]>
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> |
f8a22fc2 | 44 | #include <linux/idr.h> |
bc1ecc65 | 45 | #include <linux/workqueue.h> |
602adf40 YS |
46 | |
47 | #include "rbd_types.h" | |
48 | ||
aafb230e AE |
49 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
50 | ||
593a9e7b AE |
51 | /* |
52 | * The basic unit of block I/O is a sector. It is interpreted in a | |
53 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
54 | * universally 512 bytes. These symbols are just slightly more | |
55 | * meaningful than the bare numbers they represent. | |
56 | */ | |
57 | #define SECTOR_SHIFT 9 | |
58 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
59 | ||
a2acd00e AE |
60 | /* |
61 | * Increment the given counter and return its updated value. | |
62 | * If the counter is already 0 it will not be incremented. | |
63 | * If the counter is already at its maximum value returns | |
64 | * -EINVAL without updating it. | |
65 | */ | |
66 | static int atomic_inc_return_safe(atomic_t *v) | |
67 | { | |
68 | unsigned int counter; | |
69 | ||
70 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
71 | if (counter <= (unsigned int)INT_MAX) | |
72 | return (int)counter; | |
73 | ||
74 | atomic_dec(v); | |
75 | ||
76 | return -EINVAL; | |
77 | } | |
78 | ||
79 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
80 | static int atomic_dec_return_safe(atomic_t *v) | |
81 | { | |
82 | int counter; | |
83 | ||
84 | counter = atomic_dec_return(v); | |
85 | if (counter >= 0) | |
86 | return counter; | |
87 | ||
88 | atomic_inc(v); | |
89 | ||
90 | return -EINVAL; | |
91 | } | |
92 | ||
f0f8cef5 | 93 | #define RBD_DRV_NAME "rbd" |
602adf40 | 94 | |
7e513d43 ID |
95 | #define RBD_MINORS_PER_MAJOR 256 |
96 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 97 | |
d4b125e9 AE |
98 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
99 | #define RBD_MAX_SNAP_NAME_LEN \ | |
100 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
101 | ||
35d489f9 | 102 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
103 | |
104 | #define RBD_SNAP_HEAD_NAME "-" | |
105 | ||
9682fc6d AE |
106 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
107 | ||
9e15b77d AE |
108 | /* This allows a single page to hold an image name sent by OSD */ |
109 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 110 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 111 | |
1e130199 | 112 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 113 | |
d889140c AE |
114 | /* Feature bits */ |
115 | ||
5cbf6f12 AE |
116 | #define RBD_FEATURE_LAYERING (1<<0) |
117 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
118 | #define RBD_FEATURES_ALL \ | |
119 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
120 | |
121 | /* Features supported by this (client software) implementation. */ | |
122 | ||
770eba6e | 123 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 124 | |
81a89793 AE |
125 | /* |
126 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
127 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
128 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
129 | * enough to hold all possible device names. | |
130 | */ | |
602adf40 | 131 | #define DEV_NAME_LEN 32 |
81a89793 | 132 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
133 | |
134 | /* | |
135 | * block device image metadata (in-memory version) | |
136 | */ | |
137 | struct rbd_image_header { | |
f35a4dee | 138 | /* These six fields never change for a given rbd image */ |
849b4260 | 139 | char *object_prefix; |
602adf40 YS |
140 | __u8 obj_order; |
141 | __u8 crypt_type; | |
142 | __u8 comp_type; | |
f35a4dee AE |
143 | u64 stripe_unit; |
144 | u64 stripe_count; | |
145 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 146 | |
f84344f3 AE |
147 | /* The remaining fields need to be updated occasionally */ |
148 | u64 image_size; | |
149 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
150 | char *snap_names; /* format 1 only */ |
151 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
152 | }; |
153 | ||
0d7dbfce AE |
154 | /* |
155 | * An rbd image specification. | |
156 | * | |
157 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
158 | * identify an image. Each rbd_dev structure includes a pointer to |
159 | * an rbd_spec structure that encapsulates this identity. | |
160 | * | |
161 | * Each of the id's in an rbd_spec has an associated name. For a | |
162 | * user-mapped image, the names are supplied and the id's associated | |
163 | * with them are looked up. For a layered image, a parent image is | |
164 | * defined by the tuple, and the names are looked up. | |
165 | * | |
166 | * An rbd_dev structure contains a parent_spec pointer which is | |
167 | * non-null if the image it represents is a child in a layered | |
168 | * image. This pointer will refer to the rbd_spec structure used | |
169 | * by the parent rbd_dev for its own identity (i.e., the structure | |
170 | * is shared between the parent and child). | |
171 | * | |
172 | * Since these structures are populated once, during the discovery | |
173 | * phase of image construction, they are effectively immutable so | |
174 | * we make no effort to synchronize access to them. | |
175 | * | |
176 | * Note that code herein does not assume the image name is known (it | |
177 | * could be a null pointer). | |
0d7dbfce AE |
178 | */ |
179 | struct rbd_spec { | |
180 | u64 pool_id; | |
ecb4dc22 | 181 | const char *pool_name; |
0d7dbfce | 182 | |
ecb4dc22 AE |
183 | const char *image_id; |
184 | const char *image_name; | |
0d7dbfce AE |
185 | |
186 | u64 snap_id; | |
ecb4dc22 | 187 | const char *snap_name; |
0d7dbfce AE |
188 | |
189 | struct kref kref; | |
190 | }; | |
191 | ||
602adf40 | 192 | /* |
f0f8cef5 | 193 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
194 | */ |
195 | struct rbd_client { | |
196 | struct ceph_client *client; | |
197 | struct kref kref; | |
198 | struct list_head node; | |
199 | }; | |
200 | ||
bf0d5f50 AE |
201 | struct rbd_img_request; |
202 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
203 | ||
204 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
205 | ||
206 | struct rbd_obj_request; | |
207 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
208 | ||
9969ebc5 AE |
209 | enum obj_request_type { |
210 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
211 | }; | |
bf0d5f50 | 212 | |
6d2940c8 GZ |
213 | enum obj_operation_type { |
214 | OBJ_OP_WRITE, | |
215 | OBJ_OP_READ, | |
90e98c52 | 216 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
217 | }; |
218 | ||
926f9b3f AE |
219 | enum obj_req_flags { |
220 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 221 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
222 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
223 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
224 | }; |
225 | ||
bf0d5f50 AE |
226 | struct rbd_obj_request { |
227 | const char *object_name; | |
228 | u64 offset; /* object start byte */ | |
229 | u64 length; /* bytes from offset */ | |
926f9b3f | 230 | unsigned long flags; |
bf0d5f50 | 231 | |
c5b5ef6c AE |
232 | /* |
233 | * An object request associated with an image will have its | |
234 | * img_data flag set; a standalone object request will not. | |
235 | * | |
236 | * A standalone object request will have which == BAD_WHICH | |
237 | * and a null obj_request pointer. | |
238 | * | |
239 | * An object request initiated in support of a layered image | |
240 | * object (to check for its existence before a write) will | |
241 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
242 | * | |
243 | * Finally, an object request for rbd image data will have | |
244 | * which != BAD_WHICH, and will have a non-null img_request | |
245 | * pointer. The value of which will be in the range | |
246 | * 0..(img_request->obj_request_count-1). | |
247 | */ | |
248 | union { | |
249 | struct rbd_obj_request *obj_request; /* STAT op */ | |
250 | struct { | |
251 | struct rbd_img_request *img_request; | |
252 | u64 img_offset; | |
253 | /* links for img_request->obj_requests list */ | |
254 | struct list_head links; | |
255 | }; | |
256 | }; | |
bf0d5f50 AE |
257 | u32 which; /* posn image request list */ |
258 | ||
259 | enum obj_request_type type; | |
788e2df3 AE |
260 | union { |
261 | struct bio *bio_list; | |
262 | struct { | |
263 | struct page **pages; | |
264 | u32 page_count; | |
265 | }; | |
266 | }; | |
0eefd470 | 267 | struct page **copyup_pages; |
ebda6408 | 268 | u32 copyup_page_count; |
bf0d5f50 AE |
269 | |
270 | struct ceph_osd_request *osd_req; | |
271 | ||
272 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 273 | int result; |
bf0d5f50 AE |
274 | |
275 | rbd_obj_callback_t callback; | |
788e2df3 | 276 | struct completion completion; |
bf0d5f50 AE |
277 | |
278 | struct kref kref; | |
279 | }; | |
280 | ||
0c425248 | 281 | enum img_req_flags { |
9849e986 AE |
282 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
283 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 284 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 285 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
286 | }; |
287 | ||
bf0d5f50 | 288 | struct rbd_img_request { |
bf0d5f50 AE |
289 | struct rbd_device *rbd_dev; |
290 | u64 offset; /* starting image byte offset */ | |
291 | u64 length; /* byte count from offset */ | |
0c425248 | 292 | unsigned long flags; |
bf0d5f50 | 293 | union { |
9849e986 | 294 | u64 snap_id; /* for reads */ |
bf0d5f50 | 295 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
296 | }; |
297 | union { | |
298 | struct request *rq; /* block request */ | |
299 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 300 | }; |
3d7efd18 | 301 | struct page **copyup_pages; |
ebda6408 | 302 | u32 copyup_page_count; |
bf0d5f50 AE |
303 | spinlock_t completion_lock;/* protects next_completion */ |
304 | u32 next_completion; | |
305 | rbd_img_callback_t callback; | |
55f27e09 | 306 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 307 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
308 | |
309 | u32 obj_request_count; | |
310 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
311 | ||
312 | struct kref kref; | |
313 | }; | |
314 | ||
315 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 316 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 317 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 318 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 319 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 320 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 321 | |
f84344f3 | 322 | struct rbd_mapping { |
99c1f08f | 323 | u64 size; |
34b13184 | 324 | u64 features; |
f84344f3 AE |
325 | bool read_only; |
326 | }; | |
327 | ||
602adf40 YS |
328 | /* |
329 | * a single device | |
330 | */ | |
331 | struct rbd_device { | |
de71a297 | 332 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
333 | |
334 | int major; /* blkdev assigned major */ | |
dd82fff1 | 335 | int minor; |
602adf40 | 336 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 337 | |
a30b71b9 | 338 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
339 | struct rbd_client *rbd_client; |
340 | ||
341 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
342 | ||
bc1ecc65 | 343 | struct list_head rq_queue; /* incoming rq queue */ |
b82d167b | 344 | spinlock_t lock; /* queue, flags, open_count */ |
bc1ecc65 ID |
345 | struct workqueue_struct *rq_wq; |
346 | struct work_struct rq_work; | |
602adf40 YS |
347 | |
348 | struct rbd_image_header header; | |
b82d167b | 349 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 350 | struct rbd_spec *spec; |
602adf40 | 351 | |
0d7dbfce | 352 | char *header_name; |
971f839a | 353 | |
0903e875 AE |
354 | struct ceph_file_layout layout; |
355 | ||
59c2be1e | 356 | struct ceph_osd_event *watch_event; |
975241af | 357 | struct rbd_obj_request *watch_request; |
59c2be1e | 358 | |
86b00e0d AE |
359 | struct rbd_spec *parent_spec; |
360 | u64 parent_overlap; | |
a2acd00e | 361 | atomic_t parent_ref; |
2f82ee54 | 362 | struct rbd_device *parent; |
86b00e0d | 363 | |
c666601a JD |
364 | /* protects updating the header */ |
365 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
366 | |
367 | struct rbd_mapping mapping; | |
602adf40 YS |
368 | |
369 | struct list_head node; | |
dfc5606d | 370 | |
dfc5606d YS |
371 | /* sysfs related */ |
372 | struct device dev; | |
b82d167b | 373 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
374 | }; |
375 | ||
b82d167b AE |
376 | /* |
377 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
378 | * rbd_dev->lock is used to protect access. | |
379 | * | |
380 | * Currently, only the "removing" flag (which is coupled with the | |
381 | * "open_count" field) requires atomic access. | |
382 | */ | |
6d292906 AE |
383 | enum rbd_dev_flags { |
384 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 385 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
386 | }; |
387 | ||
cfbf6377 | 388 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 389 | |
602adf40 | 390 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
391 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
392 | ||
432b8587 AE |
393 | static LIST_HEAD(rbd_client_list); /* clients */ |
394 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 395 | |
78c2a44a AE |
396 | /* Slab caches for frequently-allocated structures */ |
397 | ||
1c2a9dfe | 398 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 399 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 400 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 401 | |
9b60e70b | 402 | static int rbd_major; |
f8a22fc2 ID |
403 | static DEFINE_IDA(rbd_dev_id_ida); |
404 | ||
9b60e70b ID |
405 | /* |
406 | * Default to false for now, as single-major requires >= 0.75 version of | |
407 | * userspace rbd utility. | |
408 | */ | |
409 | static bool single_major = false; | |
410 | module_param(single_major, bool, S_IRUGO); | |
411 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
412 | ||
3d7efd18 AE |
413 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
414 | ||
200a6a8b | 415 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 416 | |
f0f8cef5 AE |
417 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
418 | size_t count); | |
419 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
420 | size_t count); | |
9b60e70b ID |
421 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
422 | size_t count); | |
423 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
424 | size_t count); | |
1f3ef788 | 425 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping); |
a2acd00e | 426 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 427 | |
9b60e70b ID |
428 | static int rbd_dev_id_to_minor(int dev_id) |
429 | { | |
7e513d43 | 430 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
431 | } |
432 | ||
433 | static int minor_to_rbd_dev_id(int minor) | |
434 | { | |
7e513d43 | 435 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
436 | } |
437 | ||
b15a21dd GKH |
438 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
439 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
440 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
441 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
442 | |
443 | static struct attribute *rbd_bus_attrs[] = { | |
444 | &bus_attr_add.attr, | |
445 | &bus_attr_remove.attr, | |
9b60e70b ID |
446 | &bus_attr_add_single_major.attr, |
447 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 448 | NULL, |
f0f8cef5 | 449 | }; |
92c76dc0 ID |
450 | |
451 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
452 | struct attribute *attr, int index) | |
453 | { | |
9b60e70b ID |
454 | if (!single_major && |
455 | (attr == &bus_attr_add_single_major.attr || | |
456 | attr == &bus_attr_remove_single_major.attr)) | |
457 | return 0; | |
458 | ||
92c76dc0 ID |
459 | return attr->mode; |
460 | } | |
461 | ||
462 | static const struct attribute_group rbd_bus_group = { | |
463 | .attrs = rbd_bus_attrs, | |
464 | .is_visible = rbd_bus_is_visible, | |
465 | }; | |
466 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
467 | |
468 | static struct bus_type rbd_bus_type = { | |
469 | .name = "rbd", | |
b15a21dd | 470 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
471 | }; |
472 | ||
473 | static void rbd_root_dev_release(struct device *dev) | |
474 | { | |
475 | } | |
476 | ||
477 | static struct device rbd_root_dev = { | |
478 | .init_name = "rbd", | |
479 | .release = rbd_root_dev_release, | |
480 | }; | |
481 | ||
06ecc6cb AE |
482 | static __printf(2, 3) |
483 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
484 | { | |
485 | struct va_format vaf; | |
486 | va_list args; | |
487 | ||
488 | va_start(args, fmt); | |
489 | vaf.fmt = fmt; | |
490 | vaf.va = &args; | |
491 | ||
492 | if (!rbd_dev) | |
493 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
494 | else if (rbd_dev->disk) | |
495 | printk(KERN_WARNING "%s: %s: %pV\n", | |
496 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
497 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
498 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
499 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
500 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
501 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
502 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
503 | else /* punt */ | |
504 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
505 | RBD_DRV_NAME, rbd_dev, &vaf); | |
506 | va_end(args); | |
507 | } | |
508 | ||
aafb230e AE |
509 | #ifdef RBD_DEBUG |
510 | #define rbd_assert(expr) \ | |
511 | if (unlikely(!(expr))) { \ | |
512 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
513 | "at line %d:\n\n" \ | |
514 | "\trbd_assert(%s);\n\n", \ | |
515 | __func__, __LINE__, #expr); \ | |
516 | BUG(); \ | |
517 | } | |
518 | #else /* !RBD_DEBUG */ | |
519 | # define rbd_assert(expr) ((void) 0) | |
520 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 521 | |
b454e36d | 522 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
523 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
524 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 525 | |
cc4a38bd | 526 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 527 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 528 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 529 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
530 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
531 | u64 snap_id); | |
2ad3d716 AE |
532 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
533 | u8 *order, u64 *snap_size); | |
534 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
535 | u64 *snap_features); | |
536 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 537 | |
602adf40 YS |
538 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
539 | { | |
f0f8cef5 | 540 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 541 | bool removing = false; |
602adf40 | 542 | |
f84344f3 | 543 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
544 | return -EROFS; |
545 | ||
a14ea269 | 546 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
547 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
548 | removing = true; | |
549 | else | |
550 | rbd_dev->open_count++; | |
a14ea269 | 551 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
552 | if (removing) |
553 | return -ENOENT; | |
554 | ||
c3e946ce | 555 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 556 | |
602adf40 YS |
557 | return 0; |
558 | } | |
559 | ||
db2a144b | 560 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
561 | { |
562 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
563 | unsigned long open_count_before; |
564 | ||
a14ea269 | 565 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 566 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 567 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 568 | rbd_assert(open_count_before > 0); |
dfc5606d | 569 | |
c3e946ce | 570 | put_device(&rbd_dev->dev); |
dfc5606d YS |
571 | } |
572 | ||
131fd9f6 GZ |
573 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
574 | { | |
77f33c03 | 575 | int ret = 0; |
131fd9f6 GZ |
576 | int val; |
577 | bool ro; | |
77f33c03 | 578 | bool ro_changed = false; |
131fd9f6 | 579 | |
77f33c03 | 580 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
581 | if (get_user(val, (int __user *)(arg))) |
582 | return -EFAULT; | |
583 | ||
584 | ro = val ? true : false; | |
585 | /* Snapshot doesn't allow to write*/ | |
586 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
587 | return -EROFS; | |
588 | ||
77f33c03 JD |
589 | spin_lock_irq(&rbd_dev->lock); |
590 | /* prevent others open this device */ | |
591 | if (rbd_dev->open_count > 1) { | |
592 | ret = -EBUSY; | |
593 | goto out; | |
594 | } | |
595 | ||
131fd9f6 GZ |
596 | if (rbd_dev->mapping.read_only != ro) { |
597 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 598 | ro_changed = true; |
131fd9f6 GZ |
599 | } |
600 | ||
77f33c03 JD |
601 | out: |
602 | spin_unlock_irq(&rbd_dev->lock); | |
603 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
604 | if (ret == 0 && ro_changed) | |
605 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
606 | ||
607 | return ret; | |
131fd9f6 GZ |
608 | } |
609 | ||
610 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
611 | unsigned int cmd, unsigned long arg) | |
612 | { | |
613 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
614 | int ret = 0; | |
615 | ||
131fd9f6 GZ |
616 | switch (cmd) { |
617 | case BLKROSET: | |
618 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
619 | break; | |
620 | default: | |
621 | ret = -ENOTTY; | |
622 | } | |
623 | ||
131fd9f6 GZ |
624 | return ret; |
625 | } | |
626 | ||
627 | #ifdef CONFIG_COMPAT | |
628 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
629 | unsigned int cmd, unsigned long arg) | |
630 | { | |
631 | return rbd_ioctl(bdev, mode, cmd, arg); | |
632 | } | |
633 | #endif /* CONFIG_COMPAT */ | |
634 | ||
602adf40 YS |
635 | static const struct block_device_operations rbd_bd_ops = { |
636 | .owner = THIS_MODULE, | |
637 | .open = rbd_open, | |
dfc5606d | 638 | .release = rbd_release, |
131fd9f6 GZ |
639 | .ioctl = rbd_ioctl, |
640 | #ifdef CONFIG_COMPAT | |
641 | .compat_ioctl = rbd_compat_ioctl, | |
642 | #endif | |
602adf40 YS |
643 | }; |
644 | ||
645 | /* | |
7262cfca | 646 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 647 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 648 | */ |
f8c38929 | 649 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
650 | { |
651 | struct rbd_client *rbdc; | |
652 | int ret = -ENOMEM; | |
653 | ||
37206ee5 | 654 | dout("%s:\n", __func__); |
602adf40 YS |
655 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
656 | if (!rbdc) | |
657 | goto out_opt; | |
658 | ||
659 | kref_init(&rbdc->kref); | |
660 | INIT_LIST_HEAD(&rbdc->node); | |
661 | ||
43ae4701 | 662 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 663 | if (IS_ERR(rbdc->client)) |
08f75463 | 664 | goto out_rbdc; |
43ae4701 | 665 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
666 | |
667 | ret = ceph_open_session(rbdc->client); | |
668 | if (ret < 0) | |
08f75463 | 669 | goto out_client; |
602adf40 | 670 | |
432b8587 | 671 | spin_lock(&rbd_client_list_lock); |
602adf40 | 672 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 673 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 674 | |
37206ee5 | 675 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 676 | |
602adf40 | 677 | return rbdc; |
08f75463 | 678 | out_client: |
602adf40 | 679 | ceph_destroy_client(rbdc->client); |
08f75463 | 680 | out_rbdc: |
602adf40 YS |
681 | kfree(rbdc); |
682 | out_opt: | |
43ae4701 AE |
683 | if (ceph_opts) |
684 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
685 | dout("%s: error %d\n", __func__, ret); |
686 | ||
28f259b7 | 687 | return ERR_PTR(ret); |
602adf40 YS |
688 | } |
689 | ||
2f82ee54 AE |
690 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
691 | { | |
692 | kref_get(&rbdc->kref); | |
693 | ||
694 | return rbdc; | |
695 | } | |
696 | ||
602adf40 | 697 | /* |
1f7ba331 AE |
698 | * Find a ceph client with specific addr and configuration. If |
699 | * found, bump its reference count. | |
602adf40 | 700 | */ |
1f7ba331 | 701 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
702 | { |
703 | struct rbd_client *client_node; | |
1f7ba331 | 704 | bool found = false; |
602adf40 | 705 | |
43ae4701 | 706 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
707 | return NULL; |
708 | ||
1f7ba331 AE |
709 | spin_lock(&rbd_client_list_lock); |
710 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
711 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
712 | __rbd_get_client(client_node); |
713 | ||
1f7ba331 AE |
714 | found = true; |
715 | break; | |
716 | } | |
717 | } | |
718 | spin_unlock(&rbd_client_list_lock); | |
719 | ||
720 | return found ? client_node : NULL; | |
602adf40 YS |
721 | } |
722 | ||
59c2be1e YS |
723 | /* |
724 | * mount options | |
725 | */ | |
726 | enum { | |
59c2be1e YS |
727 | Opt_last_int, |
728 | /* int args above */ | |
729 | Opt_last_string, | |
730 | /* string args above */ | |
cc0538b6 AE |
731 | Opt_read_only, |
732 | Opt_read_write, | |
733 | /* Boolean args above */ | |
734 | Opt_last_bool, | |
59c2be1e YS |
735 | }; |
736 | ||
43ae4701 | 737 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
738 | /* int args above */ |
739 | /* string args above */ | |
be466c1c | 740 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
741 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
742 | {Opt_read_write, "read_write"}, | |
743 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
744 | /* Boolean args above */ | |
59c2be1e YS |
745 | {-1, NULL} |
746 | }; | |
747 | ||
98571b5a AE |
748 | struct rbd_options { |
749 | bool read_only; | |
750 | }; | |
751 | ||
752 | #define RBD_READ_ONLY_DEFAULT false | |
753 | ||
59c2be1e YS |
754 | static int parse_rbd_opts_token(char *c, void *private) |
755 | { | |
43ae4701 | 756 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
757 | substring_t argstr[MAX_OPT_ARGS]; |
758 | int token, intval, ret; | |
759 | ||
43ae4701 | 760 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
761 | if (token < 0) |
762 | return -EINVAL; | |
763 | ||
764 | if (token < Opt_last_int) { | |
765 | ret = match_int(&argstr[0], &intval); | |
766 | if (ret < 0) { | |
767 | pr_err("bad mount option arg (not int) " | |
768 | "at '%s'\n", c); | |
769 | return ret; | |
770 | } | |
771 | dout("got int token %d val %d\n", token, intval); | |
772 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
773 | dout("got string token %d val %s\n", token, | |
774 | argstr[0].from); | |
cc0538b6 AE |
775 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
776 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
777 | } else { |
778 | dout("got token %d\n", token); | |
779 | } | |
780 | ||
781 | switch (token) { | |
cc0538b6 AE |
782 | case Opt_read_only: |
783 | rbd_opts->read_only = true; | |
784 | break; | |
785 | case Opt_read_write: | |
786 | rbd_opts->read_only = false; | |
787 | break; | |
59c2be1e | 788 | default: |
aafb230e AE |
789 | rbd_assert(false); |
790 | break; | |
59c2be1e YS |
791 | } |
792 | return 0; | |
793 | } | |
794 | ||
6d2940c8 GZ |
795 | static char* obj_op_name(enum obj_operation_type op_type) |
796 | { | |
797 | switch (op_type) { | |
798 | case OBJ_OP_READ: | |
799 | return "read"; | |
800 | case OBJ_OP_WRITE: | |
801 | return "write"; | |
90e98c52 GZ |
802 | case OBJ_OP_DISCARD: |
803 | return "discard"; | |
6d2940c8 GZ |
804 | default: |
805 | return "???"; | |
806 | } | |
807 | } | |
808 | ||
602adf40 YS |
809 | /* |
810 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
811 | * not exist create it. Either way, ceph_opts is consumed by this |
812 | * function. | |
602adf40 | 813 | */ |
9d3997fd | 814 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 815 | { |
f8c38929 | 816 | struct rbd_client *rbdc; |
59c2be1e | 817 | |
cfbf6377 | 818 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 819 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 820 | if (rbdc) /* using an existing client */ |
43ae4701 | 821 | ceph_destroy_options(ceph_opts); |
9d3997fd | 822 | else |
f8c38929 | 823 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 824 | mutex_unlock(&client_mutex); |
602adf40 | 825 | |
9d3997fd | 826 | return rbdc; |
602adf40 YS |
827 | } |
828 | ||
829 | /* | |
830 | * Destroy ceph client | |
d23a4b3f | 831 | * |
432b8587 | 832 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
833 | */ |
834 | static void rbd_client_release(struct kref *kref) | |
835 | { | |
836 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
837 | ||
37206ee5 | 838 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 839 | spin_lock(&rbd_client_list_lock); |
602adf40 | 840 | list_del(&rbdc->node); |
cd9d9f5d | 841 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
842 | |
843 | ceph_destroy_client(rbdc->client); | |
844 | kfree(rbdc); | |
845 | } | |
846 | ||
847 | /* | |
848 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
849 | * it. | |
850 | */ | |
9d3997fd | 851 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 852 | { |
c53d5893 AE |
853 | if (rbdc) |
854 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
855 | } |
856 | ||
a30b71b9 AE |
857 | static bool rbd_image_format_valid(u32 image_format) |
858 | { | |
859 | return image_format == 1 || image_format == 2; | |
860 | } | |
861 | ||
8e94af8e AE |
862 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
863 | { | |
103a150f AE |
864 | size_t size; |
865 | u32 snap_count; | |
866 | ||
867 | /* The header has to start with the magic rbd header text */ | |
868 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
869 | return false; | |
870 | ||
db2388b6 AE |
871 | /* The bio layer requires at least sector-sized I/O */ |
872 | ||
873 | if (ondisk->options.order < SECTOR_SHIFT) | |
874 | return false; | |
875 | ||
876 | /* If we use u64 in a few spots we may be able to loosen this */ | |
877 | ||
878 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
879 | return false; | |
880 | ||
103a150f AE |
881 | /* |
882 | * The size of a snapshot header has to fit in a size_t, and | |
883 | * that limits the number of snapshots. | |
884 | */ | |
885 | snap_count = le32_to_cpu(ondisk->snap_count); | |
886 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
887 | if (snap_count > size / sizeof (__le64)) | |
888 | return false; | |
889 | ||
890 | /* | |
891 | * Not only that, but the size of the entire the snapshot | |
892 | * header must also be representable in a size_t. | |
893 | */ | |
894 | size -= snap_count * sizeof (__le64); | |
895 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
896 | return false; | |
897 | ||
898 | return true; | |
8e94af8e AE |
899 | } |
900 | ||
602adf40 | 901 | /* |
bb23e37a AE |
902 | * Fill an rbd image header with information from the given format 1 |
903 | * on-disk header. | |
602adf40 | 904 | */ |
662518b1 | 905 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 906 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 907 | { |
662518b1 | 908 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
909 | bool first_time = header->object_prefix == NULL; |
910 | struct ceph_snap_context *snapc; | |
911 | char *object_prefix = NULL; | |
912 | char *snap_names = NULL; | |
913 | u64 *snap_sizes = NULL; | |
ccece235 | 914 | u32 snap_count; |
d2bb24e5 | 915 | size_t size; |
bb23e37a | 916 | int ret = -ENOMEM; |
621901d6 | 917 | u32 i; |
602adf40 | 918 | |
bb23e37a | 919 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 920 | |
bb23e37a AE |
921 | if (first_time) { |
922 | size_t len; | |
103a150f | 923 | |
bb23e37a AE |
924 | len = strnlen(ondisk->object_prefix, |
925 | sizeof (ondisk->object_prefix)); | |
926 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
927 | if (!object_prefix) | |
928 | return -ENOMEM; | |
929 | memcpy(object_prefix, ondisk->object_prefix, len); | |
930 | object_prefix[len] = '\0'; | |
931 | } | |
00f1f36f | 932 | |
bb23e37a | 933 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 934 | |
bb23e37a AE |
935 | snap_count = le32_to_cpu(ondisk->snap_count); |
936 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
937 | if (!snapc) | |
938 | goto out_err; | |
939 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 940 | if (snap_count) { |
bb23e37a | 941 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
942 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
943 | ||
bb23e37a | 944 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 945 | |
bb23e37a AE |
946 | if (snap_names_len > (u64)SIZE_MAX) |
947 | goto out_2big; | |
948 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
949 | if (!snap_names) | |
6a52325f AE |
950 | goto out_err; |
951 | ||
bb23e37a | 952 | /* ...as well as the array of their sizes. */ |
621901d6 | 953 | |
d2bb24e5 | 954 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
955 | snap_sizes = kmalloc(size, GFP_KERNEL); |
956 | if (!snap_sizes) | |
6a52325f | 957 | goto out_err; |
bb23e37a | 958 | |
f785cc1d | 959 | /* |
bb23e37a AE |
960 | * Copy the names, and fill in each snapshot's id |
961 | * and size. | |
962 | * | |
99a41ebc | 963 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 964 | * ondisk buffer we're working with has |
f785cc1d AE |
965 | * snap_names_len bytes beyond the end of the |
966 | * snapshot id array, this memcpy() is safe. | |
967 | */ | |
bb23e37a AE |
968 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
969 | snaps = ondisk->snaps; | |
970 | for (i = 0; i < snap_count; i++) { | |
971 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
972 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
973 | } | |
602adf40 | 974 | } |
6a52325f | 975 | |
bb23e37a | 976 | /* We won't fail any more, fill in the header */ |
621901d6 | 977 | |
bb23e37a AE |
978 | if (first_time) { |
979 | header->object_prefix = object_prefix; | |
980 | header->obj_order = ondisk->options.order; | |
981 | header->crypt_type = ondisk->options.crypt_type; | |
982 | header->comp_type = ondisk->options.comp_type; | |
983 | /* The rest aren't used for format 1 images */ | |
984 | header->stripe_unit = 0; | |
985 | header->stripe_count = 0; | |
986 | header->features = 0; | |
602adf40 | 987 | } else { |
662518b1 AE |
988 | ceph_put_snap_context(header->snapc); |
989 | kfree(header->snap_names); | |
990 | kfree(header->snap_sizes); | |
602adf40 | 991 | } |
849b4260 | 992 | |
bb23e37a | 993 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 994 | |
f84344f3 | 995 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
996 | header->snapc = snapc; |
997 | header->snap_names = snap_names; | |
998 | header->snap_sizes = snap_sizes; | |
468521c1 | 999 | |
602adf40 | 1000 | return 0; |
bb23e37a AE |
1001 | out_2big: |
1002 | ret = -EIO; | |
6a52325f | 1003 | out_err: |
bb23e37a AE |
1004 | kfree(snap_sizes); |
1005 | kfree(snap_names); | |
1006 | ceph_put_snap_context(snapc); | |
1007 | kfree(object_prefix); | |
ccece235 | 1008 | |
bb23e37a | 1009 | return ret; |
602adf40 YS |
1010 | } |
1011 | ||
9682fc6d AE |
1012 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1013 | { | |
1014 | const char *snap_name; | |
1015 | ||
1016 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1017 | ||
1018 | /* Skip over names until we find the one we are looking for */ | |
1019 | ||
1020 | snap_name = rbd_dev->header.snap_names; | |
1021 | while (which--) | |
1022 | snap_name += strlen(snap_name) + 1; | |
1023 | ||
1024 | return kstrdup(snap_name, GFP_KERNEL); | |
1025 | } | |
1026 | ||
30d1cff8 AE |
1027 | /* |
1028 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1029 | * Note that result is for snapshots in *descending* order. | |
1030 | */ | |
1031 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1032 | { | |
1033 | u64 snap_id1 = *(u64 *)s1; | |
1034 | u64 snap_id2 = *(u64 *)s2; | |
1035 | ||
1036 | if (snap_id1 < snap_id2) | |
1037 | return 1; | |
1038 | return snap_id1 == snap_id2 ? 0 : -1; | |
1039 | } | |
1040 | ||
1041 | /* | |
1042 | * Search a snapshot context to see if the given snapshot id is | |
1043 | * present. | |
1044 | * | |
1045 | * Returns the position of the snapshot id in the array if it's found, | |
1046 | * or BAD_SNAP_INDEX otherwise. | |
1047 | * | |
1048 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1049 | * reverse order, highest snapshot id first. | |
1050 | */ | |
9682fc6d AE |
1051 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1052 | { | |
1053 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1054 | u64 *found; |
9682fc6d | 1055 | |
30d1cff8 AE |
1056 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1057 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1058 | |
30d1cff8 | 1059 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1060 | } |
1061 | ||
2ad3d716 AE |
1062 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1063 | u64 snap_id) | |
9e15b77d | 1064 | { |
54cac61f | 1065 | u32 which; |
da6a6b63 | 1066 | const char *snap_name; |
9e15b77d | 1067 | |
54cac61f AE |
1068 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1069 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1070 | return ERR_PTR(-ENOENT); |
54cac61f | 1071 | |
da6a6b63 JD |
1072 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1073 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1074 | } |
1075 | ||
1076 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1077 | { | |
9e15b77d AE |
1078 | if (snap_id == CEPH_NOSNAP) |
1079 | return RBD_SNAP_HEAD_NAME; | |
1080 | ||
54cac61f AE |
1081 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1082 | if (rbd_dev->image_format == 1) | |
1083 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1084 | |
54cac61f | 1085 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1086 | } |
1087 | ||
2ad3d716 AE |
1088 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1089 | u64 *snap_size) | |
602adf40 | 1090 | { |
2ad3d716 AE |
1091 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1092 | if (snap_id == CEPH_NOSNAP) { | |
1093 | *snap_size = rbd_dev->header.image_size; | |
1094 | } else if (rbd_dev->image_format == 1) { | |
1095 | u32 which; | |
602adf40 | 1096 | |
2ad3d716 AE |
1097 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1098 | if (which == BAD_SNAP_INDEX) | |
1099 | return -ENOENT; | |
e86924a8 | 1100 | |
2ad3d716 AE |
1101 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1102 | } else { | |
1103 | u64 size = 0; | |
1104 | int ret; | |
1105 | ||
1106 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1107 | if (ret) | |
1108 | return ret; | |
1109 | ||
1110 | *snap_size = size; | |
1111 | } | |
1112 | return 0; | |
602adf40 YS |
1113 | } |
1114 | ||
2ad3d716 AE |
1115 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1116 | u64 *snap_features) | |
602adf40 | 1117 | { |
2ad3d716 AE |
1118 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1119 | if (snap_id == CEPH_NOSNAP) { | |
1120 | *snap_features = rbd_dev->header.features; | |
1121 | } else if (rbd_dev->image_format == 1) { | |
1122 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1123 | } else { |
2ad3d716 AE |
1124 | u64 features = 0; |
1125 | int ret; | |
8b0241f8 | 1126 | |
2ad3d716 AE |
1127 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1128 | if (ret) | |
1129 | return ret; | |
1130 | ||
1131 | *snap_features = features; | |
1132 | } | |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1137 | { | |
8f4b7d98 | 1138 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1139 | u64 size = 0; |
1140 | u64 features = 0; | |
1141 | int ret; | |
1142 | ||
2ad3d716 AE |
1143 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1144 | if (ret) | |
1145 | return ret; | |
1146 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1147 | if (ret) | |
1148 | return ret; | |
1149 | ||
1150 | rbd_dev->mapping.size = size; | |
1151 | rbd_dev->mapping.features = features; | |
1152 | ||
8b0241f8 | 1153 | return 0; |
602adf40 YS |
1154 | } |
1155 | ||
d1cf5788 AE |
1156 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1157 | { | |
1158 | rbd_dev->mapping.size = 0; | |
1159 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1160 | } |
1161 | ||
7d5079aa HS |
1162 | static void rbd_segment_name_free(const char *name) |
1163 | { | |
1164 | /* The explicit cast here is needed to drop the const qualifier */ | |
1165 | ||
1166 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1167 | } | |
1168 | ||
98571b5a | 1169 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1170 | { |
65ccfe21 AE |
1171 | char *name; |
1172 | u64 segment; | |
1173 | int ret; | |
3a96d5cd | 1174 | char *name_format; |
602adf40 | 1175 | |
78c2a44a | 1176 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1177 | if (!name) |
1178 | return NULL; | |
1179 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1180 | name_format = "%s.%012llx"; |
1181 | if (rbd_dev->image_format == 2) | |
1182 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1183 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1184 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1185 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1186 | pr_err("error formatting segment name for #%llu (%d)\n", |
1187 | segment, ret); | |
7d5079aa | 1188 | rbd_segment_name_free(name); |
65ccfe21 AE |
1189 | name = NULL; |
1190 | } | |
602adf40 | 1191 | |
65ccfe21 AE |
1192 | return name; |
1193 | } | |
602adf40 | 1194 | |
65ccfe21 AE |
1195 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1196 | { | |
1197 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1198 | |
65ccfe21 AE |
1199 | return offset & (segment_size - 1); |
1200 | } | |
1201 | ||
1202 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1203 | u64 offset, u64 length) | |
1204 | { | |
1205 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1206 | ||
1207 | offset &= segment_size - 1; | |
1208 | ||
aafb230e | 1209 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1210 | if (offset + length > segment_size) |
1211 | length = segment_size - offset; | |
1212 | ||
1213 | return length; | |
602adf40 YS |
1214 | } |
1215 | ||
029bcbd8 JD |
1216 | /* |
1217 | * returns the size of an object in the image | |
1218 | */ | |
1219 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1220 | { | |
1221 | return 1 << header->obj_order; | |
1222 | } | |
1223 | ||
602adf40 YS |
1224 | /* |
1225 | * bio helpers | |
1226 | */ | |
1227 | ||
1228 | static void bio_chain_put(struct bio *chain) | |
1229 | { | |
1230 | struct bio *tmp; | |
1231 | ||
1232 | while (chain) { | |
1233 | tmp = chain; | |
1234 | chain = chain->bi_next; | |
1235 | bio_put(tmp); | |
1236 | } | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * zeros a bio chain, starting at specific offset | |
1241 | */ | |
1242 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1243 | { | |
7988613b KO |
1244 | struct bio_vec bv; |
1245 | struct bvec_iter iter; | |
602adf40 YS |
1246 | unsigned long flags; |
1247 | void *buf; | |
602adf40 YS |
1248 | int pos = 0; |
1249 | ||
1250 | while (chain) { | |
7988613b KO |
1251 | bio_for_each_segment(bv, chain, iter) { |
1252 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1253 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1254 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1255 | memset(buf + remainder, 0, |
7988613b KO |
1256 | bv.bv_len - remainder); |
1257 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1258 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1259 | } |
7988613b | 1260 | pos += bv.bv_len; |
602adf40 YS |
1261 | } |
1262 | ||
1263 | chain = chain->bi_next; | |
1264 | } | |
1265 | } | |
1266 | ||
b9434c5b AE |
1267 | /* |
1268 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1269 | * starting at the given byte offset from the start of the array and | |
1270 | * continuing up to the given end offset. The pages array is | |
1271 | * assumed to be big enough to hold all bytes up to the end. | |
1272 | */ | |
1273 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1274 | { | |
1275 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1276 | ||
1277 | rbd_assert(end > offset); | |
1278 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1279 | while (offset < end) { | |
1280 | size_t page_offset; | |
1281 | size_t length; | |
1282 | unsigned long flags; | |
1283 | void *kaddr; | |
1284 | ||
491205a8 GU |
1285 | page_offset = offset & ~PAGE_MASK; |
1286 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1287 | local_irq_save(flags); |
1288 | kaddr = kmap_atomic(*page); | |
1289 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1290 | flush_dcache_page(*page); |
b9434c5b AE |
1291 | kunmap_atomic(kaddr); |
1292 | local_irq_restore(flags); | |
1293 | ||
1294 | offset += length; | |
1295 | page++; | |
1296 | } | |
1297 | } | |
1298 | ||
602adf40 | 1299 | /* |
f7760dad AE |
1300 | * Clone a portion of a bio, starting at the given byte offset |
1301 | * and continuing for the number of bytes indicated. | |
602adf40 | 1302 | */ |
f7760dad AE |
1303 | static struct bio *bio_clone_range(struct bio *bio_src, |
1304 | unsigned int offset, | |
1305 | unsigned int len, | |
1306 | gfp_t gfpmask) | |
602adf40 | 1307 | { |
f7760dad AE |
1308 | struct bio *bio; |
1309 | ||
5341a627 | 1310 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1311 | if (!bio) |
1312 | return NULL; /* ENOMEM */ | |
602adf40 | 1313 | |
5341a627 | 1314 | bio_advance(bio, offset); |
4f024f37 | 1315 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1316 | |
1317 | return bio; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Clone a portion of a bio chain, starting at the given byte offset | |
1322 | * into the first bio in the source chain and continuing for the | |
1323 | * number of bytes indicated. The result is another bio chain of | |
1324 | * exactly the given length, or a null pointer on error. | |
1325 | * | |
1326 | * The bio_src and offset parameters are both in-out. On entry they | |
1327 | * refer to the first source bio and the offset into that bio where | |
1328 | * the start of data to be cloned is located. | |
1329 | * | |
1330 | * On return, bio_src is updated to refer to the bio in the source | |
1331 | * chain that contains first un-cloned byte, and *offset will | |
1332 | * contain the offset of that byte within that bio. | |
1333 | */ | |
1334 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1335 | unsigned int *offset, | |
1336 | unsigned int len, | |
1337 | gfp_t gfpmask) | |
1338 | { | |
1339 | struct bio *bi = *bio_src; | |
1340 | unsigned int off = *offset; | |
1341 | struct bio *chain = NULL; | |
1342 | struct bio **end; | |
1343 | ||
1344 | /* Build up a chain of clone bios up to the limit */ | |
1345 | ||
4f024f37 | 1346 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1347 | return NULL; /* Nothing to clone */ |
602adf40 | 1348 | |
f7760dad AE |
1349 | end = &chain; |
1350 | while (len) { | |
1351 | unsigned int bi_size; | |
1352 | struct bio *bio; | |
1353 | ||
f5400b7a AE |
1354 | if (!bi) { |
1355 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1356 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1357 | } |
4f024f37 | 1358 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1359 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1360 | if (!bio) | |
1361 | goto out_err; /* ENOMEM */ | |
1362 | ||
1363 | *end = bio; | |
1364 | end = &bio->bi_next; | |
602adf40 | 1365 | |
f7760dad | 1366 | off += bi_size; |
4f024f37 | 1367 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1368 | bi = bi->bi_next; |
1369 | off = 0; | |
1370 | } | |
1371 | len -= bi_size; | |
1372 | } | |
1373 | *bio_src = bi; | |
1374 | *offset = off; | |
1375 | ||
1376 | return chain; | |
1377 | out_err: | |
1378 | bio_chain_put(chain); | |
602adf40 | 1379 | |
602adf40 YS |
1380 | return NULL; |
1381 | } | |
1382 | ||
926f9b3f AE |
1383 | /* |
1384 | * The default/initial value for all object request flags is 0. For | |
1385 | * each flag, once its value is set to 1 it is never reset to 0 | |
1386 | * again. | |
1387 | */ | |
57acbaa7 | 1388 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1389 | { |
57acbaa7 | 1390 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1391 | struct rbd_device *rbd_dev; |
1392 | ||
57acbaa7 | 1393 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1394 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1395 | obj_request); |
1396 | } | |
1397 | } | |
1398 | ||
57acbaa7 | 1399 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1400 | { |
1401 | smp_mb(); | |
57acbaa7 | 1402 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1403 | } |
1404 | ||
57acbaa7 | 1405 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1406 | { |
57acbaa7 AE |
1407 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1408 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1409 | |
57acbaa7 AE |
1410 | if (obj_request_img_data_test(obj_request)) |
1411 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1412 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1413 | obj_request); |
1414 | } | |
1415 | } | |
1416 | ||
57acbaa7 | 1417 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1418 | { |
1419 | smp_mb(); | |
57acbaa7 | 1420 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1421 | } |
1422 | ||
5679c59f AE |
1423 | /* |
1424 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1425 | * flag. The latter is set based on the "exists" value provided. | |
1426 | * | |
1427 | * Note that for our purposes once an object exists it never goes | |
1428 | * away again. It's possible that the response from two existence | |
1429 | * checks are separated by the creation of the target object, and | |
1430 | * the first ("doesn't exist") response arrives *after* the second | |
1431 | * ("does exist"). In that case we ignore the second one. | |
1432 | */ | |
1433 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1434 | bool exists) | |
1435 | { | |
1436 | if (exists) | |
1437 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1438 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1439 | smp_mb(); | |
1440 | } | |
1441 | ||
1442 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1443 | { | |
1444 | smp_mb(); | |
1445 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1446 | } | |
1447 | ||
1448 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1449 | { | |
1450 | smp_mb(); | |
1451 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1452 | } | |
1453 | ||
9638556a ID |
1454 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1455 | { | |
1456 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1457 | ||
1458 | return obj_request->img_offset < | |
1459 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1460 | } | |
1461 | ||
bf0d5f50 AE |
1462 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1463 | { | |
37206ee5 AE |
1464 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1465 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1466 | kref_get(&obj_request->kref); |
1467 | } | |
1468 | ||
1469 | static void rbd_obj_request_destroy(struct kref *kref); | |
1470 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1471 | { | |
1472 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1473 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1474 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1475 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1476 | } | |
1477 | ||
0f2d5be7 AE |
1478 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1479 | { | |
1480 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1481 | atomic_read(&img_request->kref.refcount)); | |
1482 | kref_get(&img_request->kref); | |
1483 | } | |
1484 | ||
e93f3152 AE |
1485 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1486 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1487 | static void rbd_img_request_destroy(struct kref *kref); |
1488 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1489 | { | |
1490 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1491 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1492 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1493 | if (img_request_child_test(img_request)) |
1494 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1495 | else | |
1496 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1497 | } |
1498 | ||
1499 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1500 | struct rbd_obj_request *obj_request) | |
1501 | { | |
25dcf954 AE |
1502 | rbd_assert(obj_request->img_request == NULL); |
1503 | ||
b155e86c | 1504 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1505 | obj_request->img_request = img_request; |
25dcf954 | 1506 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1507 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1508 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1509 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1510 | img_request->obj_request_count++; |
1511 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1512 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1513 | obj_request->which); | |
bf0d5f50 AE |
1514 | } |
1515 | ||
1516 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1517 | struct rbd_obj_request *obj_request) | |
1518 | { | |
1519 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1520 | |
37206ee5 AE |
1521 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1522 | obj_request->which); | |
bf0d5f50 | 1523 | list_del(&obj_request->links); |
25dcf954 AE |
1524 | rbd_assert(img_request->obj_request_count > 0); |
1525 | img_request->obj_request_count--; | |
1526 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1527 | obj_request->which = BAD_WHICH; | |
6365d33a | 1528 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1529 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1530 | obj_request->img_request = NULL; |
25dcf954 | 1531 | obj_request->callback = NULL; |
bf0d5f50 AE |
1532 | rbd_obj_request_put(obj_request); |
1533 | } | |
1534 | ||
1535 | static bool obj_request_type_valid(enum obj_request_type type) | |
1536 | { | |
1537 | switch (type) { | |
9969ebc5 | 1538 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1539 | case OBJ_REQUEST_BIO: |
788e2df3 | 1540 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1541 | return true; |
1542 | default: | |
1543 | return false; | |
1544 | } | |
1545 | } | |
1546 | ||
bf0d5f50 AE |
1547 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1548 | struct rbd_obj_request *obj_request) | |
1549 | { | |
71c20a06 | 1550 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1551 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1552 | } | |
1553 | ||
71c20a06 ID |
1554 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1555 | { | |
1556 | dout("%s %p\n", __func__, obj_request); | |
1557 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * Wait for an object request to complete. If interrupted, cancel the | |
1562 | * underlying osd request. | |
1563 | */ | |
1564 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1565 | { | |
1566 | int ret; | |
1567 | ||
1568 | dout("%s %p\n", __func__, obj_request); | |
1569 | ||
1570 | ret = wait_for_completion_interruptible(&obj_request->completion); | |
1571 | if (ret < 0) { | |
1572 | dout("%s %p interrupted\n", __func__, obj_request); | |
1573 | rbd_obj_request_end(obj_request); | |
1574 | return ret; | |
1575 | } | |
1576 | ||
1577 | dout("%s %p done\n", __func__, obj_request); | |
1578 | return 0; | |
1579 | } | |
1580 | ||
bf0d5f50 AE |
1581 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1582 | { | |
55f27e09 | 1583 | |
37206ee5 | 1584 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1585 | |
1586 | /* | |
1587 | * If no error occurred, compute the aggregate transfer | |
1588 | * count for the image request. We could instead use | |
1589 | * atomic64_cmpxchg() to update it as each object request | |
1590 | * completes; not clear which way is better off hand. | |
1591 | */ | |
1592 | if (!img_request->result) { | |
1593 | struct rbd_obj_request *obj_request; | |
1594 | u64 xferred = 0; | |
1595 | ||
1596 | for_each_obj_request(img_request, obj_request) | |
1597 | xferred += obj_request->xferred; | |
1598 | img_request->xferred = xferred; | |
1599 | } | |
1600 | ||
bf0d5f50 AE |
1601 | if (img_request->callback) |
1602 | img_request->callback(img_request); | |
1603 | else | |
1604 | rbd_img_request_put(img_request); | |
1605 | } | |
1606 | ||
0c425248 AE |
1607 | /* |
1608 | * The default/initial value for all image request flags is 0. Each | |
1609 | * is conditionally set to 1 at image request initialization time | |
1610 | * and currently never change thereafter. | |
1611 | */ | |
1612 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1613 | { | |
1614 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1615 | smp_mb(); | |
1616 | } | |
1617 | ||
1618 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1619 | { | |
1620 | smp_mb(); | |
1621 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1622 | } | |
1623 | ||
90e98c52 GZ |
1624 | /* |
1625 | * Set the discard flag when the img_request is an discard request | |
1626 | */ | |
1627 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1628 | { | |
1629 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1630 | smp_mb(); | |
1631 | } | |
1632 | ||
1633 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1634 | { | |
1635 | smp_mb(); | |
1636 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1637 | } | |
1638 | ||
9849e986 AE |
1639 | static void img_request_child_set(struct rbd_img_request *img_request) |
1640 | { | |
1641 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1642 | smp_mb(); | |
1643 | } | |
1644 | ||
e93f3152 AE |
1645 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1646 | { | |
1647 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1648 | smp_mb(); | |
1649 | } | |
1650 | ||
9849e986 AE |
1651 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1652 | { | |
1653 | smp_mb(); | |
1654 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1655 | } | |
1656 | ||
d0b2e944 AE |
1657 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1658 | { | |
1659 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1660 | smp_mb(); | |
1661 | } | |
1662 | ||
a2acd00e AE |
1663 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1664 | { | |
1665 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1666 | smp_mb(); | |
1667 | } | |
1668 | ||
d0b2e944 AE |
1669 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1670 | { | |
1671 | smp_mb(); | |
1672 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1673 | } | |
1674 | ||
6e2a4505 AE |
1675 | static void |
1676 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1677 | { | |
b9434c5b AE |
1678 | u64 xferred = obj_request->xferred; |
1679 | u64 length = obj_request->length; | |
1680 | ||
6e2a4505 AE |
1681 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1682 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1683 | xferred, length); |
6e2a4505 | 1684 | /* |
17c1cc1d JD |
1685 | * ENOENT means a hole in the image. We zero-fill the entire |
1686 | * length of the request. A short read also implies zero-fill | |
1687 | * to the end of the request. An error requires the whole | |
1688 | * length of the request to be reported finished with an error | |
1689 | * to the block layer. In each case we update the xferred | |
1690 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1691 | */ |
b9434c5b | 1692 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1693 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1694 | if (obj_request->type == OBJ_REQUEST_BIO) |
1695 | zero_bio_chain(obj_request->bio_list, 0); | |
1696 | else | |
1697 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1698 | obj_request->result = 0; |
b9434c5b AE |
1699 | } else if (xferred < length && !obj_request->result) { |
1700 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1701 | zero_bio_chain(obj_request->bio_list, xferred); | |
1702 | else | |
1703 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1704 | } |
17c1cc1d | 1705 | obj_request->xferred = length; |
6e2a4505 AE |
1706 | obj_request_done_set(obj_request); |
1707 | } | |
1708 | ||
bf0d5f50 AE |
1709 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1710 | { | |
37206ee5 AE |
1711 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1712 | obj_request->callback); | |
bf0d5f50 AE |
1713 | if (obj_request->callback) |
1714 | obj_request->callback(obj_request); | |
788e2df3 AE |
1715 | else |
1716 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1717 | } |
1718 | ||
c47f9371 | 1719 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1720 | { |
1721 | dout("%s: obj %p\n", __func__, obj_request); | |
1722 | obj_request_done_set(obj_request); | |
1723 | } | |
1724 | ||
c47f9371 | 1725 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1726 | { |
57acbaa7 | 1727 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1728 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1729 | bool layered = false; |
1730 | ||
1731 | if (obj_request_img_data_test(obj_request)) { | |
1732 | img_request = obj_request->img_request; | |
1733 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1734 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1735 | } |
8b3e1a56 AE |
1736 | |
1737 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1738 | obj_request, img_request, obj_request->result, | |
1739 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1740 | if (layered && obj_request->result == -ENOENT && |
1741 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1742 | rbd_img_parent_read(obj_request); |
1743 | else if (img_request) | |
6e2a4505 AE |
1744 | rbd_img_obj_request_read_callback(obj_request); |
1745 | else | |
1746 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1747 | } |
1748 | ||
c47f9371 | 1749 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1750 | { |
1b83bef2 SW |
1751 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1752 | obj_request->result, obj_request->length); | |
1753 | /* | |
8b3e1a56 AE |
1754 | * There is no such thing as a successful short write. Set |
1755 | * it to our originally-requested length. | |
1b83bef2 SW |
1756 | */ |
1757 | obj_request->xferred = obj_request->length; | |
07741308 | 1758 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1759 | } |
1760 | ||
90e98c52 GZ |
1761 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1762 | { | |
1763 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1764 | obj_request->result, obj_request->length); | |
1765 | /* | |
1766 | * There is no such thing as a successful short discard. Set | |
1767 | * it to our originally-requested length. | |
1768 | */ | |
1769 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1770 | /* discarding a non-existent object is not a problem */ |
1771 | if (obj_request->result == -ENOENT) | |
1772 | obj_request->result = 0; | |
90e98c52 GZ |
1773 | obj_request_done_set(obj_request); |
1774 | } | |
1775 | ||
fbfab539 AE |
1776 | /* |
1777 | * For a simple stat call there's nothing to do. We'll do more if | |
1778 | * this is part of a write sequence for a layered image. | |
1779 | */ | |
c47f9371 | 1780 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1781 | { |
37206ee5 | 1782 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1783 | obj_request_done_set(obj_request); |
1784 | } | |
1785 | ||
bf0d5f50 AE |
1786 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1787 | struct ceph_msg *msg) | |
1788 | { | |
1789 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1790 | u16 opcode; |
1791 | ||
37206ee5 | 1792 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1793 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1794 | if (obj_request_img_data_test(obj_request)) { |
1795 | rbd_assert(obj_request->img_request); | |
1796 | rbd_assert(obj_request->which != BAD_WHICH); | |
1797 | } else { | |
1798 | rbd_assert(obj_request->which == BAD_WHICH); | |
1799 | } | |
bf0d5f50 | 1800 | |
1b83bef2 SW |
1801 | if (osd_req->r_result < 0) |
1802 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1803 | |
7cc69d42 | 1804 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1805 | |
c47f9371 AE |
1806 | /* |
1807 | * We support a 64-bit length, but ultimately it has to be | |
1808 | * passed to blk_end_request(), which takes an unsigned int. | |
1809 | */ | |
1b83bef2 | 1810 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1811 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1812 | |
79528734 | 1813 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1814 | switch (opcode) { |
1815 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1816 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1817 | break; |
0ccd5926 ID |
1818 | case CEPH_OSD_OP_SETALLOCHINT: |
1819 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE); | |
1820 | /* fall through */ | |
bf0d5f50 | 1821 | case CEPH_OSD_OP_WRITE: |
c47f9371 | 1822 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1823 | break; |
fbfab539 | 1824 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1825 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1826 | break; |
90e98c52 GZ |
1827 | case CEPH_OSD_OP_DELETE: |
1828 | case CEPH_OSD_OP_TRUNCATE: | |
1829 | case CEPH_OSD_OP_ZERO: | |
1830 | rbd_osd_discard_callback(obj_request); | |
1831 | break; | |
36be9a76 | 1832 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1833 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1834 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1835 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1836 | break; |
bf0d5f50 | 1837 | default: |
9584d508 | 1838 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1839 | obj_request->object_name, (unsigned short) opcode); |
1840 | break; | |
1841 | } | |
1842 | ||
07741308 | 1843 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1844 | rbd_obj_request_complete(obj_request); |
1845 | } | |
1846 | ||
9d4df01f | 1847 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1848 | { |
1849 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1850 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1851 | u64 snap_id; |
430c28c3 | 1852 | |
8c042b0d | 1853 | rbd_assert(osd_req != NULL); |
430c28c3 | 1854 | |
9d4df01f | 1855 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1856 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1857 | NULL, snap_id, NULL); |
1858 | } | |
1859 | ||
1860 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1861 | { | |
1862 | struct rbd_img_request *img_request = obj_request->img_request; | |
1863 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1864 | struct ceph_snap_context *snapc; | |
1865 | struct timespec mtime = CURRENT_TIME; | |
1866 | ||
1867 | rbd_assert(osd_req != NULL); | |
1868 | ||
1869 | snapc = img_request ? img_request->snapc : NULL; | |
1870 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1871 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1872 | } |
1873 | ||
0ccd5926 ID |
1874 | /* |
1875 | * Create an osd request. A read request has one osd op (read). | |
1876 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1877 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1878 | * technically osd watch is a write request, hence this distinction.) | |
1879 | */ | |
bf0d5f50 AE |
1880 | static struct ceph_osd_request *rbd_osd_req_create( |
1881 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1882 | enum obj_operation_type op_type, |
deb236b3 | 1883 | unsigned int num_ops, |
430c28c3 | 1884 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1885 | { |
bf0d5f50 AE |
1886 | struct ceph_snap_context *snapc = NULL; |
1887 | struct ceph_osd_client *osdc; | |
1888 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1889 | |
90e98c52 GZ |
1890 | if (obj_request_img_data_test(obj_request) && |
1891 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1892 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1893 | if (op_type == OBJ_OP_WRITE) { |
1894 | rbd_assert(img_request_write_test(img_request)); | |
1895 | } else { | |
1896 | rbd_assert(img_request_discard_test(img_request)); | |
1897 | } | |
6d2940c8 | 1898 | snapc = img_request->snapc; |
bf0d5f50 AE |
1899 | } |
1900 | ||
6d2940c8 | 1901 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1902 | |
1903 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1904 | |
1905 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1906 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1907 | GFP_ATOMIC); | |
bf0d5f50 AE |
1908 | if (!osd_req) |
1909 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1910 | |
90e98c52 | 1911 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1912 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1913 | else |
bf0d5f50 | 1914 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1915 | |
1916 | osd_req->r_callback = rbd_osd_req_callback; | |
1917 | osd_req->r_priv = obj_request; | |
1918 | ||
3c972c95 ID |
1919 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1920 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1921 | |
bf0d5f50 AE |
1922 | return osd_req; |
1923 | } | |
1924 | ||
0eefd470 AE |
1925 | /* |
1926 | * Create a copyup osd request based on the information in the | |
0ccd5926 ID |
1927 | * object request supplied. A copyup request has three osd ops, |
1928 | * a copyup method call, a hint op, and a write op. | |
0eefd470 AE |
1929 | */ |
1930 | static struct ceph_osd_request * | |
1931 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1932 | { | |
1933 | struct rbd_img_request *img_request; | |
1934 | struct ceph_snap_context *snapc; | |
1935 | struct rbd_device *rbd_dev; | |
1936 | struct ceph_osd_client *osdc; | |
1937 | struct ceph_osd_request *osd_req; | |
1938 | ||
1939 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1940 | img_request = obj_request->img_request; | |
1941 | rbd_assert(img_request); | |
1942 | rbd_assert(img_request_write_test(img_request)); | |
1943 | ||
0ccd5926 | 1944 | /* Allocate and initialize the request, for the three ops */ |
0eefd470 AE |
1945 | |
1946 | snapc = img_request->snapc; | |
1947 | rbd_dev = img_request->rbd_dev; | |
1948 | osdc = &rbd_dev->rbd_client->client->osdc; | |
0ccd5926 | 1949 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 3, false, GFP_ATOMIC); |
0eefd470 AE |
1950 | if (!osd_req) |
1951 | return NULL; /* ENOMEM */ | |
1952 | ||
1953 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1954 | osd_req->r_callback = rbd_osd_req_callback; | |
1955 | osd_req->r_priv = obj_request; | |
1956 | ||
3c972c95 ID |
1957 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1958 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 1959 | |
0eefd470 AE |
1960 | return osd_req; |
1961 | } | |
1962 | ||
1963 | ||
bf0d5f50 AE |
1964 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1965 | { | |
1966 | ceph_osdc_put_request(osd_req); | |
1967 | } | |
1968 | ||
1969 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1970 | ||
1971 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1972 | u64 offset, u64 length, | |
1973 | enum obj_request_type type) | |
1974 | { | |
1975 | struct rbd_obj_request *obj_request; | |
1976 | size_t size; | |
1977 | char *name; | |
1978 | ||
1979 | rbd_assert(obj_request_type_valid(type)); | |
1980 | ||
1981 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1982 | name = kmalloc(size, GFP_KERNEL); |
1983 | if (!name) | |
bf0d5f50 AE |
1984 | return NULL; |
1985 | ||
868311b1 | 1986 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1987 | if (!obj_request) { |
1988 | kfree(name); | |
1989 | return NULL; | |
1990 | } | |
1991 | ||
bf0d5f50 AE |
1992 | obj_request->object_name = memcpy(name, object_name, size); |
1993 | obj_request->offset = offset; | |
1994 | obj_request->length = length; | |
926f9b3f | 1995 | obj_request->flags = 0; |
bf0d5f50 AE |
1996 | obj_request->which = BAD_WHICH; |
1997 | obj_request->type = type; | |
1998 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1999 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2000 | kref_init(&obj_request->kref); |
2001 | ||
37206ee5 AE |
2002 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2003 | offset, length, (int)type, obj_request); | |
2004 | ||
bf0d5f50 AE |
2005 | return obj_request; |
2006 | } | |
2007 | ||
2008 | static void rbd_obj_request_destroy(struct kref *kref) | |
2009 | { | |
2010 | struct rbd_obj_request *obj_request; | |
2011 | ||
2012 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2013 | ||
37206ee5 AE |
2014 | dout("%s: obj %p\n", __func__, obj_request); |
2015 | ||
bf0d5f50 AE |
2016 | rbd_assert(obj_request->img_request == NULL); |
2017 | rbd_assert(obj_request->which == BAD_WHICH); | |
2018 | ||
2019 | if (obj_request->osd_req) | |
2020 | rbd_osd_req_destroy(obj_request->osd_req); | |
2021 | ||
2022 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2023 | switch (obj_request->type) { | |
9969ebc5 AE |
2024 | case OBJ_REQUEST_NODATA: |
2025 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2026 | case OBJ_REQUEST_BIO: |
2027 | if (obj_request->bio_list) | |
2028 | bio_chain_put(obj_request->bio_list); | |
2029 | break; | |
788e2df3 AE |
2030 | case OBJ_REQUEST_PAGES: |
2031 | if (obj_request->pages) | |
2032 | ceph_release_page_vector(obj_request->pages, | |
2033 | obj_request->page_count); | |
2034 | break; | |
bf0d5f50 AE |
2035 | } |
2036 | ||
f907ad55 | 2037 | kfree(obj_request->object_name); |
868311b1 AE |
2038 | obj_request->object_name = NULL; |
2039 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2040 | } |
2041 | ||
fb65d228 AE |
2042 | /* It's OK to call this for a device with no parent */ |
2043 | ||
2044 | static void rbd_spec_put(struct rbd_spec *spec); | |
2045 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2046 | { | |
2047 | rbd_dev_remove_parent(rbd_dev); | |
2048 | rbd_spec_put(rbd_dev->parent_spec); | |
2049 | rbd_dev->parent_spec = NULL; | |
2050 | rbd_dev->parent_overlap = 0; | |
2051 | } | |
2052 | ||
a2acd00e AE |
2053 | /* |
2054 | * Parent image reference counting is used to determine when an | |
2055 | * image's parent fields can be safely torn down--after there are no | |
2056 | * more in-flight requests to the parent image. When the last | |
2057 | * reference is dropped, cleaning them up is safe. | |
2058 | */ | |
2059 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2060 | { | |
2061 | int counter; | |
2062 | ||
2063 | if (!rbd_dev->parent_spec) | |
2064 | return; | |
2065 | ||
2066 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2067 | if (counter > 0) | |
2068 | return; | |
2069 | ||
2070 | /* Last reference; clean up parent data structures */ | |
2071 | ||
2072 | if (!counter) | |
2073 | rbd_dev_unparent(rbd_dev); | |
2074 | else | |
9584d508 | 2075 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2076 | } |
2077 | ||
2078 | /* | |
2079 | * If an image has a non-zero parent overlap, get a reference to its | |
2080 | * parent. | |
2081 | * | |
392a9dad AE |
2082 | * We must get the reference before checking for the overlap to |
2083 | * coordinate properly with zeroing the parent overlap in | |
2084 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
2085 | * drop it again if there is no overlap. | |
2086 | * | |
a2acd00e AE |
2087 | * Returns true if the rbd device has a parent with a non-zero |
2088 | * overlap and a reference for it was successfully taken, or | |
2089 | * false otherwise. | |
2090 | */ | |
2091 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2092 | { | |
2093 | int counter; | |
2094 | ||
2095 | if (!rbd_dev->parent_spec) | |
2096 | return false; | |
2097 | ||
2098 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2099 | if (counter > 0 && rbd_dev->parent_overlap) | |
2100 | return true; | |
2101 | ||
2102 | /* Image was flattened, but parent is not yet torn down */ | |
2103 | ||
2104 | if (counter < 0) | |
9584d508 | 2105 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e AE |
2106 | |
2107 | return false; | |
2108 | } | |
2109 | ||
bf0d5f50 AE |
2110 | /* |
2111 | * Caller is responsible for filling in the list of object requests | |
2112 | * that comprises the image request, and the Linux request pointer | |
2113 | * (if there is one). | |
2114 | */ | |
cc344fa1 AE |
2115 | static struct rbd_img_request *rbd_img_request_create( |
2116 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2117 | u64 offset, u64 length, |
6d2940c8 | 2118 | enum obj_operation_type op_type, |
4e752f0a | 2119 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2120 | { |
2121 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2122 | |
7a716aac | 2123 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2124 | if (!img_request) |
2125 | return NULL; | |
2126 | ||
bf0d5f50 AE |
2127 | img_request->rq = NULL; |
2128 | img_request->rbd_dev = rbd_dev; | |
2129 | img_request->offset = offset; | |
2130 | img_request->length = length; | |
0c425248 | 2131 | img_request->flags = 0; |
90e98c52 GZ |
2132 | if (op_type == OBJ_OP_DISCARD) { |
2133 | img_request_discard_set(img_request); | |
2134 | img_request->snapc = snapc; | |
2135 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2136 | img_request_write_set(img_request); |
4e752f0a | 2137 | img_request->snapc = snapc; |
0c425248 | 2138 | } else { |
bf0d5f50 | 2139 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2140 | } |
a2acd00e | 2141 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2142 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2143 | spin_lock_init(&img_request->completion_lock); |
2144 | img_request->next_completion = 0; | |
2145 | img_request->callback = NULL; | |
a5a337d4 | 2146 | img_request->result = 0; |
bf0d5f50 AE |
2147 | img_request->obj_request_count = 0; |
2148 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2149 | kref_init(&img_request->kref); | |
2150 | ||
37206ee5 | 2151 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2152 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2153 | |
bf0d5f50 AE |
2154 | return img_request; |
2155 | } | |
2156 | ||
2157 | static void rbd_img_request_destroy(struct kref *kref) | |
2158 | { | |
2159 | struct rbd_img_request *img_request; | |
2160 | struct rbd_obj_request *obj_request; | |
2161 | struct rbd_obj_request *next_obj_request; | |
2162 | ||
2163 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2164 | ||
37206ee5 AE |
2165 | dout("%s: img %p\n", __func__, img_request); |
2166 | ||
bf0d5f50 AE |
2167 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2168 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2169 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2170 | |
a2acd00e AE |
2171 | if (img_request_layered_test(img_request)) { |
2172 | img_request_layered_clear(img_request); | |
2173 | rbd_dev_parent_put(img_request->rbd_dev); | |
2174 | } | |
2175 | ||
bef95455 JD |
2176 | if (img_request_write_test(img_request) || |
2177 | img_request_discard_test(img_request)) | |
812164f8 | 2178 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2179 | |
1c2a9dfe | 2180 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2181 | } |
2182 | ||
e93f3152 AE |
2183 | static struct rbd_img_request *rbd_parent_request_create( |
2184 | struct rbd_obj_request *obj_request, | |
2185 | u64 img_offset, u64 length) | |
2186 | { | |
2187 | struct rbd_img_request *parent_request; | |
2188 | struct rbd_device *rbd_dev; | |
2189 | ||
2190 | rbd_assert(obj_request->img_request); | |
2191 | rbd_dev = obj_request->img_request->rbd_dev; | |
2192 | ||
4e752f0a | 2193 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2194 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2195 | if (!parent_request) |
2196 | return NULL; | |
2197 | ||
2198 | img_request_child_set(parent_request); | |
2199 | rbd_obj_request_get(obj_request); | |
2200 | parent_request->obj_request = obj_request; | |
2201 | ||
2202 | return parent_request; | |
2203 | } | |
2204 | ||
2205 | static void rbd_parent_request_destroy(struct kref *kref) | |
2206 | { | |
2207 | struct rbd_img_request *parent_request; | |
2208 | struct rbd_obj_request *orig_request; | |
2209 | ||
2210 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2211 | orig_request = parent_request->obj_request; | |
2212 | ||
2213 | parent_request->obj_request = NULL; | |
2214 | rbd_obj_request_put(orig_request); | |
2215 | img_request_child_clear(parent_request); | |
2216 | ||
2217 | rbd_img_request_destroy(kref); | |
2218 | } | |
2219 | ||
1217857f AE |
2220 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2221 | { | |
6365d33a | 2222 | struct rbd_img_request *img_request; |
1217857f AE |
2223 | unsigned int xferred; |
2224 | int result; | |
8b3e1a56 | 2225 | bool more; |
1217857f | 2226 | |
6365d33a AE |
2227 | rbd_assert(obj_request_img_data_test(obj_request)); |
2228 | img_request = obj_request->img_request; | |
2229 | ||
1217857f AE |
2230 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2231 | xferred = (unsigned int)obj_request->xferred; | |
2232 | result = obj_request->result; | |
2233 | if (result) { | |
2234 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2235 | enum obj_operation_type op_type; |
2236 | ||
90e98c52 GZ |
2237 | if (img_request_discard_test(img_request)) |
2238 | op_type = OBJ_OP_DISCARD; | |
2239 | else if (img_request_write_test(img_request)) | |
2240 | op_type = OBJ_OP_WRITE; | |
2241 | else | |
2242 | op_type = OBJ_OP_READ; | |
1217857f | 2243 | |
9584d508 | 2244 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2245 | obj_op_name(op_type), obj_request->length, |
2246 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2247 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2248 | result, xferred); |
2249 | if (!img_request->result) | |
2250 | img_request->result = result; | |
2251 | } | |
2252 | ||
f1a4739f AE |
2253 | /* Image object requests don't own their page array */ |
2254 | ||
2255 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2256 | obj_request->pages = NULL; | |
2257 | obj_request->page_count = 0; | |
2258 | } | |
2259 | ||
8b3e1a56 AE |
2260 | if (img_request_child_test(img_request)) { |
2261 | rbd_assert(img_request->obj_request != NULL); | |
2262 | more = obj_request->which < img_request->obj_request_count - 1; | |
2263 | } else { | |
2264 | rbd_assert(img_request->rq != NULL); | |
2265 | more = blk_end_request(img_request->rq, result, xferred); | |
2266 | } | |
2267 | ||
2268 | return more; | |
1217857f AE |
2269 | } |
2270 | ||
2169238d AE |
2271 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2272 | { | |
2273 | struct rbd_img_request *img_request; | |
2274 | u32 which = obj_request->which; | |
2275 | bool more = true; | |
2276 | ||
6365d33a | 2277 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2278 | img_request = obj_request->img_request; |
2279 | ||
2280 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2281 | rbd_assert(img_request != NULL); | |
2169238d AE |
2282 | rbd_assert(img_request->obj_request_count > 0); |
2283 | rbd_assert(which != BAD_WHICH); | |
2284 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2285 | |
2286 | spin_lock_irq(&img_request->completion_lock); | |
2287 | if (which != img_request->next_completion) | |
2288 | goto out; | |
2289 | ||
2290 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2291 | rbd_assert(more); |
2292 | rbd_assert(which < img_request->obj_request_count); | |
2293 | ||
2294 | if (!obj_request_done_test(obj_request)) | |
2295 | break; | |
1217857f | 2296 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2297 | which++; |
2298 | } | |
2299 | ||
2300 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2301 | img_request->next_completion = which; | |
2302 | out: | |
2303 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2304 | rbd_img_request_put(img_request); |
2169238d AE |
2305 | |
2306 | if (!more) | |
2307 | rbd_img_request_complete(img_request); | |
2308 | } | |
2309 | ||
f1a4739f AE |
2310 | /* |
2311 | * Split up an image request into one or more object requests, each | |
2312 | * to a different object. The "type" parameter indicates whether | |
2313 | * "data_desc" is the pointer to the head of a list of bio | |
2314 | * structures, or the base of a page array. In either case this | |
2315 | * function assumes data_desc describes memory sufficient to hold | |
2316 | * all data described by the image request. | |
2317 | */ | |
2318 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2319 | enum obj_request_type type, | |
2320 | void *data_desc) | |
bf0d5f50 AE |
2321 | { |
2322 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2323 | struct rbd_obj_request *obj_request = NULL; | |
2324 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2325 | struct bio *bio_list = NULL; |
f1a4739f | 2326 | unsigned int bio_offset = 0; |
a158073c | 2327 | struct page **pages = NULL; |
6d2940c8 | 2328 | enum obj_operation_type op_type; |
90e98c52 | 2329 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); |
7da22d29 | 2330 | u64 img_offset; |
90e98c52 | 2331 | u64 img_end; |
bf0d5f50 AE |
2332 | u64 resid; |
2333 | u16 opcode; | |
2334 | ||
f1a4739f AE |
2335 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2336 | (int)type, data_desc); | |
37206ee5 | 2337 | |
7da22d29 | 2338 | img_offset = img_request->offset; |
bf0d5f50 | 2339 | resid = img_request->length; |
4dda41d3 | 2340 | rbd_assert(resid > 0); |
f1a4739f AE |
2341 | |
2342 | if (type == OBJ_REQUEST_BIO) { | |
2343 | bio_list = data_desc; | |
4f024f37 KO |
2344 | rbd_assert(img_offset == |
2345 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2346 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2347 | pages = data_desc; |
2348 | } | |
2349 | ||
bf0d5f50 | 2350 | while (resid) { |
2fa12320 | 2351 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2352 | const char *object_name; |
bf0d5f50 AE |
2353 | u64 offset; |
2354 | u64 length; | |
0ccd5926 | 2355 | unsigned int which = 0; |
bf0d5f50 | 2356 | |
7da22d29 | 2357 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2358 | if (!object_name) |
2359 | goto out_unwind; | |
7da22d29 AE |
2360 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2361 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2362 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2363 | offset, length, type); |
78c2a44a AE |
2364 | /* object request has its own copy of the object name */ |
2365 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2366 | if (!obj_request) |
2367 | goto out_unwind; | |
62054da6 | 2368 | |
03507db6 JD |
2369 | /* |
2370 | * set obj_request->img_request before creating the | |
2371 | * osd_request so that it gets the right snapc | |
2372 | */ | |
2373 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2374 | |
f1a4739f AE |
2375 | if (type == OBJ_REQUEST_BIO) { |
2376 | unsigned int clone_size; | |
2377 | ||
2378 | rbd_assert(length <= (u64)UINT_MAX); | |
2379 | clone_size = (unsigned int)length; | |
2380 | obj_request->bio_list = | |
2381 | bio_chain_clone_range(&bio_list, | |
2382 | &bio_offset, | |
2383 | clone_size, | |
2384 | GFP_ATOMIC); | |
2385 | if (!obj_request->bio_list) | |
62054da6 | 2386 | goto out_unwind; |
90e98c52 | 2387 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2388 | unsigned int page_count; |
2389 | ||
2390 | obj_request->pages = pages; | |
2391 | page_count = (u32)calc_pages_for(offset, length); | |
2392 | obj_request->page_count = page_count; | |
2393 | if ((offset + length) & ~PAGE_MASK) | |
2394 | page_count--; /* more on last page */ | |
2395 | pages += page_count; | |
2396 | } | |
bf0d5f50 | 2397 | |
90e98c52 GZ |
2398 | if (img_request_discard_test(img_request)) { |
2399 | op_type = OBJ_OP_DISCARD; | |
2400 | if (!offset && (length == object_size) | |
2401 | && (!img_request_layered_test(img_request) || | |
2402 | (rbd_dev->parent_overlap <= | |
3c5df893 | 2403 | obj_request->img_offset))) { |
90e98c52 | 2404 | opcode = CEPH_OSD_OP_DELETE; |
3c5df893 | 2405 | } else if ((offset + length == object_size)) { |
90e98c52 | 2406 | opcode = CEPH_OSD_OP_TRUNCATE; |
3c5df893 JD |
2407 | } else { |
2408 | down_read(&rbd_dev->header_rwsem); | |
2409 | img_end = rbd_dev->header.image_size; | |
2410 | up_read(&rbd_dev->header_rwsem); | |
2411 | ||
2412 | if (obj_request->img_offset + length == img_end) | |
2413 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2414 | else | |
2415 | opcode = CEPH_OSD_OP_ZERO; | |
2416 | } | |
90e98c52 | 2417 | } else if (img_request_write_test(img_request)) { |
6d2940c8 GZ |
2418 | op_type = OBJ_OP_WRITE; |
2419 | opcode = CEPH_OSD_OP_WRITE; | |
2420 | } else { | |
2421 | op_type = OBJ_OP_READ; | |
2422 | opcode = CEPH_OSD_OP_READ; | |
2423 | } | |
2424 | ||
2425 | osd_req = rbd_osd_req_create(rbd_dev, op_type, | |
2426 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2427 | obj_request); | |
2fa12320 | 2428 | if (!osd_req) |
62054da6 | 2429 | goto out_unwind; |
2fa12320 | 2430 | obj_request->osd_req = osd_req; |
2169238d | 2431 | obj_request->callback = rbd_img_obj_callback; |
0f2d5be7 | 2432 | rbd_img_request_get(img_request); |
430c28c3 | 2433 | |
6d2940c8 | 2434 | if (op_type == OBJ_OP_WRITE) { |
0ccd5926 ID |
2435 | osd_req_op_alloc_hint_init(osd_req, which, |
2436 | rbd_obj_bytes(&rbd_dev->header), | |
2437 | rbd_obj_bytes(&rbd_dev->header)); | |
2438 | which++; | |
2439 | } | |
2440 | ||
2441 | osd_req_op_extent_init(osd_req, which, opcode, offset, length, | |
2442 | 0, 0); | |
f1a4739f | 2443 | if (type == OBJ_REQUEST_BIO) |
0ccd5926 | 2444 | osd_req_op_extent_osd_data_bio(osd_req, which, |
f1a4739f | 2445 | obj_request->bio_list, length); |
90e98c52 | 2446 | else if (type == OBJ_REQUEST_PAGES) |
0ccd5926 | 2447 | osd_req_op_extent_osd_data_pages(osd_req, which, |
f1a4739f AE |
2448 | obj_request->pages, length, |
2449 | offset & ~PAGE_MASK, false, false); | |
9d4df01f | 2450 | |
90e98c52 GZ |
2451 | /* Discards are also writes */ |
2452 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
9d4df01f AE |
2453 | rbd_osd_req_format_write(obj_request); |
2454 | else | |
2455 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2456 | |
7da22d29 | 2457 | obj_request->img_offset = img_offset; |
bf0d5f50 | 2458 | |
7da22d29 | 2459 | img_offset += length; |
bf0d5f50 AE |
2460 | resid -= length; |
2461 | } | |
2462 | ||
2463 | return 0; | |
2464 | ||
bf0d5f50 AE |
2465 | out_unwind: |
2466 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2467 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2468 | |
2469 | return -ENOMEM; | |
2470 | } | |
2471 | ||
0eefd470 AE |
2472 | static void |
2473 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2474 | { | |
2475 | struct rbd_img_request *img_request; | |
2476 | struct rbd_device *rbd_dev; | |
ebda6408 | 2477 | struct page **pages; |
0eefd470 AE |
2478 | u32 page_count; |
2479 | ||
2480 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2481 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2482 | img_request = obj_request->img_request; | |
2483 | rbd_assert(img_request); | |
2484 | ||
2485 | rbd_dev = img_request->rbd_dev; | |
2486 | rbd_assert(rbd_dev); | |
0eefd470 | 2487 | |
ebda6408 AE |
2488 | pages = obj_request->copyup_pages; |
2489 | rbd_assert(pages != NULL); | |
0eefd470 | 2490 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2491 | page_count = obj_request->copyup_page_count; |
2492 | rbd_assert(page_count); | |
2493 | obj_request->copyup_page_count = 0; | |
2494 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2495 | |
2496 | /* | |
2497 | * We want the transfer count to reflect the size of the | |
2498 | * original write request. There is no such thing as a | |
2499 | * successful short write, so if the request was successful | |
2500 | * we can just set it to the originally-requested length. | |
2501 | */ | |
2502 | if (!obj_request->result) | |
2503 | obj_request->xferred = obj_request->length; | |
2504 | ||
2505 | /* Finish up with the normal image object callback */ | |
2506 | ||
2507 | rbd_img_obj_callback(obj_request); | |
2508 | } | |
2509 | ||
3d7efd18 AE |
2510 | static void |
2511 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2512 | { | |
2513 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2514 | struct ceph_osd_request *osd_req; |
2515 | struct ceph_osd_client *osdc; | |
2516 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2517 | struct page **pages; |
ebda6408 | 2518 | u32 page_count; |
bbea1c1a | 2519 | int img_result; |
ebda6408 | 2520 | u64 parent_length; |
b91f09f1 AE |
2521 | u64 offset; |
2522 | u64 length; | |
3d7efd18 AE |
2523 | |
2524 | rbd_assert(img_request_child_test(img_request)); | |
2525 | ||
2526 | /* First get what we need from the image request */ | |
2527 | ||
2528 | pages = img_request->copyup_pages; | |
2529 | rbd_assert(pages != NULL); | |
2530 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2531 | page_count = img_request->copyup_page_count; |
2532 | rbd_assert(page_count); | |
2533 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2534 | |
2535 | orig_request = img_request->obj_request; | |
2536 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2537 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2538 | img_result = img_request->result; |
ebda6408 AE |
2539 | parent_length = img_request->length; |
2540 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2541 | rbd_img_request_put(img_request); |
3d7efd18 | 2542 | |
91c6febb AE |
2543 | rbd_assert(orig_request->img_request); |
2544 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2545 | rbd_assert(rbd_dev); |
0eefd470 | 2546 | |
bbea1c1a AE |
2547 | /* |
2548 | * If the overlap has become 0 (most likely because the | |
2549 | * image has been flattened) we need to free the pages | |
2550 | * and re-submit the original write request. | |
2551 | */ | |
2552 | if (!rbd_dev->parent_overlap) { | |
2553 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2554 | |
bbea1c1a AE |
2555 | ceph_release_page_vector(pages, page_count); |
2556 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2557 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2558 | if (!img_result) | |
2559 | return; | |
2560 | } | |
0eefd470 | 2561 | |
bbea1c1a | 2562 | if (img_result) |
0eefd470 | 2563 | goto out_err; |
0eefd470 | 2564 | |
8785b1d4 AE |
2565 | /* |
2566 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2567 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2568 | * request. Allocate the new copyup osd request for the |
2569 | * original request, and release the old one. | |
2570 | */ | |
bbea1c1a | 2571 | img_result = -ENOMEM; |
0eefd470 AE |
2572 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2573 | if (!osd_req) | |
2574 | goto out_err; | |
8785b1d4 | 2575 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2576 | orig_request->osd_req = osd_req; |
2577 | orig_request->copyup_pages = pages; | |
ebda6408 | 2578 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2579 | |
0eefd470 | 2580 | /* Initialize the copyup op */ |
3d7efd18 | 2581 | |
0eefd470 | 2582 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2583 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2584 | false, false); |
3d7efd18 | 2585 | |
0ccd5926 ID |
2586 | /* Then the hint op */ |
2587 | ||
2588 | osd_req_op_alloc_hint_init(osd_req, 1, rbd_obj_bytes(&rbd_dev->header), | |
2589 | rbd_obj_bytes(&rbd_dev->header)); | |
2590 | ||
2591 | /* And the original write request op */ | |
0eefd470 | 2592 | |
b91f09f1 AE |
2593 | offset = orig_request->offset; |
2594 | length = orig_request->length; | |
0ccd5926 | 2595 | osd_req_op_extent_init(osd_req, 2, CEPH_OSD_OP_WRITE, |
b91f09f1 AE |
2596 | offset, length, 0, 0); |
2597 | if (orig_request->type == OBJ_REQUEST_BIO) | |
0ccd5926 | 2598 | osd_req_op_extent_osd_data_bio(osd_req, 2, |
b91f09f1 AE |
2599 | orig_request->bio_list, length); |
2600 | else | |
0ccd5926 | 2601 | osd_req_op_extent_osd_data_pages(osd_req, 2, |
b91f09f1 AE |
2602 | orig_request->pages, length, |
2603 | offset & ~PAGE_MASK, false, false); | |
0eefd470 AE |
2604 | |
2605 | rbd_osd_req_format_write(orig_request); | |
2606 | ||
2607 | /* All set, send it off. */ | |
2608 | ||
2609 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2610 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2611 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2612 | if (!img_result) | |
0eefd470 AE |
2613 | return; |
2614 | out_err: | |
2615 | /* Record the error code and complete the request */ | |
2616 | ||
bbea1c1a | 2617 | orig_request->result = img_result; |
0eefd470 AE |
2618 | orig_request->xferred = 0; |
2619 | obj_request_done_set(orig_request); | |
2620 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2621 | } |
2622 | ||
2623 | /* | |
2624 | * Read from the parent image the range of data that covers the | |
2625 | * entire target of the given object request. This is used for | |
2626 | * satisfying a layered image write request when the target of an | |
2627 | * object request from the image request does not exist. | |
2628 | * | |
2629 | * A page array big enough to hold the returned data is allocated | |
2630 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2631 | * When the read completes, this page array will be transferred to | |
2632 | * the original object request for the copyup operation. | |
2633 | * | |
2634 | * If an error occurs, record it as the result of the original | |
2635 | * object request and mark it done so it gets completed. | |
2636 | */ | |
2637 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2638 | { | |
2639 | struct rbd_img_request *img_request = NULL; | |
2640 | struct rbd_img_request *parent_request = NULL; | |
2641 | struct rbd_device *rbd_dev; | |
2642 | u64 img_offset; | |
2643 | u64 length; | |
2644 | struct page **pages = NULL; | |
2645 | u32 page_count; | |
2646 | int result; | |
2647 | ||
2648 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2649 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2650 | |
2651 | img_request = obj_request->img_request; | |
2652 | rbd_assert(img_request != NULL); | |
2653 | rbd_dev = img_request->rbd_dev; | |
2654 | rbd_assert(rbd_dev->parent != NULL); | |
2655 | ||
2656 | /* | |
2657 | * Determine the byte range covered by the object in the | |
2658 | * child image to which the original request was to be sent. | |
2659 | */ | |
2660 | img_offset = obj_request->img_offset - obj_request->offset; | |
2661 | length = (u64)1 << rbd_dev->header.obj_order; | |
2662 | ||
a9e8ba2c AE |
2663 | /* |
2664 | * There is no defined parent data beyond the parent | |
2665 | * overlap, so limit what we read at that boundary if | |
2666 | * necessary. | |
2667 | */ | |
2668 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2669 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2670 | length = rbd_dev->parent_overlap - img_offset; | |
2671 | } | |
2672 | ||
3d7efd18 AE |
2673 | /* |
2674 | * Allocate a page array big enough to receive the data read | |
2675 | * from the parent. | |
2676 | */ | |
2677 | page_count = (u32)calc_pages_for(0, length); | |
2678 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2679 | if (IS_ERR(pages)) { | |
2680 | result = PTR_ERR(pages); | |
2681 | pages = NULL; | |
2682 | goto out_err; | |
2683 | } | |
2684 | ||
2685 | result = -ENOMEM; | |
e93f3152 AE |
2686 | parent_request = rbd_parent_request_create(obj_request, |
2687 | img_offset, length); | |
3d7efd18 AE |
2688 | if (!parent_request) |
2689 | goto out_err; | |
3d7efd18 AE |
2690 | |
2691 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2692 | if (result) | |
2693 | goto out_err; | |
2694 | parent_request->copyup_pages = pages; | |
ebda6408 | 2695 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2696 | |
2697 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2698 | result = rbd_img_request_submit(parent_request); | |
2699 | if (!result) | |
2700 | return 0; | |
2701 | ||
2702 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2703 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2704 | parent_request->obj_request = NULL; |
2705 | rbd_obj_request_put(obj_request); | |
2706 | out_err: | |
2707 | if (pages) | |
2708 | ceph_release_page_vector(pages, page_count); | |
2709 | if (parent_request) | |
2710 | rbd_img_request_put(parent_request); | |
2711 | obj_request->result = result; | |
2712 | obj_request->xferred = 0; | |
2713 | obj_request_done_set(obj_request); | |
2714 | ||
2715 | return result; | |
2716 | } | |
2717 | ||
c5b5ef6c AE |
2718 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2719 | { | |
c5b5ef6c | 2720 | struct rbd_obj_request *orig_request; |
638f5abe | 2721 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2722 | int result; |
2723 | ||
2724 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2725 | ||
2726 | /* | |
2727 | * All we need from the object request is the original | |
2728 | * request and the result of the STAT op. Grab those, then | |
2729 | * we're done with the request. | |
2730 | */ | |
2731 | orig_request = obj_request->obj_request; | |
2732 | obj_request->obj_request = NULL; | |
912c317d | 2733 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2734 | rbd_assert(orig_request); |
2735 | rbd_assert(orig_request->img_request); | |
2736 | ||
2737 | result = obj_request->result; | |
2738 | obj_request->result = 0; | |
2739 | ||
2740 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2741 | obj_request, orig_request, result, | |
2742 | obj_request->xferred, obj_request->length); | |
2743 | rbd_obj_request_put(obj_request); | |
2744 | ||
638f5abe AE |
2745 | /* |
2746 | * If the overlap has become 0 (most likely because the | |
2747 | * image has been flattened) we need to free the pages | |
2748 | * and re-submit the original write request. | |
2749 | */ | |
2750 | rbd_dev = orig_request->img_request->rbd_dev; | |
2751 | if (!rbd_dev->parent_overlap) { | |
2752 | struct ceph_osd_client *osdc; | |
2753 | ||
638f5abe AE |
2754 | osdc = &rbd_dev->rbd_client->client->osdc; |
2755 | result = rbd_obj_request_submit(osdc, orig_request); | |
2756 | if (!result) | |
2757 | return; | |
2758 | } | |
c5b5ef6c AE |
2759 | |
2760 | /* | |
2761 | * Our only purpose here is to determine whether the object | |
2762 | * exists, and we don't want to treat the non-existence as | |
2763 | * an error. If something else comes back, transfer the | |
2764 | * error to the original request and complete it now. | |
2765 | */ | |
2766 | if (!result) { | |
2767 | obj_request_existence_set(orig_request, true); | |
2768 | } else if (result == -ENOENT) { | |
2769 | obj_request_existence_set(orig_request, false); | |
2770 | } else if (result) { | |
2771 | orig_request->result = result; | |
3d7efd18 | 2772 | goto out; |
c5b5ef6c AE |
2773 | } |
2774 | ||
2775 | /* | |
2776 | * Resubmit the original request now that we have recorded | |
2777 | * whether the target object exists. | |
2778 | */ | |
b454e36d | 2779 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2780 | out: |
c5b5ef6c AE |
2781 | if (orig_request->result) |
2782 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2783 | } |
2784 | ||
2785 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2786 | { | |
2787 | struct rbd_obj_request *stat_request; | |
2788 | struct rbd_device *rbd_dev; | |
2789 | struct ceph_osd_client *osdc; | |
2790 | struct page **pages = NULL; | |
2791 | u32 page_count; | |
2792 | size_t size; | |
2793 | int ret; | |
2794 | ||
2795 | /* | |
2796 | * The response data for a STAT call consists of: | |
2797 | * le64 length; | |
2798 | * struct { | |
2799 | * le32 tv_sec; | |
2800 | * le32 tv_nsec; | |
2801 | * } mtime; | |
2802 | */ | |
2803 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2804 | page_count = (u32)calc_pages_for(0, size); | |
2805 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2806 | if (IS_ERR(pages)) | |
2807 | return PTR_ERR(pages); | |
2808 | ||
2809 | ret = -ENOMEM; | |
2810 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2811 | OBJ_REQUEST_PAGES); | |
2812 | if (!stat_request) | |
2813 | goto out; | |
2814 | ||
2815 | rbd_obj_request_get(obj_request); | |
2816 | stat_request->obj_request = obj_request; | |
2817 | stat_request->pages = pages; | |
2818 | stat_request->page_count = page_count; | |
2819 | ||
2820 | rbd_assert(obj_request->img_request); | |
2821 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2822 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2823 | stat_request); |
c5b5ef6c AE |
2824 | if (!stat_request->osd_req) |
2825 | goto out; | |
2826 | stat_request->callback = rbd_img_obj_exists_callback; | |
2827 | ||
2828 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2829 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2830 | false, false); | |
9d4df01f | 2831 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2832 | |
2833 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2834 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2835 | out: | |
2836 | if (ret) | |
2837 | rbd_obj_request_put(obj_request); | |
2838 | ||
2839 | return ret; | |
2840 | } | |
2841 | ||
70d045f6 | 2842 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2843 | { |
2844 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2845 | struct rbd_device *rbd_dev; |
b454e36d AE |
2846 | |
2847 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2848 | ||
2849 | img_request = obj_request->img_request; | |
2850 | rbd_assert(img_request); | |
a9e8ba2c | 2851 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2852 | |
70d045f6 | 2853 | /* Reads */ |
1c220881 JD |
2854 | if (!img_request_write_test(img_request) && |
2855 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2856 | return true; |
2857 | ||
2858 | /* Non-layered writes */ | |
2859 | if (!img_request_layered_test(img_request)) | |
2860 | return true; | |
2861 | ||
b454e36d | 2862 | /* |
70d045f6 ID |
2863 | * Layered writes outside of the parent overlap range don't |
2864 | * share any data with the parent. | |
b454e36d | 2865 | */ |
70d045f6 ID |
2866 | if (!obj_request_overlaps_parent(obj_request)) |
2867 | return true; | |
b454e36d | 2868 | |
c622d226 GZ |
2869 | /* |
2870 | * Entire-object layered writes - we will overwrite whatever | |
2871 | * parent data there is anyway. | |
2872 | */ | |
2873 | if (!obj_request->offset && | |
2874 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2875 | return true; | |
2876 | ||
70d045f6 ID |
2877 | /* |
2878 | * If the object is known to already exist, its parent data has | |
2879 | * already been copied. | |
2880 | */ | |
2881 | if (obj_request_known_test(obj_request) && | |
2882 | obj_request_exists_test(obj_request)) | |
2883 | return true; | |
2884 | ||
2885 | return false; | |
2886 | } | |
2887 | ||
2888 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2889 | { | |
2890 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2891 | struct rbd_device *rbd_dev; |
2892 | struct ceph_osd_client *osdc; | |
2893 | ||
2894 | rbd_dev = obj_request->img_request->rbd_dev; | |
2895 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2896 | ||
2897 | return rbd_obj_request_submit(osdc, obj_request); | |
2898 | } | |
2899 | ||
2900 | /* | |
3d7efd18 AE |
2901 | * It's a layered write. The target object might exist but |
2902 | * we may not know that yet. If we know it doesn't exist, | |
2903 | * start by reading the data for the full target object from | |
2904 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2905 | */ |
70d045f6 | 2906 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2907 | return rbd_img_obj_parent_read_full(obj_request); |
2908 | ||
2909 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2910 | |
2911 | return rbd_img_obj_exists_submit(obj_request); | |
2912 | } | |
2913 | ||
bf0d5f50 AE |
2914 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2915 | { | |
bf0d5f50 | 2916 | struct rbd_obj_request *obj_request; |
46faeed4 | 2917 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2918 | |
37206ee5 | 2919 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2920 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2921 | int ret; |
2922 | ||
b454e36d | 2923 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2924 | if (ret) |
2925 | return ret; | |
bf0d5f50 AE |
2926 | } |
2927 | ||
2928 | return 0; | |
2929 | } | |
8b3e1a56 AE |
2930 | |
2931 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2932 | { | |
2933 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2934 | struct rbd_device *rbd_dev; |
2935 | u64 obj_end; | |
02c74fba AE |
2936 | u64 img_xferred; |
2937 | int img_result; | |
8b3e1a56 AE |
2938 | |
2939 | rbd_assert(img_request_child_test(img_request)); | |
2940 | ||
02c74fba AE |
2941 | /* First get what we need from the image request and release it */ |
2942 | ||
8b3e1a56 | 2943 | obj_request = img_request->obj_request; |
02c74fba AE |
2944 | img_xferred = img_request->xferred; |
2945 | img_result = img_request->result; | |
2946 | rbd_img_request_put(img_request); | |
2947 | ||
2948 | /* | |
2949 | * If the overlap has become 0 (most likely because the | |
2950 | * image has been flattened) we need to re-submit the | |
2951 | * original request. | |
2952 | */ | |
a9e8ba2c AE |
2953 | rbd_assert(obj_request); |
2954 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2955 | rbd_dev = obj_request->img_request->rbd_dev; |
2956 | if (!rbd_dev->parent_overlap) { | |
2957 | struct ceph_osd_client *osdc; | |
2958 | ||
2959 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2960 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2961 | if (!img_result) | |
2962 | return; | |
2963 | } | |
a9e8ba2c | 2964 | |
02c74fba | 2965 | obj_request->result = img_result; |
a9e8ba2c AE |
2966 | if (obj_request->result) |
2967 | goto out; | |
2968 | ||
2969 | /* | |
2970 | * We need to zero anything beyond the parent overlap | |
2971 | * boundary. Since rbd_img_obj_request_read_callback() | |
2972 | * will zero anything beyond the end of a short read, an | |
2973 | * easy way to do this is to pretend the data from the | |
2974 | * parent came up short--ending at the overlap boundary. | |
2975 | */ | |
2976 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2977 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2978 | if (obj_end > rbd_dev->parent_overlap) { |
2979 | u64 xferred = 0; | |
2980 | ||
2981 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2982 | xferred = rbd_dev->parent_overlap - | |
2983 | obj_request->img_offset; | |
8b3e1a56 | 2984 | |
02c74fba | 2985 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2986 | } else { |
02c74fba | 2987 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2988 | } |
2989 | out: | |
8b3e1a56 AE |
2990 | rbd_img_obj_request_read_callback(obj_request); |
2991 | rbd_obj_request_complete(obj_request); | |
2992 | } | |
2993 | ||
2994 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2995 | { | |
8b3e1a56 AE |
2996 | struct rbd_img_request *img_request; |
2997 | int result; | |
2998 | ||
2999 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3000 | rbd_assert(obj_request->img_request != NULL); | |
3001 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3002 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3003 | |
8b3e1a56 | 3004 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3005 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3006 | obj_request->img_offset, |
e93f3152 | 3007 | obj_request->length); |
8b3e1a56 AE |
3008 | result = -ENOMEM; |
3009 | if (!img_request) | |
3010 | goto out_err; | |
3011 | ||
5b2ab72d AE |
3012 | if (obj_request->type == OBJ_REQUEST_BIO) |
3013 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3014 | obj_request->bio_list); | |
3015 | else | |
3016 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3017 | obj_request->pages); | |
8b3e1a56 AE |
3018 | if (result) |
3019 | goto out_err; | |
3020 | ||
3021 | img_request->callback = rbd_img_parent_read_callback; | |
3022 | result = rbd_img_request_submit(img_request); | |
3023 | if (result) | |
3024 | goto out_err; | |
3025 | ||
3026 | return; | |
3027 | out_err: | |
3028 | if (img_request) | |
3029 | rbd_img_request_put(img_request); | |
3030 | obj_request->result = result; | |
3031 | obj_request->xferred = 0; | |
3032 | obj_request_done_set(obj_request); | |
3033 | } | |
bf0d5f50 | 3034 | |
20e0af67 | 3035 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3036 | { |
3037 | struct rbd_obj_request *obj_request; | |
2169238d | 3038 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3039 | int ret; |
3040 | ||
3041 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3042 | OBJ_REQUEST_NODATA); | |
3043 | if (!obj_request) | |
3044 | return -ENOMEM; | |
3045 | ||
3046 | ret = -ENOMEM; | |
6d2940c8 | 3047 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3048 | obj_request); |
b8d70035 AE |
3049 | if (!obj_request->osd_req) |
3050 | goto out; | |
3051 | ||
c99d2d4a | 3052 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3053 | notify_id, 0, 0); |
9d4df01f | 3054 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3055 | |
b8d70035 | 3056 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3057 | if (ret) |
20e0af67 JD |
3058 | goto out; |
3059 | ret = rbd_obj_request_wait(obj_request); | |
3060 | out: | |
3061 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3062 | |
3063 | return ret; | |
3064 | } | |
3065 | ||
3066 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3067 | { | |
3068 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3069 | int ret; |
b8d70035 AE |
3070 | |
3071 | if (!rbd_dev) | |
3072 | return; | |
3073 | ||
37206ee5 | 3074 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3075 | rbd_dev->header_name, (unsigned long long)notify_id, |
3076 | (unsigned int)opcode); | |
52bb1f9b ID |
3077 | |
3078 | /* | |
3079 | * Until adequate refresh error handling is in place, there is | |
3080 | * not much we can do here, except warn. | |
3081 | * | |
3082 | * See http://tracker.ceph.com/issues/5040 | |
3083 | */ | |
e627db08 AE |
3084 | ret = rbd_dev_refresh(rbd_dev); |
3085 | if (ret) | |
9584d508 | 3086 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3087 | |
52bb1f9b ID |
3088 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3089 | if (ret) | |
9584d508 | 3090 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3091 | } |
3092 | ||
bb040aa0 ID |
3093 | /* |
3094 | * Send a (un)watch request and wait for the ack. Return a request | |
3095 | * with a ref held on success or error. | |
3096 | */ | |
3097 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3098 | struct rbd_device *rbd_dev, | |
3099 | bool watch) | |
3100 | { | |
3101 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3102 | struct rbd_obj_request *obj_request; | |
3103 | int ret; | |
3104 | ||
3105 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3106 | OBJ_REQUEST_NODATA); | |
3107 | if (!obj_request) | |
3108 | return ERR_PTR(-ENOMEM); | |
3109 | ||
6d2940c8 | 3110 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3111 | obj_request); |
3112 | if (!obj_request->osd_req) { | |
3113 | ret = -ENOMEM; | |
3114 | goto out; | |
3115 | } | |
3116 | ||
3117 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3118 | rbd_dev->watch_event->cookie, 0, watch); | |
3119 | rbd_osd_req_format_write(obj_request); | |
3120 | ||
3121 | if (watch) | |
3122 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3123 | ||
3124 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3125 | if (ret) | |
3126 | goto out; | |
3127 | ||
3128 | ret = rbd_obj_request_wait(obj_request); | |
3129 | if (ret) | |
3130 | goto out; | |
3131 | ||
3132 | ret = obj_request->result; | |
3133 | if (ret) { | |
3134 | if (watch) | |
3135 | rbd_obj_request_end(obj_request); | |
3136 | goto out; | |
3137 | } | |
3138 | ||
3139 | return obj_request; | |
3140 | ||
3141 | out: | |
3142 | rbd_obj_request_put(obj_request); | |
3143 | return ERR_PTR(ret); | |
3144 | } | |
3145 | ||
9969ebc5 | 3146 | /* |
b30a01f2 | 3147 | * Initiate a watch request, synchronously. |
9969ebc5 | 3148 | */ |
b30a01f2 | 3149 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3150 | { |
3151 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3152 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3153 | int ret; |
3154 | ||
b30a01f2 ID |
3155 | rbd_assert(!rbd_dev->watch_event); |
3156 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3157 | |
b30a01f2 ID |
3158 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3159 | &rbd_dev->watch_event); | |
3160 | if (ret < 0) | |
3161 | return ret; | |
3162 | ||
76756a51 ID |
3163 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3164 | if (IS_ERR(obj_request)) { | |
3165 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3166 | rbd_dev->watch_event = NULL; | |
3167 | return PTR_ERR(obj_request); | |
b30a01f2 | 3168 | } |
9969ebc5 | 3169 | |
8eb87565 AE |
3170 | /* |
3171 | * A watch request is set to linger, so the underlying osd | |
3172 | * request won't go away until we unregister it. We retain | |
3173 | * a pointer to the object request during that time (in | |
76756a51 ID |
3174 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3175 | * We'll drop that reference after we've unregistered it in | |
3176 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3177 | */ |
b30a01f2 | 3178 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3179 | |
b30a01f2 | 3180 | return 0; |
b30a01f2 ID |
3181 | } |
3182 | ||
3183 | /* | |
3184 | * Tear down a watch request, synchronously. | |
3185 | */ | |
76756a51 | 3186 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3187 | { |
b30a01f2 | 3188 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3189 | |
3190 | rbd_assert(rbd_dev->watch_event); | |
3191 | rbd_assert(rbd_dev->watch_request); | |
3192 | ||
76756a51 | 3193 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3194 | rbd_obj_request_put(rbd_dev->watch_request); |
3195 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3196 | |
76756a51 ID |
3197 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3198 | if (!IS_ERR(obj_request)) | |
3199 | rbd_obj_request_put(obj_request); | |
3200 | else | |
3201 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3202 | PTR_ERR(obj_request)); | |
3203 | ||
9969ebc5 AE |
3204 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3205 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3206 | } |
3207 | ||
36be9a76 | 3208 | /* |
f40eb349 AE |
3209 | * Synchronous osd object method call. Returns the number of bytes |
3210 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3211 | */ |
3212 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3213 | const char *object_name, | |
3214 | const char *class_name, | |
3215 | const char *method_name, | |
4157976b | 3216 | const void *outbound, |
36be9a76 | 3217 | size_t outbound_size, |
4157976b | 3218 | void *inbound, |
e2a58ee5 | 3219 | size_t inbound_size) |
36be9a76 | 3220 | { |
2169238d | 3221 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3222 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3223 | struct page **pages; |
3224 | u32 page_count; | |
3225 | int ret; | |
3226 | ||
3227 | /* | |
6010a451 AE |
3228 | * Method calls are ultimately read operations. The result |
3229 | * should placed into the inbound buffer provided. They | |
3230 | * also supply outbound data--parameters for the object | |
3231 | * method. Currently if this is present it will be a | |
3232 | * snapshot id. | |
36be9a76 | 3233 | */ |
57385b51 | 3234 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3235 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3236 | if (IS_ERR(pages)) | |
3237 | return PTR_ERR(pages); | |
3238 | ||
3239 | ret = -ENOMEM; | |
6010a451 | 3240 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3241 | OBJ_REQUEST_PAGES); |
3242 | if (!obj_request) | |
3243 | goto out; | |
3244 | ||
3245 | obj_request->pages = pages; | |
3246 | obj_request->page_count = page_count; | |
3247 | ||
6d2940c8 | 3248 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3249 | obj_request); |
36be9a76 AE |
3250 | if (!obj_request->osd_req) |
3251 | goto out; | |
3252 | ||
c99d2d4a | 3253 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3254 | class_name, method_name); |
3255 | if (outbound_size) { | |
3256 | struct ceph_pagelist *pagelist; | |
3257 | ||
3258 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3259 | if (!pagelist) | |
3260 | goto out; | |
3261 | ||
3262 | ceph_pagelist_init(pagelist); | |
3263 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3264 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3265 | pagelist); | |
3266 | } | |
a4ce40a9 AE |
3267 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3268 | obj_request->pages, inbound_size, | |
44cd188d | 3269 | 0, false, false); |
9d4df01f | 3270 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3271 | |
36be9a76 AE |
3272 | ret = rbd_obj_request_submit(osdc, obj_request); |
3273 | if (ret) | |
3274 | goto out; | |
3275 | ret = rbd_obj_request_wait(obj_request); | |
3276 | if (ret) | |
3277 | goto out; | |
3278 | ||
3279 | ret = obj_request->result; | |
3280 | if (ret < 0) | |
3281 | goto out; | |
57385b51 AE |
3282 | |
3283 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3284 | ret = (int)obj_request->xferred; | |
903bb32e | 3285 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3286 | out: |
3287 | if (obj_request) | |
3288 | rbd_obj_request_put(obj_request); | |
3289 | else | |
3290 | ceph_release_page_vector(pages, page_count); | |
3291 | ||
3292 | return ret; | |
3293 | } | |
3294 | ||
bc1ecc65 | 3295 | static void rbd_handle_request(struct rbd_device *rbd_dev, struct request *rq) |
bf0d5f50 | 3296 | { |
bc1ecc65 | 3297 | struct rbd_img_request *img_request; |
4e752f0a | 3298 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3299 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3300 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3301 | enum obj_operation_type op_type; |
4e752f0a | 3302 | u64 mapping_size; |
bf0d5f50 AE |
3303 | int result; |
3304 | ||
90e98c52 GZ |
3305 | if (rq->cmd_flags & REQ_DISCARD) |
3306 | op_type = OBJ_OP_DISCARD; | |
3307 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3308 | op_type = OBJ_OP_WRITE; |
3309 | else | |
3310 | op_type = OBJ_OP_READ; | |
3311 | ||
bc1ecc65 | 3312 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3313 | |
bc1ecc65 ID |
3314 | if (!length) { |
3315 | dout("%s: zero-length request\n", __func__); | |
3316 | result = 0; | |
3317 | goto err_rq; | |
3318 | } | |
bf0d5f50 | 3319 | |
6d2940c8 | 3320 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3321 | |
6d2940c8 | 3322 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3323 | if (rbd_dev->mapping.read_only) { |
3324 | result = -EROFS; | |
3325 | goto err_rq; | |
4dda41d3 | 3326 | } |
bc1ecc65 ID |
3327 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3328 | } | |
4dda41d3 | 3329 | |
bc1ecc65 ID |
3330 | /* |
3331 | * Quit early if the mapped snapshot no longer exists. It's | |
3332 | * still possible the snapshot will have disappeared by the | |
3333 | * time our request arrives at the osd, but there's no sense in | |
3334 | * sending it if we already know. | |
3335 | */ | |
3336 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3337 | dout("request for non-existent snapshot"); | |
3338 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3339 | result = -ENXIO; | |
3340 | goto err_rq; | |
3341 | } | |
4dda41d3 | 3342 | |
bc1ecc65 ID |
3343 | if (offset && length > U64_MAX - offset + 1) { |
3344 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3345 | length); | |
3346 | result = -EINVAL; | |
3347 | goto err_rq; /* Shouldn't happen */ | |
3348 | } | |
4dda41d3 | 3349 | |
4e752f0a JD |
3350 | down_read(&rbd_dev->header_rwsem); |
3351 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3352 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3353 | snapc = rbd_dev->header.snapc; |
3354 | ceph_get_snap_context(snapc); | |
3355 | } | |
3356 | up_read(&rbd_dev->header_rwsem); | |
3357 | ||
3358 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3359 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3360 | length, mapping_size); |
bc1ecc65 ID |
3361 | result = -EIO; |
3362 | goto err_rq; | |
3363 | } | |
bf0d5f50 | 3364 | |
6d2940c8 | 3365 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3366 | snapc); |
bc1ecc65 ID |
3367 | if (!img_request) { |
3368 | result = -ENOMEM; | |
3369 | goto err_rq; | |
3370 | } | |
3371 | img_request->rq = rq; | |
bf0d5f50 | 3372 | |
90e98c52 GZ |
3373 | if (op_type == OBJ_OP_DISCARD) |
3374 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3375 | NULL); | |
3376 | else | |
3377 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3378 | rq->bio); | |
bc1ecc65 ID |
3379 | if (result) |
3380 | goto err_img_request; | |
bf0d5f50 | 3381 | |
bc1ecc65 ID |
3382 | result = rbd_img_request_submit(img_request); |
3383 | if (result) | |
3384 | goto err_img_request; | |
bf0d5f50 | 3385 | |
bc1ecc65 | 3386 | return; |
bf0d5f50 | 3387 | |
bc1ecc65 ID |
3388 | err_img_request: |
3389 | rbd_img_request_put(img_request); | |
3390 | err_rq: | |
3391 | if (result) | |
3392 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3393 | obj_op_name(op_type), length, offset, result); |
4e752f0a JD |
3394 | if (snapc) |
3395 | ceph_put_snap_context(snapc); | |
bc1ecc65 ID |
3396 | blk_end_request_all(rq, result); |
3397 | } | |
bf0d5f50 | 3398 | |
bc1ecc65 ID |
3399 | static void rbd_request_workfn(struct work_struct *work) |
3400 | { | |
3401 | struct rbd_device *rbd_dev = | |
3402 | container_of(work, struct rbd_device, rq_work); | |
3403 | struct request *rq, *next; | |
3404 | LIST_HEAD(requests); | |
00a653e2 | 3405 | |
bc1ecc65 ID |
3406 | spin_lock_irq(&rbd_dev->lock); /* rq->q->queue_lock */ |
3407 | list_splice_init(&rbd_dev->rq_queue, &requests); | |
3408 | spin_unlock_irq(&rbd_dev->lock); | |
bf0d5f50 | 3409 | |
bc1ecc65 ID |
3410 | list_for_each_entry_safe(rq, next, &requests, queuelist) { |
3411 | list_del_init(&rq->queuelist); | |
3412 | rbd_handle_request(rbd_dev, rq); | |
3413 | } | |
3414 | } | |
bf0d5f50 | 3415 | |
bc1ecc65 ID |
3416 | /* |
3417 | * Called with q->queue_lock held and interrupts disabled, possibly on | |
3418 | * the way to schedule(). Do not sleep here! | |
3419 | */ | |
3420 | static void rbd_request_fn(struct request_queue *q) | |
3421 | { | |
3422 | struct rbd_device *rbd_dev = q->queuedata; | |
3423 | struct request *rq; | |
3424 | int queued = 0; | |
3425 | ||
3426 | rbd_assert(rbd_dev); | |
3427 | ||
3428 | while ((rq = blk_fetch_request(q))) { | |
3429 | /* Ignore any non-FS requests that filter through. */ | |
3430 | if (rq->cmd_type != REQ_TYPE_FS) { | |
3431 | dout("%s: non-fs request type %d\n", __func__, | |
3432 | (int) rq->cmd_type); | |
3433 | __blk_end_request_all(rq, 0); | |
3434 | continue; | |
bf0d5f50 | 3435 | } |
bc1ecc65 ID |
3436 | |
3437 | list_add_tail(&rq->queuelist, &rbd_dev->rq_queue); | |
3438 | queued++; | |
bf0d5f50 | 3439 | } |
bc1ecc65 ID |
3440 | |
3441 | if (queued) | |
3442 | queue_work(rbd_dev->rq_wq, &rbd_dev->rq_work); | |
bf0d5f50 AE |
3443 | } |
3444 | ||
602adf40 YS |
3445 | /* |
3446 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3447 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3448 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3449 | */ |
3450 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3451 | struct bio_vec *bvec) | |
3452 | { | |
3453 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3454 | sector_t sector_offset; |
3455 | sector_t sectors_per_obj; | |
3456 | sector_t obj_sector_offset; | |
3457 | int ret; | |
3458 | ||
3459 | /* | |
3460 | * Find how far into its rbd object the partition-relative | |
3461 | * bio start sector is to offset relative to the enclosing | |
3462 | * device. | |
3463 | */ | |
3464 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3465 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3466 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3467 | ||
3468 | /* | |
3469 | * Compute the number of bytes from that offset to the end | |
3470 | * of the object. Account for what's already used by the bio. | |
3471 | */ | |
3472 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3473 | if (ret > bmd->bi_size) | |
3474 | ret -= bmd->bi_size; | |
3475 | else | |
3476 | ret = 0; | |
3477 | ||
3478 | /* | |
3479 | * Don't send back more than was asked for. And if the bio | |
3480 | * was empty, let the whole thing through because: "Note | |
3481 | * that a block device *must* allow a single page to be | |
3482 | * added to an empty bio." | |
3483 | */ | |
3484 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3485 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3486 | ret = (int) bvec->bv_len; | |
3487 | ||
3488 | return ret; | |
602adf40 YS |
3489 | } |
3490 | ||
3491 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3492 | { | |
3493 | struct gendisk *disk = rbd_dev->disk; | |
3494 | ||
3495 | if (!disk) | |
3496 | return; | |
3497 | ||
a0cab924 AE |
3498 | rbd_dev->disk = NULL; |
3499 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3500 | del_gendisk(disk); |
a0cab924 AE |
3501 | if (disk->queue) |
3502 | blk_cleanup_queue(disk->queue); | |
3503 | } | |
602adf40 YS |
3504 | put_disk(disk); |
3505 | } | |
3506 | ||
788e2df3 AE |
3507 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3508 | const char *object_name, | |
7097f8df | 3509 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3510 | |
3511 | { | |
2169238d | 3512 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3513 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3514 | struct page **pages = NULL; |
3515 | u32 page_count; | |
1ceae7ef | 3516 | size_t size; |
788e2df3 AE |
3517 | int ret; |
3518 | ||
3519 | page_count = (u32) calc_pages_for(offset, length); | |
3520 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3521 | if (IS_ERR(pages)) | |
3522 | ret = PTR_ERR(pages); | |
3523 | ||
3524 | ret = -ENOMEM; | |
3525 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3526 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3527 | if (!obj_request) |
3528 | goto out; | |
3529 | ||
3530 | obj_request->pages = pages; | |
3531 | obj_request->page_count = page_count; | |
3532 | ||
6d2940c8 | 3533 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3534 | obj_request); |
788e2df3 AE |
3535 | if (!obj_request->osd_req) |
3536 | goto out; | |
3537 | ||
c99d2d4a AE |
3538 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3539 | offset, length, 0, 0); | |
406e2c9f | 3540 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3541 | obj_request->pages, |
44cd188d AE |
3542 | obj_request->length, |
3543 | obj_request->offset & ~PAGE_MASK, | |
3544 | false, false); | |
9d4df01f | 3545 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3546 | |
788e2df3 AE |
3547 | ret = rbd_obj_request_submit(osdc, obj_request); |
3548 | if (ret) | |
3549 | goto out; | |
3550 | ret = rbd_obj_request_wait(obj_request); | |
3551 | if (ret) | |
3552 | goto out; | |
3553 | ||
3554 | ret = obj_request->result; | |
3555 | if (ret < 0) | |
3556 | goto out; | |
1ceae7ef AE |
3557 | |
3558 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3559 | size = (size_t) obj_request->xferred; | |
903bb32e | 3560 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3561 | rbd_assert(size <= (size_t)INT_MAX); |
3562 | ret = (int)size; | |
788e2df3 AE |
3563 | out: |
3564 | if (obj_request) | |
3565 | rbd_obj_request_put(obj_request); | |
3566 | else | |
3567 | ceph_release_page_vector(pages, page_count); | |
3568 | ||
3569 | return ret; | |
3570 | } | |
3571 | ||
602adf40 | 3572 | /* |
662518b1 AE |
3573 | * Read the complete header for the given rbd device. On successful |
3574 | * return, the rbd_dev->header field will contain up-to-date | |
3575 | * information about the image. | |
602adf40 | 3576 | */ |
99a41ebc | 3577 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3578 | { |
4156d998 | 3579 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3580 | u32 snap_count = 0; |
4156d998 AE |
3581 | u64 names_size = 0; |
3582 | u32 want_count; | |
3583 | int ret; | |
602adf40 | 3584 | |
00f1f36f | 3585 | /* |
4156d998 AE |
3586 | * The complete header will include an array of its 64-bit |
3587 | * snapshot ids, followed by the names of those snapshots as | |
3588 | * a contiguous block of NUL-terminated strings. Note that | |
3589 | * the number of snapshots could change by the time we read | |
3590 | * it in, in which case we re-read it. | |
00f1f36f | 3591 | */ |
4156d998 AE |
3592 | do { |
3593 | size_t size; | |
3594 | ||
3595 | kfree(ondisk); | |
3596 | ||
3597 | size = sizeof (*ondisk); | |
3598 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3599 | size += names_size; | |
3600 | ondisk = kmalloc(size, GFP_KERNEL); | |
3601 | if (!ondisk) | |
662518b1 | 3602 | return -ENOMEM; |
4156d998 | 3603 | |
788e2df3 | 3604 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3605 | 0, size, ondisk); |
4156d998 | 3606 | if (ret < 0) |
662518b1 | 3607 | goto out; |
c0cd10db | 3608 | if ((size_t)ret < size) { |
4156d998 | 3609 | ret = -ENXIO; |
06ecc6cb AE |
3610 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3611 | size, ret); | |
662518b1 | 3612 | goto out; |
4156d998 AE |
3613 | } |
3614 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3615 | ret = -ENXIO; | |
06ecc6cb | 3616 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3617 | goto out; |
81e759fb | 3618 | } |
602adf40 | 3619 | |
4156d998 AE |
3620 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3621 | want_count = snap_count; | |
3622 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3623 | } while (snap_count != want_count); | |
00f1f36f | 3624 | |
662518b1 AE |
3625 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3626 | out: | |
4156d998 AE |
3627 | kfree(ondisk); |
3628 | ||
3629 | return ret; | |
602adf40 YS |
3630 | } |
3631 | ||
15228ede AE |
3632 | /* |
3633 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3634 | * has disappeared from the (just updated) snapshot context. | |
3635 | */ | |
3636 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3637 | { | |
3638 | u64 snap_id; | |
3639 | ||
3640 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3641 | return; | |
3642 | ||
3643 | snap_id = rbd_dev->spec->snap_id; | |
3644 | if (snap_id == CEPH_NOSNAP) | |
3645 | return; | |
3646 | ||
3647 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3648 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3649 | } | |
3650 | ||
9875201e JD |
3651 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3652 | { | |
3653 | sector_t size; | |
3654 | bool removing; | |
3655 | ||
3656 | /* | |
3657 | * Don't hold the lock while doing disk operations, | |
3658 | * or lock ordering will conflict with the bdev mutex via: | |
3659 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3660 | */ | |
3661 | spin_lock_irq(&rbd_dev->lock); | |
3662 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3663 | spin_unlock_irq(&rbd_dev->lock); | |
3664 | /* | |
3665 | * If the device is being removed, rbd_dev->disk has | |
3666 | * been destroyed, so don't try to update its size | |
3667 | */ | |
3668 | if (!removing) { | |
3669 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3670 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3671 | set_capacity(rbd_dev->disk, size); | |
3672 | revalidate_disk(rbd_dev->disk); | |
3673 | } | |
3674 | } | |
3675 | ||
cc4a38bd | 3676 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3677 | { |
e627db08 | 3678 | u64 mapping_size; |
1fe5e993 AE |
3679 | int ret; |
3680 | ||
cfbf6377 | 3681 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3682 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3683 | |
3684 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b ID |
3685 | if (ret) |
3686 | return ret; | |
15228ede | 3687 | |
e8f59b59 ID |
3688 | /* |
3689 | * If there is a parent, see if it has disappeared due to the | |
3690 | * mapped image getting flattened. | |
3691 | */ | |
3692 | if (rbd_dev->parent) { | |
3693 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3694 | if (ret) | |
3695 | return ret; | |
3696 | } | |
3697 | ||
5ff1108c ID |
3698 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
3699 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
3700 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
3701 | } else { | |
3702 | /* validate mapped snapshot's EXISTS flag */ | |
3703 | rbd_exists_validate(rbd_dev); | |
3704 | } | |
15228ede | 3705 | |
cfbf6377 AE |
3706 | up_write(&rbd_dev->header_rwsem); |
3707 | ||
e8f59b59 | 3708 | if (mapping_size != rbd_dev->mapping.size) |
9875201e | 3709 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3710 | |
52bb1f9b | 3711 | return 0; |
1fe5e993 AE |
3712 | } |
3713 | ||
602adf40 YS |
3714 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3715 | { | |
3716 | struct gendisk *disk; | |
3717 | struct request_queue *q; | |
593a9e7b | 3718 | u64 segment_size; |
602adf40 | 3719 | |
602adf40 | 3720 | /* create gendisk info */ |
7e513d43 ID |
3721 | disk = alloc_disk(single_major ? |
3722 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3723 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3724 | if (!disk) |
1fcdb8aa | 3725 | return -ENOMEM; |
602adf40 | 3726 | |
f0f8cef5 | 3727 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3728 | rbd_dev->dev_id); |
602adf40 | 3729 | disk->major = rbd_dev->major; |
dd82fff1 | 3730 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3731 | if (single_major) |
3732 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3733 | disk->fops = &rbd_bd_ops; |
3734 | disk->private_data = rbd_dev; | |
3735 | ||
bf0d5f50 | 3736 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3737 | if (!q) |
3738 | goto out_disk; | |
029bcbd8 | 3739 | |
593a9e7b AE |
3740 | /* We use the default size, but let's be explicit about it. */ |
3741 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3742 | ||
029bcbd8 | 3743 | /* set io sizes to object size */ |
593a9e7b AE |
3744 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3745 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3746 | blk_queue_max_segment_size(q, segment_size); | |
3747 | blk_queue_io_min(q, segment_size); | |
3748 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3749 | |
90e98c52 GZ |
3750 | /* enable the discard support */ |
3751 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3752 | q->limits.discard_granularity = segment_size; | |
3753 | q->limits.discard_alignment = segment_size; | |
3754 | ||
602adf40 YS |
3755 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3756 | disk->queue = q; | |
3757 | ||
3758 | q->queuedata = rbd_dev; | |
3759 | ||
3760 | rbd_dev->disk = disk; | |
602adf40 | 3761 | |
602adf40 | 3762 | return 0; |
602adf40 YS |
3763 | out_disk: |
3764 | put_disk(disk); | |
1fcdb8aa AE |
3765 | |
3766 | return -ENOMEM; | |
602adf40 YS |
3767 | } |
3768 | ||
dfc5606d YS |
3769 | /* |
3770 | sysfs | |
3771 | */ | |
3772 | ||
593a9e7b AE |
3773 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3774 | { | |
3775 | return container_of(dev, struct rbd_device, dev); | |
3776 | } | |
3777 | ||
dfc5606d YS |
3778 | static ssize_t rbd_size_show(struct device *dev, |
3779 | struct device_attribute *attr, char *buf) | |
3780 | { | |
593a9e7b | 3781 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3782 | |
fc71d833 AE |
3783 | return sprintf(buf, "%llu\n", |
3784 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3785 | } |
3786 | ||
34b13184 AE |
3787 | /* |
3788 | * Note this shows the features for whatever's mapped, which is not | |
3789 | * necessarily the base image. | |
3790 | */ | |
3791 | static ssize_t rbd_features_show(struct device *dev, | |
3792 | struct device_attribute *attr, char *buf) | |
3793 | { | |
3794 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3795 | ||
3796 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3797 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3798 | } |
3799 | ||
dfc5606d YS |
3800 | static ssize_t rbd_major_show(struct device *dev, |
3801 | struct device_attribute *attr, char *buf) | |
3802 | { | |
593a9e7b | 3803 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3804 | |
fc71d833 AE |
3805 | if (rbd_dev->major) |
3806 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3807 | ||
3808 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3809 | } |
3810 | ||
3811 | static ssize_t rbd_minor_show(struct device *dev, | |
3812 | struct device_attribute *attr, char *buf) | |
3813 | { | |
3814 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3815 | |
dd82fff1 | 3816 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3817 | } |
3818 | ||
3819 | static ssize_t rbd_client_id_show(struct device *dev, | |
3820 | struct device_attribute *attr, char *buf) | |
602adf40 | 3821 | { |
593a9e7b | 3822 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3823 | |
1dbb4399 AE |
3824 | return sprintf(buf, "client%lld\n", |
3825 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3826 | } |
3827 | ||
dfc5606d YS |
3828 | static ssize_t rbd_pool_show(struct device *dev, |
3829 | struct device_attribute *attr, char *buf) | |
602adf40 | 3830 | { |
593a9e7b | 3831 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3832 | |
0d7dbfce | 3833 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3834 | } |
3835 | ||
9bb2f334 AE |
3836 | static ssize_t rbd_pool_id_show(struct device *dev, |
3837 | struct device_attribute *attr, char *buf) | |
3838 | { | |
3839 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3840 | ||
0d7dbfce | 3841 | return sprintf(buf, "%llu\n", |
fc71d833 | 3842 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3843 | } |
3844 | ||
dfc5606d YS |
3845 | static ssize_t rbd_name_show(struct device *dev, |
3846 | struct device_attribute *attr, char *buf) | |
3847 | { | |
593a9e7b | 3848 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3849 | |
a92ffdf8 AE |
3850 | if (rbd_dev->spec->image_name) |
3851 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3852 | ||
3853 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3854 | } |
3855 | ||
589d30e0 AE |
3856 | static ssize_t rbd_image_id_show(struct device *dev, |
3857 | struct device_attribute *attr, char *buf) | |
3858 | { | |
3859 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3860 | ||
0d7dbfce | 3861 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3862 | } |
3863 | ||
34b13184 AE |
3864 | /* |
3865 | * Shows the name of the currently-mapped snapshot (or | |
3866 | * RBD_SNAP_HEAD_NAME for the base image). | |
3867 | */ | |
dfc5606d YS |
3868 | static ssize_t rbd_snap_show(struct device *dev, |
3869 | struct device_attribute *attr, | |
3870 | char *buf) | |
3871 | { | |
593a9e7b | 3872 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3873 | |
0d7dbfce | 3874 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3875 | } |
3876 | ||
86b00e0d | 3877 | /* |
ff96128f ID |
3878 | * For a v2 image, shows the chain of parent images, separated by empty |
3879 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3880 | * image)". | |
86b00e0d AE |
3881 | */ |
3882 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3883 | struct device_attribute *attr, |
3884 | char *buf) | |
86b00e0d AE |
3885 | { |
3886 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3887 | ssize_t count = 0; |
86b00e0d | 3888 | |
ff96128f | 3889 | if (!rbd_dev->parent) |
86b00e0d AE |
3890 | return sprintf(buf, "(no parent image)\n"); |
3891 | ||
ff96128f ID |
3892 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3893 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3894 | ||
3895 | count += sprintf(&buf[count], "%s" | |
3896 | "pool_id %llu\npool_name %s\n" | |
3897 | "image_id %s\nimage_name %s\n" | |
3898 | "snap_id %llu\nsnap_name %s\n" | |
3899 | "overlap %llu\n", | |
3900 | !count ? "" : "\n", /* first? */ | |
3901 | spec->pool_id, spec->pool_name, | |
3902 | spec->image_id, spec->image_name ?: "(unknown)", | |
3903 | spec->snap_id, spec->snap_name, | |
3904 | rbd_dev->parent_overlap); | |
3905 | } | |
3906 | ||
3907 | return count; | |
86b00e0d AE |
3908 | } |
3909 | ||
dfc5606d YS |
3910 | static ssize_t rbd_image_refresh(struct device *dev, |
3911 | struct device_attribute *attr, | |
3912 | const char *buf, | |
3913 | size_t size) | |
3914 | { | |
593a9e7b | 3915 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3916 | int ret; |
602adf40 | 3917 | |
cc4a38bd | 3918 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3919 | if (ret) |
52bb1f9b | 3920 | return ret; |
b813623a | 3921 | |
52bb1f9b | 3922 | return size; |
dfc5606d | 3923 | } |
602adf40 | 3924 | |
dfc5606d | 3925 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3926 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3927 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3928 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3929 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3930 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3931 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3932 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3933 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3934 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3935 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3936 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3937 | |
3938 | static struct attribute *rbd_attrs[] = { | |
3939 | &dev_attr_size.attr, | |
34b13184 | 3940 | &dev_attr_features.attr, |
dfc5606d | 3941 | &dev_attr_major.attr, |
dd82fff1 | 3942 | &dev_attr_minor.attr, |
dfc5606d YS |
3943 | &dev_attr_client_id.attr, |
3944 | &dev_attr_pool.attr, | |
9bb2f334 | 3945 | &dev_attr_pool_id.attr, |
dfc5606d | 3946 | &dev_attr_name.attr, |
589d30e0 | 3947 | &dev_attr_image_id.attr, |
dfc5606d | 3948 | &dev_attr_current_snap.attr, |
86b00e0d | 3949 | &dev_attr_parent.attr, |
dfc5606d | 3950 | &dev_attr_refresh.attr, |
dfc5606d YS |
3951 | NULL |
3952 | }; | |
3953 | ||
3954 | static struct attribute_group rbd_attr_group = { | |
3955 | .attrs = rbd_attrs, | |
3956 | }; | |
3957 | ||
3958 | static const struct attribute_group *rbd_attr_groups[] = { | |
3959 | &rbd_attr_group, | |
3960 | NULL | |
3961 | }; | |
3962 | ||
3963 | static void rbd_sysfs_dev_release(struct device *dev) | |
3964 | { | |
3965 | } | |
3966 | ||
3967 | static struct device_type rbd_device_type = { | |
3968 | .name = "rbd", | |
3969 | .groups = rbd_attr_groups, | |
3970 | .release = rbd_sysfs_dev_release, | |
3971 | }; | |
3972 | ||
8b8fb99c AE |
3973 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3974 | { | |
3975 | kref_get(&spec->kref); | |
3976 | ||
3977 | return spec; | |
3978 | } | |
3979 | ||
3980 | static void rbd_spec_free(struct kref *kref); | |
3981 | static void rbd_spec_put(struct rbd_spec *spec) | |
3982 | { | |
3983 | if (spec) | |
3984 | kref_put(&spec->kref, rbd_spec_free); | |
3985 | } | |
3986 | ||
3987 | static struct rbd_spec *rbd_spec_alloc(void) | |
3988 | { | |
3989 | struct rbd_spec *spec; | |
3990 | ||
3991 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3992 | if (!spec) | |
3993 | return NULL; | |
04077599 ID |
3994 | |
3995 | spec->pool_id = CEPH_NOPOOL; | |
3996 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
3997 | kref_init(&spec->kref); |
3998 | ||
8b8fb99c AE |
3999 | return spec; |
4000 | } | |
4001 | ||
4002 | static void rbd_spec_free(struct kref *kref) | |
4003 | { | |
4004 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4005 | ||
4006 | kfree(spec->pool_name); | |
4007 | kfree(spec->image_id); | |
4008 | kfree(spec->image_name); | |
4009 | kfree(spec->snap_name); | |
4010 | kfree(spec); | |
4011 | } | |
4012 | ||
cc344fa1 | 4013 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
4014 | struct rbd_spec *spec) |
4015 | { | |
4016 | struct rbd_device *rbd_dev; | |
4017 | ||
4018 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4019 | if (!rbd_dev) | |
4020 | return NULL; | |
4021 | ||
4022 | spin_lock_init(&rbd_dev->lock); | |
bc1ecc65 ID |
4023 | INIT_LIST_HEAD(&rbd_dev->rq_queue); |
4024 | INIT_WORK(&rbd_dev->rq_work, rbd_request_workfn); | |
6d292906 | 4025 | rbd_dev->flags = 0; |
a2acd00e | 4026 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4027 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4028 | init_rwsem(&rbd_dev->header_rwsem); |
4029 | ||
4030 | rbd_dev->spec = spec; | |
4031 | rbd_dev->rbd_client = rbdc; | |
4032 | ||
0903e875 AE |
4033 | /* Initialize the layout used for all rbd requests */ |
4034 | ||
4035 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4036 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4037 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4038 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4039 | ||
c53d5893 AE |
4040 | return rbd_dev; |
4041 | } | |
4042 | ||
4043 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4044 | { | |
c53d5893 AE |
4045 | rbd_put_client(rbd_dev->rbd_client); |
4046 | rbd_spec_put(rbd_dev->spec); | |
4047 | kfree(rbd_dev); | |
4048 | } | |
4049 | ||
9d475de5 AE |
4050 | /* |
4051 | * Get the size and object order for an image snapshot, or if | |
4052 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4053 | * image. | |
4054 | */ | |
4055 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4056 | u8 *order, u64 *snap_size) | |
4057 | { | |
4058 | __le64 snapid = cpu_to_le64(snap_id); | |
4059 | int ret; | |
4060 | struct { | |
4061 | u8 order; | |
4062 | __le64 size; | |
4063 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4064 | ||
36be9a76 | 4065 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4066 | "rbd", "get_size", |
4157976b | 4067 | &snapid, sizeof (snapid), |
e2a58ee5 | 4068 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4069 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4070 | if (ret < 0) |
4071 | return ret; | |
57385b51 AE |
4072 | if (ret < sizeof (size_buf)) |
4073 | return -ERANGE; | |
9d475de5 | 4074 | |
c3545579 | 4075 | if (order) { |
c86f86e9 | 4076 | *order = size_buf.order; |
c3545579 JD |
4077 | dout(" order %u", (unsigned int)*order); |
4078 | } | |
9d475de5 AE |
4079 | *snap_size = le64_to_cpu(size_buf.size); |
4080 | ||
c3545579 JD |
4081 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4082 | (unsigned long long)snap_id, | |
57385b51 | 4083 | (unsigned long long)*snap_size); |
9d475de5 AE |
4084 | |
4085 | return 0; | |
4086 | } | |
4087 | ||
4088 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4089 | { | |
4090 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4091 | &rbd_dev->header.obj_order, | |
4092 | &rbd_dev->header.image_size); | |
4093 | } | |
4094 | ||
1e130199 AE |
4095 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4096 | { | |
4097 | void *reply_buf; | |
4098 | int ret; | |
4099 | void *p; | |
4100 | ||
4101 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4102 | if (!reply_buf) | |
4103 | return -ENOMEM; | |
4104 | ||
36be9a76 | 4105 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4106 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4107 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4108 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4109 | if (ret < 0) |
4110 | goto out; | |
4111 | ||
4112 | p = reply_buf; | |
4113 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4114 | p + ret, NULL, GFP_NOIO); |
4115 | ret = 0; | |
1e130199 AE |
4116 | |
4117 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4118 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4119 | rbd_dev->header.object_prefix = NULL; | |
4120 | } else { | |
4121 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4122 | } | |
1e130199 AE |
4123 | out: |
4124 | kfree(reply_buf); | |
4125 | ||
4126 | return ret; | |
4127 | } | |
4128 | ||
b1b5402a AE |
4129 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4130 | u64 *snap_features) | |
4131 | { | |
4132 | __le64 snapid = cpu_to_le64(snap_id); | |
4133 | struct { | |
4134 | __le64 features; | |
4135 | __le64 incompat; | |
4157976b | 4136 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4137 | u64 incompat; |
b1b5402a AE |
4138 | int ret; |
4139 | ||
36be9a76 | 4140 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4141 | "rbd", "get_features", |
4157976b | 4142 | &snapid, sizeof (snapid), |
e2a58ee5 | 4143 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4144 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4145 | if (ret < 0) |
4146 | return ret; | |
57385b51 AE |
4147 | if (ret < sizeof (features_buf)) |
4148 | return -ERANGE; | |
d889140c AE |
4149 | |
4150 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4151 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4152 | return -ENXIO; |
d889140c | 4153 | |
b1b5402a AE |
4154 | *snap_features = le64_to_cpu(features_buf.features); |
4155 | ||
4156 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4157 | (unsigned long long)snap_id, |
4158 | (unsigned long long)*snap_features, | |
4159 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4160 | |
4161 | return 0; | |
4162 | } | |
4163 | ||
4164 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4165 | { | |
4166 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4167 | &rbd_dev->header.features); | |
4168 | } | |
4169 | ||
86b00e0d AE |
4170 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4171 | { | |
4172 | struct rbd_spec *parent_spec; | |
4173 | size_t size; | |
4174 | void *reply_buf = NULL; | |
4175 | __le64 snapid; | |
4176 | void *p; | |
4177 | void *end; | |
642a2537 | 4178 | u64 pool_id; |
86b00e0d | 4179 | char *image_id; |
3b5cf2a2 | 4180 | u64 snap_id; |
86b00e0d | 4181 | u64 overlap; |
86b00e0d AE |
4182 | int ret; |
4183 | ||
4184 | parent_spec = rbd_spec_alloc(); | |
4185 | if (!parent_spec) | |
4186 | return -ENOMEM; | |
4187 | ||
4188 | size = sizeof (__le64) + /* pool_id */ | |
4189 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4190 | sizeof (__le64) + /* snap_id */ | |
4191 | sizeof (__le64); /* overlap */ | |
4192 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4193 | if (!reply_buf) { | |
4194 | ret = -ENOMEM; | |
4195 | goto out_err; | |
4196 | } | |
4197 | ||
4d9b67cd | 4198 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4199 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4200 | "rbd", "get_parent", |
4157976b | 4201 | &snapid, sizeof (snapid), |
e2a58ee5 | 4202 | reply_buf, size); |
36be9a76 | 4203 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4204 | if (ret < 0) |
4205 | goto out_err; | |
4206 | ||
86b00e0d | 4207 | p = reply_buf; |
57385b51 AE |
4208 | end = reply_buf + ret; |
4209 | ret = -ERANGE; | |
642a2537 | 4210 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4211 | if (pool_id == CEPH_NOPOOL) { |
4212 | /* | |
4213 | * Either the parent never existed, or we have | |
4214 | * record of it but the image got flattened so it no | |
4215 | * longer has a parent. When the parent of a | |
4216 | * layered image disappears we immediately set the | |
4217 | * overlap to 0. The effect of this is that all new | |
4218 | * requests will be treated as if the image had no | |
4219 | * parent. | |
4220 | */ | |
4221 | if (rbd_dev->parent_overlap) { | |
4222 | rbd_dev->parent_overlap = 0; | |
4223 | smp_mb(); | |
4224 | rbd_dev_parent_put(rbd_dev); | |
4225 | pr_info("%s: clone image has been flattened\n", | |
4226 | rbd_dev->disk->disk_name); | |
4227 | } | |
4228 | ||
86b00e0d | 4229 | goto out; /* No parent? No problem. */ |
392a9dad | 4230 | } |
86b00e0d | 4231 | |
0903e875 AE |
4232 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4233 | ||
4234 | ret = -EIO; | |
642a2537 | 4235 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4236 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4237 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4238 | goto out_err; |
c0cd10db | 4239 | } |
0903e875 | 4240 | |
979ed480 | 4241 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4242 | if (IS_ERR(image_id)) { |
4243 | ret = PTR_ERR(image_id); | |
4244 | goto out_err; | |
4245 | } | |
3b5cf2a2 | 4246 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4247 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4248 | ||
3b5cf2a2 AE |
4249 | /* |
4250 | * The parent won't change (except when the clone is | |
4251 | * flattened, already handled that). So we only need to | |
4252 | * record the parent spec we have not already done so. | |
4253 | */ | |
4254 | if (!rbd_dev->parent_spec) { | |
4255 | parent_spec->pool_id = pool_id; | |
4256 | parent_spec->image_id = image_id; | |
4257 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4258 | rbd_dev->parent_spec = parent_spec; |
4259 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4260 | } else { |
4261 | kfree(image_id); | |
3b5cf2a2 AE |
4262 | } |
4263 | ||
4264 | /* | |
4265 | * We always update the parent overlap. If it's zero we | |
4266 | * treat it specially. | |
4267 | */ | |
4268 | rbd_dev->parent_overlap = overlap; | |
4269 | smp_mb(); | |
4270 | if (!overlap) { | |
4271 | ||
4272 | /* A null parent_spec indicates it's the initial probe */ | |
4273 | ||
4274 | if (parent_spec) { | |
4275 | /* | |
4276 | * The overlap has become zero, so the clone | |
4277 | * must have been resized down to 0 at some | |
4278 | * point. Treat this the same as a flatten. | |
4279 | */ | |
4280 | rbd_dev_parent_put(rbd_dev); | |
4281 | pr_info("%s: clone image now standalone\n", | |
4282 | rbd_dev->disk->disk_name); | |
4283 | } else { | |
4284 | /* | |
4285 | * For the initial probe, if we find the | |
4286 | * overlap is zero we just pretend there was | |
4287 | * no parent image. | |
4288 | */ | |
9584d508 | 4289 | rbd_warn(rbd_dev, "ignoring parent with overlap 0"); |
3b5cf2a2 | 4290 | } |
70cf49cf | 4291 | } |
86b00e0d AE |
4292 | out: |
4293 | ret = 0; | |
4294 | out_err: | |
4295 | kfree(reply_buf); | |
4296 | rbd_spec_put(parent_spec); | |
4297 | ||
4298 | return ret; | |
4299 | } | |
4300 | ||
cc070d59 AE |
4301 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4302 | { | |
4303 | struct { | |
4304 | __le64 stripe_unit; | |
4305 | __le64 stripe_count; | |
4306 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4307 | size_t size = sizeof (striping_info_buf); | |
4308 | void *p; | |
4309 | u64 obj_size; | |
4310 | u64 stripe_unit; | |
4311 | u64 stripe_count; | |
4312 | int ret; | |
4313 | ||
4314 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4315 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4316 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4317 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4318 | if (ret < 0) | |
4319 | return ret; | |
4320 | if (ret < size) | |
4321 | return -ERANGE; | |
4322 | ||
4323 | /* | |
4324 | * We don't actually support the "fancy striping" feature | |
4325 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4326 | * defaults the behavior is the same as before. So find | |
4327 | * out, and only fail if the image has non-default values. | |
4328 | */ | |
4329 | ret = -EINVAL; | |
4330 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4331 | p = &striping_info_buf; | |
4332 | stripe_unit = ceph_decode_64(&p); | |
4333 | if (stripe_unit != obj_size) { | |
4334 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4335 | "(got %llu want %llu)", | |
4336 | stripe_unit, obj_size); | |
4337 | return -EINVAL; | |
4338 | } | |
4339 | stripe_count = ceph_decode_64(&p); | |
4340 | if (stripe_count != 1) { | |
4341 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4342 | "(got %llu want 1)", stripe_count); | |
4343 | return -EINVAL; | |
4344 | } | |
500d0c0f AE |
4345 | rbd_dev->header.stripe_unit = stripe_unit; |
4346 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4347 | |
4348 | return 0; | |
4349 | } | |
4350 | ||
9e15b77d AE |
4351 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4352 | { | |
4353 | size_t image_id_size; | |
4354 | char *image_id; | |
4355 | void *p; | |
4356 | void *end; | |
4357 | size_t size; | |
4358 | void *reply_buf = NULL; | |
4359 | size_t len = 0; | |
4360 | char *image_name = NULL; | |
4361 | int ret; | |
4362 | ||
4363 | rbd_assert(!rbd_dev->spec->image_name); | |
4364 | ||
69e7a02f AE |
4365 | len = strlen(rbd_dev->spec->image_id); |
4366 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4367 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4368 | if (!image_id) | |
4369 | return NULL; | |
4370 | ||
4371 | p = image_id; | |
4157976b | 4372 | end = image_id + image_id_size; |
57385b51 | 4373 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4374 | |
4375 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4376 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4377 | if (!reply_buf) | |
4378 | goto out; | |
4379 | ||
36be9a76 | 4380 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4381 | "rbd", "dir_get_name", |
4382 | image_id, image_id_size, | |
e2a58ee5 | 4383 | reply_buf, size); |
9e15b77d AE |
4384 | if (ret < 0) |
4385 | goto out; | |
4386 | p = reply_buf; | |
f40eb349 AE |
4387 | end = reply_buf + ret; |
4388 | ||
9e15b77d AE |
4389 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4390 | if (IS_ERR(image_name)) | |
4391 | image_name = NULL; | |
4392 | else | |
4393 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4394 | out: | |
4395 | kfree(reply_buf); | |
4396 | kfree(image_id); | |
4397 | ||
4398 | return image_name; | |
4399 | } | |
4400 | ||
2ad3d716 AE |
4401 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4402 | { | |
4403 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4404 | const char *snap_name; | |
4405 | u32 which = 0; | |
4406 | ||
4407 | /* Skip over names until we find the one we are looking for */ | |
4408 | ||
4409 | snap_name = rbd_dev->header.snap_names; | |
4410 | while (which < snapc->num_snaps) { | |
4411 | if (!strcmp(name, snap_name)) | |
4412 | return snapc->snaps[which]; | |
4413 | snap_name += strlen(snap_name) + 1; | |
4414 | which++; | |
4415 | } | |
4416 | return CEPH_NOSNAP; | |
4417 | } | |
4418 | ||
4419 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4420 | { | |
4421 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4422 | u32 which; | |
4423 | bool found = false; | |
4424 | u64 snap_id; | |
4425 | ||
4426 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4427 | const char *snap_name; | |
4428 | ||
4429 | snap_id = snapc->snaps[which]; | |
4430 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4431 | if (IS_ERR(snap_name)) { |
4432 | /* ignore no-longer existing snapshots */ | |
4433 | if (PTR_ERR(snap_name) == -ENOENT) | |
4434 | continue; | |
4435 | else | |
4436 | break; | |
4437 | } | |
2ad3d716 AE |
4438 | found = !strcmp(name, snap_name); |
4439 | kfree(snap_name); | |
4440 | } | |
4441 | return found ? snap_id : CEPH_NOSNAP; | |
4442 | } | |
4443 | ||
4444 | /* | |
4445 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4446 | * no snapshot by that name is found, or if an error occurs. | |
4447 | */ | |
4448 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4449 | { | |
4450 | if (rbd_dev->image_format == 1) | |
4451 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4452 | ||
4453 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4454 | } | |
4455 | ||
9e15b77d | 4456 | /* |
04077599 ID |
4457 | * An image being mapped will have everything but the snap id. |
4458 | */ | |
4459 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4460 | { | |
4461 | struct rbd_spec *spec = rbd_dev->spec; | |
4462 | ||
4463 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4464 | rbd_assert(spec->image_id && spec->image_name); | |
4465 | rbd_assert(spec->snap_name); | |
4466 | ||
4467 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4468 | u64 snap_id; | |
4469 | ||
4470 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4471 | if (snap_id == CEPH_NOSNAP) | |
4472 | return -ENOENT; | |
4473 | ||
4474 | spec->snap_id = snap_id; | |
4475 | } else { | |
4476 | spec->snap_id = CEPH_NOSNAP; | |
4477 | } | |
4478 | ||
4479 | return 0; | |
4480 | } | |
4481 | ||
4482 | /* | |
4483 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4484 | * |
04077599 ID |
4485 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4486 | * can't figure out the name for an image id. | |
9e15b77d | 4487 | */ |
04077599 | 4488 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4489 | { |
2e9f7f1c AE |
4490 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4491 | struct rbd_spec *spec = rbd_dev->spec; | |
4492 | const char *pool_name; | |
4493 | const char *image_name; | |
4494 | const char *snap_name; | |
9e15b77d AE |
4495 | int ret; |
4496 | ||
04077599 ID |
4497 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4498 | rbd_assert(spec->image_id); | |
4499 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4500 | |
2e9f7f1c | 4501 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4502 | |
2e9f7f1c AE |
4503 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4504 | if (!pool_name) { | |
4505 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4506 | return -EIO; |
4507 | } | |
2e9f7f1c AE |
4508 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4509 | if (!pool_name) | |
9e15b77d AE |
4510 | return -ENOMEM; |
4511 | ||
4512 | /* Fetch the image name; tolerate failure here */ | |
4513 | ||
2e9f7f1c AE |
4514 | image_name = rbd_dev_image_name(rbd_dev); |
4515 | if (!image_name) | |
06ecc6cb | 4516 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4517 | |
04077599 | 4518 | /* Fetch the snapshot name */ |
9e15b77d | 4519 | |
2e9f7f1c | 4520 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4521 | if (IS_ERR(snap_name)) { |
4522 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4523 | goto out_err; |
2e9f7f1c AE |
4524 | } |
4525 | ||
4526 | spec->pool_name = pool_name; | |
4527 | spec->image_name = image_name; | |
4528 | spec->snap_name = snap_name; | |
9e15b77d AE |
4529 | |
4530 | return 0; | |
04077599 | 4531 | |
9e15b77d | 4532 | out_err: |
2e9f7f1c AE |
4533 | kfree(image_name); |
4534 | kfree(pool_name); | |
9e15b77d AE |
4535 | return ret; |
4536 | } | |
4537 | ||
cc4a38bd | 4538 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4539 | { |
4540 | size_t size; | |
4541 | int ret; | |
4542 | void *reply_buf; | |
4543 | void *p; | |
4544 | void *end; | |
4545 | u64 seq; | |
4546 | u32 snap_count; | |
4547 | struct ceph_snap_context *snapc; | |
4548 | u32 i; | |
4549 | ||
4550 | /* | |
4551 | * We'll need room for the seq value (maximum snapshot id), | |
4552 | * snapshot count, and array of that many snapshot ids. | |
4553 | * For now we have a fixed upper limit on the number we're | |
4554 | * prepared to receive. | |
4555 | */ | |
4556 | size = sizeof (__le64) + sizeof (__le32) + | |
4557 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4558 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4559 | if (!reply_buf) | |
4560 | return -ENOMEM; | |
4561 | ||
36be9a76 | 4562 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4563 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4564 | reply_buf, size); |
36be9a76 | 4565 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4566 | if (ret < 0) |
4567 | goto out; | |
4568 | ||
35d489f9 | 4569 | p = reply_buf; |
57385b51 AE |
4570 | end = reply_buf + ret; |
4571 | ret = -ERANGE; | |
35d489f9 AE |
4572 | ceph_decode_64_safe(&p, end, seq, out); |
4573 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4574 | ||
4575 | /* | |
4576 | * Make sure the reported number of snapshot ids wouldn't go | |
4577 | * beyond the end of our buffer. But before checking that, | |
4578 | * make sure the computed size of the snapshot context we | |
4579 | * allocate is representable in a size_t. | |
4580 | */ | |
4581 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4582 | / sizeof (u64)) { | |
4583 | ret = -EINVAL; | |
4584 | goto out; | |
4585 | } | |
4586 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4587 | goto out; | |
468521c1 | 4588 | ret = 0; |
35d489f9 | 4589 | |
812164f8 | 4590 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4591 | if (!snapc) { |
4592 | ret = -ENOMEM; | |
4593 | goto out; | |
4594 | } | |
35d489f9 | 4595 | snapc->seq = seq; |
35d489f9 AE |
4596 | for (i = 0; i < snap_count; i++) |
4597 | snapc->snaps[i] = ceph_decode_64(&p); | |
4598 | ||
49ece554 | 4599 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4600 | rbd_dev->header.snapc = snapc; |
4601 | ||
4602 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4603 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4604 | out: |
4605 | kfree(reply_buf); | |
4606 | ||
57385b51 | 4607 | return ret; |
35d489f9 AE |
4608 | } |
4609 | ||
54cac61f AE |
4610 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4611 | u64 snap_id) | |
b8b1e2db AE |
4612 | { |
4613 | size_t size; | |
4614 | void *reply_buf; | |
54cac61f | 4615 | __le64 snapid; |
b8b1e2db AE |
4616 | int ret; |
4617 | void *p; | |
4618 | void *end; | |
b8b1e2db AE |
4619 | char *snap_name; |
4620 | ||
4621 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4622 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4623 | if (!reply_buf) | |
4624 | return ERR_PTR(-ENOMEM); | |
4625 | ||
54cac61f | 4626 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4627 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4628 | "rbd", "get_snapshot_name", |
54cac61f | 4629 | &snapid, sizeof (snapid), |
e2a58ee5 | 4630 | reply_buf, size); |
36be9a76 | 4631 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4632 | if (ret < 0) { |
4633 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4634 | goto out; |
f40eb349 | 4635 | } |
b8b1e2db AE |
4636 | |
4637 | p = reply_buf; | |
f40eb349 | 4638 | end = reply_buf + ret; |
e5c35534 | 4639 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4640 | if (IS_ERR(snap_name)) |
b8b1e2db | 4641 | goto out; |
b8b1e2db | 4642 | |
f40eb349 | 4643 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4644 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4645 | out: |
4646 | kfree(reply_buf); | |
4647 | ||
f40eb349 | 4648 | return snap_name; |
b8b1e2db AE |
4649 | } |
4650 | ||
2df3fac7 | 4651 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4652 | { |
2df3fac7 | 4653 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4654 | int ret; |
117973fb | 4655 | |
1617e40c JD |
4656 | ret = rbd_dev_v2_image_size(rbd_dev); |
4657 | if (ret) | |
cfbf6377 | 4658 | return ret; |
1617e40c | 4659 | |
2df3fac7 AE |
4660 | if (first_time) { |
4661 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4662 | if (ret) | |
cfbf6377 | 4663 | return ret; |
2df3fac7 AE |
4664 | } |
4665 | ||
cc4a38bd | 4666 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4667 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4668 | |
4669 | return ret; | |
4670 | } | |
4671 | ||
a720ae09 ID |
4672 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4673 | { | |
4674 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4675 | ||
4676 | if (rbd_dev->image_format == 1) | |
4677 | return rbd_dev_v1_header_info(rbd_dev); | |
4678 | ||
4679 | return rbd_dev_v2_header_info(rbd_dev); | |
4680 | } | |
4681 | ||
dfc5606d YS |
4682 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4683 | { | |
dfc5606d | 4684 | struct device *dev; |
cd789ab9 | 4685 | int ret; |
dfc5606d | 4686 | |
cd789ab9 | 4687 | dev = &rbd_dev->dev; |
dfc5606d YS |
4688 | dev->bus = &rbd_bus_type; |
4689 | dev->type = &rbd_device_type; | |
4690 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4691 | dev->release = rbd_dev_device_release; |
de71a297 | 4692 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4693 | ret = device_register(dev); |
dfc5606d | 4694 | |
dfc5606d | 4695 | return ret; |
602adf40 YS |
4696 | } |
4697 | ||
dfc5606d YS |
4698 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4699 | { | |
4700 | device_unregister(&rbd_dev->dev); | |
4701 | } | |
4702 | ||
1ddbe94e | 4703 | /* |
499afd5b | 4704 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4705 | * the rbd_dev to the global list. |
1ddbe94e | 4706 | */ |
f8a22fc2 | 4707 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4708 | { |
f8a22fc2 ID |
4709 | int new_dev_id; |
4710 | ||
9b60e70b ID |
4711 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4712 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4713 | GFP_KERNEL); | |
f8a22fc2 ID |
4714 | if (new_dev_id < 0) |
4715 | return new_dev_id; | |
4716 | ||
4717 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4718 | |
4719 | spin_lock(&rbd_dev_list_lock); | |
4720 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4721 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4722 | |
70eebd20 | 4723 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4724 | |
4725 | return 0; | |
1ddbe94e | 4726 | } |
b7f23c36 | 4727 | |
1ddbe94e | 4728 | /* |
499afd5b AE |
4729 | * Remove an rbd_dev from the global list, and record that its |
4730 | * identifier is no longer in use. | |
1ddbe94e | 4731 | */ |
e2839308 | 4732 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4733 | { |
499afd5b AE |
4734 | spin_lock(&rbd_dev_list_lock); |
4735 | list_del_init(&rbd_dev->node); | |
4736 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4737 | |
f8a22fc2 ID |
4738 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4739 | ||
4740 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4741 | } |
4742 | ||
e28fff26 AE |
4743 | /* |
4744 | * Skips over white space at *buf, and updates *buf to point to the | |
4745 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4746 | * the token (string of non-white space characters) found. Note |
4747 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4748 | */ |
4749 | static inline size_t next_token(const char **buf) | |
4750 | { | |
4751 | /* | |
4752 | * These are the characters that produce nonzero for | |
4753 | * isspace() in the "C" and "POSIX" locales. | |
4754 | */ | |
4755 | const char *spaces = " \f\n\r\t\v"; | |
4756 | ||
4757 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4758 | ||
4759 | return strcspn(*buf, spaces); /* Return token length */ | |
4760 | } | |
4761 | ||
4762 | /* | |
4763 | * Finds the next token in *buf, and if the provided token buffer is | |
4764 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4765 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4766 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4767 | * |
4768 | * Returns the length of the token found (not including the '\0'). | |
4769 | * Return value will be 0 if no token is found, and it will be >= | |
4770 | * token_size if the token would not fit. | |
4771 | * | |
593a9e7b | 4772 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4773 | * found token. Note that this occurs even if the token buffer is |
4774 | * too small to hold it. | |
4775 | */ | |
4776 | static inline size_t copy_token(const char **buf, | |
4777 | char *token, | |
4778 | size_t token_size) | |
4779 | { | |
4780 | size_t len; | |
4781 | ||
4782 | len = next_token(buf); | |
4783 | if (len < token_size) { | |
4784 | memcpy(token, *buf, len); | |
4785 | *(token + len) = '\0'; | |
4786 | } | |
4787 | *buf += len; | |
4788 | ||
4789 | return len; | |
4790 | } | |
4791 | ||
ea3352f4 AE |
4792 | /* |
4793 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4794 | * enough to hold a copy of it, and copies the token into the new | |
4795 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4796 | * that a duplicate buffer is created even for a zero-length token. | |
4797 | * | |
4798 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4799 | * pointer if memory for the duplicate was not available. If | |
4800 | * the lenp argument is a non-null pointer, the length of the token | |
4801 | * (not including the '\0') is returned in *lenp. | |
4802 | * | |
4803 | * If successful, the *buf pointer will be updated to point beyond | |
4804 | * the end of the found token. | |
4805 | * | |
4806 | * Note: uses GFP_KERNEL for allocation. | |
4807 | */ | |
4808 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4809 | { | |
4810 | char *dup; | |
4811 | size_t len; | |
4812 | ||
4813 | len = next_token(buf); | |
4caf35f9 | 4814 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4815 | if (!dup) |
4816 | return NULL; | |
ea3352f4 AE |
4817 | *(dup + len) = '\0'; |
4818 | *buf += len; | |
4819 | ||
4820 | if (lenp) | |
4821 | *lenp = len; | |
4822 | ||
4823 | return dup; | |
4824 | } | |
4825 | ||
a725f65e | 4826 | /* |
859c31df AE |
4827 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4828 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4829 | * and the data written is passed here via a NUL-terminated buffer. | |
4830 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4831 | * |
859c31df AE |
4832 | * The information extracted from these options is recorded in |
4833 | * the other parameters which return dynamically-allocated | |
4834 | * structures: | |
4835 | * ceph_opts | |
4836 | * The address of a pointer that will refer to a ceph options | |
4837 | * structure. Caller must release the returned pointer using | |
4838 | * ceph_destroy_options() when it is no longer needed. | |
4839 | * rbd_opts | |
4840 | * Address of an rbd options pointer. Fully initialized by | |
4841 | * this function; caller must release with kfree(). | |
4842 | * spec | |
4843 | * Address of an rbd image specification pointer. Fully | |
4844 | * initialized by this function based on parsed options. | |
4845 | * Caller must release with rbd_spec_put(). | |
4846 | * | |
4847 | * The options passed take this form: | |
4848 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4849 | * where: | |
4850 | * <mon_addrs> | |
4851 | * A comma-separated list of one or more monitor addresses. | |
4852 | * A monitor address is an ip address, optionally followed | |
4853 | * by a port number (separated by a colon). | |
4854 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4855 | * <options> | |
4856 | * A comma-separated list of ceph and/or rbd options. | |
4857 | * <pool_name> | |
4858 | * The name of the rados pool containing the rbd image. | |
4859 | * <image_name> | |
4860 | * The name of the image in that pool to map. | |
4861 | * <snap_id> | |
4862 | * An optional snapshot id. If provided, the mapping will | |
4863 | * present data from the image at the time that snapshot was | |
4864 | * created. The image head is used if no snapshot id is | |
4865 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4866 | */ |
859c31df | 4867 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4868 | struct ceph_options **ceph_opts, |
859c31df AE |
4869 | struct rbd_options **opts, |
4870 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4871 | { |
d22f76e7 | 4872 | size_t len; |
859c31df | 4873 | char *options; |
0ddebc0c | 4874 | const char *mon_addrs; |
ecb4dc22 | 4875 | char *snap_name; |
0ddebc0c | 4876 | size_t mon_addrs_size; |
859c31df | 4877 | struct rbd_spec *spec = NULL; |
4e9afeba | 4878 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4879 | struct ceph_options *copts; |
dc79b113 | 4880 | int ret; |
e28fff26 AE |
4881 | |
4882 | /* The first four tokens are required */ | |
4883 | ||
7ef3214a | 4884 | len = next_token(&buf); |
4fb5d671 AE |
4885 | if (!len) { |
4886 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4887 | return -EINVAL; | |
4888 | } | |
0ddebc0c | 4889 | mon_addrs = buf; |
f28e565a | 4890 | mon_addrs_size = len + 1; |
7ef3214a | 4891 | buf += len; |
a725f65e | 4892 | |
dc79b113 | 4893 | ret = -EINVAL; |
f28e565a AE |
4894 | options = dup_token(&buf, NULL); |
4895 | if (!options) | |
dc79b113 | 4896 | return -ENOMEM; |
4fb5d671 AE |
4897 | if (!*options) { |
4898 | rbd_warn(NULL, "no options provided"); | |
4899 | goto out_err; | |
4900 | } | |
e28fff26 | 4901 | |
859c31df AE |
4902 | spec = rbd_spec_alloc(); |
4903 | if (!spec) | |
f28e565a | 4904 | goto out_mem; |
859c31df AE |
4905 | |
4906 | spec->pool_name = dup_token(&buf, NULL); | |
4907 | if (!spec->pool_name) | |
4908 | goto out_mem; | |
4fb5d671 AE |
4909 | if (!*spec->pool_name) { |
4910 | rbd_warn(NULL, "no pool name provided"); | |
4911 | goto out_err; | |
4912 | } | |
e28fff26 | 4913 | |
69e7a02f | 4914 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4915 | if (!spec->image_name) |
f28e565a | 4916 | goto out_mem; |
4fb5d671 AE |
4917 | if (!*spec->image_name) { |
4918 | rbd_warn(NULL, "no image name provided"); | |
4919 | goto out_err; | |
4920 | } | |
d4b125e9 | 4921 | |
f28e565a AE |
4922 | /* |
4923 | * Snapshot name is optional; default is to use "-" | |
4924 | * (indicating the head/no snapshot). | |
4925 | */ | |
3feeb894 | 4926 | len = next_token(&buf); |
820a5f3e | 4927 | if (!len) { |
3feeb894 AE |
4928 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4929 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4930 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4931 | ret = -ENAMETOOLONG; |
f28e565a | 4932 | goto out_err; |
849b4260 | 4933 | } |
ecb4dc22 AE |
4934 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4935 | if (!snap_name) | |
f28e565a | 4936 | goto out_mem; |
ecb4dc22 AE |
4937 | *(snap_name + len) = '\0'; |
4938 | spec->snap_name = snap_name; | |
e5c35534 | 4939 | |
0ddebc0c | 4940 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4941 | |
4e9afeba AE |
4942 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4943 | if (!rbd_opts) | |
4944 | goto out_mem; | |
4945 | ||
4946 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4947 | |
859c31df | 4948 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4949 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4950 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4951 | if (IS_ERR(copts)) { |
4952 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4953 | goto out_err; |
4954 | } | |
859c31df AE |
4955 | kfree(options); |
4956 | ||
4957 | *ceph_opts = copts; | |
4e9afeba | 4958 | *opts = rbd_opts; |
859c31df | 4959 | *rbd_spec = spec; |
0ddebc0c | 4960 | |
dc79b113 | 4961 | return 0; |
f28e565a | 4962 | out_mem: |
dc79b113 | 4963 | ret = -ENOMEM; |
d22f76e7 | 4964 | out_err: |
859c31df AE |
4965 | kfree(rbd_opts); |
4966 | rbd_spec_put(spec); | |
f28e565a | 4967 | kfree(options); |
d22f76e7 | 4968 | |
dc79b113 | 4969 | return ret; |
a725f65e AE |
4970 | } |
4971 | ||
30ba1f02 ID |
4972 | /* |
4973 | * Return pool id (>= 0) or a negative error code. | |
4974 | */ | |
4975 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4976 | { | |
4977 | u64 newest_epoch; | |
4978 | unsigned long timeout = rbdc->client->options->mount_timeout * HZ; | |
4979 | int tries = 0; | |
4980 | int ret; | |
4981 | ||
4982 | again: | |
4983 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4984 | if (ret == -ENOENT && tries++ < 1) { | |
4985 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4986 | &newest_epoch); | |
4987 | if (ret < 0) | |
4988 | return ret; | |
4989 | ||
4990 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4991 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4992 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
4993 | newest_epoch, timeout); | |
4994 | goto again; | |
4995 | } else { | |
4996 | /* the osdmap we have is new enough */ | |
4997 | return -ENOENT; | |
4998 | } | |
4999 | } | |
5000 | ||
5001 | return ret; | |
5002 | } | |
5003 | ||
589d30e0 AE |
5004 | /* |
5005 | * An rbd format 2 image has a unique identifier, distinct from the | |
5006 | * name given to it by the user. Internally, that identifier is | |
5007 | * what's used to specify the names of objects related to the image. | |
5008 | * | |
5009 | * A special "rbd id" object is used to map an rbd image name to its | |
5010 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5011 | * with the supplied name. | |
5012 | * | |
5013 | * This function will record the given rbd_dev's image_id field if | |
5014 | * it can be determined, and in that case will return 0. If any | |
5015 | * errors occur a negative errno will be returned and the rbd_dev's | |
5016 | * image_id field will be unchanged (and should be NULL). | |
5017 | */ | |
5018 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5019 | { | |
5020 | int ret; | |
5021 | size_t size; | |
5022 | char *object_name; | |
5023 | void *response; | |
c0fba368 | 5024 | char *image_id; |
2f82ee54 | 5025 | |
2c0d0a10 AE |
5026 | /* |
5027 | * When probing a parent image, the image id is already | |
5028 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5029 | * need to fetch the image id again in this case. We |
5030 | * do still need to set the image format though. | |
2c0d0a10 | 5031 | */ |
c0fba368 AE |
5032 | if (rbd_dev->spec->image_id) { |
5033 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5034 | ||
2c0d0a10 | 5035 | return 0; |
c0fba368 | 5036 | } |
2c0d0a10 | 5037 | |
589d30e0 AE |
5038 | /* |
5039 | * First, see if the format 2 image id file exists, and if | |
5040 | * so, get the image's persistent id from it. | |
5041 | */ | |
69e7a02f | 5042 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5043 | object_name = kmalloc(size, GFP_NOIO); |
5044 | if (!object_name) | |
5045 | return -ENOMEM; | |
0d7dbfce | 5046 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5047 | dout("rbd id object name is %s\n", object_name); |
5048 | ||
5049 | /* Response will be an encoded string, which includes a length */ | |
5050 | ||
5051 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5052 | response = kzalloc(size, GFP_NOIO); | |
5053 | if (!response) { | |
5054 | ret = -ENOMEM; | |
5055 | goto out; | |
5056 | } | |
5057 | ||
c0fba368 AE |
5058 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5059 | ||
36be9a76 | 5060 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5061 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5062 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5063 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5064 | if (ret == -ENOENT) { |
5065 | image_id = kstrdup("", GFP_KERNEL); | |
5066 | ret = image_id ? 0 : -ENOMEM; | |
5067 | if (!ret) | |
5068 | rbd_dev->image_format = 1; | |
7dd440c9 | 5069 | } else if (ret >= 0) { |
c0fba368 AE |
5070 | void *p = response; |
5071 | ||
5072 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5073 | NULL, GFP_NOIO); |
461f758a | 5074 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5075 | if (!ret) |
5076 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5077 | } |
5078 | ||
5079 | if (!ret) { | |
5080 | rbd_dev->spec->image_id = image_id; | |
5081 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5082 | } |
5083 | out: | |
5084 | kfree(response); | |
5085 | kfree(object_name); | |
5086 | ||
5087 | return ret; | |
5088 | } | |
5089 | ||
3abef3b3 AE |
5090 | /* |
5091 | * Undo whatever state changes are made by v1 or v2 header info | |
5092 | * call. | |
5093 | */ | |
6fd48b3b AE |
5094 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5095 | { | |
5096 | struct rbd_image_header *header; | |
5097 | ||
392a9dad AE |
5098 | /* Drop parent reference unless it's already been done (or none) */ |
5099 | ||
5100 | if (rbd_dev->parent_overlap) | |
5101 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
5102 | |
5103 | /* Free dynamic fields from the header, then zero it out */ | |
5104 | ||
5105 | header = &rbd_dev->header; | |
812164f8 | 5106 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5107 | kfree(header->snap_sizes); |
5108 | kfree(header->snap_names); | |
5109 | kfree(header->object_prefix); | |
5110 | memset(header, 0, sizeof (*header)); | |
5111 | } | |
5112 | ||
2df3fac7 | 5113 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5114 | { |
5115 | int ret; | |
a30b71b9 | 5116 | |
1e130199 | 5117 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5118 | if (ret) |
b1b5402a AE |
5119 | goto out_err; |
5120 | ||
2df3fac7 AE |
5121 | /* |
5122 | * Get the and check features for the image. Currently the | |
5123 | * features are assumed to never change. | |
5124 | */ | |
b1b5402a | 5125 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5126 | if (ret) |
9d475de5 | 5127 | goto out_err; |
35d489f9 | 5128 | |
cc070d59 AE |
5129 | /* If the image supports fancy striping, get its parameters */ |
5130 | ||
5131 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5132 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5133 | if (ret < 0) | |
5134 | goto out_err; | |
5135 | } | |
2df3fac7 | 5136 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5137 | |
35152979 | 5138 | return 0; |
9d475de5 | 5139 | out_err: |
642a2537 | 5140 | rbd_dev->header.features = 0; |
1e130199 AE |
5141 | kfree(rbd_dev->header.object_prefix); |
5142 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5143 | |
5144 | return ret; | |
a30b71b9 AE |
5145 | } |
5146 | ||
124afba2 | 5147 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 5148 | { |
2f82ee54 | 5149 | struct rbd_device *parent = NULL; |
124afba2 AE |
5150 | struct rbd_spec *parent_spec; |
5151 | struct rbd_client *rbdc; | |
5152 | int ret; | |
5153 | ||
5154 | if (!rbd_dev->parent_spec) | |
5155 | return 0; | |
5156 | /* | |
5157 | * We need to pass a reference to the client and the parent | |
5158 | * spec when creating the parent rbd_dev. Images related by | |
5159 | * parent/child relationships always share both. | |
5160 | */ | |
5161 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
5162 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
5163 | ||
5164 | ret = -ENOMEM; | |
5165 | parent = rbd_dev_create(rbdc, parent_spec); | |
5166 | if (!parent) | |
5167 | goto out_err; | |
5168 | ||
1f3ef788 | 5169 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
5170 | if (ret < 0) |
5171 | goto out_err; | |
5172 | rbd_dev->parent = parent; | |
a2acd00e | 5173 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
5174 | |
5175 | return 0; | |
5176 | out_err: | |
5177 | if (parent) { | |
fb65d228 | 5178 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
5179 | kfree(rbd_dev->header_name); |
5180 | rbd_dev_destroy(parent); | |
5181 | } else { | |
5182 | rbd_put_client(rbdc); | |
5183 | rbd_spec_put(parent_spec); | |
5184 | } | |
5185 | ||
5186 | return ret; | |
5187 | } | |
5188 | ||
200a6a8b | 5189 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5190 | { |
83a06263 | 5191 | int ret; |
d1cf5788 | 5192 | |
f8a22fc2 ID |
5193 | /* Get an id and fill in device name. */ |
5194 | ||
5195 | ret = rbd_dev_id_get(rbd_dev); | |
5196 | if (ret) | |
5197 | return ret; | |
83a06263 | 5198 | |
83a06263 AE |
5199 | BUILD_BUG_ON(DEV_NAME_LEN |
5200 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5201 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5202 | ||
9b60e70b | 5203 | /* Record our major and minor device numbers. */ |
83a06263 | 5204 | |
9b60e70b ID |
5205 | if (!single_major) { |
5206 | ret = register_blkdev(0, rbd_dev->name); | |
5207 | if (ret < 0) | |
5208 | goto err_out_id; | |
5209 | ||
5210 | rbd_dev->major = ret; | |
5211 | rbd_dev->minor = 0; | |
5212 | } else { | |
5213 | rbd_dev->major = rbd_major; | |
5214 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5215 | } | |
83a06263 AE |
5216 | |
5217 | /* Set up the blkdev mapping. */ | |
5218 | ||
5219 | ret = rbd_init_disk(rbd_dev); | |
5220 | if (ret) | |
5221 | goto err_out_blkdev; | |
5222 | ||
f35a4dee | 5223 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5224 | if (ret) |
5225 | goto err_out_disk; | |
bc1ecc65 | 5226 | |
f35a4dee | 5227 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5228 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 5229 | |
58d1362b | 5230 | rbd_dev->rq_wq = alloc_workqueue("%s", 0, 0, rbd_dev->disk->disk_name); |
255939e7 WY |
5231 | if (!rbd_dev->rq_wq) { |
5232 | ret = -ENOMEM; | |
bc1ecc65 | 5233 | goto err_out_mapping; |
255939e7 | 5234 | } |
bc1ecc65 | 5235 | |
f35a4dee AE |
5236 | ret = rbd_bus_add_dev(rbd_dev); |
5237 | if (ret) | |
bc1ecc65 | 5238 | goto err_out_workqueue; |
83a06263 | 5239 | |
83a06263 AE |
5240 | /* Everything's ready. Announce the disk to the world. */ |
5241 | ||
129b79d4 | 5242 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5243 | add_disk(rbd_dev->disk); |
5244 | ||
5245 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5246 | (unsigned long long) rbd_dev->mapping.size); | |
5247 | ||
5248 | return ret; | |
2f82ee54 | 5249 | |
bc1ecc65 ID |
5250 | err_out_workqueue: |
5251 | destroy_workqueue(rbd_dev->rq_wq); | |
5252 | rbd_dev->rq_wq = NULL; | |
f35a4dee AE |
5253 | err_out_mapping: |
5254 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5255 | err_out_disk: |
5256 | rbd_free_disk(rbd_dev); | |
5257 | err_out_blkdev: | |
9b60e70b ID |
5258 | if (!single_major) |
5259 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5260 | err_out_id: |
5261 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5262 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5263 | |
5264 | return ret; | |
5265 | } | |
5266 | ||
332bb12d AE |
5267 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5268 | { | |
5269 | struct rbd_spec *spec = rbd_dev->spec; | |
5270 | size_t size; | |
5271 | ||
5272 | /* Record the header object name for this rbd image. */ | |
5273 | ||
5274 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5275 | ||
5276 | if (rbd_dev->image_format == 1) | |
5277 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5278 | else | |
5279 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5280 | ||
5281 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5282 | if (!rbd_dev->header_name) | |
5283 | return -ENOMEM; | |
5284 | ||
5285 | if (rbd_dev->image_format == 1) | |
5286 | sprintf(rbd_dev->header_name, "%s%s", | |
5287 | spec->image_name, RBD_SUFFIX); | |
5288 | else | |
5289 | sprintf(rbd_dev->header_name, "%s%s", | |
5290 | RBD_HEADER_PREFIX, spec->image_id); | |
5291 | return 0; | |
5292 | } | |
5293 | ||
200a6a8b AE |
5294 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5295 | { | |
6fd48b3b | 5296 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5297 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5298 | rbd_dev->header_name = NULL; |
5299 | rbd_dev->image_format = 0; | |
5300 | kfree(rbd_dev->spec->image_id); | |
5301 | rbd_dev->spec->image_id = NULL; | |
5302 | ||
200a6a8b AE |
5303 | rbd_dev_destroy(rbd_dev); |
5304 | } | |
5305 | ||
a30b71b9 AE |
5306 | /* |
5307 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5308 | * device. If this image is the one being mapped (i.e., not a |
5309 | * parent), initiate a watch on its header object before using that | |
5310 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5311 | */ |
1f3ef788 | 5312 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
5313 | { |
5314 | int ret; | |
5315 | ||
5316 | /* | |
3abef3b3 AE |
5317 | * Get the id from the image id object. Unless there's an |
5318 | * error, rbd_dev->spec->image_id will be filled in with | |
5319 | * a dynamically-allocated string, and rbd_dev->image_format | |
5320 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5321 | */ |
5322 | ret = rbd_dev_image_id(rbd_dev); | |
5323 | if (ret) | |
c0fba368 | 5324 | return ret; |
c0fba368 | 5325 | |
332bb12d AE |
5326 | ret = rbd_dev_header_name(rbd_dev); |
5327 | if (ret) | |
5328 | goto err_out_format; | |
5329 | ||
1f3ef788 | 5330 | if (mapping) { |
fca27065 | 5331 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1f3ef788 AE |
5332 | if (ret) |
5333 | goto out_header_name; | |
5334 | } | |
b644de2b | 5335 | |
a720ae09 | 5336 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5337 | if (ret) |
b644de2b | 5338 | goto err_out_watch; |
83a06263 | 5339 | |
04077599 ID |
5340 | /* |
5341 | * If this image is the one being mapped, we have pool name and | |
5342 | * id, image name and id, and snap name - need to fill snap id. | |
5343 | * Otherwise this is a parent image, identified by pool, image | |
5344 | * and snap ids - need to fill in names for those ids. | |
5345 | */ | |
5346 | if (mapping) | |
5347 | ret = rbd_spec_fill_snap_id(rbd_dev); | |
5348 | else | |
5349 | ret = rbd_spec_fill_names(rbd_dev); | |
9bb81c9b | 5350 | if (ret) |
33dca39f | 5351 | goto err_out_probe; |
9bb81c9b | 5352 | |
e8f59b59 ID |
5353 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5354 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5355 | if (ret) | |
5356 | goto err_out_probe; | |
5357 | ||
5358 | /* | |
5359 | * Need to warn users if this image is the one being | |
5360 | * mapped and has a parent. | |
5361 | */ | |
5362 | if (mapping && rbd_dev->parent_spec) | |
5363 | rbd_warn(rbd_dev, | |
5364 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5365 | } | |
5366 | ||
9bb81c9b | 5367 | ret = rbd_dev_probe_parent(rbd_dev); |
30d60ba2 AE |
5368 | if (ret) |
5369 | goto err_out_probe; | |
5370 | ||
5371 | dout("discovered format %u image, header name is %s\n", | |
5372 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5373 | return 0; |
e8f59b59 | 5374 | |
6fd48b3b AE |
5375 | err_out_probe: |
5376 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5377 | err_out_watch: |
fca27065 ID |
5378 | if (mapping) |
5379 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5380 | out_header_name: |
5381 | kfree(rbd_dev->header_name); | |
5382 | rbd_dev->header_name = NULL; | |
5383 | err_out_format: | |
5384 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5385 | kfree(rbd_dev->spec->image_id); |
5386 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5387 | return ret; |
5388 | } | |
5389 | ||
9b60e70b ID |
5390 | static ssize_t do_rbd_add(struct bus_type *bus, |
5391 | const char *buf, | |
5392 | size_t count) | |
602adf40 | 5393 | { |
cb8627c7 | 5394 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5395 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5396 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5397 | struct rbd_spec *spec = NULL; |
9d3997fd | 5398 | struct rbd_client *rbdc; |
51344a38 | 5399 | bool read_only; |
27cc2594 | 5400 | int rc = -ENOMEM; |
602adf40 YS |
5401 | |
5402 | if (!try_module_get(THIS_MODULE)) | |
5403 | return -ENODEV; | |
5404 | ||
602adf40 | 5405 | /* parse add command */ |
859c31df | 5406 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5407 | if (rc < 0) |
bd4ba655 | 5408 | goto err_out_module; |
51344a38 AE |
5409 | read_only = rbd_opts->read_only; |
5410 | kfree(rbd_opts); | |
5411 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5412 | |
9d3997fd AE |
5413 | rbdc = rbd_get_client(ceph_opts); |
5414 | if (IS_ERR(rbdc)) { | |
5415 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5416 | goto err_out_args; |
9d3997fd | 5417 | } |
602adf40 | 5418 | |
602adf40 | 5419 | /* pick the pool */ |
30ba1f02 | 5420 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
602adf40 YS |
5421 | if (rc < 0) |
5422 | goto err_out_client; | |
c0cd10db | 5423 | spec->pool_id = (u64)rc; |
859c31df | 5424 | |
0903e875 AE |
5425 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5426 | ||
c0cd10db | 5427 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5428 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5429 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5430 | rc = -EIO; |
5431 | goto err_out_client; | |
5432 | } | |
5433 | ||
c53d5893 | 5434 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5435 | if (!rbd_dev) |
5436 | goto err_out_client; | |
c53d5893 AE |
5437 | rbdc = NULL; /* rbd_dev now owns this */ |
5438 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5439 | |
1f3ef788 | 5440 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5441 | if (rc < 0) |
c53d5893 | 5442 | goto err_out_rbd_dev; |
05fd6f6f | 5443 | |
7ce4eef7 AE |
5444 | /* If we are mapping a snapshot it must be marked read-only */ |
5445 | ||
5446 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5447 | read_only = true; | |
5448 | rbd_dev->mapping.read_only = read_only; | |
5449 | ||
b536f69a | 5450 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5451 | if (rc) { |
e37180c0 ID |
5452 | /* |
5453 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5454 | * rbd_dev_image_release() without refactoring, see | |
5455 | * commit 1f3ef78861ac. | |
5456 | */ | |
5457 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5458 | rbd_dev_image_release(rbd_dev); |
5459 | goto err_out_module; | |
5460 | } | |
5461 | ||
5462 | return count; | |
b536f69a | 5463 | |
c53d5893 AE |
5464 | err_out_rbd_dev: |
5465 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5466 | err_out_client: |
9d3997fd | 5467 | rbd_put_client(rbdc); |
0ddebc0c | 5468 | err_out_args: |
859c31df | 5469 | rbd_spec_put(spec); |
bd4ba655 AE |
5470 | err_out_module: |
5471 | module_put(THIS_MODULE); | |
27cc2594 | 5472 | |
602adf40 | 5473 | dout("Error adding device %s\n", buf); |
27cc2594 | 5474 | |
c0cd10db | 5475 | return (ssize_t)rc; |
602adf40 YS |
5476 | } |
5477 | ||
9b60e70b ID |
5478 | static ssize_t rbd_add(struct bus_type *bus, |
5479 | const char *buf, | |
5480 | size_t count) | |
5481 | { | |
5482 | if (single_major) | |
5483 | return -EINVAL; | |
5484 | ||
5485 | return do_rbd_add(bus, buf, count); | |
5486 | } | |
5487 | ||
5488 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5489 | const char *buf, | |
5490 | size_t count) | |
5491 | { | |
5492 | return do_rbd_add(bus, buf, count); | |
5493 | } | |
5494 | ||
200a6a8b | 5495 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5496 | { |
593a9e7b | 5497 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5498 | |
bc1ecc65 | 5499 | destroy_workqueue(rbd_dev->rq_wq); |
602adf40 | 5500 | rbd_free_disk(rbd_dev); |
200a6a8b | 5501 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5502 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5503 | if (!single_major) |
5504 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5505 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5506 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5507 | } |
5508 | ||
05a46afd AE |
5509 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5510 | { | |
ad945fc1 | 5511 | while (rbd_dev->parent) { |
05a46afd AE |
5512 | struct rbd_device *first = rbd_dev; |
5513 | struct rbd_device *second = first->parent; | |
5514 | struct rbd_device *third; | |
5515 | ||
5516 | /* | |
5517 | * Follow to the parent with no grandparent and | |
5518 | * remove it. | |
5519 | */ | |
5520 | while (second && (third = second->parent)) { | |
5521 | first = second; | |
5522 | second = third; | |
5523 | } | |
ad945fc1 | 5524 | rbd_assert(second); |
8ad42cd0 | 5525 | rbd_dev_image_release(second); |
ad945fc1 AE |
5526 | first->parent = NULL; |
5527 | first->parent_overlap = 0; | |
5528 | ||
5529 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5530 | rbd_spec_put(first->parent_spec); |
5531 | first->parent_spec = NULL; | |
05a46afd AE |
5532 | } |
5533 | } | |
5534 | ||
9b60e70b ID |
5535 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5536 | const char *buf, | |
5537 | size_t count) | |
602adf40 YS |
5538 | { |
5539 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5540 | struct list_head *tmp; |
5541 | int dev_id; | |
602adf40 | 5542 | unsigned long ul; |
82a442d2 | 5543 | bool already = false; |
0d8189e1 | 5544 | int ret; |
602adf40 | 5545 | |
bb8e0e84 | 5546 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5547 | if (ret) |
5548 | return ret; | |
602adf40 YS |
5549 | |
5550 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5551 | dev_id = (int)ul; |
5552 | if (dev_id != ul) | |
602adf40 YS |
5553 | return -EINVAL; |
5554 | ||
751cc0e3 AE |
5555 | ret = -ENOENT; |
5556 | spin_lock(&rbd_dev_list_lock); | |
5557 | list_for_each(tmp, &rbd_dev_list) { | |
5558 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5559 | if (rbd_dev->dev_id == dev_id) { | |
5560 | ret = 0; | |
5561 | break; | |
5562 | } | |
42382b70 | 5563 | } |
751cc0e3 AE |
5564 | if (!ret) { |
5565 | spin_lock_irq(&rbd_dev->lock); | |
5566 | if (rbd_dev->open_count) | |
5567 | ret = -EBUSY; | |
5568 | else | |
82a442d2 AE |
5569 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5570 | &rbd_dev->flags); | |
751cc0e3 AE |
5571 | spin_unlock_irq(&rbd_dev->lock); |
5572 | } | |
5573 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5574 | if (ret < 0 || already) |
1ba0f1e7 | 5575 | return ret; |
751cc0e3 | 5576 | |
fca27065 | 5577 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5578 | /* |
5579 | * flush remaining watch callbacks - these must be complete | |
5580 | * before the osd_client is shutdown | |
5581 | */ | |
5582 | dout("%s: flushing notifies", __func__); | |
5583 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5584 | |
9875201e JD |
5585 | /* |
5586 | * Don't free anything from rbd_dev->disk until after all | |
5587 | * notifies are completely processed. Otherwise | |
5588 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5589 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5590 | */ | |
5591 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5592 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5593 | module_put(THIS_MODULE); |
aafb230e | 5594 | |
1ba0f1e7 | 5595 | return count; |
602adf40 YS |
5596 | } |
5597 | ||
9b60e70b ID |
5598 | static ssize_t rbd_remove(struct bus_type *bus, |
5599 | const char *buf, | |
5600 | size_t count) | |
5601 | { | |
5602 | if (single_major) | |
5603 | return -EINVAL; | |
5604 | ||
5605 | return do_rbd_remove(bus, buf, count); | |
5606 | } | |
5607 | ||
5608 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5609 | const char *buf, | |
5610 | size_t count) | |
5611 | { | |
5612 | return do_rbd_remove(bus, buf, count); | |
5613 | } | |
5614 | ||
602adf40 YS |
5615 | /* |
5616 | * create control files in sysfs | |
dfc5606d | 5617 | * /sys/bus/rbd/... |
602adf40 YS |
5618 | */ |
5619 | static int rbd_sysfs_init(void) | |
5620 | { | |
dfc5606d | 5621 | int ret; |
602adf40 | 5622 | |
fed4c143 | 5623 | ret = device_register(&rbd_root_dev); |
21079786 | 5624 | if (ret < 0) |
dfc5606d | 5625 | return ret; |
602adf40 | 5626 | |
fed4c143 AE |
5627 | ret = bus_register(&rbd_bus_type); |
5628 | if (ret < 0) | |
5629 | device_unregister(&rbd_root_dev); | |
602adf40 | 5630 | |
602adf40 YS |
5631 | return ret; |
5632 | } | |
5633 | ||
5634 | static void rbd_sysfs_cleanup(void) | |
5635 | { | |
dfc5606d | 5636 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5637 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5638 | } |
5639 | ||
1c2a9dfe AE |
5640 | static int rbd_slab_init(void) |
5641 | { | |
5642 | rbd_assert(!rbd_img_request_cache); | |
5643 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5644 | sizeof (struct rbd_img_request), | |
5645 | __alignof__(struct rbd_img_request), | |
5646 | 0, NULL); | |
868311b1 AE |
5647 | if (!rbd_img_request_cache) |
5648 | return -ENOMEM; | |
5649 | ||
5650 | rbd_assert(!rbd_obj_request_cache); | |
5651 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5652 | sizeof (struct rbd_obj_request), | |
5653 | __alignof__(struct rbd_obj_request), | |
5654 | 0, NULL); | |
78c2a44a AE |
5655 | if (!rbd_obj_request_cache) |
5656 | goto out_err; | |
5657 | ||
5658 | rbd_assert(!rbd_segment_name_cache); | |
5659 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5660 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5661 | if (rbd_segment_name_cache) |
1c2a9dfe | 5662 | return 0; |
78c2a44a AE |
5663 | out_err: |
5664 | if (rbd_obj_request_cache) { | |
5665 | kmem_cache_destroy(rbd_obj_request_cache); | |
5666 | rbd_obj_request_cache = NULL; | |
5667 | } | |
1c2a9dfe | 5668 | |
868311b1 AE |
5669 | kmem_cache_destroy(rbd_img_request_cache); |
5670 | rbd_img_request_cache = NULL; | |
5671 | ||
1c2a9dfe AE |
5672 | return -ENOMEM; |
5673 | } | |
5674 | ||
5675 | static void rbd_slab_exit(void) | |
5676 | { | |
78c2a44a AE |
5677 | rbd_assert(rbd_segment_name_cache); |
5678 | kmem_cache_destroy(rbd_segment_name_cache); | |
5679 | rbd_segment_name_cache = NULL; | |
5680 | ||
868311b1 AE |
5681 | rbd_assert(rbd_obj_request_cache); |
5682 | kmem_cache_destroy(rbd_obj_request_cache); | |
5683 | rbd_obj_request_cache = NULL; | |
5684 | ||
1c2a9dfe AE |
5685 | rbd_assert(rbd_img_request_cache); |
5686 | kmem_cache_destroy(rbd_img_request_cache); | |
5687 | rbd_img_request_cache = NULL; | |
5688 | } | |
5689 | ||
cc344fa1 | 5690 | static int __init rbd_init(void) |
602adf40 YS |
5691 | { |
5692 | int rc; | |
5693 | ||
1e32d34c AE |
5694 | if (!libceph_compatible(NULL)) { |
5695 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5696 | return -EINVAL; |
5697 | } | |
e1b4d96d | 5698 | |
1c2a9dfe | 5699 | rc = rbd_slab_init(); |
602adf40 YS |
5700 | if (rc) |
5701 | return rc; | |
e1b4d96d | 5702 | |
9b60e70b ID |
5703 | if (single_major) { |
5704 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5705 | if (rbd_major < 0) { | |
5706 | rc = rbd_major; | |
5707 | goto err_out_slab; | |
5708 | } | |
5709 | } | |
5710 | ||
1c2a9dfe AE |
5711 | rc = rbd_sysfs_init(); |
5712 | if (rc) | |
9b60e70b ID |
5713 | goto err_out_blkdev; |
5714 | ||
5715 | if (single_major) | |
5716 | pr_info("loaded (major %d)\n", rbd_major); | |
5717 | else | |
5718 | pr_info("loaded\n"); | |
1c2a9dfe | 5719 | |
e1b4d96d ID |
5720 | return 0; |
5721 | ||
9b60e70b ID |
5722 | err_out_blkdev: |
5723 | if (single_major) | |
5724 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
e1b4d96d ID |
5725 | err_out_slab: |
5726 | rbd_slab_exit(); | |
1c2a9dfe | 5727 | return rc; |
602adf40 YS |
5728 | } |
5729 | ||
cc344fa1 | 5730 | static void __exit rbd_exit(void) |
602adf40 | 5731 | { |
ffe312cf | 5732 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5733 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5734 | if (single_major) |
5735 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
1c2a9dfe | 5736 | rbd_slab_exit(); |
602adf40 YS |
5737 | } |
5738 | ||
5739 | module_init(rbd_init); | |
5740 | module_exit(rbd_exit); | |
5741 | ||
d552c619 | 5742 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5743 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5744 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5745 | /* following authorship retained from original osdblk.c */ |
5746 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5747 | ||
90da258b | 5748 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5749 | MODULE_LICENSE("GPL"); |