | 1873 | img_request->result = 0; |
bf0d5f50 AE |
1874 | img_request->obj_request_count = 0; |
1875 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1876 | kref_init(&img_request->kref); | |
1877 | ||
1878 | rbd_img_request_get(img_request); /* Avoid a warning */ | |
1879 | rbd_img_request_put(img_request); /* TEMPORARY */ | |
1880 | ||
37206ee5 AE |
1881 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1882 | write_request ? "write" : "read", offset, length, | |
1883 | img_request); | |
1884 | ||
bf0d5f50 AE |
1885 | return img_request; |
1886 | } | |
1887 | ||
1888 | static void rbd_img_request_destroy(struct kref *kref) | |
1889 | { | |
1890 | struct rbd_img_request *img_request; | |
1891 | struct rbd_obj_request *obj_request; | |
1892 | struct rbd_obj_request *next_obj_request; | |
1893 | ||
1894 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1895 | ||
37206ee5 AE |
1896 | dout("%s: img %p\n", __func__, img_request); |
1897 | ||
bf0d5f50 AE |
1898 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
1899 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 1900 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 1901 | |
0c425248 | 1902 | if (img_request_write_test(img_request)) |
812164f8 | 1903 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 1904 | |
8b3e1a56 AE |
1905 | if (img_request_child_test(img_request)) |
1906 | rbd_obj_request_put(img_request->obj_request); | |
1907 | ||
1c2a9dfe | 1908 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
1909 | } |
1910 | ||
1217857f AE |
1911 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
1912 | { | |
6365d33a | 1913 | struct rbd_img_request *img_request; |
1217857f AE |
1914 | unsigned int xferred; |
1915 | int result; | |
8b3e1a56 | 1916 | bool more; |
1217857f | 1917 | |
6365d33a AE |
1918 | rbd_assert(obj_request_img_data_test(obj_request)); |
1919 | img_request = obj_request->img_request; | |
1920 | ||
1217857f AE |
1921 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
1922 | xferred = (unsigned int)obj_request->xferred; | |
1923 | result = obj_request->result; | |
1924 | if (result) { | |
1925 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
1926 | ||
1927 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
1928 | img_request_write_test(img_request) ? "write" : "read", | |
1929 | obj_request->length, obj_request->img_offset, | |
1930 | obj_request->offset); | |
1931 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
1932 | result, xferred); | |
1933 | if (!img_request->result) | |
1934 | img_request->result = result; | |
1935 | } | |
1936 | ||
f1a4739f AE |
1937 | /* Image object requests don't own their page array */ |
1938 | ||
1939 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
1940 | obj_request->pages = NULL; | |
1941 | obj_request->page_count = 0; | |
1942 | } | |
1943 | ||
8b3e1a56 AE |
1944 | if (img_request_child_test(img_request)) { |
1945 | rbd_assert(img_request->obj_request != NULL); | |
1946 | more = obj_request->which < img_request->obj_request_count - 1; | |
1947 | } else { | |
1948 | rbd_assert(img_request->rq != NULL); | |
1949 | more = blk_end_request(img_request->rq, result, xferred); | |
1950 | } | |
1951 | ||
1952 | return more; | |
1217857f AE |
1953 | } |
1954 | ||
2169238d AE |
1955 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
1956 | { | |
1957 | struct rbd_img_request *img_request; | |
1958 | u32 which = obj_request->which; | |
1959 | bool more = true; | |
1960 | ||
6365d33a | 1961 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
1962 | img_request = obj_request->img_request; |
1963 | ||
1964 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
1965 | rbd_assert(img_request != NULL); | |
2169238d AE |
1966 | rbd_assert(img_request->obj_request_count > 0); |
1967 | rbd_assert(which != BAD_WHICH); | |
1968 | rbd_assert(which < img_request->obj_request_count); | |
1969 | rbd_assert(which >= img_request->next_completion); | |
1970 | ||
1971 | spin_lock_irq(&img_request->completion_lock); | |
1972 | if (which != img_request->next_completion) | |
1973 | goto out; | |
1974 | ||
1975 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
1976 | rbd_assert(more); |
1977 | rbd_assert(which < img_request->obj_request_count); | |
1978 | ||
1979 | if (!obj_request_done_test(obj_request)) | |
1980 | break; | |
1217857f | 1981 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
1982 | which++; |
1983 | } | |
1984 | ||
1985 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
1986 | img_request->next_completion = which; | |
1987 | out: | |
1988 | spin_unlock_irq(&img_request->completion_lock); | |
1989 | ||
1990 | if (!more) | |
1991 | rbd_img_request_complete(img_request); | |
1992 | } | |
1993 | ||
f1a4739f AE |
1994 | /* |
1995 | * Split up an image request into one or more object requests, each | |
1996 | * to a different object. The "type" parameter indicates whether | |
1997 | * "data_desc" is the pointer to the head of a list of bio | |
1998 | * structures, or the base of a page array. In either case this | |
1999 | * function assumes data_desc describes memory sufficient to hold | |
2000 | * all data described by the image request. | |
2001 | */ | |
2002 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2003 | enum obj_request_type type, | |
2004 | void *data_desc) | |
bf0d5f50 AE |
2005 | { |
2006 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2007 | struct rbd_obj_request *obj_request = NULL; | |
2008 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 2009 | bool write_request = img_request_write_test(img_request); |
f1a4739f AE |
2010 | struct bio *bio_list; |
2011 | unsigned int bio_offset = 0; | |
2012 | struct page **pages; | |
7da22d29 | 2013 | u64 img_offset; |
bf0d5f50 AE |
2014 | u64 resid; |
2015 | u16 opcode; | |
2016 | ||
f1a4739f AE |
2017 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2018 | (int)type, data_desc); | |
37206ee5 | 2019 | |
430c28c3 | 2020 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 2021 | img_offset = img_request->offset; |
bf0d5f50 | 2022 | resid = img_request->length; |
4dda41d3 | 2023 | rbd_assert(resid > 0); |
f1a4739f AE |
2024 | |
2025 | if (type == OBJ_REQUEST_BIO) { | |
2026 | bio_list = data_desc; | |
2027 | rbd_assert(img_offset == bio_list->bi_sector << SECTOR_SHIFT); | |
2028 | } else { | |
2029 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
2030 | pages = data_desc; | |
2031 | } | |
2032 | ||
bf0d5f50 | 2033 | while (resid) { |
2fa12320 | 2034 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2035 | const char *object_name; |
bf0d5f50 AE |
2036 | u64 offset; |
2037 | u64 length; | |
2038 | ||
7da22d29 | 2039 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2040 | if (!object_name) |
2041 | goto out_unwind; | |
7da22d29 AE |
2042 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2043 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2044 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2045 | offset, length, type); |
78c2a44a AE |
2046 | /* object request has its own copy of the object name */ |
2047 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2048 | if (!obj_request) |
2049 | goto out_unwind; | |
2050 | ||
f1a4739f AE |
2051 | if (type == OBJ_REQUEST_BIO) { |
2052 | unsigned int clone_size; | |
2053 | ||
2054 | rbd_assert(length <= (u64)UINT_MAX); | |
2055 | clone_size = (unsigned int)length; | |
2056 | obj_request->bio_list = | |
2057 | bio_chain_clone_range(&bio_list, | |
2058 | &bio_offset, | |
2059 | clone_size, | |
2060 | GFP_ATOMIC); | |
2061 | if (!obj_request->bio_list) | |
2062 | goto out_partial; | |
2063 | } else { | |
2064 | unsigned int page_count; | |
2065 | ||
2066 | obj_request->pages = pages; | |
2067 | page_count = (u32)calc_pages_for(offset, length); | |
2068 | obj_request->page_count = page_count; | |
2069 | if ((offset + length) & ~PAGE_MASK) | |
2070 | page_count--; /* more on last page */ | |
2071 | pages += page_count; | |
2072 | } | |
bf0d5f50 | 2073 | |
2fa12320 AE |
2074 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
2075 | obj_request); | |
2076 | if (!osd_req) | |
bf0d5f50 | 2077 | goto out_partial; |
2fa12320 | 2078 | obj_request->osd_req = osd_req; |
2169238d | 2079 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 2080 | |
2fa12320 AE |
2081 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
2082 | 0, 0); | |
f1a4739f AE |
2083 | if (type == OBJ_REQUEST_BIO) |
2084 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
2085 | obj_request->bio_list, length); | |
2086 | else | |
2087 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
2088 | obj_request->pages, length, | |
2089 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
2090 | |
2091 | if (write_request) | |
2092 | rbd_osd_req_format_write(obj_request); | |
2093 | else | |
2094 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2095 | |
7da22d29 | 2096 | obj_request->img_offset = img_offset; |
bf0d5f50 AE |
2097 | rbd_img_obj_request_add(img_request, obj_request); |
2098 | ||
7da22d29 | 2099 | img_offset += length; |
bf0d5f50 AE |
2100 | resid -= length; |
2101 | } | |
2102 | ||
2103 | return 0; | |
2104 | ||
2105 | out_partial: | |
2106 | rbd_obj_request_put(obj_request); | |
2107 | out_unwind: | |
2108 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
2109 | rbd_obj_request_put(obj_request); | |
2110 | ||
2111 | return -ENOMEM; | |
2112 | } | |
2113 | ||
0eefd470 AE |
2114 | static void |
2115 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2116 | { | |
2117 | struct rbd_img_request *img_request; | |
2118 | struct rbd_device *rbd_dev; | |
2119 | u64 length; | |
2120 | u32 page_count; | |
2121 | ||
2122 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2123 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2124 | img_request = obj_request->img_request; | |
2125 | rbd_assert(img_request); | |
2126 | ||
2127 | rbd_dev = img_request->rbd_dev; | |
2128 | rbd_assert(rbd_dev); | |
2129 | length = (u64)1 << rbd_dev->header.obj_order; | |
2130 | page_count = (u32)calc_pages_for(0, length); | |
2131 | ||
2132 | rbd_assert(obj_request->copyup_pages); | |
2133 | ceph_release_page_vector(obj_request->copyup_pages, page_count); | |
2134 | obj_request->copyup_pages = NULL; | |
2135 | ||
2136 | /* | |
2137 | * We want the transfer count to reflect the size of the | |
2138 | * original write request. There is no such thing as a | |
2139 | * successful short write, so if the request was successful | |
2140 | * we can just set it to the originally-requested length. | |
2141 | */ | |
2142 | if (!obj_request->result) | |
2143 | obj_request->xferred = obj_request->length; | |
2144 | ||
2145 | /* Finish up with the normal image object callback */ | |
2146 | ||
2147 | rbd_img_obj_callback(obj_request); | |
2148 | } | |
2149 | ||
3d7efd18 AE |
2150 | static void |
2151 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2152 | { | |
2153 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2154 | struct ceph_osd_request *osd_req; |
2155 | struct ceph_osd_client *osdc; | |
2156 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2157 | struct page **pages; |
3d7efd18 AE |
2158 | int result; |
2159 | u64 obj_size; | |
2160 | u64 xferred; | |
2161 | ||
2162 | rbd_assert(img_request_child_test(img_request)); | |
2163 | ||
2164 | /* First get what we need from the image request */ | |
2165 | ||
2166 | pages = img_request->copyup_pages; | |
2167 | rbd_assert(pages != NULL); | |
2168 | img_request->copyup_pages = NULL; | |
2169 | ||
2170 | orig_request = img_request->obj_request; | |
2171 | rbd_assert(orig_request != NULL); | |
0eefd470 | 2172 | rbd_assert(orig_request->type == OBJ_REQUEST_BIO); |
3d7efd18 AE |
2173 | result = img_request->result; |
2174 | obj_size = img_request->length; | |
2175 | xferred = img_request->xferred; | |
2176 | ||
0eefd470 AE |
2177 | rbd_dev = img_request->rbd_dev; |
2178 | rbd_assert(rbd_dev); | |
2179 | rbd_assert(obj_size == (u64)1 << rbd_dev->header.obj_order); | |
2180 | ||
3d7efd18 AE |
2181 | rbd_img_request_put(img_request); |
2182 | ||
0eefd470 AE |
2183 | if (result) |
2184 | goto out_err; | |
2185 | ||
2186 | /* Allocate the new copyup osd request for the original request */ | |
2187 | ||
2188 | result = -ENOMEM; | |
2189 | rbd_assert(!orig_request->osd_req); | |
2190 | osd_req = rbd_osd_req_create_copyup(orig_request); | |
2191 | if (!osd_req) | |
2192 | goto out_err; | |
2193 | orig_request->osd_req = osd_req; | |
2194 | orig_request->copyup_pages = pages; | |
3d7efd18 | 2195 | |
0eefd470 | 2196 | /* Initialize the copyup op */ |
3d7efd18 | 2197 | |
0eefd470 AE |
2198 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
2199 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, obj_size, 0, | |
2200 | false, false); | |
3d7efd18 | 2201 | |
0eefd470 AE |
2202 | /* Then the original write request op */ |
2203 | ||
2204 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, | |
2205 | orig_request->offset, | |
2206 | orig_request->length, 0, 0); | |
2207 | osd_req_op_extent_osd_data_bio(osd_req, 1, orig_request->bio_list, | |
2208 | orig_request->length); | |
2209 | ||
2210 | rbd_osd_req_format_write(orig_request); | |
2211 | ||
2212 | /* All set, send it off. */ | |
2213 | ||
2214 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2215 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2216 | result = rbd_obj_request_submit(osdc, orig_request); | |
2217 | if (!result) | |
2218 | return; | |
2219 | out_err: | |
2220 | /* Record the error code and complete the request */ | |
2221 | ||
2222 | orig_request->result = result; | |
2223 | orig_request->xferred = 0; | |
2224 | obj_request_done_set(orig_request); | |
2225 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2226 | } |
2227 | ||
2228 | /* | |
2229 | * Read from the parent image the range of data that covers the | |
2230 | * entire target of the given object request. This is used for | |
2231 | * satisfying a layered image write request when the target of an | |
2232 | * object request from the image request does not exist. | |
2233 | * | |
2234 | * A page array big enough to hold the returned data is allocated | |
2235 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2236 | * When the read completes, this page array will be transferred to | |
2237 | * the original object request for the copyup operation. | |
2238 | * | |
2239 | * If an error occurs, record it as the result of the original | |
2240 | * object request and mark it done so it gets completed. | |
2241 | */ | |
2242 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2243 | { | |
2244 | struct rbd_img_request *img_request = NULL; | |
2245 | struct rbd_img_request *parent_request = NULL; | |
2246 | struct rbd_device *rbd_dev; | |
2247 | u64 img_offset; | |
2248 | u64 length; | |
2249 | struct page **pages = NULL; | |
2250 | u32 page_count; | |
2251 | int result; | |
2252 | ||
2253 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2254 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2255 | ||
2256 | img_request = obj_request->img_request; | |
2257 | rbd_assert(img_request != NULL); | |
2258 | rbd_dev = img_request->rbd_dev; | |
2259 | rbd_assert(rbd_dev->parent != NULL); | |
2260 | ||
0eefd470 AE |
2261 | /* |
2262 | * First things first. The original osd request is of no | |
2263 | * use to use any more, we'll need a new one that can hold | |
2264 | * the two ops in a copyup request. We'll get that later, | |
2265 | * but for now we can release the old one. | |
2266 | */ | |
2267 | rbd_osd_req_destroy(obj_request->osd_req); | |
2268 | obj_request->osd_req = NULL; | |
2269 | ||
3d7efd18 AE |
2270 | /* |
2271 | * Determine the byte range covered by the object in the | |
2272 | * child image to which the original request was to be sent. | |
2273 | */ | |
2274 | img_offset = obj_request->img_offset - obj_request->offset; | |
2275 | length = (u64)1 << rbd_dev->header.obj_order; | |
2276 | ||
a9e8ba2c AE |
2277 | /* |
2278 | * There is no defined parent data beyond the parent | |
2279 | * overlap, so limit what we read at that boundary if | |
2280 | * necessary. | |
2281 | */ | |
2282 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2283 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2284 | length = rbd_dev->parent_overlap - img_offset; | |
2285 | } | |
2286 | ||
3d7efd18 AE |
2287 | /* |
2288 | * Allocate a page array big enough to receive the data read | |
2289 | * from the parent. | |
2290 | */ | |
2291 | page_count = (u32)calc_pages_for(0, length); | |
2292 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2293 | if (IS_ERR(pages)) { | |
2294 | result = PTR_ERR(pages); | |
2295 | pages = NULL; | |
2296 | goto out_err; | |
2297 | } | |
2298 | ||
2299 | result = -ENOMEM; | |
2300 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2301 | img_offset, length, | |
2302 | false, true); | |
2303 | if (!parent_request) | |
2304 | goto out_err; | |
2305 | rbd_obj_request_get(obj_request); | |
2306 | parent_request->obj_request = obj_request; | |
2307 | ||
2308 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2309 | if (result) | |
2310 | goto out_err; | |
2311 | parent_request->copyup_pages = pages; | |
2312 | ||
2313 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2314 | result = rbd_img_request_submit(parent_request); | |
2315 | if (!result) | |
2316 | return 0; | |
2317 | ||
2318 | parent_request->copyup_pages = NULL; | |
2319 | parent_request->obj_request = NULL; | |
2320 | rbd_obj_request_put(obj_request); | |
2321 | out_err: | |
2322 | if (pages) | |
2323 | ceph_release_page_vector(pages, page_count); | |
2324 | if (parent_request) | |
2325 | rbd_img_request_put(parent_request); | |
2326 | obj_request->result = result; | |
2327 | obj_request->xferred = 0; | |
2328 | obj_request_done_set(obj_request); | |
2329 | ||
2330 | return result; | |
2331 | } | |
2332 | ||
c5b5ef6c AE |
2333 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2334 | { | |
c5b5ef6c AE |
2335 | struct rbd_obj_request *orig_request; |
2336 | int result; | |
2337 | ||
2338 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2339 | ||
2340 | /* | |
2341 | * All we need from the object request is the original | |
2342 | * request and the result of the STAT op. Grab those, then | |
2343 | * we're done with the request. | |
2344 | */ | |
2345 | orig_request = obj_request->obj_request; | |
2346 | obj_request->obj_request = NULL; | |
2347 | rbd_assert(orig_request); | |
2348 | rbd_assert(orig_request->img_request); | |
2349 | ||
2350 | result = obj_request->result; | |
2351 | obj_request->result = 0; | |
2352 | ||
2353 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2354 | obj_request, orig_request, result, | |
2355 | obj_request->xferred, obj_request->length); | |
2356 | rbd_obj_request_put(obj_request); | |
2357 | ||
2358 | rbd_assert(orig_request); | |
2359 | rbd_assert(orig_request->img_request); | |
c5b5ef6c AE |
2360 | |
2361 | /* | |
2362 | * Our only purpose here is to determine whether the object | |
2363 | * exists, and we don't want to treat the non-existence as | |
2364 | * an error. If something else comes back, transfer the | |
2365 | * error to the original request and complete it now. | |
2366 | */ | |
2367 | if (!result) { | |
2368 | obj_request_existence_set(orig_request, true); | |
2369 | } else if (result == -ENOENT) { | |
2370 | obj_request_existence_set(orig_request, false); | |
2371 | } else if (result) { | |
2372 | orig_request->result = result; | |
3d7efd18 | 2373 | goto out; |
c5b5ef6c AE |
2374 | } |
2375 | ||
2376 | /* | |
2377 | * Resubmit the original request now that we have recorded | |
2378 | * whether the target object exists. | |
2379 | */ | |
b454e36d | 2380 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2381 | out: |
c5b5ef6c AE |
2382 | if (orig_request->result) |
2383 | rbd_obj_request_complete(orig_request); | |
2384 | rbd_obj_request_put(orig_request); | |
2385 | } | |
2386 | ||
2387 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2388 | { | |
2389 | struct rbd_obj_request *stat_request; | |
2390 | struct rbd_device *rbd_dev; | |
2391 | struct ceph_osd_client *osdc; | |
2392 | struct page **pages = NULL; | |
2393 | u32 page_count; | |
2394 | size_t size; | |
2395 | int ret; | |
2396 | ||
2397 | /* | |
2398 | * The response data for a STAT call consists of: | |
2399 | * le64 length; | |
2400 | * struct { | |
2401 | * le32 tv_sec; | |
2402 | * le32 tv_nsec; | |
2403 | * } mtime; | |
2404 | */ | |
2405 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2406 | page_count = (u32)calc_pages_for(0, size); | |
2407 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2408 | if (IS_ERR(pages)) | |
2409 | return PTR_ERR(pages); | |
2410 | ||
2411 | ret = -ENOMEM; | |
2412 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2413 | OBJ_REQUEST_PAGES); | |
2414 | if (!stat_request) | |
2415 | goto out; | |
2416 | ||
2417 | rbd_obj_request_get(obj_request); | |
2418 | stat_request->obj_request = obj_request; | |
2419 | stat_request->pages = pages; | |
2420 | stat_request->page_count = page_count; | |
2421 | ||
2422 | rbd_assert(obj_request->img_request); | |
2423 | rbd_dev = obj_request->img_request->rbd_dev; | |
2424 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2425 | stat_request); | |
2426 | if (!stat_request->osd_req) | |
2427 | goto out; | |
2428 | stat_request->callback = rbd_img_obj_exists_callback; | |
2429 | ||
2430 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2431 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2432 | false, false); | |
9d4df01f | 2433 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2434 | |
2435 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2436 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2437 | out: | |
2438 | if (ret) | |
2439 | rbd_obj_request_put(obj_request); | |
2440 | ||
2441 | return ret; | |
2442 | } | |
2443 | ||
b454e36d AE |
2444 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2445 | { | |
2446 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2447 | struct rbd_device *rbd_dev; |
3d7efd18 | 2448 | bool known; |
b454e36d AE |
2449 | |
2450 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2451 | ||
2452 | img_request = obj_request->img_request; | |
2453 | rbd_assert(img_request); | |
a9e8ba2c | 2454 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2455 | |
b454e36d | 2456 | /* |
a9e8ba2c AE |
2457 | * Only writes to layered images need special handling. |
2458 | * Reads and non-layered writes are simple object requests. | |
2459 | * Layered writes that start beyond the end of the overlap | |
2460 | * with the parent have no parent data, so they too are | |
2461 | * simple object requests. Finally, if the target object is | |
2462 | * known to already exist, its parent data has already been | |
2463 | * copied, so a write to the object can also be handled as a | |
2464 | * simple object request. | |
b454e36d AE |
2465 | */ |
2466 | if (!img_request_write_test(img_request) || | |
2467 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2468 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2469 | ((known = obj_request_known_test(obj_request)) && |
2470 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2471 | |
2472 | struct rbd_device *rbd_dev; | |
2473 | struct ceph_osd_client *osdc; | |
2474 | ||
2475 | rbd_dev = obj_request->img_request->rbd_dev; | |
2476 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2477 | ||
2478 | return rbd_obj_request_submit(osdc, obj_request); | |
2479 | } | |
2480 | ||
2481 | /* | |
3d7efd18 AE |
2482 | * It's a layered write. The target object might exist but |
2483 | * we may not know that yet. If we know it doesn't exist, | |
2484 | * start by reading the data for the full target object from | |
2485 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2486 | */ |
3d7efd18 AE |
2487 | if (known) |
2488 | return rbd_img_obj_parent_read_full(obj_request); | |
2489 | ||
2490 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2491 | |
2492 | return rbd_img_obj_exists_submit(obj_request); | |
2493 | } | |
2494 | ||
bf0d5f50 AE |
2495 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2496 | { | |
bf0d5f50 | 2497 | struct rbd_obj_request *obj_request; |
46faeed4 | 2498 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2499 | |
37206ee5 | 2500 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2501 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2502 | int ret; |
2503 | ||
b454e36d | 2504 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2505 | if (ret) |
2506 | return ret; | |
bf0d5f50 AE |
2507 | } |
2508 | ||
2509 | return 0; | |
2510 | } | |
8b3e1a56 AE |
2511 | |
2512 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2513 | { | |
2514 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2515 | struct rbd_device *rbd_dev; |
2516 | u64 obj_end; | |
8b3e1a56 AE |
2517 | |
2518 | rbd_assert(img_request_child_test(img_request)); | |
2519 | ||
2520 | obj_request = img_request->obj_request; | |
a9e8ba2c AE |
2521 | rbd_assert(obj_request); |
2522 | rbd_assert(obj_request->img_request); | |
2523 | ||
8b3e1a56 | 2524 | obj_request->result = img_request->result; |
a9e8ba2c AE |
2525 | if (obj_request->result) |
2526 | goto out; | |
2527 | ||
2528 | /* | |
2529 | * We need to zero anything beyond the parent overlap | |
2530 | * boundary. Since rbd_img_obj_request_read_callback() | |
2531 | * will zero anything beyond the end of a short read, an | |
2532 | * easy way to do this is to pretend the data from the | |
2533 | * parent came up short--ending at the overlap boundary. | |
2534 | */ | |
2535 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2536 | obj_end = obj_request->img_offset + obj_request->length; | |
2537 | rbd_dev = obj_request->img_request->rbd_dev; | |
2538 | if (obj_end > rbd_dev->parent_overlap) { | |
2539 | u64 xferred = 0; | |
2540 | ||
2541 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2542 | xferred = rbd_dev->parent_overlap - | |
2543 | obj_request->img_offset; | |
8b3e1a56 | 2544 | |
a9e8ba2c AE |
2545 | obj_request->xferred = min(img_request->xferred, xferred); |
2546 | } else { | |
2547 | obj_request->xferred = img_request->xferred; | |
2548 | } | |
2549 | out: | |
b5b09be3 | 2550 | rbd_img_request_put(img_request); |
8b3e1a56 AE |
2551 | rbd_img_obj_request_read_callback(obj_request); |
2552 | rbd_obj_request_complete(obj_request); | |
2553 | } | |
2554 | ||
2555 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2556 | { | |
2557 | struct rbd_device *rbd_dev; | |
2558 | struct rbd_img_request *img_request; | |
2559 | int result; | |
2560 | ||
2561 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2562 | rbd_assert(obj_request->img_request != NULL); | |
2563 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
2564 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2565 | ||
2566 | rbd_dev = obj_request->img_request->rbd_dev; | |
2567 | rbd_assert(rbd_dev->parent != NULL); | |
2568 | /* rbd_read_finish(obj_request, obj_request->length); */ | |
2569 | img_request = rbd_img_request_create(rbd_dev->parent, | |
2570 | obj_request->img_offset, | |
2571 | obj_request->length, | |
2572 | false, true); | |
2573 | result = -ENOMEM; | |
2574 | if (!img_request) | |
2575 | goto out_err; | |
2576 | ||
2577 | rbd_obj_request_get(obj_request); | |
2578 | img_request->obj_request = obj_request; | |
2579 | ||
f1a4739f AE |
2580 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2581 | obj_request->bio_list); | |
8b3e1a56 AE |
2582 | if (result) |
2583 | goto out_err; | |
2584 | ||
2585 | img_request->callback = rbd_img_parent_read_callback; | |
2586 | result = rbd_img_request_submit(img_request); | |
2587 | if (result) | |
2588 | goto out_err; | |
2589 | ||
2590 | return; | |
2591 | out_err: | |
2592 | if (img_request) | |
2593 | rbd_img_request_put(img_request); | |
2594 | obj_request->result = result; | |
2595 | obj_request->xferred = 0; | |
2596 | obj_request_done_set(obj_request); | |
2597 | } | |
bf0d5f50 | 2598 | |
cc4a38bd | 2599 | static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2600 | { |
2601 | struct rbd_obj_request *obj_request; | |
2169238d | 2602 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2603 | int ret; |
2604 | ||
2605 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2606 | OBJ_REQUEST_NODATA); | |
2607 | if (!obj_request) | |
2608 | return -ENOMEM; | |
2609 | ||
2610 | ret = -ENOMEM; | |
430c28c3 | 2611 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2612 | if (!obj_request->osd_req) |
2613 | goto out; | |
2169238d | 2614 | obj_request->callback = rbd_obj_request_put; |
b8d70035 | 2615 | |
c99d2d4a | 2616 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2617 | notify_id, 0, 0); |
9d4df01f | 2618 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2619 | |
b8d70035 | 2620 | ret = rbd_obj_request_submit(osdc, obj_request); |
b8d70035 | 2621 | out: |
cf81b60e AE |
2622 | if (ret) |
2623 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2624 | |
2625 | return ret; | |
2626 | } | |
2627 | ||
2628 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2629 | { | |
2630 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
b8d70035 AE |
2631 | |
2632 | if (!rbd_dev) | |
2633 | return; | |
2634 | ||
37206ee5 | 2635 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2636 | rbd_dev->header_name, (unsigned long long)notify_id, |
2637 | (unsigned int)opcode); | |
2638 | (void)rbd_dev_refresh(rbd_dev); | |
b8d70035 | 2639 | |
cc4a38bd | 2640 | rbd_obj_notify_ack(rbd_dev, notify_id); |
b8d70035 AE |
2641 | } |
2642 | ||
9969ebc5 AE |
2643 | /* |
2644 | * Request sync osd watch/unwatch. The value of "start" determines | |
2645 | * whether a watch request is being initiated or torn down. | |
2646 | */ | |
2647 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, int start) | |
2648 | { | |
2649 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2650 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2651 | int ret; |
2652 | ||
2653 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2654 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2655 | ||
2656 | if (start) { | |
3c663bbd | 2657 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2658 | &rbd_dev->watch_event); |
2659 | if (ret < 0) | |
2660 | return ret; | |
8eb87565 | 2661 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2662 | } |
2663 | ||
2664 | ret = -ENOMEM; | |
2665 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2666 | OBJ_REQUEST_NODATA); | |
2667 | if (!obj_request) | |
2668 | goto out_cancel; | |
2669 | ||
430c28c3 AE |
2670 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2671 | if (!obj_request->osd_req) | |
2672 | goto out_cancel; | |
2673 | ||
8eb87565 | 2674 | if (start) |
975241af | 2675 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2676 | else |
6977c3f9 | 2677 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2678 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2679 | |
2680 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
b21ebddd | 2681 | rbd_dev->watch_event->cookie, 0, start); |
9d4df01f | 2682 | rbd_osd_req_format_write(obj_request); |
2169238d | 2683 | |
9969ebc5 AE |
2684 | ret = rbd_obj_request_submit(osdc, obj_request); |
2685 | if (ret) | |
2686 | goto out_cancel; | |
2687 | ret = rbd_obj_request_wait(obj_request); | |
2688 | if (ret) | |
2689 | goto out_cancel; | |
9969ebc5 AE |
2690 | ret = obj_request->result; |
2691 | if (ret) | |
2692 | goto out_cancel; | |
2693 | ||
8eb87565 AE |
2694 | /* |
2695 | * A watch request is set to linger, so the underlying osd | |
2696 | * request won't go away until we unregister it. We retain | |
2697 | * a pointer to the object request during that time (in | |
2698 | * rbd_dev->watch_request), so we'll keep a reference to | |
2699 | * it. We'll drop that reference (below) after we've | |
2700 | * unregistered it. | |
2701 | */ | |
2702 | if (start) { | |
2703 | rbd_dev->watch_request = obj_request; | |
2704 | ||
2705 | return 0; | |
2706 | } | |
2707 | ||
2708 | /* We have successfully torn down the watch request */ | |
2709 | ||
2710 | rbd_obj_request_put(rbd_dev->watch_request); | |
2711 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2712 | out_cancel: |
2713 | /* Cancel the event if we're tearing down, or on error */ | |
2714 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2715 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2716 | if (obj_request) |
2717 | rbd_obj_request_put(obj_request); | |
2718 | ||
2719 | return ret; | |
2720 | } | |
2721 | ||
36be9a76 | 2722 | /* |
f40eb349 AE |
2723 | * Synchronous osd object method call. Returns the number of bytes |
2724 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2725 | */ |
2726 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2727 | const char *object_name, | |
2728 | const char *class_name, | |
2729 | const char *method_name, | |
4157976b | 2730 | const void *outbound, |
36be9a76 | 2731 | size_t outbound_size, |
4157976b | 2732 | void *inbound, |
e2a58ee5 | 2733 | size_t inbound_size) |
36be9a76 | 2734 | { |
2169238d | 2735 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2736 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2737 | struct page **pages; |
2738 | u32 page_count; | |
2739 | int ret; | |
2740 | ||
2741 | /* | |
6010a451 AE |
2742 | * Method calls are ultimately read operations. The result |
2743 | * should placed into the inbound buffer provided. They | |
2744 | * also supply outbound data--parameters for the object | |
2745 | * method. Currently if this is present it will be a | |
2746 | * snapshot id. | |
36be9a76 | 2747 | */ |
57385b51 | 2748 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2749 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2750 | if (IS_ERR(pages)) | |
2751 | return PTR_ERR(pages); | |
2752 | ||
2753 | ret = -ENOMEM; | |
6010a451 | 2754 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2755 | OBJ_REQUEST_PAGES); |
2756 | if (!obj_request) | |
2757 | goto out; | |
2758 | ||
2759 | obj_request->pages = pages; | |
2760 | obj_request->page_count = page_count; | |
2761 | ||
430c28c3 | 2762 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2763 | if (!obj_request->osd_req) |
2764 | goto out; | |
2765 | ||
c99d2d4a | 2766 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2767 | class_name, method_name); |
2768 | if (outbound_size) { | |
2769 | struct ceph_pagelist *pagelist; | |
2770 | ||
2771 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2772 | if (!pagelist) | |
2773 | goto out; | |
2774 | ||
2775 | ceph_pagelist_init(pagelist); | |
2776 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
2777 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
2778 | pagelist); | |
2779 | } | |
a4ce40a9 AE |
2780 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
2781 | obj_request->pages, inbound_size, | |
44cd188d | 2782 | 0, false, false); |
9d4df01f | 2783 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2784 | |
36be9a76 AE |
2785 | ret = rbd_obj_request_submit(osdc, obj_request); |
2786 | if (ret) | |
2787 | goto out; | |
2788 | ret = rbd_obj_request_wait(obj_request); | |
2789 | if (ret) | |
2790 | goto out; | |
2791 | ||
2792 | ret = obj_request->result; | |
2793 | if (ret < 0) | |
2794 | goto out; | |
57385b51 AE |
2795 | |
2796 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
2797 | ret = (int)obj_request->xferred; | |
903bb32e | 2798 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
2799 | out: |
2800 | if (obj_request) | |
2801 | rbd_obj_request_put(obj_request); | |
2802 | else | |
2803 | ceph_release_page_vector(pages, page_count); | |
2804 | ||
2805 | return ret; | |
2806 | } | |
2807 | ||
bf0d5f50 | 2808 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 2809 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
2810 | { |
2811 | struct rbd_device *rbd_dev = q->queuedata; | |
2812 | bool read_only = rbd_dev->mapping.read_only; | |
2813 | struct request *rq; | |
2814 | int result; | |
2815 | ||
2816 | while ((rq = blk_fetch_request(q))) { | |
2817 | bool write_request = rq_data_dir(rq) == WRITE; | |
2818 | struct rbd_img_request *img_request; | |
2819 | u64 offset; | |
2820 | u64 length; | |
2821 | ||
2822 | /* Ignore any non-FS requests that filter through. */ | |
2823 | ||
2824 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
2825 | dout("%s: non-fs request type %d\n", __func__, |
2826 | (int) rq->cmd_type); | |
2827 | __blk_end_request_all(rq, 0); | |
2828 | continue; | |
2829 | } | |
2830 | ||
2831 | /* Ignore/skip any zero-length requests */ | |
2832 | ||
2833 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
2834 | length = (u64) blk_rq_bytes(rq); | |
2835 | ||
2836 | if (!length) { | |
2837 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
2838 | __blk_end_request_all(rq, 0); |
2839 | continue; | |
2840 | } | |
2841 | ||
2842 | spin_unlock_irq(q->queue_lock); | |
2843 | ||
2844 | /* Disallow writes to a read-only device */ | |
2845 | ||
2846 | if (write_request) { | |
2847 | result = -EROFS; | |
2848 | if (read_only) | |
2849 | goto end_request; | |
2850 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
2851 | } | |
2852 | ||
6d292906 AE |
2853 | /* |
2854 | * Quit early if the mapped snapshot no longer | |
2855 | * exists. It's still possible the snapshot will | |
2856 | * have disappeared by the time our request arrives | |
2857 | * at the osd, but there's no sense in sending it if | |
2858 | * we already know. | |
2859 | */ | |
2860 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
2861 | dout("request for non-existent snapshot"); |
2862 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
2863 | result = -ENXIO; | |
2864 | goto end_request; | |
2865 | } | |
2866 | ||
bf0d5f50 | 2867 | result = -EINVAL; |
c0cd10db AE |
2868 | if (offset && length > U64_MAX - offset + 1) { |
2869 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
2870 | offset, length); | |
bf0d5f50 | 2871 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 2872 | } |
bf0d5f50 AE |
2873 | |
2874 | result = -ENOMEM; | |
2875 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
9849e986 | 2876 | write_request, false); |
bf0d5f50 AE |
2877 | if (!img_request) |
2878 | goto end_request; | |
2879 | ||
2880 | img_request->rq = rq; | |
2881 | ||
f1a4739f AE |
2882 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
2883 | rq->bio); | |
bf0d5f50 AE |
2884 | if (!result) |
2885 | result = rbd_img_request_submit(img_request); | |
2886 | if (result) | |
2887 | rbd_img_request_put(img_request); | |
2888 | end_request: | |
2889 | spin_lock_irq(q->queue_lock); | |
2890 | if (result < 0) { | |
7da22d29 AE |
2891 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
2892 | write_request ? "write" : "read", | |
2893 | length, offset, result); | |
2894 | ||
bf0d5f50 AE |
2895 | __blk_end_request_all(rq, result); |
2896 | } | |
2897 | } | |
2898 | } | |
2899 | ||
602adf40 YS |
2900 | /* |
2901 | * a queue callback. Makes sure that we don't create a bio that spans across | |
2902 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 2903 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
2904 | */ |
2905 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
2906 | struct bio_vec *bvec) | |
2907 | { | |
2908 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
2909 | sector_t sector_offset; |
2910 | sector_t sectors_per_obj; | |
2911 | sector_t obj_sector_offset; | |
2912 | int ret; | |
2913 | ||
2914 | /* | |
2915 | * Find how far into its rbd object the partition-relative | |
2916 | * bio start sector is to offset relative to the enclosing | |
2917 | * device. | |
2918 | */ | |
2919 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
2920 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
2921 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
2922 | ||
2923 | /* | |
2924 | * Compute the number of bytes from that offset to the end | |
2925 | * of the object. Account for what's already used by the bio. | |
2926 | */ | |
2927 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
2928 | if (ret > bmd->bi_size) | |
2929 | ret -= bmd->bi_size; | |
2930 | else | |
2931 | ret = 0; | |
2932 | ||
2933 | /* | |
2934 | * Don't send back more than was asked for. And if the bio | |
2935 | * was empty, let the whole thing through because: "Note | |
2936 | * that a block device *must* allow a single page to be | |
2937 | * added to an empty bio." | |
2938 | */ | |
2939 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
2940 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
2941 | ret = (int) bvec->bv_len; | |
2942 | ||
2943 | return ret; | |
602adf40 YS |
2944 | } |
2945 | ||
2946 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
2947 | { | |
2948 | struct gendisk *disk = rbd_dev->disk; | |
2949 | ||
2950 | if (!disk) | |
2951 | return; | |
2952 | ||
a0cab924 AE |
2953 | rbd_dev->disk = NULL; |
2954 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 2955 | del_gendisk(disk); |
a0cab924 AE |
2956 | if (disk->queue) |
2957 | blk_cleanup_queue(disk->queue); | |
2958 | } | |
602adf40 YS |
2959 | put_disk(disk); |
2960 | } | |
2961 | ||
788e2df3 AE |
2962 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
2963 | const char *object_name, | |
7097f8df | 2964 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
2965 | |
2966 | { | |
2169238d | 2967 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 2968 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
2969 | struct page **pages = NULL; |
2970 | u32 page_count; | |
1ceae7ef | 2971 | size_t size; |
788e2df3 AE |
2972 | int ret; |
2973 | ||
2974 | page_count = (u32) calc_pages_for(offset, length); | |
2975 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2976 | if (IS_ERR(pages)) | |
2977 | ret = PTR_ERR(pages); | |
2978 | ||
2979 | ret = -ENOMEM; | |
2980 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 2981 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
2982 | if (!obj_request) |
2983 | goto out; | |
2984 | ||
2985 | obj_request->pages = pages; | |
2986 | obj_request->page_count = page_count; | |
2987 | ||
430c28c3 | 2988 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
2989 | if (!obj_request->osd_req) |
2990 | goto out; | |
2991 | ||
c99d2d4a AE |
2992 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
2993 | offset, length, 0, 0); | |
406e2c9f | 2994 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 2995 | obj_request->pages, |
44cd188d AE |
2996 | obj_request->length, |
2997 | obj_request->offset & ~PAGE_MASK, | |
2998 | false, false); | |
9d4df01f | 2999 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3000 | |
788e2df3 AE |
3001 | ret = rbd_obj_request_submit(osdc, obj_request); |
3002 | if (ret) | |
3003 | goto out; | |
3004 | ret = rbd_obj_request_wait(obj_request); | |
3005 | if (ret) | |
3006 | goto out; | |
3007 | ||
3008 | ret = obj_request->result; | |
3009 | if (ret < 0) | |
3010 | goto out; | |
1ceae7ef AE |
3011 | |
3012 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3013 | size = (size_t) obj_request->xferred; | |
903bb32e | 3014 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3015 | rbd_assert(size <= (size_t)INT_MAX); |
3016 | ret = (int)size; | |
788e2df3 AE |
3017 | out: |
3018 | if (obj_request) | |
3019 | rbd_obj_request_put(obj_request); | |
3020 | else | |
3021 | ceph_release_page_vector(pages, page_count); | |
3022 | ||
3023 | return ret; | |
3024 | } | |
3025 | ||
602adf40 | 3026 | /* |
4156d998 AE |
3027 | * Read the complete header for the given rbd device. |
3028 | * | |
3029 | * Returns a pointer to a dynamically-allocated buffer containing | |
3030 | * the complete and validated header. Caller can pass the address | |
3031 | * of a variable that will be filled in with the version of the | |
3032 | * header object at the time it was read. | |
3033 | * | |
3034 | * Returns a pointer-coded errno if a failure occurs. | |
602adf40 | 3035 | */ |
4156d998 | 3036 | static struct rbd_image_header_ondisk * |
7097f8df | 3037 | rbd_dev_v1_header_read(struct rbd_device *rbd_dev) |
602adf40 | 3038 | { |
4156d998 | 3039 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3040 | u32 snap_count = 0; |
4156d998 AE |
3041 | u64 names_size = 0; |
3042 | u32 want_count; | |
3043 | int ret; | |
602adf40 | 3044 | |
00f1f36f | 3045 | /* |
4156d998 AE |
3046 | * The complete header will include an array of its 64-bit |
3047 | * snapshot ids, followed by the names of those snapshots as | |
3048 | * a contiguous block of NUL-terminated strings. Note that | |
3049 | * the number of snapshots could change by the time we read | |
3050 | * it in, in which case we re-read it. | |
00f1f36f | 3051 | */ |
4156d998 AE |
3052 | do { |
3053 | size_t size; | |
3054 | ||
3055 | kfree(ondisk); | |
3056 | ||
3057 | size = sizeof (*ondisk); | |
3058 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3059 | size += names_size; | |
3060 | ondisk = kmalloc(size, GFP_KERNEL); | |
3061 | if (!ondisk) | |
3062 | return ERR_PTR(-ENOMEM); | |
3063 | ||
788e2df3 | 3064 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3065 | 0, size, ondisk); |
4156d998 AE |
3066 | if (ret < 0) |
3067 | goto out_err; | |
c0cd10db | 3068 | if ((size_t)ret < size) { |
4156d998 | 3069 | ret = -ENXIO; |
06ecc6cb AE |
3070 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3071 | size, ret); | |
4156d998 AE |
3072 | goto out_err; |
3073 | } | |
3074 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3075 | ret = -ENXIO; | |
06ecc6cb | 3076 | rbd_warn(rbd_dev, "invalid header"); |
4156d998 | 3077 | goto out_err; |
81e759fb | 3078 | } |
602adf40 | 3079 | |
4156d998 AE |
3080 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3081 | want_count = snap_count; | |
3082 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3083 | } while (snap_count != want_count); | |
00f1f36f | 3084 | |
4156d998 | 3085 | return ondisk; |
00f1f36f | 3086 | |
4156d998 AE |
3087 | out_err: |
3088 | kfree(ondisk); | |
3089 | ||
3090 | return ERR_PTR(ret); | |
3091 | } | |
3092 | ||
3093 | /* | |
3094 | * reload the ondisk the header | |
3095 | */ | |
3096 | static int rbd_read_header(struct rbd_device *rbd_dev, | |
3097 | struct rbd_image_header *header) | |
3098 | { | |
3099 | struct rbd_image_header_ondisk *ondisk; | |
4156d998 | 3100 | int ret; |
602adf40 | 3101 | |
7097f8df | 3102 | ondisk = rbd_dev_v1_header_read(rbd_dev); |
4156d998 AE |
3103 | if (IS_ERR(ondisk)) |
3104 | return PTR_ERR(ondisk); | |
3105 | ret = rbd_header_from_disk(header, ondisk); | |
4156d998 AE |
3106 | kfree(ondisk); |
3107 | ||
3108 | return ret; | |
602adf40 YS |
3109 | } |
3110 | ||
9478554a AE |
3111 | static void rbd_update_mapping_size(struct rbd_device *rbd_dev) |
3112 | { | |
0d7dbfce | 3113 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
9478554a AE |
3114 | return; |
3115 | ||
e28626a0 AE |
3116 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) { |
3117 | sector_t size; | |
3118 | ||
3119 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
3120 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3121 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3122 | set_capacity(rbd_dev->disk, size); | |
3123 | } | |
9478554a AE |
3124 | } |
3125 | ||
602adf40 YS |
3126 | /* |
3127 | * only read the first part of the ondisk header, without the snaps info | |
3128 | */ | |
cc4a38bd | 3129 | static int rbd_dev_v1_refresh(struct rbd_device *rbd_dev) |
602adf40 YS |
3130 | { |
3131 | int ret; | |
3132 | struct rbd_image_header h; | |
602adf40 YS |
3133 | |
3134 | ret = rbd_read_header(rbd_dev, &h); | |
3135 | if (ret < 0) | |
3136 | return ret; | |
3137 | ||
a51aa0c0 JD |
3138 | down_write(&rbd_dev->header_rwsem); |
3139 | ||
9478554a AE |
3140 | /* Update image size, and check for resize of mapped image */ |
3141 | rbd_dev->header.image_size = h.image_size; | |
3142 | rbd_update_mapping_size(rbd_dev); | |
9db4b3e3 | 3143 | |
849b4260 | 3144 | /* rbd_dev->header.object_prefix shouldn't change */ |
602adf40 | 3145 | kfree(rbd_dev->header.snap_sizes); |
849b4260 | 3146 | kfree(rbd_dev->header.snap_names); |
d1d25646 | 3147 | /* osd requests may still refer to snapc */ |
812164f8 | 3148 | ceph_put_snap_context(rbd_dev->header.snapc); |
602adf40 | 3149 | |
93a24e08 | 3150 | rbd_dev->header.image_size = h.image_size; |
602adf40 YS |
3151 | rbd_dev->header.snapc = h.snapc; |
3152 | rbd_dev->header.snap_names = h.snap_names; | |
3153 | rbd_dev->header.snap_sizes = h.snap_sizes; | |
849b4260 | 3154 | /* Free the extra copy of the object prefix */ |
c0cd10db AE |
3155 | if (strcmp(rbd_dev->header.object_prefix, h.object_prefix)) |
3156 | rbd_warn(rbd_dev, "object prefix changed (ignoring)"); | |
849b4260 AE |
3157 | kfree(h.object_prefix); |
3158 | ||
c666601a | 3159 | up_write(&rbd_dev->header_rwsem); |
602adf40 | 3160 | |
dfc5606d | 3161 | return ret; |
602adf40 YS |
3162 | } |
3163 | ||
15228ede AE |
3164 | /* |
3165 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3166 | * has disappeared from the (just updated) snapshot context. | |
3167 | */ | |
3168 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3169 | { | |
3170 | u64 snap_id; | |
3171 | ||
3172 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3173 | return; | |
3174 | ||
3175 | snap_id = rbd_dev->spec->snap_id; | |
3176 | if (snap_id == CEPH_NOSNAP) | |
3177 | return; | |
3178 | ||
3179 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3180 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3181 | } | |
3182 | ||
cc4a38bd | 3183 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3184 | { |
a3fbe5d4 | 3185 | u64 image_size; |
1fe5e993 AE |
3186 | int ret; |
3187 | ||
117973fb | 3188 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
a3fbe5d4 | 3189 | image_size = rbd_dev->header.image_size; |
1fe5e993 | 3190 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); |
117973fb | 3191 | if (rbd_dev->image_format == 1) |
cc4a38bd | 3192 | ret = rbd_dev_v1_refresh(rbd_dev); |
117973fb | 3193 | else |
cc4a38bd | 3194 | ret = rbd_dev_v2_refresh(rbd_dev); |
15228ede AE |
3195 | |
3196 | /* If it's a mapped snapshot, validate its EXISTS flag */ | |
3197 | ||
3198 | rbd_exists_validate(rbd_dev); | |
1fe5e993 | 3199 | mutex_unlock(&ctl_mutex); |
522a0cc0 AE |
3200 | if (ret) |
3201 | rbd_warn(rbd_dev, "got notification but failed to " | |
3202 | " update snaps: %d\n", ret); | |
a3fbe5d4 AE |
3203 | if (image_size != rbd_dev->header.image_size) |
3204 | revalidate_disk(rbd_dev->disk); | |
1fe5e993 AE |
3205 | |
3206 | return ret; | |
3207 | } | |
3208 | ||
602adf40 YS |
3209 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3210 | { | |
3211 | struct gendisk *disk; | |
3212 | struct request_queue *q; | |
593a9e7b | 3213 | u64 segment_size; |
602adf40 | 3214 | |
602adf40 | 3215 | /* create gendisk info */ |
602adf40 YS |
3216 | disk = alloc_disk(RBD_MINORS_PER_MAJOR); |
3217 | if (!disk) | |
1fcdb8aa | 3218 | return -ENOMEM; |
602adf40 | 3219 | |
f0f8cef5 | 3220 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3221 | rbd_dev->dev_id); |
602adf40 YS |
3222 | disk->major = rbd_dev->major; |
3223 | disk->first_minor = 0; | |
3224 | disk->fops = &rbd_bd_ops; | |
3225 | disk->private_data = rbd_dev; | |
3226 | ||
bf0d5f50 | 3227 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3228 | if (!q) |
3229 | goto out_disk; | |
029bcbd8 | 3230 | |
593a9e7b AE |
3231 | /* We use the default size, but let's be explicit about it. */ |
3232 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3233 | ||
029bcbd8 | 3234 | /* set io sizes to object size */ |
593a9e7b AE |
3235 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3236 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3237 | blk_queue_max_segment_size(q, segment_size); | |
3238 | blk_queue_io_min(q, segment_size); | |
3239 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3240 | |
602adf40 YS |
3241 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3242 | disk->queue = q; | |
3243 | ||
3244 | q->queuedata = rbd_dev; | |
3245 | ||
3246 | rbd_dev->disk = disk; | |
602adf40 | 3247 | |
602adf40 | 3248 | return 0; |
602adf40 YS |
3249 | out_disk: |
3250 | put_disk(disk); | |
1fcdb8aa AE |
3251 | |
3252 | return -ENOMEM; | |
602adf40 YS |
3253 | } |
3254 | ||
dfc5606d YS |
3255 | /* |
3256 | sysfs | |
3257 | */ | |
3258 | ||
593a9e7b AE |
3259 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3260 | { | |
3261 | return container_of(dev, struct rbd_device, dev); | |
3262 | } | |
3263 | ||
dfc5606d YS |
3264 | static ssize_t rbd_size_show(struct device *dev, |
3265 | struct device_attribute *attr, char *buf) | |
3266 | { | |
593a9e7b | 3267 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3268 | |
fc71d833 AE |
3269 | return sprintf(buf, "%llu\n", |
3270 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3271 | } |
3272 | ||
34b13184 AE |
3273 | /* |
3274 | * Note this shows the features for whatever's mapped, which is not | |
3275 | * necessarily the base image. | |
3276 | */ | |
3277 | static ssize_t rbd_features_show(struct device *dev, | |
3278 | struct device_attribute *attr, char *buf) | |
3279 | { | |
3280 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3281 | ||
3282 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3283 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3284 | } |
3285 | ||
dfc5606d YS |
3286 | static ssize_t rbd_major_show(struct device *dev, |
3287 | struct device_attribute *attr, char *buf) | |
3288 | { | |
593a9e7b | 3289 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3290 | |
fc71d833 AE |
3291 | if (rbd_dev->major) |
3292 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3293 | ||
3294 | return sprintf(buf, "(none)\n"); | |
3295 | ||
dfc5606d YS |
3296 | } |
3297 | ||
3298 | static ssize_t rbd_client_id_show(struct device *dev, | |
3299 | struct device_attribute *attr, char *buf) | |
602adf40 | 3300 | { |
593a9e7b | 3301 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3302 | |
1dbb4399 AE |
3303 | return sprintf(buf, "client%lld\n", |
3304 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3305 | } |
3306 | ||
dfc5606d YS |
3307 | static ssize_t rbd_pool_show(struct device *dev, |
3308 | struct device_attribute *attr, char *buf) | |
602adf40 | 3309 | { |
593a9e7b | 3310 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3311 | |
0d7dbfce | 3312 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3313 | } |
3314 | ||
9bb2f334 AE |
3315 | static ssize_t rbd_pool_id_show(struct device *dev, |
3316 | struct device_attribute *attr, char *buf) | |
3317 | { | |
3318 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3319 | ||
0d7dbfce | 3320 | return sprintf(buf, "%llu\n", |
fc71d833 | 3321 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3322 | } |
3323 | ||
dfc5606d YS |
3324 | static ssize_t rbd_name_show(struct device *dev, |
3325 | struct device_attribute *attr, char *buf) | |
3326 | { | |
593a9e7b | 3327 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3328 | |
a92ffdf8 AE |
3329 | if (rbd_dev->spec->image_name) |
3330 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3331 | ||
3332 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3333 | } |
3334 | ||
589d30e0 AE |
3335 | static ssize_t rbd_image_id_show(struct device *dev, |
3336 | struct device_attribute *attr, char *buf) | |
3337 | { | |
3338 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3339 | ||
0d7dbfce | 3340 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3341 | } |
3342 | ||
34b13184 AE |
3343 | /* |
3344 | * Shows the name of the currently-mapped snapshot (or | |
3345 | * RBD_SNAP_HEAD_NAME for the base image). | |
3346 | */ | |
dfc5606d YS |
3347 | static ssize_t rbd_snap_show(struct device *dev, |
3348 | struct device_attribute *attr, | |
3349 | char *buf) | |
3350 | { | |
593a9e7b | 3351 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3352 | |
0d7dbfce | 3353 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3354 | } |
3355 | ||
86b00e0d AE |
3356 | /* |
3357 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3358 | * for the parent image. If there is no parent, simply shows | |
3359 | * "(no parent image)". | |
3360 | */ | |
3361 | static ssize_t rbd_parent_show(struct device *dev, | |
3362 | struct device_attribute *attr, | |
3363 | char *buf) | |
3364 | { | |
3365 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3366 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3367 | int count; | |
3368 | char *bufp = buf; | |
3369 | ||
3370 | if (!spec) | |
3371 | return sprintf(buf, "(no parent image)\n"); | |
3372 | ||
3373 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3374 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3375 | if (count < 0) | |
3376 | return count; | |
3377 | bufp += count; | |
3378 | ||
3379 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3380 | spec->image_name ? spec->image_name : "(unknown)"); | |
3381 | if (count < 0) | |
3382 | return count; | |
3383 | bufp += count; | |
3384 | ||
3385 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3386 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3387 | if (count < 0) | |
3388 | return count; | |
3389 | bufp += count; | |
3390 | ||
3391 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3392 | if (count < 0) | |
3393 | return count; | |
3394 | bufp += count; | |
3395 | ||
3396 | return (ssize_t) (bufp - buf); | |
3397 | } | |
3398 | ||
dfc5606d YS |
3399 | static ssize_t rbd_image_refresh(struct device *dev, |
3400 | struct device_attribute *attr, | |
3401 | const char *buf, | |
3402 | size_t size) | |
3403 | { | |
593a9e7b | 3404 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3405 | int ret; |
602adf40 | 3406 | |
cc4a38bd | 3407 | ret = rbd_dev_refresh(rbd_dev); |
b813623a AE |
3408 | |
3409 | return ret < 0 ? ret : size; | |
dfc5606d | 3410 | } |
602adf40 | 3411 | |
dfc5606d | 3412 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3413 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d YS |
3414 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
3415 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); | |
3416 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3417 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3418 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3419 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3420 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3421 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3422 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3423 | |
3424 | static struct attribute *rbd_attrs[] = { | |
3425 | &dev_attr_size.attr, | |
34b13184 | 3426 | &dev_attr_features.attr, |
dfc5606d YS |
3427 | &dev_attr_major.attr, |
3428 | &dev_attr_client_id.attr, | |
3429 | &dev_attr_pool.attr, | |
9bb2f334 | 3430 | &dev_attr_pool_id.attr, |
dfc5606d | 3431 | &dev_attr_name.attr, |
589d30e0 | 3432 | &dev_attr_image_id.attr, |
dfc5606d | 3433 | &dev_attr_current_snap.attr, |
86b00e0d | 3434 | &dev_attr_parent.attr, |
dfc5606d | 3435 | &dev_attr_refresh.attr, |
dfc5606d YS |
3436 | NULL |
3437 | }; | |
3438 | ||
3439 | static struct attribute_group rbd_attr_group = { | |
3440 | .attrs = rbd_attrs, | |
3441 | }; | |
3442 | ||
3443 | static const struct attribute_group *rbd_attr_groups[] = { | |
3444 | &rbd_attr_group, | |
3445 | NULL | |
3446 | }; | |
3447 | ||
3448 | static void rbd_sysfs_dev_release(struct device *dev) | |
3449 | { | |
3450 | } | |
3451 | ||
3452 | static struct device_type rbd_device_type = { | |
3453 | .name = "rbd", | |
3454 | .groups = rbd_attr_groups, | |
3455 | .release = rbd_sysfs_dev_release, | |
3456 | }; | |
3457 | ||
8b8fb99c AE |
3458 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3459 | { | |
3460 | kref_get(&spec->kref); | |
3461 | ||
3462 | return spec; | |
3463 | } | |
3464 | ||
3465 | static void rbd_spec_free(struct kref *kref); | |
3466 | static void rbd_spec_put(struct rbd_spec *spec) | |
3467 | { | |
3468 | if (spec) | |
3469 | kref_put(&spec->kref, rbd_spec_free); | |
3470 | } | |
3471 | ||
3472 | static struct rbd_spec *rbd_spec_alloc(void) | |
3473 | { | |
3474 | struct rbd_spec *spec; | |
3475 | ||
3476 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3477 | if (!spec) | |
3478 | return NULL; | |
3479 | kref_init(&spec->kref); | |
3480 | ||
8b8fb99c AE |
3481 | return spec; |
3482 | } | |
3483 | ||
3484 | static void rbd_spec_free(struct kref *kref) | |
3485 | { | |
3486 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3487 | ||
3488 | kfree(spec->pool_name); | |
3489 | kfree(spec->image_id); | |
3490 | kfree(spec->image_name); | |
3491 | kfree(spec->snap_name); | |
3492 | kfree(spec); | |
3493 | } | |
3494 | ||
cc344fa1 | 3495 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3496 | struct rbd_spec *spec) |
3497 | { | |
3498 | struct rbd_device *rbd_dev; | |
3499 | ||
3500 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3501 | if (!rbd_dev) | |
3502 | return NULL; | |
3503 | ||
3504 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3505 | rbd_dev->flags = 0; |
c53d5893 | 3506 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
3507 | init_rwsem(&rbd_dev->header_rwsem); |
3508 | ||
3509 | rbd_dev->spec = spec; | |
3510 | rbd_dev->rbd_client = rbdc; | |
3511 | ||
0903e875 AE |
3512 | /* Initialize the layout used for all rbd requests */ |
3513 | ||
3514 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3515 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3516 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3517 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3518 | ||
c53d5893 AE |
3519 | return rbd_dev; |
3520 | } | |
3521 | ||
3522 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3523 | { | |
c53d5893 AE |
3524 | rbd_put_client(rbd_dev->rbd_client); |
3525 | rbd_spec_put(rbd_dev->spec); | |
3526 | kfree(rbd_dev); | |
3527 | } | |
3528 | ||
9d475de5 AE |
3529 | /* |
3530 | * Get the size and object order for an image snapshot, or if | |
3531 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3532 | * image. | |
3533 | */ | |
3534 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3535 | u8 *order, u64 *snap_size) | |
3536 | { | |
3537 | __le64 snapid = cpu_to_le64(snap_id); | |
3538 | int ret; | |
3539 | struct { | |
3540 | u8 order; | |
3541 | __le64 size; | |
3542 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3543 | ||
36be9a76 | 3544 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3545 | "rbd", "get_size", |
4157976b | 3546 | &snapid, sizeof (snapid), |
e2a58ee5 | 3547 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3548 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3549 | if (ret < 0) |
3550 | return ret; | |
57385b51 AE |
3551 | if (ret < sizeof (size_buf)) |
3552 | return -ERANGE; | |
9d475de5 | 3553 | |
c86f86e9 AE |
3554 | if (order) |
3555 | *order = size_buf.order; | |
9d475de5 AE |
3556 | *snap_size = le64_to_cpu(size_buf.size); |
3557 | ||
3558 | dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n", | |
57385b51 AE |
3559 | (unsigned long long)snap_id, (unsigned int)*order, |
3560 | (unsigned long long)*snap_size); | |
9d475de5 AE |
3561 | |
3562 | return 0; | |
3563 | } | |
3564 | ||
3565 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3566 | { | |
3567 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3568 | &rbd_dev->header.obj_order, | |
3569 | &rbd_dev->header.image_size); | |
3570 | } | |
3571 | ||
1e130199 AE |
3572 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3573 | { | |
3574 | void *reply_buf; | |
3575 | int ret; | |
3576 | void *p; | |
3577 | ||
3578 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3579 | if (!reply_buf) | |
3580 | return -ENOMEM; | |
3581 | ||
36be9a76 | 3582 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3583 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3584 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3585 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3586 | if (ret < 0) |
3587 | goto out; | |
3588 | ||
3589 | p = reply_buf; | |
3590 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3591 | p + ret, NULL, GFP_NOIO); |
3592 | ret = 0; | |
1e130199 AE |
3593 | |
3594 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3595 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3596 | rbd_dev->header.object_prefix = NULL; | |
3597 | } else { | |
3598 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3599 | } | |
1e130199 AE |
3600 | out: |
3601 | kfree(reply_buf); | |
3602 | ||
3603 | return ret; | |
3604 | } | |
3605 | ||
b1b5402a AE |
3606 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3607 | u64 *snap_features) | |
3608 | { | |
3609 | __le64 snapid = cpu_to_le64(snap_id); | |
3610 | struct { | |
3611 | __le64 features; | |
3612 | __le64 incompat; | |
4157976b | 3613 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3614 | u64 incompat; |
b1b5402a AE |
3615 | int ret; |
3616 | ||
36be9a76 | 3617 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3618 | "rbd", "get_features", |
4157976b | 3619 | &snapid, sizeof (snapid), |
e2a58ee5 | 3620 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3621 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3622 | if (ret < 0) |
3623 | return ret; | |
57385b51 AE |
3624 | if (ret < sizeof (features_buf)) |
3625 | return -ERANGE; | |
d889140c AE |
3626 | |
3627 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3628 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3629 | return -ENXIO; |
d889140c | 3630 | |
b1b5402a AE |
3631 | *snap_features = le64_to_cpu(features_buf.features); |
3632 | ||
3633 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3634 | (unsigned long long)snap_id, |
3635 | (unsigned long long)*snap_features, | |
3636 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3637 | |
3638 | return 0; | |
3639 | } | |
3640 | ||
3641 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3642 | { | |
3643 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3644 | &rbd_dev->header.features); | |
3645 | } | |
3646 | ||
86b00e0d AE |
3647 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3648 | { | |
3649 | struct rbd_spec *parent_spec; | |
3650 | size_t size; | |
3651 | void *reply_buf = NULL; | |
3652 | __le64 snapid; | |
3653 | void *p; | |
3654 | void *end; | |
3655 | char *image_id; | |
3656 | u64 overlap; | |
86b00e0d AE |
3657 | int ret; |
3658 | ||
3659 | parent_spec = rbd_spec_alloc(); | |
3660 | if (!parent_spec) | |
3661 | return -ENOMEM; | |
3662 | ||
3663 | size = sizeof (__le64) + /* pool_id */ | |
3664 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3665 | sizeof (__le64) + /* snap_id */ | |
3666 | sizeof (__le64); /* overlap */ | |
3667 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3668 | if (!reply_buf) { | |
3669 | ret = -ENOMEM; | |
3670 | goto out_err; | |
3671 | } | |
3672 | ||
3673 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3674 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3675 | "rbd", "get_parent", |
4157976b | 3676 | &snapid, sizeof (snapid), |
e2a58ee5 | 3677 | reply_buf, size); |
36be9a76 | 3678 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3679 | if (ret < 0) |
3680 | goto out_err; | |
3681 | ||
86b00e0d | 3682 | p = reply_buf; |
57385b51 AE |
3683 | end = reply_buf + ret; |
3684 | ret = -ERANGE; | |
86b00e0d AE |
3685 | ceph_decode_64_safe(&p, end, parent_spec->pool_id, out_err); |
3686 | if (parent_spec->pool_id == CEPH_NOPOOL) | |
3687 | goto out; /* No parent? No problem. */ | |
3688 | ||
0903e875 AE |
3689 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3690 | ||
3691 | ret = -EIO; | |
c0cd10db AE |
3692 | if (parent_spec->pool_id > (u64)U32_MAX) { |
3693 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", | |
3694 | (unsigned long long)parent_spec->pool_id, U32_MAX); | |
57385b51 | 3695 | goto out_err; |
c0cd10db | 3696 | } |
0903e875 | 3697 | |
979ed480 | 3698 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3699 | if (IS_ERR(image_id)) { |
3700 | ret = PTR_ERR(image_id); | |
3701 | goto out_err; | |
3702 | } | |
3703 | parent_spec->image_id = image_id; | |
3704 | ceph_decode_64_safe(&p, end, parent_spec->snap_id, out_err); | |
3705 | ceph_decode_64_safe(&p, end, overlap, out_err); | |
3706 | ||
3707 | rbd_dev->parent_overlap = overlap; | |
3708 | rbd_dev->parent_spec = parent_spec; | |
3709 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3710 | out: | |
3711 | ret = 0; | |
3712 | out_err: | |
3713 | kfree(reply_buf); | |
3714 | rbd_spec_put(parent_spec); | |
3715 | ||
3716 | return ret; | |
3717 | } | |
3718 | ||
cc070d59 AE |
3719 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3720 | { | |
3721 | struct { | |
3722 | __le64 stripe_unit; | |
3723 | __le64 stripe_count; | |
3724 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3725 | size_t size = sizeof (striping_info_buf); | |
3726 | void *p; | |
3727 | u64 obj_size; | |
3728 | u64 stripe_unit; | |
3729 | u64 stripe_count; | |
3730 | int ret; | |
3731 | ||
3732 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3733 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 3734 | (char *)&striping_info_buf, size); |
cc070d59 AE |
3735 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
3736 | if (ret < 0) | |
3737 | return ret; | |
3738 | if (ret < size) | |
3739 | return -ERANGE; | |
3740 | ||
3741 | /* | |
3742 | * We don't actually support the "fancy striping" feature | |
3743 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3744 | * defaults the behavior is the same as before. So find | |
3745 | * out, and only fail if the image has non-default values. | |
3746 | */ | |
3747 | ret = -EINVAL; | |
3748 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3749 | p = &striping_info_buf; | |
3750 | stripe_unit = ceph_decode_64(&p); | |
3751 | if (stripe_unit != obj_size) { | |
3752 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
3753 | "(got %llu want %llu)", | |
3754 | stripe_unit, obj_size); | |
3755 | return -EINVAL; | |
3756 | } | |
3757 | stripe_count = ceph_decode_64(&p); | |
3758 | if (stripe_count != 1) { | |
3759 | rbd_warn(rbd_dev, "unsupported stripe count " | |
3760 | "(got %llu want 1)", stripe_count); | |
3761 | return -EINVAL; | |
3762 | } | |
500d0c0f AE |
3763 | rbd_dev->header.stripe_unit = stripe_unit; |
3764 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
3765 | |
3766 | return 0; | |
3767 | } | |
3768 | ||
9e15b77d AE |
3769 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
3770 | { | |
3771 | size_t image_id_size; | |
3772 | char *image_id; | |
3773 | void *p; | |
3774 | void *end; | |
3775 | size_t size; | |
3776 | void *reply_buf = NULL; | |
3777 | size_t len = 0; | |
3778 | char *image_name = NULL; | |
3779 | int ret; | |
3780 | ||
3781 | rbd_assert(!rbd_dev->spec->image_name); | |
3782 | ||
69e7a02f AE |
3783 | len = strlen(rbd_dev->spec->image_id); |
3784 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
3785 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
3786 | if (!image_id) | |
3787 | return NULL; | |
3788 | ||
3789 | p = image_id; | |
4157976b | 3790 | end = image_id + image_id_size; |
57385b51 | 3791 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
3792 | |
3793 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
3794 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3795 | if (!reply_buf) | |
3796 | goto out; | |
3797 | ||
36be9a76 | 3798 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
3799 | "rbd", "dir_get_name", |
3800 | image_id, image_id_size, | |
e2a58ee5 | 3801 | reply_buf, size); |
9e15b77d AE |
3802 | if (ret < 0) |
3803 | goto out; | |
3804 | p = reply_buf; | |
f40eb349 AE |
3805 | end = reply_buf + ret; |
3806 | ||
9e15b77d AE |
3807 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
3808 | if (IS_ERR(image_name)) | |
3809 | image_name = NULL; | |
3810 | else | |
3811 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
3812 | out: | |
3813 | kfree(reply_buf); | |
3814 | kfree(image_id); | |
3815 | ||
3816 | return image_name; | |
3817 | } | |
3818 | ||
2ad3d716 AE |
3819 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
3820 | { | |
3821 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
3822 | const char *snap_name; | |
3823 | u32 which = 0; | |
3824 | ||
3825 | /* Skip over names until we find the one we are looking for */ | |
3826 | ||
3827 | snap_name = rbd_dev->header.snap_names; | |
3828 | while (which < snapc->num_snaps) { | |
3829 | if (!strcmp(name, snap_name)) | |
3830 | return snapc->snaps[which]; | |
3831 | snap_name += strlen(snap_name) + 1; | |
3832 | which++; | |
3833 | } | |
3834 | return CEPH_NOSNAP; | |
3835 | } | |
3836 | ||
3837 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
3838 | { | |
3839 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
3840 | u32 which; | |
3841 | bool found = false; | |
3842 | u64 snap_id; | |
3843 | ||
3844 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
3845 | const char *snap_name; | |
3846 | ||
3847 | snap_id = snapc->snaps[which]; | |
3848 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
3849 | if (IS_ERR(snap_name)) | |
3850 | break; | |
3851 | found = !strcmp(name, snap_name); | |
3852 | kfree(snap_name); | |
3853 | } | |
3854 | return found ? snap_id : CEPH_NOSNAP; | |
3855 | } | |
3856 | ||
3857 | /* | |
3858 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
3859 | * no snapshot by that name is found, or if an error occurs. | |
3860 | */ | |
3861 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
3862 | { | |
3863 | if (rbd_dev->image_format == 1) | |
3864 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
3865 | ||
3866 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
3867 | } | |
3868 | ||
9e15b77d | 3869 | /* |
2e9f7f1c AE |
3870 | * When an rbd image has a parent image, it is identified by the |
3871 | * pool, image, and snapshot ids (not names). This function fills | |
3872 | * in the names for those ids. (It's OK if we can't figure out the | |
3873 | * name for an image id, but the pool and snapshot ids should always | |
3874 | * exist and have names.) All names in an rbd spec are dynamically | |
3875 | * allocated. | |
e1d4213f AE |
3876 | * |
3877 | * When an image being mapped (not a parent) is probed, we have the | |
3878 | * pool name and pool id, image name and image id, and the snapshot | |
3879 | * name. The only thing we're missing is the snapshot id. | |
9e15b77d | 3880 | */ |
2e9f7f1c | 3881 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 3882 | { |
2e9f7f1c AE |
3883 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
3884 | struct rbd_spec *spec = rbd_dev->spec; | |
3885 | const char *pool_name; | |
3886 | const char *image_name; | |
3887 | const char *snap_name; | |
9e15b77d AE |
3888 | int ret; |
3889 | ||
e1d4213f AE |
3890 | /* |
3891 | * An image being mapped will have the pool name (etc.), but | |
3892 | * we need to look up the snapshot id. | |
3893 | */ | |
2e9f7f1c AE |
3894 | if (spec->pool_name) { |
3895 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
2ad3d716 | 3896 | u64 snap_id; |
e1d4213f | 3897 | |
2ad3d716 AE |
3898 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); |
3899 | if (snap_id == CEPH_NOSNAP) | |
e1d4213f | 3900 | return -ENOENT; |
2ad3d716 | 3901 | spec->snap_id = snap_id; |
e1d4213f | 3902 | } else { |
2e9f7f1c | 3903 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
3904 | } |
3905 | ||
3906 | return 0; | |
3907 | } | |
9e15b77d | 3908 | |
2e9f7f1c | 3909 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 3910 | |
2e9f7f1c AE |
3911 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
3912 | if (!pool_name) { | |
3913 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
3914 | return -EIO; |
3915 | } | |
2e9f7f1c AE |
3916 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
3917 | if (!pool_name) | |
9e15b77d AE |
3918 | return -ENOMEM; |
3919 | ||
3920 | /* Fetch the image name; tolerate failure here */ | |
3921 | ||
2e9f7f1c AE |
3922 | image_name = rbd_dev_image_name(rbd_dev); |
3923 | if (!image_name) | |
06ecc6cb | 3924 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 3925 | |
2e9f7f1c | 3926 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 3927 | |
2e9f7f1c | 3928 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
2e9f7f1c AE |
3929 | if (!snap_name) { |
3930 | ret = -ENOMEM; | |
9e15b77d | 3931 | goto out_err; |
2e9f7f1c AE |
3932 | } |
3933 | ||
3934 | spec->pool_name = pool_name; | |
3935 | spec->image_name = image_name; | |
3936 | spec->snap_name = snap_name; | |
9e15b77d AE |
3937 | |
3938 | return 0; | |
3939 | out_err: | |
2e9f7f1c AE |
3940 | kfree(image_name); |
3941 | kfree(pool_name); | |
9e15b77d AE |
3942 | |
3943 | return ret; | |
3944 | } | |
3945 | ||
cc4a38bd | 3946 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
3947 | { |
3948 | size_t size; | |
3949 | int ret; | |
3950 | void *reply_buf; | |
3951 | void *p; | |
3952 | void *end; | |
3953 | u64 seq; | |
3954 | u32 snap_count; | |
3955 | struct ceph_snap_context *snapc; | |
3956 | u32 i; | |
3957 | ||
3958 | /* | |
3959 | * We'll need room for the seq value (maximum snapshot id), | |
3960 | * snapshot count, and array of that many snapshot ids. | |
3961 | * For now we have a fixed upper limit on the number we're | |
3962 | * prepared to receive. | |
3963 | */ | |
3964 | size = sizeof (__le64) + sizeof (__le32) + | |
3965 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
3966 | reply_buf = kzalloc(size, GFP_KERNEL); | |
3967 | if (!reply_buf) | |
3968 | return -ENOMEM; | |
3969 | ||
36be9a76 | 3970 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3971 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 3972 | reply_buf, size); |
36be9a76 | 3973 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
3974 | if (ret < 0) |
3975 | goto out; | |
3976 | ||
35d489f9 | 3977 | p = reply_buf; |
57385b51 AE |
3978 | end = reply_buf + ret; |
3979 | ret = -ERANGE; | |
35d489f9 AE |
3980 | ceph_decode_64_safe(&p, end, seq, out); |
3981 | ceph_decode_32_safe(&p, end, snap_count, out); | |
3982 | ||
3983 | /* | |
3984 | * Make sure the reported number of snapshot ids wouldn't go | |
3985 | * beyond the end of our buffer. But before checking that, | |
3986 | * make sure the computed size of the snapshot context we | |
3987 | * allocate is representable in a size_t. | |
3988 | */ | |
3989 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
3990 | / sizeof (u64)) { | |
3991 | ret = -EINVAL; | |
3992 | goto out; | |
3993 | } | |
3994 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
3995 | goto out; | |
468521c1 | 3996 | ret = 0; |
35d489f9 | 3997 | |
812164f8 | 3998 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
3999 | if (!snapc) { |
4000 | ret = -ENOMEM; | |
4001 | goto out; | |
4002 | } | |
35d489f9 | 4003 | snapc->seq = seq; |
35d489f9 AE |
4004 | for (i = 0; i < snap_count; i++) |
4005 | snapc->snaps[i] = ceph_decode_64(&p); | |
4006 | ||
49ece554 | 4007 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4008 | rbd_dev->header.snapc = snapc; |
4009 | ||
4010 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4011 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4012 | out: |
4013 | kfree(reply_buf); | |
4014 | ||
57385b51 | 4015 | return ret; |
35d489f9 AE |
4016 | } |
4017 | ||
54cac61f AE |
4018 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4019 | u64 snap_id) | |
b8b1e2db AE |
4020 | { |
4021 | size_t size; | |
4022 | void *reply_buf; | |
54cac61f | 4023 | __le64 snapid; |
b8b1e2db AE |
4024 | int ret; |
4025 | void *p; | |
4026 | void *end; | |
b8b1e2db AE |
4027 | char *snap_name; |
4028 | ||
4029 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4030 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4031 | if (!reply_buf) | |
4032 | return ERR_PTR(-ENOMEM); | |
4033 | ||
54cac61f | 4034 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4035 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4036 | "rbd", "get_snapshot_name", |
54cac61f | 4037 | &snapid, sizeof (snapid), |
e2a58ee5 | 4038 | reply_buf, size); |
36be9a76 | 4039 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4040 | if (ret < 0) { |
4041 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4042 | goto out; |
f40eb349 | 4043 | } |
b8b1e2db AE |
4044 | |
4045 | p = reply_buf; | |
f40eb349 | 4046 | end = reply_buf + ret; |
e5c35534 | 4047 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4048 | if (IS_ERR(snap_name)) |
b8b1e2db | 4049 | goto out; |
b8b1e2db | 4050 | |
f40eb349 | 4051 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4052 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4053 | out: |
4054 | kfree(reply_buf); | |
4055 | ||
f40eb349 | 4056 | return snap_name; |
b8b1e2db AE |
4057 | } |
4058 | ||
cc4a38bd | 4059 | static int rbd_dev_v2_refresh(struct rbd_device *rbd_dev) |
117973fb AE |
4060 | { |
4061 | int ret; | |
117973fb AE |
4062 | |
4063 | down_write(&rbd_dev->header_rwsem); | |
4064 | ||
117973fb AE |
4065 | ret = rbd_dev_v2_image_size(rbd_dev); |
4066 | if (ret) | |
4067 | goto out; | |
117973fb AE |
4068 | rbd_update_mapping_size(rbd_dev); |
4069 | ||
cc4a38bd | 4070 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb AE |
4071 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
4072 | if (ret) | |
4073 | goto out; | |
117973fb AE |
4074 | out: |
4075 | up_write(&rbd_dev->header_rwsem); | |
4076 | ||
4077 | return ret; | |
4078 | } | |
4079 | ||
dfc5606d YS |
4080 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4081 | { | |
dfc5606d | 4082 | struct device *dev; |
cd789ab9 | 4083 | int ret; |
dfc5606d YS |
4084 | |
4085 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
dfc5606d | 4086 | |
cd789ab9 | 4087 | dev = &rbd_dev->dev; |
dfc5606d YS |
4088 | dev->bus = &rbd_bus_type; |
4089 | dev->type = &rbd_device_type; | |
4090 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4091 | dev->release = rbd_dev_device_release; |
de71a297 | 4092 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4093 | ret = device_register(dev); |
dfc5606d | 4094 | |
dfc5606d | 4095 | mutex_unlock(&ctl_mutex); |
cd789ab9 | 4096 | |
dfc5606d | 4097 | return ret; |
602adf40 YS |
4098 | } |
4099 | ||
dfc5606d YS |
4100 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4101 | { | |
4102 | device_unregister(&rbd_dev->dev); | |
4103 | } | |
4104 | ||
e2839308 | 4105 | static atomic64_t rbd_dev_id_max = ATOMIC64_INIT(0); |
1ddbe94e AE |
4106 | |
4107 | /* | |
499afd5b AE |
4108 | * Get a unique rbd identifier for the given new rbd_dev, and add |
4109 | * the rbd_dev to the global list. The minimum rbd id is 1. | |
1ddbe94e | 4110 | */ |
e2839308 | 4111 | static void rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4112 | { |
e2839308 | 4113 | rbd_dev->dev_id = atomic64_inc_return(&rbd_dev_id_max); |
499afd5b AE |
4114 | |
4115 | spin_lock(&rbd_dev_list_lock); | |
4116 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4117 | spin_unlock(&rbd_dev_list_lock); | |
e2839308 AE |
4118 | dout("rbd_dev %p given dev id %llu\n", rbd_dev, |
4119 | (unsigned long long) rbd_dev->dev_id); | |
1ddbe94e | 4120 | } |
b7f23c36 | 4121 | |
1ddbe94e | 4122 | /* |
499afd5b AE |
4123 | * Remove an rbd_dev from the global list, and record that its |
4124 | * identifier is no longer in use. | |
1ddbe94e | 4125 | */ |
e2839308 | 4126 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4127 | { |
d184f6bf | 4128 | struct list_head *tmp; |
de71a297 | 4129 | int rbd_id = rbd_dev->dev_id; |
d184f6bf AE |
4130 | int max_id; |
4131 | ||
aafb230e | 4132 | rbd_assert(rbd_id > 0); |
499afd5b | 4133 | |
e2839308 AE |
4134 | dout("rbd_dev %p released dev id %llu\n", rbd_dev, |
4135 | (unsigned long long) rbd_dev->dev_id); | |
499afd5b AE |
4136 | spin_lock(&rbd_dev_list_lock); |
4137 | list_del_init(&rbd_dev->node); | |
d184f6bf AE |
4138 | |
4139 | /* | |
4140 | * If the id being "put" is not the current maximum, there | |
4141 | * is nothing special we need to do. | |
4142 | */ | |
e2839308 | 4143 | if (rbd_id != atomic64_read(&rbd_dev_id_max)) { |
d184f6bf AE |
4144 | spin_unlock(&rbd_dev_list_lock); |
4145 | return; | |
4146 | } | |
4147 | ||
4148 | /* | |
4149 | * We need to update the current maximum id. Search the | |
4150 | * list to find out what it is. We're more likely to find | |
4151 | * the maximum at the end, so search the list backward. | |
4152 | */ | |
4153 | max_id = 0; | |
4154 | list_for_each_prev(tmp, &rbd_dev_list) { | |
4155 | struct rbd_device *rbd_dev; | |
4156 | ||
4157 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
b213e0b1 AE |
4158 | if (rbd_dev->dev_id > max_id) |
4159 | max_id = rbd_dev->dev_id; | |
d184f6bf | 4160 | } |
499afd5b | 4161 | spin_unlock(&rbd_dev_list_lock); |
b7f23c36 | 4162 | |
1ddbe94e | 4163 | /* |
e2839308 | 4164 | * The max id could have been updated by rbd_dev_id_get(), in |
d184f6bf AE |
4165 | * which case it now accurately reflects the new maximum. |
4166 | * Be careful not to overwrite the maximum value in that | |
4167 | * case. | |
1ddbe94e | 4168 | */ |
e2839308 AE |
4169 | atomic64_cmpxchg(&rbd_dev_id_max, rbd_id, max_id); |
4170 | dout(" max dev id has been reset\n"); | |
b7f23c36 AE |
4171 | } |
4172 | ||
e28fff26 AE |
4173 | /* |
4174 | * Skips over white space at *buf, and updates *buf to point to the | |
4175 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4176 | * the token (string of non-white space characters) found. Note |
4177 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4178 | */ |
4179 | static inline size_t next_token(const char **buf) | |
4180 | { | |
4181 | /* | |
4182 | * These are the characters that produce nonzero for | |
4183 | * isspace() in the "C" and "POSIX" locales. | |
4184 | */ | |
4185 | const char *spaces = " \f\n\r\t\v"; | |
4186 | ||
4187 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4188 | ||
4189 | return strcspn(*buf, spaces); /* Return token length */ | |
4190 | } | |
4191 | ||
4192 | /* | |
4193 | * Finds the next token in *buf, and if the provided token buffer is | |
4194 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4195 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4196 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4197 | * |
4198 | * Returns the length of the token found (not including the '\0'). | |
4199 | * Return value will be 0 if no token is found, and it will be >= | |
4200 | * token_size if the token would not fit. | |
4201 | * | |
593a9e7b | 4202 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4203 | * found token. Note that this occurs even if the token buffer is |
4204 | * too small to hold it. | |
4205 | */ | |
4206 | static inline size_t copy_token(const char **buf, | |
4207 | char *token, | |
4208 | size_t token_size) | |
4209 | { | |
4210 | size_t len; | |
4211 | ||
4212 | len = next_token(buf); | |
4213 | if (len < token_size) { | |
4214 | memcpy(token, *buf, len); | |
4215 | *(token + len) = '\0'; | |
4216 | } | |
4217 | *buf += len; | |
4218 | ||
4219 | return len; | |
4220 | } | |
4221 | ||
ea3352f4 AE |
4222 | /* |
4223 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4224 | * enough to hold a copy of it, and copies the token into the new | |
4225 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4226 | * that a duplicate buffer is created even for a zero-length token. | |
4227 | * | |
4228 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4229 | * pointer if memory for the duplicate was not available. If | |
4230 | * the lenp argument is a non-null pointer, the length of the token | |
4231 | * (not including the '\0') is returned in *lenp. | |
4232 | * | |
4233 | * If successful, the *buf pointer will be updated to point beyond | |
4234 | * the end of the found token. | |
4235 | * | |
4236 | * Note: uses GFP_KERNEL for allocation. | |
4237 | */ | |
4238 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4239 | { | |
4240 | char *dup; | |
4241 | size_t len; | |
4242 | ||
4243 | len = next_token(buf); | |
4caf35f9 | 4244 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4245 | if (!dup) |
4246 | return NULL; | |
ea3352f4 AE |
4247 | *(dup + len) = '\0'; |
4248 | *buf += len; | |
4249 | ||
4250 | if (lenp) | |
4251 | *lenp = len; | |
4252 | ||
4253 | return dup; | |
4254 | } | |
4255 | ||
a725f65e | 4256 | /* |
859c31df AE |
4257 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4258 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4259 | * and the data written is passed here via a NUL-terminated buffer. | |
4260 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4261 | * |
859c31df AE |
4262 | * The information extracted from these options is recorded in |
4263 | * the other parameters which return dynamically-allocated | |
4264 | * structures: | |
4265 | * ceph_opts | |
4266 | * The address of a pointer that will refer to a ceph options | |
4267 | * structure. Caller must release the returned pointer using | |
4268 | * ceph_destroy_options() when it is no longer needed. | |
4269 | * rbd_opts | |
4270 | * Address of an rbd options pointer. Fully initialized by | |
4271 | * this function; caller must release with kfree(). | |
4272 | * spec | |
4273 | * Address of an rbd image specification pointer. Fully | |
4274 | * initialized by this function based on parsed options. | |
4275 | * Caller must release with rbd_spec_put(). | |
4276 | * | |
4277 | * The options passed take this form: | |
4278 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4279 | * where: | |
4280 | * <mon_addrs> | |
4281 | * A comma-separated list of one or more monitor addresses. | |
4282 | * A monitor address is an ip address, optionally followed | |
4283 | * by a port number (separated by a colon). | |
4284 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4285 | * <options> | |
4286 | * A comma-separated list of ceph and/or rbd options. | |
4287 | * <pool_name> | |
4288 | * The name of the rados pool containing the rbd image. | |
4289 | * <image_name> | |
4290 | * The name of the image in that pool to map. | |
4291 | * <snap_id> | |
4292 | * An optional snapshot id. If provided, the mapping will | |
4293 | * present data from the image at the time that snapshot was | |
4294 | * created. The image head is used if no snapshot id is | |
4295 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4296 | */ |
859c31df | 4297 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4298 | struct ceph_options **ceph_opts, |
859c31df AE |
4299 | struct rbd_options **opts, |
4300 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4301 | { |
d22f76e7 | 4302 | size_t len; |
859c31df | 4303 | char *options; |
0ddebc0c | 4304 | const char *mon_addrs; |
ecb4dc22 | 4305 | char *snap_name; |
0ddebc0c | 4306 | size_t mon_addrs_size; |
859c31df | 4307 | struct rbd_spec *spec = NULL; |
4e9afeba | 4308 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4309 | struct ceph_options *copts; |
dc79b113 | 4310 | int ret; |
e28fff26 AE |
4311 | |
4312 | /* The first four tokens are required */ | |
4313 | ||
7ef3214a | 4314 | len = next_token(&buf); |
4fb5d671 AE |
4315 | if (!len) { |
4316 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4317 | return -EINVAL; | |
4318 | } | |
0ddebc0c | 4319 | mon_addrs = buf; |
f28e565a | 4320 | mon_addrs_size = len + 1; |
7ef3214a | 4321 | buf += len; |
a725f65e | 4322 | |
dc79b113 | 4323 | ret = -EINVAL; |
f28e565a AE |
4324 | options = dup_token(&buf, NULL); |
4325 | if (!options) | |
dc79b113 | 4326 | return -ENOMEM; |
4fb5d671 AE |
4327 | if (!*options) { |
4328 | rbd_warn(NULL, "no options provided"); | |
4329 | goto out_err; | |
4330 | } | |
e28fff26 | 4331 | |
859c31df AE |
4332 | spec = rbd_spec_alloc(); |
4333 | if (!spec) | |
f28e565a | 4334 | goto out_mem; |
859c31df AE |
4335 | |
4336 | spec->pool_name = dup_token(&buf, NULL); | |
4337 | if (!spec->pool_name) | |
4338 | goto out_mem; | |
4fb5d671 AE |
4339 | if (!*spec->pool_name) { |
4340 | rbd_warn(NULL, "no pool name provided"); | |
4341 | goto out_err; | |
4342 | } | |
e28fff26 | 4343 | |
69e7a02f | 4344 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4345 | if (!spec->image_name) |
f28e565a | 4346 | goto out_mem; |
4fb5d671 AE |
4347 | if (!*spec->image_name) { |
4348 | rbd_warn(NULL, "no image name provided"); | |
4349 | goto out_err; | |
4350 | } | |
d4b125e9 | 4351 | |
f28e565a AE |
4352 | /* |
4353 | * Snapshot name is optional; default is to use "-" | |
4354 | * (indicating the head/no snapshot). | |
4355 | */ | |
3feeb894 | 4356 | len = next_token(&buf); |
820a5f3e | 4357 | if (!len) { |
3feeb894 AE |
4358 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4359 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4360 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4361 | ret = -ENAMETOOLONG; |
f28e565a | 4362 | goto out_err; |
849b4260 | 4363 | } |
ecb4dc22 AE |
4364 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4365 | if (!snap_name) | |
f28e565a | 4366 | goto out_mem; |
ecb4dc22 AE |
4367 | *(snap_name + len) = '\0'; |
4368 | spec->snap_name = snap_name; | |
e5c35534 | 4369 | |
0ddebc0c | 4370 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4371 | |
4e9afeba AE |
4372 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4373 | if (!rbd_opts) | |
4374 | goto out_mem; | |
4375 | ||
4376 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4377 | |
859c31df | 4378 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4379 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4380 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4381 | if (IS_ERR(copts)) { |
4382 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4383 | goto out_err; |
4384 | } | |
859c31df AE |
4385 | kfree(options); |
4386 | ||
4387 | *ceph_opts = copts; | |
4e9afeba | 4388 | *opts = rbd_opts; |
859c31df | 4389 | *rbd_spec = spec; |
0ddebc0c | 4390 | |
dc79b113 | 4391 | return 0; |
f28e565a | 4392 | out_mem: |
dc79b113 | 4393 | ret = -ENOMEM; |
d22f76e7 | 4394 | out_err: |
859c31df AE |
4395 | kfree(rbd_opts); |
4396 | rbd_spec_put(spec); | |
f28e565a | 4397 | kfree(options); |
d22f76e7 | 4398 | |
dc79b113 | 4399 | return ret; |
a725f65e AE |
4400 | } |
4401 | ||
589d30e0 AE |
4402 | /* |
4403 | * An rbd format 2 image has a unique identifier, distinct from the | |
4404 | * name given to it by the user. Internally, that identifier is | |
4405 | * what's used to specify the names of objects related to the image. | |
4406 | * | |
4407 | * A special "rbd id" object is used to map an rbd image name to its | |
4408 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4409 | * with the supplied name. | |
4410 | * | |
4411 | * This function will record the given rbd_dev's image_id field if | |
4412 | * it can be determined, and in that case will return 0. If any | |
4413 | * errors occur a negative errno will be returned and the rbd_dev's | |
4414 | * image_id field will be unchanged (and should be NULL). | |
4415 | */ | |
4416 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4417 | { | |
4418 | int ret; | |
4419 | size_t size; | |
4420 | char *object_name; | |
4421 | void *response; | |
c0fba368 | 4422 | char *image_id; |
2f82ee54 | 4423 | |
2c0d0a10 AE |
4424 | /* |
4425 | * When probing a parent image, the image id is already | |
4426 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4427 | * need to fetch the image id again in this case. We |
4428 | * do still need to set the image format though. | |
2c0d0a10 | 4429 | */ |
c0fba368 AE |
4430 | if (rbd_dev->spec->image_id) { |
4431 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4432 | ||
2c0d0a10 | 4433 | return 0; |
c0fba368 | 4434 | } |
2c0d0a10 | 4435 | |
589d30e0 AE |
4436 | /* |
4437 | * First, see if the format 2 image id file exists, and if | |
4438 | * so, get the image's persistent id from it. | |
4439 | */ | |
69e7a02f | 4440 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4441 | object_name = kmalloc(size, GFP_NOIO); |
4442 | if (!object_name) | |
4443 | return -ENOMEM; | |
0d7dbfce | 4444 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4445 | dout("rbd id object name is %s\n", object_name); |
4446 | ||
4447 | /* Response will be an encoded string, which includes a length */ | |
4448 | ||
4449 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4450 | response = kzalloc(size, GFP_NOIO); | |
4451 | if (!response) { | |
4452 | ret = -ENOMEM; | |
4453 | goto out; | |
4454 | } | |
4455 | ||
c0fba368 AE |
4456 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4457 | ||
36be9a76 | 4458 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4459 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4460 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4461 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4462 | if (ret == -ENOENT) { |
4463 | image_id = kstrdup("", GFP_KERNEL); | |
4464 | ret = image_id ? 0 : -ENOMEM; | |
4465 | if (!ret) | |
4466 | rbd_dev->image_format = 1; | |
4467 | } else if (ret > sizeof (__le32)) { | |
4468 | void *p = response; | |
4469 | ||
4470 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4471 | NULL, GFP_NOIO); |
c0fba368 AE |
4472 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4473 | if (!ret) | |
4474 | rbd_dev->image_format = 2; | |
589d30e0 | 4475 | } else { |
c0fba368 AE |
4476 | ret = -EINVAL; |
4477 | } | |
4478 | ||
4479 | if (!ret) { | |
4480 | rbd_dev->spec->image_id = image_id; | |
4481 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4482 | } |
4483 | out: | |
4484 | kfree(response); | |
4485 | kfree(object_name); | |
4486 | ||
4487 | return ret; | |
4488 | } | |
4489 | ||
6fd48b3b AE |
4490 | /* Undo whatever state changes are made by v1 or v2 image probe */ |
4491 | ||
4492 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) | |
4493 | { | |
4494 | struct rbd_image_header *header; | |
4495 | ||
4496 | rbd_dev_remove_parent(rbd_dev); | |
4497 | rbd_spec_put(rbd_dev->parent_spec); | |
4498 | rbd_dev->parent_spec = NULL; | |
4499 | rbd_dev->parent_overlap = 0; | |
4500 | ||
4501 | /* Free dynamic fields from the header, then zero it out */ | |
4502 | ||
4503 | header = &rbd_dev->header; | |
812164f8 | 4504 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4505 | kfree(header->snap_sizes); |
4506 | kfree(header->snap_names); | |
4507 | kfree(header->object_prefix); | |
4508 | memset(header, 0, sizeof (*header)); | |
4509 | } | |
4510 | ||
a30b71b9 AE |
4511 | static int rbd_dev_v1_probe(struct rbd_device *rbd_dev) |
4512 | { | |
4513 | int ret; | |
a30b71b9 AE |
4514 | |
4515 | /* Populate rbd image metadata */ | |
4516 | ||
4517 | ret = rbd_read_header(rbd_dev, &rbd_dev->header); | |
4518 | if (ret < 0) | |
4519 | goto out_err; | |
86b00e0d AE |
4520 | |
4521 | /* Version 1 images have no parent (no layering) */ | |
4522 | ||
4523 | rbd_dev->parent_spec = NULL; | |
4524 | rbd_dev->parent_overlap = 0; | |
4525 | ||
a30b71b9 AE |
4526 | dout("discovered version 1 image, header name is %s\n", |
4527 | rbd_dev->header_name); | |
4528 | ||
4529 | return 0; | |
4530 | ||
4531 | out_err: | |
4532 | kfree(rbd_dev->header_name); | |
4533 | rbd_dev->header_name = NULL; | |
0d7dbfce AE |
4534 | kfree(rbd_dev->spec->image_id); |
4535 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
4536 | |
4537 | return ret; | |
4538 | } | |
4539 | ||
4540 | static int rbd_dev_v2_probe(struct rbd_device *rbd_dev) | |
4541 | { | |
9d475de5 | 4542 | int ret; |
a30b71b9 | 4543 | |
9d475de5 | 4544 | ret = rbd_dev_v2_image_size(rbd_dev); |
57385b51 | 4545 | if (ret) |
1e130199 AE |
4546 | goto out_err; |
4547 | ||
4548 | /* Get the object prefix (a.k.a. block_name) for the image */ | |
4549 | ||
4550 | ret = rbd_dev_v2_object_prefix(rbd_dev); | |
57385b51 | 4551 | if (ret) |
b1b5402a AE |
4552 | goto out_err; |
4553 | ||
d889140c | 4554 | /* Get the and check features for the image */ |
b1b5402a AE |
4555 | |
4556 | ret = rbd_dev_v2_features(rbd_dev); | |
57385b51 | 4557 | if (ret) |
9d475de5 | 4558 | goto out_err; |
35d489f9 | 4559 | |
86b00e0d AE |
4560 | /* If the image supports layering, get the parent info */ |
4561 | ||
4562 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { | |
4563 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
57385b51 | 4564 | if (ret) |
86b00e0d | 4565 | goto out_err; |
96882f55 AE |
4566 | |
4567 | /* | |
4568 | * Don't print a warning for parent images. We can | |
4569 | * tell this point because we won't know its pool | |
4570 | * name yet (just its pool id). | |
4571 | */ | |
4572 | if (rbd_dev->spec->pool_name) | |
4573 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4574 | "is EXPERIMENTAL!"); | |
86b00e0d AE |
4575 | } |
4576 | ||
cc070d59 AE |
4577 | /* If the image supports fancy striping, get its parameters */ |
4578 | ||
4579 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4580 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4581 | if (ret < 0) | |
4582 | goto out_err; | |
4583 | } | |
4584 | ||
6e14b1a6 AE |
4585 | /* crypto and compression type aren't (yet) supported for v2 images */ |
4586 | ||
4587 | rbd_dev->header.crypt_type = 0; | |
4588 | rbd_dev->header.comp_type = 0; | |
35d489f9 | 4589 | |
6e14b1a6 AE |
4590 | /* Get the snapshot context, plus the header version */ |
4591 | ||
cc4a38bd | 4592 | ret = rbd_dev_v2_snap_context(rbd_dev); |
35d489f9 AE |
4593 | if (ret) |
4594 | goto out_err; | |
6e14b1a6 | 4595 | |
a30b71b9 AE |
4596 | dout("discovered version 2 image, header name is %s\n", |
4597 | rbd_dev->header_name); | |
4598 | ||
35152979 | 4599 | return 0; |
9d475de5 | 4600 | out_err: |
86b00e0d AE |
4601 | rbd_dev->parent_overlap = 0; |
4602 | rbd_spec_put(rbd_dev->parent_spec); | |
4603 | rbd_dev->parent_spec = NULL; | |
9d475de5 AE |
4604 | kfree(rbd_dev->header_name); |
4605 | rbd_dev->header_name = NULL; | |
1e130199 AE |
4606 | kfree(rbd_dev->header.object_prefix); |
4607 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4608 | |
4609 | return ret; | |
a30b71b9 AE |
4610 | } |
4611 | ||
124afba2 | 4612 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4613 | { |
2f82ee54 | 4614 | struct rbd_device *parent = NULL; |
124afba2 AE |
4615 | struct rbd_spec *parent_spec; |
4616 | struct rbd_client *rbdc; | |
4617 | int ret; | |
4618 | ||
4619 | if (!rbd_dev->parent_spec) | |
4620 | return 0; | |
4621 | /* | |
4622 | * We need to pass a reference to the client and the parent | |
4623 | * spec when creating the parent rbd_dev. Images related by | |
4624 | * parent/child relationships always share both. | |
4625 | */ | |
4626 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4627 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4628 | ||
4629 | ret = -ENOMEM; | |
4630 | parent = rbd_dev_create(rbdc, parent_spec); | |
4631 | if (!parent) | |
4632 | goto out_err; | |
4633 | ||
4634 | ret = rbd_dev_image_probe(parent); | |
4635 | if (ret < 0) | |
4636 | goto out_err; | |
4637 | rbd_dev->parent = parent; | |
4638 | ||
4639 | return 0; | |
4640 | out_err: | |
4641 | if (parent) { | |
4642 | rbd_spec_put(rbd_dev->parent_spec); | |
4643 | kfree(rbd_dev->header_name); | |
4644 | rbd_dev_destroy(parent); | |
4645 | } else { | |
4646 | rbd_put_client(rbdc); | |
4647 | rbd_spec_put(parent_spec); | |
4648 | } | |
4649 | ||
4650 | return ret; | |
4651 | } | |
4652 | ||
200a6a8b | 4653 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4654 | { |
83a06263 | 4655 | int ret; |
d1cf5788 AE |
4656 | |
4657 | ret = rbd_dev_mapping_set(rbd_dev); | |
83a06263 | 4658 | if (ret) |
9bb81c9b | 4659 | return ret; |
5de10f3b | 4660 | |
83a06263 AE |
4661 | /* generate unique id: find highest unique id, add one */ |
4662 | rbd_dev_id_get(rbd_dev); | |
4663 | ||
4664 | /* Fill in the device name, now that we have its id. */ | |
4665 | BUILD_BUG_ON(DEV_NAME_LEN | |
4666 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4667 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4668 | ||
4669 | /* Get our block major device number. */ | |
4670 | ||
4671 | ret = register_blkdev(0, rbd_dev->name); | |
4672 | if (ret < 0) | |
4673 | goto err_out_id; | |
4674 | rbd_dev->major = ret; | |
4675 | ||
4676 | /* Set up the blkdev mapping. */ | |
4677 | ||
4678 | ret = rbd_init_disk(rbd_dev); | |
4679 | if (ret) | |
4680 | goto err_out_blkdev; | |
4681 | ||
4682 | ret = rbd_bus_add_dev(rbd_dev); | |
4683 | if (ret) | |
4684 | goto err_out_disk; | |
4685 | ||
83a06263 AE |
4686 | /* Everything's ready. Announce the disk to the world. */ |
4687 | ||
b5156e76 | 4688 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
129b79d4 | 4689 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4690 | add_disk(rbd_dev->disk); |
4691 | ||
4692 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4693 | (unsigned long long) rbd_dev->mapping.size); | |
4694 | ||
4695 | return ret; | |
2f82ee54 | 4696 | |
83a06263 AE |
4697 | err_out_disk: |
4698 | rbd_free_disk(rbd_dev); | |
4699 | err_out_blkdev: | |
4700 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
4701 | err_out_id: | |
4702 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4703 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4704 | |
4705 | return ret; | |
4706 | } | |
4707 | ||
332bb12d AE |
4708 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4709 | { | |
4710 | struct rbd_spec *spec = rbd_dev->spec; | |
4711 | size_t size; | |
4712 | ||
4713 | /* Record the header object name for this rbd image. */ | |
4714 | ||
4715 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4716 | ||
4717 | if (rbd_dev->image_format == 1) | |
4718 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4719 | else | |
4720 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4721 | ||
4722 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4723 | if (!rbd_dev->header_name) | |
4724 | return -ENOMEM; | |
4725 | ||
4726 | if (rbd_dev->image_format == 1) | |
4727 | sprintf(rbd_dev->header_name, "%s%s", | |
4728 | spec->image_name, RBD_SUFFIX); | |
4729 | else | |
4730 | sprintf(rbd_dev->header_name, "%s%s", | |
4731 | RBD_HEADER_PREFIX, spec->image_id); | |
4732 | return 0; | |
4733 | } | |
4734 | ||
200a6a8b AE |
4735 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4736 | { | |
6fd48b3b AE |
4737 | int ret; |
4738 | ||
6fd48b3b AE |
4739 | rbd_dev_unprobe(rbd_dev); |
4740 | ret = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4741 | if (ret) | |
4742 | rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret); | |
200a6a8b | 4743 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4744 | rbd_dev->header_name = NULL; |
4745 | rbd_dev->image_format = 0; | |
4746 | kfree(rbd_dev->spec->image_id); | |
4747 | rbd_dev->spec->image_id = NULL; | |
4748 | ||
200a6a8b AE |
4749 | rbd_dev_destroy(rbd_dev); |
4750 | } | |
4751 | ||
a30b71b9 AE |
4752 | /* |
4753 | * Probe for the existence of the header object for the given rbd | |
4754 | * device. For format 2 images this includes determining the image | |
4755 | * id. | |
4756 | */ | |
71f293e2 | 4757 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4758 | { |
4759 | int ret; | |
b644de2b | 4760 | int tmp; |
a30b71b9 AE |
4761 | |
4762 | /* | |
4763 | * Get the id from the image id object. If it's not a | |
4764 | * format 2 image, we'll get ENOENT back, and we'll assume | |
4765 | * it's a format 1 image. | |
4766 | */ | |
4767 | ret = rbd_dev_image_id(rbd_dev); | |
4768 | if (ret) | |
c0fba368 AE |
4769 | return ret; |
4770 | rbd_assert(rbd_dev->spec->image_id); | |
4771 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4772 | ||
332bb12d AE |
4773 | ret = rbd_dev_header_name(rbd_dev); |
4774 | if (ret) | |
4775 | goto err_out_format; | |
4776 | ||
b644de2b AE |
4777 | ret = rbd_dev_header_watch_sync(rbd_dev, 1); |
4778 | if (ret) | |
4779 | goto out_header_name; | |
4780 | ||
c0fba368 | 4781 | if (rbd_dev->image_format == 1) |
a30b71b9 AE |
4782 | ret = rbd_dev_v1_probe(rbd_dev); |
4783 | else | |
4784 | ret = rbd_dev_v2_probe(rbd_dev); | |
5655c4d9 | 4785 | if (ret) |
b644de2b | 4786 | goto err_out_watch; |
83a06263 | 4787 | |
9bb81c9b AE |
4788 | ret = rbd_dev_spec_update(rbd_dev); |
4789 | if (ret) | |
33dca39f | 4790 | goto err_out_probe; |
9bb81c9b AE |
4791 | |
4792 | ret = rbd_dev_probe_parent(rbd_dev); | |
6fd48b3b AE |
4793 | if (!ret) |
4794 | return 0; | |
83a06263 | 4795 | |
6fd48b3b AE |
4796 | err_out_probe: |
4797 | rbd_dev_unprobe(rbd_dev); | |
b644de2b AE |
4798 | err_out_watch: |
4799 | tmp = rbd_dev_header_watch_sync(rbd_dev, 0); | |
4800 | if (tmp) | |
4801 | rbd_warn(rbd_dev, "unable to tear down watch request\n"); | |
332bb12d AE |
4802 | out_header_name: |
4803 | kfree(rbd_dev->header_name); | |
4804 | rbd_dev->header_name = NULL; | |
4805 | err_out_format: | |
4806 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
4807 | kfree(rbd_dev->spec->image_id); |
4808 | rbd_dev->spec->image_id = NULL; | |
4809 | ||
4810 | dout("probe failed, returning %d\n", ret); | |
4811 | ||
a30b71b9 AE |
4812 | return ret; |
4813 | } | |
4814 | ||
59c2be1e YS |
4815 | static ssize_t rbd_add(struct bus_type *bus, |
4816 | const char *buf, | |
4817 | size_t count) | |
602adf40 | 4818 | { |
cb8627c7 | 4819 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 4820 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 4821 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4822 | struct rbd_spec *spec = NULL; |
9d3997fd | 4823 | struct rbd_client *rbdc; |
27cc2594 AE |
4824 | struct ceph_osd_client *osdc; |
4825 | int rc = -ENOMEM; | |
602adf40 YS |
4826 | |
4827 | if (!try_module_get(THIS_MODULE)) | |
4828 | return -ENODEV; | |
4829 | ||
602adf40 | 4830 | /* parse add command */ |
859c31df | 4831 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 4832 | if (rc < 0) |
bd4ba655 | 4833 | goto err_out_module; |
78cea76e | 4834 | |
9d3997fd AE |
4835 | rbdc = rbd_get_client(ceph_opts); |
4836 | if (IS_ERR(rbdc)) { | |
4837 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 4838 | goto err_out_args; |
9d3997fd | 4839 | } |
c53d5893 | 4840 | ceph_opts = NULL; /* rbd_dev client now owns this */ |
602adf40 | 4841 | |
602adf40 | 4842 | /* pick the pool */ |
9d3997fd | 4843 | osdc = &rbdc->client->osdc; |
859c31df | 4844 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
4845 | if (rc < 0) |
4846 | goto err_out_client; | |
c0cd10db | 4847 | spec->pool_id = (u64)rc; |
859c31df | 4848 | |
0903e875 AE |
4849 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4850 | ||
c0cd10db AE |
4851 | if (spec->pool_id > (u64)U32_MAX) { |
4852 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
4853 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
4854 | rc = -EIO; |
4855 | goto err_out_client; | |
4856 | } | |
4857 | ||
c53d5893 | 4858 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
4859 | if (!rbd_dev) |
4860 | goto err_out_client; | |
c53d5893 AE |
4861 | rbdc = NULL; /* rbd_dev now owns this */ |
4862 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 4863 | |
bd4ba655 | 4864 | rbd_dev->mapping.read_only = rbd_opts->read_only; |
c53d5893 AE |
4865 | kfree(rbd_opts); |
4866 | rbd_opts = NULL; /* done with this */ | |
bd4ba655 | 4867 | |
71f293e2 | 4868 | rc = rbd_dev_image_probe(rbd_dev); |
a30b71b9 | 4869 | if (rc < 0) |
c53d5893 | 4870 | goto err_out_rbd_dev; |
05fd6f6f | 4871 | |
b536f69a AE |
4872 | rc = rbd_dev_device_setup(rbd_dev); |
4873 | if (!rc) | |
4874 | return count; | |
4875 | ||
4876 | rbd_dev_image_release(rbd_dev); | |
c53d5893 AE |
4877 | err_out_rbd_dev: |
4878 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 4879 | err_out_client: |
9d3997fd | 4880 | rbd_put_client(rbdc); |
0ddebc0c | 4881 | err_out_args: |
78cea76e AE |
4882 | if (ceph_opts) |
4883 | ceph_destroy_options(ceph_opts); | |
4e9afeba | 4884 | kfree(rbd_opts); |
859c31df | 4885 | rbd_spec_put(spec); |
bd4ba655 AE |
4886 | err_out_module: |
4887 | module_put(THIS_MODULE); | |
27cc2594 | 4888 | |
602adf40 | 4889 | dout("Error adding device %s\n", buf); |
27cc2594 | 4890 | |
c0cd10db | 4891 | return (ssize_t)rc; |
602adf40 YS |
4892 | } |
4893 | ||
de71a297 | 4894 | static struct rbd_device *__rbd_get_dev(unsigned long dev_id) |
602adf40 YS |
4895 | { |
4896 | struct list_head *tmp; | |
4897 | struct rbd_device *rbd_dev; | |
4898 | ||
e124a82f | 4899 | spin_lock(&rbd_dev_list_lock); |
602adf40 YS |
4900 | list_for_each(tmp, &rbd_dev_list) { |
4901 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
de71a297 | 4902 | if (rbd_dev->dev_id == dev_id) { |
e124a82f | 4903 | spin_unlock(&rbd_dev_list_lock); |
602adf40 | 4904 | return rbd_dev; |
e124a82f | 4905 | } |
602adf40 | 4906 | } |
e124a82f | 4907 | spin_unlock(&rbd_dev_list_lock); |
602adf40 YS |
4908 | return NULL; |
4909 | } | |
4910 | ||
200a6a8b | 4911 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 4912 | { |
593a9e7b | 4913 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 4914 | |
602adf40 | 4915 | rbd_free_disk(rbd_dev); |
200a6a8b AE |
4916 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
4917 | rbd_dev_clear_mapping(rbd_dev); | |
602adf40 | 4918 | unregister_blkdev(rbd_dev->major, rbd_dev->name); |
200a6a8b | 4919 | rbd_dev->major = 0; |
e2839308 | 4920 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 4921 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
4922 | } |
4923 | ||
05a46afd AE |
4924 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
4925 | { | |
ad945fc1 | 4926 | while (rbd_dev->parent) { |
05a46afd AE |
4927 | struct rbd_device *first = rbd_dev; |
4928 | struct rbd_device *second = first->parent; | |
4929 | struct rbd_device *third; | |
4930 | ||
4931 | /* | |
4932 | * Follow to the parent with no grandparent and | |
4933 | * remove it. | |
4934 | */ | |
4935 | while (second && (third = second->parent)) { | |
4936 | first = second; | |
4937 | second = third; | |
4938 | } | |
ad945fc1 | 4939 | rbd_assert(second); |
8ad42cd0 | 4940 | rbd_dev_image_release(second); |
ad945fc1 AE |
4941 | first->parent = NULL; |
4942 | first->parent_overlap = 0; | |
4943 | ||
4944 | rbd_assert(first->parent_spec); | |
05a46afd AE |
4945 | rbd_spec_put(first->parent_spec); |
4946 | first->parent_spec = NULL; | |
05a46afd AE |
4947 | } |
4948 | } | |
4949 | ||
dfc5606d YS |
4950 | static ssize_t rbd_remove(struct bus_type *bus, |
4951 | const char *buf, | |
4952 | size_t count) | |
602adf40 YS |
4953 | { |
4954 | struct rbd_device *rbd_dev = NULL; | |
0d8189e1 | 4955 | int target_id; |
602adf40 | 4956 | unsigned long ul; |
0d8189e1 | 4957 | int ret; |
602adf40 | 4958 | |
0d8189e1 AE |
4959 | ret = strict_strtoul(buf, 10, &ul); |
4960 | if (ret) | |
4961 | return ret; | |
602adf40 YS |
4962 | |
4963 | /* convert to int; abort if we lost anything in the conversion */ | |
4964 | target_id = (int) ul; | |
4965 | if (target_id != ul) | |
4966 | return -EINVAL; | |
4967 | ||
4968 | mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING); | |
4969 | ||
4970 | rbd_dev = __rbd_get_dev(target_id); | |
4971 | if (!rbd_dev) { | |
4972 | ret = -ENOENT; | |
4973 | goto done; | |
42382b70 AE |
4974 | } |
4975 | ||
a14ea269 | 4976 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 4977 | if (rbd_dev->open_count) |
42382b70 | 4978 | ret = -EBUSY; |
b82d167b AE |
4979 | else |
4980 | set_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
a14ea269 | 4981 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 4982 | if (ret < 0) |
42382b70 | 4983 | goto done; |
0d8189e1 | 4984 | ret = count; |
b480815a | 4985 | rbd_bus_del_dev(rbd_dev); |
8ad42cd0 | 4986 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 4987 | module_put(THIS_MODULE); |
602adf40 YS |
4988 | done: |
4989 | mutex_unlock(&ctl_mutex); | |
aafb230e | 4990 | |
602adf40 YS |
4991 | return ret; |
4992 | } | |
4993 | ||
602adf40 YS |
4994 | /* |
4995 | * create control files in sysfs | |
dfc5606d | 4996 | * /sys/bus/rbd/... |
602adf40 YS |
4997 | */ |
4998 | static int rbd_sysfs_init(void) | |
4999 | { | |
dfc5606d | 5000 | int ret; |
602adf40 | 5001 | |
fed4c143 | 5002 | ret = device_register(&rbd_root_dev); |
21079786 | 5003 | if (ret < 0) |
dfc5606d | 5004 | return ret; |
602adf40 | 5005 | |
fed4c143 AE |
5006 | ret = bus_register(&rbd_bus_type); |
5007 | if (ret < 0) | |
5008 | device_unregister(&rbd_root_dev); | |
602adf40 | 5009 | |
602adf40 YS |
5010 | return ret; |
5011 | } | |
5012 | ||
5013 | static void rbd_sysfs_cleanup(void) | |
5014 | { | |
dfc5606d | 5015 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5016 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5017 | } |
5018 | ||
1c2a9dfe AE |
5019 | static int rbd_slab_init(void) |
5020 | { | |
5021 | rbd_assert(!rbd_img_request_cache); | |
5022 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5023 | sizeof (struct rbd_img_request), | |
5024 | __alignof__(struct rbd_img_request), | |
5025 | 0, NULL); | |
868311b1 AE |
5026 | if (!rbd_img_request_cache) |
5027 | return -ENOMEM; | |
5028 | ||
5029 | rbd_assert(!rbd_obj_request_cache); | |
5030 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5031 | sizeof (struct rbd_obj_request), | |
5032 | __alignof__(struct rbd_obj_request), | |
5033 | 0, NULL); | |
78c2a44a AE |
5034 | if (!rbd_obj_request_cache) |
5035 | goto out_err; | |
5036 | ||
5037 | rbd_assert(!rbd_segment_name_cache); | |
5038 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
5039 | MAX_OBJ_NAME_SIZE + 1, 1, 0, NULL); | |
5040 | if (rbd_segment_name_cache) | |
1c2a9dfe | 5041 | return 0; |
78c2a44a AE |
5042 | out_err: |
5043 | if (rbd_obj_request_cache) { | |
5044 | kmem_cache_destroy(rbd_obj_request_cache); | |
5045 | rbd_obj_request_cache = NULL; | |
5046 | } | |
1c2a9dfe | 5047 | |
868311b1 AE |
5048 | kmem_cache_destroy(rbd_img_request_cache); |
5049 | rbd_img_request_cache = NULL; | |
5050 | ||
1c2a9dfe AE |
5051 | return -ENOMEM; |
5052 | } | |
5053 | ||
5054 | static void rbd_slab_exit(void) | |
5055 | { | |
78c2a44a AE |
5056 | rbd_assert(rbd_segment_name_cache); |
5057 | kmem_cache_destroy(rbd_segment_name_cache); | |
5058 | rbd_segment_name_cache = NULL; | |
5059 | ||
868311b1 AE |
5060 | rbd_assert(rbd_obj_request_cache); |
5061 | kmem_cache_destroy(rbd_obj_request_cache); | |
5062 | rbd_obj_request_cache = NULL; | |
5063 | ||
1c2a9dfe AE |
5064 | rbd_assert(rbd_img_request_cache); |
5065 | kmem_cache_destroy(rbd_img_request_cache); | |
5066 | rbd_img_request_cache = NULL; | |
5067 | } | |
5068 | ||
cc344fa1 | 5069 | static int __init rbd_init(void) |
602adf40 YS |
5070 | { |
5071 | int rc; | |
5072 | ||
1e32d34c AE |
5073 | if (!libceph_compatible(NULL)) { |
5074 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
5075 | ||
5076 | return -EINVAL; | |
5077 | } | |
1c2a9dfe | 5078 | rc = rbd_slab_init(); |
602adf40 YS |
5079 | if (rc) |
5080 | return rc; | |
1c2a9dfe AE |
5081 | rc = rbd_sysfs_init(); |
5082 | if (rc) | |
5083 | rbd_slab_exit(); | |
5084 | else | |
5085 | pr_info("loaded " RBD_DRV_NAME_LONG "\n"); | |
5086 | ||
5087 | return rc; | |
602adf40 YS |
5088 | } |
5089 | ||
cc344fa1 | 5090 | static void __exit rbd_exit(void) |
602adf40 YS |
5091 | { |
5092 | rbd_sysfs_cleanup(); | |
1c2a9dfe | 5093 | rbd_slab_exit(); |
602adf40 YS |
5094 | } |
5095 | ||
5096 | module_init(rbd_init); | |
5097 | module_exit(rbd_exit); | |
5098 | ||
5099 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
5100 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
5101 | MODULE_DESCRIPTION("rados block device"); | |
5102 | ||
5103 | /* following authorship retained from original osdblk.c */ | |
5104 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5105 | ||
5106 | MODULE_LICENSE("GPL"); |