Commit | Line | Data |
---|---|---|
e2a58ee5 | 1 | |
602adf40 YS |
2 | /* |
3 | rbd.c -- Export ceph rados objects as a Linux block device | |
4 | ||
5 | ||
6 | based on drivers/block/osdblk.c: | |
7 | ||
8 | Copyright 2009 Red Hat, Inc. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; see the file COPYING. If not, write to | |
21 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
22 | ||
23 | ||
24 | ||
dfc5606d | 25 | For usage instructions, please refer to: |
602adf40 | 26 | |
dfc5606d | 27 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
28 | |
29 | */ | |
30 | ||
31 | #include <linux/ceph/libceph.h> | |
32 | #include <linux/ceph/osd_client.h> | |
33 | #include <linux/ceph/mon_client.h> | |
34 | #include <linux/ceph/decode.h> | |
59c2be1e | 35 | #include <linux/parser.h> |
30d1cff8 | 36 | #include <linux/bsearch.h> |
602adf40 YS |
37 | |
38 | #include <linux/kernel.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/module.h> | |
41 | #include <linux/fs.h> | |
42 | #include <linux/blkdev.h> | |
1c2a9dfe | 43 | #include <linux/slab.h> |
f8a22fc2 | 44 | #include <linux/idr.h> |
602adf40 YS |
45 | |
46 | #include "rbd_types.h" | |
47 | ||
aafb230e AE |
48 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
49 | ||
593a9e7b AE |
50 | /* |
51 | * The basic unit of block I/O is a sector. It is interpreted in a | |
52 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
53 | * universally 512 bytes. These symbols are just slightly more | |
54 | * meaningful than the bare numbers they represent. | |
55 | */ | |
56 | #define SECTOR_SHIFT 9 | |
57 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
58 | ||
a2acd00e AE |
59 | /* |
60 | * Increment the given counter and return its updated value. | |
61 | * If the counter is already 0 it will not be incremented. | |
62 | * If the counter is already at its maximum value returns | |
63 | * -EINVAL without updating it. | |
64 | */ | |
65 | static int atomic_inc_return_safe(atomic_t *v) | |
66 | { | |
67 | unsigned int counter; | |
68 | ||
69 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
70 | if (counter <= (unsigned int)INT_MAX) | |
71 | return (int)counter; | |
72 | ||
73 | atomic_dec(v); | |
74 | ||
75 | return -EINVAL; | |
76 | } | |
77 | ||
78 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
79 | static int atomic_dec_return_safe(atomic_t *v) | |
80 | { | |
81 | int counter; | |
82 | ||
83 | counter = atomic_dec_return(v); | |
84 | if (counter >= 0) | |
85 | return counter; | |
86 | ||
87 | atomic_inc(v); | |
88 | ||
89 | return -EINVAL; | |
90 | } | |
91 | ||
f0f8cef5 | 92 | #define RBD_DRV_NAME "rbd" |
602adf40 | 93 | |
7e513d43 ID |
94 | #define RBD_MINORS_PER_MAJOR 256 |
95 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 96 | |
d4b125e9 AE |
97 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
98 | #define RBD_MAX_SNAP_NAME_LEN \ | |
99 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
100 | ||
35d489f9 | 101 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
102 | |
103 | #define RBD_SNAP_HEAD_NAME "-" | |
104 | ||
9682fc6d AE |
105 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
106 | ||
9e15b77d AE |
107 | /* This allows a single page to hold an image name sent by OSD */ |
108 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 109 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 110 | |
1e130199 | 111 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 112 | |
d889140c AE |
113 | /* Feature bits */ |
114 | ||
5cbf6f12 AE |
115 | #define RBD_FEATURE_LAYERING (1<<0) |
116 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
117 | #define RBD_FEATURES_ALL \ | |
118 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
119 | |
120 | /* Features supported by this (client software) implementation. */ | |
121 | ||
770eba6e | 122 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 123 | |
81a89793 AE |
124 | /* |
125 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
126 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
127 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
128 | * enough to hold all possible device names. | |
129 | */ | |
602adf40 | 130 | #define DEV_NAME_LEN 32 |
81a89793 | 131 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
132 | |
133 | /* | |
134 | * block device image metadata (in-memory version) | |
135 | */ | |
136 | struct rbd_image_header { | |
f35a4dee | 137 | /* These six fields never change for a given rbd image */ |
849b4260 | 138 | char *object_prefix; |
602adf40 YS |
139 | __u8 obj_order; |
140 | __u8 crypt_type; | |
141 | __u8 comp_type; | |
f35a4dee AE |
142 | u64 stripe_unit; |
143 | u64 stripe_count; | |
144 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 145 | |
f84344f3 AE |
146 | /* The remaining fields need to be updated occasionally */ |
147 | u64 image_size; | |
148 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
149 | char *snap_names; /* format 1 only */ |
150 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
151 | }; |
152 | ||
0d7dbfce AE |
153 | /* |
154 | * An rbd image specification. | |
155 | * | |
156 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
157 | * identify an image. Each rbd_dev structure includes a pointer to |
158 | * an rbd_spec structure that encapsulates this identity. | |
159 | * | |
160 | * Each of the id's in an rbd_spec has an associated name. For a | |
161 | * user-mapped image, the names are supplied and the id's associated | |
162 | * with them are looked up. For a layered image, a parent image is | |
163 | * defined by the tuple, and the names are looked up. | |
164 | * | |
165 | * An rbd_dev structure contains a parent_spec pointer which is | |
166 | * non-null if the image it represents is a child in a layered | |
167 | * image. This pointer will refer to the rbd_spec structure used | |
168 | * by the parent rbd_dev for its own identity (i.e., the structure | |
169 | * is shared between the parent and child). | |
170 | * | |
171 | * Since these structures are populated once, during the discovery | |
172 | * phase of image construction, they are effectively immutable so | |
173 | * we make no effort to synchronize access to them. | |
174 | * | |
175 | * Note that code herein does not assume the image name is known (it | |
176 | * could be a null pointer). | |
0d7dbfce AE |
177 | */ |
178 | struct rbd_spec { | |
179 | u64 pool_id; | |
ecb4dc22 | 180 | const char *pool_name; |
0d7dbfce | 181 | |
ecb4dc22 AE |
182 | const char *image_id; |
183 | const char *image_name; | |
0d7dbfce AE |
184 | |
185 | u64 snap_id; | |
ecb4dc22 | 186 | const char *snap_name; |
0d7dbfce AE |
187 | |
188 | struct kref kref; | |
189 | }; | |
190 | ||
602adf40 | 191 | /* |
f0f8cef5 | 192 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
193 | */ |
194 | struct rbd_client { | |
195 | struct ceph_client *client; | |
196 | struct kref kref; | |
197 | struct list_head node; | |
198 | }; | |
199 | ||
bf0d5f50 AE |
200 | struct rbd_img_request; |
201 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
202 | ||
203 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
204 | ||
205 | struct rbd_obj_request; | |
206 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
207 | ||
9969ebc5 AE |
208 | enum obj_request_type { |
209 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
210 | }; | |
bf0d5f50 | 211 | |
926f9b3f AE |
212 | enum obj_req_flags { |
213 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 214 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
215 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
216 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
217 | }; |
218 | ||
bf0d5f50 AE |
219 | struct rbd_obj_request { |
220 | const char *object_name; | |
221 | u64 offset; /* object start byte */ | |
222 | u64 length; /* bytes from offset */ | |
926f9b3f | 223 | unsigned long flags; |
bf0d5f50 | 224 | |
c5b5ef6c AE |
225 | /* |
226 | * An object request associated with an image will have its | |
227 | * img_data flag set; a standalone object request will not. | |
228 | * | |
229 | * A standalone object request will have which == BAD_WHICH | |
230 | * and a null obj_request pointer. | |
231 | * | |
232 | * An object request initiated in support of a layered image | |
233 | * object (to check for its existence before a write) will | |
234 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
235 | * | |
236 | * Finally, an object request for rbd image data will have | |
237 | * which != BAD_WHICH, and will have a non-null img_request | |
238 | * pointer. The value of which will be in the range | |
239 | * 0..(img_request->obj_request_count-1). | |
240 | */ | |
241 | union { | |
242 | struct rbd_obj_request *obj_request; /* STAT op */ | |
243 | struct { | |
244 | struct rbd_img_request *img_request; | |
245 | u64 img_offset; | |
246 | /* links for img_request->obj_requests list */ | |
247 | struct list_head links; | |
248 | }; | |
249 | }; | |
bf0d5f50 AE |
250 | u32 which; /* posn image request list */ |
251 | ||
252 | enum obj_request_type type; | |
788e2df3 AE |
253 | union { |
254 | struct bio *bio_list; | |
255 | struct { | |
256 | struct page **pages; | |
257 | u32 page_count; | |
258 | }; | |
259 | }; | |
0eefd470 | 260 | struct page **copyup_pages; |
ebda6408 | 261 | u32 copyup_page_count; |
bf0d5f50 AE |
262 | |
263 | struct ceph_osd_request *osd_req; | |
264 | ||
265 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 266 | int result; |
bf0d5f50 AE |
267 | |
268 | rbd_obj_callback_t callback; | |
788e2df3 | 269 | struct completion completion; |
bf0d5f50 AE |
270 | |
271 | struct kref kref; | |
272 | }; | |
273 | ||
0c425248 | 274 | enum img_req_flags { |
9849e986 AE |
275 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
276 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 277 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
0c425248 AE |
278 | }; |
279 | ||
bf0d5f50 | 280 | struct rbd_img_request { |
bf0d5f50 AE |
281 | struct rbd_device *rbd_dev; |
282 | u64 offset; /* starting image byte offset */ | |
283 | u64 length; /* byte count from offset */ | |
0c425248 | 284 | unsigned long flags; |
bf0d5f50 | 285 | union { |
9849e986 | 286 | u64 snap_id; /* for reads */ |
bf0d5f50 | 287 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
288 | }; |
289 | union { | |
290 | struct request *rq; /* block request */ | |
291 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 292 | }; |
3d7efd18 | 293 | struct page **copyup_pages; |
ebda6408 | 294 | u32 copyup_page_count; |
bf0d5f50 AE |
295 | spinlock_t completion_lock;/* protects next_completion */ |
296 | u32 next_completion; | |
297 | rbd_img_callback_t callback; | |
55f27e09 | 298 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 299 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
300 | |
301 | u32 obj_request_count; | |
302 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
303 | ||
304 | struct kref kref; | |
305 | }; | |
306 | ||
307 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 308 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 309 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 310 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 311 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 312 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 313 | |
f84344f3 | 314 | struct rbd_mapping { |
99c1f08f | 315 | u64 size; |
34b13184 | 316 | u64 features; |
f84344f3 AE |
317 | bool read_only; |
318 | }; | |
319 | ||
602adf40 YS |
320 | /* |
321 | * a single device | |
322 | */ | |
323 | struct rbd_device { | |
de71a297 | 324 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
325 | |
326 | int major; /* blkdev assigned major */ | |
dd82fff1 | 327 | int minor; |
602adf40 | 328 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 329 | |
a30b71b9 | 330 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
331 | struct rbd_client *rbd_client; |
332 | ||
333 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
334 | ||
b82d167b | 335 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
336 | |
337 | struct rbd_image_header header; | |
b82d167b | 338 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 339 | struct rbd_spec *spec; |
602adf40 | 340 | |
0d7dbfce | 341 | char *header_name; |
971f839a | 342 | |
0903e875 AE |
343 | struct ceph_file_layout layout; |
344 | ||
59c2be1e | 345 | struct ceph_osd_event *watch_event; |
975241af | 346 | struct rbd_obj_request *watch_request; |
59c2be1e | 347 | |
86b00e0d AE |
348 | struct rbd_spec *parent_spec; |
349 | u64 parent_overlap; | |
a2acd00e | 350 | atomic_t parent_ref; |
2f82ee54 | 351 | struct rbd_device *parent; |
86b00e0d | 352 | |
c666601a JD |
353 | /* protects updating the header */ |
354 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
355 | |
356 | struct rbd_mapping mapping; | |
602adf40 YS |
357 | |
358 | struct list_head node; | |
dfc5606d | 359 | |
dfc5606d YS |
360 | /* sysfs related */ |
361 | struct device dev; | |
b82d167b | 362 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
363 | }; |
364 | ||
b82d167b AE |
365 | /* |
366 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
367 | * rbd_dev->lock is used to protect access. | |
368 | * | |
369 | * Currently, only the "removing" flag (which is coupled with the | |
370 | * "open_count" field) requires atomic access. | |
371 | */ | |
6d292906 AE |
372 | enum rbd_dev_flags { |
373 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 374 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
375 | }; |
376 | ||
cfbf6377 | 377 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 378 | |
602adf40 | 379 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
380 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
381 | ||
432b8587 AE |
382 | static LIST_HEAD(rbd_client_list); /* clients */ |
383 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 384 | |
78c2a44a AE |
385 | /* Slab caches for frequently-allocated structures */ |
386 | ||
1c2a9dfe | 387 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 388 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 389 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 390 | |
9b60e70b | 391 | static int rbd_major; |
f8a22fc2 ID |
392 | static DEFINE_IDA(rbd_dev_id_ida); |
393 | ||
9b60e70b ID |
394 | /* |
395 | * Default to false for now, as single-major requires >= 0.75 version of | |
396 | * userspace rbd utility. | |
397 | */ | |
398 | static bool single_major = false; | |
399 | module_param(single_major, bool, S_IRUGO); | |
400 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
401 | ||
3d7efd18 AE |
402 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
403 | ||
200a6a8b | 404 | static void rbd_dev_device_release(struct device *dev); |
dfc5606d | 405 | |
f0f8cef5 AE |
406 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
407 | size_t count); | |
408 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
409 | size_t count); | |
9b60e70b ID |
410 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
411 | size_t count); | |
412 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
413 | size_t count); | |
1f3ef788 | 414 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping); |
a2acd00e | 415 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 416 | |
9b60e70b ID |
417 | static int rbd_dev_id_to_minor(int dev_id) |
418 | { | |
7e513d43 | 419 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
420 | } |
421 | ||
422 | static int minor_to_rbd_dev_id(int minor) | |
423 | { | |
7e513d43 | 424 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
425 | } |
426 | ||
b15a21dd GKH |
427 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
428 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
429 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
430 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
431 | |
432 | static struct attribute *rbd_bus_attrs[] = { | |
433 | &bus_attr_add.attr, | |
434 | &bus_attr_remove.attr, | |
9b60e70b ID |
435 | &bus_attr_add_single_major.attr, |
436 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 437 | NULL, |
f0f8cef5 | 438 | }; |
92c76dc0 ID |
439 | |
440 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
441 | struct attribute *attr, int index) | |
442 | { | |
9b60e70b ID |
443 | if (!single_major && |
444 | (attr == &bus_attr_add_single_major.attr || | |
445 | attr == &bus_attr_remove_single_major.attr)) | |
446 | return 0; | |
447 | ||
92c76dc0 ID |
448 | return attr->mode; |
449 | } | |
450 | ||
451 | static const struct attribute_group rbd_bus_group = { | |
452 | .attrs = rbd_bus_attrs, | |
453 | .is_visible = rbd_bus_is_visible, | |
454 | }; | |
455 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
456 | |
457 | static struct bus_type rbd_bus_type = { | |
458 | .name = "rbd", | |
b15a21dd | 459 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
460 | }; |
461 | ||
462 | static void rbd_root_dev_release(struct device *dev) | |
463 | { | |
464 | } | |
465 | ||
466 | static struct device rbd_root_dev = { | |
467 | .init_name = "rbd", | |
468 | .release = rbd_root_dev_release, | |
469 | }; | |
470 | ||
06ecc6cb AE |
471 | static __printf(2, 3) |
472 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
473 | { | |
474 | struct va_format vaf; | |
475 | va_list args; | |
476 | ||
477 | va_start(args, fmt); | |
478 | vaf.fmt = fmt; | |
479 | vaf.va = &args; | |
480 | ||
481 | if (!rbd_dev) | |
482 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
483 | else if (rbd_dev->disk) | |
484 | printk(KERN_WARNING "%s: %s: %pV\n", | |
485 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
486 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
487 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
488 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
489 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
490 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
491 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
492 | else /* punt */ | |
493 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
494 | RBD_DRV_NAME, rbd_dev, &vaf); | |
495 | va_end(args); | |
496 | } | |
497 | ||
aafb230e AE |
498 | #ifdef RBD_DEBUG |
499 | #define rbd_assert(expr) \ | |
500 | if (unlikely(!(expr))) { \ | |
501 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
502 | "at line %d:\n\n" \ | |
503 | "\trbd_assert(%s);\n\n", \ | |
504 | __func__, __LINE__, #expr); \ | |
505 | BUG(); \ | |
506 | } | |
507 | #else /* !RBD_DEBUG */ | |
508 | # define rbd_assert(expr) ((void) 0) | |
509 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 510 | |
b454e36d | 511 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
512 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
513 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 514 | |
cc4a38bd | 515 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 AE |
516 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
517 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev); | |
54cac61f AE |
518 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
519 | u64 snap_id); | |
2ad3d716 AE |
520 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
521 | u8 *order, u64 *snap_size); | |
522 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
523 | u64 *snap_features); | |
524 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 525 | |
602adf40 YS |
526 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
527 | { | |
f0f8cef5 | 528 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 529 | bool removing = false; |
602adf40 | 530 | |
f84344f3 | 531 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
532 | return -EROFS; |
533 | ||
a14ea269 | 534 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
535 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
536 | removing = true; | |
537 | else | |
538 | rbd_dev->open_count++; | |
a14ea269 | 539 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
540 | if (removing) |
541 | return -ENOENT; | |
542 | ||
c3e946ce | 543 | (void) get_device(&rbd_dev->dev); |
f84344f3 | 544 | set_device_ro(bdev, rbd_dev->mapping.read_only); |
340c7a2b | 545 | |
602adf40 YS |
546 | return 0; |
547 | } | |
548 | ||
db2a144b | 549 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
550 | { |
551 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
552 | unsigned long open_count_before; |
553 | ||
a14ea269 | 554 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 555 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 556 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 557 | rbd_assert(open_count_before > 0); |
dfc5606d | 558 | |
c3e946ce | 559 | put_device(&rbd_dev->dev); |
dfc5606d YS |
560 | } |
561 | ||
131fd9f6 GZ |
562 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
563 | { | |
77f33c03 | 564 | int ret = 0; |
131fd9f6 GZ |
565 | int val; |
566 | bool ro; | |
77f33c03 | 567 | bool ro_changed = false; |
131fd9f6 | 568 | |
77f33c03 | 569 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
570 | if (get_user(val, (int __user *)(arg))) |
571 | return -EFAULT; | |
572 | ||
573 | ro = val ? true : false; | |
574 | /* Snapshot doesn't allow to write*/ | |
575 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
576 | return -EROFS; | |
577 | ||
77f33c03 JD |
578 | spin_lock_irq(&rbd_dev->lock); |
579 | /* prevent others open this device */ | |
580 | if (rbd_dev->open_count > 1) { | |
581 | ret = -EBUSY; | |
582 | goto out; | |
583 | } | |
584 | ||
131fd9f6 GZ |
585 | if (rbd_dev->mapping.read_only != ro) { |
586 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 587 | ro_changed = true; |
131fd9f6 GZ |
588 | } |
589 | ||
77f33c03 JD |
590 | out: |
591 | spin_unlock_irq(&rbd_dev->lock); | |
592 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
593 | if (ret == 0 && ro_changed) | |
594 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
595 | ||
596 | return ret; | |
131fd9f6 GZ |
597 | } |
598 | ||
599 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
600 | unsigned int cmd, unsigned long arg) | |
601 | { | |
602 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
603 | int ret = 0; | |
604 | ||
131fd9f6 GZ |
605 | switch (cmd) { |
606 | case BLKROSET: | |
607 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
608 | break; | |
609 | default: | |
610 | ret = -ENOTTY; | |
611 | } | |
612 | ||
131fd9f6 GZ |
613 | return ret; |
614 | } | |
615 | ||
616 | #ifdef CONFIG_COMPAT | |
617 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
618 | unsigned int cmd, unsigned long arg) | |
619 | { | |
620 | return rbd_ioctl(bdev, mode, cmd, arg); | |
621 | } | |
622 | #endif /* CONFIG_COMPAT */ | |
623 | ||
602adf40 YS |
624 | static const struct block_device_operations rbd_bd_ops = { |
625 | .owner = THIS_MODULE, | |
626 | .open = rbd_open, | |
dfc5606d | 627 | .release = rbd_release, |
131fd9f6 GZ |
628 | .ioctl = rbd_ioctl, |
629 | #ifdef CONFIG_COMPAT | |
630 | .compat_ioctl = rbd_compat_ioctl, | |
631 | #endif | |
602adf40 YS |
632 | }; |
633 | ||
634 | /* | |
7262cfca | 635 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 636 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 637 | */ |
f8c38929 | 638 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
639 | { |
640 | struct rbd_client *rbdc; | |
641 | int ret = -ENOMEM; | |
642 | ||
37206ee5 | 643 | dout("%s:\n", __func__); |
602adf40 YS |
644 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
645 | if (!rbdc) | |
646 | goto out_opt; | |
647 | ||
648 | kref_init(&rbdc->kref); | |
649 | INIT_LIST_HEAD(&rbdc->node); | |
650 | ||
43ae4701 | 651 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 652 | if (IS_ERR(rbdc->client)) |
08f75463 | 653 | goto out_rbdc; |
43ae4701 | 654 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
655 | |
656 | ret = ceph_open_session(rbdc->client); | |
657 | if (ret < 0) | |
08f75463 | 658 | goto out_client; |
602adf40 | 659 | |
432b8587 | 660 | spin_lock(&rbd_client_list_lock); |
602adf40 | 661 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 662 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 663 | |
37206ee5 | 664 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 665 | |
602adf40 | 666 | return rbdc; |
08f75463 | 667 | out_client: |
602adf40 | 668 | ceph_destroy_client(rbdc->client); |
08f75463 | 669 | out_rbdc: |
602adf40 YS |
670 | kfree(rbdc); |
671 | out_opt: | |
43ae4701 AE |
672 | if (ceph_opts) |
673 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
674 | dout("%s: error %d\n", __func__, ret); |
675 | ||
28f259b7 | 676 | return ERR_PTR(ret); |
602adf40 YS |
677 | } |
678 | ||
2f82ee54 AE |
679 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
680 | { | |
681 | kref_get(&rbdc->kref); | |
682 | ||
683 | return rbdc; | |
684 | } | |
685 | ||
602adf40 | 686 | /* |
1f7ba331 AE |
687 | * Find a ceph client with specific addr and configuration. If |
688 | * found, bump its reference count. | |
602adf40 | 689 | */ |
1f7ba331 | 690 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
691 | { |
692 | struct rbd_client *client_node; | |
1f7ba331 | 693 | bool found = false; |
602adf40 | 694 | |
43ae4701 | 695 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
696 | return NULL; |
697 | ||
1f7ba331 AE |
698 | spin_lock(&rbd_client_list_lock); |
699 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
700 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
701 | __rbd_get_client(client_node); |
702 | ||
1f7ba331 AE |
703 | found = true; |
704 | break; | |
705 | } | |
706 | } | |
707 | spin_unlock(&rbd_client_list_lock); | |
708 | ||
709 | return found ? client_node : NULL; | |
602adf40 YS |
710 | } |
711 | ||
59c2be1e YS |
712 | /* |
713 | * mount options | |
714 | */ | |
715 | enum { | |
59c2be1e YS |
716 | Opt_last_int, |
717 | /* int args above */ | |
718 | Opt_last_string, | |
719 | /* string args above */ | |
cc0538b6 AE |
720 | Opt_read_only, |
721 | Opt_read_write, | |
722 | /* Boolean args above */ | |
723 | Opt_last_bool, | |
59c2be1e YS |
724 | }; |
725 | ||
43ae4701 | 726 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
727 | /* int args above */ |
728 | /* string args above */ | |
be466c1c | 729 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
730 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
731 | {Opt_read_write, "read_write"}, | |
732 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
733 | /* Boolean args above */ | |
59c2be1e YS |
734 | {-1, NULL} |
735 | }; | |
736 | ||
98571b5a AE |
737 | struct rbd_options { |
738 | bool read_only; | |
739 | }; | |
740 | ||
741 | #define RBD_READ_ONLY_DEFAULT false | |
742 | ||
59c2be1e YS |
743 | static int parse_rbd_opts_token(char *c, void *private) |
744 | { | |
43ae4701 | 745 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
746 | substring_t argstr[MAX_OPT_ARGS]; |
747 | int token, intval, ret; | |
748 | ||
43ae4701 | 749 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
750 | if (token < 0) |
751 | return -EINVAL; | |
752 | ||
753 | if (token < Opt_last_int) { | |
754 | ret = match_int(&argstr[0], &intval); | |
755 | if (ret < 0) { | |
756 | pr_err("bad mount option arg (not int) " | |
757 | "at '%s'\n", c); | |
758 | return ret; | |
759 | } | |
760 | dout("got int token %d val %d\n", token, intval); | |
761 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
762 | dout("got string token %d val %s\n", token, | |
763 | argstr[0].from); | |
cc0538b6 AE |
764 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
765 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
766 | } else { |
767 | dout("got token %d\n", token); | |
768 | } | |
769 | ||
770 | switch (token) { | |
cc0538b6 AE |
771 | case Opt_read_only: |
772 | rbd_opts->read_only = true; | |
773 | break; | |
774 | case Opt_read_write: | |
775 | rbd_opts->read_only = false; | |
776 | break; | |
59c2be1e | 777 | default: |
aafb230e AE |
778 | rbd_assert(false); |
779 | break; | |
59c2be1e YS |
780 | } |
781 | return 0; | |
782 | } | |
783 | ||
602adf40 YS |
784 | /* |
785 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
786 | * not exist create it. Either way, ceph_opts is consumed by this |
787 | * function. | |
602adf40 | 788 | */ |
9d3997fd | 789 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 790 | { |
f8c38929 | 791 | struct rbd_client *rbdc; |
59c2be1e | 792 | |
cfbf6377 | 793 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 794 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 795 | if (rbdc) /* using an existing client */ |
43ae4701 | 796 | ceph_destroy_options(ceph_opts); |
9d3997fd | 797 | else |
f8c38929 | 798 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 799 | mutex_unlock(&client_mutex); |
602adf40 | 800 | |
9d3997fd | 801 | return rbdc; |
602adf40 YS |
802 | } |
803 | ||
804 | /* | |
805 | * Destroy ceph client | |
d23a4b3f | 806 | * |
432b8587 | 807 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
808 | */ |
809 | static void rbd_client_release(struct kref *kref) | |
810 | { | |
811 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
812 | ||
37206ee5 | 813 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 814 | spin_lock(&rbd_client_list_lock); |
602adf40 | 815 | list_del(&rbdc->node); |
cd9d9f5d | 816 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
817 | |
818 | ceph_destroy_client(rbdc->client); | |
819 | kfree(rbdc); | |
820 | } | |
821 | ||
822 | /* | |
823 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
824 | * it. | |
825 | */ | |
9d3997fd | 826 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 827 | { |
c53d5893 AE |
828 | if (rbdc) |
829 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
830 | } |
831 | ||
a30b71b9 AE |
832 | static bool rbd_image_format_valid(u32 image_format) |
833 | { | |
834 | return image_format == 1 || image_format == 2; | |
835 | } | |
836 | ||
8e94af8e AE |
837 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
838 | { | |
103a150f AE |
839 | size_t size; |
840 | u32 snap_count; | |
841 | ||
842 | /* The header has to start with the magic rbd header text */ | |
843 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
844 | return false; | |
845 | ||
db2388b6 AE |
846 | /* The bio layer requires at least sector-sized I/O */ |
847 | ||
848 | if (ondisk->options.order < SECTOR_SHIFT) | |
849 | return false; | |
850 | ||
851 | /* If we use u64 in a few spots we may be able to loosen this */ | |
852 | ||
853 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
854 | return false; | |
855 | ||
103a150f AE |
856 | /* |
857 | * The size of a snapshot header has to fit in a size_t, and | |
858 | * that limits the number of snapshots. | |
859 | */ | |
860 | snap_count = le32_to_cpu(ondisk->snap_count); | |
861 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
862 | if (snap_count > size / sizeof (__le64)) | |
863 | return false; | |
864 | ||
865 | /* | |
866 | * Not only that, but the size of the entire the snapshot | |
867 | * header must also be representable in a size_t. | |
868 | */ | |
869 | size -= snap_count * sizeof (__le64); | |
870 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
871 | return false; | |
872 | ||
873 | return true; | |
8e94af8e AE |
874 | } |
875 | ||
602adf40 | 876 | /* |
bb23e37a AE |
877 | * Fill an rbd image header with information from the given format 1 |
878 | * on-disk header. | |
602adf40 | 879 | */ |
662518b1 | 880 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 881 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 882 | { |
662518b1 | 883 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
884 | bool first_time = header->object_prefix == NULL; |
885 | struct ceph_snap_context *snapc; | |
886 | char *object_prefix = NULL; | |
887 | char *snap_names = NULL; | |
888 | u64 *snap_sizes = NULL; | |
ccece235 | 889 | u32 snap_count; |
d2bb24e5 | 890 | size_t size; |
bb23e37a | 891 | int ret = -ENOMEM; |
621901d6 | 892 | u32 i; |
602adf40 | 893 | |
bb23e37a | 894 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 895 | |
bb23e37a AE |
896 | if (first_time) { |
897 | size_t len; | |
103a150f | 898 | |
bb23e37a AE |
899 | len = strnlen(ondisk->object_prefix, |
900 | sizeof (ondisk->object_prefix)); | |
901 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
902 | if (!object_prefix) | |
903 | return -ENOMEM; | |
904 | memcpy(object_prefix, ondisk->object_prefix, len); | |
905 | object_prefix[len] = '\0'; | |
906 | } | |
00f1f36f | 907 | |
bb23e37a | 908 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 909 | |
bb23e37a AE |
910 | snap_count = le32_to_cpu(ondisk->snap_count); |
911 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
912 | if (!snapc) | |
913 | goto out_err; | |
914 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 915 | if (snap_count) { |
bb23e37a | 916 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
917 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
918 | ||
bb23e37a | 919 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 920 | |
bb23e37a AE |
921 | if (snap_names_len > (u64)SIZE_MAX) |
922 | goto out_2big; | |
923 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
924 | if (!snap_names) | |
6a52325f AE |
925 | goto out_err; |
926 | ||
bb23e37a | 927 | /* ...as well as the array of their sizes. */ |
621901d6 | 928 | |
d2bb24e5 | 929 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
930 | snap_sizes = kmalloc(size, GFP_KERNEL); |
931 | if (!snap_sizes) | |
6a52325f | 932 | goto out_err; |
bb23e37a | 933 | |
f785cc1d | 934 | /* |
bb23e37a AE |
935 | * Copy the names, and fill in each snapshot's id |
936 | * and size. | |
937 | * | |
99a41ebc | 938 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 939 | * ondisk buffer we're working with has |
f785cc1d AE |
940 | * snap_names_len bytes beyond the end of the |
941 | * snapshot id array, this memcpy() is safe. | |
942 | */ | |
bb23e37a AE |
943 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
944 | snaps = ondisk->snaps; | |
945 | for (i = 0; i < snap_count; i++) { | |
946 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
947 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
948 | } | |
602adf40 | 949 | } |
6a52325f | 950 | |
bb23e37a | 951 | /* We won't fail any more, fill in the header */ |
621901d6 | 952 | |
bb23e37a AE |
953 | if (first_time) { |
954 | header->object_prefix = object_prefix; | |
955 | header->obj_order = ondisk->options.order; | |
956 | header->crypt_type = ondisk->options.crypt_type; | |
957 | header->comp_type = ondisk->options.comp_type; | |
958 | /* The rest aren't used for format 1 images */ | |
959 | header->stripe_unit = 0; | |
960 | header->stripe_count = 0; | |
961 | header->features = 0; | |
602adf40 | 962 | } else { |
662518b1 AE |
963 | ceph_put_snap_context(header->snapc); |
964 | kfree(header->snap_names); | |
965 | kfree(header->snap_sizes); | |
602adf40 | 966 | } |
849b4260 | 967 | |
bb23e37a | 968 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 969 | |
f84344f3 | 970 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
971 | header->snapc = snapc; |
972 | header->snap_names = snap_names; | |
973 | header->snap_sizes = snap_sizes; | |
468521c1 | 974 | |
662518b1 | 975 | /* Make sure mapping size is consistent with header info */ |
602adf40 | 976 | |
662518b1 AE |
977 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time) |
978 | if (rbd_dev->mapping.size != header->image_size) | |
979 | rbd_dev->mapping.size = header->image_size; | |
980 | ||
602adf40 | 981 | return 0; |
bb23e37a AE |
982 | out_2big: |
983 | ret = -EIO; | |
6a52325f | 984 | out_err: |
bb23e37a AE |
985 | kfree(snap_sizes); |
986 | kfree(snap_names); | |
987 | ceph_put_snap_context(snapc); | |
988 | kfree(object_prefix); | |
ccece235 | 989 | |
bb23e37a | 990 | return ret; |
602adf40 YS |
991 | } |
992 | ||
9682fc6d AE |
993 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
994 | { | |
995 | const char *snap_name; | |
996 | ||
997 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
998 | ||
999 | /* Skip over names until we find the one we are looking for */ | |
1000 | ||
1001 | snap_name = rbd_dev->header.snap_names; | |
1002 | while (which--) | |
1003 | snap_name += strlen(snap_name) + 1; | |
1004 | ||
1005 | return kstrdup(snap_name, GFP_KERNEL); | |
1006 | } | |
1007 | ||
30d1cff8 AE |
1008 | /* |
1009 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1010 | * Note that result is for snapshots in *descending* order. | |
1011 | */ | |
1012 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1013 | { | |
1014 | u64 snap_id1 = *(u64 *)s1; | |
1015 | u64 snap_id2 = *(u64 *)s2; | |
1016 | ||
1017 | if (snap_id1 < snap_id2) | |
1018 | return 1; | |
1019 | return snap_id1 == snap_id2 ? 0 : -1; | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Search a snapshot context to see if the given snapshot id is | |
1024 | * present. | |
1025 | * | |
1026 | * Returns the position of the snapshot id in the array if it's found, | |
1027 | * or BAD_SNAP_INDEX otherwise. | |
1028 | * | |
1029 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1030 | * reverse order, highest snapshot id first. | |
1031 | */ | |
9682fc6d AE |
1032 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1033 | { | |
1034 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1035 | u64 *found; |
9682fc6d | 1036 | |
30d1cff8 AE |
1037 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1038 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1039 | |
30d1cff8 | 1040 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1041 | } |
1042 | ||
2ad3d716 AE |
1043 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1044 | u64 snap_id) | |
9e15b77d | 1045 | { |
54cac61f | 1046 | u32 which; |
da6a6b63 | 1047 | const char *snap_name; |
9e15b77d | 1048 | |
54cac61f AE |
1049 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1050 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1051 | return ERR_PTR(-ENOENT); |
54cac61f | 1052 | |
da6a6b63 JD |
1053 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1054 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1055 | } |
1056 | ||
1057 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1058 | { | |
9e15b77d AE |
1059 | if (snap_id == CEPH_NOSNAP) |
1060 | return RBD_SNAP_HEAD_NAME; | |
1061 | ||
54cac61f AE |
1062 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1063 | if (rbd_dev->image_format == 1) | |
1064 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1065 | |
54cac61f | 1066 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1067 | } |
1068 | ||
2ad3d716 AE |
1069 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1070 | u64 *snap_size) | |
602adf40 | 1071 | { |
2ad3d716 AE |
1072 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1073 | if (snap_id == CEPH_NOSNAP) { | |
1074 | *snap_size = rbd_dev->header.image_size; | |
1075 | } else if (rbd_dev->image_format == 1) { | |
1076 | u32 which; | |
602adf40 | 1077 | |
2ad3d716 AE |
1078 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1079 | if (which == BAD_SNAP_INDEX) | |
1080 | return -ENOENT; | |
e86924a8 | 1081 | |
2ad3d716 AE |
1082 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1083 | } else { | |
1084 | u64 size = 0; | |
1085 | int ret; | |
1086 | ||
1087 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1088 | if (ret) | |
1089 | return ret; | |
1090 | ||
1091 | *snap_size = size; | |
1092 | } | |
1093 | return 0; | |
602adf40 YS |
1094 | } |
1095 | ||
2ad3d716 AE |
1096 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1097 | u64 *snap_features) | |
602adf40 | 1098 | { |
2ad3d716 AE |
1099 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1100 | if (snap_id == CEPH_NOSNAP) { | |
1101 | *snap_features = rbd_dev->header.features; | |
1102 | } else if (rbd_dev->image_format == 1) { | |
1103 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1104 | } else { |
2ad3d716 AE |
1105 | u64 features = 0; |
1106 | int ret; | |
8b0241f8 | 1107 | |
2ad3d716 AE |
1108 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1109 | if (ret) | |
1110 | return ret; | |
1111 | ||
1112 | *snap_features = features; | |
1113 | } | |
1114 | return 0; | |
1115 | } | |
1116 | ||
1117 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1118 | { | |
8f4b7d98 | 1119 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1120 | u64 size = 0; |
1121 | u64 features = 0; | |
1122 | int ret; | |
1123 | ||
2ad3d716 AE |
1124 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1125 | if (ret) | |
1126 | return ret; | |
1127 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1128 | if (ret) | |
1129 | return ret; | |
1130 | ||
1131 | rbd_dev->mapping.size = size; | |
1132 | rbd_dev->mapping.features = features; | |
1133 | ||
8b0241f8 | 1134 | return 0; |
602adf40 YS |
1135 | } |
1136 | ||
d1cf5788 AE |
1137 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1138 | { | |
1139 | rbd_dev->mapping.size = 0; | |
1140 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1141 | } |
1142 | ||
98571b5a | 1143 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1144 | { |
65ccfe21 AE |
1145 | char *name; |
1146 | u64 segment; | |
1147 | int ret; | |
3a96d5cd | 1148 | char *name_format; |
602adf40 | 1149 | |
78c2a44a | 1150 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1151 | if (!name) |
1152 | return NULL; | |
1153 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1154 | name_format = "%s.%012llx"; |
1155 | if (rbd_dev->image_format == 2) | |
1156 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1157 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1158 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1159 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1160 | pr_err("error formatting segment name for #%llu (%d)\n", |
1161 | segment, ret); | |
1162 | kfree(name); | |
1163 | name = NULL; | |
1164 | } | |
602adf40 | 1165 | |
65ccfe21 AE |
1166 | return name; |
1167 | } | |
602adf40 | 1168 | |
78c2a44a AE |
1169 | static void rbd_segment_name_free(const char *name) |
1170 | { | |
1171 | /* The explicit cast here is needed to drop the const qualifier */ | |
1172 | ||
1173 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1174 | } | |
1175 | ||
65ccfe21 AE |
1176 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1177 | { | |
1178 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1179 | |
65ccfe21 AE |
1180 | return offset & (segment_size - 1); |
1181 | } | |
1182 | ||
1183 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1184 | u64 offset, u64 length) | |
1185 | { | |
1186 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1187 | ||
1188 | offset &= segment_size - 1; | |
1189 | ||
aafb230e | 1190 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1191 | if (offset + length > segment_size) |
1192 | length = segment_size - offset; | |
1193 | ||
1194 | return length; | |
602adf40 YS |
1195 | } |
1196 | ||
029bcbd8 JD |
1197 | /* |
1198 | * returns the size of an object in the image | |
1199 | */ | |
1200 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1201 | { | |
1202 | return 1 << header->obj_order; | |
1203 | } | |
1204 | ||
602adf40 YS |
1205 | /* |
1206 | * bio helpers | |
1207 | */ | |
1208 | ||
1209 | static void bio_chain_put(struct bio *chain) | |
1210 | { | |
1211 | struct bio *tmp; | |
1212 | ||
1213 | while (chain) { | |
1214 | tmp = chain; | |
1215 | chain = chain->bi_next; | |
1216 | bio_put(tmp); | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * zeros a bio chain, starting at specific offset | |
1222 | */ | |
1223 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1224 | { | |
7988613b KO |
1225 | struct bio_vec bv; |
1226 | struct bvec_iter iter; | |
602adf40 YS |
1227 | unsigned long flags; |
1228 | void *buf; | |
602adf40 YS |
1229 | int pos = 0; |
1230 | ||
1231 | while (chain) { | |
7988613b KO |
1232 | bio_for_each_segment(bv, chain, iter) { |
1233 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1234 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1235 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1236 | memset(buf + remainder, 0, |
7988613b KO |
1237 | bv.bv_len - remainder); |
1238 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1239 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1240 | } |
7988613b | 1241 | pos += bv.bv_len; |
602adf40 YS |
1242 | } |
1243 | ||
1244 | chain = chain->bi_next; | |
1245 | } | |
1246 | } | |
1247 | ||
b9434c5b AE |
1248 | /* |
1249 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1250 | * starting at the given byte offset from the start of the array and | |
1251 | * continuing up to the given end offset. The pages array is | |
1252 | * assumed to be big enough to hold all bytes up to the end. | |
1253 | */ | |
1254 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1255 | { | |
1256 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1257 | ||
1258 | rbd_assert(end > offset); | |
1259 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1260 | while (offset < end) { | |
1261 | size_t page_offset; | |
1262 | size_t length; | |
1263 | unsigned long flags; | |
1264 | void *kaddr; | |
1265 | ||
491205a8 GU |
1266 | page_offset = offset & ~PAGE_MASK; |
1267 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1268 | local_irq_save(flags); |
1269 | kaddr = kmap_atomic(*page); | |
1270 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1271 | flush_dcache_page(*page); |
b9434c5b AE |
1272 | kunmap_atomic(kaddr); |
1273 | local_irq_restore(flags); | |
1274 | ||
1275 | offset += length; | |
1276 | page++; | |
1277 | } | |
1278 | } | |
1279 | ||
602adf40 | 1280 | /* |
f7760dad AE |
1281 | * Clone a portion of a bio, starting at the given byte offset |
1282 | * and continuing for the number of bytes indicated. | |
602adf40 | 1283 | */ |
f7760dad AE |
1284 | static struct bio *bio_clone_range(struct bio *bio_src, |
1285 | unsigned int offset, | |
1286 | unsigned int len, | |
1287 | gfp_t gfpmask) | |
602adf40 | 1288 | { |
f7760dad AE |
1289 | struct bio *bio; |
1290 | ||
5341a627 | 1291 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1292 | if (!bio) |
1293 | return NULL; /* ENOMEM */ | |
602adf40 | 1294 | |
5341a627 | 1295 | bio_advance(bio, offset); |
4f024f37 | 1296 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1297 | |
1298 | return bio; | |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * Clone a portion of a bio chain, starting at the given byte offset | |
1303 | * into the first bio in the source chain and continuing for the | |
1304 | * number of bytes indicated. The result is another bio chain of | |
1305 | * exactly the given length, or a null pointer on error. | |
1306 | * | |
1307 | * The bio_src and offset parameters are both in-out. On entry they | |
1308 | * refer to the first source bio and the offset into that bio where | |
1309 | * the start of data to be cloned is located. | |
1310 | * | |
1311 | * On return, bio_src is updated to refer to the bio in the source | |
1312 | * chain that contains first un-cloned byte, and *offset will | |
1313 | * contain the offset of that byte within that bio. | |
1314 | */ | |
1315 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1316 | unsigned int *offset, | |
1317 | unsigned int len, | |
1318 | gfp_t gfpmask) | |
1319 | { | |
1320 | struct bio *bi = *bio_src; | |
1321 | unsigned int off = *offset; | |
1322 | struct bio *chain = NULL; | |
1323 | struct bio **end; | |
1324 | ||
1325 | /* Build up a chain of clone bios up to the limit */ | |
1326 | ||
4f024f37 | 1327 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1328 | return NULL; /* Nothing to clone */ |
602adf40 | 1329 | |
f7760dad AE |
1330 | end = &chain; |
1331 | while (len) { | |
1332 | unsigned int bi_size; | |
1333 | struct bio *bio; | |
1334 | ||
f5400b7a AE |
1335 | if (!bi) { |
1336 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1337 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1338 | } |
4f024f37 | 1339 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1340 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1341 | if (!bio) | |
1342 | goto out_err; /* ENOMEM */ | |
1343 | ||
1344 | *end = bio; | |
1345 | end = &bio->bi_next; | |
602adf40 | 1346 | |
f7760dad | 1347 | off += bi_size; |
4f024f37 | 1348 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1349 | bi = bi->bi_next; |
1350 | off = 0; | |
1351 | } | |
1352 | len -= bi_size; | |
1353 | } | |
1354 | *bio_src = bi; | |
1355 | *offset = off; | |
1356 | ||
1357 | return chain; | |
1358 | out_err: | |
1359 | bio_chain_put(chain); | |
602adf40 | 1360 | |
602adf40 YS |
1361 | return NULL; |
1362 | } | |
1363 | ||
926f9b3f AE |
1364 | /* |
1365 | * The default/initial value for all object request flags is 0. For | |
1366 | * each flag, once its value is set to 1 it is never reset to 0 | |
1367 | * again. | |
1368 | */ | |
57acbaa7 | 1369 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1370 | { |
57acbaa7 | 1371 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1372 | struct rbd_device *rbd_dev; |
1373 | ||
57acbaa7 AE |
1374 | rbd_dev = obj_request->img_request->rbd_dev; |
1375 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1376 | obj_request); |
1377 | } | |
1378 | } | |
1379 | ||
57acbaa7 | 1380 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1381 | { |
1382 | smp_mb(); | |
57acbaa7 | 1383 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1384 | } |
1385 | ||
57acbaa7 | 1386 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1387 | { |
57acbaa7 AE |
1388 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1389 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1390 | |
57acbaa7 AE |
1391 | if (obj_request_img_data_test(obj_request)) |
1392 | rbd_dev = obj_request->img_request->rbd_dev; | |
1393 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1394 | obj_request); |
1395 | } | |
1396 | } | |
1397 | ||
57acbaa7 | 1398 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1399 | { |
1400 | smp_mb(); | |
57acbaa7 | 1401 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1402 | } |
1403 | ||
5679c59f AE |
1404 | /* |
1405 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1406 | * flag. The latter is set based on the "exists" value provided. | |
1407 | * | |
1408 | * Note that for our purposes once an object exists it never goes | |
1409 | * away again. It's possible that the response from two existence | |
1410 | * checks are separated by the creation of the target object, and | |
1411 | * the first ("doesn't exist") response arrives *after* the second | |
1412 | * ("does exist"). In that case we ignore the second one. | |
1413 | */ | |
1414 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1415 | bool exists) | |
1416 | { | |
1417 | if (exists) | |
1418 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1419 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1420 | smp_mb(); | |
1421 | } | |
1422 | ||
1423 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1424 | { | |
1425 | smp_mb(); | |
1426 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1427 | } | |
1428 | ||
1429 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1430 | { | |
1431 | smp_mb(); | |
1432 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1433 | } | |
1434 | ||
bf0d5f50 AE |
1435 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1436 | { | |
37206ee5 AE |
1437 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1438 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1439 | kref_get(&obj_request->kref); |
1440 | } | |
1441 | ||
1442 | static void rbd_obj_request_destroy(struct kref *kref); | |
1443 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1444 | { | |
1445 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1446 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1447 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1448 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1449 | } | |
1450 | ||
0f2d5be7 AE |
1451 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1452 | { | |
1453 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1454 | atomic_read(&img_request->kref.refcount)); | |
1455 | kref_get(&img_request->kref); | |
1456 | } | |
1457 | ||
e93f3152 AE |
1458 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1459 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1460 | static void rbd_img_request_destroy(struct kref *kref); |
1461 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1462 | { | |
1463 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1464 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1465 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1466 | if (img_request_child_test(img_request)) |
1467 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1468 | else | |
1469 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1470 | } |
1471 | ||
1472 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1473 | struct rbd_obj_request *obj_request) | |
1474 | { | |
25dcf954 AE |
1475 | rbd_assert(obj_request->img_request == NULL); |
1476 | ||
b155e86c | 1477 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1478 | obj_request->img_request = img_request; |
25dcf954 | 1479 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1480 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1481 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1482 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1483 | img_request->obj_request_count++; |
1484 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1485 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1486 | obj_request->which); | |
bf0d5f50 AE |
1487 | } |
1488 | ||
1489 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1490 | struct rbd_obj_request *obj_request) | |
1491 | { | |
1492 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1493 | |
37206ee5 AE |
1494 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1495 | obj_request->which); | |
bf0d5f50 | 1496 | list_del(&obj_request->links); |
25dcf954 AE |
1497 | rbd_assert(img_request->obj_request_count > 0); |
1498 | img_request->obj_request_count--; | |
1499 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1500 | obj_request->which = BAD_WHICH; | |
6365d33a | 1501 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1502 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1503 | obj_request->img_request = NULL; |
25dcf954 | 1504 | obj_request->callback = NULL; |
bf0d5f50 AE |
1505 | rbd_obj_request_put(obj_request); |
1506 | } | |
1507 | ||
1508 | static bool obj_request_type_valid(enum obj_request_type type) | |
1509 | { | |
1510 | switch (type) { | |
9969ebc5 | 1511 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1512 | case OBJ_REQUEST_BIO: |
788e2df3 | 1513 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1514 | return true; |
1515 | default: | |
1516 | return false; | |
1517 | } | |
1518 | } | |
1519 | ||
bf0d5f50 AE |
1520 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1521 | struct rbd_obj_request *obj_request) | |
1522 | { | |
37206ee5 AE |
1523 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1524 | ||
bf0d5f50 AE |
1525 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1526 | } | |
1527 | ||
1528 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1529 | { | |
55f27e09 | 1530 | |
37206ee5 | 1531 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1532 | |
1533 | /* | |
1534 | * If no error occurred, compute the aggregate transfer | |
1535 | * count for the image request. We could instead use | |
1536 | * atomic64_cmpxchg() to update it as each object request | |
1537 | * completes; not clear which way is better off hand. | |
1538 | */ | |
1539 | if (!img_request->result) { | |
1540 | struct rbd_obj_request *obj_request; | |
1541 | u64 xferred = 0; | |
1542 | ||
1543 | for_each_obj_request(img_request, obj_request) | |
1544 | xferred += obj_request->xferred; | |
1545 | img_request->xferred = xferred; | |
1546 | } | |
1547 | ||
bf0d5f50 AE |
1548 | if (img_request->callback) |
1549 | img_request->callback(img_request); | |
1550 | else | |
1551 | rbd_img_request_put(img_request); | |
1552 | } | |
1553 | ||
788e2df3 AE |
1554 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1555 | ||
1556 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1557 | { | |
37206ee5 AE |
1558 | dout("%s: obj %p\n", __func__, obj_request); |
1559 | ||
788e2df3 AE |
1560 | return wait_for_completion_interruptible(&obj_request->completion); |
1561 | } | |
1562 | ||
0c425248 AE |
1563 | /* |
1564 | * The default/initial value for all image request flags is 0. Each | |
1565 | * is conditionally set to 1 at image request initialization time | |
1566 | * and currently never change thereafter. | |
1567 | */ | |
1568 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1569 | { | |
1570 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1571 | smp_mb(); | |
1572 | } | |
1573 | ||
1574 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1575 | { | |
1576 | smp_mb(); | |
1577 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1578 | } | |
1579 | ||
9849e986 AE |
1580 | static void img_request_child_set(struct rbd_img_request *img_request) |
1581 | { | |
1582 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1583 | smp_mb(); | |
1584 | } | |
1585 | ||
e93f3152 AE |
1586 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1587 | { | |
1588 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1589 | smp_mb(); | |
1590 | } | |
1591 | ||
9849e986 AE |
1592 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1593 | { | |
1594 | smp_mb(); | |
1595 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1596 | } | |
1597 | ||
d0b2e944 AE |
1598 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1599 | { | |
1600 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1601 | smp_mb(); | |
1602 | } | |
1603 | ||
a2acd00e AE |
1604 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1605 | { | |
1606 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1607 | smp_mb(); | |
1608 | } | |
1609 | ||
d0b2e944 AE |
1610 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1611 | { | |
1612 | smp_mb(); | |
1613 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1614 | } | |
1615 | ||
6e2a4505 AE |
1616 | static void |
1617 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1618 | { | |
b9434c5b AE |
1619 | u64 xferred = obj_request->xferred; |
1620 | u64 length = obj_request->length; | |
1621 | ||
6e2a4505 AE |
1622 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1623 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1624 | xferred, length); |
6e2a4505 | 1625 | /* |
17c1cc1d JD |
1626 | * ENOENT means a hole in the image. We zero-fill the entire |
1627 | * length of the request. A short read also implies zero-fill | |
1628 | * to the end of the request. An error requires the whole | |
1629 | * length of the request to be reported finished with an error | |
1630 | * to the block layer. In each case we update the xferred | |
1631 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1632 | */ |
b9434c5b | 1633 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1634 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1635 | if (obj_request->type == OBJ_REQUEST_BIO) |
1636 | zero_bio_chain(obj_request->bio_list, 0); | |
1637 | else | |
1638 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1639 | obj_request->result = 0; |
b9434c5b AE |
1640 | } else if (xferred < length && !obj_request->result) { |
1641 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1642 | zero_bio_chain(obj_request->bio_list, xferred); | |
1643 | else | |
1644 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1645 | } |
17c1cc1d | 1646 | obj_request->xferred = length; |
6e2a4505 AE |
1647 | obj_request_done_set(obj_request); |
1648 | } | |
1649 | ||
bf0d5f50 AE |
1650 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1651 | { | |
37206ee5 AE |
1652 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1653 | obj_request->callback); | |
bf0d5f50 AE |
1654 | if (obj_request->callback) |
1655 | obj_request->callback(obj_request); | |
788e2df3 AE |
1656 | else |
1657 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1658 | } |
1659 | ||
c47f9371 | 1660 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1661 | { |
1662 | dout("%s: obj %p\n", __func__, obj_request); | |
1663 | obj_request_done_set(obj_request); | |
1664 | } | |
1665 | ||
c47f9371 | 1666 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1667 | { |
57acbaa7 | 1668 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1669 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1670 | bool layered = false; |
1671 | ||
1672 | if (obj_request_img_data_test(obj_request)) { | |
1673 | img_request = obj_request->img_request; | |
1674 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1675 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1676 | } |
8b3e1a56 AE |
1677 | |
1678 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1679 | obj_request, img_request, obj_request->result, | |
1680 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1681 | if (layered && obj_request->result == -ENOENT && |
1682 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1683 | rbd_img_parent_read(obj_request); |
1684 | else if (img_request) | |
6e2a4505 AE |
1685 | rbd_img_obj_request_read_callback(obj_request); |
1686 | else | |
1687 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1688 | } |
1689 | ||
c47f9371 | 1690 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1691 | { |
1b83bef2 SW |
1692 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1693 | obj_request->result, obj_request->length); | |
1694 | /* | |
8b3e1a56 AE |
1695 | * There is no such thing as a successful short write. Set |
1696 | * it to our originally-requested length. | |
1b83bef2 SW |
1697 | */ |
1698 | obj_request->xferred = obj_request->length; | |
07741308 | 1699 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1700 | } |
1701 | ||
fbfab539 AE |
1702 | /* |
1703 | * For a simple stat call there's nothing to do. We'll do more if | |
1704 | * this is part of a write sequence for a layered image. | |
1705 | */ | |
c47f9371 | 1706 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1707 | { |
37206ee5 | 1708 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1709 | obj_request_done_set(obj_request); |
1710 | } | |
1711 | ||
bf0d5f50 AE |
1712 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1713 | struct ceph_msg *msg) | |
1714 | { | |
1715 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1716 | u16 opcode; |
1717 | ||
37206ee5 | 1718 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1719 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1720 | if (obj_request_img_data_test(obj_request)) { |
1721 | rbd_assert(obj_request->img_request); | |
1722 | rbd_assert(obj_request->which != BAD_WHICH); | |
1723 | } else { | |
1724 | rbd_assert(obj_request->which == BAD_WHICH); | |
1725 | } | |
bf0d5f50 | 1726 | |
1b83bef2 SW |
1727 | if (osd_req->r_result < 0) |
1728 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1729 | |
7cc69d42 | 1730 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1731 | |
c47f9371 AE |
1732 | /* |
1733 | * We support a 64-bit length, but ultimately it has to be | |
1734 | * passed to blk_end_request(), which takes an unsigned int. | |
1735 | */ | |
1b83bef2 | 1736 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1737 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1738 | |
79528734 | 1739 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1740 | switch (opcode) { |
1741 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1742 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1743 | break; |
0ccd5926 ID |
1744 | case CEPH_OSD_OP_SETALLOCHINT: |
1745 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE); | |
1746 | /* fall through */ | |
bf0d5f50 | 1747 | case CEPH_OSD_OP_WRITE: |
c47f9371 | 1748 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1749 | break; |
fbfab539 | 1750 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1751 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1752 | break; |
36be9a76 | 1753 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1754 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1755 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1756 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1757 | break; |
bf0d5f50 AE |
1758 | default: |
1759 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1760 | obj_request->object_name, (unsigned short) opcode); | |
1761 | break; | |
1762 | } | |
1763 | ||
07741308 | 1764 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1765 | rbd_obj_request_complete(obj_request); |
1766 | } | |
1767 | ||
9d4df01f | 1768 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1769 | { |
1770 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1771 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1772 | u64 snap_id; |
430c28c3 | 1773 | |
8c042b0d | 1774 | rbd_assert(osd_req != NULL); |
430c28c3 | 1775 | |
9d4df01f | 1776 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1777 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1778 | NULL, snap_id, NULL); |
1779 | } | |
1780 | ||
1781 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1782 | { | |
1783 | struct rbd_img_request *img_request = obj_request->img_request; | |
1784 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1785 | struct ceph_snap_context *snapc; | |
1786 | struct timespec mtime = CURRENT_TIME; | |
1787 | ||
1788 | rbd_assert(osd_req != NULL); | |
1789 | ||
1790 | snapc = img_request ? img_request->snapc : NULL; | |
1791 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1792 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1793 | } |
1794 | ||
0ccd5926 ID |
1795 | /* |
1796 | * Create an osd request. A read request has one osd op (read). | |
1797 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1798 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1799 | * technically osd watch is a write request, hence this distinction.) | |
1800 | */ | |
bf0d5f50 AE |
1801 | static struct ceph_osd_request *rbd_osd_req_create( |
1802 | struct rbd_device *rbd_dev, | |
1803 | bool write_request, | |
deb236b3 | 1804 | unsigned int num_ops, |
430c28c3 | 1805 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1806 | { |
bf0d5f50 AE |
1807 | struct ceph_snap_context *snapc = NULL; |
1808 | struct ceph_osd_client *osdc; | |
1809 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1810 | |
6365d33a AE |
1811 | if (obj_request_img_data_test(obj_request)) { |
1812 | struct rbd_img_request *img_request = obj_request->img_request; | |
1813 | ||
0c425248 AE |
1814 | rbd_assert(write_request == |
1815 | img_request_write_test(img_request)); | |
1816 | if (write_request) | |
bf0d5f50 | 1817 | snapc = img_request->snapc; |
bf0d5f50 AE |
1818 | } |
1819 | ||
0ccd5926 | 1820 | rbd_assert(num_ops == 1 || (write_request && num_ops == 2)); |
deb236b3 ID |
1821 | |
1822 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1823 | |
1824 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1825 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1826 | GFP_ATOMIC); | |
bf0d5f50 AE |
1827 | if (!osd_req) |
1828 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1829 | |
430c28c3 | 1830 | if (write_request) |
bf0d5f50 | 1831 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1832 | else |
bf0d5f50 | 1833 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1834 | |
1835 | osd_req->r_callback = rbd_osd_req_callback; | |
1836 | osd_req->r_priv = obj_request; | |
1837 | ||
3c972c95 ID |
1838 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1839 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1840 | |
bf0d5f50 AE |
1841 | return osd_req; |
1842 | } | |
1843 | ||
0eefd470 AE |
1844 | /* |
1845 | * Create a copyup osd request based on the information in the | |
0ccd5926 ID |
1846 | * object request supplied. A copyup request has three osd ops, |
1847 | * a copyup method call, a hint op, and a write op. | |
0eefd470 AE |
1848 | */ |
1849 | static struct ceph_osd_request * | |
1850 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1851 | { | |
1852 | struct rbd_img_request *img_request; | |
1853 | struct ceph_snap_context *snapc; | |
1854 | struct rbd_device *rbd_dev; | |
1855 | struct ceph_osd_client *osdc; | |
1856 | struct ceph_osd_request *osd_req; | |
1857 | ||
1858 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1859 | img_request = obj_request->img_request; | |
1860 | rbd_assert(img_request); | |
1861 | rbd_assert(img_request_write_test(img_request)); | |
1862 | ||
0ccd5926 | 1863 | /* Allocate and initialize the request, for the three ops */ |
0eefd470 AE |
1864 | |
1865 | snapc = img_request->snapc; | |
1866 | rbd_dev = img_request->rbd_dev; | |
1867 | osdc = &rbd_dev->rbd_client->client->osdc; | |
0ccd5926 | 1868 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 3, false, GFP_ATOMIC); |
0eefd470 AE |
1869 | if (!osd_req) |
1870 | return NULL; /* ENOMEM */ | |
1871 | ||
1872 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1873 | osd_req->r_callback = rbd_osd_req_callback; | |
1874 | osd_req->r_priv = obj_request; | |
1875 | ||
3c972c95 ID |
1876 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1877 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 1878 | |
0eefd470 AE |
1879 | return osd_req; |
1880 | } | |
1881 | ||
1882 | ||
bf0d5f50 AE |
1883 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1884 | { | |
1885 | ceph_osdc_put_request(osd_req); | |
1886 | } | |
1887 | ||
1888 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1889 | ||
1890 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1891 | u64 offset, u64 length, | |
1892 | enum obj_request_type type) | |
1893 | { | |
1894 | struct rbd_obj_request *obj_request; | |
1895 | size_t size; | |
1896 | char *name; | |
1897 | ||
1898 | rbd_assert(obj_request_type_valid(type)); | |
1899 | ||
1900 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1901 | name = kmalloc(size, GFP_KERNEL); |
1902 | if (!name) | |
bf0d5f50 AE |
1903 | return NULL; |
1904 | ||
868311b1 | 1905 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1906 | if (!obj_request) { |
1907 | kfree(name); | |
1908 | return NULL; | |
1909 | } | |
1910 | ||
bf0d5f50 AE |
1911 | obj_request->object_name = memcpy(name, object_name, size); |
1912 | obj_request->offset = offset; | |
1913 | obj_request->length = length; | |
926f9b3f | 1914 | obj_request->flags = 0; |
bf0d5f50 AE |
1915 | obj_request->which = BAD_WHICH; |
1916 | obj_request->type = type; | |
1917 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1918 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1919 | kref_init(&obj_request->kref); |
1920 | ||
37206ee5 AE |
1921 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1922 | offset, length, (int)type, obj_request); | |
1923 | ||
bf0d5f50 AE |
1924 | return obj_request; |
1925 | } | |
1926 | ||
1927 | static void rbd_obj_request_destroy(struct kref *kref) | |
1928 | { | |
1929 | struct rbd_obj_request *obj_request; | |
1930 | ||
1931 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1932 | ||
37206ee5 AE |
1933 | dout("%s: obj %p\n", __func__, obj_request); |
1934 | ||
bf0d5f50 AE |
1935 | rbd_assert(obj_request->img_request == NULL); |
1936 | rbd_assert(obj_request->which == BAD_WHICH); | |
1937 | ||
1938 | if (obj_request->osd_req) | |
1939 | rbd_osd_req_destroy(obj_request->osd_req); | |
1940 | ||
1941 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1942 | switch (obj_request->type) { | |
9969ebc5 AE |
1943 | case OBJ_REQUEST_NODATA: |
1944 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1945 | case OBJ_REQUEST_BIO: |
1946 | if (obj_request->bio_list) | |
1947 | bio_chain_put(obj_request->bio_list); | |
1948 | break; | |
788e2df3 AE |
1949 | case OBJ_REQUEST_PAGES: |
1950 | if (obj_request->pages) | |
1951 | ceph_release_page_vector(obj_request->pages, | |
1952 | obj_request->page_count); | |
1953 | break; | |
bf0d5f50 AE |
1954 | } |
1955 | ||
f907ad55 | 1956 | kfree(obj_request->object_name); |
868311b1 AE |
1957 | obj_request->object_name = NULL; |
1958 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
1959 | } |
1960 | ||
fb65d228 AE |
1961 | /* It's OK to call this for a device with no parent */ |
1962 | ||
1963 | static void rbd_spec_put(struct rbd_spec *spec); | |
1964 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
1965 | { | |
1966 | rbd_dev_remove_parent(rbd_dev); | |
1967 | rbd_spec_put(rbd_dev->parent_spec); | |
1968 | rbd_dev->parent_spec = NULL; | |
1969 | rbd_dev->parent_overlap = 0; | |
1970 | } | |
1971 | ||
a2acd00e AE |
1972 | /* |
1973 | * Parent image reference counting is used to determine when an | |
1974 | * image's parent fields can be safely torn down--after there are no | |
1975 | * more in-flight requests to the parent image. When the last | |
1976 | * reference is dropped, cleaning them up is safe. | |
1977 | */ | |
1978 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
1979 | { | |
1980 | int counter; | |
1981 | ||
1982 | if (!rbd_dev->parent_spec) | |
1983 | return; | |
1984 | ||
1985 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
1986 | if (counter > 0) | |
1987 | return; | |
1988 | ||
1989 | /* Last reference; clean up parent data structures */ | |
1990 | ||
1991 | if (!counter) | |
1992 | rbd_dev_unparent(rbd_dev); | |
1993 | else | |
1994 | rbd_warn(rbd_dev, "parent reference underflow\n"); | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * If an image has a non-zero parent overlap, get a reference to its | |
1999 | * parent. | |
2000 | * | |
392a9dad AE |
2001 | * We must get the reference before checking for the overlap to |
2002 | * coordinate properly with zeroing the parent overlap in | |
2003 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
2004 | * drop it again if there is no overlap. | |
2005 | * | |
a2acd00e AE |
2006 | * Returns true if the rbd device has a parent with a non-zero |
2007 | * overlap and a reference for it was successfully taken, or | |
2008 | * false otherwise. | |
2009 | */ | |
2010 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2011 | { | |
2012 | int counter; | |
2013 | ||
2014 | if (!rbd_dev->parent_spec) | |
2015 | return false; | |
2016 | ||
2017 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2018 | if (counter > 0 && rbd_dev->parent_overlap) | |
2019 | return true; | |
2020 | ||
2021 | /* Image was flattened, but parent is not yet torn down */ | |
2022 | ||
2023 | if (counter < 0) | |
2024 | rbd_warn(rbd_dev, "parent reference overflow\n"); | |
2025 | ||
2026 | return false; | |
2027 | } | |
2028 | ||
bf0d5f50 AE |
2029 | /* |
2030 | * Caller is responsible for filling in the list of object requests | |
2031 | * that comprises the image request, and the Linux request pointer | |
2032 | * (if there is one). | |
2033 | */ | |
cc344fa1 AE |
2034 | static struct rbd_img_request *rbd_img_request_create( |
2035 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2036 | u64 offset, u64 length, |
e93f3152 | 2037 | bool write_request) |
bf0d5f50 AE |
2038 | { |
2039 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2040 | |
1c2a9dfe | 2041 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC); |
bf0d5f50 AE |
2042 | if (!img_request) |
2043 | return NULL; | |
2044 | ||
2045 | if (write_request) { | |
2046 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 2047 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 2048 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
2049 | } |
2050 | ||
2051 | img_request->rq = NULL; | |
2052 | img_request->rbd_dev = rbd_dev; | |
2053 | img_request->offset = offset; | |
2054 | img_request->length = length; | |
0c425248 AE |
2055 | img_request->flags = 0; |
2056 | if (write_request) { | |
2057 | img_request_write_set(img_request); | |
468521c1 | 2058 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 2059 | } else { |
bf0d5f50 | 2060 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2061 | } |
a2acd00e | 2062 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2063 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2064 | spin_lock_init(&img_request->completion_lock); |
2065 | img_request->next_completion = 0; | |
2066 | img_request->callback = NULL; | |
a5a337d4 | 2067 | img_request->result = 0; |
bf0d5f50 AE |
2068 | img_request->obj_request_count = 0; |
2069 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2070 | kref_init(&img_request->kref); | |
2071 | ||
37206ee5 AE |
2072 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
2073 | write_request ? "write" : "read", offset, length, | |
2074 | img_request); | |
2075 | ||
bf0d5f50 AE |
2076 | return img_request; |
2077 | } | |
2078 | ||
2079 | static void rbd_img_request_destroy(struct kref *kref) | |
2080 | { | |
2081 | struct rbd_img_request *img_request; | |
2082 | struct rbd_obj_request *obj_request; | |
2083 | struct rbd_obj_request *next_obj_request; | |
2084 | ||
2085 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2086 | ||
37206ee5 AE |
2087 | dout("%s: img %p\n", __func__, img_request); |
2088 | ||
bf0d5f50 AE |
2089 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2090 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2091 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2092 | |
a2acd00e AE |
2093 | if (img_request_layered_test(img_request)) { |
2094 | img_request_layered_clear(img_request); | |
2095 | rbd_dev_parent_put(img_request->rbd_dev); | |
2096 | } | |
2097 | ||
0c425248 | 2098 | if (img_request_write_test(img_request)) |
812164f8 | 2099 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2100 | |
1c2a9dfe | 2101 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2102 | } |
2103 | ||
e93f3152 AE |
2104 | static struct rbd_img_request *rbd_parent_request_create( |
2105 | struct rbd_obj_request *obj_request, | |
2106 | u64 img_offset, u64 length) | |
2107 | { | |
2108 | struct rbd_img_request *parent_request; | |
2109 | struct rbd_device *rbd_dev; | |
2110 | ||
2111 | rbd_assert(obj_request->img_request); | |
2112 | rbd_dev = obj_request->img_request->rbd_dev; | |
2113 | ||
2114 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2115 | img_offset, length, false); | |
2116 | if (!parent_request) | |
2117 | return NULL; | |
2118 | ||
2119 | img_request_child_set(parent_request); | |
2120 | rbd_obj_request_get(obj_request); | |
2121 | parent_request->obj_request = obj_request; | |
2122 | ||
2123 | return parent_request; | |
2124 | } | |
2125 | ||
2126 | static void rbd_parent_request_destroy(struct kref *kref) | |
2127 | { | |
2128 | struct rbd_img_request *parent_request; | |
2129 | struct rbd_obj_request *orig_request; | |
2130 | ||
2131 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2132 | orig_request = parent_request->obj_request; | |
2133 | ||
2134 | parent_request->obj_request = NULL; | |
2135 | rbd_obj_request_put(orig_request); | |
2136 | img_request_child_clear(parent_request); | |
2137 | ||
2138 | rbd_img_request_destroy(kref); | |
2139 | } | |
2140 | ||
1217857f AE |
2141 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2142 | { | |
6365d33a | 2143 | struct rbd_img_request *img_request; |
1217857f AE |
2144 | unsigned int xferred; |
2145 | int result; | |
8b3e1a56 | 2146 | bool more; |
1217857f | 2147 | |
6365d33a AE |
2148 | rbd_assert(obj_request_img_data_test(obj_request)); |
2149 | img_request = obj_request->img_request; | |
2150 | ||
1217857f AE |
2151 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2152 | xferred = (unsigned int)obj_request->xferred; | |
2153 | result = obj_request->result; | |
2154 | if (result) { | |
2155 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2156 | ||
2157 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
2158 | img_request_write_test(img_request) ? "write" : "read", | |
2159 | obj_request->length, obj_request->img_offset, | |
2160 | obj_request->offset); | |
2161 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
2162 | result, xferred); | |
2163 | if (!img_request->result) | |
2164 | img_request->result = result; | |
2165 | } | |
2166 | ||
f1a4739f AE |
2167 | /* Image object requests don't own their page array */ |
2168 | ||
2169 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2170 | obj_request->pages = NULL; | |
2171 | obj_request->page_count = 0; | |
2172 | } | |
2173 | ||
8b3e1a56 AE |
2174 | if (img_request_child_test(img_request)) { |
2175 | rbd_assert(img_request->obj_request != NULL); | |
2176 | more = obj_request->which < img_request->obj_request_count - 1; | |
2177 | } else { | |
2178 | rbd_assert(img_request->rq != NULL); | |
2179 | more = blk_end_request(img_request->rq, result, xferred); | |
2180 | } | |
2181 | ||
2182 | return more; | |
1217857f AE |
2183 | } |
2184 | ||
2169238d AE |
2185 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2186 | { | |
2187 | struct rbd_img_request *img_request; | |
2188 | u32 which = obj_request->which; | |
2189 | bool more = true; | |
2190 | ||
6365d33a | 2191 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2192 | img_request = obj_request->img_request; |
2193 | ||
2194 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2195 | rbd_assert(img_request != NULL); | |
2169238d AE |
2196 | rbd_assert(img_request->obj_request_count > 0); |
2197 | rbd_assert(which != BAD_WHICH); | |
2198 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2199 | |
2200 | spin_lock_irq(&img_request->completion_lock); | |
2201 | if (which != img_request->next_completion) | |
2202 | goto out; | |
2203 | ||
2204 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2205 | rbd_assert(more); |
2206 | rbd_assert(which < img_request->obj_request_count); | |
2207 | ||
2208 | if (!obj_request_done_test(obj_request)) | |
2209 | break; | |
1217857f | 2210 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2211 | which++; |
2212 | } | |
2213 | ||
2214 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2215 | img_request->next_completion = which; | |
2216 | out: | |
2217 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2218 | rbd_img_request_put(img_request); |
2169238d AE |
2219 | |
2220 | if (!more) | |
2221 | rbd_img_request_complete(img_request); | |
2222 | } | |
2223 | ||
f1a4739f AE |
2224 | /* |
2225 | * Split up an image request into one or more object requests, each | |
2226 | * to a different object. The "type" parameter indicates whether | |
2227 | * "data_desc" is the pointer to the head of a list of bio | |
2228 | * structures, or the base of a page array. In either case this | |
2229 | * function assumes data_desc describes memory sufficient to hold | |
2230 | * all data described by the image request. | |
2231 | */ | |
2232 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2233 | enum obj_request_type type, | |
2234 | void *data_desc) | |
bf0d5f50 AE |
2235 | { |
2236 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2237 | struct rbd_obj_request *obj_request = NULL; | |
2238 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 2239 | bool write_request = img_request_write_test(img_request); |
a158073c | 2240 | struct bio *bio_list = NULL; |
f1a4739f | 2241 | unsigned int bio_offset = 0; |
a158073c | 2242 | struct page **pages = NULL; |
7da22d29 | 2243 | u64 img_offset; |
bf0d5f50 AE |
2244 | u64 resid; |
2245 | u16 opcode; | |
2246 | ||
f1a4739f AE |
2247 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2248 | (int)type, data_desc); | |
37206ee5 | 2249 | |
430c28c3 | 2250 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 2251 | img_offset = img_request->offset; |
bf0d5f50 | 2252 | resid = img_request->length; |
4dda41d3 | 2253 | rbd_assert(resid > 0); |
f1a4739f AE |
2254 | |
2255 | if (type == OBJ_REQUEST_BIO) { | |
2256 | bio_list = data_desc; | |
4f024f37 KO |
2257 | rbd_assert(img_offset == |
2258 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
f1a4739f AE |
2259 | } else { |
2260 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
2261 | pages = data_desc; | |
2262 | } | |
2263 | ||
bf0d5f50 | 2264 | while (resid) { |
2fa12320 | 2265 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2266 | const char *object_name; |
bf0d5f50 AE |
2267 | u64 offset; |
2268 | u64 length; | |
0ccd5926 | 2269 | unsigned int which = 0; |
bf0d5f50 | 2270 | |
7da22d29 | 2271 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2272 | if (!object_name) |
2273 | goto out_unwind; | |
7da22d29 AE |
2274 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2275 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2276 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2277 | offset, length, type); |
78c2a44a AE |
2278 | /* object request has its own copy of the object name */ |
2279 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2280 | if (!obj_request) |
2281 | goto out_unwind; | |
62054da6 | 2282 | |
03507db6 JD |
2283 | /* |
2284 | * set obj_request->img_request before creating the | |
2285 | * osd_request so that it gets the right snapc | |
2286 | */ | |
2287 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2288 | |
f1a4739f AE |
2289 | if (type == OBJ_REQUEST_BIO) { |
2290 | unsigned int clone_size; | |
2291 | ||
2292 | rbd_assert(length <= (u64)UINT_MAX); | |
2293 | clone_size = (unsigned int)length; | |
2294 | obj_request->bio_list = | |
2295 | bio_chain_clone_range(&bio_list, | |
2296 | &bio_offset, | |
2297 | clone_size, | |
2298 | GFP_ATOMIC); | |
2299 | if (!obj_request->bio_list) | |
62054da6 | 2300 | goto out_unwind; |
f1a4739f AE |
2301 | } else { |
2302 | unsigned int page_count; | |
2303 | ||
2304 | obj_request->pages = pages; | |
2305 | page_count = (u32)calc_pages_for(offset, length); | |
2306 | obj_request->page_count = page_count; | |
2307 | if ((offset + length) & ~PAGE_MASK) | |
2308 | page_count--; /* more on last page */ | |
2309 | pages += page_count; | |
2310 | } | |
bf0d5f50 | 2311 | |
0ccd5926 ID |
2312 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
2313 | (write_request ? 2 : 1), | |
deb236b3 | 2314 | obj_request); |
2fa12320 | 2315 | if (!osd_req) |
62054da6 | 2316 | goto out_unwind; |
2fa12320 | 2317 | obj_request->osd_req = osd_req; |
2169238d | 2318 | obj_request->callback = rbd_img_obj_callback; |
0f2d5be7 | 2319 | rbd_img_request_get(img_request); |
430c28c3 | 2320 | |
0ccd5926 ID |
2321 | if (write_request) { |
2322 | osd_req_op_alloc_hint_init(osd_req, which, | |
2323 | rbd_obj_bytes(&rbd_dev->header), | |
2324 | rbd_obj_bytes(&rbd_dev->header)); | |
2325 | which++; | |
2326 | } | |
2327 | ||
2328 | osd_req_op_extent_init(osd_req, which, opcode, offset, length, | |
2329 | 0, 0); | |
f1a4739f | 2330 | if (type == OBJ_REQUEST_BIO) |
0ccd5926 | 2331 | osd_req_op_extent_osd_data_bio(osd_req, which, |
f1a4739f AE |
2332 | obj_request->bio_list, length); |
2333 | else | |
0ccd5926 | 2334 | osd_req_op_extent_osd_data_pages(osd_req, which, |
f1a4739f AE |
2335 | obj_request->pages, length, |
2336 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
2337 | |
2338 | if (write_request) | |
2339 | rbd_osd_req_format_write(obj_request); | |
2340 | else | |
2341 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2342 | |
7da22d29 | 2343 | obj_request->img_offset = img_offset; |
bf0d5f50 | 2344 | |
7da22d29 | 2345 | img_offset += length; |
bf0d5f50 AE |
2346 | resid -= length; |
2347 | } | |
2348 | ||
2349 | return 0; | |
2350 | ||
bf0d5f50 AE |
2351 | out_unwind: |
2352 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2353 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2354 | |
2355 | return -ENOMEM; | |
2356 | } | |
2357 | ||
0eefd470 AE |
2358 | static void |
2359 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2360 | { | |
2361 | struct rbd_img_request *img_request; | |
2362 | struct rbd_device *rbd_dev; | |
ebda6408 | 2363 | struct page **pages; |
0eefd470 AE |
2364 | u32 page_count; |
2365 | ||
2366 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2367 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2368 | img_request = obj_request->img_request; | |
2369 | rbd_assert(img_request); | |
2370 | ||
2371 | rbd_dev = img_request->rbd_dev; | |
2372 | rbd_assert(rbd_dev); | |
0eefd470 | 2373 | |
ebda6408 AE |
2374 | pages = obj_request->copyup_pages; |
2375 | rbd_assert(pages != NULL); | |
0eefd470 | 2376 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2377 | page_count = obj_request->copyup_page_count; |
2378 | rbd_assert(page_count); | |
2379 | obj_request->copyup_page_count = 0; | |
2380 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2381 | |
2382 | /* | |
2383 | * We want the transfer count to reflect the size of the | |
2384 | * original write request. There is no such thing as a | |
2385 | * successful short write, so if the request was successful | |
2386 | * we can just set it to the originally-requested length. | |
2387 | */ | |
2388 | if (!obj_request->result) | |
2389 | obj_request->xferred = obj_request->length; | |
2390 | ||
2391 | /* Finish up with the normal image object callback */ | |
2392 | ||
2393 | rbd_img_obj_callback(obj_request); | |
2394 | } | |
2395 | ||
3d7efd18 AE |
2396 | static void |
2397 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2398 | { | |
2399 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2400 | struct ceph_osd_request *osd_req; |
2401 | struct ceph_osd_client *osdc; | |
2402 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2403 | struct page **pages; |
ebda6408 | 2404 | u32 page_count; |
bbea1c1a | 2405 | int img_result; |
ebda6408 | 2406 | u64 parent_length; |
b91f09f1 AE |
2407 | u64 offset; |
2408 | u64 length; | |
3d7efd18 AE |
2409 | |
2410 | rbd_assert(img_request_child_test(img_request)); | |
2411 | ||
2412 | /* First get what we need from the image request */ | |
2413 | ||
2414 | pages = img_request->copyup_pages; | |
2415 | rbd_assert(pages != NULL); | |
2416 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2417 | page_count = img_request->copyup_page_count; |
2418 | rbd_assert(page_count); | |
2419 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2420 | |
2421 | orig_request = img_request->obj_request; | |
2422 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2423 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2424 | img_result = img_request->result; |
ebda6408 AE |
2425 | parent_length = img_request->length; |
2426 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2427 | rbd_img_request_put(img_request); |
3d7efd18 | 2428 | |
91c6febb AE |
2429 | rbd_assert(orig_request->img_request); |
2430 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2431 | rbd_assert(rbd_dev); |
0eefd470 | 2432 | |
bbea1c1a AE |
2433 | /* |
2434 | * If the overlap has become 0 (most likely because the | |
2435 | * image has been flattened) we need to free the pages | |
2436 | * and re-submit the original write request. | |
2437 | */ | |
2438 | if (!rbd_dev->parent_overlap) { | |
2439 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2440 | |
bbea1c1a AE |
2441 | ceph_release_page_vector(pages, page_count); |
2442 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2443 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2444 | if (!img_result) | |
2445 | return; | |
2446 | } | |
0eefd470 | 2447 | |
bbea1c1a | 2448 | if (img_result) |
0eefd470 | 2449 | goto out_err; |
0eefd470 | 2450 | |
8785b1d4 AE |
2451 | /* |
2452 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2453 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2454 | * request. Allocate the new copyup osd request for the |
2455 | * original request, and release the old one. | |
2456 | */ | |
bbea1c1a | 2457 | img_result = -ENOMEM; |
0eefd470 AE |
2458 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2459 | if (!osd_req) | |
2460 | goto out_err; | |
8785b1d4 | 2461 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2462 | orig_request->osd_req = osd_req; |
2463 | orig_request->copyup_pages = pages; | |
ebda6408 | 2464 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2465 | |
0eefd470 | 2466 | /* Initialize the copyup op */ |
3d7efd18 | 2467 | |
0eefd470 | 2468 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2469 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2470 | false, false); |
3d7efd18 | 2471 | |
0ccd5926 ID |
2472 | /* Then the hint op */ |
2473 | ||
2474 | osd_req_op_alloc_hint_init(osd_req, 1, rbd_obj_bytes(&rbd_dev->header), | |
2475 | rbd_obj_bytes(&rbd_dev->header)); | |
2476 | ||
2477 | /* And the original write request op */ | |
0eefd470 | 2478 | |
b91f09f1 AE |
2479 | offset = orig_request->offset; |
2480 | length = orig_request->length; | |
0ccd5926 | 2481 | osd_req_op_extent_init(osd_req, 2, CEPH_OSD_OP_WRITE, |
b91f09f1 AE |
2482 | offset, length, 0, 0); |
2483 | if (orig_request->type == OBJ_REQUEST_BIO) | |
0ccd5926 | 2484 | osd_req_op_extent_osd_data_bio(osd_req, 2, |
b91f09f1 AE |
2485 | orig_request->bio_list, length); |
2486 | else | |
0ccd5926 | 2487 | osd_req_op_extent_osd_data_pages(osd_req, 2, |
b91f09f1 AE |
2488 | orig_request->pages, length, |
2489 | offset & ~PAGE_MASK, false, false); | |
0eefd470 AE |
2490 | |
2491 | rbd_osd_req_format_write(orig_request); | |
2492 | ||
2493 | /* All set, send it off. */ | |
2494 | ||
2495 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2496 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2497 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2498 | if (!img_result) | |
0eefd470 AE |
2499 | return; |
2500 | out_err: | |
2501 | /* Record the error code and complete the request */ | |
2502 | ||
bbea1c1a | 2503 | orig_request->result = img_result; |
0eefd470 AE |
2504 | orig_request->xferred = 0; |
2505 | obj_request_done_set(orig_request); | |
2506 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2507 | } |
2508 | ||
2509 | /* | |
2510 | * Read from the parent image the range of data that covers the | |
2511 | * entire target of the given object request. This is used for | |
2512 | * satisfying a layered image write request when the target of an | |
2513 | * object request from the image request does not exist. | |
2514 | * | |
2515 | * A page array big enough to hold the returned data is allocated | |
2516 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2517 | * When the read completes, this page array will be transferred to | |
2518 | * the original object request for the copyup operation. | |
2519 | * | |
2520 | * If an error occurs, record it as the result of the original | |
2521 | * object request and mark it done so it gets completed. | |
2522 | */ | |
2523 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2524 | { | |
2525 | struct rbd_img_request *img_request = NULL; | |
2526 | struct rbd_img_request *parent_request = NULL; | |
2527 | struct rbd_device *rbd_dev; | |
2528 | u64 img_offset; | |
2529 | u64 length; | |
2530 | struct page **pages = NULL; | |
2531 | u32 page_count; | |
2532 | int result; | |
2533 | ||
2534 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2535 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2536 | |
2537 | img_request = obj_request->img_request; | |
2538 | rbd_assert(img_request != NULL); | |
2539 | rbd_dev = img_request->rbd_dev; | |
2540 | rbd_assert(rbd_dev->parent != NULL); | |
2541 | ||
2542 | /* | |
2543 | * Determine the byte range covered by the object in the | |
2544 | * child image to which the original request was to be sent. | |
2545 | */ | |
2546 | img_offset = obj_request->img_offset - obj_request->offset; | |
2547 | length = (u64)1 << rbd_dev->header.obj_order; | |
2548 | ||
a9e8ba2c AE |
2549 | /* |
2550 | * There is no defined parent data beyond the parent | |
2551 | * overlap, so limit what we read at that boundary if | |
2552 | * necessary. | |
2553 | */ | |
2554 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2555 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2556 | length = rbd_dev->parent_overlap - img_offset; | |
2557 | } | |
2558 | ||
3d7efd18 AE |
2559 | /* |
2560 | * Allocate a page array big enough to receive the data read | |
2561 | * from the parent. | |
2562 | */ | |
2563 | page_count = (u32)calc_pages_for(0, length); | |
2564 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2565 | if (IS_ERR(pages)) { | |
2566 | result = PTR_ERR(pages); | |
2567 | pages = NULL; | |
2568 | goto out_err; | |
2569 | } | |
2570 | ||
2571 | result = -ENOMEM; | |
e93f3152 AE |
2572 | parent_request = rbd_parent_request_create(obj_request, |
2573 | img_offset, length); | |
3d7efd18 AE |
2574 | if (!parent_request) |
2575 | goto out_err; | |
3d7efd18 AE |
2576 | |
2577 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2578 | if (result) | |
2579 | goto out_err; | |
2580 | parent_request->copyup_pages = pages; | |
ebda6408 | 2581 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2582 | |
2583 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2584 | result = rbd_img_request_submit(parent_request); | |
2585 | if (!result) | |
2586 | return 0; | |
2587 | ||
2588 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2589 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2590 | parent_request->obj_request = NULL; |
2591 | rbd_obj_request_put(obj_request); | |
2592 | out_err: | |
2593 | if (pages) | |
2594 | ceph_release_page_vector(pages, page_count); | |
2595 | if (parent_request) | |
2596 | rbd_img_request_put(parent_request); | |
2597 | obj_request->result = result; | |
2598 | obj_request->xferred = 0; | |
2599 | obj_request_done_set(obj_request); | |
2600 | ||
2601 | return result; | |
2602 | } | |
2603 | ||
c5b5ef6c AE |
2604 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2605 | { | |
c5b5ef6c | 2606 | struct rbd_obj_request *orig_request; |
638f5abe | 2607 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2608 | int result; |
2609 | ||
2610 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2611 | ||
2612 | /* | |
2613 | * All we need from the object request is the original | |
2614 | * request and the result of the STAT op. Grab those, then | |
2615 | * we're done with the request. | |
2616 | */ | |
2617 | orig_request = obj_request->obj_request; | |
2618 | obj_request->obj_request = NULL; | |
912c317d | 2619 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2620 | rbd_assert(orig_request); |
2621 | rbd_assert(orig_request->img_request); | |
2622 | ||
2623 | result = obj_request->result; | |
2624 | obj_request->result = 0; | |
2625 | ||
2626 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2627 | obj_request, orig_request, result, | |
2628 | obj_request->xferred, obj_request->length); | |
2629 | rbd_obj_request_put(obj_request); | |
2630 | ||
638f5abe AE |
2631 | /* |
2632 | * If the overlap has become 0 (most likely because the | |
2633 | * image has been flattened) we need to free the pages | |
2634 | * and re-submit the original write request. | |
2635 | */ | |
2636 | rbd_dev = orig_request->img_request->rbd_dev; | |
2637 | if (!rbd_dev->parent_overlap) { | |
2638 | struct ceph_osd_client *osdc; | |
2639 | ||
638f5abe AE |
2640 | osdc = &rbd_dev->rbd_client->client->osdc; |
2641 | result = rbd_obj_request_submit(osdc, orig_request); | |
2642 | if (!result) | |
2643 | return; | |
2644 | } | |
c5b5ef6c AE |
2645 | |
2646 | /* | |
2647 | * Our only purpose here is to determine whether the object | |
2648 | * exists, and we don't want to treat the non-existence as | |
2649 | * an error. If something else comes back, transfer the | |
2650 | * error to the original request and complete it now. | |
2651 | */ | |
2652 | if (!result) { | |
2653 | obj_request_existence_set(orig_request, true); | |
2654 | } else if (result == -ENOENT) { | |
2655 | obj_request_existence_set(orig_request, false); | |
2656 | } else if (result) { | |
2657 | orig_request->result = result; | |
3d7efd18 | 2658 | goto out; |
c5b5ef6c AE |
2659 | } |
2660 | ||
2661 | /* | |
2662 | * Resubmit the original request now that we have recorded | |
2663 | * whether the target object exists. | |
2664 | */ | |
b454e36d | 2665 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2666 | out: |
c5b5ef6c AE |
2667 | if (orig_request->result) |
2668 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2669 | } |
2670 | ||
2671 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2672 | { | |
2673 | struct rbd_obj_request *stat_request; | |
2674 | struct rbd_device *rbd_dev; | |
2675 | struct ceph_osd_client *osdc; | |
2676 | struct page **pages = NULL; | |
2677 | u32 page_count; | |
2678 | size_t size; | |
2679 | int ret; | |
2680 | ||
2681 | /* | |
2682 | * The response data for a STAT call consists of: | |
2683 | * le64 length; | |
2684 | * struct { | |
2685 | * le32 tv_sec; | |
2686 | * le32 tv_nsec; | |
2687 | * } mtime; | |
2688 | */ | |
2689 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2690 | page_count = (u32)calc_pages_for(0, size); | |
2691 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2692 | if (IS_ERR(pages)) | |
2693 | return PTR_ERR(pages); | |
2694 | ||
2695 | ret = -ENOMEM; | |
2696 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2697 | OBJ_REQUEST_PAGES); | |
2698 | if (!stat_request) | |
2699 | goto out; | |
2700 | ||
2701 | rbd_obj_request_get(obj_request); | |
2702 | stat_request->obj_request = obj_request; | |
2703 | stat_request->pages = pages; | |
2704 | stat_request->page_count = page_count; | |
2705 | ||
2706 | rbd_assert(obj_request->img_request); | |
2707 | rbd_dev = obj_request->img_request->rbd_dev; | |
deb236b3 ID |
2708 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, 1, |
2709 | stat_request); | |
c5b5ef6c AE |
2710 | if (!stat_request->osd_req) |
2711 | goto out; | |
2712 | stat_request->callback = rbd_img_obj_exists_callback; | |
2713 | ||
2714 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2715 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2716 | false, false); | |
9d4df01f | 2717 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2718 | |
2719 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2720 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2721 | out: | |
2722 | if (ret) | |
2723 | rbd_obj_request_put(obj_request); | |
2724 | ||
2725 | return ret; | |
2726 | } | |
2727 | ||
b454e36d AE |
2728 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2729 | { | |
2730 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2731 | struct rbd_device *rbd_dev; |
3d7efd18 | 2732 | bool known; |
b454e36d AE |
2733 | |
2734 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2735 | ||
2736 | img_request = obj_request->img_request; | |
2737 | rbd_assert(img_request); | |
a9e8ba2c | 2738 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2739 | |
b454e36d | 2740 | /* |
a9e8ba2c AE |
2741 | * Only writes to layered images need special handling. |
2742 | * Reads and non-layered writes are simple object requests. | |
2743 | * Layered writes that start beyond the end of the overlap | |
2744 | * with the parent have no parent data, so they too are | |
2745 | * simple object requests. Finally, if the target object is | |
2746 | * known to already exist, its parent data has already been | |
2747 | * copied, so a write to the object can also be handled as a | |
2748 | * simple object request. | |
b454e36d AE |
2749 | */ |
2750 | if (!img_request_write_test(img_request) || | |
2751 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2752 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2753 | ((known = obj_request_known_test(obj_request)) && |
2754 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2755 | |
2756 | struct rbd_device *rbd_dev; | |
2757 | struct ceph_osd_client *osdc; | |
2758 | ||
2759 | rbd_dev = obj_request->img_request->rbd_dev; | |
2760 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2761 | ||
2762 | return rbd_obj_request_submit(osdc, obj_request); | |
2763 | } | |
2764 | ||
2765 | /* | |
3d7efd18 AE |
2766 | * It's a layered write. The target object might exist but |
2767 | * we may not know that yet. If we know it doesn't exist, | |
2768 | * start by reading the data for the full target object from | |
2769 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2770 | */ |
3d7efd18 AE |
2771 | if (known) |
2772 | return rbd_img_obj_parent_read_full(obj_request); | |
2773 | ||
2774 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2775 | |
2776 | return rbd_img_obj_exists_submit(obj_request); | |
2777 | } | |
2778 | ||
bf0d5f50 AE |
2779 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2780 | { | |
bf0d5f50 | 2781 | struct rbd_obj_request *obj_request; |
46faeed4 | 2782 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2783 | |
37206ee5 | 2784 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2785 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2786 | int ret; |
2787 | ||
b454e36d | 2788 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2789 | if (ret) |
2790 | return ret; | |
bf0d5f50 AE |
2791 | } |
2792 | ||
2793 | return 0; | |
2794 | } | |
8b3e1a56 AE |
2795 | |
2796 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2797 | { | |
2798 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2799 | struct rbd_device *rbd_dev; |
2800 | u64 obj_end; | |
02c74fba AE |
2801 | u64 img_xferred; |
2802 | int img_result; | |
8b3e1a56 AE |
2803 | |
2804 | rbd_assert(img_request_child_test(img_request)); | |
2805 | ||
02c74fba AE |
2806 | /* First get what we need from the image request and release it */ |
2807 | ||
8b3e1a56 | 2808 | obj_request = img_request->obj_request; |
02c74fba AE |
2809 | img_xferred = img_request->xferred; |
2810 | img_result = img_request->result; | |
2811 | rbd_img_request_put(img_request); | |
2812 | ||
2813 | /* | |
2814 | * If the overlap has become 0 (most likely because the | |
2815 | * image has been flattened) we need to re-submit the | |
2816 | * original request. | |
2817 | */ | |
a9e8ba2c AE |
2818 | rbd_assert(obj_request); |
2819 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2820 | rbd_dev = obj_request->img_request->rbd_dev; |
2821 | if (!rbd_dev->parent_overlap) { | |
2822 | struct ceph_osd_client *osdc; | |
2823 | ||
2824 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2825 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2826 | if (!img_result) | |
2827 | return; | |
2828 | } | |
a9e8ba2c | 2829 | |
02c74fba | 2830 | obj_request->result = img_result; |
a9e8ba2c AE |
2831 | if (obj_request->result) |
2832 | goto out; | |
2833 | ||
2834 | /* | |
2835 | * We need to zero anything beyond the parent overlap | |
2836 | * boundary. Since rbd_img_obj_request_read_callback() | |
2837 | * will zero anything beyond the end of a short read, an | |
2838 | * easy way to do this is to pretend the data from the | |
2839 | * parent came up short--ending at the overlap boundary. | |
2840 | */ | |
2841 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2842 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2843 | if (obj_end > rbd_dev->parent_overlap) { |
2844 | u64 xferred = 0; | |
2845 | ||
2846 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2847 | xferred = rbd_dev->parent_overlap - | |
2848 | obj_request->img_offset; | |
8b3e1a56 | 2849 | |
02c74fba | 2850 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2851 | } else { |
02c74fba | 2852 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2853 | } |
2854 | out: | |
8b3e1a56 AE |
2855 | rbd_img_obj_request_read_callback(obj_request); |
2856 | rbd_obj_request_complete(obj_request); | |
2857 | } | |
2858 | ||
2859 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2860 | { | |
8b3e1a56 AE |
2861 | struct rbd_img_request *img_request; |
2862 | int result; | |
2863 | ||
2864 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2865 | rbd_assert(obj_request->img_request != NULL); | |
2866 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 2867 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 2868 | |
8b3e1a56 | 2869 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 2870 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 2871 | obj_request->img_offset, |
e93f3152 | 2872 | obj_request->length); |
8b3e1a56 AE |
2873 | result = -ENOMEM; |
2874 | if (!img_request) | |
2875 | goto out_err; | |
2876 | ||
5b2ab72d AE |
2877 | if (obj_request->type == OBJ_REQUEST_BIO) |
2878 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
2879 | obj_request->bio_list); | |
2880 | else | |
2881 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
2882 | obj_request->pages); | |
8b3e1a56 AE |
2883 | if (result) |
2884 | goto out_err; | |
2885 | ||
2886 | img_request->callback = rbd_img_parent_read_callback; | |
2887 | result = rbd_img_request_submit(img_request); | |
2888 | if (result) | |
2889 | goto out_err; | |
2890 | ||
2891 | return; | |
2892 | out_err: | |
2893 | if (img_request) | |
2894 | rbd_img_request_put(img_request); | |
2895 | obj_request->result = result; | |
2896 | obj_request->xferred = 0; | |
2897 | obj_request_done_set(obj_request); | |
2898 | } | |
bf0d5f50 | 2899 | |
20e0af67 | 2900 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2901 | { |
2902 | struct rbd_obj_request *obj_request; | |
2169238d | 2903 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2904 | int ret; |
2905 | ||
2906 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2907 | OBJ_REQUEST_NODATA); | |
2908 | if (!obj_request) | |
2909 | return -ENOMEM; | |
2910 | ||
2911 | ret = -ENOMEM; | |
deb236b3 ID |
2912 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, 1, |
2913 | obj_request); | |
b8d70035 AE |
2914 | if (!obj_request->osd_req) |
2915 | goto out; | |
2916 | ||
c99d2d4a | 2917 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2918 | notify_id, 0, 0); |
9d4df01f | 2919 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2920 | |
b8d70035 | 2921 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 2922 | if (ret) |
20e0af67 JD |
2923 | goto out; |
2924 | ret = rbd_obj_request_wait(obj_request); | |
2925 | out: | |
2926 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2927 | |
2928 | return ret; | |
2929 | } | |
2930 | ||
2931 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2932 | { | |
2933 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 2934 | int ret; |
b8d70035 AE |
2935 | |
2936 | if (!rbd_dev) | |
2937 | return; | |
2938 | ||
37206ee5 | 2939 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2940 | rbd_dev->header_name, (unsigned long long)notify_id, |
2941 | (unsigned int)opcode); | |
e627db08 AE |
2942 | ret = rbd_dev_refresh(rbd_dev); |
2943 | if (ret) | |
3b5cf2a2 | 2944 | rbd_warn(rbd_dev, "header refresh error (%d)\n", ret); |
b8d70035 | 2945 | |
20e0af67 | 2946 | rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
b8d70035 AE |
2947 | } |
2948 | ||
9969ebc5 | 2949 | /* |
b30a01f2 | 2950 | * Initiate a watch request, synchronously. |
9969ebc5 | 2951 | */ |
b30a01f2 | 2952 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
2953 | { |
2954 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2955 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2956 | int ret; |
2957 | ||
b30a01f2 ID |
2958 | rbd_assert(!rbd_dev->watch_event); |
2959 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 2960 | |
b30a01f2 ID |
2961 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
2962 | &rbd_dev->watch_event); | |
2963 | if (ret < 0) | |
2964 | return ret; | |
2965 | ||
2966 | rbd_assert(rbd_dev->watch_event); | |
9969ebc5 | 2967 | |
9969ebc5 | 2968 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, |
b30a01f2 ID |
2969 | OBJ_REQUEST_NODATA); |
2970 | if (!obj_request) { | |
2971 | ret = -ENOMEM; | |
9969ebc5 | 2972 | goto out_cancel; |
b30a01f2 | 2973 | } |
9969ebc5 | 2974 | |
deb236b3 ID |
2975 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, 1, |
2976 | obj_request); | |
b30a01f2 ID |
2977 | if (!obj_request->osd_req) { |
2978 | ret = -ENOMEM; | |
2979 | goto out_put; | |
2980 | } | |
430c28c3 | 2981 | |
b30a01f2 | 2982 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
2169238d AE |
2983 | |
2984 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
b30a01f2 | 2985 | rbd_dev->watch_event->cookie, 0, 1); |
9d4df01f | 2986 | rbd_osd_req_format_write(obj_request); |
2169238d | 2987 | |
9969ebc5 AE |
2988 | ret = rbd_obj_request_submit(osdc, obj_request); |
2989 | if (ret) | |
b30a01f2 ID |
2990 | goto out_linger; |
2991 | ||
9969ebc5 AE |
2992 | ret = rbd_obj_request_wait(obj_request); |
2993 | if (ret) | |
b30a01f2 ID |
2994 | goto out_linger; |
2995 | ||
9969ebc5 AE |
2996 | ret = obj_request->result; |
2997 | if (ret) | |
b30a01f2 | 2998 | goto out_linger; |
9969ebc5 | 2999 | |
8eb87565 AE |
3000 | /* |
3001 | * A watch request is set to linger, so the underlying osd | |
3002 | * request won't go away until we unregister it. We retain | |
3003 | * a pointer to the object request during that time (in | |
3004 | * rbd_dev->watch_request), so we'll keep a reference to | |
3005 | * it. We'll drop that reference (below) after we've | |
3006 | * unregistered it. | |
3007 | */ | |
b30a01f2 | 3008 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3009 | |
b30a01f2 ID |
3010 | return 0; |
3011 | ||
3012 | out_linger: | |
3013 | ceph_osdc_unregister_linger_request(osdc, obj_request->osd_req); | |
3014 | out_put: | |
3015 | rbd_obj_request_put(obj_request); | |
3016 | out_cancel: | |
3017 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3018 | rbd_dev->watch_event = NULL; | |
3019 | ||
3020 | return ret; | |
3021 | } | |
3022 | ||
3023 | /* | |
3024 | * Tear down a watch request, synchronously. | |
3025 | */ | |
3026 | static int __rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) | |
3027 | { | |
3028 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3029 | struct rbd_obj_request *obj_request; | |
3030 | int ret; | |
3031 | ||
3032 | rbd_assert(rbd_dev->watch_event); | |
3033 | rbd_assert(rbd_dev->watch_request); | |
3034 | ||
3035 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3036 | OBJ_REQUEST_NODATA); | |
3037 | if (!obj_request) { | |
3038 | ret = -ENOMEM; | |
3039 | goto out_cancel; | |
3040 | } | |
3041 | ||
3042 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, 1, | |
3043 | obj_request); | |
3044 | if (!obj_request->osd_req) { | |
3045 | ret = -ENOMEM; | |
3046 | goto out_put; | |
8eb87565 AE |
3047 | } |
3048 | ||
b30a01f2 ID |
3049 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, |
3050 | rbd_dev->watch_event->cookie, 0, 0); | |
3051 | rbd_osd_req_format_write(obj_request); | |
3052 | ||
3053 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3054 | if (ret) | |
3055 | goto out_put; | |
3056 | ||
3057 | ret = rbd_obj_request_wait(obj_request); | |
3058 | if (ret) | |
3059 | goto out_put; | |
3060 | ||
3061 | ret = obj_request->result; | |
3062 | if (ret) | |
3063 | goto out_put; | |
3064 | ||
8eb87565 AE |
3065 | /* We have successfully torn down the watch request */ |
3066 | ||
b30a01f2 ID |
3067 | ceph_osdc_unregister_linger_request(osdc, |
3068 | rbd_dev->watch_request->osd_req); | |
8eb87565 AE |
3069 | rbd_obj_request_put(rbd_dev->watch_request); |
3070 | rbd_dev->watch_request = NULL; | |
b30a01f2 ID |
3071 | |
3072 | out_put: | |
3073 | rbd_obj_request_put(obj_request); | |
9969ebc5 | 3074 | out_cancel: |
9969ebc5 AE |
3075 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3076 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
3077 | |
3078 | return ret; | |
3079 | } | |
3080 | ||
fca27065 ID |
3081 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
3082 | { | |
3083 | int ret; | |
3084 | ||
b30a01f2 | 3085 | ret = __rbd_dev_header_unwatch_sync(rbd_dev); |
fca27065 ID |
3086 | if (ret) { |
3087 | rbd_warn(rbd_dev, "unable to tear down watch request: %d\n", | |
3088 | ret); | |
3089 | } | |
3090 | } | |
3091 | ||
36be9a76 | 3092 | /* |
f40eb349 AE |
3093 | * Synchronous osd object method call. Returns the number of bytes |
3094 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3095 | */ |
3096 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3097 | const char *object_name, | |
3098 | const char *class_name, | |
3099 | const char *method_name, | |
4157976b | 3100 | const void *outbound, |
36be9a76 | 3101 | size_t outbound_size, |
4157976b | 3102 | void *inbound, |
e2a58ee5 | 3103 | size_t inbound_size) |
36be9a76 | 3104 | { |
2169238d | 3105 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3106 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3107 | struct page **pages; |
3108 | u32 page_count; | |
3109 | int ret; | |
3110 | ||
3111 | /* | |
6010a451 AE |
3112 | * Method calls are ultimately read operations. The result |
3113 | * should placed into the inbound buffer provided. They | |
3114 | * also supply outbound data--parameters for the object | |
3115 | * method. Currently if this is present it will be a | |
3116 | * snapshot id. | |
36be9a76 | 3117 | */ |
57385b51 | 3118 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3119 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3120 | if (IS_ERR(pages)) | |
3121 | return PTR_ERR(pages); | |
3122 | ||
3123 | ret = -ENOMEM; | |
6010a451 | 3124 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3125 | OBJ_REQUEST_PAGES); |
3126 | if (!obj_request) | |
3127 | goto out; | |
3128 | ||
3129 | obj_request->pages = pages; | |
3130 | obj_request->page_count = page_count; | |
3131 | ||
deb236b3 ID |
3132 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, 1, |
3133 | obj_request); | |
36be9a76 AE |
3134 | if (!obj_request->osd_req) |
3135 | goto out; | |
3136 | ||
c99d2d4a | 3137 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3138 | class_name, method_name); |
3139 | if (outbound_size) { | |
3140 | struct ceph_pagelist *pagelist; | |
3141 | ||
3142 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3143 | if (!pagelist) | |
3144 | goto out; | |
3145 | ||
3146 | ceph_pagelist_init(pagelist); | |
3147 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3148 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3149 | pagelist); | |
3150 | } | |
a4ce40a9 AE |
3151 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3152 | obj_request->pages, inbound_size, | |
44cd188d | 3153 | 0, false, false); |
9d4df01f | 3154 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3155 | |
36be9a76 AE |
3156 | ret = rbd_obj_request_submit(osdc, obj_request); |
3157 | if (ret) | |
3158 | goto out; | |
3159 | ret = rbd_obj_request_wait(obj_request); | |
3160 | if (ret) | |
3161 | goto out; | |
3162 | ||
3163 | ret = obj_request->result; | |
3164 | if (ret < 0) | |
3165 | goto out; | |
57385b51 AE |
3166 | |
3167 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3168 | ret = (int)obj_request->xferred; | |
903bb32e | 3169 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3170 | out: |
3171 | if (obj_request) | |
3172 | rbd_obj_request_put(obj_request); | |
3173 | else | |
3174 | ceph_release_page_vector(pages, page_count); | |
3175 | ||
3176 | return ret; | |
3177 | } | |
3178 | ||
bf0d5f50 | 3179 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 3180 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
3181 | { |
3182 | struct rbd_device *rbd_dev = q->queuedata; | |
bf0d5f50 AE |
3183 | struct request *rq; |
3184 | int result; | |
3185 | ||
3186 | while ((rq = blk_fetch_request(q))) { | |
3187 | bool write_request = rq_data_dir(rq) == WRITE; | |
3188 | struct rbd_img_request *img_request; | |
3189 | u64 offset; | |
3190 | u64 length; | |
3191 | ||
3192 | /* Ignore any non-FS requests that filter through. */ | |
3193 | ||
3194 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
3195 | dout("%s: non-fs request type %d\n", __func__, |
3196 | (int) rq->cmd_type); | |
3197 | __blk_end_request_all(rq, 0); | |
3198 | continue; | |
3199 | } | |
3200 | ||
3201 | /* Ignore/skip any zero-length requests */ | |
3202 | ||
3203 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
3204 | length = (u64) blk_rq_bytes(rq); | |
3205 | ||
3206 | if (!length) { | |
3207 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
3208 | __blk_end_request_all(rq, 0); |
3209 | continue; | |
3210 | } | |
3211 | ||
3212 | spin_unlock_irq(q->queue_lock); | |
3213 | ||
3214 | /* Disallow writes to a read-only device */ | |
3215 | ||
3216 | if (write_request) { | |
3217 | result = -EROFS; | |
131fd9f6 | 3218 | if (rbd_dev->mapping.read_only) |
bf0d5f50 AE |
3219 | goto end_request; |
3220 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
3221 | } | |
3222 | ||
6d292906 AE |
3223 | /* |
3224 | * Quit early if the mapped snapshot no longer | |
3225 | * exists. It's still possible the snapshot will | |
3226 | * have disappeared by the time our request arrives | |
3227 | * at the osd, but there's no sense in sending it if | |
3228 | * we already know. | |
3229 | */ | |
3230 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
3231 | dout("request for non-existent snapshot"); |
3232 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3233 | result = -ENXIO; | |
3234 | goto end_request; | |
3235 | } | |
3236 | ||
bf0d5f50 | 3237 | result = -EINVAL; |
c0cd10db AE |
3238 | if (offset && length > U64_MAX - offset + 1) { |
3239 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
3240 | offset, length); | |
bf0d5f50 | 3241 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 3242 | } |
bf0d5f50 | 3243 | |
00a653e2 AE |
3244 | result = -EIO; |
3245 | if (offset + length > rbd_dev->mapping.size) { | |
3246 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n", | |
3247 | offset, length, rbd_dev->mapping.size); | |
3248 | goto end_request; | |
3249 | } | |
3250 | ||
bf0d5f50 AE |
3251 | result = -ENOMEM; |
3252 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
e93f3152 | 3253 | write_request); |
bf0d5f50 AE |
3254 | if (!img_request) |
3255 | goto end_request; | |
3256 | ||
3257 | img_request->rq = rq; | |
3258 | ||
f1a4739f AE |
3259 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
3260 | rq->bio); | |
bf0d5f50 AE |
3261 | if (!result) |
3262 | result = rbd_img_request_submit(img_request); | |
3263 | if (result) | |
3264 | rbd_img_request_put(img_request); | |
3265 | end_request: | |
3266 | spin_lock_irq(q->queue_lock); | |
3267 | if (result < 0) { | |
7da22d29 AE |
3268 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
3269 | write_request ? "write" : "read", | |
3270 | length, offset, result); | |
3271 | ||
bf0d5f50 AE |
3272 | __blk_end_request_all(rq, result); |
3273 | } | |
3274 | } | |
3275 | } | |
3276 | ||
602adf40 YS |
3277 | /* |
3278 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3279 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3280 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3281 | */ |
3282 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3283 | struct bio_vec *bvec) | |
3284 | { | |
3285 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3286 | sector_t sector_offset; |
3287 | sector_t sectors_per_obj; | |
3288 | sector_t obj_sector_offset; | |
3289 | int ret; | |
3290 | ||
3291 | /* | |
3292 | * Find how far into its rbd object the partition-relative | |
3293 | * bio start sector is to offset relative to the enclosing | |
3294 | * device. | |
3295 | */ | |
3296 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3297 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3298 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3299 | ||
3300 | /* | |
3301 | * Compute the number of bytes from that offset to the end | |
3302 | * of the object. Account for what's already used by the bio. | |
3303 | */ | |
3304 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3305 | if (ret > bmd->bi_size) | |
3306 | ret -= bmd->bi_size; | |
3307 | else | |
3308 | ret = 0; | |
3309 | ||
3310 | /* | |
3311 | * Don't send back more than was asked for. And if the bio | |
3312 | * was empty, let the whole thing through because: "Note | |
3313 | * that a block device *must* allow a single page to be | |
3314 | * added to an empty bio." | |
3315 | */ | |
3316 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3317 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3318 | ret = (int) bvec->bv_len; | |
3319 | ||
3320 | return ret; | |
602adf40 YS |
3321 | } |
3322 | ||
3323 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3324 | { | |
3325 | struct gendisk *disk = rbd_dev->disk; | |
3326 | ||
3327 | if (!disk) | |
3328 | return; | |
3329 | ||
a0cab924 AE |
3330 | rbd_dev->disk = NULL; |
3331 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3332 | del_gendisk(disk); |
a0cab924 AE |
3333 | if (disk->queue) |
3334 | blk_cleanup_queue(disk->queue); | |
3335 | } | |
602adf40 YS |
3336 | put_disk(disk); |
3337 | } | |
3338 | ||
788e2df3 AE |
3339 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3340 | const char *object_name, | |
7097f8df | 3341 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3342 | |
3343 | { | |
2169238d | 3344 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3345 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3346 | struct page **pages = NULL; |
3347 | u32 page_count; | |
1ceae7ef | 3348 | size_t size; |
788e2df3 AE |
3349 | int ret; |
3350 | ||
3351 | page_count = (u32) calc_pages_for(offset, length); | |
3352 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3353 | if (IS_ERR(pages)) | |
3354 | ret = PTR_ERR(pages); | |
3355 | ||
3356 | ret = -ENOMEM; | |
3357 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3358 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3359 | if (!obj_request) |
3360 | goto out; | |
3361 | ||
3362 | obj_request->pages = pages; | |
3363 | obj_request->page_count = page_count; | |
3364 | ||
deb236b3 ID |
3365 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, 1, |
3366 | obj_request); | |
788e2df3 AE |
3367 | if (!obj_request->osd_req) |
3368 | goto out; | |
3369 | ||
c99d2d4a AE |
3370 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3371 | offset, length, 0, 0); | |
406e2c9f | 3372 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3373 | obj_request->pages, |
44cd188d AE |
3374 | obj_request->length, |
3375 | obj_request->offset & ~PAGE_MASK, | |
3376 | false, false); | |
9d4df01f | 3377 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3378 | |
788e2df3 AE |
3379 | ret = rbd_obj_request_submit(osdc, obj_request); |
3380 | if (ret) | |
3381 | goto out; | |
3382 | ret = rbd_obj_request_wait(obj_request); | |
3383 | if (ret) | |
3384 | goto out; | |
3385 | ||
3386 | ret = obj_request->result; | |
3387 | if (ret < 0) | |
3388 | goto out; | |
1ceae7ef AE |
3389 | |
3390 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3391 | size = (size_t) obj_request->xferred; | |
903bb32e | 3392 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3393 | rbd_assert(size <= (size_t)INT_MAX); |
3394 | ret = (int)size; | |
788e2df3 AE |
3395 | out: |
3396 | if (obj_request) | |
3397 | rbd_obj_request_put(obj_request); | |
3398 | else | |
3399 | ceph_release_page_vector(pages, page_count); | |
3400 | ||
3401 | return ret; | |
3402 | } | |
3403 | ||
602adf40 | 3404 | /* |
662518b1 AE |
3405 | * Read the complete header for the given rbd device. On successful |
3406 | * return, the rbd_dev->header field will contain up-to-date | |
3407 | * information about the image. | |
602adf40 | 3408 | */ |
99a41ebc | 3409 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3410 | { |
4156d998 | 3411 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3412 | u32 snap_count = 0; |
4156d998 AE |
3413 | u64 names_size = 0; |
3414 | u32 want_count; | |
3415 | int ret; | |
602adf40 | 3416 | |
00f1f36f | 3417 | /* |
4156d998 AE |
3418 | * The complete header will include an array of its 64-bit |
3419 | * snapshot ids, followed by the names of those snapshots as | |
3420 | * a contiguous block of NUL-terminated strings. Note that | |
3421 | * the number of snapshots could change by the time we read | |
3422 | * it in, in which case we re-read it. | |
00f1f36f | 3423 | */ |
4156d998 AE |
3424 | do { |
3425 | size_t size; | |
3426 | ||
3427 | kfree(ondisk); | |
3428 | ||
3429 | size = sizeof (*ondisk); | |
3430 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3431 | size += names_size; | |
3432 | ondisk = kmalloc(size, GFP_KERNEL); | |
3433 | if (!ondisk) | |
662518b1 | 3434 | return -ENOMEM; |
4156d998 | 3435 | |
788e2df3 | 3436 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3437 | 0, size, ondisk); |
4156d998 | 3438 | if (ret < 0) |
662518b1 | 3439 | goto out; |
c0cd10db | 3440 | if ((size_t)ret < size) { |
4156d998 | 3441 | ret = -ENXIO; |
06ecc6cb AE |
3442 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3443 | size, ret); | |
662518b1 | 3444 | goto out; |
4156d998 AE |
3445 | } |
3446 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3447 | ret = -ENXIO; | |
06ecc6cb | 3448 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3449 | goto out; |
81e759fb | 3450 | } |
602adf40 | 3451 | |
4156d998 AE |
3452 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3453 | want_count = snap_count; | |
3454 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3455 | } while (snap_count != want_count); | |
00f1f36f | 3456 | |
662518b1 AE |
3457 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3458 | out: | |
4156d998 AE |
3459 | kfree(ondisk); |
3460 | ||
3461 | return ret; | |
602adf40 YS |
3462 | } |
3463 | ||
15228ede AE |
3464 | /* |
3465 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3466 | * has disappeared from the (just updated) snapshot context. | |
3467 | */ | |
3468 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3469 | { | |
3470 | u64 snap_id; | |
3471 | ||
3472 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3473 | return; | |
3474 | ||
3475 | snap_id = rbd_dev->spec->snap_id; | |
3476 | if (snap_id == CEPH_NOSNAP) | |
3477 | return; | |
3478 | ||
3479 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3480 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3481 | } | |
3482 | ||
9875201e JD |
3483 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3484 | { | |
3485 | sector_t size; | |
3486 | bool removing; | |
3487 | ||
3488 | /* | |
3489 | * Don't hold the lock while doing disk operations, | |
3490 | * or lock ordering will conflict with the bdev mutex via: | |
3491 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3492 | */ | |
3493 | spin_lock_irq(&rbd_dev->lock); | |
3494 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3495 | spin_unlock_irq(&rbd_dev->lock); | |
3496 | /* | |
3497 | * If the device is being removed, rbd_dev->disk has | |
3498 | * been destroyed, so don't try to update its size | |
3499 | */ | |
3500 | if (!removing) { | |
3501 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3502 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3503 | set_capacity(rbd_dev->disk, size); | |
3504 | revalidate_disk(rbd_dev->disk); | |
3505 | } | |
3506 | } | |
3507 | ||
cc4a38bd | 3508 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3509 | { |
e627db08 | 3510 | u64 mapping_size; |
1fe5e993 AE |
3511 | int ret; |
3512 | ||
117973fb | 3513 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
cfbf6377 | 3514 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3515 | mapping_size = rbd_dev->mapping.size; |
117973fb | 3516 | if (rbd_dev->image_format == 1) |
99a41ebc | 3517 | ret = rbd_dev_v1_header_info(rbd_dev); |
117973fb | 3518 | else |
2df3fac7 | 3519 | ret = rbd_dev_v2_header_info(rbd_dev); |
15228ede AE |
3520 | |
3521 | /* If it's a mapped snapshot, validate its EXISTS flag */ | |
3522 | ||
3523 | rbd_exists_validate(rbd_dev); | |
cfbf6377 AE |
3524 | up_write(&rbd_dev->header_rwsem); |
3525 | ||
00a653e2 | 3526 | if (mapping_size != rbd_dev->mapping.size) { |
9875201e | 3527 | rbd_dev_update_size(rbd_dev); |
00a653e2 | 3528 | } |
1fe5e993 AE |
3529 | |
3530 | return ret; | |
3531 | } | |
3532 | ||
602adf40 YS |
3533 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3534 | { | |
3535 | struct gendisk *disk; | |
3536 | struct request_queue *q; | |
593a9e7b | 3537 | u64 segment_size; |
602adf40 | 3538 | |
602adf40 | 3539 | /* create gendisk info */ |
7e513d43 ID |
3540 | disk = alloc_disk(single_major ? |
3541 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3542 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3543 | if (!disk) |
1fcdb8aa | 3544 | return -ENOMEM; |
602adf40 | 3545 | |
f0f8cef5 | 3546 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3547 | rbd_dev->dev_id); |
602adf40 | 3548 | disk->major = rbd_dev->major; |
dd82fff1 | 3549 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3550 | if (single_major) |
3551 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3552 | disk->fops = &rbd_bd_ops; |
3553 | disk->private_data = rbd_dev; | |
3554 | ||
bf0d5f50 | 3555 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3556 | if (!q) |
3557 | goto out_disk; | |
029bcbd8 | 3558 | |
593a9e7b AE |
3559 | /* We use the default size, but let's be explicit about it. */ |
3560 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3561 | ||
029bcbd8 | 3562 | /* set io sizes to object size */ |
593a9e7b AE |
3563 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3564 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3565 | blk_queue_max_segment_size(q, segment_size); | |
3566 | blk_queue_io_min(q, segment_size); | |
3567 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3568 | |
602adf40 YS |
3569 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3570 | disk->queue = q; | |
3571 | ||
3572 | q->queuedata = rbd_dev; | |
3573 | ||
3574 | rbd_dev->disk = disk; | |
602adf40 | 3575 | |
602adf40 | 3576 | return 0; |
602adf40 YS |
3577 | out_disk: |
3578 | put_disk(disk); | |
1fcdb8aa AE |
3579 | |
3580 | return -ENOMEM; | |
602adf40 YS |
3581 | } |
3582 | ||
dfc5606d YS |
3583 | /* |
3584 | sysfs | |
3585 | */ | |
3586 | ||
593a9e7b AE |
3587 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3588 | { | |
3589 | return container_of(dev, struct rbd_device, dev); | |
3590 | } | |
3591 | ||
dfc5606d YS |
3592 | static ssize_t rbd_size_show(struct device *dev, |
3593 | struct device_attribute *attr, char *buf) | |
3594 | { | |
593a9e7b | 3595 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3596 | |
fc71d833 AE |
3597 | return sprintf(buf, "%llu\n", |
3598 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3599 | } |
3600 | ||
34b13184 AE |
3601 | /* |
3602 | * Note this shows the features for whatever's mapped, which is not | |
3603 | * necessarily the base image. | |
3604 | */ | |
3605 | static ssize_t rbd_features_show(struct device *dev, | |
3606 | struct device_attribute *attr, char *buf) | |
3607 | { | |
3608 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3609 | ||
3610 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3611 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3612 | } |
3613 | ||
dfc5606d YS |
3614 | static ssize_t rbd_major_show(struct device *dev, |
3615 | struct device_attribute *attr, char *buf) | |
3616 | { | |
593a9e7b | 3617 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3618 | |
fc71d833 AE |
3619 | if (rbd_dev->major) |
3620 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3621 | ||
3622 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3623 | } |
3624 | ||
3625 | static ssize_t rbd_minor_show(struct device *dev, | |
3626 | struct device_attribute *attr, char *buf) | |
3627 | { | |
3628 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3629 | |
dd82fff1 | 3630 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3631 | } |
3632 | ||
3633 | static ssize_t rbd_client_id_show(struct device *dev, | |
3634 | struct device_attribute *attr, char *buf) | |
602adf40 | 3635 | { |
593a9e7b | 3636 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3637 | |
1dbb4399 AE |
3638 | return sprintf(buf, "client%lld\n", |
3639 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3640 | } |
3641 | ||
dfc5606d YS |
3642 | static ssize_t rbd_pool_show(struct device *dev, |
3643 | struct device_attribute *attr, char *buf) | |
602adf40 | 3644 | { |
593a9e7b | 3645 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3646 | |
0d7dbfce | 3647 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3648 | } |
3649 | ||
9bb2f334 AE |
3650 | static ssize_t rbd_pool_id_show(struct device *dev, |
3651 | struct device_attribute *attr, char *buf) | |
3652 | { | |
3653 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3654 | ||
0d7dbfce | 3655 | return sprintf(buf, "%llu\n", |
fc71d833 | 3656 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3657 | } |
3658 | ||
dfc5606d YS |
3659 | static ssize_t rbd_name_show(struct device *dev, |
3660 | struct device_attribute *attr, char *buf) | |
3661 | { | |
593a9e7b | 3662 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3663 | |
a92ffdf8 AE |
3664 | if (rbd_dev->spec->image_name) |
3665 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3666 | ||
3667 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3668 | } |
3669 | ||
589d30e0 AE |
3670 | static ssize_t rbd_image_id_show(struct device *dev, |
3671 | struct device_attribute *attr, char *buf) | |
3672 | { | |
3673 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3674 | ||
0d7dbfce | 3675 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3676 | } |
3677 | ||
34b13184 AE |
3678 | /* |
3679 | * Shows the name of the currently-mapped snapshot (or | |
3680 | * RBD_SNAP_HEAD_NAME for the base image). | |
3681 | */ | |
dfc5606d YS |
3682 | static ssize_t rbd_snap_show(struct device *dev, |
3683 | struct device_attribute *attr, | |
3684 | char *buf) | |
3685 | { | |
593a9e7b | 3686 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3687 | |
0d7dbfce | 3688 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3689 | } |
3690 | ||
86b00e0d AE |
3691 | /* |
3692 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3693 | * for the parent image. If there is no parent, simply shows | |
3694 | * "(no parent image)". | |
3695 | */ | |
3696 | static ssize_t rbd_parent_show(struct device *dev, | |
3697 | struct device_attribute *attr, | |
3698 | char *buf) | |
3699 | { | |
3700 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3701 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3702 | int count; | |
3703 | char *bufp = buf; | |
3704 | ||
3705 | if (!spec) | |
3706 | return sprintf(buf, "(no parent image)\n"); | |
3707 | ||
3708 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3709 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3710 | if (count < 0) | |
3711 | return count; | |
3712 | bufp += count; | |
3713 | ||
3714 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3715 | spec->image_name ? spec->image_name : "(unknown)"); | |
3716 | if (count < 0) | |
3717 | return count; | |
3718 | bufp += count; | |
3719 | ||
3720 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3721 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3722 | if (count < 0) | |
3723 | return count; | |
3724 | bufp += count; | |
3725 | ||
3726 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3727 | if (count < 0) | |
3728 | return count; | |
3729 | bufp += count; | |
3730 | ||
3731 | return (ssize_t) (bufp - buf); | |
3732 | } | |
3733 | ||
dfc5606d YS |
3734 | static ssize_t rbd_image_refresh(struct device *dev, |
3735 | struct device_attribute *attr, | |
3736 | const char *buf, | |
3737 | size_t size) | |
3738 | { | |
593a9e7b | 3739 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3740 | int ret; |
602adf40 | 3741 | |
cc4a38bd | 3742 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 AE |
3743 | if (ret) |
3744 | rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret); | |
b813623a AE |
3745 | |
3746 | return ret < 0 ? ret : size; | |
dfc5606d | 3747 | } |
602adf40 | 3748 | |
dfc5606d | 3749 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3750 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3751 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3752 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3753 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3754 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3755 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3756 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3757 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3758 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3759 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3760 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3761 | |
3762 | static struct attribute *rbd_attrs[] = { | |
3763 | &dev_attr_size.attr, | |
34b13184 | 3764 | &dev_attr_features.attr, |
dfc5606d | 3765 | &dev_attr_major.attr, |
dd82fff1 | 3766 | &dev_attr_minor.attr, |
dfc5606d YS |
3767 | &dev_attr_client_id.attr, |
3768 | &dev_attr_pool.attr, | |
9bb2f334 | 3769 | &dev_attr_pool_id.attr, |
dfc5606d | 3770 | &dev_attr_name.attr, |
589d30e0 | 3771 | &dev_attr_image_id.attr, |
dfc5606d | 3772 | &dev_attr_current_snap.attr, |
86b00e0d | 3773 | &dev_attr_parent.attr, |
dfc5606d | 3774 | &dev_attr_refresh.attr, |
dfc5606d YS |
3775 | NULL |
3776 | }; | |
3777 | ||
3778 | static struct attribute_group rbd_attr_group = { | |
3779 | .attrs = rbd_attrs, | |
3780 | }; | |
3781 | ||
3782 | static const struct attribute_group *rbd_attr_groups[] = { | |
3783 | &rbd_attr_group, | |
3784 | NULL | |
3785 | }; | |
3786 | ||
3787 | static void rbd_sysfs_dev_release(struct device *dev) | |
3788 | { | |
3789 | } | |
3790 | ||
3791 | static struct device_type rbd_device_type = { | |
3792 | .name = "rbd", | |
3793 | .groups = rbd_attr_groups, | |
3794 | .release = rbd_sysfs_dev_release, | |
3795 | }; | |
3796 | ||
8b8fb99c AE |
3797 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3798 | { | |
3799 | kref_get(&spec->kref); | |
3800 | ||
3801 | return spec; | |
3802 | } | |
3803 | ||
3804 | static void rbd_spec_free(struct kref *kref); | |
3805 | static void rbd_spec_put(struct rbd_spec *spec) | |
3806 | { | |
3807 | if (spec) | |
3808 | kref_put(&spec->kref, rbd_spec_free); | |
3809 | } | |
3810 | ||
3811 | static struct rbd_spec *rbd_spec_alloc(void) | |
3812 | { | |
3813 | struct rbd_spec *spec; | |
3814 | ||
3815 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3816 | if (!spec) | |
3817 | return NULL; | |
3818 | kref_init(&spec->kref); | |
3819 | ||
8b8fb99c AE |
3820 | return spec; |
3821 | } | |
3822 | ||
3823 | static void rbd_spec_free(struct kref *kref) | |
3824 | { | |
3825 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3826 | ||
3827 | kfree(spec->pool_name); | |
3828 | kfree(spec->image_id); | |
3829 | kfree(spec->image_name); | |
3830 | kfree(spec->snap_name); | |
3831 | kfree(spec); | |
3832 | } | |
3833 | ||
cc344fa1 | 3834 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3835 | struct rbd_spec *spec) |
3836 | { | |
3837 | struct rbd_device *rbd_dev; | |
3838 | ||
3839 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3840 | if (!rbd_dev) | |
3841 | return NULL; | |
3842 | ||
3843 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3844 | rbd_dev->flags = 0; |
a2acd00e | 3845 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 3846 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
3847 | init_rwsem(&rbd_dev->header_rwsem); |
3848 | ||
3849 | rbd_dev->spec = spec; | |
3850 | rbd_dev->rbd_client = rbdc; | |
3851 | ||
0903e875 AE |
3852 | /* Initialize the layout used for all rbd requests */ |
3853 | ||
3854 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3855 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3856 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3857 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3858 | ||
c53d5893 AE |
3859 | return rbd_dev; |
3860 | } | |
3861 | ||
3862 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3863 | { | |
c53d5893 AE |
3864 | rbd_put_client(rbd_dev->rbd_client); |
3865 | rbd_spec_put(rbd_dev->spec); | |
3866 | kfree(rbd_dev); | |
3867 | } | |
3868 | ||
9d475de5 AE |
3869 | /* |
3870 | * Get the size and object order for an image snapshot, or if | |
3871 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3872 | * image. | |
3873 | */ | |
3874 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3875 | u8 *order, u64 *snap_size) | |
3876 | { | |
3877 | __le64 snapid = cpu_to_le64(snap_id); | |
3878 | int ret; | |
3879 | struct { | |
3880 | u8 order; | |
3881 | __le64 size; | |
3882 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3883 | ||
36be9a76 | 3884 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3885 | "rbd", "get_size", |
4157976b | 3886 | &snapid, sizeof (snapid), |
e2a58ee5 | 3887 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3888 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3889 | if (ret < 0) |
3890 | return ret; | |
57385b51 AE |
3891 | if (ret < sizeof (size_buf)) |
3892 | return -ERANGE; | |
9d475de5 | 3893 | |
c3545579 | 3894 | if (order) { |
c86f86e9 | 3895 | *order = size_buf.order; |
c3545579 JD |
3896 | dout(" order %u", (unsigned int)*order); |
3897 | } | |
9d475de5 AE |
3898 | *snap_size = le64_to_cpu(size_buf.size); |
3899 | ||
c3545579 JD |
3900 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
3901 | (unsigned long long)snap_id, | |
57385b51 | 3902 | (unsigned long long)*snap_size); |
9d475de5 AE |
3903 | |
3904 | return 0; | |
3905 | } | |
3906 | ||
3907 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3908 | { | |
3909 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3910 | &rbd_dev->header.obj_order, | |
3911 | &rbd_dev->header.image_size); | |
3912 | } | |
3913 | ||
1e130199 AE |
3914 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3915 | { | |
3916 | void *reply_buf; | |
3917 | int ret; | |
3918 | void *p; | |
3919 | ||
3920 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3921 | if (!reply_buf) | |
3922 | return -ENOMEM; | |
3923 | ||
36be9a76 | 3924 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3925 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3926 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3927 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3928 | if (ret < 0) |
3929 | goto out; | |
3930 | ||
3931 | p = reply_buf; | |
3932 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3933 | p + ret, NULL, GFP_NOIO); |
3934 | ret = 0; | |
1e130199 AE |
3935 | |
3936 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3937 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3938 | rbd_dev->header.object_prefix = NULL; | |
3939 | } else { | |
3940 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3941 | } | |
1e130199 AE |
3942 | out: |
3943 | kfree(reply_buf); | |
3944 | ||
3945 | return ret; | |
3946 | } | |
3947 | ||
b1b5402a AE |
3948 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3949 | u64 *snap_features) | |
3950 | { | |
3951 | __le64 snapid = cpu_to_le64(snap_id); | |
3952 | struct { | |
3953 | __le64 features; | |
3954 | __le64 incompat; | |
4157976b | 3955 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3956 | u64 incompat; |
b1b5402a AE |
3957 | int ret; |
3958 | ||
36be9a76 | 3959 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3960 | "rbd", "get_features", |
4157976b | 3961 | &snapid, sizeof (snapid), |
e2a58ee5 | 3962 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3963 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3964 | if (ret < 0) |
3965 | return ret; | |
57385b51 AE |
3966 | if (ret < sizeof (features_buf)) |
3967 | return -ERANGE; | |
d889140c AE |
3968 | |
3969 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3970 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3971 | return -ENXIO; |
d889140c | 3972 | |
b1b5402a AE |
3973 | *snap_features = le64_to_cpu(features_buf.features); |
3974 | ||
3975 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3976 | (unsigned long long)snap_id, |
3977 | (unsigned long long)*snap_features, | |
3978 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3979 | |
3980 | return 0; | |
3981 | } | |
3982 | ||
3983 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3984 | { | |
3985 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3986 | &rbd_dev->header.features); | |
3987 | } | |
3988 | ||
86b00e0d AE |
3989 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3990 | { | |
3991 | struct rbd_spec *parent_spec; | |
3992 | size_t size; | |
3993 | void *reply_buf = NULL; | |
3994 | __le64 snapid; | |
3995 | void *p; | |
3996 | void *end; | |
642a2537 | 3997 | u64 pool_id; |
86b00e0d | 3998 | char *image_id; |
3b5cf2a2 | 3999 | u64 snap_id; |
86b00e0d | 4000 | u64 overlap; |
86b00e0d AE |
4001 | int ret; |
4002 | ||
4003 | parent_spec = rbd_spec_alloc(); | |
4004 | if (!parent_spec) | |
4005 | return -ENOMEM; | |
4006 | ||
4007 | size = sizeof (__le64) + /* pool_id */ | |
4008 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4009 | sizeof (__le64) + /* snap_id */ | |
4010 | sizeof (__le64); /* overlap */ | |
4011 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4012 | if (!reply_buf) { | |
4013 | ret = -ENOMEM; | |
4014 | goto out_err; | |
4015 | } | |
4016 | ||
4017 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 4018 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4019 | "rbd", "get_parent", |
4157976b | 4020 | &snapid, sizeof (snapid), |
e2a58ee5 | 4021 | reply_buf, size); |
36be9a76 | 4022 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4023 | if (ret < 0) |
4024 | goto out_err; | |
4025 | ||
86b00e0d | 4026 | p = reply_buf; |
57385b51 AE |
4027 | end = reply_buf + ret; |
4028 | ret = -ERANGE; | |
642a2537 | 4029 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4030 | if (pool_id == CEPH_NOPOOL) { |
4031 | /* | |
4032 | * Either the parent never existed, or we have | |
4033 | * record of it but the image got flattened so it no | |
4034 | * longer has a parent. When the parent of a | |
4035 | * layered image disappears we immediately set the | |
4036 | * overlap to 0. The effect of this is that all new | |
4037 | * requests will be treated as if the image had no | |
4038 | * parent. | |
4039 | */ | |
4040 | if (rbd_dev->parent_overlap) { | |
4041 | rbd_dev->parent_overlap = 0; | |
4042 | smp_mb(); | |
4043 | rbd_dev_parent_put(rbd_dev); | |
4044 | pr_info("%s: clone image has been flattened\n", | |
4045 | rbd_dev->disk->disk_name); | |
4046 | } | |
4047 | ||
86b00e0d | 4048 | goto out; /* No parent? No problem. */ |
392a9dad | 4049 | } |
86b00e0d | 4050 | |
0903e875 AE |
4051 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4052 | ||
4053 | ret = -EIO; | |
642a2537 | 4054 | if (pool_id > (u64)U32_MAX) { |
c0cd10db | 4055 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", |
642a2537 | 4056 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4057 | goto out_err; |
c0cd10db | 4058 | } |
0903e875 | 4059 | |
979ed480 | 4060 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4061 | if (IS_ERR(image_id)) { |
4062 | ret = PTR_ERR(image_id); | |
4063 | goto out_err; | |
4064 | } | |
3b5cf2a2 | 4065 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4066 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4067 | ||
3b5cf2a2 AE |
4068 | /* |
4069 | * The parent won't change (except when the clone is | |
4070 | * flattened, already handled that). So we only need to | |
4071 | * record the parent spec we have not already done so. | |
4072 | */ | |
4073 | if (!rbd_dev->parent_spec) { | |
4074 | parent_spec->pool_id = pool_id; | |
4075 | parent_spec->image_id = image_id; | |
4076 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4077 | rbd_dev->parent_spec = parent_spec; |
4078 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3b5cf2a2 AE |
4079 | } |
4080 | ||
4081 | /* | |
4082 | * We always update the parent overlap. If it's zero we | |
4083 | * treat it specially. | |
4084 | */ | |
4085 | rbd_dev->parent_overlap = overlap; | |
4086 | smp_mb(); | |
4087 | if (!overlap) { | |
4088 | ||
4089 | /* A null parent_spec indicates it's the initial probe */ | |
4090 | ||
4091 | if (parent_spec) { | |
4092 | /* | |
4093 | * The overlap has become zero, so the clone | |
4094 | * must have been resized down to 0 at some | |
4095 | * point. Treat this the same as a flatten. | |
4096 | */ | |
4097 | rbd_dev_parent_put(rbd_dev); | |
4098 | pr_info("%s: clone image now standalone\n", | |
4099 | rbd_dev->disk->disk_name); | |
4100 | } else { | |
4101 | /* | |
4102 | * For the initial probe, if we find the | |
4103 | * overlap is zero we just pretend there was | |
4104 | * no parent image. | |
4105 | */ | |
4106 | rbd_warn(rbd_dev, "ignoring parent of " | |
4107 | "clone with overlap 0\n"); | |
4108 | } | |
70cf49cf | 4109 | } |
86b00e0d AE |
4110 | out: |
4111 | ret = 0; | |
4112 | out_err: | |
4113 | kfree(reply_buf); | |
4114 | rbd_spec_put(parent_spec); | |
4115 | ||
4116 | return ret; | |
4117 | } | |
4118 | ||
cc070d59 AE |
4119 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4120 | { | |
4121 | struct { | |
4122 | __le64 stripe_unit; | |
4123 | __le64 stripe_count; | |
4124 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4125 | size_t size = sizeof (striping_info_buf); | |
4126 | void *p; | |
4127 | u64 obj_size; | |
4128 | u64 stripe_unit; | |
4129 | u64 stripe_count; | |
4130 | int ret; | |
4131 | ||
4132 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4133 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4134 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4135 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4136 | if (ret < 0) | |
4137 | return ret; | |
4138 | if (ret < size) | |
4139 | return -ERANGE; | |
4140 | ||
4141 | /* | |
4142 | * We don't actually support the "fancy striping" feature | |
4143 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4144 | * defaults the behavior is the same as before. So find | |
4145 | * out, and only fail if the image has non-default values. | |
4146 | */ | |
4147 | ret = -EINVAL; | |
4148 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4149 | p = &striping_info_buf; | |
4150 | stripe_unit = ceph_decode_64(&p); | |
4151 | if (stripe_unit != obj_size) { | |
4152 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4153 | "(got %llu want %llu)", | |
4154 | stripe_unit, obj_size); | |
4155 | return -EINVAL; | |
4156 | } | |
4157 | stripe_count = ceph_decode_64(&p); | |
4158 | if (stripe_count != 1) { | |
4159 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4160 | "(got %llu want 1)", stripe_count); | |
4161 | return -EINVAL; | |
4162 | } | |
500d0c0f AE |
4163 | rbd_dev->header.stripe_unit = stripe_unit; |
4164 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4165 | |
4166 | return 0; | |
4167 | } | |
4168 | ||
9e15b77d AE |
4169 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4170 | { | |
4171 | size_t image_id_size; | |
4172 | char *image_id; | |
4173 | void *p; | |
4174 | void *end; | |
4175 | size_t size; | |
4176 | void *reply_buf = NULL; | |
4177 | size_t len = 0; | |
4178 | char *image_name = NULL; | |
4179 | int ret; | |
4180 | ||
4181 | rbd_assert(!rbd_dev->spec->image_name); | |
4182 | ||
69e7a02f AE |
4183 | len = strlen(rbd_dev->spec->image_id); |
4184 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4185 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4186 | if (!image_id) | |
4187 | return NULL; | |
4188 | ||
4189 | p = image_id; | |
4157976b | 4190 | end = image_id + image_id_size; |
57385b51 | 4191 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4192 | |
4193 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4194 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4195 | if (!reply_buf) | |
4196 | goto out; | |
4197 | ||
36be9a76 | 4198 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4199 | "rbd", "dir_get_name", |
4200 | image_id, image_id_size, | |
e2a58ee5 | 4201 | reply_buf, size); |
9e15b77d AE |
4202 | if (ret < 0) |
4203 | goto out; | |
4204 | p = reply_buf; | |
f40eb349 AE |
4205 | end = reply_buf + ret; |
4206 | ||
9e15b77d AE |
4207 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4208 | if (IS_ERR(image_name)) | |
4209 | image_name = NULL; | |
4210 | else | |
4211 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4212 | out: | |
4213 | kfree(reply_buf); | |
4214 | kfree(image_id); | |
4215 | ||
4216 | return image_name; | |
4217 | } | |
4218 | ||
2ad3d716 AE |
4219 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4220 | { | |
4221 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4222 | const char *snap_name; | |
4223 | u32 which = 0; | |
4224 | ||
4225 | /* Skip over names until we find the one we are looking for */ | |
4226 | ||
4227 | snap_name = rbd_dev->header.snap_names; | |
4228 | while (which < snapc->num_snaps) { | |
4229 | if (!strcmp(name, snap_name)) | |
4230 | return snapc->snaps[which]; | |
4231 | snap_name += strlen(snap_name) + 1; | |
4232 | which++; | |
4233 | } | |
4234 | return CEPH_NOSNAP; | |
4235 | } | |
4236 | ||
4237 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4238 | { | |
4239 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4240 | u32 which; | |
4241 | bool found = false; | |
4242 | u64 snap_id; | |
4243 | ||
4244 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4245 | const char *snap_name; | |
4246 | ||
4247 | snap_id = snapc->snaps[which]; | |
4248 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4249 | if (IS_ERR(snap_name)) { |
4250 | /* ignore no-longer existing snapshots */ | |
4251 | if (PTR_ERR(snap_name) == -ENOENT) | |
4252 | continue; | |
4253 | else | |
4254 | break; | |
4255 | } | |
2ad3d716 AE |
4256 | found = !strcmp(name, snap_name); |
4257 | kfree(snap_name); | |
4258 | } | |
4259 | return found ? snap_id : CEPH_NOSNAP; | |
4260 | } | |
4261 | ||
4262 | /* | |
4263 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4264 | * no snapshot by that name is found, or if an error occurs. | |
4265 | */ | |
4266 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4267 | { | |
4268 | if (rbd_dev->image_format == 1) | |
4269 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4270 | ||
4271 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4272 | } | |
4273 | ||
9e15b77d | 4274 | /* |
2e9f7f1c AE |
4275 | * When an rbd image has a parent image, it is identified by the |
4276 | * pool, image, and snapshot ids (not names). This function fills | |
4277 | * in the names for those ids. (It's OK if we can't figure out the | |
4278 | * name for an image id, but the pool and snapshot ids should always | |
4279 | * exist and have names.) All names in an rbd spec are dynamically | |
4280 | * allocated. | |
e1d4213f AE |
4281 | * |
4282 | * When an image being mapped (not a parent) is probed, we have the | |
4283 | * pool name and pool id, image name and image id, and the snapshot | |
4284 | * name. The only thing we're missing is the snapshot id. | |
9e15b77d | 4285 | */ |
2e9f7f1c | 4286 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 4287 | { |
2e9f7f1c AE |
4288 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4289 | struct rbd_spec *spec = rbd_dev->spec; | |
4290 | const char *pool_name; | |
4291 | const char *image_name; | |
4292 | const char *snap_name; | |
9e15b77d AE |
4293 | int ret; |
4294 | ||
e1d4213f AE |
4295 | /* |
4296 | * An image being mapped will have the pool name (etc.), but | |
4297 | * we need to look up the snapshot id. | |
4298 | */ | |
2e9f7f1c AE |
4299 | if (spec->pool_name) { |
4300 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
2ad3d716 | 4301 | u64 snap_id; |
e1d4213f | 4302 | |
2ad3d716 AE |
4303 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); |
4304 | if (snap_id == CEPH_NOSNAP) | |
e1d4213f | 4305 | return -ENOENT; |
2ad3d716 | 4306 | spec->snap_id = snap_id; |
e1d4213f | 4307 | } else { |
2e9f7f1c | 4308 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
4309 | } |
4310 | ||
4311 | return 0; | |
4312 | } | |
9e15b77d | 4313 | |
2e9f7f1c | 4314 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4315 | |
2e9f7f1c AE |
4316 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4317 | if (!pool_name) { | |
4318 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4319 | return -EIO; |
4320 | } | |
2e9f7f1c AE |
4321 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4322 | if (!pool_name) | |
9e15b77d AE |
4323 | return -ENOMEM; |
4324 | ||
4325 | /* Fetch the image name; tolerate failure here */ | |
4326 | ||
2e9f7f1c AE |
4327 | image_name = rbd_dev_image_name(rbd_dev); |
4328 | if (!image_name) | |
06ecc6cb | 4329 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4330 | |
2e9f7f1c | 4331 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 4332 | |
2e9f7f1c | 4333 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4334 | if (IS_ERR(snap_name)) { |
4335 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4336 | goto out_err; |
2e9f7f1c AE |
4337 | } |
4338 | ||
4339 | spec->pool_name = pool_name; | |
4340 | spec->image_name = image_name; | |
4341 | spec->snap_name = snap_name; | |
9e15b77d AE |
4342 | |
4343 | return 0; | |
4344 | out_err: | |
2e9f7f1c AE |
4345 | kfree(image_name); |
4346 | kfree(pool_name); | |
9e15b77d AE |
4347 | |
4348 | return ret; | |
4349 | } | |
4350 | ||
cc4a38bd | 4351 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4352 | { |
4353 | size_t size; | |
4354 | int ret; | |
4355 | void *reply_buf; | |
4356 | void *p; | |
4357 | void *end; | |
4358 | u64 seq; | |
4359 | u32 snap_count; | |
4360 | struct ceph_snap_context *snapc; | |
4361 | u32 i; | |
4362 | ||
4363 | /* | |
4364 | * We'll need room for the seq value (maximum snapshot id), | |
4365 | * snapshot count, and array of that many snapshot ids. | |
4366 | * For now we have a fixed upper limit on the number we're | |
4367 | * prepared to receive. | |
4368 | */ | |
4369 | size = sizeof (__le64) + sizeof (__le32) + | |
4370 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4371 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4372 | if (!reply_buf) | |
4373 | return -ENOMEM; | |
4374 | ||
36be9a76 | 4375 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4376 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4377 | reply_buf, size); |
36be9a76 | 4378 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4379 | if (ret < 0) |
4380 | goto out; | |
4381 | ||
35d489f9 | 4382 | p = reply_buf; |
57385b51 AE |
4383 | end = reply_buf + ret; |
4384 | ret = -ERANGE; | |
35d489f9 AE |
4385 | ceph_decode_64_safe(&p, end, seq, out); |
4386 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4387 | ||
4388 | /* | |
4389 | * Make sure the reported number of snapshot ids wouldn't go | |
4390 | * beyond the end of our buffer. But before checking that, | |
4391 | * make sure the computed size of the snapshot context we | |
4392 | * allocate is representable in a size_t. | |
4393 | */ | |
4394 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4395 | / sizeof (u64)) { | |
4396 | ret = -EINVAL; | |
4397 | goto out; | |
4398 | } | |
4399 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4400 | goto out; | |
468521c1 | 4401 | ret = 0; |
35d489f9 | 4402 | |
812164f8 | 4403 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4404 | if (!snapc) { |
4405 | ret = -ENOMEM; | |
4406 | goto out; | |
4407 | } | |
35d489f9 | 4408 | snapc->seq = seq; |
35d489f9 AE |
4409 | for (i = 0; i < snap_count; i++) |
4410 | snapc->snaps[i] = ceph_decode_64(&p); | |
4411 | ||
49ece554 | 4412 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4413 | rbd_dev->header.snapc = snapc; |
4414 | ||
4415 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4416 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4417 | out: |
4418 | kfree(reply_buf); | |
4419 | ||
57385b51 | 4420 | return ret; |
35d489f9 AE |
4421 | } |
4422 | ||
54cac61f AE |
4423 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4424 | u64 snap_id) | |
b8b1e2db AE |
4425 | { |
4426 | size_t size; | |
4427 | void *reply_buf; | |
54cac61f | 4428 | __le64 snapid; |
b8b1e2db AE |
4429 | int ret; |
4430 | void *p; | |
4431 | void *end; | |
b8b1e2db AE |
4432 | char *snap_name; |
4433 | ||
4434 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4435 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4436 | if (!reply_buf) | |
4437 | return ERR_PTR(-ENOMEM); | |
4438 | ||
54cac61f | 4439 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4440 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4441 | "rbd", "get_snapshot_name", |
54cac61f | 4442 | &snapid, sizeof (snapid), |
e2a58ee5 | 4443 | reply_buf, size); |
36be9a76 | 4444 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4445 | if (ret < 0) { |
4446 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4447 | goto out; |
f40eb349 | 4448 | } |
b8b1e2db AE |
4449 | |
4450 | p = reply_buf; | |
f40eb349 | 4451 | end = reply_buf + ret; |
e5c35534 | 4452 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4453 | if (IS_ERR(snap_name)) |
b8b1e2db | 4454 | goto out; |
b8b1e2db | 4455 | |
f40eb349 | 4456 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4457 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4458 | out: |
4459 | kfree(reply_buf); | |
4460 | ||
f40eb349 | 4461 | return snap_name; |
b8b1e2db AE |
4462 | } |
4463 | ||
2df3fac7 | 4464 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4465 | { |
2df3fac7 | 4466 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4467 | int ret; |
117973fb | 4468 | |
1617e40c JD |
4469 | ret = rbd_dev_v2_image_size(rbd_dev); |
4470 | if (ret) | |
cfbf6377 | 4471 | return ret; |
1617e40c | 4472 | |
2df3fac7 AE |
4473 | if (first_time) { |
4474 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4475 | if (ret) | |
cfbf6377 | 4476 | return ret; |
2df3fac7 AE |
4477 | } |
4478 | ||
642a2537 AE |
4479 | /* |
4480 | * If the image supports layering, get the parent info. We | |
4481 | * need to probe the first time regardless. Thereafter we | |
4482 | * only need to if there's a parent, to see if it has | |
4483 | * disappeared due to the mapped image getting flattened. | |
4484 | */ | |
4485 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING && | |
4486 | (first_time || rbd_dev->parent_spec)) { | |
4487 | bool warn; | |
4488 | ||
4489 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4490 | if (ret) | |
cfbf6377 | 4491 | return ret; |
642a2537 AE |
4492 | |
4493 | /* | |
4494 | * Print a warning if this is the initial probe and | |
4495 | * the image has a parent. Don't print it if the | |
4496 | * image now being probed is itself a parent. We | |
4497 | * can tell at this point because we won't know its | |
4498 | * pool name yet (just its pool id). | |
4499 | */ | |
4500 | warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name; | |
4501 | if (first_time && warn) | |
4502 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4503 | "is EXPERIMENTAL!"); | |
4504 | } | |
4505 | ||
29334ba4 AE |
4506 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) |
4507 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
4508 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
117973fb | 4509 | |
cc4a38bd | 4510 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4511 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4512 | |
4513 | return ret; | |
4514 | } | |
4515 | ||
dfc5606d YS |
4516 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4517 | { | |
dfc5606d | 4518 | struct device *dev; |
cd789ab9 | 4519 | int ret; |
dfc5606d | 4520 | |
cd789ab9 | 4521 | dev = &rbd_dev->dev; |
dfc5606d YS |
4522 | dev->bus = &rbd_bus_type; |
4523 | dev->type = &rbd_device_type; | |
4524 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4525 | dev->release = rbd_dev_device_release; |
de71a297 | 4526 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4527 | ret = device_register(dev); |
dfc5606d | 4528 | |
dfc5606d | 4529 | return ret; |
602adf40 YS |
4530 | } |
4531 | ||
dfc5606d YS |
4532 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4533 | { | |
4534 | device_unregister(&rbd_dev->dev); | |
4535 | } | |
4536 | ||
1ddbe94e | 4537 | /* |
499afd5b | 4538 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4539 | * the rbd_dev to the global list. |
1ddbe94e | 4540 | */ |
f8a22fc2 | 4541 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4542 | { |
f8a22fc2 ID |
4543 | int new_dev_id; |
4544 | ||
9b60e70b ID |
4545 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4546 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4547 | GFP_KERNEL); | |
f8a22fc2 ID |
4548 | if (new_dev_id < 0) |
4549 | return new_dev_id; | |
4550 | ||
4551 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4552 | |
4553 | spin_lock(&rbd_dev_list_lock); | |
4554 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4555 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4556 | |
70eebd20 | 4557 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4558 | |
4559 | return 0; | |
1ddbe94e | 4560 | } |
b7f23c36 | 4561 | |
1ddbe94e | 4562 | /* |
499afd5b AE |
4563 | * Remove an rbd_dev from the global list, and record that its |
4564 | * identifier is no longer in use. | |
1ddbe94e | 4565 | */ |
e2839308 | 4566 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4567 | { |
499afd5b AE |
4568 | spin_lock(&rbd_dev_list_lock); |
4569 | list_del_init(&rbd_dev->node); | |
4570 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4571 | |
f8a22fc2 ID |
4572 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4573 | ||
4574 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4575 | } |
4576 | ||
e28fff26 AE |
4577 | /* |
4578 | * Skips over white space at *buf, and updates *buf to point to the | |
4579 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4580 | * the token (string of non-white space characters) found. Note |
4581 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4582 | */ |
4583 | static inline size_t next_token(const char **buf) | |
4584 | { | |
4585 | /* | |
4586 | * These are the characters that produce nonzero for | |
4587 | * isspace() in the "C" and "POSIX" locales. | |
4588 | */ | |
4589 | const char *spaces = " \f\n\r\t\v"; | |
4590 | ||
4591 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4592 | ||
4593 | return strcspn(*buf, spaces); /* Return token length */ | |
4594 | } | |
4595 | ||
4596 | /* | |
4597 | * Finds the next token in *buf, and if the provided token buffer is | |
4598 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4599 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4600 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4601 | * |
4602 | * Returns the length of the token found (not including the '\0'). | |
4603 | * Return value will be 0 if no token is found, and it will be >= | |
4604 | * token_size if the token would not fit. | |
4605 | * | |
593a9e7b | 4606 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4607 | * found token. Note that this occurs even if the token buffer is |
4608 | * too small to hold it. | |
4609 | */ | |
4610 | static inline size_t copy_token(const char **buf, | |
4611 | char *token, | |
4612 | size_t token_size) | |
4613 | { | |
4614 | size_t len; | |
4615 | ||
4616 | len = next_token(buf); | |
4617 | if (len < token_size) { | |
4618 | memcpy(token, *buf, len); | |
4619 | *(token + len) = '\0'; | |
4620 | } | |
4621 | *buf += len; | |
4622 | ||
4623 | return len; | |
4624 | } | |
4625 | ||
ea3352f4 AE |
4626 | /* |
4627 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4628 | * enough to hold a copy of it, and copies the token into the new | |
4629 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4630 | * that a duplicate buffer is created even for a zero-length token. | |
4631 | * | |
4632 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4633 | * pointer if memory for the duplicate was not available. If | |
4634 | * the lenp argument is a non-null pointer, the length of the token | |
4635 | * (not including the '\0') is returned in *lenp. | |
4636 | * | |
4637 | * If successful, the *buf pointer will be updated to point beyond | |
4638 | * the end of the found token. | |
4639 | * | |
4640 | * Note: uses GFP_KERNEL for allocation. | |
4641 | */ | |
4642 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4643 | { | |
4644 | char *dup; | |
4645 | size_t len; | |
4646 | ||
4647 | len = next_token(buf); | |
4caf35f9 | 4648 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4649 | if (!dup) |
4650 | return NULL; | |
ea3352f4 AE |
4651 | *(dup + len) = '\0'; |
4652 | *buf += len; | |
4653 | ||
4654 | if (lenp) | |
4655 | *lenp = len; | |
4656 | ||
4657 | return dup; | |
4658 | } | |
4659 | ||
a725f65e | 4660 | /* |
859c31df AE |
4661 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4662 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4663 | * and the data written is passed here via a NUL-terminated buffer. | |
4664 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4665 | * |
859c31df AE |
4666 | * The information extracted from these options is recorded in |
4667 | * the other parameters which return dynamically-allocated | |
4668 | * structures: | |
4669 | * ceph_opts | |
4670 | * The address of a pointer that will refer to a ceph options | |
4671 | * structure. Caller must release the returned pointer using | |
4672 | * ceph_destroy_options() when it is no longer needed. | |
4673 | * rbd_opts | |
4674 | * Address of an rbd options pointer. Fully initialized by | |
4675 | * this function; caller must release with kfree(). | |
4676 | * spec | |
4677 | * Address of an rbd image specification pointer. Fully | |
4678 | * initialized by this function based on parsed options. | |
4679 | * Caller must release with rbd_spec_put(). | |
4680 | * | |
4681 | * The options passed take this form: | |
4682 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4683 | * where: | |
4684 | * <mon_addrs> | |
4685 | * A comma-separated list of one or more monitor addresses. | |
4686 | * A monitor address is an ip address, optionally followed | |
4687 | * by a port number (separated by a colon). | |
4688 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4689 | * <options> | |
4690 | * A comma-separated list of ceph and/or rbd options. | |
4691 | * <pool_name> | |
4692 | * The name of the rados pool containing the rbd image. | |
4693 | * <image_name> | |
4694 | * The name of the image in that pool to map. | |
4695 | * <snap_id> | |
4696 | * An optional snapshot id. If provided, the mapping will | |
4697 | * present data from the image at the time that snapshot was | |
4698 | * created. The image head is used if no snapshot id is | |
4699 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4700 | */ |
859c31df | 4701 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4702 | struct ceph_options **ceph_opts, |
859c31df AE |
4703 | struct rbd_options **opts, |
4704 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4705 | { |
d22f76e7 | 4706 | size_t len; |
859c31df | 4707 | char *options; |
0ddebc0c | 4708 | const char *mon_addrs; |
ecb4dc22 | 4709 | char *snap_name; |
0ddebc0c | 4710 | size_t mon_addrs_size; |
859c31df | 4711 | struct rbd_spec *spec = NULL; |
4e9afeba | 4712 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4713 | struct ceph_options *copts; |
dc79b113 | 4714 | int ret; |
e28fff26 AE |
4715 | |
4716 | /* The first four tokens are required */ | |
4717 | ||
7ef3214a | 4718 | len = next_token(&buf); |
4fb5d671 AE |
4719 | if (!len) { |
4720 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4721 | return -EINVAL; | |
4722 | } | |
0ddebc0c | 4723 | mon_addrs = buf; |
f28e565a | 4724 | mon_addrs_size = len + 1; |
7ef3214a | 4725 | buf += len; |
a725f65e | 4726 | |
dc79b113 | 4727 | ret = -EINVAL; |
f28e565a AE |
4728 | options = dup_token(&buf, NULL); |
4729 | if (!options) | |
dc79b113 | 4730 | return -ENOMEM; |
4fb5d671 AE |
4731 | if (!*options) { |
4732 | rbd_warn(NULL, "no options provided"); | |
4733 | goto out_err; | |
4734 | } | |
e28fff26 | 4735 | |
859c31df AE |
4736 | spec = rbd_spec_alloc(); |
4737 | if (!spec) | |
f28e565a | 4738 | goto out_mem; |
859c31df AE |
4739 | |
4740 | spec->pool_name = dup_token(&buf, NULL); | |
4741 | if (!spec->pool_name) | |
4742 | goto out_mem; | |
4fb5d671 AE |
4743 | if (!*spec->pool_name) { |
4744 | rbd_warn(NULL, "no pool name provided"); | |
4745 | goto out_err; | |
4746 | } | |
e28fff26 | 4747 | |
69e7a02f | 4748 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4749 | if (!spec->image_name) |
f28e565a | 4750 | goto out_mem; |
4fb5d671 AE |
4751 | if (!*spec->image_name) { |
4752 | rbd_warn(NULL, "no image name provided"); | |
4753 | goto out_err; | |
4754 | } | |
d4b125e9 | 4755 | |
f28e565a AE |
4756 | /* |
4757 | * Snapshot name is optional; default is to use "-" | |
4758 | * (indicating the head/no snapshot). | |
4759 | */ | |
3feeb894 | 4760 | len = next_token(&buf); |
820a5f3e | 4761 | if (!len) { |
3feeb894 AE |
4762 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4763 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4764 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4765 | ret = -ENAMETOOLONG; |
f28e565a | 4766 | goto out_err; |
849b4260 | 4767 | } |
ecb4dc22 AE |
4768 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4769 | if (!snap_name) | |
f28e565a | 4770 | goto out_mem; |
ecb4dc22 AE |
4771 | *(snap_name + len) = '\0'; |
4772 | spec->snap_name = snap_name; | |
e5c35534 | 4773 | |
0ddebc0c | 4774 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4775 | |
4e9afeba AE |
4776 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4777 | if (!rbd_opts) | |
4778 | goto out_mem; | |
4779 | ||
4780 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4781 | |
859c31df | 4782 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4783 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4784 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4785 | if (IS_ERR(copts)) { |
4786 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4787 | goto out_err; |
4788 | } | |
859c31df AE |
4789 | kfree(options); |
4790 | ||
4791 | *ceph_opts = copts; | |
4e9afeba | 4792 | *opts = rbd_opts; |
859c31df | 4793 | *rbd_spec = spec; |
0ddebc0c | 4794 | |
dc79b113 | 4795 | return 0; |
f28e565a | 4796 | out_mem: |
dc79b113 | 4797 | ret = -ENOMEM; |
d22f76e7 | 4798 | out_err: |
859c31df AE |
4799 | kfree(rbd_opts); |
4800 | rbd_spec_put(spec); | |
f28e565a | 4801 | kfree(options); |
d22f76e7 | 4802 | |
dc79b113 | 4803 | return ret; |
a725f65e AE |
4804 | } |
4805 | ||
30ba1f02 ID |
4806 | /* |
4807 | * Return pool id (>= 0) or a negative error code. | |
4808 | */ | |
4809 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4810 | { | |
4811 | u64 newest_epoch; | |
4812 | unsigned long timeout = rbdc->client->options->mount_timeout * HZ; | |
4813 | int tries = 0; | |
4814 | int ret; | |
4815 | ||
4816 | again: | |
4817 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4818 | if (ret == -ENOENT && tries++ < 1) { | |
4819 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4820 | &newest_epoch); | |
4821 | if (ret < 0) | |
4822 | return ret; | |
4823 | ||
4824 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4825 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4826 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
4827 | newest_epoch, timeout); | |
4828 | goto again; | |
4829 | } else { | |
4830 | /* the osdmap we have is new enough */ | |
4831 | return -ENOENT; | |
4832 | } | |
4833 | } | |
4834 | ||
4835 | return ret; | |
4836 | } | |
4837 | ||
589d30e0 AE |
4838 | /* |
4839 | * An rbd format 2 image has a unique identifier, distinct from the | |
4840 | * name given to it by the user. Internally, that identifier is | |
4841 | * what's used to specify the names of objects related to the image. | |
4842 | * | |
4843 | * A special "rbd id" object is used to map an rbd image name to its | |
4844 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4845 | * with the supplied name. | |
4846 | * | |
4847 | * This function will record the given rbd_dev's image_id field if | |
4848 | * it can be determined, and in that case will return 0. If any | |
4849 | * errors occur a negative errno will be returned and the rbd_dev's | |
4850 | * image_id field will be unchanged (and should be NULL). | |
4851 | */ | |
4852 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4853 | { | |
4854 | int ret; | |
4855 | size_t size; | |
4856 | char *object_name; | |
4857 | void *response; | |
c0fba368 | 4858 | char *image_id; |
2f82ee54 | 4859 | |
2c0d0a10 AE |
4860 | /* |
4861 | * When probing a parent image, the image id is already | |
4862 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4863 | * need to fetch the image id again in this case. We |
4864 | * do still need to set the image format though. | |
2c0d0a10 | 4865 | */ |
c0fba368 AE |
4866 | if (rbd_dev->spec->image_id) { |
4867 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4868 | ||
2c0d0a10 | 4869 | return 0; |
c0fba368 | 4870 | } |
2c0d0a10 | 4871 | |
589d30e0 AE |
4872 | /* |
4873 | * First, see if the format 2 image id file exists, and if | |
4874 | * so, get the image's persistent id from it. | |
4875 | */ | |
69e7a02f | 4876 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4877 | object_name = kmalloc(size, GFP_NOIO); |
4878 | if (!object_name) | |
4879 | return -ENOMEM; | |
0d7dbfce | 4880 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4881 | dout("rbd id object name is %s\n", object_name); |
4882 | ||
4883 | /* Response will be an encoded string, which includes a length */ | |
4884 | ||
4885 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4886 | response = kzalloc(size, GFP_NOIO); | |
4887 | if (!response) { | |
4888 | ret = -ENOMEM; | |
4889 | goto out; | |
4890 | } | |
4891 | ||
c0fba368 AE |
4892 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4893 | ||
36be9a76 | 4894 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4895 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4896 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4897 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4898 | if (ret == -ENOENT) { |
4899 | image_id = kstrdup("", GFP_KERNEL); | |
4900 | ret = image_id ? 0 : -ENOMEM; | |
4901 | if (!ret) | |
4902 | rbd_dev->image_format = 1; | |
4903 | } else if (ret > sizeof (__le32)) { | |
4904 | void *p = response; | |
4905 | ||
4906 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4907 | NULL, GFP_NOIO); |
461f758a | 4908 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
4909 | if (!ret) |
4910 | rbd_dev->image_format = 2; | |
589d30e0 | 4911 | } else { |
c0fba368 AE |
4912 | ret = -EINVAL; |
4913 | } | |
4914 | ||
4915 | if (!ret) { | |
4916 | rbd_dev->spec->image_id = image_id; | |
4917 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4918 | } |
4919 | out: | |
4920 | kfree(response); | |
4921 | kfree(object_name); | |
4922 | ||
4923 | return ret; | |
4924 | } | |
4925 | ||
3abef3b3 AE |
4926 | /* |
4927 | * Undo whatever state changes are made by v1 or v2 header info | |
4928 | * call. | |
4929 | */ | |
6fd48b3b AE |
4930 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
4931 | { | |
4932 | struct rbd_image_header *header; | |
4933 | ||
392a9dad AE |
4934 | /* Drop parent reference unless it's already been done (or none) */ |
4935 | ||
4936 | if (rbd_dev->parent_overlap) | |
4937 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
4938 | |
4939 | /* Free dynamic fields from the header, then zero it out */ | |
4940 | ||
4941 | header = &rbd_dev->header; | |
812164f8 | 4942 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4943 | kfree(header->snap_sizes); |
4944 | kfree(header->snap_names); | |
4945 | kfree(header->object_prefix); | |
4946 | memset(header, 0, sizeof (*header)); | |
4947 | } | |
4948 | ||
2df3fac7 | 4949 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4950 | { |
4951 | int ret; | |
a30b71b9 | 4952 | |
1e130199 | 4953 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 4954 | if (ret) |
b1b5402a AE |
4955 | goto out_err; |
4956 | ||
2df3fac7 AE |
4957 | /* |
4958 | * Get the and check features for the image. Currently the | |
4959 | * features are assumed to never change. | |
4960 | */ | |
b1b5402a | 4961 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 4962 | if (ret) |
9d475de5 | 4963 | goto out_err; |
35d489f9 | 4964 | |
cc070d59 AE |
4965 | /* If the image supports fancy striping, get its parameters */ |
4966 | ||
4967 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4968 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4969 | if (ret < 0) | |
4970 | goto out_err; | |
4971 | } | |
2df3fac7 | 4972 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 4973 | |
35152979 | 4974 | return 0; |
9d475de5 | 4975 | out_err: |
642a2537 | 4976 | rbd_dev->header.features = 0; |
1e130199 AE |
4977 | kfree(rbd_dev->header.object_prefix); |
4978 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4979 | |
4980 | return ret; | |
a30b71b9 AE |
4981 | } |
4982 | ||
124afba2 | 4983 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4984 | { |
2f82ee54 | 4985 | struct rbd_device *parent = NULL; |
124afba2 AE |
4986 | struct rbd_spec *parent_spec; |
4987 | struct rbd_client *rbdc; | |
4988 | int ret; | |
4989 | ||
4990 | if (!rbd_dev->parent_spec) | |
4991 | return 0; | |
4992 | /* | |
4993 | * We need to pass a reference to the client and the parent | |
4994 | * spec when creating the parent rbd_dev. Images related by | |
4995 | * parent/child relationships always share both. | |
4996 | */ | |
4997 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4998 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4999 | ||
5000 | ret = -ENOMEM; | |
5001 | parent = rbd_dev_create(rbdc, parent_spec); | |
5002 | if (!parent) | |
5003 | goto out_err; | |
5004 | ||
1f3ef788 | 5005 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
5006 | if (ret < 0) |
5007 | goto out_err; | |
5008 | rbd_dev->parent = parent; | |
a2acd00e | 5009 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
5010 | |
5011 | return 0; | |
5012 | out_err: | |
5013 | if (parent) { | |
fb65d228 | 5014 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
5015 | kfree(rbd_dev->header_name); |
5016 | rbd_dev_destroy(parent); | |
5017 | } else { | |
5018 | rbd_put_client(rbdc); | |
5019 | rbd_spec_put(parent_spec); | |
5020 | } | |
5021 | ||
5022 | return ret; | |
5023 | } | |
5024 | ||
200a6a8b | 5025 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5026 | { |
83a06263 | 5027 | int ret; |
d1cf5788 | 5028 | |
f8a22fc2 ID |
5029 | /* Get an id and fill in device name. */ |
5030 | ||
5031 | ret = rbd_dev_id_get(rbd_dev); | |
5032 | if (ret) | |
5033 | return ret; | |
83a06263 | 5034 | |
83a06263 AE |
5035 | BUILD_BUG_ON(DEV_NAME_LEN |
5036 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5037 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5038 | ||
9b60e70b | 5039 | /* Record our major and minor device numbers. */ |
83a06263 | 5040 | |
9b60e70b ID |
5041 | if (!single_major) { |
5042 | ret = register_blkdev(0, rbd_dev->name); | |
5043 | if (ret < 0) | |
5044 | goto err_out_id; | |
5045 | ||
5046 | rbd_dev->major = ret; | |
5047 | rbd_dev->minor = 0; | |
5048 | } else { | |
5049 | rbd_dev->major = rbd_major; | |
5050 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5051 | } | |
83a06263 AE |
5052 | |
5053 | /* Set up the blkdev mapping. */ | |
5054 | ||
5055 | ret = rbd_init_disk(rbd_dev); | |
5056 | if (ret) | |
5057 | goto err_out_blkdev; | |
5058 | ||
f35a4dee | 5059 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5060 | if (ret) |
5061 | goto err_out_disk; | |
f35a4dee AE |
5062 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
5063 | ||
5064 | ret = rbd_bus_add_dev(rbd_dev); | |
5065 | if (ret) | |
5066 | goto err_out_mapping; | |
83a06263 | 5067 | |
83a06263 AE |
5068 | /* Everything's ready. Announce the disk to the world. */ |
5069 | ||
129b79d4 | 5070 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5071 | add_disk(rbd_dev->disk); |
5072 | ||
5073 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5074 | (unsigned long long) rbd_dev->mapping.size); | |
5075 | ||
5076 | return ret; | |
2f82ee54 | 5077 | |
f35a4dee AE |
5078 | err_out_mapping: |
5079 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5080 | err_out_disk: |
5081 | rbd_free_disk(rbd_dev); | |
5082 | err_out_blkdev: | |
9b60e70b ID |
5083 | if (!single_major) |
5084 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5085 | err_out_id: |
5086 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 5087 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
5088 | |
5089 | return ret; | |
5090 | } | |
5091 | ||
332bb12d AE |
5092 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5093 | { | |
5094 | struct rbd_spec *spec = rbd_dev->spec; | |
5095 | size_t size; | |
5096 | ||
5097 | /* Record the header object name for this rbd image. */ | |
5098 | ||
5099 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5100 | ||
5101 | if (rbd_dev->image_format == 1) | |
5102 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5103 | else | |
5104 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5105 | ||
5106 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5107 | if (!rbd_dev->header_name) | |
5108 | return -ENOMEM; | |
5109 | ||
5110 | if (rbd_dev->image_format == 1) | |
5111 | sprintf(rbd_dev->header_name, "%s%s", | |
5112 | spec->image_name, RBD_SUFFIX); | |
5113 | else | |
5114 | sprintf(rbd_dev->header_name, "%s%s", | |
5115 | RBD_HEADER_PREFIX, spec->image_id); | |
5116 | return 0; | |
5117 | } | |
5118 | ||
200a6a8b AE |
5119 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5120 | { | |
6fd48b3b | 5121 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5122 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5123 | rbd_dev->header_name = NULL; |
5124 | rbd_dev->image_format = 0; | |
5125 | kfree(rbd_dev->spec->image_id); | |
5126 | rbd_dev->spec->image_id = NULL; | |
5127 | ||
200a6a8b AE |
5128 | rbd_dev_destroy(rbd_dev); |
5129 | } | |
5130 | ||
a30b71b9 AE |
5131 | /* |
5132 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5133 | * device. If this image is the one being mapped (i.e., not a |
5134 | * parent), initiate a watch on its header object before using that | |
5135 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5136 | */ |
1f3ef788 | 5137 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
5138 | { |
5139 | int ret; | |
5140 | ||
5141 | /* | |
3abef3b3 AE |
5142 | * Get the id from the image id object. Unless there's an |
5143 | * error, rbd_dev->spec->image_id will be filled in with | |
5144 | * a dynamically-allocated string, and rbd_dev->image_format | |
5145 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5146 | */ |
5147 | ret = rbd_dev_image_id(rbd_dev); | |
5148 | if (ret) | |
c0fba368 AE |
5149 | return ret; |
5150 | rbd_assert(rbd_dev->spec->image_id); | |
5151 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5152 | ||
332bb12d AE |
5153 | ret = rbd_dev_header_name(rbd_dev); |
5154 | if (ret) | |
5155 | goto err_out_format; | |
5156 | ||
1f3ef788 | 5157 | if (mapping) { |
fca27065 | 5158 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1f3ef788 AE |
5159 | if (ret) |
5160 | goto out_header_name; | |
5161 | } | |
b644de2b | 5162 | |
c0fba368 | 5163 | if (rbd_dev->image_format == 1) |
99a41ebc | 5164 | ret = rbd_dev_v1_header_info(rbd_dev); |
a30b71b9 | 5165 | else |
2df3fac7 | 5166 | ret = rbd_dev_v2_header_info(rbd_dev); |
5655c4d9 | 5167 | if (ret) |
b644de2b | 5168 | goto err_out_watch; |
83a06263 | 5169 | |
9bb81c9b AE |
5170 | ret = rbd_dev_spec_update(rbd_dev); |
5171 | if (ret) | |
33dca39f | 5172 | goto err_out_probe; |
9bb81c9b AE |
5173 | |
5174 | ret = rbd_dev_probe_parent(rbd_dev); | |
30d60ba2 AE |
5175 | if (ret) |
5176 | goto err_out_probe; | |
5177 | ||
5178 | dout("discovered format %u image, header name is %s\n", | |
5179 | rbd_dev->image_format, rbd_dev->header_name); | |
83a06263 | 5180 | |
30d60ba2 | 5181 | return 0; |
6fd48b3b AE |
5182 | err_out_probe: |
5183 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5184 | err_out_watch: |
fca27065 ID |
5185 | if (mapping) |
5186 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5187 | out_header_name: |
5188 | kfree(rbd_dev->header_name); | |
5189 | rbd_dev->header_name = NULL; | |
5190 | err_out_format: | |
5191 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5192 | kfree(rbd_dev->spec->image_id); |
5193 | rbd_dev->spec->image_id = NULL; | |
5194 | ||
5195 | dout("probe failed, returning %d\n", ret); | |
5196 | ||
a30b71b9 AE |
5197 | return ret; |
5198 | } | |
5199 | ||
9b60e70b ID |
5200 | static ssize_t do_rbd_add(struct bus_type *bus, |
5201 | const char *buf, | |
5202 | size_t count) | |
602adf40 | 5203 | { |
cb8627c7 | 5204 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5205 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5206 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5207 | struct rbd_spec *spec = NULL; |
9d3997fd | 5208 | struct rbd_client *rbdc; |
51344a38 | 5209 | bool read_only; |
27cc2594 | 5210 | int rc = -ENOMEM; |
602adf40 YS |
5211 | |
5212 | if (!try_module_get(THIS_MODULE)) | |
5213 | return -ENODEV; | |
5214 | ||
602adf40 | 5215 | /* parse add command */ |
859c31df | 5216 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5217 | if (rc < 0) |
bd4ba655 | 5218 | goto err_out_module; |
51344a38 AE |
5219 | read_only = rbd_opts->read_only; |
5220 | kfree(rbd_opts); | |
5221 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5222 | |
9d3997fd AE |
5223 | rbdc = rbd_get_client(ceph_opts); |
5224 | if (IS_ERR(rbdc)) { | |
5225 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5226 | goto err_out_args; |
9d3997fd | 5227 | } |
602adf40 | 5228 | |
602adf40 | 5229 | /* pick the pool */ |
30ba1f02 | 5230 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
602adf40 YS |
5231 | if (rc < 0) |
5232 | goto err_out_client; | |
c0cd10db | 5233 | spec->pool_id = (u64)rc; |
859c31df | 5234 | |
0903e875 AE |
5235 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5236 | ||
c0cd10db AE |
5237 | if (spec->pool_id > (u64)U32_MAX) { |
5238 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
5239 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
5240 | rc = -EIO; |
5241 | goto err_out_client; | |
5242 | } | |
5243 | ||
c53d5893 | 5244 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5245 | if (!rbd_dev) |
5246 | goto err_out_client; | |
c53d5893 AE |
5247 | rbdc = NULL; /* rbd_dev now owns this */ |
5248 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5249 | |
1f3ef788 | 5250 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5251 | if (rc < 0) |
c53d5893 | 5252 | goto err_out_rbd_dev; |
05fd6f6f | 5253 | |
7ce4eef7 AE |
5254 | /* If we are mapping a snapshot it must be marked read-only */ |
5255 | ||
5256 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5257 | read_only = true; | |
5258 | rbd_dev->mapping.read_only = read_only; | |
5259 | ||
b536f69a | 5260 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5261 | if (rc) { |
e37180c0 ID |
5262 | /* |
5263 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5264 | * rbd_dev_image_release() without refactoring, see | |
5265 | * commit 1f3ef78861ac. | |
5266 | */ | |
5267 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5268 | rbd_dev_image_release(rbd_dev); |
5269 | goto err_out_module; | |
5270 | } | |
5271 | ||
5272 | return count; | |
b536f69a | 5273 | |
c53d5893 AE |
5274 | err_out_rbd_dev: |
5275 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5276 | err_out_client: |
9d3997fd | 5277 | rbd_put_client(rbdc); |
0ddebc0c | 5278 | err_out_args: |
859c31df | 5279 | rbd_spec_put(spec); |
bd4ba655 AE |
5280 | err_out_module: |
5281 | module_put(THIS_MODULE); | |
27cc2594 | 5282 | |
602adf40 | 5283 | dout("Error adding device %s\n", buf); |
27cc2594 | 5284 | |
c0cd10db | 5285 | return (ssize_t)rc; |
602adf40 YS |
5286 | } |
5287 | ||
9b60e70b ID |
5288 | static ssize_t rbd_add(struct bus_type *bus, |
5289 | const char *buf, | |
5290 | size_t count) | |
5291 | { | |
5292 | if (single_major) | |
5293 | return -EINVAL; | |
5294 | ||
5295 | return do_rbd_add(bus, buf, count); | |
5296 | } | |
5297 | ||
5298 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5299 | const char *buf, | |
5300 | size_t count) | |
5301 | { | |
5302 | return do_rbd_add(bus, buf, count); | |
5303 | } | |
5304 | ||
200a6a8b | 5305 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5306 | { |
593a9e7b | 5307 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5308 | |
602adf40 | 5309 | rbd_free_disk(rbd_dev); |
200a6a8b | 5310 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5311 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5312 | if (!single_major) |
5313 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5314 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5315 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5316 | } |
5317 | ||
05a46afd AE |
5318 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5319 | { | |
ad945fc1 | 5320 | while (rbd_dev->parent) { |
05a46afd AE |
5321 | struct rbd_device *first = rbd_dev; |
5322 | struct rbd_device *second = first->parent; | |
5323 | struct rbd_device *third; | |
5324 | ||
5325 | /* | |
5326 | * Follow to the parent with no grandparent and | |
5327 | * remove it. | |
5328 | */ | |
5329 | while (second && (third = second->parent)) { | |
5330 | first = second; | |
5331 | second = third; | |
5332 | } | |
ad945fc1 | 5333 | rbd_assert(second); |
8ad42cd0 | 5334 | rbd_dev_image_release(second); |
ad945fc1 AE |
5335 | first->parent = NULL; |
5336 | first->parent_overlap = 0; | |
5337 | ||
5338 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5339 | rbd_spec_put(first->parent_spec); |
5340 | first->parent_spec = NULL; | |
05a46afd AE |
5341 | } |
5342 | } | |
5343 | ||
9b60e70b ID |
5344 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5345 | const char *buf, | |
5346 | size_t count) | |
602adf40 YS |
5347 | { |
5348 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5349 | struct list_head *tmp; |
5350 | int dev_id; | |
602adf40 | 5351 | unsigned long ul; |
82a442d2 | 5352 | bool already = false; |
0d8189e1 | 5353 | int ret; |
602adf40 | 5354 | |
bb8e0e84 | 5355 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5356 | if (ret) |
5357 | return ret; | |
602adf40 YS |
5358 | |
5359 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5360 | dev_id = (int)ul; |
5361 | if (dev_id != ul) | |
602adf40 YS |
5362 | return -EINVAL; |
5363 | ||
751cc0e3 AE |
5364 | ret = -ENOENT; |
5365 | spin_lock(&rbd_dev_list_lock); | |
5366 | list_for_each(tmp, &rbd_dev_list) { | |
5367 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5368 | if (rbd_dev->dev_id == dev_id) { | |
5369 | ret = 0; | |
5370 | break; | |
5371 | } | |
42382b70 | 5372 | } |
751cc0e3 AE |
5373 | if (!ret) { |
5374 | spin_lock_irq(&rbd_dev->lock); | |
5375 | if (rbd_dev->open_count) | |
5376 | ret = -EBUSY; | |
5377 | else | |
82a442d2 AE |
5378 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5379 | &rbd_dev->flags); | |
751cc0e3 AE |
5380 | spin_unlock_irq(&rbd_dev->lock); |
5381 | } | |
5382 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5383 | if (ret < 0 || already) |
1ba0f1e7 | 5384 | return ret; |
751cc0e3 | 5385 | |
fca27065 | 5386 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5387 | /* |
5388 | * flush remaining watch callbacks - these must be complete | |
5389 | * before the osd_client is shutdown | |
5390 | */ | |
5391 | dout("%s: flushing notifies", __func__); | |
5392 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5393 | |
9875201e JD |
5394 | /* |
5395 | * Don't free anything from rbd_dev->disk until after all | |
5396 | * notifies are completely processed. Otherwise | |
5397 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5398 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5399 | */ | |
5400 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5401 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5402 | module_put(THIS_MODULE); |
aafb230e | 5403 | |
1ba0f1e7 | 5404 | return count; |
602adf40 YS |
5405 | } |
5406 | ||
9b60e70b ID |
5407 | static ssize_t rbd_remove(struct bus_type *bus, |
5408 | const char *buf, | |
5409 | size_t count) | |
5410 | { | |
5411 | if (single_major) | |
5412 | return -EINVAL; | |
5413 | ||
5414 | return do_rbd_remove(bus, buf, count); | |
5415 | } | |
5416 | ||
5417 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5418 | const char *buf, | |
5419 | size_t count) | |
5420 | { | |
5421 | return do_rbd_remove(bus, buf, count); | |
5422 | } | |
5423 | ||
602adf40 YS |
5424 | /* |
5425 | * create control files in sysfs | |
dfc5606d | 5426 | * /sys/bus/rbd/... |
602adf40 YS |
5427 | */ |
5428 | static int rbd_sysfs_init(void) | |
5429 | { | |
dfc5606d | 5430 | int ret; |
602adf40 | 5431 | |
fed4c143 | 5432 | ret = device_register(&rbd_root_dev); |
21079786 | 5433 | if (ret < 0) |
dfc5606d | 5434 | return ret; |
602adf40 | 5435 | |
fed4c143 AE |
5436 | ret = bus_register(&rbd_bus_type); |
5437 | if (ret < 0) | |
5438 | device_unregister(&rbd_root_dev); | |
602adf40 | 5439 | |
602adf40 YS |
5440 | return ret; |
5441 | } | |
5442 | ||
5443 | static void rbd_sysfs_cleanup(void) | |
5444 | { | |
dfc5606d | 5445 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5446 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5447 | } |
5448 | ||
1c2a9dfe AE |
5449 | static int rbd_slab_init(void) |
5450 | { | |
5451 | rbd_assert(!rbd_img_request_cache); | |
5452 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5453 | sizeof (struct rbd_img_request), | |
5454 | __alignof__(struct rbd_img_request), | |
5455 | 0, NULL); | |
868311b1 AE |
5456 | if (!rbd_img_request_cache) |
5457 | return -ENOMEM; | |
5458 | ||
5459 | rbd_assert(!rbd_obj_request_cache); | |
5460 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5461 | sizeof (struct rbd_obj_request), | |
5462 | __alignof__(struct rbd_obj_request), | |
5463 | 0, NULL); | |
78c2a44a AE |
5464 | if (!rbd_obj_request_cache) |
5465 | goto out_err; | |
5466 | ||
5467 | rbd_assert(!rbd_segment_name_cache); | |
5468 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5469 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5470 | if (rbd_segment_name_cache) |
1c2a9dfe | 5471 | return 0; |
78c2a44a AE |
5472 | out_err: |
5473 | if (rbd_obj_request_cache) { | |
5474 | kmem_cache_destroy(rbd_obj_request_cache); | |
5475 | rbd_obj_request_cache = NULL; | |
5476 | } | |
1c2a9dfe | 5477 | |
868311b1 AE |
5478 | kmem_cache_destroy(rbd_img_request_cache); |
5479 | rbd_img_request_cache = NULL; | |
5480 | ||
1c2a9dfe AE |
5481 | return -ENOMEM; |
5482 | } | |
5483 | ||
5484 | static void rbd_slab_exit(void) | |
5485 | { | |
78c2a44a AE |
5486 | rbd_assert(rbd_segment_name_cache); |
5487 | kmem_cache_destroy(rbd_segment_name_cache); | |
5488 | rbd_segment_name_cache = NULL; | |
5489 | ||
868311b1 AE |
5490 | rbd_assert(rbd_obj_request_cache); |
5491 | kmem_cache_destroy(rbd_obj_request_cache); | |
5492 | rbd_obj_request_cache = NULL; | |
5493 | ||
1c2a9dfe AE |
5494 | rbd_assert(rbd_img_request_cache); |
5495 | kmem_cache_destroy(rbd_img_request_cache); | |
5496 | rbd_img_request_cache = NULL; | |
5497 | } | |
5498 | ||
cc344fa1 | 5499 | static int __init rbd_init(void) |
602adf40 YS |
5500 | { |
5501 | int rc; | |
5502 | ||
1e32d34c AE |
5503 | if (!libceph_compatible(NULL)) { |
5504 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5505 | return -EINVAL; |
5506 | } | |
e1b4d96d | 5507 | |
1c2a9dfe | 5508 | rc = rbd_slab_init(); |
602adf40 YS |
5509 | if (rc) |
5510 | return rc; | |
e1b4d96d | 5511 | |
9b60e70b ID |
5512 | if (single_major) { |
5513 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5514 | if (rbd_major < 0) { | |
5515 | rc = rbd_major; | |
5516 | goto err_out_slab; | |
5517 | } | |
5518 | } | |
5519 | ||
1c2a9dfe AE |
5520 | rc = rbd_sysfs_init(); |
5521 | if (rc) | |
9b60e70b ID |
5522 | goto err_out_blkdev; |
5523 | ||
5524 | if (single_major) | |
5525 | pr_info("loaded (major %d)\n", rbd_major); | |
5526 | else | |
5527 | pr_info("loaded\n"); | |
1c2a9dfe | 5528 | |
e1b4d96d ID |
5529 | return 0; |
5530 | ||
9b60e70b ID |
5531 | err_out_blkdev: |
5532 | if (single_major) | |
5533 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
e1b4d96d ID |
5534 | err_out_slab: |
5535 | rbd_slab_exit(); | |
1c2a9dfe | 5536 | return rc; |
602adf40 YS |
5537 | } |
5538 | ||
cc344fa1 | 5539 | static void __exit rbd_exit(void) |
602adf40 | 5540 | { |
ffe312cf | 5541 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5542 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5543 | if (single_major) |
5544 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
1c2a9dfe | 5545 | rbd_slab_exit(); |
602adf40 YS |
5546 | } |
5547 | ||
5548 | module_init(rbd_init); | |
5549 | module_exit(rbd_exit); | |
5550 | ||
d552c619 | 5551 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5552 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5553 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5554 | /* following authorship retained from original osdblk.c */ |
5555 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5556 | ||
90da258b | 5557 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5558 | MODULE_LICENSE("GPL"); |