]>
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 | ||
3b434a2a JD |
1675 | static enum obj_operation_type |
1676 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1677 | { | |
1678 | if (img_request_write_test(img_request)) | |
1679 | return OBJ_OP_WRITE; | |
1680 | else if (img_request_discard_test(img_request)) | |
1681 | return OBJ_OP_DISCARD; | |
1682 | else | |
1683 | return OBJ_OP_READ; | |
1684 | } | |
1685 | ||
6e2a4505 AE |
1686 | static void |
1687 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1688 | { | |
b9434c5b AE |
1689 | u64 xferred = obj_request->xferred; |
1690 | u64 length = obj_request->length; | |
1691 | ||
6e2a4505 AE |
1692 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1693 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1694 | xferred, length); |
6e2a4505 | 1695 | /* |
17c1cc1d JD |
1696 | * ENOENT means a hole in the image. We zero-fill the entire |
1697 | * length of the request. A short read also implies zero-fill | |
1698 | * to the end of the request. An error requires the whole | |
1699 | * length of the request to be reported finished with an error | |
1700 | * to the block layer. In each case we update the xferred | |
1701 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1702 | */ |
b9434c5b | 1703 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1704 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1705 | if (obj_request->type == OBJ_REQUEST_BIO) |
1706 | zero_bio_chain(obj_request->bio_list, 0); | |
1707 | else | |
1708 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1709 | obj_request->result = 0; |
b9434c5b AE |
1710 | } else if (xferred < length && !obj_request->result) { |
1711 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1712 | zero_bio_chain(obj_request->bio_list, xferred); | |
1713 | else | |
1714 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1715 | } |
17c1cc1d | 1716 | obj_request->xferred = length; |
6e2a4505 AE |
1717 | obj_request_done_set(obj_request); |
1718 | } | |
1719 | ||
bf0d5f50 AE |
1720 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1721 | { | |
37206ee5 AE |
1722 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1723 | obj_request->callback); | |
bf0d5f50 AE |
1724 | if (obj_request->callback) |
1725 | obj_request->callback(obj_request); | |
788e2df3 AE |
1726 | else |
1727 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1728 | } |
1729 | ||
c47f9371 | 1730 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1731 | { |
1732 | dout("%s: obj %p\n", __func__, obj_request); | |
1733 | obj_request_done_set(obj_request); | |
1734 | } | |
1735 | ||
c47f9371 | 1736 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1737 | { |
57acbaa7 | 1738 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1739 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1740 | bool layered = false; |
1741 | ||
1742 | if (obj_request_img_data_test(obj_request)) { | |
1743 | img_request = obj_request->img_request; | |
1744 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1745 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1746 | } |
8b3e1a56 AE |
1747 | |
1748 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1749 | obj_request, img_request, obj_request->result, | |
1750 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1751 | if (layered && obj_request->result == -ENOENT && |
1752 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1753 | rbd_img_parent_read(obj_request); |
1754 | else if (img_request) | |
6e2a4505 AE |
1755 | rbd_img_obj_request_read_callback(obj_request); |
1756 | else | |
1757 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1758 | } |
1759 | ||
c47f9371 | 1760 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1761 | { |
1b83bef2 SW |
1762 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1763 | obj_request->result, obj_request->length); | |
1764 | /* | |
8b3e1a56 AE |
1765 | * There is no such thing as a successful short write. Set |
1766 | * it to our originally-requested length. | |
1b83bef2 SW |
1767 | */ |
1768 | obj_request->xferred = obj_request->length; | |
07741308 | 1769 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1770 | } |
1771 | ||
90e98c52 GZ |
1772 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1773 | { | |
1774 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1775 | obj_request->result, obj_request->length); | |
1776 | /* | |
1777 | * There is no such thing as a successful short discard. Set | |
1778 | * it to our originally-requested length. | |
1779 | */ | |
1780 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1781 | /* discarding a non-existent object is not a problem */ |
1782 | if (obj_request->result == -ENOENT) | |
1783 | obj_request->result = 0; | |
90e98c52 GZ |
1784 | obj_request_done_set(obj_request); |
1785 | } | |
1786 | ||
fbfab539 AE |
1787 | /* |
1788 | * For a simple stat call there's nothing to do. We'll do more if | |
1789 | * this is part of a write sequence for a layered image. | |
1790 | */ | |
c47f9371 | 1791 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1792 | { |
37206ee5 | 1793 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1794 | obj_request_done_set(obj_request); |
1795 | } | |
1796 | ||
bf0d5f50 AE |
1797 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1798 | struct ceph_msg *msg) | |
1799 | { | |
1800 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1801 | u16 opcode; |
1802 | ||
37206ee5 | 1803 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1804 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1805 | if (obj_request_img_data_test(obj_request)) { |
1806 | rbd_assert(obj_request->img_request); | |
1807 | rbd_assert(obj_request->which != BAD_WHICH); | |
1808 | } else { | |
1809 | rbd_assert(obj_request->which == BAD_WHICH); | |
1810 | } | |
bf0d5f50 | 1811 | |
1b83bef2 SW |
1812 | if (osd_req->r_result < 0) |
1813 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1814 | |
7cc69d42 | 1815 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1816 | |
c47f9371 AE |
1817 | /* |
1818 | * We support a 64-bit length, but ultimately it has to be | |
1819 | * passed to blk_end_request(), which takes an unsigned int. | |
1820 | */ | |
1b83bef2 | 1821 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1822 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1823 | |
79528734 | 1824 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1825 | switch (opcode) { |
1826 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1827 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1828 | break; |
0ccd5926 ID |
1829 | case CEPH_OSD_OP_SETALLOCHINT: |
1830 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE); | |
1831 | /* fall through */ | |
bf0d5f50 | 1832 | case CEPH_OSD_OP_WRITE: |
c47f9371 | 1833 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1834 | break; |
fbfab539 | 1835 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1836 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1837 | break; |
90e98c52 GZ |
1838 | case CEPH_OSD_OP_DELETE: |
1839 | case CEPH_OSD_OP_TRUNCATE: | |
1840 | case CEPH_OSD_OP_ZERO: | |
1841 | rbd_osd_discard_callback(obj_request); | |
1842 | break; | |
36be9a76 | 1843 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1844 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1845 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1846 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1847 | break; |
bf0d5f50 | 1848 | default: |
9584d508 | 1849 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1850 | obj_request->object_name, (unsigned short) opcode); |
1851 | break; | |
1852 | } | |
1853 | ||
07741308 | 1854 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1855 | rbd_obj_request_complete(obj_request); |
1856 | } | |
1857 | ||
9d4df01f | 1858 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1859 | { |
1860 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1861 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1862 | u64 snap_id; |
430c28c3 | 1863 | |
8c042b0d | 1864 | rbd_assert(osd_req != NULL); |
430c28c3 | 1865 | |
9d4df01f | 1866 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1867 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1868 | NULL, snap_id, NULL); |
1869 | } | |
1870 | ||
1871 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1872 | { | |
1873 | struct rbd_img_request *img_request = obj_request->img_request; | |
1874 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1875 | struct ceph_snap_context *snapc; | |
1876 | struct timespec mtime = CURRENT_TIME; | |
1877 | ||
1878 | rbd_assert(osd_req != NULL); | |
1879 | ||
1880 | snapc = img_request ? img_request->snapc : NULL; | |
1881 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1882 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1883 | } |
1884 | ||
0ccd5926 ID |
1885 | /* |
1886 | * Create an osd request. A read request has one osd op (read). | |
1887 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1888 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1889 | * technically osd watch is a write request, hence this distinction.) | |
1890 | */ | |
bf0d5f50 AE |
1891 | static struct ceph_osd_request *rbd_osd_req_create( |
1892 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1893 | enum obj_operation_type op_type, |
deb236b3 | 1894 | unsigned int num_ops, |
430c28c3 | 1895 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1896 | { |
bf0d5f50 AE |
1897 | struct ceph_snap_context *snapc = NULL; |
1898 | struct ceph_osd_client *osdc; | |
1899 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1900 | |
90e98c52 GZ |
1901 | if (obj_request_img_data_test(obj_request) && |
1902 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1903 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1904 | if (op_type == OBJ_OP_WRITE) { |
1905 | rbd_assert(img_request_write_test(img_request)); | |
1906 | } else { | |
1907 | rbd_assert(img_request_discard_test(img_request)); | |
1908 | } | |
6d2940c8 | 1909 | snapc = img_request->snapc; |
bf0d5f50 AE |
1910 | } |
1911 | ||
6d2940c8 | 1912 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1913 | |
1914 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1915 | |
1916 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1917 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1918 | GFP_ATOMIC); | |
bf0d5f50 AE |
1919 | if (!osd_req) |
1920 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1921 | |
90e98c52 | 1922 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1923 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1924 | else |
bf0d5f50 | 1925 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1926 | |
1927 | osd_req->r_callback = rbd_osd_req_callback; | |
1928 | osd_req->r_priv = obj_request; | |
1929 | ||
3c972c95 ID |
1930 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1931 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1932 | |
bf0d5f50 AE |
1933 | return osd_req; |
1934 | } | |
1935 | ||
0eefd470 | 1936 | /* |
d3246fb0 JD |
1937 | * Create a copyup osd request based on the information in the object |
1938 | * request supplied. A copyup request has two or three osd ops, a | |
1939 | * copyup method call, potentially a hint op, and a write or truncate | |
1940 | * or zero op. | |
0eefd470 AE |
1941 | */ |
1942 | static struct ceph_osd_request * | |
1943 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1944 | { | |
1945 | struct rbd_img_request *img_request; | |
1946 | struct ceph_snap_context *snapc; | |
1947 | struct rbd_device *rbd_dev; | |
1948 | struct ceph_osd_client *osdc; | |
1949 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1950 | int num_osd_ops = 3; |
0eefd470 AE |
1951 | |
1952 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1953 | img_request = obj_request->img_request; | |
1954 | rbd_assert(img_request); | |
d3246fb0 JD |
1955 | rbd_assert(img_request_write_test(img_request) || |
1956 | img_request_discard_test(img_request)); | |
0eefd470 | 1957 | |
d3246fb0 JD |
1958 | if (img_request_discard_test(img_request)) |
1959 | num_osd_ops = 2; | |
1960 | ||
1961 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
1962 | |
1963 | snapc = img_request->snapc; | |
1964 | rbd_dev = img_request->rbd_dev; | |
1965 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 JD |
1966 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
1967 | false, GFP_ATOMIC); | |
0eefd470 AE |
1968 | if (!osd_req) |
1969 | return NULL; /* ENOMEM */ | |
1970 | ||
1971 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1972 | osd_req->r_callback = rbd_osd_req_callback; | |
1973 | osd_req->r_priv = obj_request; | |
1974 | ||
3c972c95 ID |
1975 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1976 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 1977 | |
0eefd470 AE |
1978 | return osd_req; |
1979 | } | |
1980 | ||
1981 | ||
bf0d5f50 AE |
1982 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1983 | { | |
1984 | ceph_osdc_put_request(osd_req); | |
1985 | } | |
1986 | ||
1987 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1988 | ||
1989 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1990 | u64 offset, u64 length, | |
1991 | enum obj_request_type type) | |
1992 | { | |
1993 | struct rbd_obj_request *obj_request; | |
1994 | size_t size; | |
1995 | char *name; | |
1996 | ||
1997 | rbd_assert(obj_request_type_valid(type)); | |
1998 | ||
1999 | size = strlen(object_name) + 1; | |
f907ad55 AE |
2000 | name = kmalloc(size, GFP_KERNEL); |
2001 | if (!name) | |
bf0d5f50 AE |
2002 | return NULL; |
2003 | ||
868311b1 | 2004 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
2005 | if (!obj_request) { |
2006 | kfree(name); | |
2007 | return NULL; | |
2008 | } | |
2009 | ||
bf0d5f50 AE |
2010 | obj_request->object_name = memcpy(name, object_name, size); |
2011 | obj_request->offset = offset; | |
2012 | obj_request->length = length; | |
926f9b3f | 2013 | obj_request->flags = 0; |
bf0d5f50 AE |
2014 | obj_request->which = BAD_WHICH; |
2015 | obj_request->type = type; | |
2016 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2017 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2018 | kref_init(&obj_request->kref); |
2019 | ||
37206ee5 AE |
2020 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2021 | offset, length, (int)type, obj_request); | |
2022 | ||
bf0d5f50 AE |
2023 | return obj_request; |
2024 | } | |
2025 | ||
2026 | static void rbd_obj_request_destroy(struct kref *kref) | |
2027 | { | |
2028 | struct rbd_obj_request *obj_request; | |
2029 | ||
2030 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2031 | ||
37206ee5 AE |
2032 | dout("%s: obj %p\n", __func__, obj_request); |
2033 | ||
bf0d5f50 AE |
2034 | rbd_assert(obj_request->img_request == NULL); |
2035 | rbd_assert(obj_request->which == BAD_WHICH); | |
2036 | ||
2037 | if (obj_request->osd_req) | |
2038 | rbd_osd_req_destroy(obj_request->osd_req); | |
2039 | ||
2040 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2041 | switch (obj_request->type) { | |
9969ebc5 AE |
2042 | case OBJ_REQUEST_NODATA: |
2043 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2044 | case OBJ_REQUEST_BIO: |
2045 | if (obj_request->bio_list) | |
2046 | bio_chain_put(obj_request->bio_list); | |
2047 | break; | |
788e2df3 AE |
2048 | case OBJ_REQUEST_PAGES: |
2049 | if (obj_request->pages) | |
2050 | ceph_release_page_vector(obj_request->pages, | |
2051 | obj_request->page_count); | |
2052 | break; | |
bf0d5f50 AE |
2053 | } |
2054 | ||
f907ad55 | 2055 | kfree(obj_request->object_name); |
868311b1 AE |
2056 | obj_request->object_name = NULL; |
2057 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2058 | } |
2059 | ||
fb65d228 AE |
2060 | /* It's OK to call this for a device with no parent */ |
2061 | ||
2062 | static void rbd_spec_put(struct rbd_spec *spec); | |
2063 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2064 | { | |
2065 | rbd_dev_remove_parent(rbd_dev); | |
2066 | rbd_spec_put(rbd_dev->parent_spec); | |
2067 | rbd_dev->parent_spec = NULL; | |
2068 | rbd_dev->parent_overlap = 0; | |
2069 | } | |
2070 | ||
a2acd00e AE |
2071 | /* |
2072 | * Parent image reference counting is used to determine when an | |
2073 | * image's parent fields can be safely torn down--after there are no | |
2074 | * more in-flight requests to the parent image. When the last | |
2075 | * reference is dropped, cleaning them up is safe. | |
2076 | */ | |
2077 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2078 | { | |
2079 | int counter; | |
2080 | ||
2081 | if (!rbd_dev->parent_spec) | |
2082 | return; | |
2083 | ||
2084 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2085 | if (counter > 0) | |
2086 | return; | |
2087 | ||
2088 | /* Last reference; clean up parent data structures */ | |
2089 | ||
2090 | if (!counter) | |
2091 | rbd_dev_unparent(rbd_dev); | |
2092 | else | |
9584d508 | 2093 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2094 | } |
2095 | ||
2096 | /* | |
2097 | * If an image has a non-zero parent overlap, get a reference to its | |
2098 | * parent. | |
2099 | * | |
392a9dad AE |
2100 | * We must get the reference before checking for the overlap to |
2101 | * coordinate properly with zeroing the parent overlap in | |
2102 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
2103 | * drop it again if there is no overlap. | |
2104 | * | |
a2acd00e AE |
2105 | * Returns true if the rbd device has a parent with a non-zero |
2106 | * overlap and a reference for it was successfully taken, or | |
2107 | * false otherwise. | |
2108 | */ | |
2109 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2110 | { | |
2111 | int counter; | |
2112 | ||
2113 | if (!rbd_dev->parent_spec) | |
2114 | return false; | |
2115 | ||
2116 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2117 | if (counter > 0 && rbd_dev->parent_overlap) | |
2118 | return true; | |
2119 | ||
2120 | /* Image was flattened, but parent is not yet torn down */ | |
2121 | ||
2122 | if (counter < 0) | |
9584d508 | 2123 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e AE |
2124 | |
2125 | return false; | |
2126 | } | |
2127 | ||
bf0d5f50 AE |
2128 | /* |
2129 | * Caller is responsible for filling in the list of object requests | |
2130 | * that comprises the image request, and the Linux request pointer | |
2131 | * (if there is one). | |
2132 | */ | |
cc344fa1 AE |
2133 | static struct rbd_img_request *rbd_img_request_create( |
2134 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2135 | u64 offset, u64 length, |
6d2940c8 | 2136 | enum obj_operation_type op_type, |
4e752f0a | 2137 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2138 | { |
2139 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2140 | |
7a716aac | 2141 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2142 | if (!img_request) |
2143 | return NULL; | |
2144 | ||
bf0d5f50 AE |
2145 | img_request->rq = NULL; |
2146 | img_request->rbd_dev = rbd_dev; | |
2147 | img_request->offset = offset; | |
2148 | img_request->length = length; | |
0c425248 | 2149 | img_request->flags = 0; |
90e98c52 GZ |
2150 | if (op_type == OBJ_OP_DISCARD) { |
2151 | img_request_discard_set(img_request); | |
2152 | img_request->snapc = snapc; | |
2153 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2154 | img_request_write_set(img_request); |
4e752f0a | 2155 | img_request->snapc = snapc; |
0c425248 | 2156 | } else { |
bf0d5f50 | 2157 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2158 | } |
a2acd00e | 2159 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2160 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2161 | spin_lock_init(&img_request->completion_lock); |
2162 | img_request->next_completion = 0; | |
2163 | img_request->callback = NULL; | |
a5a337d4 | 2164 | img_request->result = 0; |
bf0d5f50 AE |
2165 | img_request->obj_request_count = 0; |
2166 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2167 | kref_init(&img_request->kref); | |
2168 | ||
37206ee5 | 2169 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2170 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2171 | |
bf0d5f50 AE |
2172 | return img_request; |
2173 | } | |
2174 | ||
2175 | static void rbd_img_request_destroy(struct kref *kref) | |
2176 | { | |
2177 | struct rbd_img_request *img_request; | |
2178 | struct rbd_obj_request *obj_request; | |
2179 | struct rbd_obj_request *next_obj_request; | |
2180 | ||
2181 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2182 | ||
37206ee5 AE |
2183 | dout("%s: img %p\n", __func__, img_request); |
2184 | ||
bf0d5f50 AE |
2185 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2186 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2187 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2188 | |
a2acd00e AE |
2189 | if (img_request_layered_test(img_request)) { |
2190 | img_request_layered_clear(img_request); | |
2191 | rbd_dev_parent_put(img_request->rbd_dev); | |
2192 | } | |
2193 | ||
bef95455 JD |
2194 | if (img_request_write_test(img_request) || |
2195 | img_request_discard_test(img_request)) | |
812164f8 | 2196 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2197 | |
1c2a9dfe | 2198 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2199 | } |
2200 | ||
e93f3152 AE |
2201 | static struct rbd_img_request *rbd_parent_request_create( |
2202 | struct rbd_obj_request *obj_request, | |
2203 | u64 img_offset, u64 length) | |
2204 | { | |
2205 | struct rbd_img_request *parent_request; | |
2206 | struct rbd_device *rbd_dev; | |
2207 | ||
2208 | rbd_assert(obj_request->img_request); | |
2209 | rbd_dev = obj_request->img_request->rbd_dev; | |
2210 | ||
4e752f0a | 2211 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2212 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2213 | if (!parent_request) |
2214 | return NULL; | |
2215 | ||
2216 | img_request_child_set(parent_request); | |
2217 | rbd_obj_request_get(obj_request); | |
2218 | parent_request->obj_request = obj_request; | |
2219 | ||
2220 | return parent_request; | |
2221 | } | |
2222 | ||
2223 | static void rbd_parent_request_destroy(struct kref *kref) | |
2224 | { | |
2225 | struct rbd_img_request *parent_request; | |
2226 | struct rbd_obj_request *orig_request; | |
2227 | ||
2228 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2229 | orig_request = parent_request->obj_request; | |
2230 | ||
2231 | parent_request->obj_request = NULL; | |
2232 | rbd_obj_request_put(orig_request); | |
2233 | img_request_child_clear(parent_request); | |
2234 | ||
2235 | rbd_img_request_destroy(kref); | |
2236 | } | |
2237 | ||
1217857f AE |
2238 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2239 | { | |
6365d33a | 2240 | struct rbd_img_request *img_request; |
1217857f AE |
2241 | unsigned int xferred; |
2242 | int result; | |
8b3e1a56 | 2243 | bool more; |
1217857f | 2244 | |
6365d33a AE |
2245 | rbd_assert(obj_request_img_data_test(obj_request)); |
2246 | img_request = obj_request->img_request; | |
2247 | ||
1217857f AE |
2248 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2249 | xferred = (unsigned int)obj_request->xferred; | |
2250 | result = obj_request->result; | |
2251 | if (result) { | |
2252 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2253 | enum obj_operation_type op_type; |
2254 | ||
90e98c52 GZ |
2255 | if (img_request_discard_test(img_request)) |
2256 | op_type = OBJ_OP_DISCARD; | |
2257 | else if (img_request_write_test(img_request)) | |
2258 | op_type = OBJ_OP_WRITE; | |
2259 | else | |
2260 | op_type = OBJ_OP_READ; | |
1217857f | 2261 | |
9584d508 | 2262 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2263 | obj_op_name(op_type), obj_request->length, |
2264 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2265 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2266 | result, xferred); |
2267 | if (!img_request->result) | |
2268 | img_request->result = result; | |
2269 | } | |
2270 | ||
f1a4739f AE |
2271 | /* Image object requests don't own their page array */ |
2272 | ||
2273 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2274 | obj_request->pages = NULL; | |
2275 | obj_request->page_count = 0; | |
2276 | } | |
2277 | ||
8b3e1a56 AE |
2278 | if (img_request_child_test(img_request)) { |
2279 | rbd_assert(img_request->obj_request != NULL); | |
2280 | more = obj_request->which < img_request->obj_request_count - 1; | |
2281 | } else { | |
2282 | rbd_assert(img_request->rq != NULL); | |
2283 | more = blk_end_request(img_request->rq, result, xferred); | |
2284 | } | |
2285 | ||
2286 | return more; | |
1217857f AE |
2287 | } |
2288 | ||
2169238d AE |
2289 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2290 | { | |
2291 | struct rbd_img_request *img_request; | |
2292 | u32 which = obj_request->which; | |
2293 | bool more = true; | |
2294 | ||
6365d33a | 2295 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2296 | img_request = obj_request->img_request; |
2297 | ||
2298 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2299 | rbd_assert(img_request != NULL); | |
2169238d AE |
2300 | rbd_assert(img_request->obj_request_count > 0); |
2301 | rbd_assert(which != BAD_WHICH); | |
2302 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2303 | |
2304 | spin_lock_irq(&img_request->completion_lock); | |
2305 | if (which != img_request->next_completion) | |
2306 | goto out; | |
2307 | ||
2308 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2309 | rbd_assert(more); |
2310 | rbd_assert(which < img_request->obj_request_count); | |
2311 | ||
2312 | if (!obj_request_done_test(obj_request)) | |
2313 | break; | |
1217857f | 2314 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2315 | which++; |
2316 | } | |
2317 | ||
2318 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2319 | img_request->next_completion = which; | |
2320 | out: | |
2321 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2322 | rbd_img_request_put(img_request); |
2169238d AE |
2323 | |
2324 | if (!more) | |
2325 | rbd_img_request_complete(img_request); | |
2326 | } | |
2327 | ||
3b434a2a JD |
2328 | /* |
2329 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2330 | * them for submission. num_ops is the current number of | |
2331 | * osd operations already to the object request. | |
2332 | */ | |
2333 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2334 | struct ceph_osd_request *osd_request, | |
2335 | enum obj_operation_type op_type, | |
2336 | unsigned int num_ops) | |
2337 | { | |
2338 | struct rbd_img_request *img_request = obj_request->img_request; | |
2339 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2340 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2341 | u64 offset = obj_request->offset; | |
2342 | u64 length = obj_request->length; | |
2343 | u64 img_end; | |
2344 | u16 opcode; | |
2345 | ||
2346 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2347 | if (!offset && length == object_size && |
2348 | (!img_request_layered_test(img_request) || | |
2349 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2350 | opcode = CEPH_OSD_OP_DELETE; |
2351 | } else if ((offset + length == object_size)) { | |
2352 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2353 | } else { | |
2354 | down_read(&rbd_dev->header_rwsem); | |
2355 | img_end = rbd_dev->header.image_size; | |
2356 | up_read(&rbd_dev->header_rwsem); | |
2357 | ||
2358 | if (obj_request->img_offset + length == img_end) | |
2359 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2360 | else | |
2361 | opcode = CEPH_OSD_OP_ZERO; | |
2362 | } | |
2363 | } else if (op_type == OBJ_OP_WRITE) { | |
2364 | opcode = CEPH_OSD_OP_WRITE; | |
2365 | osd_req_op_alloc_hint_init(osd_request, num_ops, | |
2366 | object_size, object_size); | |
2367 | num_ops++; | |
2368 | } else { | |
2369 | opcode = CEPH_OSD_OP_READ; | |
2370 | } | |
2371 | ||
2372 | osd_req_op_extent_init(osd_request, num_ops, opcode, offset, length, | |
2373 | 0, 0); | |
2374 | if (obj_request->type == OBJ_REQUEST_BIO) | |
2375 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2376 | obj_request->bio_list, length); | |
2377 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2378 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2379 | obj_request->pages, length, | |
2380 | offset & ~PAGE_MASK, false, false); | |
2381 | ||
2382 | /* Discards are also writes */ | |
2383 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2384 | rbd_osd_req_format_write(obj_request); | |
2385 | else | |
2386 | rbd_osd_req_format_read(obj_request); | |
2387 | } | |
2388 | ||
f1a4739f AE |
2389 | /* |
2390 | * Split up an image request into one or more object requests, each | |
2391 | * to a different object. The "type" parameter indicates whether | |
2392 | * "data_desc" is the pointer to the head of a list of bio | |
2393 | * structures, or the base of a page array. In either case this | |
2394 | * function assumes data_desc describes memory sufficient to hold | |
2395 | * all data described by the image request. | |
2396 | */ | |
2397 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2398 | enum obj_request_type type, | |
2399 | void *data_desc) | |
bf0d5f50 AE |
2400 | { |
2401 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2402 | struct rbd_obj_request *obj_request = NULL; | |
2403 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2404 | struct bio *bio_list = NULL; |
f1a4739f | 2405 | unsigned int bio_offset = 0; |
a158073c | 2406 | struct page **pages = NULL; |
6d2940c8 | 2407 | enum obj_operation_type op_type; |
7da22d29 | 2408 | u64 img_offset; |
bf0d5f50 | 2409 | u64 resid; |
bf0d5f50 | 2410 | |
f1a4739f AE |
2411 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2412 | (int)type, data_desc); | |
37206ee5 | 2413 | |
7da22d29 | 2414 | img_offset = img_request->offset; |
bf0d5f50 | 2415 | resid = img_request->length; |
4dda41d3 | 2416 | rbd_assert(resid > 0); |
3b434a2a | 2417 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2418 | |
2419 | if (type == OBJ_REQUEST_BIO) { | |
2420 | bio_list = data_desc; | |
4f024f37 KO |
2421 | rbd_assert(img_offset == |
2422 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2423 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2424 | pages = data_desc; |
2425 | } | |
2426 | ||
bf0d5f50 | 2427 | while (resid) { |
2fa12320 | 2428 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2429 | const char *object_name; |
bf0d5f50 AE |
2430 | u64 offset; |
2431 | u64 length; | |
2432 | ||
7da22d29 | 2433 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2434 | if (!object_name) |
2435 | goto out_unwind; | |
7da22d29 AE |
2436 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2437 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2438 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2439 | offset, length, type); |
78c2a44a AE |
2440 | /* object request has its own copy of the object name */ |
2441 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2442 | if (!obj_request) |
2443 | goto out_unwind; | |
62054da6 | 2444 | |
03507db6 JD |
2445 | /* |
2446 | * set obj_request->img_request before creating the | |
2447 | * osd_request so that it gets the right snapc | |
2448 | */ | |
2449 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2450 | |
f1a4739f AE |
2451 | if (type == OBJ_REQUEST_BIO) { |
2452 | unsigned int clone_size; | |
2453 | ||
2454 | rbd_assert(length <= (u64)UINT_MAX); | |
2455 | clone_size = (unsigned int)length; | |
2456 | obj_request->bio_list = | |
2457 | bio_chain_clone_range(&bio_list, | |
2458 | &bio_offset, | |
2459 | clone_size, | |
2460 | GFP_ATOMIC); | |
2461 | if (!obj_request->bio_list) | |
62054da6 | 2462 | goto out_unwind; |
90e98c52 | 2463 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2464 | unsigned int page_count; |
2465 | ||
2466 | obj_request->pages = pages; | |
2467 | page_count = (u32)calc_pages_for(offset, length); | |
2468 | obj_request->page_count = page_count; | |
2469 | if ((offset + length) & ~PAGE_MASK) | |
2470 | page_count--; /* more on last page */ | |
2471 | pages += page_count; | |
2472 | } | |
bf0d5f50 | 2473 | |
6d2940c8 GZ |
2474 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2475 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2476 | obj_request); | |
2fa12320 | 2477 | if (!osd_req) |
62054da6 | 2478 | goto out_unwind; |
3b434a2a | 2479 | |
2fa12320 | 2480 | obj_request->osd_req = osd_req; |
2169238d | 2481 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2482 | obj_request->img_offset = img_offset; |
9d4df01f | 2483 | |
3b434a2a | 2484 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2485 | |
3b434a2a | 2486 | rbd_img_request_get(img_request); |
bf0d5f50 | 2487 | |
7da22d29 | 2488 | img_offset += length; |
bf0d5f50 AE |
2489 | resid -= length; |
2490 | } | |
2491 | ||
2492 | return 0; | |
2493 | ||
bf0d5f50 AE |
2494 | out_unwind: |
2495 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2496 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2497 | |
2498 | return -ENOMEM; | |
2499 | } | |
2500 | ||
0eefd470 AE |
2501 | static void |
2502 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2503 | { | |
2504 | struct rbd_img_request *img_request; | |
2505 | struct rbd_device *rbd_dev; | |
ebda6408 | 2506 | struct page **pages; |
0eefd470 AE |
2507 | u32 page_count; |
2508 | ||
d3246fb0 JD |
2509 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2510 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2511 | rbd_assert(obj_request_img_data_test(obj_request)); |
2512 | img_request = obj_request->img_request; | |
2513 | rbd_assert(img_request); | |
2514 | ||
2515 | rbd_dev = img_request->rbd_dev; | |
2516 | rbd_assert(rbd_dev); | |
0eefd470 | 2517 | |
ebda6408 AE |
2518 | pages = obj_request->copyup_pages; |
2519 | rbd_assert(pages != NULL); | |
0eefd470 | 2520 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2521 | page_count = obj_request->copyup_page_count; |
2522 | rbd_assert(page_count); | |
2523 | obj_request->copyup_page_count = 0; | |
2524 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2525 | |
2526 | /* | |
2527 | * We want the transfer count to reflect the size of the | |
2528 | * original write request. There is no such thing as a | |
2529 | * successful short write, so if the request was successful | |
2530 | * we can just set it to the originally-requested length. | |
2531 | */ | |
2532 | if (!obj_request->result) | |
2533 | obj_request->xferred = obj_request->length; | |
2534 | ||
2535 | /* Finish up with the normal image object callback */ | |
2536 | ||
2537 | rbd_img_obj_callback(obj_request); | |
2538 | } | |
2539 | ||
3d7efd18 AE |
2540 | static void |
2541 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2542 | { | |
2543 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2544 | struct ceph_osd_request *osd_req; |
2545 | struct ceph_osd_client *osdc; | |
2546 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2547 | struct page **pages; |
d3246fb0 | 2548 | enum obj_operation_type op_type; |
ebda6408 | 2549 | u32 page_count; |
bbea1c1a | 2550 | int img_result; |
ebda6408 | 2551 | u64 parent_length; |
3d7efd18 AE |
2552 | |
2553 | rbd_assert(img_request_child_test(img_request)); | |
2554 | ||
2555 | /* First get what we need from the image request */ | |
2556 | ||
2557 | pages = img_request->copyup_pages; | |
2558 | rbd_assert(pages != NULL); | |
2559 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2560 | page_count = img_request->copyup_page_count; |
2561 | rbd_assert(page_count); | |
2562 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2563 | |
2564 | orig_request = img_request->obj_request; | |
2565 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2566 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2567 | img_result = img_request->result; |
ebda6408 AE |
2568 | parent_length = img_request->length; |
2569 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2570 | rbd_img_request_put(img_request); |
3d7efd18 | 2571 | |
91c6febb AE |
2572 | rbd_assert(orig_request->img_request); |
2573 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2574 | rbd_assert(rbd_dev); |
0eefd470 | 2575 | |
bbea1c1a AE |
2576 | /* |
2577 | * If the overlap has become 0 (most likely because the | |
2578 | * image has been flattened) we need to free the pages | |
2579 | * and re-submit the original write request. | |
2580 | */ | |
2581 | if (!rbd_dev->parent_overlap) { | |
2582 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2583 | |
bbea1c1a AE |
2584 | ceph_release_page_vector(pages, page_count); |
2585 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2586 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2587 | if (!img_result) | |
2588 | return; | |
2589 | } | |
0eefd470 | 2590 | |
bbea1c1a | 2591 | if (img_result) |
0eefd470 | 2592 | goto out_err; |
0eefd470 | 2593 | |
8785b1d4 AE |
2594 | /* |
2595 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2596 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2597 | * request. Allocate the new copyup osd request for the |
2598 | * original request, and release the old one. | |
2599 | */ | |
bbea1c1a | 2600 | img_result = -ENOMEM; |
0eefd470 AE |
2601 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2602 | if (!osd_req) | |
2603 | goto out_err; | |
8785b1d4 | 2604 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2605 | orig_request->osd_req = osd_req; |
2606 | orig_request->copyup_pages = pages; | |
ebda6408 | 2607 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2608 | |
0eefd470 | 2609 | /* Initialize the copyup op */ |
3d7efd18 | 2610 | |
0eefd470 | 2611 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2612 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2613 | false, false); |
3d7efd18 | 2614 | |
d3246fb0 | 2615 | /* Add the other op(s) */ |
0eefd470 | 2616 | |
d3246fb0 JD |
2617 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2618 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2619 | |
2620 | /* All set, send it off. */ | |
2621 | ||
2622 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2623 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2624 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2625 | if (!img_result) | |
0eefd470 AE |
2626 | return; |
2627 | out_err: | |
2628 | /* Record the error code and complete the request */ | |
2629 | ||
bbea1c1a | 2630 | orig_request->result = img_result; |
0eefd470 AE |
2631 | orig_request->xferred = 0; |
2632 | obj_request_done_set(orig_request); | |
2633 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2634 | } |
2635 | ||
2636 | /* | |
2637 | * Read from the parent image the range of data that covers the | |
2638 | * entire target of the given object request. This is used for | |
2639 | * satisfying a layered image write request when the target of an | |
2640 | * object request from the image request does not exist. | |
2641 | * | |
2642 | * A page array big enough to hold the returned data is allocated | |
2643 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2644 | * When the read completes, this page array will be transferred to | |
2645 | * the original object request for the copyup operation. | |
2646 | * | |
2647 | * If an error occurs, record it as the result of the original | |
2648 | * object request and mark it done so it gets completed. | |
2649 | */ | |
2650 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2651 | { | |
2652 | struct rbd_img_request *img_request = NULL; | |
2653 | struct rbd_img_request *parent_request = NULL; | |
2654 | struct rbd_device *rbd_dev; | |
2655 | u64 img_offset; | |
2656 | u64 length; | |
2657 | struct page **pages = NULL; | |
2658 | u32 page_count; | |
2659 | int result; | |
2660 | ||
2661 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2662 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2663 | |
2664 | img_request = obj_request->img_request; | |
2665 | rbd_assert(img_request != NULL); | |
2666 | rbd_dev = img_request->rbd_dev; | |
2667 | rbd_assert(rbd_dev->parent != NULL); | |
2668 | ||
2669 | /* | |
2670 | * Determine the byte range covered by the object in the | |
2671 | * child image to which the original request was to be sent. | |
2672 | */ | |
2673 | img_offset = obj_request->img_offset - obj_request->offset; | |
2674 | length = (u64)1 << rbd_dev->header.obj_order; | |
2675 | ||
a9e8ba2c AE |
2676 | /* |
2677 | * There is no defined parent data beyond the parent | |
2678 | * overlap, so limit what we read at that boundary if | |
2679 | * necessary. | |
2680 | */ | |
2681 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2682 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2683 | length = rbd_dev->parent_overlap - img_offset; | |
2684 | } | |
2685 | ||
3d7efd18 AE |
2686 | /* |
2687 | * Allocate a page array big enough to receive the data read | |
2688 | * from the parent. | |
2689 | */ | |
2690 | page_count = (u32)calc_pages_for(0, length); | |
2691 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2692 | if (IS_ERR(pages)) { | |
2693 | result = PTR_ERR(pages); | |
2694 | pages = NULL; | |
2695 | goto out_err; | |
2696 | } | |
2697 | ||
2698 | result = -ENOMEM; | |
e93f3152 AE |
2699 | parent_request = rbd_parent_request_create(obj_request, |
2700 | img_offset, length); | |
3d7efd18 AE |
2701 | if (!parent_request) |
2702 | goto out_err; | |
3d7efd18 AE |
2703 | |
2704 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2705 | if (result) | |
2706 | goto out_err; | |
2707 | parent_request->copyup_pages = pages; | |
ebda6408 | 2708 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2709 | |
2710 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2711 | result = rbd_img_request_submit(parent_request); | |
2712 | if (!result) | |
2713 | return 0; | |
2714 | ||
2715 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2716 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2717 | parent_request->obj_request = NULL; |
2718 | rbd_obj_request_put(obj_request); | |
2719 | out_err: | |
2720 | if (pages) | |
2721 | ceph_release_page_vector(pages, page_count); | |
2722 | if (parent_request) | |
2723 | rbd_img_request_put(parent_request); | |
2724 | obj_request->result = result; | |
2725 | obj_request->xferred = 0; | |
2726 | obj_request_done_set(obj_request); | |
2727 | ||
2728 | return result; | |
2729 | } | |
2730 | ||
c5b5ef6c AE |
2731 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2732 | { | |
c5b5ef6c | 2733 | struct rbd_obj_request *orig_request; |
638f5abe | 2734 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2735 | int result; |
2736 | ||
2737 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2738 | ||
2739 | /* | |
2740 | * All we need from the object request is the original | |
2741 | * request and the result of the STAT op. Grab those, then | |
2742 | * we're done with the request. | |
2743 | */ | |
2744 | orig_request = obj_request->obj_request; | |
2745 | obj_request->obj_request = NULL; | |
912c317d | 2746 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2747 | rbd_assert(orig_request); |
2748 | rbd_assert(orig_request->img_request); | |
2749 | ||
2750 | result = obj_request->result; | |
2751 | obj_request->result = 0; | |
2752 | ||
2753 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2754 | obj_request, orig_request, result, | |
2755 | obj_request->xferred, obj_request->length); | |
2756 | rbd_obj_request_put(obj_request); | |
2757 | ||
638f5abe AE |
2758 | /* |
2759 | * If the overlap has become 0 (most likely because the | |
2760 | * image has been flattened) we need to free the pages | |
2761 | * and re-submit the original write request. | |
2762 | */ | |
2763 | rbd_dev = orig_request->img_request->rbd_dev; | |
2764 | if (!rbd_dev->parent_overlap) { | |
2765 | struct ceph_osd_client *osdc; | |
2766 | ||
638f5abe AE |
2767 | osdc = &rbd_dev->rbd_client->client->osdc; |
2768 | result = rbd_obj_request_submit(osdc, orig_request); | |
2769 | if (!result) | |
2770 | return; | |
2771 | } | |
c5b5ef6c AE |
2772 | |
2773 | /* | |
2774 | * Our only purpose here is to determine whether the object | |
2775 | * exists, and we don't want to treat the non-existence as | |
2776 | * an error. If something else comes back, transfer the | |
2777 | * error to the original request and complete it now. | |
2778 | */ | |
2779 | if (!result) { | |
2780 | obj_request_existence_set(orig_request, true); | |
2781 | } else if (result == -ENOENT) { | |
2782 | obj_request_existence_set(orig_request, false); | |
2783 | } else if (result) { | |
2784 | orig_request->result = result; | |
3d7efd18 | 2785 | goto out; |
c5b5ef6c AE |
2786 | } |
2787 | ||
2788 | /* | |
2789 | * Resubmit the original request now that we have recorded | |
2790 | * whether the target object exists. | |
2791 | */ | |
b454e36d | 2792 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2793 | out: |
c5b5ef6c AE |
2794 | if (orig_request->result) |
2795 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2796 | } |
2797 | ||
2798 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2799 | { | |
2800 | struct rbd_obj_request *stat_request; | |
2801 | struct rbd_device *rbd_dev; | |
2802 | struct ceph_osd_client *osdc; | |
2803 | struct page **pages = NULL; | |
2804 | u32 page_count; | |
2805 | size_t size; | |
2806 | int ret; | |
2807 | ||
2808 | /* | |
2809 | * The response data for a STAT call consists of: | |
2810 | * le64 length; | |
2811 | * struct { | |
2812 | * le32 tv_sec; | |
2813 | * le32 tv_nsec; | |
2814 | * } mtime; | |
2815 | */ | |
2816 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2817 | page_count = (u32)calc_pages_for(0, size); | |
2818 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2819 | if (IS_ERR(pages)) | |
2820 | return PTR_ERR(pages); | |
2821 | ||
2822 | ret = -ENOMEM; | |
2823 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2824 | OBJ_REQUEST_PAGES); | |
2825 | if (!stat_request) | |
2826 | goto out; | |
2827 | ||
2828 | rbd_obj_request_get(obj_request); | |
2829 | stat_request->obj_request = obj_request; | |
2830 | stat_request->pages = pages; | |
2831 | stat_request->page_count = page_count; | |
2832 | ||
2833 | rbd_assert(obj_request->img_request); | |
2834 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2835 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2836 | stat_request); |
c5b5ef6c AE |
2837 | if (!stat_request->osd_req) |
2838 | goto out; | |
2839 | stat_request->callback = rbd_img_obj_exists_callback; | |
2840 | ||
2841 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2842 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2843 | false, false); | |
9d4df01f | 2844 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2845 | |
2846 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2847 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2848 | out: | |
2849 | if (ret) | |
2850 | rbd_obj_request_put(obj_request); | |
2851 | ||
2852 | return ret; | |
2853 | } | |
2854 | ||
70d045f6 | 2855 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2856 | { |
2857 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2858 | struct rbd_device *rbd_dev; |
b454e36d AE |
2859 | |
2860 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2861 | ||
2862 | img_request = obj_request->img_request; | |
2863 | rbd_assert(img_request); | |
a9e8ba2c | 2864 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2865 | |
70d045f6 | 2866 | /* Reads */ |
1c220881 JD |
2867 | if (!img_request_write_test(img_request) && |
2868 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2869 | return true; |
2870 | ||
2871 | /* Non-layered writes */ | |
2872 | if (!img_request_layered_test(img_request)) | |
2873 | return true; | |
2874 | ||
b454e36d | 2875 | /* |
70d045f6 ID |
2876 | * Layered writes outside of the parent overlap range don't |
2877 | * share any data with the parent. | |
b454e36d | 2878 | */ |
70d045f6 ID |
2879 | if (!obj_request_overlaps_parent(obj_request)) |
2880 | return true; | |
b454e36d | 2881 | |
c622d226 GZ |
2882 | /* |
2883 | * Entire-object layered writes - we will overwrite whatever | |
2884 | * parent data there is anyway. | |
2885 | */ | |
2886 | if (!obj_request->offset && | |
2887 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2888 | return true; | |
2889 | ||
70d045f6 ID |
2890 | /* |
2891 | * If the object is known to already exist, its parent data has | |
2892 | * already been copied. | |
2893 | */ | |
2894 | if (obj_request_known_test(obj_request) && | |
2895 | obj_request_exists_test(obj_request)) | |
2896 | return true; | |
2897 | ||
2898 | return false; | |
2899 | } | |
2900 | ||
2901 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2902 | { | |
2903 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2904 | struct rbd_device *rbd_dev; |
2905 | struct ceph_osd_client *osdc; | |
2906 | ||
2907 | rbd_dev = obj_request->img_request->rbd_dev; | |
2908 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2909 | ||
2910 | return rbd_obj_request_submit(osdc, obj_request); | |
2911 | } | |
2912 | ||
2913 | /* | |
3d7efd18 AE |
2914 | * It's a layered write. The target object might exist but |
2915 | * we may not know that yet. If we know it doesn't exist, | |
2916 | * start by reading the data for the full target object from | |
2917 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2918 | */ |
70d045f6 | 2919 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2920 | return rbd_img_obj_parent_read_full(obj_request); |
2921 | ||
2922 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2923 | |
2924 | return rbd_img_obj_exists_submit(obj_request); | |
2925 | } | |
2926 | ||
bf0d5f50 AE |
2927 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2928 | { | |
bf0d5f50 | 2929 | struct rbd_obj_request *obj_request; |
46faeed4 | 2930 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2931 | |
37206ee5 | 2932 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2933 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2934 | int ret; |
2935 | ||
b454e36d | 2936 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2937 | if (ret) |
2938 | return ret; | |
bf0d5f50 AE |
2939 | } |
2940 | ||
2941 | return 0; | |
2942 | } | |
8b3e1a56 AE |
2943 | |
2944 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2945 | { | |
2946 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2947 | struct rbd_device *rbd_dev; |
2948 | u64 obj_end; | |
02c74fba AE |
2949 | u64 img_xferred; |
2950 | int img_result; | |
8b3e1a56 AE |
2951 | |
2952 | rbd_assert(img_request_child_test(img_request)); | |
2953 | ||
02c74fba AE |
2954 | /* First get what we need from the image request and release it */ |
2955 | ||
8b3e1a56 | 2956 | obj_request = img_request->obj_request; |
02c74fba AE |
2957 | img_xferred = img_request->xferred; |
2958 | img_result = img_request->result; | |
2959 | rbd_img_request_put(img_request); | |
2960 | ||
2961 | /* | |
2962 | * If the overlap has become 0 (most likely because the | |
2963 | * image has been flattened) we need to re-submit the | |
2964 | * original request. | |
2965 | */ | |
a9e8ba2c AE |
2966 | rbd_assert(obj_request); |
2967 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2968 | rbd_dev = obj_request->img_request->rbd_dev; |
2969 | if (!rbd_dev->parent_overlap) { | |
2970 | struct ceph_osd_client *osdc; | |
2971 | ||
2972 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2973 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2974 | if (!img_result) | |
2975 | return; | |
2976 | } | |
a9e8ba2c | 2977 | |
02c74fba | 2978 | obj_request->result = img_result; |
a9e8ba2c AE |
2979 | if (obj_request->result) |
2980 | goto out; | |
2981 | ||
2982 | /* | |
2983 | * We need to zero anything beyond the parent overlap | |
2984 | * boundary. Since rbd_img_obj_request_read_callback() | |
2985 | * will zero anything beyond the end of a short read, an | |
2986 | * easy way to do this is to pretend the data from the | |
2987 | * parent came up short--ending at the overlap boundary. | |
2988 | */ | |
2989 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2990 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2991 | if (obj_end > rbd_dev->parent_overlap) { |
2992 | u64 xferred = 0; | |
2993 | ||
2994 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2995 | xferred = rbd_dev->parent_overlap - | |
2996 | obj_request->img_offset; | |
8b3e1a56 | 2997 | |
02c74fba | 2998 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2999 | } else { |
02c74fba | 3000 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3001 | } |
3002 | out: | |
8b3e1a56 AE |
3003 | rbd_img_obj_request_read_callback(obj_request); |
3004 | rbd_obj_request_complete(obj_request); | |
3005 | } | |
3006 | ||
3007 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3008 | { | |
8b3e1a56 AE |
3009 | struct rbd_img_request *img_request; |
3010 | int result; | |
3011 | ||
3012 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3013 | rbd_assert(obj_request->img_request != NULL); | |
3014 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3015 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3016 | |
8b3e1a56 | 3017 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3018 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3019 | obj_request->img_offset, |
e93f3152 | 3020 | obj_request->length); |
8b3e1a56 AE |
3021 | result = -ENOMEM; |
3022 | if (!img_request) | |
3023 | goto out_err; | |
3024 | ||
5b2ab72d AE |
3025 | if (obj_request->type == OBJ_REQUEST_BIO) |
3026 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3027 | obj_request->bio_list); | |
3028 | else | |
3029 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3030 | obj_request->pages); | |
8b3e1a56 AE |
3031 | if (result) |
3032 | goto out_err; | |
3033 | ||
3034 | img_request->callback = rbd_img_parent_read_callback; | |
3035 | result = rbd_img_request_submit(img_request); | |
3036 | if (result) | |
3037 | goto out_err; | |
3038 | ||
3039 | return; | |
3040 | out_err: | |
3041 | if (img_request) | |
3042 | rbd_img_request_put(img_request); | |
3043 | obj_request->result = result; | |
3044 | obj_request->xferred = 0; | |
3045 | obj_request_done_set(obj_request); | |
3046 | } | |
bf0d5f50 | 3047 | |
20e0af67 | 3048 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3049 | { |
3050 | struct rbd_obj_request *obj_request; | |
2169238d | 3051 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3052 | int ret; |
3053 | ||
3054 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3055 | OBJ_REQUEST_NODATA); | |
3056 | if (!obj_request) | |
3057 | return -ENOMEM; | |
3058 | ||
3059 | ret = -ENOMEM; | |
6d2940c8 | 3060 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3061 | obj_request); |
b8d70035 AE |
3062 | if (!obj_request->osd_req) |
3063 | goto out; | |
3064 | ||
c99d2d4a | 3065 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3066 | notify_id, 0, 0); |
9d4df01f | 3067 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3068 | |
b8d70035 | 3069 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3070 | if (ret) |
20e0af67 JD |
3071 | goto out; |
3072 | ret = rbd_obj_request_wait(obj_request); | |
3073 | out: | |
3074 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3075 | |
3076 | return ret; | |
3077 | } | |
3078 | ||
3079 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3080 | { | |
3081 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3082 | int ret; |
b8d70035 AE |
3083 | |
3084 | if (!rbd_dev) | |
3085 | return; | |
3086 | ||
37206ee5 | 3087 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3088 | rbd_dev->header_name, (unsigned long long)notify_id, |
3089 | (unsigned int)opcode); | |
52bb1f9b ID |
3090 | |
3091 | /* | |
3092 | * Until adequate refresh error handling is in place, there is | |
3093 | * not much we can do here, except warn. | |
3094 | * | |
3095 | * See http://tracker.ceph.com/issues/5040 | |
3096 | */ | |
e627db08 AE |
3097 | ret = rbd_dev_refresh(rbd_dev); |
3098 | if (ret) | |
9584d508 | 3099 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3100 | |
52bb1f9b ID |
3101 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3102 | if (ret) | |
9584d508 | 3103 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3104 | } |
3105 | ||
bb040aa0 ID |
3106 | /* |
3107 | * Send a (un)watch request and wait for the ack. Return a request | |
3108 | * with a ref held on success or error. | |
3109 | */ | |
3110 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3111 | struct rbd_device *rbd_dev, | |
3112 | bool watch) | |
3113 | { | |
3114 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3115 | struct rbd_obj_request *obj_request; | |
3116 | int ret; | |
3117 | ||
3118 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3119 | OBJ_REQUEST_NODATA); | |
3120 | if (!obj_request) | |
3121 | return ERR_PTR(-ENOMEM); | |
3122 | ||
6d2940c8 | 3123 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3124 | obj_request); |
3125 | if (!obj_request->osd_req) { | |
3126 | ret = -ENOMEM; | |
3127 | goto out; | |
3128 | } | |
3129 | ||
3130 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3131 | rbd_dev->watch_event->cookie, 0, watch); | |
3132 | rbd_osd_req_format_write(obj_request); | |
3133 | ||
3134 | if (watch) | |
3135 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3136 | ||
3137 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3138 | if (ret) | |
3139 | goto out; | |
3140 | ||
3141 | ret = rbd_obj_request_wait(obj_request); | |
3142 | if (ret) | |
3143 | goto out; | |
3144 | ||
3145 | ret = obj_request->result; | |
3146 | if (ret) { | |
3147 | if (watch) | |
3148 | rbd_obj_request_end(obj_request); | |
3149 | goto out; | |
3150 | } | |
3151 | ||
3152 | return obj_request; | |
3153 | ||
3154 | out: | |
3155 | rbd_obj_request_put(obj_request); | |
3156 | return ERR_PTR(ret); | |
3157 | } | |
3158 | ||
9969ebc5 | 3159 | /* |
b30a01f2 | 3160 | * Initiate a watch request, synchronously. |
9969ebc5 | 3161 | */ |
b30a01f2 | 3162 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3163 | { |
3164 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3165 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3166 | int ret; |
3167 | ||
b30a01f2 ID |
3168 | rbd_assert(!rbd_dev->watch_event); |
3169 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3170 | |
b30a01f2 ID |
3171 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3172 | &rbd_dev->watch_event); | |
3173 | if (ret < 0) | |
3174 | return ret; | |
3175 | ||
76756a51 ID |
3176 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3177 | if (IS_ERR(obj_request)) { | |
3178 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3179 | rbd_dev->watch_event = NULL; | |
3180 | return PTR_ERR(obj_request); | |
b30a01f2 | 3181 | } |
9969ebc5 | 3182 | |
8eb87565 AE |
3183 | /* |
3184 | * A watch request is set to linger, so the underlying osd | |
3185 | * request won't go away until we unregister it. We retain | |
3186 | * a pointer to the object request during that time (in | |
76756a51 ID |
3187 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3188 | * We'll drop that reference after we've unregistered it in | |
3189 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3190 | */ |
b30a01f2 | 3191 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3192 | |
b30a01f2 | 3193 | return 0; |
b30a01f2 ID |
3194 | } |
3195 | ||
3196 | /* | |
3197 | * Tear down a watch request, synchronously. | |
3198 | */ | |
76756a51 | 3199 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3200 | { |
b30a01f2 | 3201 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3202 | |
3203 | rbd_assert(rbd_dev->watch_event); | |
3204 | rbd_assert(rbd_dev->watch_request); | |
3205 | ||
76756a51 | 3206 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3207 | rbd_obj_request_put(rbd_dev->watch_request); |
3208 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3209 | |
76756a51 ID |
3210 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3211 | if (!IS_ERR(obj_request)) | |
3212 | rbd_obj_request_put(obj_request); | |
3213 | else | |
3214 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3215 | PTR_ERR(obj_request)); | |
3216 | ||
9969ebc5 AE |
3217 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3218 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3219 | } |
3220 | ||
36be9a76 | 3221 | /* |
f40eb349 AE |
3222 | * Synchronous osd object method call. Returns the number of bytes |
3223 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3224 | */ |
3225 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3226 | const char *object_name, | |
3227 | const char *class_name, | |
3228 | const char *method_name, | |
4157976b | 3229 | const void *outbound, |
36be9a76 | 3230 | size_t outbound_size, |
4157976b | 3231 | void *inbound, |
e2a58ee5 | 3232 | size_t inbound_size) |
36be9a76 | 3233 | { |
2169238d | 3234 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3235 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3236 | struct page **pages; |
3237 | u32 page_count; | |
3238 | int ret; | |
3239 | ||
3240 | /* | |
6010a451 AE |
3241 | * Method calls are ultimately read operations. The result |
3242 | * should placed into the inbound buffer provided. They | |
3243 | * also supply outbound data--parameters for the object | |
3244 | * method. Currently if this is present it will be a | |
3245 | * snapshot id. | |
36be9a76 | 3246 | */ |
57385b51 | 3247 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3248 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3249 | if (IS_ERR(pages)) | |
3250 | return PTR_ERR(pages); | |
3251 | ||
3252 | ret = -ENOMEM; | |
6010a451 | 3253 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3254 | OBJ_REQUEST_PAGES); |
3255 | if (!obj_request) | |
3256 | goto out; | |
3257 | ||
3258 | obj_request->pages = pages; | |
3259 | obj_request->page_count = page_count; | |
3260 | ||
6d2940c8 | 3261 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3262 | obj_request); |
36be9a76 AE |
3263 | if (!obj_request->osd_req) |
3264 | goto out; | |
3265 | ||
c99d2d4a | 3266 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3267 | class_name, method_name); |
3268 | if (outbound_size) { | |
3269 | struct ceph_pagelist *pagelist; | |
3270 | ||
3271 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3272 | if (!pagelist) | |
3273 | goto out; | |
3274 | ||
3275 | ceph_pagelist_init(pagelist); | |
3276 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3277 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3278 | pagelist); | |
3279 | } | |
a4ce40a9 AE |
3280 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3281 | obj_request->pages, inbound_size, | |
44cd188d | 3282 | 0, false, false); |
9d4df01f | 3283 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3284 | |
36be9a76 AE |
3285 | ret = rbd_obj_request_submit(osdc, obj_request); |
3286 | if (ret) | |
3287 | goto out; | |
3288 | ret = rbd_obj_request_wait(obj_request); | |
3289 | if (ret) | |
3290 | goto out; | |
3291 | ||
3292 | ret = obj_request->result; | |
3293 | if (ret < 0) | |
3294 | goto out; | |
57385b51 AE |
3295 | |
3296 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3297 | ret = (int)obj_request->xferred; | |
903bb32e | 3298 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3299 | out: |
3300 | if (obj_request) | |
3301 | rbd_obj_request_put(obj_request); | |
3302 | else | |
3303 | ceph_release_page_vector(pages, page_count); | |
3304 | ||
3305 | return ret; | |
3306 | } | |
3307 | ||
bc1ecc65 | 3308 | static void rbd_handle_request(struct rbd_device *rbd_dev, struct request *rq) |
bf0d5f50 | 3309 | { |
bc1ecc65 | 3310 | struct rbd_img_request *img_request; |
4e752f0a | 3311 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3312 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3313 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3314 | enum obj_operation_type op_type; |
4e752f0a | 3315 | u64 mapping_size; |
bf0d5f50 AE |
3316 | int result; |
3317 | ||
90e98c52 GZ |
3318 | if (rq->cmd_flags & REQ_DISCARD) |
3319 | op_type = OBJ_OP_DISCARD; | |
3320 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3321 | op_type = OBJ_OP_WRITE; |
3322 | else | |
3323 | op_type = OBJ_OP_READ; | |
3324 | ||
bc1ecc65 | 3325 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3326 | |
bc1ecc65 ID |
3327 | if (!length) { |
3328 | dout("%s: zero-length request\n", __func__); | |
3329 | result = 0; | |
3330 | goto err_rq; | |
3331 | } | |
bf0d5f50 | 3332 | |
6d2940c8 | 3333 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3334 | |
6d2940c8 | 3335 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3336 | if (rbd_dev->mapping.read_only) { |
3337 | result = -EROFS; | |
3338 | goto err_rq; | |
4dda41d3 | 3339 | } |
bc1ecc65 ID |
3340 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3341 | } | |
4dda41d3 | 3342 | |
bc1ecc65 ID |
3343 | /* |
3344 | * Quit early if the mapped snapshot no longer exists. It's | |
3345 | * still possible the snapshot will have disappeared by the | |
3346 | * time our request arrives at the osd, but there's no sense in | |
3347 | * sending it if we already know. | |
3348 | */ | |
3349 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3350 | dout("request for non-existent snapshot"); | |
3351 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3352 | result = -ENXIO; | |
3353 | goto err_rq; | |
3354 | } | |
4dda41d3 | 3355 | |
bc1ecc65 ID |
3356 | if (offset && length > U64_MAX - offset + 1) { |
3357 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3358 | length); | |
3359 | result = -EINVAL; | |
3360 | goto err_rq; /* Shouldn't happen */ | |
3361 | } | |
4dda41d3 | 3362 | |
4e752f0a JD |
3363 | down_read(&rbd_dev->header_rwsem); |
3364 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3365 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3366 | snapc = rbd_dev->header.snapc; |
3367 | ceph_get_snap_context(snapc); | |
3368 | } | |
3369 | up_read(&rbd_dev->header_rwsem); | |
3370 | ||
3371 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3372 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3373 | length, mapping_size); |
bc1ecc65 ID |
3374 | result = -EIO; |
3375 | goto err_rq; | |
3376 | } | |
bf0d5f50 | 3377 | |
6d2940c8 | 3378 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3379 | snapc); |
bc1ecc65 ID |
3380 | if (!img_request) { |
3381 | result = -ENOMEM; | |
3382 | goto err_rq; | |
3383 | } | |
3384 | img_request->rq = rq; | |
bf0d5f50 | 3385 | |
90e98c52 GZ |
3386 | if (op_type == OBJ_OP_DISCARD) |
3387 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3388 | NULL); | |
3389 | else | |
3390 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3391 | rq->bio); | |
bc1ecc65 ID |
3392 | if (result) |
3393 | goto err_img_request; | |
bf0d5f50 | 3394 | |
bc1ecc65 ID |
3395 | result = rbd_img_request_submit(img_request); |
3396 | if (result) | |
3397 | goto err_img_request; | |
bf0d5f50 | 3398 | |
bc1ecc65 | 3399 | return; |
bf0d5f50 | 3400 | |
bc1ecc65 ID |
3401 | err_img_request: |
3402 | rbd_img_request_put(img_request); | |
3403 | err_rq: | |
3404 | if (result) | |
3405 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3406 | obj_op_name(op_type), length, offset, result); |
4e752f0a JD |
3407 | if (snapc) |
3408 | ceph_put_snap_context(snapc); | |
bc1ecc65 ID |
3409 | blk_end_request_all(rq, result); |
3410 | } | |
bf0d5f50 | 3411 | |
bc1ecc65 ID |
3412 | static void rbd_request_workfn(struct work_struct *work) |
3413 | { | |
3414 | struct rbd_device *rbd_dev = | |
3415 | container_of(work, struct rbd_device, rq_work); | |
3416 | struct request *rq, *next; | |
3417 | LIST_HEAD(requests); | |
00a653e2 | 3418 | |
bc1ecc65 ID |
3419 | spin_lock_irq(&rbd_dev->lock); /* rq->q->queue_lock */ |
3420 | list_splice_init(&rbd_dev->rq_queue, &requests); | |
3421 | spin_unlock_irq(&rbd_dev->lock); | |
bf0d5f50 | 3422 | |
bc1ecc65 ID |
3423 | list_for_each_entry_safe(rq, next, &requests, queuelist) { |
3424 | list_del_init(&rq->queuelist); | |
3425 | rbd_handle_request(rbd_dev, rq); | |
3426 | } | |
3427 | } | |
bf0d5f50 | 3428 | |
bc1ecc65 ID |
3429 | /* |
3430 | * Called with q->queue_lock held and interrupts disabled, possibly on | |
3431 | * the way to schedule(). Do not sleep here! | |
3432 | */ | |
3433 | static void rbd_request_fn(struct request_queue *q) | |
3434 | { | |
3435 | struct rbd_device *rbd_dev = q->queuedata; | |
3436 | struct request *rq; | |
3437 | int queued = 0; | |
3438 | ||
3439 | rbd_assert(rbd_dev); | |
3440 | ||
3441 | while ((rq = blk_fetch_request(q))) { | |
3442 | /* Ignore any non-FS requests that filter through. */ | |
3443 | if (rq->cmd_type != REQ_TYPE_FS) { | |
3444 | dout("%s: non-fs request type %d\n", __func__, | |
3445 | (int) rq->cmd_type); | |
3446 | __blk_end_request_all(rq, 0); | |
3447 | continue; | |
bf0d5f50 | 3448 | } |
bc1ecc65 ID |
3449 | |
3450 | list_add_tail(&rq->queuelist, &rbd_dev->rq_queue); | |
3451 | queued++; | |
bf0d5f50 | 3452 | } |
bc1ecc65 ID |
3453 | |
3454 | if (queued) | |
3455 | queue_work(rbd_dev->rq_wq, &rbd_dev->rq_work); | |
bf0d5f50 AE |
3456 | } |
3457 | ||
602adf40 YS |
3458 | /* |
3459 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3460 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3461 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3462 | */ |
3463 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3464 | struct bio_vec *bvec) | |
3465 | { | |
3466 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3467 | sector_t sector_offset; |
3468 | sector_t sectors_per_obj; | |
3469 | sector_t obj_sector_offset; | |
3470 | int ret; | |
3471 | ||
3472 | /* | |
3473 | * Find how far into its rbd object the partition-relative | |
3474 | * bio start sector is to offset relative to the enclosing | |
3475 | * device. | |
3476 | */ | |
3477 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3478 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3479 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3480 | ||
3481 | /* | |
3482 | * Compute the number of bytes from that offset to the end | |
3483 | * of the object. Account for what's already used by the bio. | |
3484 | */ | |
3485 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3486 | if (ret > bmd->bi_size) | |
3487 | ret -= bmd->bi_size; | |
3488 | else | |
3489 | ret = 0; | |
3490 | ||
3491 | /* | |
3492 | * Don't send back more than was asked for. And if the bio | |
3493 | * was empty, let the whole thing through because: "Note | |
3494 | * that a block device *must* allow a single page to be | |
3495 | * added to an empty bio." | |
3496 | */ | |
3497 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3498 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3499 | ret = (int) bvec->bv_len; | |
3500 | ||
3501 | return ret; | |
602adf40 YS |
3502 | } |
3503 | ||
3504 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3505 | { | |
3506 | struct gendisk *disk = rbd_dev->disk; | |
3507 | ||
3508 | if (!disk) | |
3509 | return; | |
3510 | ||
a0cab924 AE |
3511 | rbd_dev->disk = NULL; |
3512 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3513 | del_gendisk(disk); |
a0cab924 AE |
3514 | if (disk->queue) |
3515 | blk_cleanup_queue(disk->queue); | |
3516 | } | |
602adf40 YS |
3517 | put_disk(disk); |
3518 | } | |
3519 | ||
788e2df3 AE |
3520 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3521 | const char *object_name, | |
7097f8df | 3522 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3523 | |
3524 | { | |
2169238d | 3525 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3526 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3527 | struct page **pages = NULL; |
3528 | u32 page_count; | |
1ceae7ef | 3529 | size_t size; |
788e2df3 AE |
3530 | int ret; |
3531 | ||
3532 | page_count = (u32) calc_pages_for(offset, length); | |
3533 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3534 | if (IS_ERR(pages)) | |
3535 | ret = PTR_ERR(pages); | |
3536 | ||
3537 | ret = -ENOMEM; | |
3538 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3539 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3540 | if (!obj_request) |
3541 | goto out; | |
3542 | ||
3543 | obj_request->pages = pages; | |
3544 | obj_request->page_count = page_count; | |
3545 | ||
6d2940c8 | 3546 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3547 | obj_request); |
788e2df3 AE |
3548 | if (!obj_request->osd_req) |
3549 | goto out; | |
3550 | ||
c99d2d4a AE |
3551 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3552 | offset, length, 0, 0); | |
406e2c9f | 3553 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3554 | obj_request->pages, |
44cd188d AE |
3555 | obj_request->length, |
3556 | obj_request->offset & ~PAGE_MASK, | |
3557 | false, false); | |
9d4df01f | 3558 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3559 | |
788e2df3 AE |
3560 | ret = rbd_obj_request_submit(osdc, obj_request); |
3561 | if (ret) | |
3562 | goto out; | |
3563 | ret = rbd_obj_request_wait(obj_request); | |
3564 | if (ret) | |
3565 | goto out; | |
3566 | ||
3567 | ret = obj_request->result; | |
3568 | if (ret < 0) | |
3569 | goto out; | |
1ceae7ef AE |
3570 | |
3571 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3572 | size = (size_t) obj_request->xferred; | |
903bb32e | 3573 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3574 | rbd_assert(size <= (size_t)INT_MAX); |
3575 | ret = (int)size; | |
788e2df3 AE |
3576 | out: |
3577 | if (obj_request) | |
3578 | rbd_obj_request_put(obj_request); | |
3579 | else | |
3580 | ceph_release_page_vector(pages, page_count); | |
3581 | ||
3582 | return ret; | |
3583 | } | |
3584 | ||
602adf40 | 3585 | /* |
662518b1 AE |
3586 | * Read the complete header for the given rbd device. On successful |
3587 | * return, the rbd_dev->header field will contain up-to-date | |
3588 | * information about the image. | |
602adf40 | 3589 | */ |
99a41ebc | 3590 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3591 | { |
4156d998 | 3592 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3593 | u32 snap_count = 0; |
4156d998 AE |
3594 | u64 names_size = 0; |
3595 | u32 want_count; | |
3596 | int ret; | |
602adf40 | 3597 | |
00f1f36f | 3598 | /* |
4156d998 AE |
3599 | * The complete header will include an array of its 64-bit |
3600 | * snapshot ids, followed by the names of those snapshots as | |
3601 | * a contiguous block of NUL-terminated strings. Note that | |
3602 | * the number of snapshots could change by the time we read | |
3603 | * it in, in which case we re-read it. | |
00f1f36f | 3604 | */ |
4156d998 AE |
3605 | do { |
3606 | size_t size; | |
3607 | ||
3608 | kfree(ondisk); | |
3609 | ||
3610 | size = sizeof (*ondisk); | |
3611 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3612 | size += names_size; | |
3613 | ondisk = kmalloc(size, GFP_KERNEL); | |
3614 | if (!ondisk) | |
662518b1 | 3615 | return -ENOMEM; |
4156d998 | 3616 | |
788e2df3 | 3617 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3618 | 0, size, ondisk); |
4156d998 | 3619 | if (ret < 0) |
662518b1 | 3620 | goto out; |
c0cd10db | 3621 | if ((size_t)ret < size) { |
4156d998 | 3622 | ret = -ENXIO; |
06ecc6cb AE |
3623 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3624 | size, ret); | |
662518b1 | 3625 | goto out; |
4156d998 AE |
3626 | } |
3627 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3628 | ret = -ENXIO; | |
06ecc6cb | 3629 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3630 | goto out; |
81e759fb | 3631 | } |
602adf40 | 3632 | |
4156d998 AE |
3633 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3634 | want_count = snap_count; | |
3635 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3636 | } while (snap_count != want_count); | |
00f1f36f | 3637 | |
662518b1 AE |
3638 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3639 | out: | |
4156d998 AE |
3640 | kfree(ondisk); |
3641 | ||
3642 | return ret; | |
602adf40 YS |
3643 | } |
3644 | ||
15228ede AE |
3645 | /* |
3646 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3647 | * has disappeared from the (just updated) snapshot context. | |
3648 | */ | |
3649 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3650 | { | |
3651 | u64 snap_id; | |
3652 | ||
3653 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3654 | return; | |
3655 | ||
3656 | snap_id = rbd_dev->spec->snap_id; | |
3657 | if (snap_id == CEPH_NOSNAP) | |
3658 | return; | |
3659 | ||
3660 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3661 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3662 | } | |
3663 | ||
9875201e JD |
3664 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3665 | { | |
3666 | sector_t size; | |
3667 | bool removing; | |
3668 | ||
3669 | /* | |
3670 | * Don't hold the lock while doing disk operations, | |
3671 | * or lock ordering will conflict with the bdev mutex via: | |
3672 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3673 | */ | |
3674 | spin_lock_irq(&rbd_dev->lock); | |
3675 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3676 | spin_unlock_irq(&rbd_dev->lock); | |
3677 | /* | |
3678 | * If the device is being removed, rbd_dev->disk has | |
3679 | * been destroyed, so don't try to update its size | |
3680 | */ | |
3681 | if (!removing) { | |
3682 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3683 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3684 | set_capacity(rbd_dev->disk, size); | |
3685 | revalidate_disk(rbd_dev->disk); | |
3686 | } | |
3687 | } | |
3688 | ||
cc4a38bd | 3689 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3690 | { |
e627db08 | 3691 | u64 mapping_size; |
1fe5e993 AE |
3692 | int ret; |
3693 | ||
cfbf6377 | 3694 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3695 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3696 | |
3697 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b ID |
3698 | if (ret) |
3699 | return ret; | |
15228ede | 3700 | |
e8f59b59 ID |
3701 | /* |
3702 | * If there is a parent, see if it has disappeared due to the | |
3703 | * mapped image getting flattened. | |
3704 | */ | |
3705 | if (rbd_dev->parent) { | |
3706 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3707 | if (ret) | |
3708 | return ret; | |
3709 | } | |
3710 | ||
5ff1108c ID |
3711 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
3712 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
3713 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
3714 | } else { | |
3715 | /* validate mapped snapshot's EXISTS flag */ | |
3716 | rbd_exists_validate(rbd_dev); | |
3717 | } | |
15228ede | 3718 | |
cfbf6377 AE |
3719 | up_write(&rbd_dev->header_rwsem); |
3720 | ||
e8f59b59 | 3721 | if (mapping_size != rbd_dev->mapping.size) |
9875201e | 3722 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3723 | |
52bb1f9b | 3724 | return 0; |
1fe5e993 AE |
3725 | } |
3726 | ||
602adf40 YS |
3727 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3728 | { | |
3729 | struct gendisk *disk; | |
3730 | struct request_queue *q; | |
593a9e7b | 3731 | u64 segment_size; |
602adf40 | 3732 | |
602adf40 | 3733 | /* create gendisk info */ |
7e513d43 ID |
3734 | disk = alloc_disk(single_major ? |
3735 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3736 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3737 | if (!disk) |
1fcdb8aa | 3738 | return -ENOMEM; |
602adf40 | 3739 | |
f0f8cef5 | 3740 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3741 | rbd_dev->dev_id); |
602adf40 | 3742 | disk->major = rbd_dev->major; |
dd82fff1 | 3743 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3744 | if (single_major) |
3745 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3746 | disk->fops = &rbd_bd_ops; |
3747 | disk->private_data = rbd_dev; | |
3748 | ||
bf0d5f50 | 3749 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3750 | if (!q) |
3751 | goto out_disk; | |
029bcbd8 | 3752 | |
593a9e7b AE |
3753 | /* We use the default size, but let's be explicit about it. */ |
3754 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3755 | ||
029bcbd8 | 3756 | /* set io sizes to object size */ |
593a9e7b AE |
3757 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3758 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3759 | blk_queue_max_segment_size(q, segment_size); | |
3760 | blk_queue_io_min(q, segment_size); | |
3761 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3762 | |
90e98c52 GZ |
3763 | /* enable the discard support */ |
3764 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3765 | q->limits.discard_granularity = segment_size; | |
3766 | q->limits.discard_alignment = segment_size; | |
3767 | ||
602adf40 YS |
3768 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3769 | disk->queue = q; | |
3770 | ||
3771 | q->queuedata = rbd_dev; | |
3772 | ||
3773 | rbd_dev->disk = disk; | |
602adf40 | 3774 | |
602adf40 | 3775 | return 0; |
602adf40 YS |
3776 | out_disk: |
3777 | put_disk(disk); | |
1fcdb8aa AE |
3778 | |
3779 | return -ENOMEM; | |
602adf40 YS |
3780 | } |
3781 | ||
dfc5606d YS |
3782 | /* |
3783 | sysfs | |
3784 | */ | |
3785 | ||
593a9e7b AE |
3786 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3787 | { | |
3788 | return container_of(dev, struct rbd_device, dev); | |
3789 | } | |
3790 | ||
dfc5606d YS |
3791 | static ssize_t rbd_size_show(struct device *dev, |
3792 | struct device_attribute *attr, char *buf) | |
3793 | { | |
593a9e7b | 3794 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3795 | |
fc71d833 AE |
3796 | return sprintf(buf, "%llu\n", |
3797 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3798 | } |
3799 | ||
34b13184 AE |
3800 | /* |
3801 | * Note this shows the features for whatever's mapped, which is not | |
3802 | * necessarily the base image. | |
3803 | */ | |
3804 | static ssize_t rbd_features_show(struct device *dev, | |
3805 | struct device_attribute *attr, char *buf) | |
3806 | { | |
3807 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3808 | ||
3809 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3810 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3811 | } |
3812 | ||
dfc5606d YS |
3813 | static ssize_t rbd_major_show(struct device *dev, |
3814 | struct device_attribute *attr, char *buf) | |
3815 | { | |
593a9e7b | 3816 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3817 | |
fc71d833 AE |
3818 | if (rbd_dev->major) |
3819 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3820 | ||
3821 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3822 | } |
3823 | ||
3824 | static ssize_t rbd_minor_show(struct device *dev, | |
3825 | struct device_attribute *attr, char *buf) | |
3826 | { | |
3827 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3828 | |
dd82fff1 | 3829 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3830 | } |
3831 | ||
3832 | static ssize_t rbd_client_id_show(struct device *dev, | |
3833 | struct device_attribute *attr, char *buf) | |
602adf40 | 3834 | { |
593a9e7b | 3835 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3836 | |
1dbb4399 AE |
3837 | return sprintf(buf, "client%lld\n", |
3838 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3839 | } |
3840 | ||
dfc5606d YS |
3841 | static ssize_t rbd_pool_show(struct device *dev, |
3842 | struct device_attribute *attr, char *buf) | |
602adf40 | 3843 | { |
593a9e7b | 3844 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3845 | |
0d7dbfce | 3846 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3847 | } |
3848 | ||
9bb2f334 AE |
3849 | static ssize_t rbd_pool_id_show(struct device *dev, |
3850 | struct device_attribute *attr, char *buf) | |
3851 | { | |
3852 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3853 | ||
0d7dbfce | 3854 | return sprintf(buf, "%llu\n", |
fc71d833 | 3855 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3856 | } |
3857 | ||
dfc5606d YS |
3858 | static ssize_t rbd_name_show(struct device *dev, |
3859 | struct device_attribute *attr, char *buf) | |
3860 | { | |
593a9e7b | 3861 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3862 | |
a92ffdf8 AE |
3863 | if (rbd_dev->spec->image_name) |
3864 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3865 | ||
3866 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3867 | } |
3868 | ||
589d30e0 AE |
3869 | static ssize_t rbd_image_id_show(struct device *dev, |
3870 | struct device_attribute *attr, char *buf) | |
3871 | { | |
3872 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3873 | ||
0d7dbfce | 3874 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3875 | } |
3876 | ||
34b13184 AE |
3877 | /* |
3878 | * Shows the name of the currently-mapped snapshot (or | |
3879 | * RBD_SNAP_HEAD_NAME for the base image). | |
3880 | */ | |
dfc5606d YS |
3881 | static ssize_t rbd_snap_show(struct device *dev, |
3882 | struct device_attribute *attr, | |
3883 | char *buf) | |
3884 | { | |
593a9e7b | 3885 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3886 | |
0d7dbfce | 3887 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3888 | } |
3889 | ||
86b00e0d | 3890 | /* |
ff96128f ID |
3891 | * For a v2 image, shows the chain of parent images, separated by empty |
3892 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3893 | * image)". | |
86b00e0d AE |
3894 | */ |
3895 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3896 | struct device_attribute *attr, |
3897 | char *buf) | |
86b00e0d AE |
3898 | { |
3899 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3900 | ssize_t count = 0; |
86b00e0d | 3901 | |
ff96128f | 3902 | if (!rbd_dev->parent) |
86b00e0d AE |
3903 | return sprintf(buf, "(no parent image)\n"); |
3904 | ||
ff96128f ID |
3905 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3906 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3907 | ||
3908 | count += sprintf(&buf[count], "%s" | |
3909 | "pool_id %llu\npool_name %s\n" | |
3910 | "image_id %s\nimage_name %s\n" | |
3911 | "snap_id %llu\nsnap_name %s\n" | |
3912 | "overlap %llu\n", | |
3913 | !count ? "" : "\n", /* first? */ | |
3914 | spec->pool_id, spec->pool_name, | |
3915 | spec->image_id, spec->image_name ?: "(unknown)", | |
3916 | spec->snap_id, spec->snap_name, | |
3917 | rbd_dev->parent_overlap); | |
3918 | } | |
3919 | ||
3920 | return count; | |
86b00e0d AE |
3921 | } |
3922 | ||
dfc5606d YS |
3923 | static ssize_t rbd_image_refresh(struct device *dev, |
3924 | struct device_attribute *attr, | |
3925 | const char *buf, | |
3926 | size_t size) | |
3927 | { | |
593a9e7b | 3928 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3929 | int ret; |
602adf40 | 3930 | |
cc4a38bd | 3931 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3932 | if (ret) |
52bb1f9b | 3933 | return ret; |
b813623a | 3934 | |
52bb1f9b | 3935 | return size; |
dfc5606d | 3936 | } |
602adf40 | 3937 | |
dfc5606d | 3938 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3939 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3940 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3941 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3942 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3943 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3944 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3945 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3946 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3947 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3948 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3949 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3950 | |
3951 | static struct attribute *rbd_attrs[] = { | |
3952 | &dev_attr_size.attr, | |
34b13184 | 3953 | &dev_attr_features.attr, |
dfc5606d | 3954 | &dev_attr_major.attr, |
dd82fff1 | 3955 | &dev_attr_minor.attr, |
dfc5606d YS |
3956 | &dev_attr_client_id.attr, |
3957 | &dev_attr_pool.attr, | |
9bb2f334 | 3958 | &dev_attr_pool_id.attr, |
dfc5606d | 3959 | &dev_attr_name.attr, |
589d30e0 | 3960 | &dev_attr_image_id.attr, |
dfc5606d | 3961 | &dev_attr_current_snap.attr, |
86b00e0d | 3962 | &dev_attr_parent.attr, |
dfc5606d | 3963 | &dev_attr_refresh.attr, |
dfc5606d YS |
3964 | NULL |
3965 | }; | |
3966 | ||
3967 | static struct attribute_group rbd_attr_group = { | |
3968 | .attrs = rbd_attrs, | |
3969 | }; | |
3970 | ||
3971 | static const struct attribute_group *rbd_attr_groups[] = { | |
3972 | &rbd_attr_group, | |
3973 | NULL | |
3974 | }; | |
3975 | ||
3976 | static void rbd_sysfs_dev_release(struct device *dev) | |
3977 | { | |
3978 | } | |
3979 | ||
3980 | static struct device_type rbd_device_type = { | |
3981 | .name = "rbd", | |
3982 | .groups = rbd_attr_groups, | |
3983 | .release = rbd_sysfs_dev_release, | |
3984 | }; | |
3985 | ||
8b8fb99c AE |
3986 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3987 | { | |
3988 | kref_get(&spec->kref); | |
3989 | ||
3990 | return spec; | |
3991 | } | |
3992 | ||
3993 | static void rbd_spec_free(struct kref *kref); | |
3994 | static void rbd_spec_put(struct rbd_spec *spec) | |
3995 | { | |
3996 | if (spec) | |
3997 | kref_put(&spec->kref, rbd_spec_free); | |
3998 | } | |
3999 | ||
4000 | static struct rbd_spec *rbd_spec_alloc(void) | |
4001 | { | |
4002 | struct rbd_spec *spec; | |
4003 | ||
4004 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4005 | if (!spec) | |
4006 | return NULL; | |
04077599 ID |
4007 | |
4008 | spec->pool_id = CEPH_NOPOOL; | |
4009 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4010 | kref_init(&spec->kref); |
4011 | ||
8b8fb99c AE |
4012 | return spec; |
4013 | } | |
4014 | ||
4015 | static void rbd_spec_free(struct kref *kref) | |
4016 | { | |
4017 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4018 | ||
4019 | kfree(spec->pool_name); | |
4020 | kfree(spec->image_id); | |
4021 | kfree(spec->image_name); | |
4022 | kfree(spec->snap_name); | |
4023 | kfree(spec); | |
4024 | } | |
4025 | ||
cc344fa1 | 4026 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
4027 | struct rbd_spec *spec) |
4028 | { | |
4029 | struct rbd_device *rbd_dev; | |
4030 | ||
4031 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4032 | if (!rbd_dev) | |
4033 | return NULL; | |
4034 | ||
4035 | spin_lock_init(&rbd_dev->lock); | |
bc1ecc65 ID |
4036 | INIT_LIST_HEAD(&rbd_dev->rq_queue); |
4037 | INIT_WORK(&rbd_dev->rq_work, rbd_request_workfn); | |
6d292906 | 4038 | rbd_dev->flags = 0; |
a2acd00e | 4039 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4040 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4041 | init_rwsem(&rbd_dev->header_rwsem); |
4042 | ||
4043 | rbd_dev->spec = spec; | |
4044 | rbd_dev->rbd_client = rbdc; | |
4045 | ||
0903e875 AE |
4046 | /* Initialize the layout used for all rbd requests */ |
4047 | ||
4048 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4049 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4050 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4051 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4052 | ||
c53d5893 AE |
4053 | return rbd_dev; |
4054 | } | |
4055 | ||
4056 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4057 | { | |
c53d5893 AE |
4058 | rbd_put_client(rbd_dev->rbd_client); |
4059 | rbd_spec_put(rbd_dev->spec); | |
4060 | kfree(rbd_dev); | |
4061 | } | |
4062 | ||
9d475de5 AE |
4063 | /* |
4064 | * Get the size and object order for an image snapshot, or if | |
4065 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4066 | * image. | |
4067 | */ | |
4068 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4069 | u8 *order, u64 *snap_size) | |
4070 | { | |
4071 | __le64 snapid = cpu_to_le64(snap_id); | |
4072 | int ret; | |
4073 | struct { | |
4074 | u8 order; | |
4075 | __le64 size; | |
4076 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4077 | ||
36be9a76 | 4078 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4079 | "rbd", "get_size", |
4157976b | 4080 | &snapid, sizeof (snapid), |
e2a58ee5 | 4081 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4082 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4083 | if (ret < 0) |
4084 | return ret; | |
57385b51 AE |
4085 | if (ret < sizeof (size_buf)) |
4086 | return -ERANGE; | |
9d475de5 | 4087 | |
c3545579 | 4088 | if (order) { |
c86f86e9 | 4089 | *order = size_buf.order; |
c3545579 JD |
4090 | dout(" order %u", (unsigned int)*order); |
4091 | } | |
9d475de5 AE |
4092 | *snap_size = le64_to_cpu(size_buf.size); |
4093 | ||
c3545579 JD |
4094 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4095 | (unsigned long long)snap_id, | |
57385b51 | 4096 | (unsigned long long)*snap_size); |
9d475de5 AE |
4097 | |
4098 | return 0; | |
4099 | } | |
4100 | ||
4101 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4102 | { | |
4103 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4104 | &rbd_dev->header.obj_order, | |
4105 | &rbd_dev->header.image_size); | |
4106 | } | |
4107 | ||
1e130199 AE |
4108 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4109 | { | |
4110 | void *reply_buf; | |
4111 | int ret; | |
4112 | void *p; | |
4113 | ||
4114 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4115 | if (!reply_buf) | |
4116 | return -ENOMEM; | |
4117 | ||
36be9a76 | 4118 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4119 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4120 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4121 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4122 | if (ret < 0) |
4123 | goto out; | |
4124 | ||
4125 | p = reply_buf; | |
4126 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4127 | p + ret, NULL, GFP_NOIO); |
4128 | ret = 0; | |
1e130199 AE |
4129 | |
4130 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4131 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4132 | rbd_dev->header.object_prefix = NULL; | |
4133 | } else { | |
4134 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4135 | } | |
1e130199 AE |
4136 | out: |
4137 | kfree(reply_buf); | |
4138 | ||
4139 | return ret; | |
4140 | } | |
4141 | ||
b1b5402a AE |
4142 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4143 | u64 *snap_features) | |
4144 | { | |
4145 | __le64 snapid = cpu_to_le64(snap_id); | |
4146 | struct { | |
4147 | __le64 features; | |
4148 | __le64 incompat; | |
4157976b | 4149 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4150 | u64 incompat; |
b1b5402a AE |
4151 | int ret; |
4152 | ||
36be9a76 | 4153 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4154 | "rbd", "get_features", |
4157976b | 4155 | &snapid, sizeof (snapid), |
e2a58ee5 | 4156 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4157 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4158 | if (ret < 0) |
4159 | return ret; | |
57385b51 AE |
4160 | if (ret < sizeof (features_buf)) |
4161 | return -ERANGE; | |
d889140c AE |
4162 | |
4163 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4164 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4165 | return -ENXIO; |
d889140c | 4166 | |
b1b5402a AE |
4167 | *snap_features = le64_to_cpu(features_buf.features); |
4168 | ||
4169 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4170 | (unsigned long long)snap_id, |
4171 | (unsigned long long)*snap_features, | |
4172 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4173 | |
4174 | return 0; | |
4175 | } | |
4176 | ||
4177 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4178 | { | |
4179 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4180 | &rbd_dev->header.features); | |
4181 | } | |
4182 | ||
86b00e0d AE |
4183 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4184 | { | |
4185 | struct rbd_spec *parent_spec; | |
4186 | size_t size; | |
4187 | void *reply_buf = NULL; | |
4188 | __le64 snapid; | |
4189 | void *p; | |
4190 | void *end; | |
642a2537 | 4191 | u64 pool_id; |
86b00e0d | 4192 | char *image_id; |
3b5cf2a2 | 4193 | u64 snap_id; |
86b00e0d | 4194 | u64 overlap; |
86b00e0d AE |
4195 | int ret; |
4196 | ||
4197 | parent_spec = rbd_spec_alloc(); | |
4198 | if (!parent_spec) | |
4199 | return -ENOMEM; | |
4200 | ||
4201 | size = sizeof (__le64) + /* pool_id */ | |
4202 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4203 | sizeof (__le64) + /* snap_id */ | |
4204 | sizeof (__le64); /* overlap */ | |
4205 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4206 | if (!reply_buf) { | |
4207 | ret = -ENOMEM; | |
4208 | goto out_err; | |
4209 | } | |
4210 | ||
4d9b67cd | 4211 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4212 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4213 | "rbd", "get_parent", |
4157976b | 4214 | &snapid, sizeof (snapid), |
e2a58ee5 | 4215 | reply_buf, size); |
36be9a76 | 4216 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4217 | if (ret < 0) |
4218 | goto out_err; | |
4219 | ||
86b00e0d | 4220 | p = reply_buf; |
57385b51 AE |
4221 | end = reply_buf + ret; |
4222 | ret = -ERANGE; | |
642a2537 | 4223 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4224 | if (pool_id == CEPH_NOPOOL) { |
4225 | /* | |
4226 | * Either the parent never existed, or we have | |
4227 | * record of it but the image got flattened so it no | |
4228 | * longer has a parent. When the parent of a | |
4229 | * layered image disappears we immediately set the | |
4230 | * overlap to 0. The effect of this is that all new | |
4231 | * requests will be treated as if the image had no | |
4232 | * parent. | |
4233 | */ | |
4234 | if (rbd_dev->parent_overlap) { | |
4235 | rbd_dev->parent_overlap = 0; | |
4236 | smp_mb(); | |
4237 | rbd_dev_parent_put(rbd_dev); | |
4238 | pr_info("%s: clone image has been flattened\n", | |
4239 | rbd_dev->disk->disk_name); | |
4240 | } | |
4241 | ||
86b00e0d | 4242 | goto out; /* No parent? No problem. */ |
392a9dad | 4243 | } |
86b00e0d | 4244 | |
0903e875 AE |
4245 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4246 | ||
4247 | ret = -EIO; | |
642a2537 | 4248 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4249 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4250 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4251 | goto out_err; |
c0cd10db | 4252 | } |
0903e875 | 4253 | |
979ed480 | 4254 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4255 | if (IS_ERR(image_id)) { |
4256 | ret = PTR_ERR(image_id); | |
4257 | goto out_err; | |
4258 | } | |
3b5cf2a2 | 4259 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4260 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4261 | ||
3b5cf2a2 AE |
4262 | /* |
4263 | * The parent won't change (except when the clone is | |
4264 | * flattened, already handled that). So we only need to | |
4265 | * record the parent spec we have not already done so. | |
4266 | */ | |
4267 | if (!rbd_dev->parent_spec) { | |
4268 | parent_spec->pool_id = pool_id; | |
4269 | parent_spec->image_id = image_id; | |
4270 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4271 | rbd_dev->parent_spec = parent_spec; |
4272 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4273 | } else { |
4274 | kfree(image_id); | |
3b5cf2a2 AE |
4275 | } |
4276 | ||
4277 | /* | |
4278 | * We always update the parent overlap. If it's zero we | |
4279 | * treat it specially. | |
4280 | */ | |
4281 | rbd_dev->parent_overlap = overlap; | |
4282 | smp_mb(); | |
4283 | if (!overlap) { | |
4284 | ||
4285 | /* A null parent_spec indicates it's the initial probe */ | |
4286 | ||
4287 | if (parent_spec) { | |
4288 | /* | |
4289 | * The overlap has become zero, so the clone | |
4290 | * must have been resized down to 0 at some | |
4291 | * point. Treat this the same as a flatten. | |
4292 | */ | |
4293 | rbd_dev_parent_put(rbd_dev); | |
4294 | pr_info("%s: clone image now standalone\n", | |
4295 | rbd_dev->disk->disk_name); | |
4296 | } else { | |
4297 | /* | |
4298 | * For the initial probe, if we find the | |
4299 | * overlap is zero we just pretend there was | |
4300 | * no parent image. | |
4301 | */ | |
9584d508 | 4302 | rbd_warn(rbd_dev, "ignoring parent with overlap 0"); |
3b5cf2a2 | 4303 | } |
70cf49cf | 4304 | } |
86b00e0d AE |
4305 | out: |
4306 | ret = 0; | |
4307 | out_err: | |
4308 | kfree(reply_buf); | |
4309 | rbd_spec_put(parent_spec); | |
4310 | ||
4311 | return ret; | |
4312 | } | |
4313 | ||
cc070d59 AE |
4314 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4315 | { | |
4316 | struct { | |
4317 | __le64 stripe_unit; | |
4318 | __le64 stripe_count; | |
4319 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4320 | size_t size = sizeof (striping_info_buf); | |
4321 | void *p; | |
4322 | u64 obj_size; | |
4323 | u64 stripe_unit; | |
4324 | u64 stripe_count; | |
4325 | int ret; | |
4326 | ||
4327 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4328 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4329 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4330 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4331 | if (ret < 0) | |
4332 | return ret; | |
4333 | if (ret < size) | |
4334 | return -ERANGE; | |
4335 | ||
4336 | /* | |
4337 | * We don't actually support the "fancy striping" feature | |
4338 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4339 | * defaults the behavior is the same as before. So find | |
4340 | * out, and only fail if the image has non-default values. | |
4341 | */ | |
4342 | ret = -EINVAL; | |
4343 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4344 | p = &striping_info_buf; | |
4345 | stripe_unit = ceph_decode_64(&p); | |
4346 | if (stripe_unit != obj_size) { | |
4347 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4348 | "(got %llu want %llu)", | |
4349 | stripe_unit, obj_size); | |
4350 | return -EINVAL; | |
4351 | } | |
4352 | stripe_count = ceph_decode_64(&p); | |
4353 | if (stripe_count != 1) { | |
4354 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4355 | "(got %llu want 1)", stripe_count); | |
4356 | return -EINVAL; | |
4357 | } | |
500d0c0f AE |
4358 | rbd_dev->header.stripe_unit = stripe_unit; |
4359 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4360 | |
4361 | return 0; | |
4362 | } | |
4363 | ||
9e15b77d AE |
4364 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4365 | { | |
4366 | size_t image_id_size; | |
4367 | char *image_id; | |
4368 | void *p; | |
4369 | void *end; | |
4370 | size_t size; | |
4371 | void *reply_buf = NULL; | |
4372 | size_t len = 0; | |
4373 | char *image_name = NULL; | |
4374 | int ret; | |
4375 | ||
4376 | rbd_assert(!rbd_dev->spec->image_name); | |
4377 | ||
69e7a02f AE |
4378 | len = strlen(rbd_dev->spec->image_id); |
4379 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4380 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4381 | if (!image_id) | |
4382 | return NULL; | |
4383 | ||
4384 | p = image_id; | |
4157976b | 4385 | end = image_id + image_id_size; |
57385b51 | 4386 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4387 | |
4388 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4389 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4390 | if (!reply_buf) | |
4391 | goto out; | |
4392 | ||
36be9a76 | 4393 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4394 | "rbd", "dir_get_name", |
4395 | image_id, image_id_size, | |
e2a58ee5 | 4396 | reply_buf, size); |
9e15b77d AE |
4397 | if (ret < 0) |
4398 | goto out; | |
4399 | p = reply_buf; | |
f40eb349 AE |
4400 | end = reply_buf + ret; |
4401 | ||
9e15b77d AE |
4402 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4403 | if (IS_ERR(image_name)) | |
4404 | image_name = NULL; | |
4405 | else | |
4406 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4407 | out: | |
4408 | kfree(reply_buf); | |
4409 | kfree(image_id); | |
4410 | ||
4411 | return image_name; | |
4412 | } | |
4413 | ||
2ad3d716 AE |
4414 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4415 | { | |
4416 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4417 | const char *snap_name; | |
4418 | u32 which = 0; | |
4419 | ||
4420 | /* Skip over names until we find the one we are looking for */ | |
4421 | ||
4422 | snap_name = rbd_dev->header.snap_names; | |
4423 | while (which < snapc->num_snaps) { | |
4424 | if (!strcmp(name, snap_name)) | |
4425 | return snapc->snaps[which]; | |
4426 | snap_name += strlen(snap_name) + 1; | |
4427 | which++; | |
4428 | } | |
4429 | return CEPH_NOSNAP; | |
4430 | } | |
4431 | ||
4432 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4433 | { | |
4434 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4435 | u32 which; | |
4436 | bool found = false; | |
4437 | u64 snap_id; | |
4438 | ||
4439 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4440 | const char *snap_name; | |
4441 | ||
4442 | snap_id = snapc->snaps[which]; | |
4443 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4444 | if (IS_ERR(snap_name)) { |
4445 | /* ignore no-longer existing snapshots */ | |
4446 | if (PTR_ERR(snap_name) == -ENOENT) | |
4447 | continue; | |
4448 | else | |
4449 | break; | |
4450 | } | |
2ad3d716 AE |
4451 | found = !strcmp(name, snap_name); |
4452 | kfree(snap_name); | |
4453 | } | |
4454 | return found ? snap_id : CEPH_NOSNAP; | |
4455 | } | |
4456 | ||
4457 | /* | |
4458 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4459 | * no snapshot by that name is found, or if an error occurs. | |
4460 | */ | |
4461 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4462 | { | |
4463 | if (rbd_dev->image_format == 1) | |
4464 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4465 | ||
4466 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4467 | } | |
4468 | ||
9e15b77d | 4469 | /* |
04077599 ID |
4470 | * An image being mapped will have everything but the snap id. |
4471 | */ | |
4472 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4473 | { | |
4474 | struct rbd_spec *spec = rbd_dev->spec; | |
4475 | ||
4476 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4477 | rbd_assert(spec->image_id && spec->image_name); | |
4478 | rbd_assert(spec->snap_name); | |
4479 | ||
4480 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4481 | u64 snap_id; | |
4482 | ||
4483 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4484 | if (snap_id == CEPH_NOSNAP) | |
4485 | return -ENOENT; | |
4486 | ||
4487 | spec->snap_id = snap_id; | |
4488 | } else { | |
4489 | spec->snap_id = CEPH_NOSNAP; | |
4490 | } | |
4491 | ||
4492 | return 0; | |
4493 | } | |
4494 | ||
4495 | /* | |
4496 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4497 | * |
04077599 ID |
4498 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4499 | * can't figure out the name for an image id. | |
9e15b77d | 4500 | */ |
04077599 | 4501 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4502 | { |
2e9f7f1c AE |
4503 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4504 | struct rbd_spec *spec = rbd_dev->spec; | |
4505 | const char *pool_name; | |
4506 | const char *image_name; | |
4507 | const char *snap_name; | |
9e15b77d AE |
4508 | int ret; |
4509 | ||
04077599 ID |
4510 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4511 | rbd_assert(spec->image_id); | |
4512 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4513 | |
2e9f7f1c | 4514 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4515 | |
2e9f7f1c AE |
4516 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4517 | if (!pool_name) { | |
4518 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4519 | return -EIO; |
4520 | } | |
2e9f7f1c AE |
4521 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4522 | if (!pool_name) | |
9e15b77d AE |
4523 | return -ENOMEM; |
4524 | ||
4525 | /* Fetch the image name; tolerate failure here */ | |
4526 | ||
2e9f7f1c AE |
4527 | image_name = rbd_dev_image_name(rbd_dev); |
4528 | if (!image_name) | |
06ecc6cb | 4529 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4530 | |
04077599 | 4531 | /* Fetch the snapshot name */ |
9e15b77d | 4532 | |
2e9f7f1c | 4533 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4534 | if (IS_ERR(snap_name)) { |
4535 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4536 | goto out_err; |
2e9f7f1c AE |
4537 | } |
4538 | ||
4539 | spec->pool_name = pool_name; | |
4540 | spec->image_name = image_name; | |
4541 | spec->snap_name = snap_name; | |
9e15b77d AE |
4542 | |
4543 | return 0; | |
04077599 | 4544 | |
9e15b77d | 4545 | out_err: |
2e9f7f1c AE |
4546 | kfree(image_name); |
4547 | kfree(pool_name); | |
9e15b77d AE |
4548 | return ret; |
4549 | } | |
4550 | ||
cc4a38bd | 4551 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4552 | { |
4553 | size_t size; | |
4554 | int ret; | |
4555 | void *reply_buf; | |
4556 | void *p; | |
4557 | void *end; | |
4558 | u64 seq; | |
4559 | u32 snap_count; | |
4560 | struct ceph_snap_context *snapc; | |
4561 | u32 i; | |
4562 | ||
4563 | /* | |
4564 | * We'll need room for the seq value (maximum snapshot id), | |
4565 | * snapshot count, and array of that many snapshot ids. | |
4566 | * For now we have a fixed upper limit on the number we're | |
4567 | * prepared to receive. | |
4568 | */ | |
4569 | size = sizeof (__le64) + sizeof (__le32) + | |
4570 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4571 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4572 | if (!reply_buf) | |
4573 | return -ENOMEM; | |
4574 | ||
36be9a76 | 4575 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4576 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4577 | reply_buf, size); |
36be9a76 | 4578 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4579 | if (ret < 0) |
4580 | goto out; | |
4581 | ||
35d489f9 | 4582 | p = reply_buf; |
57385b51 AE |
4583 | end = reply_buf + ret; |
4584 | ret = -ERANGE; | |
35d489f9 AE |
4585 | ceph_decode_64_safe(&p, end, seq, out); |
4586 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4587 | ||
4588 | /* | |
4589 | * Make sure the reported number of snapshot ids wouldn't go | |
4590 | * beyond the end of our buffer. But before checking that, | |
4591 | * make sure the computed size of the snapshot context we | |
4592 | * allocate is representable in a size_t. | |
4593 | */ | |
4594 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4595 | / sizeof (u64)) { | |
4596 | ret = -EINVAL; | |
4597 | goto out; | |
4598 | } | |
4599 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4600 | goto out; | |
468521c1 | 4601 | ret = 0; |
35d489f9 | 4602 | |
812164f8 | 4603 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4604 | if (!snapc) { |
4605 | ret = -ENOMEM; | |
4606 | goto out; | |
4607 | } | |
35d489f9 | 4608 | snapc->seq = seq; |
35d489f9 AE |
4609 | for (i = 0; i < snap_count; i++) |
4610 | snapc->snaps[i] = ceph_decode_64(&p); | |
4611 | ||
49ece554 | 4612 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4613 | rbd_dev->header.snapc = snapc; |
4614 | ||
4615 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4616 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4617 | out: |
4618 | kfree(reply_buf); | |
4619 | ||
57385b51 | 4620 | return ret; |
35d489f9 AE |
4621 | } |
4622 | ||
54cac61f AE |
4623 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4624 | u64 snap_id) | |
b8b1e2db AE |
4625 | { |
4626 | size_t size; | |
4627 | void *reply_buf; | |
54cac61f | 4628 | __le64 snapid; |
b8b1e2db AE |
4629 | int ret; |
4630 | void *p; | |
4631 | void *end; | |
b8b1e2db AE |
4632 | char *snap_name; |
4633 | ||
4634 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4635 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4636 | if (!reply_buf) | |
4637 | return ERR_PTR(-ENOMEM); | |
4638 | ||
54cac61f | 4639 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4640 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4641 | "rbd", "get_snapshot_name", |
54cac61f | 4642 | &snapid, sizeof (snapid), |
e2a58ee5 | 4643 | reply_buf, size); |
36be9a76 | 4644 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4645 | if (ret < 0) { |
4646 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4647 | goto out; |
f40eb349 | 4648 | } |
b8b1e2db AE |
4649 | |
4650 | p = reply_buf; | |
f40eb349 | 4651 | end = reply_buf + ret; |
e5c35534 | 4652 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4653 | if (IS_ERR(snap_name)) |
b8b1e2db | 4654 | goto out; |
b8b1e2db | 4655 | |
f40eb349 | 4656 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4657 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4658 | out: |
4659 | kfree(reply_buf); | |
4660 | ||
f40eb349 | 4661 | return snap_name; |
b8b1e2db AE |
4662 | } |
4663 | ||
2df3fac7 | 4664 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4665 | { |
2df3fac7 | 4666 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4667 | int ret; |
117973fb | 4668 | |
1617e40c JD |
4669 | ret = rbd_dev_v2_image_size(rbd_dev); |
4670 | if (ret) | |
cfbf6377 | 4671 | return ret; |
1617e40c | 4672 | |
2df3fac7 AE |
4673 | if (first_time) { |
4674 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4675 | if (ret) | |
cfbf6377 | 4676 | return ret; |
2df3fac7 AE |
4677 | } |
4678 | ||
cc4a38bd | 4679 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4680 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4681 | |
4682 | return ret; | |
4683 | } | |
4684 | ||
a720ae09 ID |
4685 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4686 | { | |
4687 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4688 | ||
4689 | if (rbd_dev->image_format == 1) | |
4690 | return rbd_dev_v1_header_info(rbd_dev); | |
4691 | ||
4692 | return rbd_dev_v2_header_info(rbd_dev); | |
4693 | } | |
4694 | ||
dfc5606d YS |
4695 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4696 | { | |
dfc5606d | 4697 | struct device *dev; |
cd789ab9 | 4698 | int ret; |
dfc5606d | 4699 | |
cd789ab9 | 4700 | dev = &rbd_dev->dev; |
dfc5606d YS |
4701 | dev->bus = &rbd_bus_type; |
4702 | dev->type = &rbd_device_type; | |
4703 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4704 | dev->release = rbd_dev_device_release; |
de71a297 | 4705 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4706 | ret = device_register(dev); |
dfc5606d | 4707 | |
dfc5606d | 4708 | return ret; |
602adf40 YS |
4709 | } |
4710 | ||
dfc5606d YS |
4711 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4712 | { | |
4713 | device_unregister(&rbd_dev->dev); | |
4714 | } | |
4715 | ||
1ddbe94e | 4716 | /* |
499afd5b | 4717 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4718 | * the rbd_dev to the global list. |
1ddbe94e | 4719 | */ |
f8a22fc2 | 4720 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4721 | { |
f8a22fc2 ID |
4722 | int new_dev_id; |
4723 | ||
9b60e70b ID |
4724 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4725 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4726 | GFP_KERNEL); | |
f8a22fc2 ID |
4727 | if (new_dev_id < 0) |
4728 | return new_dev_id; | |
4729 | ||
4730 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4731 | |
4732 | spin_lock(&rbd_dev_list_lock); | |
4733 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4734 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4735 | |
70eebd20 | 4736 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4737 | |
4738 | return 0; | |
1ddbe94e | 4739 | } |
b7f23c36 | 4740 | |
1ddbe94e | 4741 | /* |
499afd5b AE |
4742 | * Remove an rbd_dev from the global list, and record that its |
4743 | * identifier is no longer in use. | |
1ddbe94e | 4744 | */ |
e2839308 | 4745 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4746 | { |
499afd5b AE |
4747 | spin_lock(&rbd_dev_list_lock); |
4748 | list_del_init(&rbd_dev->node); | |
4749 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4750 | |
f8a22fc2 ID |
4751 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4752 | ||
4753 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4754 | } |
4755 | ||
e28fff26 AE |
4756 | /* |
4757 | * Skips over white space at *buf, and updates *buf to point to the | |
4758 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4759 | * the token (string of non-white space characters) found. Note |
4760 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4761 | */ |
4762 | static inline size_t next_token(const char **buf) | |
4763 | { | |
4764 | /* | |
4765 | * These are the characters that produce nonzero for | |
4766 | * isspace() in the "C" and "POSIX" locales. | |
4767 | */ | |
4768 | const char *spaces = " \f\n\r\t\v"; | |
4769 | ||
4770 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4771 | ||
4772 | return strcspn(*buf, spaces); /* Return token length */ | |
4773 | } | |
4774 | ||
4775 | /* | |
4776 | * Finds the next token in *buf, and if the provided token buffer is | |
4777 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4778 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4779 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4780 | * |
4781 | * Returns the length of the token found (not including the '\0'). | |
4782 | * Return value will be 0 if no token is found, and it will be >= | |
4783 | * token_size if the token would not fit. | |
4784 | * | |
593a9e7b | 4785 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4786 | * found token. Note that this occurs even if the token buffer is |
4787 | * too small to hold it. | |
4788 | */ | |
4789 | static inline size_t copy_token(const char **buf, | |
4790 | char *token, | |
4791 | size_t token_size) | |
4792 | { | |
4793 | size_t len; | |
4794 | ||
4795 | len = next_token(buf); | |
4796 | if (len < token_size) { | |
4797 | memcpy(token, *buf, len); | |
4798 | *(token + len) = '\0'; | |
4799 | } | |
4800 | *buf += len; | |
4801 | ||
4802 | return len; | |
4803 | } | |
4804 | ||
ea3352f4 AE |
4805 | /* |
4806 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4807 | * enough to hold a copy of it, and copies the token into the new | |
4808 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4809 | * that a duplicate buffer is created even for a zero-length token. | |
4810 | * | |
4811 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4812 | * pointer if memory for the duplicate was not available. If | |
4813 | * the lenp argument is a non-null pointer, the length of the token | |
4814 | * (not including the '\0') is returned in *lenp. | |
4815 | * | |
4816 | * If successful, the *buf pointer will be updated to point beyond | |
4817 | * the end of the found token. | |
4818 | * | |
4819 | * Note: uses GFP_KERNEL for allocation. | |
4820 | */ | |
4821 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4822 | { | |
4823 | char *dup; | |
4824 | size_t len; | |
4825 | ||
4826 | len = next_token(buf); | |
4caf35f9 | 4827 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4828 | if (!dup) |
4829 | return NULL; | |
ea3352f4 AE |
4830 | *(dup + len) = '\0'; |
4831 | *buf += len; | |
4832 | ||
4833 | if (lenp) | |
4834 | *lenp = len; | |
4835 | ||
4836 | return dup; | |
4837 | } | |
4838 | ||
a725f65e | 4839 | /* |
859c31df AE |
4840 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4841 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4842 | * and the data written is passed here via a NUL-terminated buffer. | |
4843 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4844 | * |
859c31df AE |
4845 | * The information extracted from these options is recorded in |
4846 | * the other parameters which return dynamically-allocated | |
4847 | * structures: | |
4848 | * ceph_opts | |
4849 | * The address of a pointer that will refer to a ceph options | |
4850 | * structure. Caller must release the returned pointer using | |
4851 | * ceph_destroy_options() when it is no longer needed. | |
4852 | * rbd_opts | |
4853 | * Address of an rbd options pointer. Fully initialized by | |
4854 | * this function; caller must release with kfree(). | |
4855 | * spec | |
4856 | * Address of an rbd image specification pointer. Fully | |
4857 | * initialized by this function based on parsed options. | |
4858 | * Caller must release with rbd_spec_put(). | |
4859 | * | |
4860 | * The options passed take this form: | |
4861 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4862 | * where: | |
4863 | * <mon_addrs> | |
4864 | * A comma-separated list of one or more monitor addresses. | |
4865 | * A monitor address is an ip address, optionally followed | |
4866 | * by a port number (separated by a colon). | |
4867 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4868 | * <options> | |
4869 | * A comma-separated list of ceph and/or rbd options. | |
4870 | * <pool_name> | |
4871 | * The name of the rados pool containing the rbd image. | |
4872 | * <image_name> | |
4873 | * The name of the image in that pool to map. | |
4874 | * <snap_id> | |
4875 | * An optional snapshot id. If provided, the mapping will | |
4876 | * present data from the image at the time that snapshot was | |
4877 | * created. The image head is used if no snapshot id is | |
4878 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4879 | */ |
859c31df | 4880 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4881 | struct ceph_options **ceph_opts, |
859c31df AE |
4882 | struct rbd_options **opts, |
4883 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4884 | { |
d22f76e7 | 4885 | size_t len; |
859c31df | 4886 | char *options; |
0ddebc0c | 4887 | const char *mon_addrs; |
ecb4dc22 | 4888 | char *snap_name; |
0ddebc0c | 4889 | size_t mon_addrs_size; |
859c31df | 4890 | struct rbd_spec *spec = NULL; |
4e9afeba | 4891 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4892 | struct ceph_options *copts; |
dc79b113 | 4893 | int ret; |
e28fff26 AE |
4894 | |
4895 | /* The first four tokens are required */ | |
4896 | ||
7ef3214a | 4897 | len = next_token(&buf); |
4fb5d671 AE |
4898 | if (!len) { |
4899 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4900 | return -EINVAL; | |
4901 | } | |
0ddebc0c | 4902 | mon_addrs = buf; |
f28e565a | 4903 | mon_addrs_size = len + 1; |
7ef3214a | 4904 | buf += len; |
a725f65e | 4905 | |
dc79b113 | 4906 | ret = -EINVAL; |
f28e565a AE |
4907 | options = dup_token(&buf, NULL); |
4908 | if (!options) | |
dc79b113 | 4909 | return -ENOMEM; |
4fb5d671 AE |
4910 | if (!*options) { |
4911 | rbd_warn(NULL, "no options provided"); | |
4912 | goto out_err; | |
4913 | } | |
e28fff26 | 4914 | |
859c31df AE |
4915 | spec = rbd_spec_alloc(); |
4916 | if (!spec) | |
f28e565a | 4917 | goto out_mem; |
859c31df AE |
4918 | |
4919 | spec->pool_name = dup_token(&buf, NULL); | |
4920 | if (!spec->pool_name) | |
4921 | goto out_mem; | |
4fb5d671 AE |
4922 | if (!*spec->pool_name) { |
4923 | rbd_warn(NULL, "no pool name provided"); | |
4924 | goto out_err; | |
4925 | } | |
e28fff26 | 4926 | |
69e7a02f | 4927 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4928 | if (!spec->image_name) |
f28e565a | 4929 | goto out_mem; |
4fb5d671 AE |
4930 | if (!*spec->image_name) { |
4931 | rbd_warn(NULL, "no image name provided"); | |
4932 | goto out_err; | |
4933 | } | |
d4b125e9 | 4934 | |
f28e565a AE |
4935 | /* |
4936 | * Snapshot name is optional; default is to use "-" | |
4937 | * (indicating the head/no snapshot). | |
4938 | */ | |
3feeb894 | 4939 | len = next_token(&buf); |
820a5f3e | 4940 | if (!len) { |
3feeb894 AE |
4941 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4942 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4943 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4944 | ret = -ENAMETOOLONG; |
f28e565a | 4945 | goto out_err; |
849b4260 | 4946 | } |
ecb4dc22 AE |
4947 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4948 | if (!snap_name) | |
f28e565a | 4949 | goto out_mem; |
ecb4dc22 AE |
4950 | *(snap_name + len) = '\0'; |
4951 | spec->snap_name = snap_name; | |
e5c35534 | 4952 | |
0ddebc0c | 4953 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4954 | |
4e9afeba AE |
4955 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4956 | if (!rbd_opts) | |
4957 | goto out_mem; | |
4958 | ||
4959 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4960 | |
859c31df | 4961 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4962 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4963 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4964 | if (IS_ERR(copts)) { |
4965 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4966 | goto out_err; |
4967 | } | |
859c31df AE |
4968 | kfree(options); |
4969 | ||
4970 | *ceph_opts = copts; | |
4e9afeba | 4971 | *opts = rbd_opts; |
859c31df | 4972 | *rbd_spec = spec; |
0ddebc0c | 4973 | |
dc79b113 | 4974 | return 0; |
f28e565a | 4975 | out_mem: |
dc79b113 | 4976 | ret = -ENOMEM; |
d22f76e7 | 4977 | out_err: |
859c31df AE |
4978 | kfree(rbd_opts); |
4979 | rbd_spec_put(spec); | |
f28e565a | 4980 | kfree(options); |
d22f76e7 | 4981 | |
dc79b113 | 4982 | return ret; |
a725f65e AE |
4983 | } |
4984 | ||
30ba1f02 ID |
4985 | /* |
4986 | * Return pool id (>= 0) or a negative error code. | |
4987 | */ | |
4988 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4989 | { | |
4990 | u64 newest_epoch; | |
4991 | unsigned long timeout = rbdc->client->options->mount_timeout * HZ; | |
4992 | int tries = 0; | |
4993 | int ret; | |
4994 | ||
4995 | again: | |
4996 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4997 | if (ret == -ENOENT && tries++ < 1) { | |
4998 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4999 | &newest_epoch); | |
5000 | if (ret < 0) | |
5001 | return ret; | |
5002 | ||
5003 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
5004 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
5005 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
5006 | newest_epoch, timeout); | |
5007 | goto again; | |
5008 | } else { | |
5009 | /* the osdmap we have is new enough */ | |
5010 | return -ENOENT; | |
5011 | } | |
5012 | } | |
5013 | ||
5014 | return ret; | |
5015 | } | |
5016 | ||
589d30e0 AE |
5017 | /* |
5018 | * An rbd format 2 image has a unique identifier, distinct from the | |
5019 | * name given to it by the user. Internally, that identifier is | |
5020 | * what's used to specify the names of objects related to the image. | |
5021 | * | |
5022 | * A special "rbd id" object is used to map an rbd image name to its | |
5023 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5024 | * with the supplied name. | |
5025 | * | |
5026 | * This function will record the given rbd_dev's image_id field if | |
5027 | * it can be determined, and in that case will return 0. If any | |
5028 | * errors occur a negative errno will be returned and the rbd_dev's | |
5029 | * image_id field will be unchanged (and should be NULL). | |
5030 | */ | |
5031 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5032 | { | |
5033 | int ret; | |
5034 | size_t size; | |
5035 | char *object_name; | |
5036 | void *response; | |
c0fba368 | 5037 | char *image_id; |
2f82ee54 | 5038 | |
2c0d0a10 AE |
5039 | /* |
5040 | * When probing a parent image, the image id is already | |
5041 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5042 | * need to fetch the image id again in this case. We |
5043 | * do still need to set the image format though. | |
2c0d0a10 | 5044 | */ |
c0fba368 AE |
5045 | if (rbd_dev->spec->image_id) { |
5046 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5047 | ||
2c0d0a10 | 5048 | return 0; |
c0fba368 | 5049 | } |
2c0d0a10 | 5050 | |
589d30e0 AE |
5051 | /* |
5052 | * First, see if the format 2 image id file exists, and if | |
5053 | * so, get the image's persistent id from it. | |
5054 | */ | |
69e7a02f | 5055 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5056 | object_name = kmalloc(size, GFP_NOIO); |
5057 | if (!object_name) | |
5058 | return -ENOMEM; | |
0d7dbfce | 5059 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5060 | dout("rbd id object name is %s\n", object_name); |
5061 | ||
5062 | /* Response will be an encoded string, which includes a length */ | |
5063 | ||
5064 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5065 | response = kzalloc(size, GFP_NOIO); | |
5066 | if (!response) { | |
5067 | ret = -ENOMEM; | |
5068 | goto out; | |
5069 | } | |
5070 | ||
c0fba368 AE |
5071 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5072 | ||
36be9a76 | 5073 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5074 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5075 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5076 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5077 | if (ret == -ENOENT) { |
5078 | image_id = kstrdup("", GFP_KERNEL); | |
5079 | ret = image_id ? 0 : -ENOMEM; | |
5080 | if (!ret) | |
5081 | rbd_dev->image_format = 1; | |
7dd440c9 | 5082 | } else if (ret >= 0) { |
c0fba368 AE |
5083 | void *p = response; |
5084 | ||
5085 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5086 | NULL, GFP_NOIO); |
461f758a | 5087 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5088 | if (!ret) |
5089 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5090 | } |
5091 | ||
5092 | if (!ret) { | |
5093 | rbd_dev->spec->image_id = image_id; | |
5094 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5095 | } |
5096 | out: | |
5097 | kfree(response); | |
5098 | kfree(object_name); | |
5099 | ||
5100 | return ret; | |
5101 | } | |
5102 | ||
3abef3b3 AE |
5103 | /* |
5104 | * Undo whatever state changes are made by v1 or v2 header info | |
5105 | * call. | |
5106 | */ | |
6fd48b3b AE |
5107 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5108 | { | |
5109 | struct rbd_image_header *header; | |
5110 | ||
392a9dad AE |
5111 | /* Drop parent reference unless it's already been done (or none) */ |
5112 | ||
5113 | if (rbd_dev->parent_overlap) | |
5114 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
5115 | |
5116 | /* Free dynamic fields from the header, then zero it out */ | |
5117 | ||
5118 | header = &rbd_dev->header; | |
812164f8 | 5119 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5120 | kfree(header->snap_sizes); |
5121 | kfree(header->snap_names); | |
5122 | kfree(header->object_prefix); | |
5123 | memset(header, 0, sizeof (*header)); | |
5124 | } | |
5125 | ||
2df3fac7 | 5126 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5127 | { |
5128 | int ret; | |
a30b71b9 | 5129 | |
1e130199 | 5130 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5131 | if (ret) |
b1b5402a AE |
5132 | goto out_err; |
5133 | ||
2df3fac7 AE |
5134 | /* |
5135 | * Get the and check features for the image. Currently the | |
5136 | * features are assumed to never change. | |
5137 | */ | |
b1b5402a | 5138 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5139 | if (ret) |
9d475de5 | 5140 | goto out_err; |
35d489f9 | 5141 | |
cc070d59 AE |
5142 | /* If the image supports fancy striping, get its parameters */ |
5143 | ||
5144 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5145 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5146 | if (ret < 0) | |
5147 | goto out_err; | |
5148 | } | |
2df3fac7 | 5149 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5150 | |
35152979 | 5151 | return 0; |
9d475de5 | 5152 | out_err: |
642a2537 | 5153 | rbd_dev->header.features = 0; |
1e130199 AE |
5154 | kfree(rbd_dev->header.object_prefix); |
5155 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5156 | |
5157 | return ret; | |
a30b71b9 AE |
5158 | } |
5159 | ||
124afba2 | 5160 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 5161 | { |
2f82ee54 | 5162 | struct rbd_device *parent = NULL; |
124afba2 AE |
5163 | struct rbd_spec *parent_spec; |
5164 | struct rbd_client *rbdc; | |
5165 | int ret; | |
5166 | ||
5167 | if (!rbd_dev->parent_spec) | |
5168 | return 0; | |
5169 | /* | |
5170 | * We need to pass a reference to the client and the parent | |
5171 | * spec when creating the parent rbd_dev. Images related by | |
5172 | * parent/child relationships always share both. | |
5173 | */ | |
5174 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
5175 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
5176 | ||
5177 | ret = -ENOMEM; | |
5178 | parent = rbd_dev_create(rbdc, parent_spec); | |
5179 | if (!parent) | |
5180 | goto out_err; | |
5181 | ||
1f3ef788 | 5182 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
5183 | if (ret < 0) |
5184 | goto out_err; | |
5185 | rbd_dev->parent = parent; | |
a2acd00e | 5186 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
5187 | |
5188 | return 0; | |
5189 | out_err: | |
5190 | if (parent) { | |
fb65d228 | 5191 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
5192 | kfree(rbd_dev->header_name); |
5193 | rbd_dev_destroy(parent); | |
5194 | } else { | |
5195 | rbd_put_client(rbdc); | |
5196 | rbd_spec_put(parent_spec); | |
5197 | } | |
5198 | ||
5199 | return ret; | |
5200 | } | |
5201 | ||
200a6a8b | 5202 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5203 | { |
83a06263 | 5204 | int ret; |
d1cf5788 | 5205 | |
f8a22fc2 ID |
5206 | /* Get an id and fill in device name. */ |
5207 | ||
5208 | ret = rbd_dev_id_get(rbd_dev); | |
5209 | if (ret) | |
5210 | return ret; | |
83a06263 | 5211 | |
83a06263 AE |
5212 | BUILD_BUG_ON(DEV_NAME_LEN |
5213 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5214 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5215 | ||
9b60e70b | 5216 | /* Record our major and minor device numbers. */ |
83a06263 | 5217 | |
9b60e70b ID |
5218 | if (!single_major) { |
5219 | ret = register_blkdev(0, rbd_dev->name); | |
5220 | if (ret < 0) | |
5221 | goto err_out_id; | |
5222 | ||
5223 | rbd_dev->major = ret; | |
5224 | rbd_dev->minor = 0; | |
5225 | } else { | |
5226 | rbd_dev->major = rbd_major; | |
5227 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5228 | } | |
83a06263 AE |
5229 | |
5230 | /* Set up the blkdev mapping. */ | |
5231 | ||
5232 | ret = rbd_init_disk(rbd_dev); | |
5233 | if (ret) | |
5234 | goto err_out_blkdev; | |
5235 | ||
f35a4dee | 5236 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5237 | if (ret) |
5238 | goto err_out_disk; | |
bc1ecc65 | 5239 | |
f35a4dee | 5240 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5241 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 5242 | |
58d1362b | 5243 | rbd_dev->rq_wq = alloc_workqueue("%s", 0, 0, rbd_dev->disk->disk_name); |
255939e7 WY |
5244 | if (!rbd_dev->rq_wq) { |
5245 | ret = -ENOMEM; | |
bc1ecc65 | 5246 | goto err_out_mapping; |
255939e7 | 5247 | } |
bc1ecc65 | 5248 | |
f35a4dee AE |
5249 | ret = rbd_bus_add_dev(rbd_dev); |
5250 | if (ret) | |
bc1ecc65 | 5251 | goto err_out_workqueue; |
83a06263 | 5252 | |
83a06263 AE |
5253 | /* Everything's ready. Announce the disk to the world. */ |
5254 | ||
129b79d4 | 5255 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5256 | add_disk(rbd_dev->disk); |
5257 | ||
5258 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5259 | (unsigned long long) rbd_dev->mapping.size); | |
5260 | ||
5261 | return ret; | |
2f82ee54 | 5262 | |
bc1ecc65 ID |
5263 | err_out_workqueue: |
5264 | destroy_workqueue(rbd_dev->rq_wq); | |
5265 | rbd_dev->rq_wq = NULL; | |
f35a4dee AE |
5266 | err_out_mapping: |
5267 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5268 | err_out_disk: |
5269 | rbd_free_disk(rbd_dev); | |
5270 | err_out_blkdev: | |
9b60e70b ID |
5271 | if (!single_major) |
5272 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5273 | err_out_id: |
5274 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5275 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5276 | |
5277 | return ret; | |
5278 | } | |
5279 | ||
332bb12d AE |
5280 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5281 | { | |
5282 | struct rbd_spec *spec = rbd_dev->spec; | |
5283 | size_t size; | |
5284 | ||
5285 | /* Record the header object name for this rbd image. */ | |
5286 | ||
5287 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5288 | ||
5289 | if (rbd_dev->image_format == 1) | |
5290 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5291 | else | |
5292 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5293 | ||
5294 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5295 | if (!rbd_dev->header_name) | |
5296 | return -ENOMEM; | |
5297 | ||
5298 | if (rbd_dev->image_format == 1) | |
5299 | sprintf(rbd_dev->header_name, "%s%s", | |
5300 | spec->image_name, RBD_SUFFIX); | |
5301 | else | |
5302 | sprintf(rbd_dev->header_name, "%s%s", | |
5303 | RBD_HEADER_PREFIX, spec->image_id); | |
5304 | return 0; | |
5305 | } | |
5306 | ||
200a6a8b AE |
5307 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5308 | { | |
6fd48b3b | 5309 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5310 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5311 | rbd_dev->header_name = NULL; |
5312 | rbd_dev->image_format = 0; | |
5313 | kfree(rbd_dev->spec->image_id); | |
5314 | rbd_dev->spec->image_id = NULL; | |
5315 | ||
200a6a8b AE |
5316 | rbd_dev_destroy(rbd_dev); |
5317 | } | |
5318 | ||
a30b71b9 AE |
5319 | /* |
5320 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5321 | * device. If this image is the one being mapped (i.e., not a |
5322 | * parent), initiate a watch on its header object before using that | |
5323 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5324 | */ |
1f3ef788 | 5325 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
5326 | { |
5327 | int ret; | |
5328 | ||
5329 | /* | |
3abef3b3 AE |
5330 | * Get the id from the image id object. Unless there's an |
5331 | * error, rbd_dev->spec->image_id will be filled in with | |
5332 | * a dynamically-allocated string, and rbd_dev->image_format | |
5333 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5334 | */ |
5335 | ret = rbd_dev_image_id(rbd_dev); | |
5336 | if (ret) | |
c0fba368 | 5337 | return ret; |
c0fba368 | 5338 | |
332bb12d AE |
5339 | ret = rbd_dev_header_name(rbd_dev); |
5340 | if (ret) | |
5341 | goto err_out_format; | |
5342 | ||
1f3ef788 | 5343 | if (mapping) { |
fca27065 | 5344 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1f3ef788 AE |
5345 | if (ret) |
5346 | goto out_header_name; | |
5347 | } | |
b644de2b | 5348 | |
a720ae09 | 5349 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5350 | if (ret) |
b644de2b | 5351 | goto err_out_watch; |
83a06263 | 5352 | |
04077599 ID |
5353 | /* |
5354 | * If this image is the one being mapped, we have pool name and | |
5355 | * id, image name and id, and snap name - need to fill snap id. | |
5356 | * Otherwise this is a parent image, identified by pool, image | |
5357 | * and snap ids - need to fill in names for those ids. | |
5358 | */ | |
5359 | if (mapping) | |
5360 | ret = rbd_spec_fill_snap_id(rbd_dev); | |
5361 | else | |
5362 | ret = rbd_spec_fill_names(rbd_dev); | |
9bb81c9b | 5363 | if (ret) |
33dca39f | 5364 | goto err_out_probe; |
9bb81c9b | 5365 | |
e8f59b59 ID |
5366 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5367 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5368 | if (ret) | |
5369 | goto err_out_probe; | |
5370 | ||
5371 | /* | |
5372 | * Need to warn users if this image is the one being | |
5373 | * mapped and has a parent. | |
5374 | */ | |
5375 | if (mapping && rbd_dev->parent_spec) | |
5376 | rbd_warn(rbd_dev, | |
5377 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5378 | } | |
5379 | ||
9bb81c9b | 5380 | ret = rbd_dev_probe_parent(rbd_dev); |
30d60ba2 AE |
5381 | if (ret) |
5382 | goto err_out_probe; | |
5383 | ||
5384 | dout("discovered format %u image, header name is %s\n", | |
5385 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5386 | return 0; |
e8f59b59 | 5387 | |
6fd48b3b AE |
5388 | err_out_probe: |
5389 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5390 | err_out_watch: |
fca27065 ID |
5391 | if (mapping) |
5392 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5393 | out_header_name: |
5394 | kfree(rbd_dev->header_name); | |
5395 | rbd_dev->header_name = NULL; | |
5396 | err_out_format: | |
5397 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5398 | kfree(rbd_dev->spec->image_id); |
5399 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5400 | return ret; |
5401 | } | |
5402 | ||
9b60e70b ID |
5403 | static ssize_t do_rbd_add(struct bus_type *bus, |
5404 | const char *buf, | |
5405 | size_t count) | |
602adf40 | 5406 | { |
cb8627c7 | 5407 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5408 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5409 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5410 | struct rbd_spec *spec = NULL; |
9d3997fd | 5411 | struct rbd_client *rbdc; |
51344a38 | 5412 | bool read_only; |
27cc2594 | 5413 | int rc = -ENOMEM; |
602adf40 YS |
5414 | |
5415 | if (!try_module_get(THIS_MODULE)) | |
5416 | return -ENODEV; | |
5417 | ||
602adf40 | 5418 | /* parse add command */ |
859c31df | 5419 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5420 | if (rc < 0) |
bd4ba655 | 5421 | goto err_out_module; |
51344a38 AE |
5422 | read_only = rbd_opts->read_only; |
5423 | kfree(rbd_opts); | |
5424 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5425 | |
9d3997fd AE |
5426 | rbdc = rbd_get_client(ceph_opts); |
5427 | if (IS_ERR(rbdc)) { | |
5428 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5429 | goto err_out_args; |
9d3997fd | 5430 | } |
602adf40 | 5431 | |
602adf40 | 5432 | /* pick the pool */ |
30ba1f02 | 5433 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
602adf40 YS |
5434 | if (rc < 0) |
5435 | goto err_out_client; | |
c0cd10db | 5436 | spec->pool_id = (u64)rc; |
859c31df | 5437 | |
0903e875 AE |
5438 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5439 | ||
c0cd10db | 5440 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5441 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5442 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5443 | rc = -EIO; |
5444 | goto err_out_client; | |
5445 | } | |
5446 | ||
c53d5893 | 5447 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5448 | if (!rbd_dev) |
5449 | goto err_out_client; | |
c53d5893 AE |
5450 | rbdc = NULL; /* rbd_dev now owns this */ |
5451 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5452 | |
1f3ef788 | 5453 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5454 | if (rc < 0) |
c53d5893 | 5455 | goto err_out_rbd_dev; |
05fd6f6f | 5456 | |
7ce4eef7 AE |
5457 | /* If we are mapping a snapshot it must be marked read-only */ |
5458 | ||
5459 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5460 | read_only = true; | |
5461 | rbd_dev->mapping.read_only = read_only; | |
5462 | ||
b536f69a | 5463 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5464 | if (rc) { |
e37180c0 ID |
5465 | /* |
5466 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5467 | * rbd_dev_image_release() without refactoring, see | |
5468 | * commit 1f3ef78861ac. | |
5469 | */ | |
5470 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5471 | rbd_dev_image_release(rbd_dev); |
5472 | goto err_out_module; | |
5473 | } | |
5474 | ||
5475 | return count; | |
b536f69a | 5476 | |
c53d5893 AE |
5477 | err_out_rbd_dev: |
5478 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5479 | err_out_client: |
9d3997fd | 5480 | rbd_put_client(rbdc); |
0ddebc0c | 5481 | err_out_args: |
859c31df | 5482 | rbd_spec_put(spec); |
bd4ba655 AE |
5483 | err_out_module: |
5484 | module_put(THIS_MODULE); | |
27cc2594 | 5485 | |
602adf40 | 5486 | dout("Error adding device %s\n", buf); |
27cc2594 | 5487 | |
c0cd10db | 5488 | return (ssize_t)rc; |
602adf40 YS |
5489 | } |
5490 | ||
9b60e70b ID |
5491 | static ssize_t rbd_add(struct bus_type *bus, |
5492 | const char *buf, | |
5493 | size_t count) | |
5494 | { | |
5495 | if (single_major) | |
5496 | return -EINVAL; | |
5497 | ||
5498 | return do_rbd_add(bus, buf, count); | |
5499 | } | |
5500 | ||
5501 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5502 | const char *buf, | |
5503 | size_t count) | |
5504 | { | |
5505 | return do_rbd_add(bus, buf, count); | |
5506 | } | |
5507 | ||
200a6a8b | 5508 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5509 | { |
593a9e7b | 5510 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5511 | |
bc1ecc65 | 5512 | destroy_workqueue(rbd_dev->rq_wq); |
602adf40 | 5513 | rbd_free_disk(rbd_dev); |
200a6a8b | 5514 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5515 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5516 | if (!single_major) |
5517 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5518 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5519 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5520 | } |
5521 | ||
05a46afd AE |
5522 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5523 | { | |
ad945fc1 | 5524 | while (rbd_dev->parent) { |
05a46afd AE |
5525 | struct rbd_device *first = rbd_dev; |
5526 | struct rbd_device *second = first->parent; | |
5527 | struct rbd_device *third; | |
5528 | ||
5529 | /* | |
5530 | * Follow to the parent with no grandparent and | |
5531 | * remove it. | |
5532 | */ | |
5533 | while (second && (third = second->parent)) { | |
5534 | first = second; | |
5535 | second = third; | |
5536 | } | |
ad945fc1 | 5537 | rbd_assert(second); |
8ad42cd0 | 5538 | rbd_dev_image_release(second); |
ad945fc1 AE |
5539 | first->parent = NULL; |
5540 | first->parent_overlap = 0; | |
5541 | ||
5542 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5543 | rbd_spec_put(first->parent_spec); |
5544 | first->parent_spec = NULL; | |
05a46afd AE |
5545 | } |
5546 | } | |
5547 | ||
9b60e70b ID |
5548 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5549 | const char *buf, | |
5550 | size_t count) | |
602adf40 YS |
5551 | { |
5552 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5553 | struct list_head *tmp; |
5554 | int dev_id; | |
602adf40 | 5555 | unsigned long ul; |
82a442d2 | 5556 | bool already = false; |
0d8189e1 | 5557 | int ret; |
602adf40 | 5558 | |
bb8e0e84 | 5559 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5560 | if (ret) |
5561 | return ret; | |
602adf40 YS |
5562 | |
5563 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5564 | dev_id = (int)ul; |
5565 | if (dev_id != ul) | |
602adf40 YS |
5566 | return -EINVAL; |
5567 | ||
751cc0e3 AE |
5568 | ret = -ENOENT; |
5569 | spin_lock(&rbd_dev_list_lock); | |
5570 | list_for_each(tmp, &rbd_dev_list) { | |
5571 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5572 | if (rbd_dev->dev_id == dev_id) { | |
5573 | ret = 0; | |
5574 | break; | |
5575 | } | |
42382b70 | 5576 | } |
751cc0e3 AE |
5577 | if (!ret) { |
5578 | spin_lock_irq(&rbd_dev->lock); | |
5579 | if (rbd_dev->open_count) | |
5580 | ret = -EBUSY; | |
5581 | else | |
82a442d2 AE |
5582 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5583 | &rbd_dev->flags); | |
751cc0e3 AE |
5584 | spin_unlock_irq(&rbd_dev->lock); |
5585 | } | |
5586 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5587 | if (ret < 0 || already) |
1ba0f1e7 | 5588 | return ret; |
751cc0e3 | 5589 | |
fca27065 | 5590 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5591 | /* |
5592 | * flush remaining watch callbacks - these must be complete | |
5593 | * before the osd_client is shutdown | |
5594 | */ | |
5595 | dout("%s: flushing notifies", __func__); | |
5596 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5597 | |
9875201e JD |
5598 | /* |
5599 | * Don't free anything from rbd_dev->disk until after all | |
5600 | * notifies are completely processed. Otherwise | |
5601 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5602 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5603 | */ | |
5604 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5605 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5606 | module_put(THIS_MODULE); |
aafb230e | 5607 | |
1ba0f1e7 | 5608 | return count; |
602adf40 YS |
5609 | } |
5610 | ||
9b60e70b ID |
5611 | static ssize_t rbd_remove(struct bus_type *bus, |
5612 | const char *buf, | |
5613 | size_t count) | |
5614 | { | |
5615 | if (single_major) | |
5616 | return -EINVAL; | |
5617 | ||
5618 | return do_rbd_remove(bus, buf, count); | |
5619 | } | |
5620 | ||
5621 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5622 | const char *buf, | |
5623 | size_t count) | |
5624 | { | |
5625 | return do_rbd_remove(bus, buf, count); | |
5626 | } | |
5627 | ||
602adf40 YS |
5628 | /* |
5629 | * create control files in sysfs | |
dfc5606d | 5630 | * /sys/bus/rbd/... |
602adf40 YS |
5631 | */ |
5632 | static int rbd_sysfs_init(void) | |
5633 | { | |
dfc5606d | 5634 | int ret; |
602adf40 | 5635 | |
fed4c143 | 5636 | ret = device_register(&rbd_root_dev); |
21079786 | 5637 | if (ret < 0) |
dfc5606d | 5638 | return ret; |
602adf40 | 5639 | |
fed4c143 AE |
5640 | ret = bus_register(&rbd_bus_type); |
5641 | if (ret < 0) | |
5642 | device_unregister(&rbd_root_dev); | |
602adf40 | 5643 | |
602adf40 YS |
5644 | return ret; |
5645 | } | |
5646 | ||
5647 | static void rbd_sysfs_cleanup(void) | |
5648 | { | |
dfc5606d | 5649 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5650 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5651 | } |
5652 | ||
1c2a9dfe AE |
5653 | static int rbd_slab_init(void) |
5654 | { | |
5655 | rbd_assert(!rbd_img_request_cache); | |
5656 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5657 | sizeof (struct rbd_img_request), | |
5658 | __alignof__(struct rbd_img_request), | |
5659 | 0, NULL); | |
868311b1 AE |
5660 | if (!rbd_img_request_cache) |
5661 | return -ENOMEM; | |
5662 | ||
5663 | rbd_assert(!rbd_obj_request_cache); | |
5664 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5665 | sizeof (struct rbd_obj_request), | |
5666 | __alignof__(struct rbd_obj_request), | |
5667 | 0, NULL); | |
78c2a44a AE |
5668 | if (!rbd_obj_request_cache) |
5669 | goto out_err; | |
5670 | ||
5671 | rbd_assert(!rbd_segment_name_cache); | |
5672 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5673 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5674 | if (rbd_segment_name_cache) |
1c2a9dfe | 5675 | return 0; |
78c2a44a AE |
5676 | out_err: |
5677 | if (rbd_obj_request_cache) { | |
5678 | kmem_cache_destroy(rbd_obj_request_cache); | |
5679 | rbd_obj_request_cache = NULL; | |
5680 | } | |
1c2a9dfe | 5681 | |
868311b1 AE |
5682 | kmem_cache_destroy(rbd_img_request_cache); |
5683 | rbd_img_request_cache = NULL; | |
5684 | ||
1c2a9dfe AE |
5685 | return -ENOMEM; |
5686 | } | |
5687 | ||
5688 | static void rbd_slab_exit(void) | |
5689 | { | |
78c2a44a AE |
5690 | rbd_assert(rbd_segment_name_cache); |
5691 | kmem_cache_destroy(rbd_segment_name_cache); | |
5692 | rbd_segment_name_cache = NULL; | |
5693 | ||
868311b1 AE |
5694 | rbd_assert(rbd_obj_request_cache); |
5695 | kmem_cache_destroy(rbd_obj_request_cache); | |
5696 | rbd_obj_request_cache = NULL; | |
5697 | ||
1c2a9dfe AE |
5698 | rbd_assert(rbd_img_request_cache); |
5699 | kmem_cache_destroy(rbd_img_request_cache); | |
5700 | rbd_img_request_cache = NULL; | |
5701 | } | |
5702 | ||
cc344fa1 | 5703 | static int __init rbd_init(void) |
602adf40 YS |
5704 | { |
5705 | int rc; | |
5706 | ||
1e32d34c AE |
5707 | if (!libceph_compatible(NULL)) { |
5708 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5709 | return -EINVAL; |
5710 | } | |
e1b4d96d | 5711 | |
1c2a9dfe | 5712 | rc = rbd_slab_init(); |
602adf40 YS |
5713 | if (rc) |
5714 | return rc; | |
e1b4d96d | 5715 | |
9b60e70b ID |
5716 | if (single_major) { |
5717 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5718 | if (rbd_major < 0) { | |
5719 | rc = rbd_major; | |
5720 | goto err_out_slab; | |
5721 | } | |
5722 | } | |
5723 | ||
1c2a9dfe AE |
5724 | rc = rbd_sysfs_init(); |
5725 | if (rc) | |
9b60e70b ID |
5726 | goto err_out_blkdev; |
5727 | ||
5728 | if (single_major) | |
5729 | pr_info("loaded (major %d)\n", rbd_major); | |
5730 | else | |
5731 | pr_info("loaded\n"); | |
1c2a9dfe | 5732 | |
e1b4d96d ID |
5733 | return 0; |
5734 | ||
9b60e70b ID |
5735 | err_out_blkdev: |
5736 | if (single_major) | |
5737 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
e1b4d96d ID |
5738 | err_out_slab: |
5739 | rbd_slab_exit(); | |
1c2a9dfe | 5740 | return rc; |
602adf40 YS |
5741 | } |
5742 | ||
cc344fa1 | 5743 | static void __exit rbd_exit(void) |
602adf40 | 5744 | { |
ffe312cf | 5745 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5746 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5747 | if (single_major) |
5748 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
1c2a9dfe | 5749 | rbd_slab_exit(); |
602adf40 YS |
5750 | } |
5751 | ||
5752 | module_init(rbd_init); | |
5753 | module_exit(rbd_exit); | |
5754 | ||
d552c619 | 5755 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5756 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5757 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5758 | /* following authorship retained from original osdblk.c */ |
5759 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5760 | ||
90da258b | 5761 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5762 | MODULE_LICENSE("GPL"); |