]>
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> | |
7ad18afa | 41 | #include <linux/blk-mq.h> |
602adf40 YS |
42 | #include <linux/fs.h> |
43 | #include <linux/blkdev.h> | |
1c2a9dfe | 44 | #include <linux/slab.h> |
f8a22fc2 | 45 | #include <linux/idr.h> |
bc1ecc65 | 46 | #include <linux/workqueue.h> |
602adf40 YS |
47 | |
48 | #include "rbd_types.h" | |
49 | ||
aafb230e AE |
50 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
51 | ||
593a9e7b AE |
52 | /* |
53 | * The basic unit of block I/O is a sector. It is interpreted in a | |
54 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
55 | * universally 512 bytes. These symbols are just slightly more | |
56 | * meaningful than the bare numbers they represent. | |
57 | */ | |
58 | #define SECTOR_SHIFT 9 | |
59 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
60 | ||
a2acd00e AE |
61 | /* |
62 | * Increment the given counter and return its updated value. | |
63 | * If the counter is already 0 it will not be incremented. | |
64 | * If the counter is already at its maximum value returns | |
65 | * -EINVAL without updating it. | |
66 | */ | |
67 | static int atomic_inc_return_safe(atomic_t *v) | |
68 | { | |
69 | unsigned int counter; | |
70 | ||
71 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
72 | if (counter <= (unsigned int)INT_MAX) | |
73 | return (int)counter; | |
74 | ||
75 | atomic_dec(v); | |
76 | ||
77 | return -EINVAL; | |
78 | } | |
79 | ||
80 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
81 | static int atomic_dec_return_safe(atomic_t *v) | |
82 | { | |
83 | int counter; | |
84 | ||
85 | counter = atomic_dec_return(v); | |
86 | if (counter >= 0) | |
87 | return counter; | |
88 | ||
89 | atomic_inc(v); | |
90 | ||
91 | return -EINVAL; | |
92 | } | |
93 | ||
f0f8cef5 | 94 | #define RBD_DRV_NAME "rbd" |
602adf40 | 95 | |
7e513d43 ID |
96 | #define RBD_MINORS_PER_MAJOR 256 |
97 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 98 | |
6d69bb53 ID |
99 | #define RBD_MAX_PARENT_CHAIN_LEN 16 |
100 | ||
d4b125e9 AE |
101 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
102 | #define RBD_MAX_SNAP_NAME_LEN \ | |
103 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
104 | ||
35d489f9 | 105 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
106 | |
107 | #define RBD_SNAP_HEAD_NAME "-" | |
108 | ||
9682fc6d AE |
109 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
110 | ||
9e15b77d AE |
111 | /* This allows a single page to hold an image name sent by OSD */ |
112 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 113 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 114 | |
1e130199 | 115 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 116 | |
d889140c AE |
117 | /* Feature bits */ |
118 | ||
5cbf6f12 AE |
119 | #define RBD_FEATURE_LAYERING (1<<0) |
120 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
121 | #define RBD_FEATURES_ALL \ | |
122 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
123 | |
124 | /* Features supported by this (client software) implementation. */ | |
125 | ||
770eba6e | 126 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 127 | |
81a89793 AE |
128 | /* |
129 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
130 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
131 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
132 | * enough to hold all possible device names. | |
133 | */ | |
602adf40 | 134 | #define DEV_NAME_LEN 32 |
81a89793 | 135 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
136 | |
137 | /* | |
138 | * block device image metadata (in-memory version) | |
139 | */ | |
140 | struct rbd_image_header { | |
f35a4dee | 141 | /* These six fields never change for a given rbd image */ |
849b4260 | 142 | char *object_prefix; |
602adf40 YS |
143 | __u8 obj_order; |
144 | __u8 crypt_type; | |
145 | __u8 comp_type; | |
f35a4dee AE |
146 | u64 stripe_unit; |
147 | u64 stripe_count; | |
148 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 149 | |
f84344f3 AE |
150 | /* The remaining fields need to be updated occasionally */ |
151 | u64 image_size; | |
152 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
153 | char *snap_names; /* format 1 only */ |
154 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
155 | }; |
156 | ||
0d7dbfce AE |
157 | /* |
158 | * An rbd image specification. | |
159 | * | |
160 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
161 | * identify an image. Each rbd_dev structure includes a pointer to |
162 | * an rbd_spec structure that encapsulates this identity. | |
163 | * | |
164 | * Each of the id's in an rbd_spec has an associated name. For a | |
165 | * user-mapped image, the names are supplied and the id's associated | |
166 | * with them are looked up. For a layered image, a parent image is | |
167 | * defined by the tuple, and the names are looked up. | |
168 | * | |
169 | * An rbd_dev structure contains a parent_spec pointer which is | |
170 | * non-null if the image it represents is a child in a layered | |
171 | * image. This pointer will refer to the rbd_spec structure used | |
172 | * by the parent rbd_dev for its own identity (i.e., the structure | |
173 | * is shared between the parent and child). | |
174 | * | |
175 | * Since these structures are populated once, during the discovery | |
176 | * phase of image construction, they are effectively immutable so | |
177 | * we make no effort to synchronize access to them. | |
178 | * | |
179 | * Note that code herein does not assume the image name is known (it | |
180 | * could be a null pointer). | |
0d7dbfce AE |
181 | */ |
182 | struct rbd_spec { | |
183 | u64 pool_id; | |
ecb4dc22 | 184 | const char *pool_name; |
0d7dbfce | 185 | |
ecb4dc22 AE |
186 | const char *image_id; |
187 | const char *image_name; | |
0d7dbfce AE |
188 | |
189 | u64 snap_id; | |
ecb4dc22 | 190 | const char *snap_name; |
0d7dbfce AE |
191 | |
192 | struct kref kref; | |
193 | }; | |
194 | ||
602adf40 | 195 | /* |
f0f8cef5 | 196 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
197 | */ |
198 | struct rbd_client { | |
199 | struct ceph_client *client; | |
200 | struct kref kref; | |
201 | struct list_head node; | |
202 | }; | |
203 | ||
bf0d5f50 AE |
204 | struct rbd_img_request; |
205 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
206 | ||
207 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
208 | ||
209 | struct rbd_obj_request; | |
210 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
211 | ||
9969ebc5 AE |
212 | enum obj_request_type { |
213 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
214 | }; | |
bf0d5f50 | 215 | |
6d2940c8 GZ |
216 | enum obj_operation_type { |
217 | OBJ_OP_WRITE, | |
218 | OBJ_OP_READ, | |
90e98c52 | 219 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
220 | }; |
221 | ||
926f9b3f AE |
222 | enum obj_req_flags { |
223 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 224 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
225 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
226 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
227 | }; |
228 | ||
bf0d5f50 AE |
229 | struct rbd_obj_request { |
230 | const char *object_name; | |
231 | u64 offset; /* object start byte */ | |
232 | u64 length; /* bytes from offset */ | |
926f9b3f | 233 | unsigned long flags; |
bf0d5f50 | 234 | |
c5b5ef6c AE |
235 | /* |
236 | * An object request associated with an image will have its | |
237 | * img_data flag set; a standalone object request will not. | |
238 | * | |
239 | * A standalone object request will have which == BAD_WHICH | |
240 | * and a null obj_request pointer. | |
241 | * | |
242 | * An object request initiated in support of a layered image | |
243 | * object (to check for its existence before a write) will | |
244 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
245 | * | |
246 | * Finally, an object request for rbd image data will have | |
247 | * which != BAD_WHICH, and will have a non-null img_request | |
248 | * pointer. The value of which will be in the range | |
249 | * 0..(img_request->obj_request_count-1). | |
250 | */ | |
251 | union { | |
252 | struct rbd_obj_request *obj_request; /* STAT op */ | |
253 | struct { | |
254 | struct rbd_img_request *img_request; | |
255 | u64 img_offset; | |
256 | /* links for img_request->obj_requests list */ | |
257 | struct list_head links; | |
258 | }; | |
259 | }; | |
bf0d5f50 AE |
260 | u32 which; /* posn image request list */ |
261 | ||
262 | enum obj_request_type type; | |
788e2df3 AE |
263 | union { |
264 | struct bio *bio_list; | |
265 | struct { | |
266 | struct page **pages; | |
267 | u32 page_count; | |
268 | }; | |
269 | }; | |
0eefd470 | 270 | struct page **copyup_pages; |
ebda6408 | 271 | u32 copyup_page_count; |
bf0d5f50 AE |
272 | |
273 | struct ceph_osd_request *osd_req; | |
274 | ||
275 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 276 | int result; |
bf0d5f50 AE |
277 | |
278 | rbd_obj_callback_t callback; | |
788e2df3 | 279 | struct completion completion; |
bf0d5f50 AE |
280 | |
281 | struct kref kref; | |
282 | }; | |
283 | ||
0c425248 | 284 | enum img_req_flags { |
9849e986 AE |
285 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
286 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 287 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 288 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
289 | }; |
290 | ||
bf0d5f50 | 291 | struct rbd_img_request { |
bf0d5f50 AE |
292 | struct rbd_device *rbd_dev; |
293 | u64 offset; /* starting image byte offset */ | |
294 | u64 length; /* byte count from offset */ | |
0c425248 | 295 | unsigned long flags; |
bf0d5f50 | 296 | union { |
9849e986 | 297 | u64 snap_id; /* for reads */ |
bf0d5f50 | 298 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
299 | }; |
300 | union { | |
301 | struct request *rq; /* block request */ | |
302 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 303 | }; |
3d7efd18 | 304 | struct page **copyup_pages; |
ebda6408 | 305 | u32 copyup_page_count; |
bf0d5f50 AE |
306 | spinlock_t completion_lock;/* protects next_completion */ |
307 | u32 next_completion; | |
308 | rbd_img_callback_t callback; | |
55f27e09 | 309 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 310 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
311 | |
312 | u32 obj_request_count; | |
313 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
314 | ||
315 | struct kref kref; | |
316 | }; | |
317 | ||
318 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 319 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 320 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 321 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 322 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 323 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 324 | |
f84344f3 | 325 | struct rbd_mapping { |
99c1f08f | 326 | u64 size; |
34b13184 | 327 | u64 features; |
f84344f3 AE |
328 | bool read_only; |
329 | }; | |
330 | ||
602adf40 YS |
331 | /* |
332 | * a single device | |
333 | */ | |
334 | struct rbd_device { | |
de71a297 | 335 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
336 | |
337 | int major; /* blkdev assigned major */ | |
dd82fff1 | 338 | int minor; |
602adf40 | 339 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 340 | |
a30b71b9 | 341 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
342 | struct rbd_client *rbd_client; |
343 | ||
344 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
345 | ||
b82d167b | 346 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
347 | |
348 | struct rbd_image_header header; | |
b82d167b | 349 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 350 | struct rbd_spec *spec; |
d147543d | 351 | struct rbd_options *opts; |
602adf40 | 352 | |
0d7dbfce | 353 | char *header_name; |
971f839a | 354 | |
0903e875 AE |
355 | struct ceph_file_layout layout; |
356 | ||
59c2be1e | 357 | struct ceph_osd_event *watch_event; |
975241af | 358 | struct rbd_obj_request *watch_request; |
59c2be1e | 359 | |
86b00e0d AE |
360 | struct rbd_spec *parent_spec; |
361 | u64 parent_overlap; | |
a2acd00e | 362 | atomic_t parent_ref; |
2f82ee54 | 363 | struct rbd_device *parent; |
86b00e0d | 364 | |
7ad18afa CH |
365 | /* Block layer tags. */ |
366 | struct blk_mq_tag_set tag_set; | |
367 | ||
c666601a JD |
368 | /* protects updating the header */ |
369 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
370 | |
371 | struct rbd_mapping mapping; | |
602adf40 YS |
372 | |
373 | struct list_head node; | |
dfc5606d | 374 | |
dfc5606d YS |
375 | /* sysfs related */ |
376 | struct device dev; | |
b82d167b | 377 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
378 | }; |
379 | ||
b82d167b AE |
380 | /* |
381 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
382 | * rbd_dev->lock is used to protect access. | |
383 | * | |
384 | * Currently, only the "removing" flag (which is coupled with the | |
385 | * "open_count" field) requires atomic access. | |
386 | */ | |
6d292906 AE |
387 | enum rbd_dev_flags { |
388 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 389 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
390 | }; |
391 | ||
cfbf6377 | 392 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 393 | |
602adf40 | 394 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
395 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
396 | ||
432b8587 AE |
397 | static LIST_HEAD(rbd_client_list); /* clients */ |
398 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 399 | |
78c2a44a AE |
400 | /* Slab caches for frequently-allocated structures */ |
401 | ||
1c2a9dfe | 402 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 403 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 404 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 405 | |
9b60e70b | 406 | static int rbd_major; |
f8a22fc2 ID |
407 | static DEFINE_IDA(rbd_dev_id_ida); |
408 | ||
f5ee37bd ID |
409 | static struct workqueue_struct *rbd_wq; |
410 | ||
9b60e70b ID |
411 | /* |
412 | * Default to false for now, as single-major requires >= 0.75 version of | |
413 | * userspace rbd utility. | |
414 | */ | |
415 | static bool single_major = false; | |
416 | module_param(single_major, bool, S_IRUGO); | |
417 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
418 | ||
3d7efd18 AE |
419 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
420 | ||
200a6a8b | 421 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 422 | |
f0f8cef5 AE |
423 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
424 | size_t count); | |
425 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
426 | size_t count); | |
9b60e70b ID |
427 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
428 | size_t count); | |
429 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
430 | size_t count); | |
6d69bb53 | 431 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
a2acd00e | 432 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 433 | |
9b60e70b ID |
434 | static int rbd_dev_id_to_minor(int dev_id) |
435 | { | |
7e513d43 | 436 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
437 | } |
438 | ||
439 | static int minor_to_rbd_dev_id(int minor) | |
440 | { | |
7e513d43 | 441 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
442 | } |
443 | ||
b15a21dd GKH |
444 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
445 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
446 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
447 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
448 | |
449 | static struct attribute *rbd_bus_attrs[] = { | |
450 | &bus_attr_add.attr, | |
451 | &bus_attr_remove.attr, | |
9b60e70b ID |
452 | &bus_attr_add_single_major.attr, |
453 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 454 | NULL, |
f0f8cef5 | 455 | }; |
92c76dc0 ID |
456 | |
457 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
458 | struct attribute *attr, int index) | |
459 | { | |
9b60e70b ID |
460 | if (!single_major && |
461 | (attr == &bus_attr_add_single_major.attr || | |
462 | attr == &bus_attr_remove_single_major.attr)) | |
463 | return 0; | |
464 | ||
92c76dc0 ID |
465 | return attr->mode; |
466 | } | |
467 | ||
468 | static const struct attribute_group rbd_bus_group = { | |
469 | .attrs = rbd_bus_attrs, | |
470 | .is_visible = rbd_bus_is_visible, | |
471 | }; | |
472 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
473 | |
474 | static struct bus_type rbd_bus_type = { | |
475 | .name = "rbd", | |
b15a21dd | 476 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
477 | }; |
478 | ||
479 | static void rbd_root_dev_release(struct device *dev) | |
480 | { | |
481 | } | |
482 | ||
483 | static struct device rbd_root_dev = { | |
484 | .init_name = "rbd", | |
485 | .release = rbd_root_dev_release, | |
486 | }; | |
487 | ||
06ecc6cb AE |
488 | static __printf(2, 3) |
489 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
490 | { | |
491 | struct va_format vaf; | |
492 | va_list args; | |
493 | ||
494 | va_start(args, fmt); | |
495 | vaf.fmt = fmt; | |
496 | vaf.va = &args; | |
497 | ||
498 | if (!rbd_dev) | |
499 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
500 | else if (rbd_dev->disk) | |
501 | printk(KERN_WARNING "%s: %s: %pV\n", | |
502 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
503 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
504 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
505 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
506 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
507 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
508 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
509 | else /* punt */ | |
510 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
511 | RBD_DRV_NAME, rbd_dev, &vaf); | |
512 | va_end(args); | |
513 | } | |
514 | ||
aafb230e AE |
515 | #ifdef RBD_DEBUG |
516 | #define rbd_assert(expr) \ | |
517 | if (unlikely(!(expr))) { \ | |
518 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
519 | "at line %d:\n\n" \ | |
520 | "\trbd_assert(%s);\n\n", \ | |
521 | __func__, __LINE__, #expr); \ | |
522 | BUG(); \ | |
523 | } | |
524 | #else /* !RBD_DEBUG */ | |
525 | # define rbd_assert(expr) ((void) 0) | |
526 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 527 | |
2761713d | 528 | static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request); |
b454e36d | 529 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
530 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
531 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 532 | |
cc4a38bd | 533 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 534 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 535 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 536 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
537 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
538 | u64 snap_id); | |
2ad3d716 AE |
539 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
540 | u8 *order, u64 *snap_size); | |
541 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
542 | u64 *snap_features); | |
543 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 544 | |
602adf40 YS |
545 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
546 | { | |
f0f8cef5 | 547 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 548 | bool removing = false; |
602adf40 | 549 | |
f84344f3 | 550 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
551 | return -EROFS; |
552 | ||
a14ea269 | 553 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
554 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
555 | removing = true; | |
556 | else | |
557 | rbd_dev->open_count++; | |
a14ea269 | 558 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
559 | if (removing) |
560 | return -ENOENT; | |
561 | ||
c3e946ce | 562 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 563 | |
602adf40 YS |
564 | return 0; |
565 | } | |
566 | ||
db2a144b | 567 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
568 | { |
569 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
570 | unsigned long open_count_before; |
571 | ||
a14ea269 | 572 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 573 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 574 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 575 | rbd_assert(open_count_before > 0); |
dfc5606d | 576 | |
c3e946ce | 577 | put_device(&rbd_dev->dev); |
dfc5606d YS |
578 | } |
579 | ||
131fd9f6 GZ |
580 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
581 | { | |
77f33c03 | 582 | int ret = 0; |
131fd9f6 GZ |
583 | int val; |
584 | bool ro; | |
77f33c03 | 585 | bool ro_changed = false; |
131fd9f6 | 586 | |
77f33c03 | 587 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
588 | if (get_user(val, (int __user *)(arg))) |
589 | return -EFAULT; | |
590 | ||
591 | ro = val ? true : false; | |
592 | /* Snapshot doesn't allow to write*/ | |
593 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
594 | return -EROFS; | |
595 | ||
77f33c03 JD |
596 | spin_lock_irq(&rbd_dev->lock); |
597 | /* prevent others open this device */ | |
598 | if (rbd_dev->open_count > 1) { | |
599 | ret = -EBUSY; | |
600 | goto out; | |
601 | } | |
602 | ||
131fd9f6 GZ |
603 | if (rbd_dev->mapping.read_only != ro) { |
604 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 605 | ro_changed = true; |
131fd9f6 GZ |
606 | } |
607 | ||
77f33c03 JD |
608 | out: |
609 | spin_unlock_irq(&rbd_dev->lock); | |
610 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
611 | if (ret == 0 && ro_changed) | |
612 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
613 | ||
614 | return ret; | |
131fd9f6 GZ |
615 | } |
616 | ||
617 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
618 | unsigned int cmd, unsigned long arg) | |
619 | { | |
620 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
621 | int ret = 0; | |
622 | ||
131fd9f6 GZ |
623 | switch (cmd) { |
624 | case BLKROSET: | |
625 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
626 | break; | |
627 | default: | |
628 | ret = -ENOTTY; | |
629 | } | |
630 | ||
131fd9f6 GZ |
631 | return ret; |
632 | } | |
633 | ||
634 | #ifdef CONFIG_COMPAT | |
635 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
636 | unsigned int cmd, unsigned long arg) | |
637 | { | |
638 | return rbd_ioctl(bdev, mode, cmd, arg); | |
639 | } | |
640 | #endif /* CONFIG_COMPAT */ | |
641 | ||
602adf40 YS |
642 | static const struct block_device_operations rbd_bd_ops = { |
643 | .owner = THIS_MODULE, | |
644 | .open = rbd_open, | |
dfc5606d | 645 | .release = rbd_release, |
131fd9f6 GZ |
646 | .ioctl = rbd_ioctl, |
647 | #ifdef CONFIG_COMPAT | |
648 | .compat_ioctl = rbd_compat_ioctl, | |
649 | #endif | |
602adf40 YS |
650 | }; |
651 | ||
652 | /* | |
7262cfca | 653 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 654 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 655 | */ |
f8c38929 | 656 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
657 | { |
658 | struct rbd_client *rbdc; | |
659 | int ret = -ENOMEM; | |
660 | ||
37206ee5 | 661 | dout("%s:\n", __func__); |
602adf40 YS |
662 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
663 | if (!rbdc) | |
664 | goto out_opt; | |
665 | ||
666 | kref_init(&rbdc->kref); | |
667 | INIT_LIST_HEAD(&rbdc->node); | |
668 | ||
43ae4701 | 669 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 670 | if (IS_ERR(rbdc->client)) |
08f75463 | 671 | goto out_rbdc; |
43ae4701 | 672 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
673 | |
674 | ret = ceph_open_session(rbdc->client); | |
675 | if (ret < 0) | |
08f75463 | 676 | goto out_client; |
602adf40 | 677 | |
432b8587 | 678 | spin_lock(&rbd_client_list_lock); |
602adf40 | 679 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 680 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 681 | |
37206ee5 | 682 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 683 | |
602adf40 | 684 | return rbdc; |
08f75463 | 685 | out_client: |
602adf40 | 686 | ceph_destroy_client(rbdc->client); |
08f75463 | 687 | out_rbdc: |
602adf40 YS |
688 | kfree(rbdc); |
689 | out_opt: | |
43ae4701 AE |
690 | if (ceph_opts) |
691 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
692 | dout("%s: error %d\n", __func__, ret); |
693 | ||
28f259b7 | 694 | return ERR_PTR(ret); |
602adf40 YS |
695 | } |
696 | ||
2f82ee54 AE |
697 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
698 | { | |
699 | kref_get(&rbdc->kref); | |
700 | ||
701 | return rbdc; | |
702 | } | |
703 | ||
602adf40 | 704 | /* |
1f7ba331 AE |
705 | * Find a ceph client with specific addr and configuration. If |
706 | * found, bump its reference count. | |
602adf40 | 707 | */ |
1f7ba331 | 708 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
709 | { |
710 | struct rbd_client *client_node; | |
1f7ba331 | 711 | bool found = false; |
602adf40 | 712 | |
43ae4701 | 713 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
714 | return NULL; |
715 | ||
1f7ba331 AE |
716 | spin_lock(&rbd_client_list_lock); |
717 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
718 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
719 | __rbd_get_client(client_node); |
720 | ||
1f7ba331 AE |
721 | found = true; |
722 | break; | |
723 | } | |
724 | } | |
725 | spin_unlock(&rbd_client_list_lock); | |
726 | ||
727 | return found ? client_node : NULL; | |
602adf40 YS |
728 | } |
729 | ||
59c2be1e | 730 | /* |
210c104c | 731 | * (Per device) rbd map options |
59c2be1e YS |
732 | */ |
733 | enum { | |
b5584180 | 734 | Opt_queue_depth, |
59c2be1e YS |
735 | Opt_last_int, |
736 | /* int args above */ | |
737 | Opt_last_string, | |
738 | /* string args above */ | |
cc0538b6 AE |
739 | Opt_read_only, |
740 | Opt_read_write, | |
210c104c | 741 | Opt_err |
59c2be1e YS |
742 | }; |
743 | ||
43ae4701 | 744 | static match_table_t rbd_opts_tokens = { |
b5584180 | 745 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
746 | /* int args above */ |
747 | /* string args above */ | |
be466c1c | 748 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
749 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
750 | {Opt_read_write, "read_write"}, | |
751 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 752 | {Opt_err, NULL} |
59c2be1e YS |
753 | }; |
754 | ||
98571b5a | 755 | struct rbd_options { |
b5584180 | 756 | int queue_depth; |
98571b5a AE |
757 | bool read_only; |
758 | }; | |
759 | ||
b5584180 | 760 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a AE |
761 | #define RBD_READ_ONLY_DEFAULT false |
762 | ||
59c2be1e YS |
763 | static int parse_rbd_opts_token(char *c, void *private) |
764 | { | |
43ae4701 | 765 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
766 | substring_t argstr[MAX_OPT_ARGS]; |
767 | int token, intval, ret; | |
768 | ||
43ae4701 | 769 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
770 | if (token < Opt_last_int) { |
771 | ret = match_int(&argstr[0], &intval); | |
772 | if (ret < 0) { | |
210c104c | 773 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
774 | return ret; |
775 | } | |
776 | dout("got int token %d val %d\n", token, intval); | |
777 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 778 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
779 | } else { |
780 | dout("got token %d\n", token); | |
781 | } | |
782 | ||
783 | switch (token) { | |
b5584180 ID |
784 | case Opt_queue_depth: |
785 | if (intval < 1) { | |
786 | pr_err("queue_depth out of range\n"); | |
787 | return -EINVAL; | |
788 | } | |
789 | rbd_opts->queue_depth = intval; | |
790 | break; | |
cc0538b6 AE |
791 | case Opt_read_only: |
792 | rbd_opts->read_only = true; | |
793 | break; | |
794 | case Opt_read_write: | |
795 | rbd_opts->read_only = false; | |
796 | break; | |
59c2be1e | 797 | default: |
210c104c ID |
798 | /* libceph prints "bad option" msg */ |
799 | return -EINVAL; | |
59c2be1e | 800 | } |
210c104c | 801 | |
59c2be1e YS |
802 | return 0; |
803 | } | |
804 | ||
6d2940c8 GZ |
805 | static char* obj_op_name(enum obj_operation_type op_type) |
806 | { | |
807 | switch (op_type) { | |
808 | case OBJ_OP_READ: | |
809 | return "read"; | |
810 | case OBJ_OP_WRITE: | |
811 | return "write"; | |
90e98c52 GZ |
812 | case OBJ_OP_DISCARD: |
813 | return "discard"; | |
6d2940c8 GZ |
814 | default: |
815 | return "???"; | |
816 | } | |
817 | } | |
818 | ||
602adf40 YS |
819 | /* |
820 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
821 | * not exist create it. Either way, ceph_opts is consumed by this |
822 | * function. | |
602adf40 | 823 | */ |
9d3997fd | 824 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 825 | { |
f8c38929 | 826 | struct rbd_client *rbdc; |
59c2be1e | 827 | |
cfbf6377 | 828 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 829 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 830 | if (rbdc) /* using an existing client */ |
43ae4701 | 831 | ceph_destroy_options(ceph_opts); |
9d3997fd | 832 | else |
f8c38929 | 833 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 834 | mutex_unlock(&client_mutex); |
602adf40 | 835 | |
9d3997fd | 836 | return rbdc; |
602adf40 YS |
837 | } |
838 | ||
839 | /* | |
840 | * Destroy ceph client | |
d23a4b3f | 841 | * |
432b8587 | 842 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
843 | */ |
844 | static void rbd_client_release(struct kref *kref) | |
845 | { | |
846 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
847 | ||
37206ee5 | 848 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 849 | spin_lock(&rbd_client_list_lock); |
602adf40 | 850 | list_del(&rbdc->node); |
cd9d9f5d | 851 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
852 | |
853 | ceph_destroy_client(rbdc->client); | |
854 | kfree(rbdc); | |
855 | } | |
856 | ||
857 | /* | |
858 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
859 | * it. | |
860 | */ | |
9d3997fd | 861 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 862 | { |
c53d5893 AE |
863 | if (rbdc) |
864 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
865 | } |
866 | ||
a30b71b9 AE |
867 | static bool rbd_image_format_valid(u32 image_format) |
868 | { | |
869 | return image_format == 1 || image_format == 2; | |
870 | } | |
871 | ||
8e94af8e AE |
872 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
873 | { | |
103a150f AE |
874 | size_t size; |
875 | u32 snap_count; | |
876 | ||
877 | /* The header has to start with the magic rbd header text */ | |
878 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
879 | return false; | |
880 | ||
db2388b6 AE |
881 | /* The bio layer requires at least sector-sized I/O */ |
882 | ||
883 | if (ondisk->options.order < SECTOR_SHIFT) | |
884 | return false; | |
885 | ||
886 | /* If we use u64 in a few spots we may be able to loosen this */ | |
887 | ||
888 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
889 | return false; | |
890 | ||
103a150f AE |
891 | /* |
892 | * The size of a snapshot header has to fit in a size_t, and | |
893 | * that limits the number of snapshots. | |
894 | */ | |
895 | snap_count = le32_to_cpu(ondisk->snap_count); | |
896 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
897 | if (snap_count > size / sizeof (__le64)) | |
898 | return false; | |
899 | ||
900 | /* | |
901 | * Not only that, but the size of the entire the snapshot | |
902 | * header must also be representable in a size_t. | |
903 | */ | |
904 | size -= snap_count * sizeof (__le64); | |
905 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
906 | return false; | |
907 | ||
908 | return true; | |
8e94af8e AE |
909 | } |
910 | ||
602adf40 | 911 | /* |
bb23e37a AE |
912 | * Fill an rbd image header with information from the given format 1 |
913 | * on-disk header. | |
602adf40 | 914 | */ |
662518b1 | 915 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 916 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 917 | { |
662518b1 | 918 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
919 | bool first_time = header->object_prefix == NULL; |
920 | struct ceph_snap_context *snapc; | |
921 | char *object_prefix = NULL; | |
922 | char *snap_names = NULL; | |
923 | u64 *snap_sizes = NULL; | |
ccece235 | 924 | u32 snap_count; |
d2bb24e5 | 925 | size_t size; |
bb23e37a | 926 | int ret = -ENOMEM; |
621901d6 | 927 | u32 i; |
602adf40 | 928 | |
bb23e37a | 929 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 930 | |
bb23e37a AE |
931 | if (first_time) { |
932 | size_t len; | |
103a150f | 933 | |
bb23e37a AE |
934 | len = strnlen(ondisk->object_prefix, |
935 | sizeof (ondisk->object_prefix)); | |
936 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
937 | if (!object_prefix) | |
938 | return -ENOMEM; | |
939 | memcpy(object_prefix, ondisk->object_prefix, len); | |
940 | object_prefix[len] = '\0'; | |
941 | } | |
00f1f36f | 942 | |
bb23e37a | 943 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 944 | |
bb23e37a AE |
945 | snap_count = le32_to_cpu(ondisk->snap_count); |
946 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
947 | if (!snapc) | |
948 | goto out_err; | |
949 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 950 | if (snap_count) { |
bb23e37a | 951 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
952 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
953 | ||
bb23e37a | 954 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 955 | |
bb23e37a AE |
956 | if (snap_names_len > (u64)SIZE_MAX) |
957 | goto out_2big; | |
958 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
959 | if (!snap_names) | |
6a52325f AE |
960 | goto out_err; |
961 | ||
bb23e37a | 962 | /* ...as well as the array of their sizes. */ |
621901d6 | 963 | |
d2bb24e5 | 964 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
965 | snap_sizes = kmalloc(size, GFP_KERNEL); |
966 | if (!snap_sizes) | |
6a52325f | 967 | goto out_err; |
bb23e37a | 968 | |
f785cc1d | 969 | /* |
bb23e37a AE |
970 | * Copy the names, and fill in each snapshot's id |
971 | * and size. | |
972 | * | |
99a41ebc | 973 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 974 | * ondisk buffer we're working with has |
f785cc1d AE |
975 | * snap_names_len bytes beyond the end of the |
976 | * snapshot id array, this memcpy() is safe. | |
977 | */ | |
bb23e37a AE |
978 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
979 | snaps = ondisk->snaps; | |
980 | for (i = 0; i < snap_count; i++) { | |
981 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
982 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
983 | } | |
602adf40 | 984 | } |
6a52325f | 985 | |
bb23e37a | 986 | /* We won't fail any more, fill in the header */ |
621901d6 | 987 | |
bb23e37a AE |
988 | if (first_time) { |
989 | header->object_prefix = object_prefix; | |
990 | header->obj_order = ondisk->options.order; | |
991 | header->crypt_type = ondisk->options.crypt_type; | |
992 | header->comp_type = ondisk->options.comp_type; | |
993 | /* The rest aren't used for format 1 images */ | |
994 | header->stripe_unit = 0; | |
995 | header->stripe_count = 0; | |
996 | header->features = 0; | |
602adf40 | 997 | } else { |
662518b1 AE |
998 | ceph_put_snap_context(header->snapc); |
999 | kfree(header->snap_names); | |
1000 | kfree(header->snap_sizes); | |
602adf40 | 1001 | } |
849b4260 | 1002 | |
bb23e37a | 1003 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1004 | |
f84344f3 | 1005 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1006 | header->snapc = snapc; |
1007 | header->snap_names = snap_names; | |
1008 | header->snap_sizes = snap_sizes; | |
468521c1 | 1009 | |
602adf40 | 1010 | return 0; |
bb23e37a AE |
1011 | out_2big: |
1012 | ret = -EIO; | |
6a52325f | 1013 | out_err: |
bb23e37a AE |
1014 | kfree(snap_sizes); |
1015 | kfree(snap_names); | |
1016 | ceph_put_snap_context(snapc); | |
1017 | kfree(object_prefix); | |
ccece235 | 1018 | |
bb23e37a | 1019 | return ret; |
602adf40 YS |
1020 | } |
1021 | ||
9682fc6d AE |
1022 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1023 | { | |
1024 | const char *snap_name; | |
1025 | ||
1026 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1027 | ||
1028 | /* Skip over names until we find the one we are looking for */ | |
1029 | ||
1030 | snap_name = rbd_dev->header.snap_names; | |
1031 | while (which--) | |
1032 | snap_name += strlen(snap_name) + 1; | |
1033 | ||
1034 | return kstrdup(snap_name, GFP_KERNEL); | |
1035 | } | |
1036 | ||
30d1cff8 AE |
1037 | /* |
1038 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1039 | * Note that result is for snapshots in *descending* order. | |
1040 | */ | |
1041 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1042 | { | |
1043 | u64 snap_id1 = *(u64 *)s1; | |
1044 | u64 snap_id2 = *(u64 *)s2; | |
1045 | ||
1046 | if (snap_id1 < snap_id2) | |
1047 | return 1; | |
1048 | return snap_id1 == snap_id2 ? 0 : -1; | |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * Search a snapshot context to see if the given snapshot id is | |
1053 | * present. | |
1054 | * | |
1055 | * Returns the position of the snapshot id in the array if it's found, | |
1056 | * or BAD_SNAP_INDEX otherwise. | |
1057 | * | |
1058 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1059 | * reverse order, highest snapshot id first. | |
1060 | */ | |
9682fc6d AE |
1061 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1062 | { | |
1063 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1064 | u64 *found; |
9682fc6d | 1065 | |
30d1cff8 AE |
1066 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1067 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1068 | |
30d1cff8 | 1069 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1070 | } |
1071 | ||
2ad3d716 AE |
1072 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1073 | u64 snap_id) | |
9e15b77d | 1074 | { |
54cac61f | 1075 | u32 which; |
da6a6b63 | 1076 | const char *snap_name; |
9e15b77d | 1077 | |
54cac61f AE |
1078 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1079 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1080 | return ERR_PTR(-ENOENT); |
54cac61f | 1081 | |
da6a6b63 JD |
1082 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1083 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1084 | } |
1085 | ||
1086 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1087 | { | |
9e15b77d AE |
1088 | if (snap_id == CEPH_NOSNAP) |
1089 | return RBD_SNAP_HEAD_NAME; | |
1090 | ||
54cac61f AE |
1091 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1092 | if (rbd_dev->image_format == 1) | |
1093 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1094 | |
54cac61f | 1095 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1096 | } |
1097 | ||
2ad3d716 AE |
1098 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1099 | u64 *snap_size) | |
602adf40 | 1100 | { |
2ad3d716 AE |
1101 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1102 | if (snap_id == CEPH_NOSNAP) { | |
1103 | *snap_size = rbd_dev->header.image_size; | |
1104 | } else if (rbd_dev->image_format == 1) { | |
1105 | u32 which; | |
602adf40 | 1106 | |
2ad3d716 AE |
1107 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1108 | if (which == BAD_SNAP_INDEX) | |
1109 | return -ENOENT; | |
e86924a8 | 1110 | |
2ad3d716 AE |
1111 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1112 | } else { | |
1113 | u64 size = 0; | |
1114 | int ret; | |
1115 | ||
1116 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1117 | if (ret) | |
1118 | return ret; | |
1119 | ||
1120 | *snap_size = size; | |
1121 | } | |
1122 | return 0; | |
602adf40 YS |
1123 | } |
1124 | ||
2ad3d716 AE |
1125 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1126 | u64 *snap_features) | |
602adf40 | 1127 | { |
2ad3d716 AE |
1128 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1129 | if (snap_id == CEPH_NOSNAP) { | |
1130 | *snap_features = rbd_dev->header.features; | |
1131 | } else if (rbd_dev->image_format == 1) { | |
1132 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1133 | } else { |
2ad3d716 AE |
1134 | u64 features = 0; |
1135 | int ret; | |
8b0241f8 | 1136 | |
2ad3d716 AE |
1137 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1138 | if (ret) | |
1139 | return ret; | |
1140 | ||
1141 | *snap_features = features; | |
1142 | } | |
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1147 | { | |
8f4b7d98 | 1148 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1149 | u64 size = 0; |
1150 | u64 features = 0; | |
1151 | int ret; | |
1152 | ||
2ad3d716 AE |
1153 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1154 | if (ret) | |
1155 | return ret; | |
1156 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1157 | if (ret) | |
1158 | return ret; | |
1159 | ||
1160 | rbd_dev->mapping.size = size; | |
1161 | rbd_dev->mapping.features = features; | |
1162 | ||
8b0241f8 | 1163 | return 0; |
602adf40 YS |
1164 | } |
1165 | ||
d1cf5788 AE |
1166 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1167 | { | |
1168 | rbd_dev->mapping.size = 0; | |
1169 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1170 | } |
1171 | ||
7d5079aa HS |
1172 | static void rbd_segment_name_free(const char *name) |
1173 | { | |
1174 | /* The explicit cast here is needed to drop the const qualifier */ | |
1175 | ||
1176 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1177 | } | |
1178 | ||
98571b5a | 1179 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1180 | { |
65ccfe21 AE |
1181 | char *name; |
1182 | u64 segment; | |
1183 | int ret; | |
3a96d5cd | 1184 | char *name_format; |
602adf40 | 1185 | |
78c2a44a | 1186 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1187 | if (!name) |
1188 | return NULL; | |
1189 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1190 | name_format = "%s.%012llx"; |
1191 | if (rbd_dev->image_format == 2) | |
1192 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1193 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1194 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1195 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1196 | pr_err("error formatting segment name for #%llu (%d)\n", |
1197 | segment, ret); | |
7d5079aa | 1198 | rbd_segment_name_free(name); |
65ccfe21 AE |
1199 | name = NULL; |
1200 | } | |
602adf40 | 1201 | |
65ccfe21 AE |
1202 | return name; |
1203 | } | |
602adf40 | 1204 | |
65ccfe21 AE |
1205 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1206 | { | |
1207 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1208 | |
65ccfe21 AE |
1209 | return offset & (segment_size - 1); |
1210 | } | |
1211 | ||
1212 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1213 | u64 offset, u64 length) | |
1214 | { | |
1215 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1216 | ||
1217 | offset &= segment_size - 1; | |
1218 | ||
aafb230e | 1219 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1220 | if (offset + length > segment_size) |
1221 | length = segment_size - offset; | |
1222 | ||
1223 | return length; | |
602adf40 YS |
1224 | } |
1225 | ||
029bcbd8 JD |
1226 | /* |
1227 | * returns the size of an object in the image | |
1228 | */ | |
1229 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1230 | { | |
1231 | return 1 << header->obj_order; | |
1232 | } | |
1233 | ||
602adf40 YS |
1234 | /* |
1235 | * bio helpers | |
1236 | */ | |
1237 | ||
1238 | static void bio_chain_put(struct bio *chain) | |
1239 | { | |
1240 | struct bio *tmp; | |
1241 | ||
1242 | while (chain) { | |
1243 | tmp = chain; | |
1244 | chain = chain->bi_next; | |
1245 | bio_put(tmp); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * zeros a bio chain, starting at specific offset | |
1251 | */ | |
1252 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1253 | { | |
7988613b KO |
1254 | struct bio_vec bv; |
1255 | struct bvec_iter iter; | |
602adf40 YS |
1256 | unsigned long flags; |
1257 | void *buf; | |
602adf40 YS |
1258 | int pos = 0; |
1259 | ||
1260 | while (chain) { | |
7988613b KO |
1261 | bio_for_each_segment(bv, chain, iter) { |
1262 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1263 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1264 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1265 | memset(buf + remainder, 0, |
7988613b KO |
1266 | bv.bv_len - remainder); |
1267 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1268 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1269 | } |
7988613b | 1270 | pos += bv.bv_len; |
602adf40 YS |
1271 | } |
1272 | ||
1273 | chain = chain->bi_next; | |
1274 | } | |
1275 | } | |
1276 | ||
b9434c5b AE |
1277 | /* |
1278 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1279 | * starting at the given byte offset from the start of the array and | |
1280 | * continuing up to the given end offset. The pages array is | |
1281 | * assumed to be big enough to hold all bytes up to the end. | |
1282 | */ | |
1283 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1284 | { | |
1285 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1286 | ||
1287 | rbd_assert(end > offset); | |
1288 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1289 | while (offset < end) { | |
1290 | size_t page_offset; | |
1291 | size_t length; | |
1292 | unsigned long flags; | |
1293 | void *kaddr; | |
1294 | ||
491205a8 GU |
1295 | page_offset = offset & ~PAGE_MASK; |
1296 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1297 | local_irq_save(flags); |
1298 | kaddr = kmap_atomic(*page); | |
1299 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1300 | flush_dcache_page(*page); |
b9434c5b AE |
1301 | kunmap_atomic(kaddr); |
1302 | local_irq_restore(flags); | |
1303 | ||
1304 | offset += length; | |
1305 | page++; | |
1306 | } | |
1307 | } | |
1308 | ||
602adf40 | 1309 | /* |
f7760dad AE |
1310 | * Clone a portion of a bio, starting at the given byte offset |
1311 | * and continuing for the number of bytes indicated. | |
602adf40 | 1312 | */ |
f7760dad AE |
1313 | static struct bio *bio_clone_range(struct bio *bio_src, |
1314 | unsigned int offset, | |
1315 | unsigned int len, | |
1316 | gfp_t gfpmask) | |
602adf40 | 1317 | { |
f7760dad AE |
1318 | struct bio *bio; |
1319 | ||
5341a627 | 1320 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1321 | if (!bio) |
1322 | return NULL; /* ENOMEM */ | |
602adf40 | 1323 | |
5341a627 | 1324 | bio_advance(bio, offset); |
4f024f37 | 1325 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1326 | |
1327 | return bio; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Clone a portion of a bio chain, starting at the given byte offset | |
1332 | * into the first bio in the source chain and continuing for the | |
1333 | * number of bytes indicated. The result is another bio chain of | |
1334 | * exactly the given length, or a null pointer on error. | |
1335 | * | |
1336 | * The bio_src and offset parameters are both in-out. On entry they | |
1337 | * refer to the first source bio and the offset into that bio where | |
1338 | * the start of data to be cloned is located. | |
1339 | * | |
1340 | * On return, bio_src is updated to refer to the bio in the source | |
1341 | * chain that contains first un-cloned byte, and *offset will | |
1342 | * contain the offset of that byte within that bio. | |
1343 | */ | |
1344 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1345 | unsigned int *offset, | |
1346 | unsigned int len, | |
1347 | gfp_t gfpmask) | |
1348 | { | |
1349 | struct bio *bi = *bio_src; | |
1350 | unsigned int off = *offset; | |
1351 | struct bio *chain = NULL; | |
1352 | struct bio **end; | |
1353 | ||
1354 | /* Build up a chain of clone bios up to the limit */ | |
1355 | ||
4f024f37 | 1356 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1357 | return NULL; /* Nothing to clone */ |
602adf40 | 1358 | |
f7760dad AE |
1359 | end = &chain; |
1360 | while (len) { | |
1361 | unsigned int bi_size; | |
1362 | struct bio *bio; | |
1363 | ||
f5400b7a AE |
1364 | if (!bi) { |
1365 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1366 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1367 | } |
4f024f37 | 1368 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1369 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1370 | if (!bio) | |
1371 | goto out_err; /* ENOMEM */ | |
1372 | ||
1373 | *end = bio; | |
1374 | end = &bio->bi_next; | |
602adf40 | 1375 | |
f7760dad | 1376 | off += bi_size; |
4f024f37 | 1377 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1378 | bi = bi->bi_next; |
1379 | off = 0; | |
1380 | } | |
1381 | len -= bi_size; | |
1382 | } | |
1383 | *bio_src = bi; | |
1384 | *offset = off; | |
1385 | ||
1386 | return chain; | |
1387 | out_err: | |
1388 | bio_chain_put(chain); | |
602adf40 | 1389 | |
602adf40 YS |
1390 | return NULL; |
1391 | } | |
1392 | ||
926f9b3f AE |
1393 | /* |
1394 | * The default/initial value for all object request flags is 0. For | |
1395 | * each flag, once its value is set to 1 it is never reset to 0 | |
1396 | * again. | |
1397 | */ | |
57acbaa7 | 1398 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1399 | { |
57acbaa7 | 1400 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1401 | struct rbd_device *rbd_dev; |
1402 | ||
57acbaa7 | 1403 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1404 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1405 | obj_request); |
1406 | } | |
1407 | } | |
1408 | ||
57acbaa7 | 1409 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1410 | { |
1411 | smp_mb(); | |
57acbaa7 | 1412 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1413 | } |
1414 | ||
57acbaa7 | 1415 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1416 | { |
57acbaa7 AE |
1417 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1418 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1419 | |
57acbaa7 AE |
1420 | if (obj_request_img_data_test(obj_request)) |
1421 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1422 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1423 | obj_request); |
1424 | } | |
1425 | } | |
1426 | ||
57acbaa7 | 1427 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1428 | { |
1429 | smp_mb(); | |
57acbaa7 | 1430 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1431 | } |
1432 | ||
5679c59f AE |
1433 | /* |
1434 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1435 | * flag. The latter is set based on the "exists" value provided. | |
1436 | * | |
1437 | * Note that for our purposes once an object exists it never goes | |
1438 | * away again. It's possible that the response from two existence | |
1439 | * checks are separated by the creation of the target object, and | |
1440 | * the first ("doesn't exist") response arrives *after* the second | |
1441 | * ("does exist"). In that case we ignore the second one. | |
1442 | */ | |
1443 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1444 | bool exists) | |
1445 | { | |
1446 | if (exists) | |
1447 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1448 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1449 | smp_mb(); | |
1450 | } | |
1451 | ||
1452 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1453 | { | |
1454 | smp_mb(); | |
1455 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1456 | } | |
1457 | ||
1458 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1459 | { | |
1460 | smp_mb(); | |
1461 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1462 | } | |
1463 | ||
9638556a ID |
1464 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1465 | { | |
1466 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1467 | ||
1468 | return obj_request->img_offset < | |
1469 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1470 | } | |
1471 | ||
bf0d5f50 AE |
1472 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1473 | { | |
37206ee5 AE |
1474 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1475 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1476 | kref_get(&obj_request->kref); |
1477 | } | |
1478 | ||
1479 | static void rbd_obj_request_destroy(struct kref *kref); | |
1480 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1481 | { | |
1482 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1483 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1484 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1485 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1486 | } | |
1487 | ||
0f2d5be7 AE |
1488 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1489 | { | |
1490 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1491 | atomic_read(&img_request->kref.refcount)); | |
1492 | kref_get(&img_request->kref); | |
1493 | } | |
1494 | ||
e93f3152 AE |
1495 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1496 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1497 | static void rbd_img_request_destroy(struct kref *kref); |
1498 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1499 | { | |
1500 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1501 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1502 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1503 | if (img_request_child_test(img_request)) |
1504 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1505 | else | |
1506 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1507 | } |
1508 | ||
1509 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1510 | struct rbd_obj_request *obj_request) | |
1511 | { | |
25dcf954 AE |
1512 | rbd_assert(obj_request->img_request == NULL); |
1513 | ||
b155e86c | 1514 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1515 | obj_request->img_request = img_request; |
25dcf954 | 1516 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1517 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1518 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1519 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1520 | img_request->obj_request_count++; |
1521 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1522 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1523 | obj_request->which); | |
bf0d5f50 AE |
1524 | } |
1525 | ||
1526 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1527 | struct rbd_obj_request *obj_request) | |
1528 | { | |
1529 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1530 | |
37206ee5 AE |
1531 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1532 | obj_request->which); | |
bf0d5f50 | 1533 | list_del(&obj_request->links); |
25dcf954 AE |
1534 | rbd_assert(img_request->obj_request_count > 0); |
1535 | img_request->obj_request_count--; | |
1536 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1537 | obj_request->which = BAD_WHICH; | |
6365d33a | 1538 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1539 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1540 | obj_request->img_request = NULL; |
25dcf954 | 1541 | obj_request->callback = NULL; |
bf0d5f50 AE |
1542 | rbd_obj_request_put(obj_request); |
1543 | } | |
1544 | ||
1545 | static bool obj_request_type_valid(enum obj_request_type type) | |
1546 | { | |
1547 | switch (type) { | |
9969ebc5 | 1548 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1549 | case OBJ_REQUEST_BIO: |
788e2df3 | 1550 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1551 | return true; |
1552 | default: | |
1553 | return false; | |
1554 | } | |
1555 | } | |
1556 | ||
bf0d5f50 AE |
1557 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1558 | struct rbd_obj_request *obj_request) | |
1559 | { | |
71c20a06 | 1560 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1561 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1562 | } | |
1563 | ||
71c20a06 ID |
1564 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1565 | { | |
1566 | dout("%s %p\n", __func__, obj_request); | |
1567 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Wait for an object request to complete. If interrupted, cancel the | |
1572 | * underlying osd request. | |
2894e1d7 ID |
1573 | * |
1574 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1575 | */ |
2894e1d7 ID |
1576 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1577 | unsigned long timeout) | |
71c20a06 | 1578 | { |
2894e1d7 | 1579 | long ret; |
71c20a06 ID |
1580 | |
1581 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1582 | ret = wait_for_completion_interruptible_timeout( |
1583 | &obj_request->completion, | |
1584 | ceph_timeout_jiffies(timeout)); | |
1585 | if (ret <= 0) { | |
1586 | if (ret == 0) | |
1587 | ret = -ETIMEDOUT; | |
71c20a06 | 1588 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1589 | } else { |
1590 | ret = 0; | |
71c20a06 ID |
1591 | } |
1592 | ||
2894e1d7 ID |
1593 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1598 | { | |
1599 | return __rbd_obj_request_wait(obj_request, 0); | |
1600 | } | |
1601 | ||
1602 | static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request, | |
1603 | unsigned long timeout) | |
1604 | { | |
1605 | return __rbd_obj_request_wait(obj_request, timeout); | |
71c20a06 ID |
1606 | } |
1607 | ||
bf0d5f50 AE |
1608 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1609 | { | |
55f27e09 | 1610 | |
37206ee5 | 1611 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1612 | |
1613 | /* | |
1614 | * If no error occurred, compute the aggregate transfer | |
1615 | * count for the image request. We could instead use | |
1616 | * atomic64_cmpxchg() to update it as each object request | |
1617 | * completes; not clear which way is better off hand. | |
1618 | */ | |
1619 | if (!img_request->result) { | |
1620 | struct rbd_obj_request *obj_request; | |
1621 | u64 xferred = 0; | |
1622 | ||
1623 | for_each_obj_request(img_request, obj_request) | |
1624 | xferred += obj_request->xferred; | |
1625 | img_request->xferred = xferred; | |
1626 | } | |
1627 | ||
bf0d5f50 AE |
1628 | if (img_request->callback) |
1629 | img_request->callback(img_request); | |
1630 | else | |
1631 | rbd_img_request_put(img_request); | |
1632 | } | |
1633 | ||
0c425248 AE |
1634 | /* |
1635 | * The default/initial value for all image request flags is 0. Each | |
1636 | * is conditionally set to 1 at image request initialization time | |
1637 | * and currently never change thereafter. | |
1638 | */ | |
1639 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1640 | { | |
1641 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1642 | smp_mb(); | |
1643 | } | |
1644 | ||
1645 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1646 | { | |
1647 | smp_mb(); | |
1648 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1649 | } | |
1650 | ||
90e98c52 GZ |
1651 | /* |
1652 | * Set the discard flag when the img_request is an discard request | |
1653 | */ | |
1654 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1655 | { | |
1656 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1657 | smp_mb(); | |
1658 | } | |
1659 | ||
1660 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1661 | { | |
1662 | smp_mb(); | |
1663 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1664 | } | |
1665 | ||
9849e986 AE |
1666 | static void img_request_child_set(struct rbd_img_request *img_request) |
1667 | { | |
1668 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1669 | smp_mb(); | |
1670 | } | |
1671 | ||
e93f3152 AE |
1672 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1673 | { | |
1674 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1675 | smp_mb(); | |
1676 | } | |
1677 | ||
9849e986 AE |
1678 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1679 | { | |
1680 | smp_mb(); | |
1681 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1682 | } | |
1683 | ||
d0b2e944 AE |
1684 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1685 | { | |
1686 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1687 | smp_mb(); | |
1688 | } | |
1689 | ||
a2acd00e AE |
1690 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1691 | { | |
1692 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1693 | smp_mb(); | |
1694 | } | |
1695 | ||
d0b2e944 AE |
1696 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1697 | { | |
1698 | smp_mb(); | |
1699 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1700 | } | |
1701 | ||
3b434a2a JD |
1702 | static enum obj_operation_type |
1703 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1704 | { | |
1705 | if (img_request_write_test(img_request)) | |
1706 | return OBJ_OP_WRITE; | |
1707 | else if (img_request_discard_test(img_request)) | |
1708 | return OBJ_OP_DISCARD; | |
1709 | else | |
1710 | return OBJ_OP_READ; | |
1711 | } | |
1712 | ||
6e2a4505 AE |
1713 | static void |
1714 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1715 | { | |
b9434c5b AE |
1716 | u64 xferred = obj_request->xferred; |
1717 | u64 length = obj_request->length; | |
1718 | ||
6e2a4505 AE |
1719 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1720 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1721 | xferred, length); |
6e2a4505 | 1722 | /* |
17c1cc1d JD |
1723 | * ENOENT means a hole in the image. We zero-fill the entire |
1724 | * length of the request. A short read also implies zero-fill | |
1725 | * to the end of the request. An error requires the whole | |
1726 | * length of the request to be reported finished with an error | |
1727 | * to the block layer. In each case we update the xferred | |
1728 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1729 | */ |
b9434c5b | 1730 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1731 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1732 | if (obj_request->type == OBJ_REQUEST_BIO) |
1733 | zero_bio_chain(obj_request->bio_list, 0); | |
1734 | else | |
1735 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1736 | obj_request->result = 0; |
b9434c5b AE |
1737 | } else if (xferred < length && !obj_request->result) { |
1738 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1739 | zero_bio_chain(obj_request->bio_list, xferred); | |
1740 | else | |
1741 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1742 | } |
17c1cc1d | 1743 | obj_request->xferred = length; |
6e2a4505 AE |
1744 | obj_request_done_set(obj_request); |
1745 | } | |
1746 | ||
bf0d5f50 AE |
1747 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1748 | { | |
37206ee5 AE |
1749 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1750 | obj_request->callback); | |
bf0d5f50 AE |
1751 | if (obj_request->callback) |
1752 | obj_request->callback(obj_request); | |
788e2df3 AE |
1753 | else |
1754 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1755 | } |
1756 | ||
c47f9371 | 1757 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1758 | { |
1759 | dout("%s: obj %p\n", __func__, obj_request); | |
1760 | obj_request_done_set(obj_request); | |
1761 | } | |
1762 | ||
c47f9371 | 1763 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1764 | { |
57acbaa7 | 1765 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1766 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1767 | bool layered = false; |
1768 | ||
1769 | if (obj_request_img_data_test(obj_request)) { | |
1770 | img_request = obj_request->img_request; | |
1771 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1772 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1773 | } |
8b3e1a56 AE |
1774 | |
1775 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1776 | obj_request, img_request, obj_request->result, | |
1777 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1778 | if (layered && obj_request->result == -ENOENT && |
1779 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1780 | rbd_img_parent_read(obj_request); |
1781 | else if (img_request) | |
6e2a4505 AE |
1782 | rbd_img_obj_request_read_callback(obj_request); |
1783 | else | |
1784 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1785 | } |
1786 | ||
c47f9371 | 1787 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1788 | { |
1b83bef2 SW |
1789 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1790 | obj_request->result, obj_request->length); | |
1791 | /* | |
8b3e1a56 AE |
1792 | * There is no such thing as a successful short write. Set |
1793 | * it to our originally-requested length. | |
1b83bef2 SW |
1794 | */ |
1795 | obj_request->xferred = obj_request->length; | |
07741308 | 1796 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1797 | } |
1798 | ||
90e98c52 GZ |
1799 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1800 | { | |
1801 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1802 | obj_request->result, obj_request->length); | |
1803 | /* | |
1804 | * There is no such thing as a successful short discard. Set | |
1805 | * it to our originally-requested length. | |
1806 | */ | |
1807 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1808 | /* discarding a non-existent object is not a problem */ |
1809 | if (obj_request->result == -ENOENT) | |
1810 | obj_request->result = 0; | |
90e98c52 GZ |
1811 | obj_request_done_set(obj_request); |
1812 | } | |
1813 | ||
fbfab539 AE |
1814 | /* |
1815 | * For a simple stat call there's nothing to do. We'll do more if | |
1816 | * this is part of a write sequence for a layered image. | |
1817 | */ | |
c47f9371 | 1818 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1819 | { |
37206ee5 | 1820 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1821 | obj_request_done_set(obj_request); |
1822 | } | |
1823 | ||
2761713d ID |
1824 | static void rbd_osd_call_callback(struct rbd_obj_request *obj_request) |
1825 | { | |
1826 | dout("%s: obj %p\n", __func__, obj_request); | |
1827 | ||
1828 | if (obj_request_img_data_test(obj_request)) | |
1829 | rbd_osd_copyup_callback(obj_request); | |
1830 | else | |
1831 | obj_request_done_set(obj_request); | |
1832 | } | |
1833 | ||
bf0d5f50 AE |
1834 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1835 | struct ceph_msg *msg) | |
1836 | { | |
1837 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1838 | u16 opcode; |
1839 | ||
37206ee5 | 1840 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1841 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1842 | if (obj_request_img_data_test(obj_request)) { |
1843 | rbd_assert(obj_request->img_request); | |
1844 | rbd_assert(obj_request->which != BAD_WHICH); | |
1845 | } else { | |
1846 | rbd_assert(obj_request->which == BAD_WHICH); | |
1847 | } | |
bf0d5f50 | 1848 | |
1b83bef2 SW |
1849 | if (osd_req->r_result < 0) |
1850 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1851 | |
7cc69d42 | 1852 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1853 | |
c47f9371 AE |
1854 | /* |
1855 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1856 | * passed to the block layer, which just supports a 32-bit |
1857 | * length field. | |
c47f9371 | 1858 | */ |
1b83bef2 | 1859 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1860 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1861 | |
79528734 | 1862 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1863 | switch (opcode) { |
1864 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1865 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1866 | break; |
0ccd5926 | 1867 | case CEPH_OSD_OP_SETALLOCHINT: |
e30b7577 ID |
1868 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE || |
1869 | osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL); | |
0ccd5926 | 1870 | /* fall through */ |
bf0d5f50 | 1871 | case CEPH_OSD_OP_WRITE: |
e30b7577 | 1872 | case CEPH_OSD_OP_WRITEFULL: |
c47f9371 | 1873 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1874 | break; |
fbfab539 | 1875 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1876 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1877 | break; |
90e98c52 GZ |
1878 | case CEPH_OSD_OP_DELETE: |
1879 | case CEPH_OSD_OP_TRUNCATE: | |
1880 | case CEPH_OSD_OP_ZERO: | |
1881 | rbd_osd_discard_callback(obj_request); | |
1882 | break; | |
36be9a76 | 1883 | case CEPH_OSD_OP_CALL: |
2761713d ID |
1884 | rbd_osd_call_callback(obj_request); |
1885 | break; | |
b8d70035 | 1886 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1887 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1888 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1889 | break; |
bf0d5f50 | 1890 | default: |
9584d508 | 1891 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1892 | obj_request->object_name, (unsigned short) opcode); |
1893 | break; | |
1894 | } | |
1895 | ||
07741308 | 1896 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1897 | rbd_obj_request_complete(obj_request); |
1898 | } | |
1899 | ||
9d4df01f | 1900 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1901 | { |
1902 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1903 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1904 | u64 snap_id; |
430c28c3 | 1905 | |
8c042b0d | 1906 | rbd_assert(osd_req != NULL); |
430c28c3 | 1907 | |
9d4df01f | 1908 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1909 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1910 | NULL, snap_id, NULL); |
1911 | } | |
1912 | ||
1913 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1914 | { | |
1915 | struct rbd_img_request *img_request = obj_request->img_request; | |
1916 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1917 | struct ceph_snap_context *snapc; | |
1918 | struct timespec mtime = CURRENT_TIME; | |
1919 | ||
1920 | rbd_assert(osd_req != NULL); | |
1921 | ||
1922 | snapc = img_request ? img_request->snapc : NULL; | |
1923 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1924 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1925 | } |
1926 | ||
0ccd5926 ID |
1927 | /* |
1928 | * Create an osd request. A read request has one osd op (read). | |
1929 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1930 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1931 | * technically osd watch is a write request, hence this distinction.) | |
1932 | */ | |
bf0d5f50 AE |
1933 | static struct ceph_osd_request *rbd_osd_req_create( |
1934 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1935 | enum obj_operation_type op_type, |
deb236b3 | 1936 | unsigned int num_ops, |
430c28c3 | 1937 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1938 | { |
bf0d5f50 AE |
1939 | struct ceph_snap_context *snapc = NULL; |
1940 | struct ceph_osd_client *osdc; | |
1941 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1942 | |
90e98c52 GZ |
1943 | if (obj_request_img_data_test(obj_request) && |
1944 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1945 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1946 | if (op_type == OBJ_OP_WRITE) { |
1947 | rbd_assert(img_request_write_test(img_request)); | |
1948 | } else { | |
1949 | rbd_assert(img_request_discard_test(img_request)); | |
1950 | } | |
6d2940c8 | 1951 | snapc = img_request->snapc; |
bf0d5f50 AE |
1952 | } |
1953 | ||
6d2940c8 | 1954 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1955 | |
1956 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1957 | |
1958 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1959 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1960 | GFP_ATOMIC); | |
bf0d5f50 AE |
1961 | if (!osd_req) |
1962 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1963 | |
90e98c52 | 1964 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1965 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1966 | else |
bf0d5f50 | 1967 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1968 | |
1969 | osd_req->r_callback = rbd_osd_req_callback; | |
1970 | osd_req->r_priv = obj_request; | |
1971 | ||
3c972c95 ID |
1972 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1973 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1974 | |
bf0d5f50 AE |
1975 | return osd_req; |
1976 | } | |
1977 | ||
0eefd470 | 1978 | /* |
d3246fb0 JD |
1979 | * Create a copyup osd request based on the information in the object |
1980 | * request supplied. A copyup request has two or three osd ops, a | |
1981 | * copyup method call, potentially a hint op, and a write or truncate | |
1982 | * or zero op. | |
0eefd470 AE |
1983 | */ |
1984 | static struct ceph_osd_request * | |
1985 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1986 | { | |
1987 | struct rbd_img_request *img_request; | |
1988 | struct ceph_snap_context *snapc; | |
1989 | struct rbd_device *rbd_dev; | |
1990 | struct ceph_osd_client *osdc; | |
1991 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1992 | int num_osd_ops = 3; |
0eefd470 AE |
1993 | |
1994 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1995 | img_request = obj_request->img_request; | |
1996 | rbd_assert(img_request); | |
d3246fb0 JD |
1997 | rbd_assert(img_request_write_test(img_request) || |
1998 | img_request_discard_test(img_request)); | |
0eefd470 | 1999 | |
d3246fb0 JD |
2000 | if (img_request_discard_test(img_request)) |
2001 | num_osd_ops = 2; | |
2002 | ||
2003 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
2004 | |
2005 | snapc = img_request->snapc; | |
2006 | rbd_dev = img_request->rbd_dev; | |
2007 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 JD |
2008 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
2009 | false, GFP_ATOMIC); | |
0eefd470 AE |
2010 | if (!osd_req) |
2011 | return NULL; /* ENOMEM */ | |
2012 | ||
2013 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
2014 | osd_req->r_callback = rbd_osd_req_callback; | |
2015 | osd_req->r_priv = obj_request; | |
2016 | ||
3c972c95 ID |
2017 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
2018 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 2019 | |
0eefd470 AE |
2020 | return osd_req; |
2021 | } | |
2022 | ||
2023 | ||
bf0d5f50 AE |
2024 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2025 | { | |
2026 | ceph_osdc_put_request(osd_req); | |
2027 | } | |
2028 | ||
2029 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2030 | ||
2031 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2032 | u64 offset, u64 length, | |
2033 | enum obj_request_type type) | |
2034 | { | |
2035 | struct rbd_obj_request *obj_request; | |
2036 | size_t size; | |
2037 | char *name; | |
2038 | ||
2039 | rbd_assert(obj_request_type_valid(type)); | |
2040 | ||
2041 | size = strlen(object_name) + 1; | |
5a60e876 | 2042 | name = kmalloc(size, GFP_NOIO); |
f907ad55 | 2043 | if (!name) |
bf0d5f50 AE |
2044 | return NULL; |
2045 | ||
5a60e876 | 2046 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
f907ad55 AE |
2047 | if (!obj_request) { |
2048 | kfree(name); | |
2049 | return NULL; | |
2050 | } | |
2051 | ||
bf0d5f50 AE |
2052 | obj_request->object_name = memcpy(name, object_name, size); |
2053 | obj_request->offset = offset; | |
2054 | obj_request->length = length; | |
926f9b3f | 2055 | obj_request->flags = 0; |
bf0d5f50 AE |
2056 | obj_request->which = BAD_WHICH; |
2057 | obj_request->type = type; | |
2058 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2059 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2060 | kref_init(&obj_request->kref); |
2061 | ||
37206ee5 AE |
2062 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2063 | offset, length, (int)type, obj_request); | |
2064 | ||
bf0d5f50 AE |
2065 | return obj_request; |
2066 | } | |
2067 | ||
2068 | static void rbd_obj_request_destroy(struct kref *kref) | |
2069 | { | |
2070 | struct rbd_obj_request *obj_request; | |
2071 | ||
2072 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2073 | ||
37206ee5 AE |
2074 | dout("%s: obj %p\n", __func__, obj_request); |
2075 | ||
bf0d5f50 AE |
2076 | rbd_assert(obj_request->img_request == NULL); |
2077 | rbd_assert(obj_request->which == BAD_WHICH); | |
2078 | ||
2079 | if (obj_request->osd_req) | |
2080 | rbd_osd_req_destroy(obj_request->osd_req); | |
2081 | ||
2082 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2083 | switch (obj_request->type) { | |
9969ebc5 AE |
2084 | case OBJ_REQUEST_NODATA: |
2085 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2086 | case OBJ_REQUEST_BIO: |
2087 | if (obj_request->bio_list) | |
2088 | bio_chain_put(obj_request->bio_list); | |
2089 | break; | |
788e2df3 AE |
2090 | case OBJ_REQUEST_PAGES: |
2091 | if (obj_request->pages) | |
2092 | ceph_release_page_vector(obj_request->pages, | |
2093 | obj_request->page_count); | |
2094 | break; | |
bf0d5f50 AE |
2095 | } |
2096 | ||
f907ad55 | 2097 | kfree(obj_request->object_name); |
868311b1 AE |
2098 | obj_request->object_name = NULL; |
2099 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2100 | } |
2101 | ||
fb65d228 AE |
2102 | /* It's OK to call this for a device with no parent */ |
2103 | ||
2104 | static void rbd_spec_put(struct rbd_spec *spec); | |
2105 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2106 | { | |
2107 | rbd_dev_remove_parent(rbd_dev); | |
2108 | rbd_spec_put(rbd_dev->parent_spec); | |
2109 | rbd_dev->parent_spec = NULL; | |
2110 | rbd_dev->parent_overlap = 0; | |
2111 | } | |
2112 | ||
a2acd00e AE |
2113 | /* |
2114 | * Parent image reference counting is used to determine when an | |
2115 | * image's parent fields can be safely torn down--after there are no | |
2116 | * more in-flight requests to the parent image. When the last | |
2117 | * reference is dropped, cleaning them up is safe. | |
2118 | */ | |
2119 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2120 | { | |
2121 | int counter; | |
2122 | ||
2123 | if (!rbd_dev->parent_spec) | |
2124 | return; | |
2125 | ||
2126 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2127 | if (counter > 0) | |
2128 | return; | |
2129 | ||
2130 | /* Last reference; clean up parent data structures */ | |
2131 | ||
2132 | if (!counter) | |
2133 | rbd_dev_unparent(rbd_dev); | |
2134 | else | |
9584d508 | 2135 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2136 | } |
2137 | ||
2138 | /* | |
2139 | * If an image has a non-zero parent overlap, get a reference to its | |
2140 | * parent. | |
2141 | * | |
2142 | * Returns true if the rbd device has a parent with a non-zero | |
2143 | * overlap and a reference for it was successfully taken, or | |
2144 | * false otherwise. | |
2145 | */ | |
2146 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2147 | { | |
ae43e9d0 | 2148 | int counter = 0; |
a2acd00e AE |
2149 | |
2150 | if (!rbd_dev->parent_spec) | |
2151 | return false; | |
2152 | ||
ae43e9d0 ID |
2153 | down_read(&rbd_dev->header_rwsem); |
2154 | if (rbd_dev->parent_overlap) | |
2155 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2156 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2157 | |
2158 | if (counter < 0) | |
9584d508 | 2159 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2160 | |
ae43e9d0 | 2161 | return counter > 0; |
a2acd00e AE |
2162 | } |
2163 | ||
bf0d5f50 AE |
2164 | /* |
2165 | * Caller is responsible for filling in the list of object requests | |
2166 | * that comprises the image request, and the Linux request pointer | |
2167 | * (if there is one). | |
2168 | */ | |
cc344fa1 AE |
2169 | static struct rbd_img_request *rbd_img_request_create( |
2170 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2171 | u64 offset, u64 length, |
6d2940c8 | 2172 | enum obj_operation_type op_type, |
4e752f0a | 2173 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2174 | { |
2175 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2176 | |
7a716aac | 2177 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2178 | if (!img_request) |
2179 | return NULL; | |
2180 | ||
bf0d5f50 AE |
2181 | img_request->rq = NULL; |
2182 | img_request->rbd_dev = rbd_dev; | |
2183 | img_request->offset = offset; | |
2184 | img_request->length = length; | |
0c425248 | 2185 | img_request->flags = 0; |
90e98c52 GZ |
2186 | if (op_type == OBJ_OP_DISCARD) { |
2187 | img_request_discard_set(img_request); | |
2188 | img_request->snapc = snapc; | |
2189 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2190 | img_request_write_set(img_request); |
4e752f0a | 2191 | img_request->snapc = snapc; |
0c425248 | 2192 | } else { |
bf0d5f50 | 2193 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2194 | } |
a2acd00e | 2195 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2196 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2197 | spin_lock_init(&img_request->completion_lock); |
2198 | img_request->next_completion = 0; | |
2199 | img_request->callback = NULL; | |
a5a337d4 | 2200 | img_request->result = 0; |
bf0d5f50 AE |
2201 | img_request->obj_request_count = 0; |
2202 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2203 | kref_init(&img_request->kref); | |
2204 | ||
37206ee5 | 2205 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2206 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2207 | |
bf0d5f50 AE |
2208 | return img_request; |
2209 | } | |
2210 | ||
2211 | static void rbd_img_request_destroy(struct kref *kref) | |
2212 | { | |
2213 | struct rbd_img_request *img_request; | |
2214 | struct rbd_obj_request *obj_request; | |
2215 | struct rbd_obj_request *next_obj_request; | |
2216 | ||
2217 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2218 | ||
37206ee5 AE |
2219 | dout("%s: img %p\n", __func__, img_request); |
2220 | ||
bf0d5f50 AE |
2221 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2222 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2223 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2224 | |
a2acd00e AE |
2225 | if (img_request_layered_test(img_request)) { |
2226 | img_request_layered_clear(img_request); | |
2227 | rbd_dev_parent_put(img_request->rbd_dev); | |
2228 | } | |
2229 | ||
bef95455 JD |
2230 | if (img_request_write_test(img_request) || |
2231 | img_request_discard_test(img_request)) | |
812164f8 | 2232 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2233 | |
1c2a9dfe | 2234 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2235 | } |
2236 | ||
e93f3152 AE |
2237 | static struct rbd_img_request *rbd_parent_request_create( |
2238 | struct rbd_obj_request *obj_request, | |
2239 | u64 img_offset, u64 length) | |
2240 | { | |
2241 | struct rbd_img_request *parent_request; | |
2242 | struct rbd_device *rbd_dev; | |
2243 | ||
2244 | rbd_assert(obj_request->img_request); | |
2245 | rbd_dev = obj_request->img_request->rbd_dev; | |
2246 | ||
4e752f0a | 2247 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2248 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2249 | if (!parent_request) |
2250 | return NULL; | |
2251 | ||
2252 | img_request_child_set(parent_request); | |
2253 | rbd_obj_request_get(obj_request); | |
2254 | parent_request->obj_request = obj_request; | |
2255 | ||
2256 | return parent_request; | |
2257 | } | |
2258 | ||
2259 | static void rbd_parent_request_destroy(struct kref *kref) | |
2260 | { | |
2261 | struct rbd_img_request *parent_request; | |
2262 | struct rbd_obj_request *orig_request; | |
2263 | ||
2264 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2265 | orig_request = parent_request->obj_request; | |
2266 | ||
2267 | parent_request->obj_request = NULL; | |
2268 | rbd_obj_request_put(orig_request); | |
2269 | img_request_child_clear(parent_request); | |
2270 | ||
2271 | rbd_img_request_destroy(kref); | |
2272 | } | |
2273 | ||
1217857f AE |
2274 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2275 | { | |
6365d33a | 2276 | struct rbd_img_request *img_request; |
1217857f AE |
2277 | unsigned int xferred; |
2278 | int result; | |
8b3e1a56 | 2279 | bool more; |
1217857f | 2280 | |
6365d33a AE |
2281 | rbd_assert(obj_request_img_data_test(obj_request)); |
2282 | img_request = obj_request->img_request; | |
2283 | ||
1217857f AE |
2284 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2285 | xferred = (unsigned int)obj_request->xferred; | |
2286 | result = obj_request->result; | |
2287 | if (result) { | |
2288 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2289 | enum obj_operation_type op_type; |
2290 | ||
90e98c52 GZ |
2291 | if (img_request_discard_test(img_request)) |
2292 | op_type = OBJ_OP_DISCARD; | |
2293 | else if (img_request_write_test(img_request)) | |
2294 | op_type = OBJ_OP_WRITE; | |
2295 | else | |
2296 | op_type = OBJ_OP_READ; | |
1217857f | 2297 | |
9584d508 | 2298 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2299 | obj_op_name(op_type), obj_request->length, |
2300 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2301 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2302 | result, xferred); |
2303 | if (!img_request->result) | |
2304 | img_request->result = result; | |
082a75da ID |
2305 | /* |
2306 | * Need to end I/O on the entire obj_request worth of | |
2307 | * bytes in case of error. | |
2308 | */ | |
2309 | xferred = obj_request->length; | |
1217857f AE |
2310 | } |
2311 | ||
f1a4739f AE |
2312 | /* Image object requests don't own their page array */ |
2313 | ||
2314 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2315 | obj_request->pages = NULL; | |
2316 | obj_request->page_count = 0; | |
2317 | } | |
2318 | ||
8b3e1a56 AE |
2319 | if (img_request_child_test(img_request)) { |
2320 | rbd_assert(img_request->obj_request != NULL); | |
2321 | more = obj_request->which < img_request->obj_request_count - 1; | |
2322 | } else { | |
2323 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2324 | |
2325 | more = blk_update_request(img_request->rq, result, xferred); | |
2326 | if (!more) | |
2327 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2328 | } |
2329 | ||
2330 | return more; | |
1217857f AE |
2331 | } |
2332 | ||
2169238d AE |
2333 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2334 | { | |
2335 | struct rbd_img_request *img_request; | |
2336 | u32 which = obj_request->which; | |
2337 | bool more = true; | |
2338 | ||
6365d33a | 2339 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2340 | img_request = obj_request->img_request; |
2341 | ||
2342 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2343 | rbd_assert(img_request != NULL); | |
2169238d AE |
2344 | rbd_assert(img_request->obj_request_count > 0); |
2345 | rbd_assert(which != BAD_WHICH); | |
2346 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2347 | |
2348 | spin_lock_irq(&img_request->completion_lock); | |
2349 | if (which != img_request->next_completion) | |
2350 | goto out; | |
2351 | ||
2352 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2353 | rbd_assert(more); |
2354 | rbd_assert(which < img_request->obj_request_count); | |
2355 | ||
2356 | if (!obj_request_done_test(obj_request)) | |
2357 | break; | |
1217857f | 2358 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2359 | which++; |
2360 | } | |
2361 | ||
2362 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2363 | img_request->next_completion = which; | |
2364 | out: | |
2365 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2366 | rbd_img_request_put(img_request); |
2169238d AE |
2367 | |
2368 | if (!more) | |
2369 | rbd_img_request_complete(img_request); | |
2370 | } | |
2371 | ||
3b434a2a JD |
2372 | /* |
2373 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2374 | * them for submission. num_ops is the current number of | |
2375 | * osd operations already to the object request. | |
2376 | */ | |
2377 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2378 | struct ceph_osd_request *osd_request, | |
2379 | enum obj_operation_type op_type, | |
2380 | unsigned int num_ops) | |
2381 | { | |
2382 | struct rbd_img_request *img_request = obj_request->img_request; | |
2383 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2384 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2385 | u64 offset = obj_request->offset; | |
2386 | u64 length = obj_request->length; | |
2387 | u64 img_end; | |
2388 | u16 opcode; | |
2389 | ||
2390 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2391 | if (!offset && length == object_size && |
2392 | (!img_request_layered_test(img_request) || | |
2393 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2394 | opcode = CEPH_OSD_OP_DELETE; |
2395 | } else if ((offset + length == object_size)) { | |
2396 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2397 | } else { | |
2398 | down_read(&rbd_dev->header_rwsem); | |
2399 | img_end = rbd_dev->header.image_size; | |
2400 | up_read(&rbd_dev->header_rwsem); | |
2401 | ||
2402 | if (obj_request->img_offset + length == img_end) | |
2403 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2404 | else | |
2405 | opcode = CEPH_OSD_OP_ZERO; | |
2406 | } | |
2407 | } else if (op_type == OBJ_OP_WRITE) { | |
e30b7577 ID |
2408 | if (!offset && length == object_size) |
2409 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2410 | else | |
2411 | opcode = CEPH_OSD_OP_WRITE; | |
3b434a2a JD |
2412 | osd_req_op_alloc_hint_init(osd_request, num_ops, |
2413 | object_size, object_size); | |
2414 | num_ops++; | |
2415 | } else { | |
2416 | opcode = CEPH_OSD_OP_READ; | |
2417 | } | |
2418 | ||
7e868b6e | 2419 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2420 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2421 | else |
2422 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2423 | offset, length, 0, 0); | |
2424 | ||
3b434a2a JD |
2425 | if (obj_request->type == OBJ_REQUEST_BIO) |
2426 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2427 | obj_request->bio_list, length); | |
2428 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2429 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2430 | obj_request->pages, length, | |
2431 | offset & ~PAGE_MASK, false, false); | |
2432 | ||
2433 | /* Discards are also writes */ | |
2434 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2435 | rbd_osd_req_format_write(obj_request); | |
2436 | else | |
2437 | rbd_osd_req_format_read(obj_request); | |
2438 | } | |
2439 | ||
f1a4739f AE |
2440 | /* |
2441 | * Split up an image request into one or more object requests, each | |
2442 | * to a different object. The "type" parameter indicates whether | |
2443 | * "data_desc" is the pointer to the head of a list of bio | |
2444 | * structures, or the base of a page array. In either case this | |
2445 | * function assumes data_desc describes memory sufficient to hold | |
2446 | * all data described by the image request. | |
2447 | */ | |
2448 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2449 | enum obj_request_type type, | |
2450 | void *data_desc) | |
bf0d5f50 AE |
2451 | { |
2452 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2453 | struct rbd_obj_request *obj_request = NULL; | |
2454 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2455 | struct bio *bio_list = NULL; |
f1a4739f | 2456 | unsigned int bio_offset = 0; |
a158073c | 2457 | struct page **pages = NULL; |
6d2940c8 | 2458 | enum obj_operation_type op_type; |
7da22d29 | 2459 | u64 img_offset; |
bf0d5f50 | 2460 | u64 resid; |
bf0d5f50 | 2461 | |
f1a4739f AE |
2462 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2463 | (int)type, data_desc); | |
37206ee5 | 2464 | |
7da22d29 | 2465 | img_offset = img_request->offset; |
bf0d5f50 | 2466 | resid = img_request->length; |
4dda41d3 | 2467 | rbd_assert(resid > 0); |
3b434a2a | 2468 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2469 | |
2470 | if (type == OBJ_REQUEST_BIO) { | |
2471 | bio_list = data_desc; | |
4f024f37 KO |
2472 | rbd_assert(img_offset == |
2473 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2474 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2475 | pages = data_desc; |
2476 | } | |
2477 | ||
bf0d5f50 | 2478 | while (resid) { |
2fa12320 | 2479 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2480 | const char *object_name; |
bf0d5f50 AE |
2481 | u64 offset; |
2482 | u64 length; | |
2483 | ||
7da22d29 | 2484 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2485 | if (!object_name) |
2486 | goto out_unwind; | |
7da22d29 AE |
2487 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2488 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2489 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2490 | offset, length, type); |
78c2a44a AE |
2491 | /* object request has its own copy of the object name */ |
2492 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2493 | if (!obj_request) |
2494 | goto out_unwind; | |
62054da6 | 2495 | |
03507db6 JD |
2496 | /* |
2497 | * set obj_request->img_request before creating the | |
2498 | * osd_request so that it gets the right snapc | |
2499 | */ | |
2500 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2501 | |
f1a4739f AE |
2502 | if (type == OBJ_REQUEST_BIO) { |
2503 | unsigned int clone_size; | |
2504 | ||
2505 | rbd_assert(length <= (u64)UINT_MAX); | |
2506 | clone_size = (unsigned int)length; | |
2507 | obj_request->bio_list = | |
2508 | bio_chain_clone_range(&bio_list, | |
2509 | &bio_offset, | |
2510 | clone_size, | |
2511 | GFP_ATOMIC); | |
2512 | if (!obj_request->bio_list) | |
62054da6 | 2513 | goto out_unwind; |
90e98c52 | 2514 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2515 | unsigned int page_count; |
2516 | ||
2517 | obj_request->pages = pages; | |
2518 | page_count = (u32)calc_pages_for(offset, length); | |
2519 | obj_request->page_count = page_count; | |
2520 | if ((offset + length) & ~PAGE_MASK) | |
2521 | page_count--; /* more on last page */ | |
2522 | pages += page_count; | |
2523 | } | |
bf0d5f50 | 2524 | |
6d2940c8 GZ |
2525 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2526 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2527 | obj_request); | |
2fa12320 | 2528 | if (!osd_req) |
62054da6 | 2529 | goto out_unwind; |
3b434a2a | 2530 | |
2fa12320 | 2531 | obj_request->osd_req = osd_req; |
2169238d | 2532 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2533 | obj_request->img_offset = img_offset; |
9d4df01f | 2534 | |
3b434a2a | 2535 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2536 | |
3b434a2a | 2537 | rbd_img_request_get(img_request); |
bf0d5f50 | 2538 | |
7da22d29 | 2539 | img_offset += length; |
bf0d5f50 AE |
2540 | resid -= length; |
2541 | } | |
2542 | ||
2543 | return 0; | |
2544 | ||
bf0d5f50 AE |
2545 | out_unwind: |
2546 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2547 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2548 | |
2549 | return -ENOMEM; | |
2550 | } | |
2551 | ||
0eefd470 | 2552 | static void |
2761713d | 2553 | rbd_osd_copyup_callback(struct rbd_obj_request *obj_request) |
0eefd470 AE |
2554 | { |
2555 | struct rbd_img_request *img_request; | |
2556 | struct rbd_device *rbd_dev; | |
ebda6408 | 2557 | struct page **pages; |
0eefd470 AE |
2558 | u32 page_count; |
2559 | ||
2761713d ID |
2560 | dout("%s: obj %p\n", __func__, obj_request); |
2561 | ||
d3246fb0 JD |
2562 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2563 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2564 | rbd_assert(obj_request_img_data_test(obj_request)); |
2565 | img_request = obj_request->img_request; | |
2566 | rbd_assert(img_request); | |
2567 | ||
2568 | rbd_dev = img_request->rbd_dev; | |
2569 | rbd_assert(rbd_dev); | |
0eefd470 | 2570 | |
ebda6408 AE |
2571 | pages = obj_request->copyup_pages; |
2572 | rbd_assert(pages != NULL); | |
0eefd470 | 2573 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2574 | page_count = obj_request->copyup_page_count; |
2575 | rbd_assert(page_count); | |
2576 | obj_request->copyup_page_count = 0; | |
2577 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2578 | |
2579 | /* | |
2580 | * We want the transfer count to reflect the size of the | |
2581 | * original write request. There is no such thing as a | |
2582 | * successful short write, so if the request was successful | |
2583 | * we can just set it to the originally-requested length. | |
2584 | */ | |
2585 | if (!obj_request->result) | |
2586 | obj_request->xferred = obj_request->length; | |
2587 | ||
2761713d | 2588 | obj_request_done_set(obj_request); |
0eefd470 AE |
2589 | } |
2590 | ||
3d7efd18 AE |
2591 | static void |
2592 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2593 | { | |
2594 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2595 | struct ceph_osd_request *osd_req; |
2596 | struct ceph_osd_client *osdc; | |
2597 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2598 | struct page **pages; |
d3246fb0 | 2599 | enum obj_operation_type op_type; |
ebda6408 | 2600 | u32 page_count; |
bbea1c1a | 2601 | int img_result; |
ebda6408 | 2602 | u64 parent_length; |
3d7efd18 AE |
2603 | |
2604 | rbd_assert(img_request_child_test(img_request)); | |
2605 | ||
2606 | /* First get what we need from the image request */ | |
2607 | ||
2608 | pages = img_request->copyup_pages; | |
2609 | rbd_assert(pages != NULL); | |
2610 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2611 | page_count = img_request->copyup_page_count; |
2612 | rbd_assert(page_count); | |
2613 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2614 | |
2615 | orig_request = img_request->obj_request; | |
2616 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2617 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2618 | img_result = img_request->result; |
ebda6408 AE |
2619 | parent_length = img_request->length; |
2620 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2621 | rbd_img_request_put(img_request); |
3d7efd18 | 2622 | |
91c6febb AE |
2623 | rbd_assert(orig_request->img_request); |
2624 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2625 | rbd_assert(rbd_dev); |
0eefd470 | 2626 | |
bbea1c1a AE |
2627 | /* |
2628 | * If the overlap has become 0 (most likely because the | |
2629 | * image has been flattened) we need to free the pages | |
2630 | * and re-submit the original write request. | |
2631 | */ | |
2632 | if (!rbd_dev->parent_overlap) { | |
2633 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2634 | |
bbea1c1a AE |
2635 | ceph_release_page_vector(pages, page_count); |
2636 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2637 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2638 | if (!img_result) | |
2639 | return; | |
2640 | } | |
0eefd470 | 2641 | |
bbea1c1a | 2642 | if (img_result) |
0eefd470 | 2643 | goto out_err; |
0eefd470 | 2644 | |
8785b1d4 AE |
2645 | /* |
2646 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2647 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2648 | * request. Allocate the new copyup osd request for the |
2649 | * original request, and release the old one. | |
2650 | */ | |
bbea1c1a | 2651 | img_result = -ENOMEM; |
0eefd470 AE |
2652 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2653 | if (!osd_req) | |
2654 | goto out_err; | |
8785b1d4 | 2655 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2656 | orig_request->osd_req = osd_req; |
2657 | orig_request->copyup_pages = pages; | |
ebda6408 | 2658 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2659 | |
0eefd470 | 2660 | /* Initialize the copyup op */ |
3d7efd18 | 2661 | |
0eefd470 | 2662 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2663 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2664 | false, false); |
3d7efd18 | 2665 | |
d3246fb0 | 2666 | /* Add the other op(s) */ |
0eefd470 | 2667 | |
d3246fb0 JD |
2668 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2669 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2670 | |
2671 | /* All set, send it off. */ | |
2672 | ||
0eefd470 | 2673 | osdc = &rbd_dev->rbd_client->client->osdc; |
bbea1c1a AE |
2674 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2675 | if (!img_result) | |
0eefd470 AE |
2676 | return; |
2677 | out_err: | |
2678 | /* Record the error code and complete the request */ | |
2679 | ||
bbea1c1a | 2680 | orig_request->result = img_result; |
0eefd470 AE |
2681 | orig_request->xferred = 0; |
2682 | obj_request_done_set(orig_request); | |
2683 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2684 | } |
2685 | ||
2686 | /* | |
2687 | * Read from the parent image the range of data that covers the | |
2688 | * entire target of the given object request. This is used for | |
2689 | * satisfying a layered image write request when the target of an | |
2690 | * object request from the image request does not exist. | |
2691 | * | |
2692 | * A page array big enough to hold the returned data is allocated | |
2693 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2694 | * When the read completes, this page array will be transferred to | |
2695 | * the original object request for the copyup operation. | |
2696 | * | |
2697 | * If an error occurs, record it as the result of the original | |
2698 | * object request and mark it done so it gets completed. | |
2699 | */ | |
2700 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2701 | { | |
2702 | struct rbd_img_request *img_request = NULL; | |
2703 | struct rbd_img_request *parent_request = NULL; | |
2704 | struct rbd_device *rbd_dev; | |
2705 | u64 img_offset; | |
2706 | u64 length; | |
2707 | struct page **pages = NULL; | |
2708 | u32 page_count; | |
2709 | int result; | |
2710 | ||
2711 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2712 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2713 | |
2714 | img_request = obj_request->img_request; | |
2715 | rbd_assert(img_request != NULL); | |
2716 | rbd_dev = img_request->rbd_dev; | |
2717 | rbd_assert(rbd_dev->parent != NULL); | |
2718 | ||
2719 | /* | |
2720 | * Determine the byte range covered by the object in the | |
2721 | * child image to which the original request was to be sent. | |
2722 | */ | |
2723 | img_offset = obj_request->img_offset - obj_request->offset; | |
2724 | length = (u64)1 << rbd_dev->header.obj_order; | |
2725 | ||
a9e8ba2c AE |
2726 | /* |
2727 | * There is no defined parent data beyond the parent | |
2728 | * overlap, so limit what we read at that boundary if | |
2729 | * necessary. | |
2730 | */ | |
2731 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2732 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2733 | length = rbd_dev->parent_overlap - img_offset; | |
2734 | } | |
2735 | ||
3d7efd18 AE |
2736 | /* |
2737 | * Allocate a page array big enough to receive the data read | |
2738 | * from the parent. | |
2739 | */ | |
2740 | page_count = (u32)calc_pages_for(0, length); | |
2741 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2742 | if (IS_ERR(pages)) { | |
2743 | result = PTR_ERR(pages); | |
2744 | pages = NULL; | |
2745 | goto out_err; | |
2746 | } | |
2747 | ||
2748 | result = -ENOMEM; | |
e93f3152 AE |
2749 | parent_request = rbd_parent_request_create(obj_request, |
2750 | img_offset, length); | |
3d7efd18 AE |
2751 | if (!parent_request) |
2752 | goto out_err; | |
3d7efd18 AE |
2753 | |
2754 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2755 | if (result) | |
2756 | goto out_err; | |
2757 | parent_request->copyup_pages = pages; | |
ebda6408 | 2758 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2759 | |
2760 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2761 | result = rbd_img_request_submit(parent_request); | |
2762 | if (!result) | |
2763 | return 0; | |
2764 | ||
2765 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2766 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2767 | parent_request->obj_request = NULL; |
2768 | rbd_obj_request_put(obj_request); | |
2769 | out_err: | |
2770 | if (pages) | |
2771 | ceph_release_page_vector(pages, page_count); | |
2772 | if (parent_request) | |
2773 | rbd_img_request_put(parent_request); | |
2774 | obj_request->result = result; | |
2775 | obj_request->xferred = 0; | |
2776 | obj_request_done_set(obj_request); | |
2777 | ||
2778 | return result; | |
2779 | } | |
2780 | ||
c5b5ef6c AE |
2781 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2782 | { | |
c5b5ef6c | 2783 | struct rbd_obj_request *orig_request; |
638f5abe | 2784 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2785 | int result; |
2786 | ||
2787 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2788 | ||
2789 | /* | |
2790 | * All we need from the object request is the original | |
2791 | * request and the result of the STAT op. Grab those, then | |
2792 | * we're done with the request. | |
2793 | */ | |
2794 | orig_request = obj_request->obj_request; | |
2795 | obj_request->obj_request = NULL; | |
912c317d | 2796 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2797 | rbd_assert(orig_request); |
2798 | rbd_assert(orig_request->img_request); | |
2799 | ||
2800 | result = obj_request->result; | |
2801 | obj_request->result = 0; | |
2802 | ||
2803 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2804 | obj_request, orig_request, result, | |
2805 | obj_request->xferred, obj_request->length); | |
2806 | rbd_obj_request_put(obj_request); | |
2807 | ||
638f5abe AE |
2808 | /* |
2809 | * If the overlap has become 0 (most likely because the | |
2810 | * image has been flattened) we need to free the pages | |
2811 | * and re-submit the original write request. | |
2812 | */ | |
2813 | rbd_dev = orig_request->img_request->rbd_dev; | |
2814 | if (!rbd_dev->parent_overlap) { | |
2815 | struct ceph_osd_client *osdc; | |
2816 | ||
638f5abe AE |
2817 | osdc = &rbd_dev->rbd_client->client->osdc; |
2818 | result = rbd_obj_request_submit(osdc, orig_request); | |
2819 | if (!result) | |
2820 | return; | |
2821 | } | |
c5b5ef6c AE |
2822 | |
2823 | /* | |
2824 | * Our only purpose here is to determine whether the object | |
2825 | * exists, and we don't want to treat the non-existence as | |
2826 | * an error. If something else comes back, transfer the | |
2827 | * error to the original request and complete it now. | |
2828 | */ | |
2829 | if (!result) { | |
2830 | obj_request_existence_set(orig_request, true); | |
2831 | } else if (result == -ENOENT) { | |
2832 | obj_request_existence_set(orig_request, false); | |
2833 | } else if (result) { | |
2834 | orig_request->result = result; | |
3d7efd18 | 2835 | goto out; |
c5b5ef6c AE |
2836 | } |
2837 | ||
2838 | /* | |
2839 | * Resubmit the original request now that we have recorded | |
2840 | * whether the target object exists. | |
2841 | */ | |
b454e36d | 2842 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2843 | out: |
c5b5ef6c AE |
2844 | if (orig_request->result) |
2845 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2846 | } |
2847 | ||
2848 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2849 | { | |
2850 | struct rbd_obj_request *stat_request; | |
2851 | struct rbd_device *rbd_dev; | |
2852 | struct ceph_osd_client *osdc; | |
2853 | struct page **pages = NULL; | |
2854 | u32 page_count; | |
2855 | size_t size; | |
2856 | int ret; | |
2857 | ||
2858 | /* | |
2859 | * The response data for a STAT call consists of: | |
2860 | * le64 length; | |
2861 | * struct { | |
2862 | * le32 tv_sec; | |
2863 | * le32 tv_nsec; | |
2864 | * } mtime; | |
2865 | */ | |
2866 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2867 | page_count = (u32)calc_pages_for(0, size); | |
2868 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2869 | if (IS_ERR(pages)) | |
2870 | return PTR_ERR(pages); | |
2871 | ||
2872 | ret = -ENOMEM; | |
2873 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2874 | OBJ_REQUEST_PAGES); | |
2875 | if (!stat_request) | |
2876 | goto out; | |
2877 | ||
2878 | rbd_obj_request_get(obj_request); | |
2879 | stat_request->obj_request = obj_request; | |
2880 | stat_request->pages = pages; | |
2881 | stat_request->page_count = page_count; | |
2882 | ||
2883 | rbd_assert(obj_request->img_request); | |
2884 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2885 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2886 | stat_request); |
c5b5ef6c AE |
2887 | if (!stat_request->osd_req) |
2888 | goto out; | |
2889 | stat_request->callback = rbd_img_obj_exists_callback; | |
2890 | ||
144cba14 | 2891 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2892 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2893 | false, false); | |
9d4df01f | 2894 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2895 | |
2896 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2897 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2898 | out: | |
2899 | if (ret) | |
2900 | rbd_obj_request_put(obj_request); | |
2901 | ||
2902 | return ret; | |
2903 | } | |
2904 | ||
70d045f6 | 2905 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2906 | { |
2907 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2908 | struct rbd_device *rbd_dev; |
b454e36d AE |
2909 | |
2910 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2911 | ||
2912 | img_request = obj_request->img_request; | |
2913 | rbd_assert(img_request); | |
a9e8ba2c | 2914 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2915 | |
70d045f6 | 2916 | /* Reads */ |
1c220881 JD |
2917 | if (!img_request_write_test(img_request) && |
2918 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2919 | return true; |
2920 | ||
2921 | /* Non-layered writes */ | |
2922 | if (!img_request_layered_test(img_request)) | |
2923 | return true; | |
2924 | ||
b454e36d | 2925 | /* |
70d045f6 ID |
2926 | * Layered writes outside of the parent overlap range don't |
2927 | * share any data with the parent. | |
b454e36d | 2928 | */ |
70d045f6 ID |
2929 | if (!obj_request_overlaps_parent(obj_request)) |
2930 | return true; | |
b454e36d | 2931 | |
c622d226 GZ |
2932 | /* |
2933 | * Entire-object layered writes - we will overwrite whatever | |
2934 | * parent data there is anyway. | |
2935 | */ | |
2936 | if (!obj_request->offset && | |
2937 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2938 | return true; | |
2939 | ||
70d045f6 ID |
2940 | /* |
2941 | * If the object is known to already exist, its parent data has | |
2942 | * already been copied. | |
2943 | */ | |
2944 | if (obj_request_known_test(obj_request) && | |
2945 | obj_request_exists_test(obj_request)) | |
2946 | return true; | |
2947 | ||
2948 | return false; | |
2949 | } | |
2950 | ||
2951 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2952 | { | |
2953 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2954 | struct rbd_device *rbd_dev; |
2955 | struct ceph_osd_client *osdc; | |
2956 | ||
2957 | rbd_dev = obj_request->img_request->rbd_dev; | |
2958 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2959 | ||
2960 | return rbd_obj_request_submit(osdc, obj_request); | |
2961 | } | |
2962 | ||
2963 | /* | |
3d7efd18 AE |
2964 | * It's a layered write. The target object might exist but |
2965 | * we may not know that yet. If we know it doesn't exist, | |
2966 | * start by reading the data for the full target object from | |
2967 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2968 | */ |
70d045f6 | 2969 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2970 | return rbd_img_obj_parent_read_full(obj_request); |
2971 | ||
2972 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2973 | |
2974 | return rbd_img_obj_exists_submit(obj_request); | |
2975 | } | |
2976 | ||
bf0d5f50 AE |
2977 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2978 | { | |
bf0d5f50 | 2979 | struct rbd_obj_request *obj_request; |
46faeed4 | 2980 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2981 | |
37206ee5 | 2982 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2983 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2984 | int ret; |
2985 | ||
b454e36d | 2986 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2987 | if (ret) |
2988 | return ret; | |
bf0d5f50 AE |
2989 | } |
2990 | ||
2991 | return 0; | |
2992 | } | |
8b3e1a56 AE |
2993 | |
2994 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2995 | { | |
2996 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2997 | struct rbd_device *rbd_dev; |
2998 | u64 obj_end; | |
02c74fba AE |
2999 | u64 img_xferred; |
3000 | int img_result; | |
8b3e1a56 AE |
3001 | |
3002 | rbd_assert(img_request_child_test(img_request)); | |
3003 | ||
02c74fba AE |
3004 | /* First get what we need from the image request and release it */ |
3005 | ||
8b3e1a56 | 3006 | obj_request = img_request->obj_request; |
02c74fba AE |
3007 | img_xferred = img_request->xferred; |
3008 | img_result = img_request->result; | |
3009 | rbd_img_request_put(img_request); | |
3010 | ||
3011 | /* | |
3012 | * If the overlap has become 0 (most likely because the | |
3013 | * image has been flattened) we need to re-submit the | |
3014 | * original request. | |
3015 | */ | |
a9e8ba2c AE |
3016 | rbd_assert(obj_request); |
3017 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
3018 | rbd_dev = obj_request->img_request->rbd_dev; |
3019 | if (!rbd_dev->parent_overlap) { | |
3020 | struct ceph_osd_client *osdc; | |
3021 | ||
3022 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3023 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
3024 | if (!img_result) | |
3025 | return; | |
3026 | } | |
a9e8ba2c | 3027 | |
02c74fba | 3028 | obj_request->result = img_result; |
a9e8ba2c AE |
3029 | if (obj_request->result) |
3030 | goto out; | |
3031 | ||
3032 | /* | |
3033 | * We need to zero anything beyond the parent overlap | |
3034 | * boundary. Since rbd_img_obj_request_read_callback() | |
3035 | * will zero anything beyond the end of a short read, an | |
3036 | * easy way to do this is to pretend the data from the | |
3037 | * parent came up short--ending at the overlap boundary. | |
3038 | */ | |
3039 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3040 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3041 | if (obj_end > rbd_dev->parent_overlap) { |
3042 | u64 xferred = 0; | |
3043 | ||
3044 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3045 | xferred = rbd_dev->parent_overlap - | |
3046 | obj_request->img_offset; | |
8b3e1a56 | 3047 | |
02c74fba | 3048 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3049 | } else { |
02c74fba | 3050 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3051 | } |
3052 | out: | |
8b3e1a56 AE |
3053 | rbd_img_obj_request_read_callback(obj_request); |
3054 | rbd_obj_request_complete(obj_request); | |
3055 | } | |
3056 | ||
3057 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3058 | { | |
8b3e1a56 AE |
3059 | struct rbd_img_request *img_request; |
3060 | int result; | |
3061 | ||
3062 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3063 | rbd_assert(obj_request->img_request != NULL); | |
3064 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3065 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3066 | |
8b3e1a56 | 3067 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3068 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3069 | obj_request->img_offset, |
e93f3152 | 3070 | obj_request->length); |
8b3e1a56 AE |
3071 | result = -ENOMEM; |
3072 | if (!img_request) | |
3073 | goto out_err; | |
3074 | ||
5b2ab72d AE |
3075 | if (obj_request->type == OBJ_REQUEST_BIO) |
3076 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3077 | obj_request->bio_list); | |
3078 | else | |
3079 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3080 | obj_request->pages); | |
8b3e1a56 AE |
3081 | if (result) |
3082 | goto out_err; | |
3083 | ||
3084 | img_request->callback = rbd_img_parent_read_callback; | |
3085 | result = rbd_img_request_submit(img_request); | |
3086 | if (result) | |
3087 | goto out_err; | |
3088 | ||
3089 | return; | |
3090 | out_err: | |
3091 | if (img_request) | |
3092 | rbd_img_request_put(img_request); | |
3093 | obj_request->result = result; | |
3094 | obj_request->xferred = 0; | |
3095 | obj_request_done_set(obj_request); | |
3096 | } | |
bf0d5f50 | 3097 | |
20e0af67 | 3098 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3099 | { |
3100 | struct rbd_obj_request *obj_request; | |
2169238d | 3101 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3102 | int ret; |
3103 | ||
3104 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3105 | OBJ_REQUEST_NODATA); | |
3106 | if (!obj_request) | |
3107 | return -ENOMEM; | |
3108 | ||
3109 | ret = -ENOMEM; | |
6d2940c8 | 3110 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3111 | obj_request); |
b8d70035 AE |
3112 | if (!obj_request->osd_req) |
3113 | goto out; | |
3114 | ||
c99d2d4a | 3115 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3116 | notify_id, 0, 0); |
9d4df01f | 3117 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3118 | |
b8d70035 | 3119 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3120 | if (ret) |
20e0af67 JD |
3121 | goto out; |
3122 | ret = rbd_obj_request_wait(obj_request); | |
3123 | out: | |
3124 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3125 | |
3126 | return ret; | |
3127 | } | |
3128 | ||
3129 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3130 | { | |
3131 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3132 | int ret; |
b8d70035 AE |
3133 | |
3134 | if (!rbd_dev) | |
3135 | return; | |
3136 | ||
37206ee5 | 3137 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3138 | rbd_dev->header_name, (unsigned long long)notify_id, |
3139 | (unsigned int)opcode); | |
52bb1f9b ID |
3140 | |
3141 | /* | |
3142 | * Until adequate refresh error handling is in place, there is | |
3143 | * not much we can do here, except warn. | |
3144 | * | |
3145 | * See http://tracker.ceph.com/issues/5040 | |
3146 | */ | |
e627db08 AE |
3147 | ret = rbd_dev_refresh(rbd_dev); |
3148 | if (ret) | |
9584d508 | 3149 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3150 | |
52bb1f9b ID |
3151 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3152 | if (ret) | |
9584d508 | 3153 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3154 | } |
3155 | ||
bb040aa0 ID |
3156 | /* |
3157 | * Send a (un)watch request and wait for the ack. Return a request | |
3158 | * with a ref held on success or error. | |
3159 | */ | |
3160 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3161 | struct rbd_device *rbd_dev, | |
3162 | bool watch) | |
3163 | { | |
3164 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2894e1d7 | 3165 | struct ceph_options *opts = osdc->client->options; |
bb040aa0 ID |
3166 | struct rbd_obj_request *obj_request; |
3167 | int ret; | |
3168 | ||
3169 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3170 | OBJ_REQUEST_NODATA); | |
3171 | if (!obj_request) | |
3172 | return ERR_PTR(-ENOMEM); | |
3173 | ||
6d2940c8 | 3174 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3175 | obj_request); |
3176 | if (!obj_request->osd_req) { | |
3177 | ret = -ENOMEM; | |
3178 | goto out; | |
3179 | } | |
3180 | ||
3181 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3182 | rbd_dev->watch_event->cookie, 0, watch); | |
3183 | rbd_osd_req_format_write(obj_request); | |
3184 | ||
3185 | if (watch) | |
3186 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3187 | ||
3188 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3189 | if (ret) | |
3190 | goto out; | |
3191 | ||
2894e1d7 | 3192 | ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout); |
bb040aa0 ID |
3193 | if (ret) |
3194 | goto out; | |
3195 | ||
3196 | ret = obj_request->result; | |
3197 | if (ret) { | |
3198 | if (watch) | |
3199 | rbd_obj_request_end(obj_request); | |
3200 | goto out; | |
3201 | } | |
3202 | ||
3203 | return obj_request; | |
3204 | ||
3205 | out: | |
3206 | rbd_obj_request_put(obj_request); | |
3207 | return ERR_PTR(ret); | |
3208 | } | |
3209 | ||
9969ebc5 | 3210 | /* |
b30a01f2 | 3211 | * Initiate a watch request, synchronously. |
9969ebc5 | 3212 | */ |
b30a01f2 | 3213 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3214 | { |
3215 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3216 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3217 | int ret; |
3218 | ||
b30a01f2 ID |
3219 | rbd_assert(!rbd_dev->watch_event); |
3220 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3221 | |
b30a01f2 ID |
3222 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3223 | &rbd_dev->watch_event); | |
3224 | if (ret < 0) | |
3225 | return ret; | |
3226 | ||
76756a51 ID |
3227 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3228 | if (IS_ERR(obj_request)) { | |
3229 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3230 | rbd_dev->watch_event = NULL; | |
3231 | return PTR_ERR(obj_request); | |
b30a01f2 | 3232 | } |
9969ebc5 | 3233 | |
8eb87565 AE |
3234 | /* |
3235 | * A watch request is set to linger, so the underlying osd | |
3236 | * request won't go away until we unregister it. We retain | |
3237 | * a pointer to the object request during that time (in | |
76756a51 ID |
3238 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3239 | * We'll drop that reference after we've unregistered it in | |
3240 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3241 | */ |
b30a01f2 | 3242 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3243 | |
b30a01f2 | 3244 | return 0; |
b30a01f2 ID |
3245 | } |
3246 | ||
3247 | /* | |
3248 | * Tear down a watch request, synchronously. | |
3249 | */ | |
76756a51 | 3250 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3251 | { |
b30a01f2 | 3252 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3253 | |
3254 | rbd_assert(rbd_dev->watch_event); | |
3255 | rbd_assert(rbd_dev->watch_request); | |
3256 | ||
76756a51 | 3257 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3258 | rbd_obj_request_put(rbd_dev->watch_request); |
3259 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3260 | |
76756a51 ID |
3261 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3262 | if (!IS_ERR(obj_request)) | |
3263 | rbd_obj_request_put(obj_request); | |
3264 | else | |
3265 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3266 | PTR_ERR(obj_request)); | |
3267 | ||
9969ebc5 AE |
3268 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3269 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3270 | } |
3271 | ||
36be9a76 | 3272 | /* |
f40eb349 AE |
3273 | * Synchronous osd object method call. Returns the number of bytes |
3274 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3275 | */ |
3276 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3277 | const char *object_name, | |
3278 | const char *class_name, | |
3279 | const char *method_name, | |
4157976b | 3280 | const void *outbound, |
36be9a76 | 3281 | size_t outbound_size, |
4157976b | 3282 | void *inbound, |
e2a58ee5 | 3283 | size_t inbound_size) |
36be9a76 | 3284 | { |
2169238d | 3285 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3286 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3287 | struct page **pages; |
3288 | u32 page_count; | |
3289 | int ret; | |
3290 | ||
3291 | /* | |
6010a451 AE |
3292 | * Method calls are ultimately read operations. The result |
3293 | * should placed into the inbound buffer provided. They | |
3294 | * also supply outbound data--parameters for the object | |
3295 | * method. Currently if this is present it will be a | |
3296 | * snapshot id. | |
36be9a76 | 3297 | */ |
57385b51 | 3298 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3299 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3300 | if (IS_ERR(pages)) | |
3301 | return PTR_ERR(pages); | |
3302 | ||
3303 | ret = -ENOMEM; | |
6010a451 | 3304 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3305 | OBJ_REQUEST_PAGES); |
3306 | if (!obj_request) | |
3307 | goto out; | |
3308 | ||
3309 | obj_request->pages = pages; | |
3310 | obj_request->page_count = page_count; | |
3311 | ||
6d2940c8 | 3312 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3313 | obj_request); |
36be9a76 AE |
3314 | if (!obj_request->osd_req) |
3315 | goto out; | |
3316 | ||
c99d2d4a | 3317 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3318 | class_name, method_name); |
3319 | if (outbound_size) { | |
3320 | struct ceph_pagelist *pagelist; | |
3321 | ||
3322 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3323 | if (!pagelist) | |
3324 | goto out; | |
3325 | ||
3326 | ceph_pagelist_init(pagelist); | |
3327 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3328 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3329 | pagelist); | |
3330 | } | |
a4ce40a9 AE |
3331 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3332 | obj_request->pages, inbound_size, | |
44cd188d | 3333 | 0, false, false); |
9d4df01f | 3334 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3335 | |
36be9a76 AE |
3336 | ret = rbd_obj_request_submit(osdc, obj_request); |
3337 | if (ret) | |
3338 | goto out; | |
3339 | ret = rbd_obj_request_wait(obj_request); | |
3340 | if (ret) | |
3341 | goto out; | |
3342 | ||
3343 | ret = obj_request->result; | |
3344 | if (ret < 0) | |
3345 | goto out; | |
57385b51 AE |
3346 | |
3347 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3348 | ret = (int)obj_request->xferred; | |
903bb32e | 3349 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3350 | out: |
3351 | if (obj_request) | |
3352 | rbd_obj_request_put(obj_request); | |
3353 | else | |
3354 | ceph_release_page_vector(pages, page_count); | |
3355 | ||
3356 | return ret; | |
3357 | } | |
3358 | ||
7ad18afa | 3359 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3360 | { |
7ad18afa CH |
3361 | struct request *rq = blk_mq_rq_from_pdu(work); |
3362 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3363 | struct rbd_img_request *img_request; |
4e752f0a | 3364 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3365 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3366 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3367 | enum obj_operation_type op_type; |
4e752f0a | 3368 | u64 mapping_size; |
bf0d5f50 AE |
3369 | int result; |
3370 | ||
7ad18afa CH |
3371 | if (rq->cmd_type != REQ_TYPE_FS) { |
3372 | dout("%s: non-fs request type %d\n", __func__, | |
3373 | (int) rq->cmd_type); | |
3374 | result = -EIO; | |
3375 | goto err; | |
3376 | } | |
3377 | ||
90e98c52 GZ |
3378 | if (rq->cmd_flags & REQ_DISCARD) |
3379 | op_type = OBJ_OP_DISCARD; | |
3380 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3381 | op_type = OBJ_OP_WRITE; |
3382 | else | |
3383 | op_type = OBJ_OP_READ; | |
3384 | ||
bc1ecc65 | 3385 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3386 | |
bc1ecc65 ID |
3387 | if (!length) { |
3388 | dout("%s: zero-length request\n", __func__); | |
3389 | result = 0; | |
3390 | goto err_rq; | |
3391 | } | |
bf0d5f50 | 3392 | |
6d2940c8 | 3393 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3394 | |
6d2940c8 | 3395 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3396 | if (rbd_dev->mapping.read_only) { |
3397 | result = -EROFS; | |
3398 | goto err_rq; | |
4dda41d3 | 3399 | } |
bc1ecc65 ID |
3400 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3401 | } | |
4dda41d3 | 3402 | |
bc1ecc65 ID |
3403 | /* |
3404 | * Quit early if the mapped snapshot no longer exists. It's | |
3405 | * still possible the snapshot will have disappeared by the | |
3406 | * time our request arrives at the osd, but there's no sense in | |
3407 | * sending it if we already know. | |
3408 | */ | |
3409 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3410 | dout("request for non-existent snapshot"); | |
3411 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3412 | result = -ENXIO; | |
3413 | goto err_rq; | |
3414 | } | |
4dda41d3 | 3415 | |
bc1ecc65 ID |
3416 | if (offset && length > U64_MAX - offset + 1) { |
3417 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3418 | length); | |
3419 | result = -EINVAL; | |
3420 | goto err_rq; /* Shouldn't happen */ | |
3421 | } | |
4dda41d3 | 3422 | |
7ad18afa CH |
3423 | blk_mq_start_request(rq); |
3424 | ||
4e752f0a JD |
3425 | down_read(&rbd_dev->header_rwsem); |
3426 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3427 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3428 | snapc = rbd_dev->header.snapc; |
3429 | ceph_get_snap_context(snapc); | |
3430 | } | |
3431 | up_read(&rbd_dev->header_rwsem); | |
3432 | ||
3433 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3434 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3435 | length, mapping_size); |
bc1ecc65 ID |
3436 | result = -EIO; |
3437 | goto err_rq; | |
3438 | } | |
bf0d5f50 | 3439 | |
6d2940c8 | 3440 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3441 | snapc); |
bc1ecc65 ID |
3442 | if (!img_request) { |
3443 | result = -ENOMEM; | |
3444 | goto err_rq; | |
3445 | } | |
3446 | img_request->rq = rq; | |
bf0d5f50 | 3447 | |
90e98c52 GZ |
3448 | if (op_type == OBJ_OP_DISCARD) |
3449 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3450 | NULL); | |
3451 | else | |
3452 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3453 | rq->bio); | |
bc1ecc65 ID |
3454 | if (result) |
3455 | goto err_img_request; | |
bf0d5f50 | 3456 | |
bc1ecc65 ID |
3457 | result = rbd_img_request_submit(img_request); |
3458 | if (result) | |
3459 | goto err_img_request; | |
bf0d5f50 | 3460 | |
bc1ecc65 | 3461 | return; |
bf0d5f50 | 3462 | |
bc1ecc65 ID |
3463 | err_img_request: |
3464 | rbd_img_request_put(img_request); | |
3465 | err_rq: | |
3466 | if (result) | |
3467 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3468 | obj_op_name(op_type), length, offset, result); |
e96a650a | 3469 | ceph_put_snap_context(snapc); |
7ad18afa CH |
3470 | err: |
3471 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 3472 | } |
bf0d5f50 | 3473 | |
7ad18afa CH |
3474 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
3475 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 3476 | { |
7ad18afa CH |
3477 | struct request *rq = bd->rq; |
3478 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 3479 | |
7ad18afa CH |
3480 | queue_work(rbd_wq, work); |
3481 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
3482 | } |
3483 | ||
602adf40 YS |
3484 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
3485 | { | |
3486 | struct gendisk *disk = rbd_dev->disk; | |
3487 | ||
3488 | if (!disk) | |
3489 | return; | |
3490 | ||
a0cab924 AE |
3491 | rbd_dev->disk = NULL; |
3492 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3493 | del_gendisk(disk); |
a0cab924 AE |
3494 | if (disk->queue) |
3495 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 3496 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 3497 | } |
602adf40 YS |
3498 | put_disk(disk); |
3499 | } | |
3500 | ||
788e2df3 AE |
3501 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3502 | const char *object_name, | |
7097f8df | 3503 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3504 | |
3505 | { | |
2169238d | 3506 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3507 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3508 | struct page **pages = NULL; |
3509 | u32 page_count; | |
1ceae7ef | 3510 | size_t size; |
788e2df3 AE |
3511 | int ret; |
3512 | ||
3513 | page_count = (u32) calc_pages_for(offset, length); | |
3514 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3515 | if (IS_ERR(pages)) | |
a8d42056 | 3516 | return PTR_ERR(pages); |
788e2df3 AE |
3517 | |
3518 | ret = -ENOMEM; | |
3519 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3520 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3521 | if (!obj_request) |
3522 | goto out; | |
3523 | ||
3524 | obj_request->pages = pages; | |
3525 | obj_request->page_count = page_count; | |
3526 | ||
6d2940c8 | 3527 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3528 | obj_request); |
788e2df3 AE |
3529 | if (!obj_request->osd_req) |
3530 | goto out; | |
3531 | ||
c99d2d4a AE |
3532 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3533 | offset, length, 0, 0); | |
406e2c9f | 3534 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3535 | obj_request->pages, |
44cd188d AE |
3536 | obj_request->length, |
3537 | obj_request->offset & ~PAGE_MASK, | |
3538 | false, false); | |
9d4df01f | 3539 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3540 | |
788e2df3 AE |
3541 | ret = rbd_obj_request_submit(osdc, obj_request); |
3542 | if (ret) | |
3543 | goto out; | |
3544 | ret = rbd_obj_request_wait(obj_request); | |
3545 | if (ret) | |
3546 | goto out; | |
3547 | ||
3548 | ret = obj_request->result; | |
3549 | if (ret < 0) | |
3550 | goto out; | |
1ceae7ef AE |
3551 | |
3552 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3553 | size = (size_t) obj_request->xferred; | |
903bb32e | 3554 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3555 | rbd_assert(size <= (size_t)INT_MAX); |
3556 | ret = (int)size; | |
788e2df3 AE |
3557 | out: |
3558 | if (obj_request) | |
3559 | rbd_obj_request_put(obj_request); | |
3560 | else | |
3561 | ceph_release_page_vector(pages, page_count); | |
3562 | ||
3563 | return ret; | |
3564 | } | |
3565 | ||
602adf40 | 3566 | /* |
662518b1 AE |
3567 | * Read the complete header for the given rbd device. On successful |
3568 | * return, the rbd_dev->header field will contain up-to-date | |
3569 | * information about the image. | |
602adf40 | 3570 | */ |
99a41ebc | 3571 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3572 | { |
4156d998 | 3573 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3574 | u32 snap_count = 0; |
4156d998 AE |
3575 | u64 names_size = 0; |
3576 | u32 want_count; | |
3577 | int ret; | |
602adf40 | 3578 | |
00f1f36f | 3579 | /* |
4156d998 AE |
3580 | * The complete header will include an array of its 64-bit |
3581 | * snapshot ids, followed by the names of those snapshots as | |
3582 | * a contiguous block of NUL-terminated strings. Note that | |
3583 | * the number of snapshots could change by the time we read | |
3584 | * it in, in which case we re-read it. | |
00f1f36f | 3585 | */ |
4156d998 AE |
3586 | do { |
3587 | size_t size; | |
3588 | ||
3589 | kfree(ondisk); | |
3590 | ||
3591 | size = sizeof (*ondisk); | |
3592 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3593 | size += names_size; | |
3594 | ondisk = kmalloc(size, GFP_KERNEL); | |
3595 | if (!ondisk) | |
662518b1 | 3596 | return -ENOMEM; |
4156d998 | 3597 | |
788e2df3 | 3598 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3599 | 0, size, ondisk); |
4156d998 | 3600 | if (ret < 0) |
662518b1 | 3601 | goto out; |
c0cd10db | 3602 | if ((size_t)ret < size) { |
4156d998 | 3603 | ret = -ENXIO; |
06ecc6cb AE |
3604 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3605 | size, ret); | |
662518b1 | 3606 | goto out; |
4156d998 AE |
3607 | } |
3608 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3609 | ret = -ENXIO; | |
06ecc6cb | 3610 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3611 | goto out; |
81e759fb | 3612 | } |
602adf40 | 3613 | |
4156d998 AE |
3614 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3615 | want_count = snap_count; | |
3616 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3617 | } while (snap_count != want_count); | |
00f1f36f | 3618 | |
662518b1 AE |
3619 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3620 | out: | |
4156d998 AE |
3621 | kfree(ondisk); |
3622 | ||
3623 | return ret; | |
602adf40 YS |
3624 | } |
3625 | ||
15228ede AE |
3626 | /* |
3627 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3628 | * has disappeared from the (just updated) snapshot context. | |
3629 | */ | |
3630 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3631 | { | |
3632 | u64 snap_id; | |
3633 | ||
3634 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3635 | return; | |
3636 | ||
3637 | snap_id = rbd_dev->spec->snap_id; | |
3638 | if (snap_id == CEPH_NOSNAP) | |
3639 | return; | |
3640 | ||
3641 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3642 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3643 | } | |
3644 | ||
9875201e JD |
3645 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3646 | { | |
3647 | sector_t size; | |
3648 | bool removing; | |
3649 | ||
3650 | /* | |
3651 | * Don't hold the lock while doing disk operations, | |
3652 | * or lock ordering will conflict with the bdev mutex via: | |
3653 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3654 | */ | |
3655 | spin_lock_irq(&rbd_dev->lock); | |
3656 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3657 | spin_unlock_irq(&rbd_dev->lock); | |
3658 | /* | |
3659 | * If the device is being removed, rbd_dev->disk has | |
3660 | * been destroyed, so don't try to update its size | |
3661 | */ | |
3662 | if (!removing) { | |
3663 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3664 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3665 | set_capacity(rbd_dev->disk, size); | |
3666 | revalidate_disk(rbd_dev->disk); | |
3667 | } | |
3668 | } | |
3669 | ||
cc4a38bd | 3670 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3671 | { |
e627db08 | 3672 | u64 mapping_size; |
1fe5e993 AE |
3673 | int ret; |
3674 | ||
cfbf6377 | 3675 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3676 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3677 | |
3678 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 3679 | if (ret) |
73e39e4d | 3680 | goto out; |
15228ede | 3681 | |
e8f59b59 ID |
3682 | /* |
3683 | * If there is a parent, see if it has disappeared due to the | |
3684 | * mapped image getting flattened. | |
3685 | */ | |
3686 | if (rbd_dev->parent) { | |
3687 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3688 | if (ret) | |
73e39e4d | 3689 | goto out; |
e8f59b59 ID |
3690 | } |
3691 | ||
5ff1108c | 3692 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 3693 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
3694 | } else { |
3695 | /* validate mapped snapshot's EXISTS flag */ | |
3696 | rbd_exists_validate(rbd_dev); | |
3697 | } | |
15228ede | 3698 | |
73e39e4d | 3699 | out: |
cfbf6377 | 3700 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 3701 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 3702 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3703 | |
73e39e4d | 3704 | return ret; |
1fe5e993 AE |
3705 | } |
3706 | ||
7ad18afa CH |
3707 | static int rbd_init_request(void *data, struct request *rq, |
3708 | unsigned int hctx_idx, unsigned int request_idx, | |
3709 | unsigned int numa_node) | |
3710 | { | |
3711 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
3712 | ||
3713 | INIT_WORK(work, rbd_queue_workfn); | |
3714 | return 0; | |
3715 | } | |
3716 | ||
3717 | static struct blk_mq_ops rbd_mq_ops = { | |
3718 | .queue_rq = rbd_queue_rq, | |
3719 | .map_queue = blk_mq_map_queue, | |
3720 | .init_request = rbd_init_request, | |
3721 | }; | |
3722 | ||
602adf40 YS |
3723 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3724 | { | |
3725 | struct gendisk *disk; | |
3726 | struct request_queue *q; | |
593a9e7b | 3727 | u64 segment_size; |
7ad18afa | 3728 | int err; |
602adf40 | 3729 | |
602adf40 | 3730 | /* create gendisk info */ |
7e513d43 ID |
3731 | disk = alloc_disk(single_major ? |
3732 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3733 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3734 | if (!disk) |
1fcdb8aa | 3735 | return -ENOMEM; |
602adf40 | 3736 | |
f0f8cef5 | 3737 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3738 | rbd_dev->dev_id); |
602adf40 | 3739 | disk->major = rbd_dev->major; |
dd82fff1 | 3740 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3741 | if (single_major) |
3742 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3743 | disk->fops = &rbd_bd_ops; |
3744 | disk->private_data = rbd_dev; | |
3745 | ||
7ad18afa CH |
3746 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
3747 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 3748 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 3749 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 3750 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
3751 | rbd_dev->tag_set.nr_hw_queues = 1; |
3752 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
3753 | ||
3754 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
3755 | if (err) | |
602adf40 | 3756 | goto out_disk; |
029bcbd8 | 3757 | |
7ad18afa CH |
3758 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
3759 | if (IS_ERR(q)) { | |
3760 | err = PTR_ERR(q); | |
3761 | goto out_tag_set; | |
3762 | } | |
3763 | ||
d8a2c89c ID |
3764 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
3765 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 3766 | |
029bcbd8 | 3767 | /* set io sizes to object size */ |
593a9e7b AE |
3768 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3769 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
0d9fde4f | 3770 | q->limits.max_sectors = queue_max_hw_sectors(q); |
d3834fef | 3771 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
3772 | blk_queue_max_segment_size(q, segment_size); |
3773 | blk_queue_io_min(q, segment_size); | |
3774 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3775 | |
90e98c52 GZ |
3776 | /* enable the discard support */ |
3777 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3778 | q->limits.discard_granularity = segment_size; | |
3779 | q->limits.discard_alignment = segment_size; | |
2bb4cd5c | 3780 | blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE); |
b76f8239 | 3781 | q->limits.discard_zeroes_data = 1; |
90e98c52 | 3782 | |
bae818ee RH |
3783 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
3784 | q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES; | |
3785 | ||
602adf40 YS |
3786 | disk->queue = q; |
3787 | ||
3788 | q->queuedata = rbd_dev; | |
3789 | ||
3790 | rbd_dev->disk = disk; | |
602adf40 | 3791 | |
602adf40 | 3792 | return 0; |
7ad18afa CH |
3793 | out_tag_set: |
3794 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
3795 | out_disk: |
3796 | put_disk(disk); | |
7ad18afa | 3797 | return err; |
602adf40 YS |
3798 | } |
3799 | ||
dfc5606d YS |
3800 | /* |
3801 | sysfs | |
3802 | */ | |
3803 | ||
593a9e7b AE |
3804 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3805 | { | |
3806 | return container_of(dev, struct rbd_device, dev); | |
3807 | } | |
3808 | ||
dfc5606d YS |
3809 | static ssize_t rbd_size_show(struct device *dev, |
3810 | struct device_attribute *attr, char *buf) | |
3811 | { | |
593a9e7b | 3812 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3813 | |
fc71d833 AE |
3814 | return sprintf(buf, "%llu\n", |
3815 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3816 | } |
3817 | ||
34b13184 AE |
3818 | /* |
3819 | * Note this shows the features for whatever's mapped, which is not | |
3820 | * necessarily the base image. | |
3821 | */ | |
3822 | static ssize_t rbd_features_show(struct device *dev, | |
3823 | struct device_attribute *attr, char *buf) | |
3824 | { | |
3825 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3826 | ||
3827 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3828 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3829 | } |
3830 | ||
dfc5606d YS |
3831 | static ssize_t rbd_major_show(struct device *dev, |
3832 | struct device_attribute *attr, char *buf) | |
3833 | { | |
593a9e7b | 3834 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3835 | |
fc71d833 AE |
3836 | if (rbd_dev->major) |
3837 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3838 | ||
3839 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3840 | } |
3841 | ||
3842 | static ssize_t rbd_minor_show(struct device *dev, | |
3843 | struct device_attribute *attr, char *buf) | |
3844 | { | |
3845 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3846 | |
dd82fff1 | 3847 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3848 | } |
3849 | ||
3850 | static ssize_t rbd_client_id_show(struct device *dev, | |
3851 | struct device_attribute *attr, char *buf) | |
602adf40 | 3852 | { |
593a9e7b | 3853 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3854 | |
1dbb4399 AE |
3855 | return sprintf(buf, "client%lld\n", |
3856 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3857 | } |
3858 | ||
dfc5606d YS |
3859 | static ssize_t rbd_pool_show(struct device *dev, |
3860 | struct device_attribute *attr, char *buf) | |
602adf40 | 3861 | { |
593a9e7b | 3862 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3863 | |
0d7dbfce | 3864 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3865 | } |
3866 | ||
9bb2f334 AE |
3867 | static ssize_t rbd_pool_id_show(struct device *dev, |
3868 | struct device_attribute *attr, char *buf) | |
3869 | { | |
3870 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3871 | ||
0d7dbfce | 3872 | return sprintf(buf, "%llu\n", |
fc71d833 | 3873 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3874 | } |
3875 | ||
dfc5606d YS |
3876 | static ssize_t rbd_name_show(struct device *dev, |
3877 | struct device_attribute *attr, char *buf) | |
3878 | { | |
593a9e7b | 3879 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3880 | |
a92ffdf8 AE |
3881 | if (rbd_dev->spec->image_name) |
3882 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3883 | ||
3884 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3885 | } |
3886 | ||
589d30e0 AE |
3887 | static ssize_t rbd_image_id_show(struct device *dev, |
3888 | struct device_attribute *attr, char *buf) | |
3889 | { | |
3890 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3891 | ||
0d7dbfce | 3892 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3893 | } |
3894 | ||
34b13184 AE |
3895 | /* |
3896 | * Shows the name of the currently-mapped snapshot (or | |
3897 | * RBD_SNAP_HEAD_NAME for the base image). | |
3898 | */ | |
dfc5606d YS |
3899 | static ssize_t rbd_snap_show(struct device *dev, |
3900 | struct device_attribute *attr, | |
3901 | char *buf) | |
3902 | { | |
593a9e7b | 3903 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3904 | |
0d7dbfce | 3905 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3906 | } |
3907 | ||
86b00e0d | 3908 | /* |
ff96128f ID |
3909 | * For a v2 image, shows the chain of parent images, separated by empty |
3910 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3911 | * image)". | |
86b00e0d AE |
3912 | */ |
3913 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3914 | struct device_attribute *attr, |
3915 | char *buf) | |
86b00e0d AE |
3916 | { |
3917 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3918 | ssize_t count = 0; |
86b00e0d | 3919 | |
ff96128f | 3920 | if (!rbd_dev->parent) |
86b00e0d AE |
3921 | return sprintf(buf, "(no parent image)\n"); |
3922 | ||
ff96128f ID |
3923 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3924 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3925 | ||
3926 | count += sprintf(&buf[count], "%s" | |
3927 | "pool_id %llu\npool_name %s\n" | |
3928 | "image_id %s\nimage_name %s\n" | |
3929 | "snap_id %llu\nsnap_name %s\n" | |
3930 | "overlap %llu\n", | |
3931 | !count ? "" : "\n", /* first? */ | |
3932 | spec->pool_id, spec->pool_name, | |
3933 | spec->image_id, spec->image_name ?: "(unknown)", | |
3934 | spec->snap_id, spec->snap_name, | |
3935 | rbd_dev->parent_overlap); | |
3936 | } | |
3937 | ||
3938 | return count; | |
86b00e0d AE |
3939 | } |
3940 | ||
dfc5606d YS |
3941 | static ssize_t rbd_image_refresh(struct device *dev, |
3942 | struct device_attribute *attr, | |
3943 | const char *buf, | |
3944 | size_t size) | |
3945 | { | |
593a9e7b | 3946 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3947 | int ret; |
602adf40 | 3948 | |
cc4a38bd | 3949 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3950 | if (ret) |
52bb1f9b | 3951 | return ret; |
b813623a | 3952 | |
52bb1f9b | 3953 | return size; |
dfc5606d | 3954 | } |
602adf40 | 3955 | |
dfc5606d | 3956 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3957 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3958 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3959 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3960 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3961 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3962 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3963 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3964 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3965 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3966 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3967 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3968 | |
3969 | static struct attribute *rbd_attrs[] = { | |
3970 | &dev_attr_size.attr, | |
34b13184 | 3971 | &dev_attr_features.attr, |
dfc5606d | 3972 | &dev_attr_major.attr, |
dd82fff1 | 3973 | &dev_attr_minor.attr, |
dfc5606d YS |
3974 | &dev_attr_client_id.attr, |
3975 | &dev_attr_pool.attr, | |
9bb2f334 | 3976 | &dev_attr_pool_id.attr, |
dfc5606d | 3977 | &dev_attr_name.attr, |
589d30e0 | 3978 | &dev_attr_image_id.attr, |
dfc5606d | 3979 | &dev_attr_current_snap.attr, |
86b00e0d | 3980 | &dev_attr_parent.attr, |
dfc5606d | 3981 | &dev_attr_refresh.attr, |
dfc5606d YS |
3982 | NULL |
3983 | }; | |
3984 | ||
3985 | static struct attribute_group rbd_attr_group = { | |
3986 | .attrs = rbd_attrs, | |
3987 | }; | |
3988 | ||
3989 | static const struct attribute_group *rbd_attr_groups[] = { | |
3990 | &rbd_attr_group, | |
3991 | NULL | |
3992 | }; | |
3993 | ||
3994 | static void rbd_sysfs_dev_release(struct device *dev) | |
3995 | { | |
3996 | } | |
3997 | ||
3998 | static struct device_type rbd_device_type = { | |
3999 | .name = "rbd", | |
4000 | .groups = rbd_attr_groups, | |
4001 | .release = rbd_sysfs_dev_release, | |
4002 | }; | |
4003 | ||
8b8fb99c AE |
4004 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4005 | { | |
4006 | kref_get(&spec->kref); | |
4007 | ||
4008 | return spec; | |
4009 | } | |
4010 | ||
4011 | static void rbd_spec_free(struct kref *kref); | |
4012 | static void rbd_spec_put(struct rbd_spec *spec) | |
4013 | { | |
4014 | if (spec) | |
4015 | kref_put(&spec->kref, rbd_spec_free); | |
4016 | } | |
4017 | ||
4018 | static struct rbd_spec *rbd_spec_alloc(void) | |
4019 | { | |
4020 | struct rbd_spec *spec; | |
4021 | ||
4022 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4023 | if (!spec) | |
4024 | return NULL; | |
04077599 ID |
4025 | |
4026 | spec->pool_id = CEPH_NOPOOL; | |
4027 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4028 | kref_init(&spec->kref); |
4029 | ||
8b8fb99c AE |
4030 | return spec; |
4031 | } | |
4032 | ||
4033 | static void rbd_spec_free(struct kref *kref) | |
4034 | { | |
4035 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4036 | ||
4037 | kfree(spec->pool_name); | |
4038 | kfree(spec->image_id); | |
4039 | kfree(spec->image_name); | |
4040 | kfree(spec->snap_name); | |
4041 | kfree(spec); | |
4042 | } | |
4043 | ||
cc344fa1 | 4044 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
d147543d ID |
4045 | struct rbd_spec *spec, |
4046 | struct rbd_options *opts) | |
c53d5893 AE |
4047 | { |
4048 | struct rbd_device *rbd_dev; | |
4049 | ||
4050 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4051 | if (!rbd_dev) | |
4052 | return NULL; | |
4053 | ||
4054 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 4055 | rbd_dev->flags = 0; |
a2acd00e | 4056 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4057 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4058 | init_rwsem(&rbd_dev->header_rwsem); |
4059 | ||
c53d5893 | 4060 | rbd_dev->rbd_client = rbdc; |
d147543d ID |
4061 | rbd_dev->spec = spec; |
4062 | rbd_dev->opts = opts; | |
c53d5893 | 4063 | |
0903e875 AE |
4064 | /* Initialize the layout used for all rbd requests */ |
4065 | ||
4066 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4067 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4068 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4069 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4070 | ||
c53d5893 AE |
4071 | return rbd_dev; |
4072 | } | |
4073 | ||
4074 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4075 | { | |
c53d5893 AE |
4076 | rbd_put_client(rbd_dev->rbd_client); |
4077 | rbd_spec_put(rbd_dev->spec); | |
d147543d | 4078 | kfree(rbd_dev->opts); |
c53d5893 AE |
4079 | kfree(rbd_dev); |
4080 | } | |
4081 | ||
9d475de5 AE |
4082 | /* |
4083 | * Get the size and object order for an image snapshot, or if | |
4084 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4085 | * image. | |
4086 | */ | |
4087 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4088 | u8 *order, u64 *snap_size) | |
4089 | { | |
4090 | __le64 snapid = cpu_to_le64(snap_id); | |
4091 | int ret; | |
4092 | struct { | |
4093 | u8 order; | |
4094 | __le64 size; | |
4095 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4096 | ||
36be9a76 | 4097 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4098 | "rbd", "get_size", |
4157976b | 4099 | &snapid, sizeof (snapid), |
e2a58ee5 | 4100 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4101 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4102 | if (ret < 0) |
4103 | return ret; | |
57385b51 AE |
4104 | if (ret < sizeof (size_buf)) |
4105 | return -ERANGE; | |
9d475de5 | 4106 | |
c3545579 | 4107 | if (order) { |
c86f86e9 | 4108 | *order = size_buf.order; |
c3545579 JD |
4109 | dout(" order %u", (unsigned int)*order); |
4110 | } | |
9d475de5 AE |
4111 | *snap_size = le64_to_cpu(size_buf.size); |
4112 | ||
c3545579 JD |
4113 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4114 | (unsigned long long)snap_id, | |
57385b51 | 4115 | (unsigned long long)*snap_size); |
9d475de5 AE |
4116 | |
4117 | return 0; | |
4118 | } | |
4119 | ||
4120 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4121 | { | |
4122 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4123 | &rbd_dev->header.obj_order, | |
4124 | &rbd_dev->header.image_size); | |
4125 | } | |
4126 | ||
1e130199 AE |
4127 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4128 | { | |
4129 | void *reply_buf; | |
4130 | int ret; | |
4131 | void *p; | |
4132 | ||
4133 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4134 | if (!reply_buf) | |
4135 | return -ENOMEM; | |
4136 | ||
36be9a76 | 4137 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4138 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4139 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4140 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4141 | if (ret < 0) |
4142 | goto out; | |
4143 | ||
4144 | p = reply_buf; | |
4145 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4146 | p + ret, NULL, GFP_NOIO); |
4147 | ret = 0; | |
1e130199 AE |
4148 | |
4149 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4150 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4151 | rbd_dev->header.object_prefix = NULL; | |
4152 | } else { | |
4153 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4154 | } | |
1e130199 AE |
4155 | out: |
4156 | kfree(reply_buf); | |
4157 | ||
4158 | return ret; | |
4159 | } | |
4160 | ||
b1b5402a AE |
4161 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4162 | u64 *snap_features) | |
4163 | { | |
4164 | __le64 snapid = cpu_to_le64(snap_id); | |
4165 | struct { | |
4166 | __le64 features; | |
4167 | __le64 incompat; | |
4157976b | 4168 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4169 | u64 incompat; |
b1b5402a AE |
4170 | int ret; |
4171 | ||
36be9a76 | 4172 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4173 | "rbd", "get_features", |
4157976b | 4174 | &snapid, sizeof (snapid), |
e2a58ee5 | 4175 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4176 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4177 | if (ret < 0) |
4178 | return ret; | |
57385b51 AE |
4179 | if (ret < sizeof (features_buf)) |
4180 | return -ERANGE; | |
d889140c AE |
4181 | |
4182 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4183 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4184 | return -ENXIO; |
d889140c | 4185 | |
b1b5402a AE |
4186 | *snap_features = le64_to_cpu(features_buf.features); |
4187 | ||
4188 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4189 | (unsigned long long)snap_id, |
4190 | (unsigned long long)*snap_features, | |
4191 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4192 | |
4193 | return 0; | |
4194 | } | |
4195 | ||
4196 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4197 | { | |
4198 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4199 | &rbd_dev->header.features); | |
4200 | } | |
4201 | ||
86b00e0d AE |
4202 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4203 | { | |
4204 | struct rbd_spec *parent_spec; | |
4205 | size_t size; | |
4206 | void *reply_buf = NULL; | |
4207 | __le64 snapid; | |
4208 | void *p; | |
4209 | void *end; | |
642a2537 | 4210 | u64 pool_id; |
86b00e0d | 4211 | char *image_id; |
3b5cf2a2 | 4212 | u64 snap_id; |
86b00e0d | 4213 | u64 overlap; |
86b00e0d AE |
4214 | int ret; |
4215 | ||
4216 | parent_spec = rbd_spec_alloc(); | |
4217 | if (!parent_spec) | |
4218 | return -ENOMEM; | |
4219 | ||
4220 | size = sizeof (__le64) + /* pool_id */ | |
4221 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4222 | sizeof (__le64) + /* snap_id */ | |
4223 | sizeof (__le64); /* overlap */ | |
4224 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4225 | if (!reply_buf) { | |
4226 | ret = -ENOMEM; | |
4227 | goto out_err; | |
4228 | } | |
4229 | ||
4d9b67cd | 4230 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4231 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4232 | "rbd", "get_parent", |
4157976b | 4233 | &snapid, sizeof (snapid), |
e2a58ee5 | 4234 | reply_buf, size); |
36be9a76 | 4235 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4236 | if (ret < 0) |
4237 | goto out_err; | |
4238 | ||
86b00e0d | 4239 | p = reply_buf; |
57385b51 AE |
4240 | end = reply_buf + ret; |
4241 | ret = -ERANGE; | |
642a2537 | 4242 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4243 | if (pool_id == CEPH_NOPOOL) { |
4244 | /* | |
4245 | * Either the parent never existed, or we have | |
4246 | * record of it but the image got flattened so it no | |
4247 | * longer has a parent. When the parent of a | |
4248 | * layered image disappears we immediately set the | |
4249 | * overlap to 0. The effect of this is that all new | |
4250 | * requests will be treated as if the image had no | |
4251 | * parent. | |
4252 | */ | |
4253 | if (rbd_dev->parent_overlap) { | |
4254 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4255 | rbd_dev_parent_put(rbd_dev); |
4256 | pr_info("%s: clone image has been flattened\n", | |
4257 | rbd_dev->disk->disk_name); | |
4258 | } | |
4259 | ||
86b00e0d | 4260 | goto out; /* No parent? No problem. */ |
392a9dad | 4261 | } |
86b00e0d | 4262 | |
0903e875 AE |
4263 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4264 | ||
4265 | ret = -EIO; | |
642a2537 | 4266 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4267 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4268 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4269 | goto out_err; |
c0cd10db | 4270 | } |
0903e875 | 4271 | |
979ed480 | 4272 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4273 | if (IS_ERR(image_id)) { |
4274 | ret = PTR_ERR(image_id); | |
4275 | goto out_err; | |
4276 | } | |
3b5cf2a2 | 4277 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4278 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4279 | ||
3b5cf2a2 AE |
4280 | /* |
4281 | * The parent won't change (except when the clone is | |
4282 | * flattened, already handled that). So we only need to | |
4283 | * record the parent spec we have not already done so. | |
4284 | */ | |
4285 | if (!rbd_dev->parent_spec) { | |
4286 | parent_spec->pool_id = pool_id; | |
4287 | parent_spec->image_id = image_id; | |
4288 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4289 | rbd_dev->parent_spec = parent_spec; |
4290 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4291 | } else { |
4292 | kfree(image_id); | |
3b5cf2a2 AE |
4293 | } |
4294 | ||
4295 | /* | |
cf32bd9c ID |
4296 | * We always update the parent overlap. If it's zero we issue |
4297 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 4298 | */ |
3b5cf2a2 | 4299 | if (!overlap) { |
3b5cf2a2 | 4300 | if (parent_spec) { |
cf32bd9c ID |
4301 | /* refresh, careful to warn just once */ |
4302 | if (rbd_dev->parent_overlap) | |
4303 | rbd_warn(rbd_dev, | |
4304 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 4305 | } else { |
cf32bd9c ID |
4306 | /* initial probe */ |
4307 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 4308 | } |
70cf49cf | 4309 | } |
cf32bd9c ID |
4310 | rbd_dev->parent_overlap = overlap; |
4311 | ||
86b00e0d AE |
4312 | out: |
4313 | ret = 0; | |
4314 | out_err: | |
4315 | kfree(reply_buf); | |
4316 | rbd_spec_put(parent_spec); | |
4317 | ||
4318 | return ret; | |
4319 | } | |
4320 | ||
cc070d59 AE |
4321 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4322 | { | |
4323 | struct { | |
4324 | __le64 stripe_unit; | |
4325 | __le64 stripe_count; | |
4326 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4327 | size_t size = sizeof (striping_info_buf); | |
4328 | void *p; | |
4329 | u64 obj_size; | |
4330 | u64 stripe_unit; | |
4331 | u64 stripe_count; | |
4332 | int ret; | |
4333 | ||
4334 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4335 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4336 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4337 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4338 | if (ret < 0) | |
4339 | return ret; | |
4340 | if (ret < size) | |
4341 | return -ERANGE; | |
4342 | ||
4343 | /* | |
4344 | * We don't actually support the "fancy striping" feature | |
4345 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4346 | * defaults the behavior is the same as before. So find | |
4347 | * out, and only fail if the image has non-default values. | |
4348 | */ | |
4349 | ret = -EINVAL; | |
4350 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4351 | p = &striping_info_buf; | |
4352 | stripe_unit = ceph_decode_64(&p); | |
4353 | if (stripe_unit != obj_size) { | |
4354 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4355 | "(got %llu want %llu)", | |
4356 | stripe_unit, obj_size); | |
4357 | return -EINVAL; | |
4358 | } | |
4359 | stripe_count = ceph_decode_64(&p); | |
4360 | if (stripe_count != 1) { | |
4361 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4362 | "(got %llu want 1)", stripe_count); | |
4363 | return -EINVAL; | |
4364 | } | |
500d0c0f AE |
4365 | rbd_dev->header.stripe_unit = stripe_unit; |
4366 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4367 | |
4368 | return 0; | |
4369 | } | |
4370 | ||
9e15b77d AE |
4371 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4372 | { | |
4373 | size_t image_id_size; | |
4374 | char *image_id; | |
4375 | void *p; | |
4376 | void *end; | |
4377 | size_t size; | |
4378 | void *reply_buf = NULL; | |
4379 | size_t len = 0; | |
4380 | char *image_name = NULL; | |
4381 | int ret; | |
4382 | ||
4383 | rbd_assert(!rbd_dev->spec->image_name); | |
4384 | ||
69e7a02f AE |
4385 | len = strlen(rbd_dev->spec->image_id); |
4386 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4387 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4388 | if (!image_id) | |
4389 | return NULL; | |
4390 | ||
4391 | p = image_id; | |
4157976b | 4392 | end = image_id + image_id_size; |
57385b51 | 4393 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4394 | |
4395 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4396 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4397 | if (!reply_buf) | |
4398 | goto out; | |
4399 | ||
36be9a76 | 4400 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4401 | "rbd", "dir_get_name", |
4402 | image_id, image_id_size, | |
e2a58ee5 | 4403 | reply_buf, size); |
9e15b77d AE |
4404 | if (ret < 0) |
4405 | goto out; | |
4406 | p = reply_buf; | |
f40eb349 AE |
4407 | end = reply_buf + ret; |
4408 | ||
9e15b77d AE |
4409 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4410 | if (IS_ERR(image_name)) | |
4411 | image_name = NULL; | |
4412 | else | |
4413 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4414 | out: | |
4415 | kfree(reply_buf); | |
4416 | kfree(image_id); | |
4417 | ||
4418 | return image_name; | |
4419 | } | |
4420 | ||
2ad3d716 AE |
4421 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4422 | { | |
4423 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4424 | const char *snap_name; | |
4425 | u32 which = 0; | |
4426 | ||
4427 | /* Skip over names until we find the one we are looking for */ | |
4428 | ||
4429 | snap_name = rbd_dev->header.snap_names; | |
4430 | while (which < snapc->num_snaps) { | |
4431 | if (!strcmp(name, snap_name)) | |
4432 | return snapc->snaps[which]; | |
4433 | snap_name += strlen(snap_name) + 1; | |
4434 | which++; | |
4435 | } | |
4436 | return CEPH_NOSNAP; | |
4437 | } | |
4438 | ||
4439 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4440 | { | |
4441 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4442 | u32 which; | |
4443 | bool found = false; | |
4444 | u64 snap_id; | |
4445 | ||
4446 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4447 | const char *snap_name; | |
4448 | ||
4449 | snap_id = snapc->snaps[which]; | |
4450 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4451 | if (IS_ERR(snap_name)) { |
4452 | /* ignore no-longer existing snapshots */ | |
4453 | if (PTR_ERR(snap_name) == -ENOENT) | |
4454 | continue; | |
4455 | else | |
4456 | break; | |
4457 | } | |
2ad3d716 AE |
4458 | found = !strcmp(name, snap_name); |
4459 | kfree(snap_name); | |
4460 | } | |
4461 | return found ? snap_id : CEPH_NOSNAP; | |
4462 | } | |
4463 | ||
4464 | /* | |
4465 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4466 | * no snapshot by that name is found, or if an error occurs. | |
4467 | */ | |
4468 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4469 | { | |
4470 | if (rbd_dev->image_format == 1) | |
4471 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4472 | ||
4473 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4474 | } | |
4475 | ||
9e15b77d | 4476 | /* |
04077599 ID |
4477 | * An image being mapped will have everything but the snap id. |
4478 | */ | |
4479 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4480 | { | |
4481 | struct rbd_spec *spec = rbd_dev->spec; | |
4482 | ||
4483 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4484 | rbd_assert(spec->image_id && spec->image_name); | |
4485 | rbd_assert(spec->snap_name); | |
4486 | ||
4487 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4488 | u64 snap_id; | |
4489 | ||
4490 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4491 | if (snap_id == CEPH_NOSNAP) | |
4492 | return -ENOENT; | |
4493 | ||
4494 | spec->snap_id = snap_id; | |
4495 | } else { | |
4496 | spec->snap_id = CEPH_NOSNAP; | |
4497 | } | |
4498 | ||
4499 | return 0; | |
4500 | } | |
4501 | ||
4502 | /* | |
4503 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4504 | * |
04077599 ID |
4505 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4506 | * can't figure out the name for an image id. | |
9e15b77d | 4507 | */ |
04077599 | 4508 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4509 | { |
2e9f7f1c AE |
4510 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4511 | struct rbd_spec *spec = rbd_dev->spec; | |
4512 | const char *pool_name; | |
4513 | const char *image_name; | |
4514 | const char *snap_name; | |
9e15b77d AE |
4515 | int ret; |
4516 | ||
04077599 ID |
4517 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4518 | rbd_assert(spec->image_id); | |
4519 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4520 | |
2e9f7f1c | 4521 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4522 | |
2e9f7f1c AE |
4523 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4524 | if (!pool_name) { | |
4525 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4526 | return -EIO; |
4527 | } | |
2e9f7f1c AE |
4528 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4529 | if (!pool_name) | |
9e15b77d AE |
4530 | return -ENOMEM; |
4531 | ||
4532 | /* Fetch the image name; tolerate failure here */ | |
4533 | ||
2e9f7f1c AE |
4534 | image_name = rbd_dev_image_name(rbd_dev); |
4535 | if (!image_name) | |
06ecc6cb | 4536 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4537 | |
04077599 | 4538 | /* Fetch the snapshot name */ |
9e15b77d | 4539 | |
2e9f7f1c | 4540 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4541 | if (IS_ERR(snap_name)) { |
4542 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4543 | goto out_err; |
2e9f7f1c AE |
4544 | } |
4545 | ||
4546 | spec->pool_name = pool_name; | |
4547 | spec->image_name = image_name; | |
4548 | spec->snap_name = snap_name; | |
9e15b77d AE |
4549 | |
4550 | return 0; | |
04077599 | 4551 | |
9e15b77d | 4552 | out_err: |
2e9f7f1c AE |
4553 | kfree(image_name); |
4554 | kfree(pool_name); | |
9e15b77d AE |
4555 | return ret; |
4556 | } | |
4557 | ||
cc4a38bd | 4558 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4559 | { |
4560 | size_t size; | |
4561 | int ret; | |
4562 | void *reply_buf; | |
4563 | void *p; | |
4564 | void *end; | |
4565 | u64 seq; | |
4566 | u32 snap_count; | |
4567 | struct ceph_snap_context *snapc; | |
4568 | u32 i; | |
4569 | ||
4570 | /* | |
4571 | * We'll need room for the seq value (maximum snapshot id), | |
4572 | * snapshot count, and array of that many snapshot ids. | |
4573 | * For now we have a fixed upper limit on the number we're | |
4574 | * prepared to receive. | |
4575 | */ | |
4576 | size = sizeof (__le64) + sizeof (__le32) + | |
4577 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4578 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4579 | if (!reply_buf) | |
4580 | return -ENOMEM; | |
4581 | ||
36be9a76 | 4582 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4583 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4584 | reply_buf, size); |
36be9a76 | 4585 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4586 | if (ret < 0) |
4587 | goto out; | |
4588 | ||
35d489f9 | 4589 | p = reply_buf; |
57385b51 AE |
4590 | end = reply_buf + ret; |
4591 | ret = -ERANGE; | |
35d489f9 AE |
4592 | ceph_decode_64_safe(&p, end, seq, out); |
4593 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4594 | ||
4595 | /* | |
4596 | * Make sure the reported number of snapshot ids wouldn't go | |
4597 | * beyond the end of our buffer. But before checking that, | |
4598 | * make sure the computed size of the snapshot context we | |
4599 | * allocate is representable in a size_t. | |
4600 | */ | |
4601 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4602 | / sizeof (u64)) { | |
4603 | ret = -EINVAL; | |
4604 | goto out; | |
4605 | } | |
4606 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4607 | goto out; | |
468521c1 | 4608 | ret = 0; |
35d489f9 | 4609 | |
812164f8 | 4610 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4611 | if (!snapc) { |
4612 | ret = -ENOMEM; | |
4613 | goto out; | |
4614 | } | |
35d489f9 | 4615 | snapc->seq = seq; |
35d489f9 AE |
4616 | for (i = 0; i < snap_count; i++) |
4617 | snapc->snaps[i] = ceph_decode_64(&p); | |
4618 | ||
49ece554 | 4619 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4620 | rbd_dev->header.snapc = snapc; |
4621 | ||
4622 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4623 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4624 | out: |
4625 | kfree(reply_buf); | |
4626 | ||
57385b51 | 4627 | return ret; |
35d489f9 AE |
4628 | } |
4629 | ||
54cac61f AE |
4630 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4631 | u64 snap_id) | |
b8b1e2db AE |
4632 | { |
4633 | size_t size; | |
4634 | void *reply_buf; | |
54cac61f | 4635 | __le64 snapid; |
b8b1e2db AE |
4636 | int ret; |
4637 | void *p; | |
4638 | void *end; | |
b8b1e2db AE |
4639 | char *snap_name; |
4640 | ||
4641 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4642 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4643 | if (!reply_buf) | |
4644 | return ERR_PTR(-ENOMEM); | |
4645 | ||
54cac61f | 4646 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4647 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4648 | "rbd", "get_snapshot_name", |
54cac61f | 4649 | &snapid, sizeof (snapid), |
e2a58ee5 | 4650 | reply_buf, size); |
36be9a76 | 4651 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4652 | if (ret < 0) { |
4653 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4654 | goto out; |
f40eb349 | 4655 | } |
b8b1e2db AE |
4656 | |
4657 | p = reply_buf; | |
f40eb349 | 4658 | end = reply_buf + ret; |
e5c35534 | 4659 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4660 | if (IS_ERR(snap_name)) |
b8b1e2db | 4661 | goto out; |
b8b1e2db | 4662 | |
f40eb349 | 4663 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4664 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4665 | out: |
4666 | kfree(reply_buf); | |
4667 | ||
f40eb349 | 4668 | return snap_name; |
b8b1e2db AE |
4669 | } |
4670 | ||
2df3fac7 | 4671 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4672 | { |
2df3fac7 | 4673 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4674 | int ret; |
117973fb | 4675 | |
1617e40c JD |
4676 | ret = rbd_dev_v2_image_size(rbd_dev); |
4677 | if (ret) | |
cfbf6377 | 4678 | return ret; |
1617e40c | 4679 | |
2df3fac7 AE |
4680 | if (first_time) { |
4681 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4682 | if (ret) | |
cfbf6377 | 4683 | return ret; |
2df3fac7 AE |
4684 | } |
4685 | ||
cc4a38bd | 4686 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
4687 | if (ret && first_time) { |
4688 | kfree(rbd_dev->header.object_prefix); | |
4689 | rbd_dev->header.object_prefix = NULL; | |
4690 | } | |
117973fb AE |
4691 | |
4692 | return ret; | |
4693 | } | |
4694 | ||
a720ae09 ID |
4695 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4696 | { | |
4697 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4698 | ||
4699 | if (rbd_dev->image_format == 1) | |
4700 | return rbd_dev_v1_header_info(rbd_dev); | |
4701 | ||
4702 | return rbd_dev_v2_header_info(rbd_dev); | |
4703 | } | |
4704 | ||
dfc5606d YS |
4705 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4706 | { | |
dfc5606d | 4707 | struct device *dev; |
cd789ab9 | 4708 | int ret; |
dfc5606d | 4709 | |
cd789ab9 | 4710 | dev = &rbd_dev->dev; |
dfc5606d YS |
4711 | dev->bus = &rbd_bus_type; |
4712 | dev->type = &rbd_device_type; | |
4713 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4714 | dev->release = rbd_dev_device_release; |
de71a297 | 4715 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4716 | ret = device_register(dev); |
dfc5606d | 4717 | |
dfc5606d | 4718 | return ret; |
602adf40 YS |
4719 | } |
4720 | ||
dfc5606d YS |
4721 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4722 | { | |
4723 | device_unregister(&rbd_dev->dev); | |
4724 | } | |
4725 | ||
1ddbe94e | 4726 | /* |
499afd5b | 4727 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4728 | * the rbd_dev to the global list. |
1ddbe94e | 4729 | */ |
f8a22fc2 | 4730 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4731 | { |
f8a22fc2 ID |
4732 | int new_dev_id; |
4733 | ||
9b60e70b ID |
4734 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4735 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4736 | GFP_KERNEL); | |
f8a22fc2 ID |
4737 | if (new_dev_id < 0) |
4738 | return new_dev_id; | |
4739 | ||
4740 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4741 | |
4742 | spin_lock(&rbd_dev_list_lock); | |
4743 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4744 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4745 | |
70eebd20 | 4746 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4747 | |
4748 | return 0; | |
1ddbe94e | 4749 | } |
b7f23c36 | 4750 | |
1ddbe94e | 4751 | /* |
499afd5b AE |
4752 | * Remove an rbd_dev from the global list, and record that its |
4753 | * identifier is no longer in use. | |
1ddbe94e | 4754 | */ |
e2839308 | 4755 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4756 | { |
499afd5b AE |
4757 | spin_lock(&rbd_dev_list_lock); |
4758 | list_del_init(&rbd_dev->node); | |
4759 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4760 | |
f8a22fc2 ID |
4761 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4762 | ||
4763 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4764 | } |
4765 | ||
e28fff26 AE |
4766 | /* |
4767 | * Skips over white space at *buf, and updates *buf to point to the | |
4768 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4769 | * the token (string of non-white space characters) found. Note |
4770 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4771 | */ |
4772 | static inline size_t next_token(const char **buf) | |
4773 | { | |
4774 | /* | |
4775 | * These are the characters that produce nonzero for | |
4776 | * isspace() in the "C" and "POSIX" locales. | |
4777 | */ | |
4778 | const char *spaces = " \f\n\r\t\v"; | |
4779 | ||
4780 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4781 | ||
4782 | return strcspn(*buf, spaces); /* Return token length */ | |
4783 | } | |
4784 | ||
ea3352f4 AE |
4785 | /* |
4786 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4787 | * enough to hold a copy of it, and copies the token into the new | |
4788 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4789 | * that a duplicate buffer is created even for a zero-length token. | |
4790 | * | |
4791 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4792 | * pointer if memory for the duplicate was not available. If | |
4793 | * the lenp argument is a non-null pointer, the length of the token | |
4794 | * (not including the '\0') is returned in *lenp. | |
4795 | * | |
4796 | * If successful, the *buf pointer will be updated to point beyond | |
4797 | * the end of the found token. | |
4798 | * | |
4799 | * Note: uses GFP_KERNEL for allocation. | |
4800 | */ | |
4801 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4802 | { | |
4803 | char *dup; | |
4804 | size_t len; | |
4805 | ||
4806 | len = next_token(buf); | |
4caf35f9 | 4807 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4808 | if (!dup) |
4809 | return NULL; | |
ea3352f4 AE |
4810 | *(dup + len) = '\0'; |
4811 | *buf += len; | |
4812 | ||
4813 | if (lenp) | |
4814 | *lenp = len; | |
4815 | ||
4816 | return dup; | |
4817 | } | |
4818 | ||
a725f65e | 4819 | /* |
859c31df AE |
4820 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4821 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4822 | * and the data written is passed here via a NUL-terminated buffer. | |
4823 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4824 | * |
859c31df AE |
4825 | * The information extracted from these options is recorded in |
4826 | * the other parameters which return dynamically-allocated | |
4827 | * structures: | |
4828 | * ceph_opts | |
4829 | * The address of a pointer that will refer to a ceph options | |
4830 | * structure. Caller must release the returned pointer using | |
4831 | * ceph_destroy_options() when it is no longer needed. | |
4832 | * rbd_opts | |
4833 | * Address of an rbd options pointer. Fully initialized by | |
4834 | * this function; caller must release with kfree(). | |
4835 | * spec | |
4836 | * Address of an rbd image specification pointer. Fully | |
4837 | * initialized by this function based on parsed options. | |
4838 | * Caller must release with rbd_spec_put(). | |
4839 | * | |
4840 | * The options passed take this form: | |
4841 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4842 | * where: | |
4843 | * <mon_addrs> | |
4844 | * A comma-separated list of one or more monitor addresses. | |
4845 | * A monitor address is an ip address, optionally followed | |
4846 | * by a port number (separated by a colon). | |
4847 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4848 | * <options> | |
4849 | * A comma-separated list of ceph and/or rbd options. | |
4850 | * <pool_name> | |
4851 | * The name of the rados pool containing the rbd image. | |
4852 | * <image_name> | |
4853 | * The name of the image in that pool to map. | |
4854 | * <snap_id> | |
4855 | * An optional snapshot id. If provided, the mapping will | |
4856 | * present data from the image at the time that snapshot was | |
4857 | * created. The image head is used if no snapshot id is | |
4858 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4859 | */ |
859c31df | 4860 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4861 | struct ceph_options **ceph_opts, |
859c31df AE |
4862 | struct rbd_options **opts, |
4863 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4864 | { |
d22f76e7 | 4865 | size_t len; |
859c31df | 4866 | char *options; |
0ddebc0c | 4867 | const char *mon_addrs; |
ecb4dc22 | 4868 | char *snap_name; |
0ddebc0c | 4869 | size_t mon_addrs_size; |
859c31df | 4870 | struct rbd_spec *spec = NULL; |
4e9afeba | 4871 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4872 | struct ceph_options *copts; |
dc79b113 | 4873 | int ret; |
e28fff26 AE |
4874 | |
4875 | /* The first four tokens are required */ | |
4876 | ||
7ef3214a | 4877 | len = next_token(&buf); |
4fb5d671 AE |
4878 | if (!len) { |
4879 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4880 | return -EINVAL; | |
4881 | } | |
0ddebc0c | 4882 | mon_addrs = buf; |
f28e565a | 4883 | mon_addrs_size = len + 1; |
7ef3214a | 4884 | buf += len; |
a725f65e | 4885 | |
dc79b113 | 4886 | ret = -EINVAL; |
f28e565a AE |
4887 | options = dup_token(&buf, NULL); |
4888 | if (!options) | |
dc79b113 | 4889 | return -ENOMEM; |
4fb5d671 AE |
4890 | if (!*options) { |
4891 | rbd_warn(NULL, "no options provided"); | |
4892 | goto out_err; | |
4893 | } | |
e28fff26 | 4894 | |
859c31df AE |
4895 | spec = rbd_spec_alloc(); |
4896 | if (!spec) | |
f28e565a | 4897 | goto out_mem; |
859c31df AE |
4898 | |
4899 | spec->pool_name = dup_token(&buf, NULL); | |
4900 | if (!spec->pool_name) | |
4901 | goto out_mem; | |
4fb5d671 AE |
4902 | if (!*spec->pool_name) { |
4903 | rbd_warn(NULL, "no pool name provided"); | |
4904 | goto out_err; | |
4905 | } | |
e28fff26 | 4906 | |
69e7a02f | 4907 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4908 | if (!spec->image_name) |
f28e565a | 4909 | goto out_mem; |
4fb5d671 AE |
4910 | if (!*spec->image_name) { |
4911 | rbd_warn(NULL, "no image name provided"); | |
4912 | goto out_err; | |
4913 | } | |
d4b125e9 | 4914 | |
f28e565a AE |
4915 | /* |
4916 | * Snapshot name is optional; default is to use "-" | |
4917 | * (indicating the head/no snapshot). | |
4918 | */ | |
3feeb894 | 4919 | len = next_token(&buf); |
820a5f3e | 4920 | if (!len) { |
3feeb894 AE |
4921 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4922 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4923 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4924 | ret = -ENAMETOOLONG; |
f28e565a | 4925 | goto out_err; |
849b4260 | 4926 | } |
ecb4dc22 AE |
4927 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4928 | if (!snap_name) | |
f28e565a | 4929 | goto out_mem; |
ecb4dc22 AE |
4930 | *(snap_name + len) = '\0'; |
4931 | spec->snap_name = snap_name; | |
e5c35534 | 4932 | |
0ddebc0c | 4933 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4934 | |
4e9afeba AE |
4935 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4936 | if (!rbd_opts) | |
4937 | goto out_mem; | |
4938 | ||
4939 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 4940 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
d22f76e7 | 4941 | |
859c31df | 4942 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4943 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4944 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4945 | if (IS_ERR(copts)) { |
4946 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4947 | goto out_err; |
4948 | } | |
859c31df AE |
4949 | kfree(options); |
4950 | ||
4951 | *ceph_opts = copts; | |
4e9afeba | 4952 | *opts = rbd_opts; |
859c31df | 4953 | *rbd_spec = spec; |
0ddebc0c | 4954 | |
dc79b113 | 4955 | return 0; |
f28e565a | 4956 | out_mem: |
dc79b113 | 4957 | ret = -ENOMEM; |
d22f76e7 | 4958 | out_err: |
859c31df AE |
4959 | kfree(rbd_opts); |
4960 | rbd_spec_put(spec); | |
f28e565a | 4961 | kfree(options); |
d22f76e7 | 4962 | |
dc79b113 | 4963 | return ret; |
a725f65e AE |
4964 | } |
4965 | ||
30ba1f02 ID |
4966 | /* |
4967 | * Return pool id (>= 0) or a negative error code. | |
4968 | */ | |
4969 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4970 | { | |
a319bf56 | 4971 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 4972 | u64 newest_epoch; |
30ba1f02 ID |
4973 | int tries = 0; |
4974 | int ret; | |
4975 | ||
4976 | again: | |
4977 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4978 | if (ret == -ENOENT && tries++ < 1) { | |
4979 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4980 | &newest_epoch); | |
4981 | if (ret < 0) | |
4982 | return ret; | |
4983 | ||
4984 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4985 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4986 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
a319bf56 ID |
4987 | newest_epoch, |
4988 | opts->mount_timeout); | |
30ba1f02 ID |
4989 | goto again; |
4990 | } else { | |
4991 | /* the osdmap we have is new enough */ | |
4992 | return -ENOENT; | |
4993 | } | |
4994 | } | |
4995 | ||
4996 | return ret; | |
4997 | } | |
4998 | ||
589d30e0 AE |
4999 | /* |
5000 | * An rbd format 2 image has a unique identifier, distinct from the | |
5001 | * name given to it by the user. Internally, that identifier is | |
5002 | * what's used to specify the names of objects related to the image. | |
5003 | * | |
5004 | * A special "rbd id" object is used to map an rbd image name to its | |
5005 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5006 | * with the supplied name. | |
5007 | * | |
5008 | * This function will record the given rbd_dev's image_id field if | |
5009 | * it can be determined, and in that case will return 0. If any | |
5010 | * errors occur a negative errno will be returned and the rbd_dev's | |
5011 | * image_id field will be unchanged (and should be NULL). | |
5012 | */ | |
5013 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5014 | { | |
5015 | int ret; | |
5016 | size_t size; | |
5017 | char *object_name; | |
5018 | void *response; | |
c0fba368 | 5019 | char *image_id; |
2f82ee54 | 5020 | |
2c0d0a10 AE |
5021 | /* |
5022 | * When probing a parent image, the image id is already | |
5023 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5024 | * need to fetch the image id again in this case. We |
5025 | * do still need to set the image format though. | |
2c0d0a10 | 5026 | */ |
c0fba368 AE |
5027 | if (rbd_dev->spec->image_id) { |
5028 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5029 | ||
2c0d0a10 | 5030 | return 0; |
c0fba368 | 5031 | } |
2c0d0a10 | 5032 | |
589d30e0 AE |
5033 | /* |
5034 | * First, see if the format 2 image id file exists, and if | |
5035 | * so, get the image's persistent id from it. | |
5036 | */ | |
69e7a02f | 5037 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5038 | object_name = kmalloc(size, GFP_NOIO); |
5039 | if (!object_name) | |
5040 | return -ENOMEM; | |
0d7dbfce | 5041 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5042 | dout("rbd id object name is %s\n", object_name); |
5043 | ||
5044 | /* Response will be an encoded string, which includes a length */ | |
5045 | ||
5046 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5047 | response = kzalloc(size, GFP_NOIO); | |
5048 | if (!response) { | |
5049 | ret = -ENOMEM; | |
5050 | goto out; | |
5051 | } | |
5052 | ||
c0fba368 AE |
5053 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5054 | ||
36be9a76 | 5055 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5056 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5057 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5058 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5059 | if (ret == -ENOENT) { |
5060 | image_id = kstrdup("", GFP_KERNEL); | |
5061 | ret = image_id ? 0 : -ENOMEM; | |
5062 | if (!ret) | |
5063 | rbd_dev->image_format = 1; | |
7dd440c9 | 5064 | } else if (ret >= 0) { |
c0fba368 AE |
5065 | void *p = response; |
5066 | ||
5067 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5068 | NULL, GFP_NOIO); |
461f758a | 5069 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5070 | if (!ret) |
5071 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5072 | } |
5073 | ||
5074 | if (!ret) { | |
5075 | rbd_dev->spec->image_id = image_id; | |
5076 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5077 | } |
5078 | out: | |
5079 | kfree(response); | |
5080 | kfree(object_name); | |
5081 | ||
5082 | return ret; | |
5083 | } | |
5084 | ||
3abef3b3 AE |
5085 | /* |
5086 | * Undo whatever state changes are made by v1 or v2 header info | |
5087 | * call. | |
5088 | */ | |
6fd48b3b AE |
5089 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5090 | { | |
5091 | struct rbd_image_header *header; | |
5092 | ||
e69b8d41 | 5093 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5094 | |
5095 | /* Free dynamic fields from the header, then zero it out */ | |
5096 | ||
5097 | header = &rbd_dev->header; | |
812164f8 | 5098 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5099 | kfree(header->snap_sizes); |
5100 | kfree(header->snap_names); | |
5101 | kfree(header->object_prefix); | |
5102 | memset(header, 0, sizeof (*header)); | |
5103 | } | |
5104 | ||
2df3fac7 | 5105 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5106 | { |
5107 | int ret; | |
a30b71b9 | 5108 | |
1e130199 | 5109 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5110 | if (ret) |
b1b5402a AE |
5111 | goto out_err; |
5112 | ||
2df3fac7 AE |
5113 | /* |
5114 | * Get the and check features for the image. Currently the | |
5115 | * features are assumed to never change. | |
5116 | */ | |
b1b5402a | 5117 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5118 | if (ret) |
9d475de5 | 5119 | goto out_err; |
35d489f9 | 5120 | |
cc070d59 AE |
5121 | /* If the image supports fancy striping, get its parameters */ |
5122 | ||
5123 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5124 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5125 | if (ret < 0) | |
5126 | goto out_err; | |
5127 | } | |
2df3fac7 | 5128 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5129 | |
35152979 | 5130 | return 0; |
9d475de5 | 5131 | out_err: |
642a2537 | 5132 | rbd_dev->header.features = 0; |
1e130199 AE |
5133 | kfree(rbd_dev->header.object_prefix); |
5134 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5135 | |
5136 | return ret; | |
a30b71b9 AE |
5137 | } |
5138 | ||
6d69bb53 ID |
5139 | /* |
5140 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
5141 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
5142 | * length of the already discovered part of the parent chain. | |
5143 | */ | |
5144 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 5145 | { |
2f82ee54 | 5146 | struct rbd_device *parent = NULL; |
124afba2 AE |
5147 | int ret; |
5148 | ||
5149 | if (!rbd_dev->parent_spec) | |
5150 | return 0; | |
124afba2 | 5151 | |
6d69bb53 ID |
5152 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
5153 | pr_info("parent chain is too long (%d)\n", depth); | |
5154 | ret = -EINVAL; | |
5155 | goto out_err; | |
5156 | } | |
5157 | ||
1f2c6651 ID |
5158 | parent = rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec, |
5159 | NULL); | |
5160 | if (!parent) { | |
5161 | ret = -ENOMEM; | |
124afba2 | 5162 | goto out_err; |
1f2c6651 ID |
5163 | } |
5164 | ||
5165 | /* | |
5166 | * Images related by parent/child relationships always share | |
5167 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
5168 | */ | |
5169 | __rbd_get_client(rbd_dev->rbd_client); | |
5170 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 5171 | |
6d69bb53 | 5172 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
5173 | if (ret < 0) |
5174 | goto out_err; | |
1f2c6651 | 5175 | |
124afba2 | 5176 | rbd_dev->parent = parent; |
a2acd00e | 5177 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 5178 | return 0; |
1f2c6651 | 5179 | |
124afba2 | 5180 | out_err: |
1f2c6651 ID |
5181 | rbd_dev_unparent(rbd_dev); |
5182 | if (parent) | |
124afba2 | 5183 | rbd_dev_destroy(parent); |
124afba2 AE |
5184 | return ret; |
5185 | } | |
5186 | ||
200a6a8b | 5187 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5188 | { |
83a06263 | 5189 | int ret; |
d1cf5788 | 5190 | |
f8a22fc2 ID |
5191 | /* Get an id and fill in device name. */ |
5192 | ||
5193 | ret = rbd_dev_id_get(rbd_dev); | |
5194 | if (ret) | |
5195 | return ret; | |
83a06263 | 5196 | |
83a06263 AE |
5197 | BUILD_BUG_ON(DEV_NAME_LEN |
5198 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5199 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5200 | ||
9b60e70b | 5201 | /* Record our major and minor device numbers. */ |
83a06263 | 5202 | |
9b60e70b ID |
5203 | if (!single_major) { |
5204 | ret = register_blkdev(0, rbd_dev->name); | |
5205 | if (ret < 0) | |
5206 | goto err_out_id; | |
5207 | ||
5208 | rbd_dev->major = ret; | |
5209 | rbd_dev->minor = 0; | |
5210 | } else { | |
5211 | rbd_dev->major = rbd_major; | |
5212 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5213 | } | |
83a06263 AE |
5214 | |
5215 | /* Set up the blkdev mapping. */ | |
5216 | ||
5217 | ret = rbd_init_disk(rbd_dev); | |
5218 | if (ret) | |
5219 | goto err_out_blkdev; | |
5220 | ||
f35a4dee | 5221 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5222 | if (ret) |
5223 | goto err_out_disk; | |
bc1ecc65 | 5224 | |
f35a4dee | 5225 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5226 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee AE |
5227 | |
5228 | ret = rbd_bus_add_dev(rbd_dev); | |
5229 | if (ret) | |
f5ee37bd | 5230 | goto err_out_mapping; |
83a06263 | 5231 | |
83a06263 AE |
5232 | /* Everything's ready. Announce the disk to the world. */ |
5233 | ||
129b79d4 | 5234 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5235 | add_disk(rbd_dev->disk); |
5236 | ||
5237 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5238 | (unsigned long long) rbd_dev->mapping.size); | |
5239 | ||
5240 | return ret; | |
2f82ee54 | 5241 | |
f35a4dee AE |
5242 | err_out_mapping: |
5243 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5244 | err_out_disk: |
5245 | rbd_free_disk(rbd_dev); | |
5246 | err_out_blkdev: | |
9b60e70b ID |
5247 | if (!single_major) |
5248 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5249 | err_out_id: |
5250 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5251 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5252 | |
5253 | return ret; | |
5254 | } | |
5255 | ||
332bb12d AE |
5256 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5257 | { | |
5258 | struct rbd_spec *spec = rbd_dev->spec; | |
5259 | size_t size; | |
5260 | ||
5261 | /* Record the header object name for this rbd image. */ | |
5262 | ||
5263 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5264 | ||
5265 | if (rbd_dev->image_format == 1) | |
5266 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5267 | else | |
5268 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5269 | ||
5270 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5271 | if (!rbd_dev->header_name) | |
5272 | return -ENOMEM; | |
5273 | ||
5274 | if (rbd_dev->image_format == 1) | |
5275 | sprintf(rbd_dev->header_name, "%s%s", | |
5276 | spec->image_name, RBD_SUFFIX); | |
5277 | else | |
5278 | sprintf(rbd_dev->header_name, "%s%s", | |
5279 | RBD_HEADER_PREFIX, spec->image_id); | |
5280 | return 0; | |
5281 | } | |
5282 | ||
200a6a8b AE |
5283 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5284 | { | |
6fd48b3b | 5285 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5286 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5287 | rbd_dev->header_name = NULL; |
5288 | rbd_dev->image_format = 0; | |
5289 | kfree(rbd_dev->spec->image_id); | |
5290 | rbd_dev->spec->image_id = NULL; | |
5291 | ||
200a6a8b AE |
5292 | rbd_dev_destroy(rbd_dev); |
5293 | } | |
5294 | ||
a30b71b9 AE |
5295 | /* |
5296 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5297 | * device. If this image is the one being mapped (i.e., not a |
5298 | * parent), initiate a watch on its header object before using that | |
5299 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5300 | */ |
6d69bb53 | 5301 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth) |
a30b71b9 AE |
5302 | { |
5303 | int ret; | |
5304 | ||
5305 | /* | |
3abef3b3 AE |
5306 | * Get the id from the image id object. Unless there's an |
5307 | * error, rbd_dev->spec->image_id will be filled in with | |
5308 | * a dynamically-allocated string, and rbd_dev->image_format | |
5309 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5310 | */ |
5311 | ret = rbd_dev_image_id(rbd_dev); | |
5312 | if (ret) | |
c0fba368 | 5313 | return ret; |
c0fba368 | 5314 | |
332bb12d AE |
5315 | ret = rbd_dev_header_name(rbd_dev); |
5316 | if (ret) | |
5317 | goto err_out_format; | |
5318 | ||
6d69bb53 | 5319 | if (!depth) { |
fca27065 | 5320 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1fe48023 ID |
5321 | if (ret) { |
5322 | if (ret == -ENOENT) | |
5323 | pr_info("image %s/%s does not exist\n", | |
5324 | rbd_dev->spec->pool_name, | |
5325 | rbd_dev->spec->image_name); | |
1f3ef788 | 5326 | goto out_header_name; |
1fe48023 | 5327 | } |
1f3ef788 | 5328 | } |
b644de2b | 5329 | |
a720ae09 | 5330 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5331 | if (ret) |
b644de2b | 5332 | goto err_out_watch; |
83a06263 | 5333 | |
04077599 ID |
5334 | /* |
5335 | * If this image is the one being mapped, we have pool name and | |
5336 | * id, image name and id, and snap name - need to fill snap id. | |
5337 | * Otherwise this is a parent image, identified by pool, image | |
5338 | * and snap ids - need to fill in names for those ids. | |
5339 | */ | |
6d69bb53 | 5340 | if (!depth) |
04077599 ID |
5341 | ret = rbd_spec_fill_snap_id(rbd_dev); |
5342 | else | |
5343 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
5344 | if (ret) { |
5345 | if (ret == -ENOENT) | |
5346 | pr_info("snap %s/%s@%s does not exist\n", | |
5347 | rbd_dev->spec->pool_name, | |
5348 | rbd_dev->spec->image_name, | |
5349 | rbd_dev->spec->snap_name); | |
33dca39f | 5350 | goto err_out_probe; |
1fe48023 | 5351 | } |
9bb81c9b | 5352 | |
e8f59b59 ID |
5353 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5354 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5355 | if (ret) | |
5356 | goto err_out_probe; | |
5357 | ||
5358 | /* | |
5359 | * Need to warn users if this image is the one being | |
5360 | * mapped and has a parent. | |
5361 | */ | |
6d69bb53 | 5362 | if (!depth && rbd_dev->parent_spec) |
e8f59b59 ID |
5363 | rbd_warn(rbd_dev, |
5364 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5365 | } | |
5366 | ||
6d69bb53 | 5367 | ret = rbd_dev_probe_parent(rbd_dev, depth); |
30d60ba2 AE |
5368 | if (ret) |
5369 | goto err_out_probe; | |
5370 | ||
5371 | dout("discovered format %u image, header name is %s\n", | |
5372 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5373 | return 0; |
e8f59b59 | 5374 | |
6fd48b3b AE |
5375 | err_out_probe: |
5376 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5377 | err_out_watch: |
6d69bb53 | 5378 | if (!depth) |
fca27065 | 5379 | rbd_dev_header_unwatch_sync(rbd_dev); |
332bb12d AE |
5380 | out_header_name: |
5381 | kfree(rbd_dev->header_name); | |
5382 | rbd_dev->header_name = NULL; | |
5383 | err_out_format: | |
5384 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5385 | kfree(rbd_dev->spec->image_id); |
5386 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5387 | return ret; |
5388 | } | |
5389 | ||
9b60e70b ID |
5390 | static ssize_t do_rbd_add(struct bus_type *bus, |
5391 | const char *buf, | |
5392 | size_t count) | |
602adf40 | 5393 | { |
cb8627c7 | 5394 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5395 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5396 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5397 | struct rbd_spec *spec = NULL; |
9d3997fd | 5398 | struct rbd_client *rbdc; |
51344a38 | 5399 | bool read_only; |
27cc2594 | 5400 | int rc = -ENOMEM; |
602adf40 YS |
5401 | |
5402 | if (!try_module_get(THIS_MODULE)) | |
5403 | return -ENODEV; | |
5404 | ||
602adf40 | 5405 | /* parse add command */ |
859c31df | 5406 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5407 | if (rc < 0) |
bd4ba655 | 5408 | goto err_out_module; |
78cea76e | 5409 | |
9d3997fd AE |
5410 | rbdc = rbd_get_client(ceph_opts); |
5411 | if (IS_ERR(rbdc)) { | |
5412 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5413 | goto err_out_args; |
9d3997fd | 5414 | } |
602adf40 | 5415 | |
602adf40 | 5416 | /* pick the pool */ |
30ba1f02 | 5417 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
5418 | if (rc < 0) { |
5419 | if (rc == -ENOENT) | |
5420 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 5421 | goto err_out_client; |
1fe48023 | 5422 | } |
c0cd10db | 5423 | spec->pool_id = (u64)rc; |
859c31df | 5424 | |
0903e875 AE |
5425 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5426 | ||
c0cd10db | 5427 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5428 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5429 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5430 | rc = -EIO; |
5431 | goto err_out_client; | |
5432 | } | |
5433 | ||
d147543d | 5434 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
bd4ba655 AE |
5435 | if (!rbd_dev) |
5436 | goto err_out_client; | |
c53d5893 AE |
5437 | rbdc = NULL; /* rbd_dev now owns this */ |
5438 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 5439 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 5440 | |
6d69bb53 | 5441 | rc = rbd_dev_image_probe(rbd_dev, 0); |
a30b71b9 | 5442 | if (rc < 0) |
c53d5893 | 5443 | goto err_out_rbd_dev; |
05fd6f6f | 5444 | |
7ce4eef7 AE |
5445 | /* If we are mapping a snapshot it must be marked read-only */ |
5446 | ||
d147543d | 5447 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
5448 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
5449 | read_only = true; | |
5450 | rbd_dev->mapping.read_only = read_only; | |
5451 | ||
b536f69a | 5452 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5453 | if (rc) { |
e37180c0 ID |
5454 | /* |
5455 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5456 | * rbd_dev_image_release() without refactoring, see | |
5457 | * commit 1f3ef78861ac. | |
5458 | */ | |
5459 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5460 | rbd_dev_image_release(rbd_dev); |
5461 | goto err_out_module; | |
5462 | } | |
5463 | ||
5464 | return count; | |
b536f69a | 5465 | |
c53d5893 AE |
5466 | err_out_rbd_dev: |
5467 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5468 | err_out_client: |
9d3997fd | 5469 | rbd_put_client(rbdc); |
0ddebc0c | 5470 | err_out_args: |
859c31df | 5471 | rbd_spec_put(spec); |
d147543d | 5472 | kfree(rbd_opts); |
bd4ba655 AE |
5473 | err_out_module: |
5474 | module_put(THIS_MODULE); | |
27cc2594 | 5475 | |
602adf40 | 5476 | dout("Error adding device %s\n", buf); |
27cc2594 | 5477 | |
c0cd10db | 5478 | return (ssize_t)rc; |
602adf40 YS |
5479 | } |
5480 | ||
9b60e70b ID |
5481 | static ssize_t rbd_add(struct bus_type *bus, |
5482 | const char *buf, | |
5483 | size_t count) | |
5484 | { | |
5485 | if (single_major) | |
5486 | return -EINVAL; | |
5487 | ||
5488 | return do_rbd_add(bus, buf, count); | |
5489 | } | |
5490 | ||
5491 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5492 | const char *buf, | |
5493 | size_t count) | |
5494 | { | |
5495 | return do_rbd_add(bus, buf, count); | |
5496 | } | |
5497 | ||
200a6a8b | 5498 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5499 | { |
593a9e7b | 5500 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5501 | |
602adf40 | 5502 | rbd_free_disk(rbd_dev); |
200a6a8b | 5503 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5504 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5505 | if (!single_major) |
5506 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5507 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5508 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5509 | } |
5510 | ||
05a46afd AE |
5511 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5512 | { | |
ad945fc1 | 5513 | while (rbd_dev->parent) { |
05a46afd AE |
5514 | struct rbd_device *first = rbd_dev; |
5515 | struct rbd_device *second = first->parent; | |
5516 | struct rbd_device *third; | |
5517 | ||
5518 | /* | |
5519 | * Follow to the parent with no grandparent and | |
5520 | * remove it. | |
5521 | */ | |
5522 | while (second && (third = second->parent)) { | |
5523 | first = second; | |
5524 | second = third; | |
5525 | } | |
ad945fc1 | 5526 | rbd_assert(second); |
8ad42cd0 | 5527 | rbd_dev_image_release(second); |
ad945fc1 AE |
5528 | first->parent = NULL; |
5529 | first->parent_overlap = 0; | |
5530 | ||
5531 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5532 | rbd_spec_put(first->parent_spec); |
5533 | first->parent_spec = NULL; | |
05a46afd AE |
5534 | } |
5535 | } | |
5536 | ||
9b60e70b ID |
5537 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5538 | const char *buf, | |
5539 | size_t count) | |
602adf40 YS |
5540 | { |
5541 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5542 | struct list_head *tmp; |
5543 | int dev_id; | |
602adf40 | 5544 | unsigned long ul; |
82a442d2 | 5545 | bool already = false; |
0d8189e1 | 5546 | int ret; |
602adf40 | 5547 | |
bb8e0e84 | 5548 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5549 | if (ret) |
5550 | return ret; | |
602adf40 YS |
5551 | |
5552 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5553 | dev_id = (int)ul; |
5554 | if (dev_id != ul) | |
602adf40 YS |
5555 | return -EINVAL; |
5556 | ||
751cc0e3 AE |
5557 | ret = -ENOENT; |
5558 | spin_lock(&rbd_dev_list_lock); | |
5559 | list_for_each(tmp, &rbd_dev_list) { | |
5560 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5561 | if (rbd_dev->dev_id == dev_id) { | |
5562 | ret = 0; | |
5563 | break; | |
5564 | } | |
42382b70 | 5565 | } |
751cc0e3 AE |
5566 | if (!ret) { |
5567 | spin_lock_irq(&rbd_dev->lock); | |
5568 | if (rbd_dev->open_count) | |
5569 | ret = -EBUSY; | |
5570 | else | |
82a442d2 AE |
5571 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5572 | &rbd_dev->flags); | |
751cc0e3 AE |
5573 | spin_unlock_irq(&rbd_dev->lock); |
5574 | } | |
5575 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5576 | if (ret < 0 || already) |
1ba0f1e7 | 5577 | return ret; |
751cc0e3 | 5578 | |
fca27065 | 5579 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5580 | /* |
5581 | * flush remaining watch callbacks - these must be complete | |
5582 | * before the osd_client is shutdown | |
5583 | */ | |
5584 | dout("%s: flushing notifies", __func__); | |
5585 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5586 | |
9875201e JD |
5587 | /* |
5588 | * Don't free anything from rbd_dev->disk until after all | |
5589 | * notifies are completely processed. Otherwise | |
5590 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5591 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5592 | */ | |
5593 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5594 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5595 | module_put(THIS_MODULE); |
aafb230e | 5596 | |
1ba0f1e7 | 5597 | return count; |
602adf40 YS |
5598 | } |
5599 | ||
9b60e70b ID |
5600 | static ssize_t rbd_remove(struct bus_type *bus, |
5601 | const char *buf, | |
5602 | size_t count) | |
5603 | { | |
5604 | if (single_major) | |
5605 | return -EINVAL; | |
5606 | ||
5607 | return do_rbd_remove(bus, buf, count); | |
5608 | } | |
5609 | ||
5610 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5611 | const char *buf, | |
5612 | size_t count) | |
5613 | { | |
5614 | return do_rbd_remove(bus, buf, count); | |
5615 | } | |
5616 | ||
602adf40 YS |
5617 | /* |
5618 | * create control files in sysfs | |
dfc5606d | 5619 | * /sys/bus/rbd/... |
602adf40 YS |
5620 | */ |
5621 | static int rbd_sysfs_init(void) | |
5622 | { | |
dfc5606d | 5623 | int ret; |
602adf40 | 5624 | |
fed4c143 | 5625 | ret = device_register(&rbd_root_dev); |
21079786 | 5626 | if (ret < 0) |
dfc5606d | 5627 | return ret; |
602adf40 | 5628 | |
fed4c143 AE |
5629 | ret = bus_register(&rbd_bus_type); |
5630 | if (ret < 0) | |
5631 | device_unregister(&rbd_root_dev); | |
602adf40 | 5632 | |
602adf40 YS |
5633 | return ret; |
5634 | } | |
5635 | ||
5636 | static void rbd_sysfs_cleanup(void) | |
5637 | { | |
dfc5606d | 5638 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5639 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5640 | } |
5641 | ||
1c2a9dfe AE |
5642 | static int rbd_slab_init(void) |
5643 | { | |
5644 | rbd_assert(!rbd_img_request_cache); | |
5645 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5646 | sizeof (struct rbd_img_request), | |
5647 | __alignof__(struct rbd_img_request), | |
5648 | 0, NULL); | |
868311b1 AE |
5649 | if (!rbd_img_request_cache) |
5650 | return -ENOMEM; | |
5651 | ||
5652 | rbd_assert(!rbd_obj_request_cache); | |
5653 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5654 | sizeof (struct rbd_obj_request), | |
5655 | __alignof__(struct rbd_obj_request), | |
5656 | 0, NULL); | |
78c2a44a AE |
5657 | if (!rbd_obj_request_cache) |
5658 | goto out_err; | |
5659 | ||
5660 | rbd_assert(!rbd_segment_name_cache); | |
5661 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5662 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5663 | if (rbd_segment_name_cache) |
1c2a9dfe | 5664 | return 0; |
78c2a44a AE |
5665 | out_err: |
5666 | if (rbd_obj_request_cache) { | |
5667 | kmem_cache_destroy(rbd_obj_request_cache); | |
5668 | rbd_obj_request_cache = NULL; | |
5669 | } | |
1c2a9dfe | 5670 | |
868311b1 AE |
5671 | kmem_cache_destroy(rbd_img_request_cache); |
5672 | rbd_img_request_cache = NULL; | |
5673 | ||
1c2a9dfe AE |
5674 | return -ENOMEM; |
5675 | } | |
5676 | ||
5677 | static void rbd_slab_exit(void) | |
5678 | { | |
78c2a44a AE |
5679 | rbd_assert(rbd_segment_name_cache); |
5680 | kmem_cache_destroy(rbd_segment_name_cache); | |
5681 | rbd_segment_name_cache = NULL; | |
5682 | ||
868311b1 AE |
5683 | rbd_assert(rbd_obj_request_cache); |
5684 | kmem_cache_destroy(rbd_obj_request_cache); | |
5685 | rbd_obj_request_cache = NULL; | |
5686 | ||
1c2a9dfe AE |
5687 | rbd_assert(rbd_img_request_cache); |
5688 | kmem_cache_destroy(rbd_img_request_cache); | |
5689 | rbd_img_request_cache = NULL; | |
5690 | } | |
5691 | ||
cc344fa1 | 5692 | static int __init rbd_init(void) |
602adf40 YS |
5693 | { |
5694 | int rc; | |
5695 | ||
1e32d34c AE |
5696 | if (!libceph_compatible(NULL)) { |
5697 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5698 | return -EINVAL; |
5699 | } | |
e1b4d96d | 5700 | |
1c2a9dfe | 5701 | rc = rbd_slab_init(); |
602adf40 YS |
5702 | if (rc) |
5703 | return rc; | |
e1b4d96d | 5704 | |
f5ee37bd ID |
5705 | /* |
5706 | * The number of active work items is limited by the number of | |
f77303bd | 5707 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
5708 | */ |
5709 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
5710 | if (!rbd_wq) { | |
5711 | rc = -ENOMEM; | |
5712 | goto err_out_slab; | |
5713 | } | |
5714 | ||
9b60e70b ID |
5715 | if (single_major) { |
5716 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5717 | if (rbd_major < 0) { | |
5718 | rc = rbd_major; | |
f5ee37bd | 5719 | goto err_out_wq; |
9b60e70b ID |
5720 | } |
5721 | } | |
5722 | ||
1c2a9dfe AE |
5723 | rc = rbd_sysfs_init(); |
5724 | if (rc) | |
9b60e70b ID |
5725 | goto err_out_blkdev; |
5726 | ||
5727 | if (single_major) | |
5728 | pr_info("loaded (major %d)\n", rbd_major); | |
5729 | else | |
5730 | pr_info("loaded\n"); | |
1c2a9dfe | 5731 | |
e1b4d96d ID |
5732 | return 0; |
5733 | ||
9b60e70b ID |
5734 | err_out_blkdev: |
5735 | if (single_major) | |
5736 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
5737 | err_out_wq: |
5738 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
5739 | err_out_slab: |
5740 | rbd_slab_exit(); | |
1c2a9dfe | 5741 | return rc; |
602adf40 YS |
5742 | } |
5743 | ||
cc344fa1 | 5744 | static void __exit rbd_exit(void) |
602adf40 | 5745 | { |
ffe312cf | 5746 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5747 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5748 | if (single_major) |
5749 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 5750 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 5751 | rbd_slab_exit(); |
602adf40 YS |
5752 | } |
5753 | ||
5754 | module_init(rbd_init); | |
5755 | module_exit(rbd_exit); | |
5756 | ||
d552c619 | 5757 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5758 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5759 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5760 | /* following authorship retained from original osdblk.c */ |
5761 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5762 | ||
90da258b | 5763 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5764 | MODULE_LICENSE("GPL"); |