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 | ||
602adf40 YS |
562 | static const struct block_device_operations rbd_bd_ops = { |
563 | .owner = THIS_MODULE, | |
564 | .open = rbd_open, | |
dfc5606d | 565 | .release = rbd_release, |
602adf40 YS |
566 | }; |
567 | ||
568 | /* | |
7262cfca | 569 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 570 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 571 | */ |
f8c38929 | 572 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
573 | { |
574 | struct rbd_client *rbdc; | |
575 | int ret = -ENOMEM; | |
576 | ||
37206ee5 | 577 | dout("%s:\n", __func__); |
602adf40 YS |
578 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
579 | if (!rbdc) | |
580 | goto out_opt; | |
581 | ||
582 | kref_init(&rbdc->kref); | |
583 | INIT_LIST_HEAD(&rbdc->node); | |
584 | ||
43ae4701 | 585 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 586 | if (IS_ERR(rbdc->client)) |
08f75463 | 587 | goto out_rbdc; |
43ae4701 | 588 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
589 | |
590 | ret = ceph_open_session(rbdc->client); | |
591 | if (ret < 0) | |
08f75463 | 592 | goto out_client; |
602adf40 | 593 | |
432b8587 | 594 | spin_lock(&rbd_client_list_lock); |
602adf40 | 595 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 596 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 597 | |
37206ee5 | 598 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 599 | |
602adf40 | 600 | return rbdc; |
08f75463 | 601 | out_client: |
602adf40 | 602 | ceph_destroy_client(rbdc->client); |
08f75463 | 603 | out_rbdc: |
602adf40 YS |
604 | kfree(rbdc); |
605 | out_opt: | |
43ae4701 AE |
606 | if (ceph_opts) |
607 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
608 | dout("%s: error %d\n", __func__, ret); |
609 | ||
28f259b7 | 610 | return ERR_PTR(ret); |
602adf40 YS |
611 | } |
612 | ||
2f82ee54 AE |
613 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
614 | { | |
615 | kref_get(&rbdc->kref); | |
616 | ||
617 | return rbdc; | |
618 | } | |
619 | ||
602adf40 | 620 | /* |
1f7ba331 AE |
621 | * Find a ceph client with specific addr and configuration. If |
622 | * found, bump its reference count. | |
602adf40 | 623 | */ |
1f7ba331 | 624 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
625 | { |
626 | struct rbd_client *client_node; | |
1f7ba331 | 627 | bool found = false; |
602adf40 | 628 | |
43ae4701 | 629 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
630 | return NULL; |
631 | ||
1f7ba331 AE |
632 | spin_lock(&rbd_client_list_lock); |
633 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
634 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
635 | __rbd_get_client(client_node); |
636 | ||
1f7ba331 AE |
637 | found = true; |
638 | break; | |
639 | } | |
640 | } | |
641 | spin_unlock(&rbd_client_list_lock); | |
642 | ||
643 | return found ? client_node : NULL; | |
602adf40 YS |
644 | } |
645 | ||
59c2be1e YS |
646 | /* |
647 | * mount options | |
648 | */ | |
649 | enum { | |
59c2be1e YS |
650 | Opt_last_int, |
651 | /* int args above */ | |
652 | Opt_last_string, | |
653 | /* string args above */ | |
cc0538b6 AE |
654 | Opt_read_only, |
655 | Opt_read_write, | |
656 | /* Boolean args above */ | |
657 | Opt_last_bool, | |
59c2be1e YS |
658 | }; |
659 | ||
43ae4701 | 660 | static match_table_t rbd_opts_tokens = { |
59c2be1e YS |
661 | /* int args above */ |
662 | /* string args above */ | |
be466c1c | 663 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
664 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
665 | {Opt_read_write, "read_write"}, | |
666 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
667 | /* Boolean args above */ | |
59c2be1e YS |
668 | {-1, NULL} |
669 | }; | |
670 | ||
98571b5a AE |
671 | struct rbd_options { |
672 | bool read_only; | |
673 | }; | |
674 | ||
675 | #define RBD_READ_ONLY_DEFAULT false | |
676 | ||
59c2be1e YS |
677 | static int parse_rbd_opts_token(char *c, void *private) |
678 | { | |
43ae4701 | 679 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
680 | substring_t argstr[MAX_OPT_ARGS]; |
681 | int token, intval, ret; | |
682 | ||
43ae4701 | 683 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
684 | if (token < 0) |
685 | return -EINVAL; | |
686 | ||
687 | if (token < Opt_last_int) { | |
688 | ret = match_int(&argstr[0], &intval); | |
689 | if (ret < 0) { | |
690 | pr_err("bad mount option arg (not int) " | |
691 | "at '%s'\n", c); | |
692 | return ret; | |
693 | } | |
694 | dout("got int token %d val %d\n", token, intval); | |
695 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
696 | dout("got string token %d val %s\n", token, | |
697 | argstr[0].from); | |
cc0538b6 AE |
698 | } else if (token > Opt_last_string && token < Opt_last_bool) { |
699 | dout("got Boolean token %d\n", token); | |
59c2be1e YS |
700 | } else { |
701 | dout("got token %d\n", token); | |
702 | } | |
703 | ||
704 | switch (token) { | |
cc0538b6 AE |
705 | case Opt_read_only: |
706 | rbd_opts->read_only = true; | |
707 | break; | |
708 | case Opt_read_write: | |
709 | rbd_opts->read_only = false; | |
710 | break; | |
59c2be1e | 711 | default: |
aafb230e AE |
712 | rbd_assert(false); |
713 | break; | |
59c2be1e YS |
714 | } |
715 | return 0; | |
716 | } | |
717 | ||
602adf40 YS |
718 | /* |
719 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
720 | * not exist create it. Either way, ceph_opts is consumed by this |
721 | * function. | |
602adf40 | 722 | */ |
9d3997fd | 723 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 724 | { |
f8c38929 | 725 | struct rbd_client *rbdc; |
59c2be1e | 726 | |
cfbf6377 | 727 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 728 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 729 | if (rbdc) /* using an existing client */ |
43ae4701 | 730 | ceph_destroy_options(ceph_opts); |
9d3997fd | 731 | else |
f8c38929 | 732 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 733 | mutex_unlock(&client_mutex); |
602adf40 | 734 | |
9d3997fd | 735 | return rbdc; |
602adf40 YS |
736 | } |
737 | ||
738 | /* | |
739 | * Destroy ceph client | |
d23a4b3f | 740 | * |
432b8587 | 741 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
742 | */ |
743 | static void rbd_client_release(struct kref *kref) | |
744 | { | |
745 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
746 | ||
37206ee5 | 747 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 748 | spin_lock(&rbd_client_list_lock); |
602adf40 | 749 | list_del(&rbdc->node); |
cd9d9f5d | 750 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
751 | |
752 | ceph_destroy_client(rbdc->client); | |
753 | kfree(rbdc); | |
754 | } | |
755 | ||
756 | /* | |
757 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
758 | * it. | |
759 | */ | |
9d3997fd | 760 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 761 | { |
c53d5893 AE |
762 | if (rbdc) |
763 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
764 | } |
765 | ||
a30b71b9 AE |
766 | static bool rbd_image_format_valid(u32 image_format) |
767 | { | |
768 | return image_format == 1 || image_format == 2; | |
769 | } | |
770 | ||
8e94af8e AE |
771 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
772 | { | |
103a150f AE |
773 | size_t size; |
774 | u32 snap_count; | |
775 | ||
776 | /* The header has to start with the magic rbd header text */ | |
777 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
778 | return false; | |
779 | ||
db2388b6 AE |
780 | /* The bio layer requires at least sector-sized I/O */ |
781 | ||
782 | if (ondisk->options.order < SECTOR_SHIFT) | |
783 | return false; | |
784 | ||
785 | /* If we use u64 in a few spots we may be able to loosen this */ | |
786 | ||
787 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
788 | return false; | |
789 | ||
103a150f AE |
790 | /* |
791 | * The size of a snapshot header has to fit in a size_t, and | |
792 | * that limits the number of snapshots. | |
793 | */ | |
794 | snap_count = le32_to_cpu(ondisk->snap_count); | |
795 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
796 | if (snap_count > size / sizeof (__le64)) | |
797 | return false; | |
798 | ||
799 | /* | |
800 | * Not only that, but the size of the entire the snapshot | |
801 | * header must also be representable in a size_t. | |
802 | */ | |
803 | size -= snap_count * sizeof (__le64); | |
804 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
805 | return false; | |
806 | ||
807 | return true; | |
8e94af8e AE |
808 | } |
809 | ||
602adf40 | 810 | /* |
bb23e37a AE |
811 | * Fill an rbd image header with information from the given format 1 |
812 | * on-disk header. | |
602adf40 | 813 | */ |
662518b1 | 814 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 815 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 816 | { |
662518b1 | 817 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
818 | bool first_time = header->object_prefix == NULL; |
819 | struct ceph_snap_context *snapc; | |
820 | char *object_prefix = NULL; | |
821 | char *snap_names = NULL; | |
822 | u64 *snap_sizes = NULL; | |
ccece235 | 823 | u32 snap_count; |
d2bb24e5 | 824 | size_t size; |
bb23e37a | 825 | int ret = -ENOMEM; |
621901d6 | 826 | u32 i; |
602adf40 | 827 | |
bb23e37a | 828 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 829 | |
bb23e37a AE |
830 | if (first_time) { |
831 | size_t len; | |
103a150f | 832 | |
bb23e37a AE |
833 | len = strnlen(ondisk->object_prefix, |
834 | sizeof (ondisk->object_prefix)); | |
835 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
836 | if (!object_prefix) | |
837 | return -ENOMEM; | |
838 | memcpy(object_prefix, ondisk->object_prefix, len); | |
839 | object_prefix[len] = '\0'; | |
840 | } | |
00f1f36f | 841 | |
bb23e37a | 842 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 843 | |
bb23e37a AE |
844 | snap_count = le32_to_cpu(ondisk->snap_count); |
845 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
846 | if (!snapc) | |
847 | goto out_err; | |
848 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 849 | if (snap_count) { |
bb23e37a | 850 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
851 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
852 | ||
bb23e37a | 853 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 854 | |
bb23e37a AE |
855 | if (snap_names_len > (u64)SIZE_MAX) |
856 | goto out_2big; | |
857 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
858 | if (!snap_names) | |
6a52325f AE |
859 | goto out_err; |
860 | ||
bb23e37a | 861 | /* ...as well as the array of their sizes. */ |
621901d6 | 862 | |
d2bb24e5 | 863 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
864 | snap_sizes = kmalloc(size, GFP_KERNEL); |
865 | if (!snap_sizes) | |
6a52325f | 866 | goto out_err; |
bb23e37a | 867 | |
f785cc1d | 868 | /* |
bb23e37a AE |
869 | * Copy the names, and fill in each snapshot's id |
870 | * and size. | |
871 | * | |
99a41ebc | 872 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 873 | * ondisk buffer we're working with has |
f785cc1d AE |
874 | * snap_names_len bytes beyond the end of the |
875 | * snapshot id array, this memcpy() is safe. | |
876 | */ | |
bb23e37a AE |
877 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
878 | snaps = ondisk->snaps; | |
879 | for (i = 0; i < snap_count; i++) { | |
880 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
881 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
882 | } | |
602adf40 | 883 | } |
6a52325f | 884 | |
bb23e37a | 885 | /* We won't fail any more, fill in the header */ |
621901d6 | 886 | |
bb23e37a AE |
887 | if (first_time) { |
888 | header->object_prefix = object_prefix; | |
889 | header->obj_order = ondisk->options.order; | |
890 | header->crypt_type = ondisk->options.crypt_type; | |
891 | header->comp_type = ondisk->options.comp_type; | |
892 | /* The rest aren't used for format 1 images */ | |
893 | header->stripe_unit = 0; | |
894 | header->stripe_count = 0; | |
895 | header->features = 0; | |
602adf40 | 896 | } else { |
662518b1 AE |
897 | ceph_put_snap_context(header->snapc); |
898 | kfree(header->snap_names); | |
899 | kfree(header->snap_sizes); | |
602adf40 | 900 | } |
849b4260 | 901 | |
bb23e37a | 902 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 903 | |
f84344f3 | 904 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
905 | header->snapc = snapc; |
906 | header->snap_names = snap_names; | |
907 | header->snap_sizes = snap_sizes; | |
468521c1 | 908 | |
662518b1 | 909 | /* Make sure mapping size is consistent with header info */ |
602adf40 | 910 | |
662518b1 AE |
911 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP || first_time) |
912 | if (rbd_dev->mapping.size != header->image_size) | |
913 | rbd_dev->mapping.size = header->image_size; | |
914 | ||
602adf40 | 915 | return 0; |
bb23e37a AE |
916 | out_2big: |
917 | ret = -EIO; | |
6a52325f | 918 | out_err: |
bb23e37a AE |
919 | kfree(snap_sizes); |
920 | kfree(snap_names); | |
921 | ceph_put_snap_context(snapc); | |
922 | kfree(object_prefix); | |
ccece235 | 923 | |
bb23e37a | 924 | return ret; |
602adf40 YS |
925 | } |
926 | ||
9682fc6d AE |
927 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
928 | { | |
929 | const char *snap_name; | |
930 | ||
931 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
932 | ||
933 | /* Skip over names until we find the one we are looking for */ | |
934 | ||
935 | snap_name = rbd_dev->header.snap_names; | |
936 | while (which--) | |
937 | snap_name += strlen(snap_name) + 1; | |
938 | ||
939 | return kstrdup(snap_name, GFP_KERNEL); | |
940 | } | |
941 | ||
30d1cff8 AE |
942 | /* |
943 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
944 | * Note that result is for snapshots in *descending* order. | |
945 | */ | |
946 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
947 | { | |
948 | u64 snap_id1 = *(u64 *)s1; | |
949 | u64 snap_id2 = *(u64 *)s2; | |
950 | ||
951 | if (snap_id1 < snap_id2) | |
952 | return 1; | |
953 | return snap_id1 == snap_id2 ? 0 : -1; | |
954 | } | |
955 | ||
956 | /* | |
957 | * Search a snapshot context to see if the given snapshot id is | |
958 | * present. | |
959 | * | |
960 | * Returns the position of the snapshot id in the array if it's found, | |
961 | * or BAD_SNAP_INDEX otherwise. | |
962 | * | |
963 | * Note: The snapshot array is in kept sorted (by the osd) in | |
964 | * reverse order, highest snapshot id first. | |
965 | */ | |
9682fc6d AE |
966 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
967 | { | |
968 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 969 | u64 *found; |
9682fc6d | 970 | |
30d1cff8 AE |
971 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
972 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 973 | |
30d1cff8 | 974 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
975 | } |
976 | ||
2ad3d716 AE |
977 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
978 | u64 snap_id) | |
9e15b77d | 979 | { |
54cac61f | 980 | u32 which; |
da6a6b63 | 981 | const char *snap_name; |
9e15b77d | 982 | |
54cac61f AE |
983 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
984 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 985 | return ERR_PTR(-ENOENT); |
54cac61f | 986 | |
da6a6b63 JD |
987 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
988 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
989 | } |
990 | ||
991 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
992 | { | |
9e15b77d AE |
993 | if (snap_id == CEPH_NOSNAP) |
994 | return RBD_SNAP_HEAD_NAME; | |
995 | ||
54cac61f AE |
996 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
997 | if (rbd_dev->image_format == 1) | |
998 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 999 | |
54cac61f | 1000 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1001 | } |
1002 | ||
2ad3d716 AE |
1003 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1004 | u64 *snap_size) | |
602adf40 | 1005 | { |
2ad3d716 AE |
1006 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1007 | if (snap_id == CEPH_NOSNAP) { | |
1008 | *snap_size = rbd_dev->header.image_size; | |
1009 | } else if (rbd_dev->image_format == 1) { | |
1010 | u32 which; | |
602adf40 | 1011 | |
2ad3d716 AE |
1012 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1013 | if (which == BAD_SNAP_INDEX) | |
1014 | return -ENOENT; | |
e86924a8 | 1015 | |
2ad3d716 AE |
1016 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1017 | } else { | |
1018 | u64 size = 0; | |
1019 | int ret; | |
1020 | ||
1021 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1022 | if (ret) | |
1023 | return ret; | |
1024 | ||
1025 | *snap_size = size; | |
1026 | } | |
1027 | return 0; | |
602adf40 YS |
1028 | } |
1029 | ||
2ad3d716 AE |
1030 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1031 | u64 *snap_features) | |
602adf40 | 1032 | { |
2ad3d716 AE |
1033 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1034 | if (snap_id == CEPH_NOSNAP) { | |
1035 | *snap_features = rbd_dev->header.features; | |
1036 | } else if (rbd_dev->image_format == 1) { | |
1037 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1038 | } else { |
2ad3d716 AE |
1039 | u64 features = 0; |
1040 | int ret; | |
8b0241f8 | 1041 | |
2ad3d716 AE |
1042 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1043 | if (ret) | |
1044 | return ret; | |
1045 | ||
1046 | *snap_features = features; | |
1047 | } | |
1048 | return 0; | |
1049 | } | |
1050 | ||
1051 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1052 | { | |
8f4b7d98 | 1053 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1054 | u64 size = 0; |
1055 | u64 features = 0; | |
1056 | int ret; | |
1057 | ||
2ad3d716 AE |
1058 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1059 | if (ret) | |
1060 | return ret; | |
1061 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1062 | if (ret) | |
1063 | return ret; | |
1064 | ||
1065 | rbd_dev->mapping.size = size; | |
1066 | rbd_dev->mapping.features = features; | |
1067 | ||
8b0241f8 | 1068 | return 0; |
602adf40 YS |
1069 | } |
1070 | ||
d1cf5788 AE |
1071 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1072 | { | |
1073 | rbd_dev->mapping.size = 0; | |
1074 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1075 | } |
1076 | ||
98571b5a | 1077 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1078 | { |
65ccfe21 AE |
1079 | char *name; |
1080 | u64 segment; | |
1081 | int ret; | |
3a96d5cd | 1082 | char *name_format; |
602adf40 | 1083 | |
78c2a44a | 1084 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1085 | if (!name) |
1086 | return NULL; | |
1087 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1088 | name_format = "%s.%012llx"; |
1089 | if (rbd_dev->image_format == 2) | |
1090 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1091 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1092 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1093 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1094 | pr_err("error formatting segment name for #%llu (%d)\n", |
1095 | segment, ret); | |
1096 | kfree(name); | |
1097 | name = NULL; | |
1098 | } | |
602adf40 | 1099 | |
65ccfe21 AE |
1100 | return name; |
1101 | } | |
602adf40 | 1102 | |
78c2a44a AE |
1103 | static void rbd_segment_name_free(const char *name) |
1104 | { | |
1105 | /* The explicit cast here is needed to drop the const qualifier */ | |
1106 | ||
1107 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1108 | } | |
1109 | ||
65ccfe21 AE |
1110 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1111 | { | |
1112 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1113 | |
65ccfe21 AE |
1114 | return offset & (segment_size - 1); |
1115 | } | |
1116 | ||
1117 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1118 | u64 offset, u64 length) | |
1119 | { | |
1120 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1121 | ||
1122 | offset &= segment_size - 1; | |
1123 | ||
aafb230e | 1124 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1125 | if (offset + length > segment_size) |
1126 | length = segment_size - offset; | |
1127 | ||
1128 | return length; | |
602adf40 YS |
1129 | } |
1130 | ||
029bcbd8 JD |
1131 | /* |
1132 | * returns the size of an object in the image | |
1133 | */ | |
1134 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1135 | { | |
1136 | return 1 << header->obj_order; | |
1137 | } | |
1138 | ||
602adf40 YS |
1139 | /* |
1140 | * bio helpers | |
1141 | */ | |
1142 | ||
1143 | static void bio_chain_put(struct bio *chain) | |
1144 | { | |
1145 | struct bio *tmp; | |
1146 | ||
1147 | while (chain) { | |
1148 | tmp = chain; | |
1149 | chain = chain->bi_next; | |
1150 | bio_put(tmp); | |
1151 | } | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * zeros a bio chain, starting at specific offset | |
1156 | */ | |
1157 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1158 | { | |
7988613b KO |
1159 | struct bio_vec bv; |
1160 | struct bvec_iter iter; | |
602adf40 YS |
1161 | unsigned long flags; |
1162 | void *buf; | |
602adf40 YS |
1163 | int pos = 0; |
1164 | ||
1165 | while (chain) { | |
7988613b KO |
1166 | bio_for_each_segment(bv, chain, iter) { |
1167 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1168 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1169 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1170 | memset(buf + remainder, 0, |
7988613b KO |
1171 | bv.bv_len - remainder); |
1172 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1173 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1174 | } |
7988613b | 1175 | pos += bv.bv_len; |
602adf40 YS |
1176 | } |
1177 | ||
1178 | chain = chain->bi_next; | |
1179 | } | |
1180 | } | |
1181 | ||
b9434c5b AE |
1182 | /* |
1183 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1184 | * starting at the given byte offset from the start of the array and | |
1185 | * continuing up to the given end offset. The pages array is | |
1186 | * assumed to be big enough to hold all bytes up to the end. | |
1187 | */ | |
1188 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1189 | { | |
1190 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1191 | ||
1192 | rbd_assert(end > offset); | |
1193 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1194 | while (offset < end) { | |
1195 | size_t page_offset; | |
1196 | size_t length; | |
1197 | unsigned long flags; | |
1198 | void *kaddr; | |
1199 | ||
491205a8 GU |
1200 | page_offset = offset & ~PAGE_MASK; |
1201 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1202 | local_irq_save(flags); |
1203 | kaddr = kmap_atomic(*page); | |
1204 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1205 | flush_dcache_page(*page); |
b9434c5b AE |
1206 | kunmap_atomic(kaddr); |
1207 | local_irq_restore(flags); | |
1208 | ||
1209 | offset += length; | |
1210 | page++; | |
1211 | } | |
1212 | } | |
1213 | ||
602adf40 | 1214 | /* |
f7760dad AE |
1215 | * Clone a portion of a bio, starting at the given byte offset |
1216 | * and continuing for the number of bytes indicated. | |
602adf40 | 1217 | */ |
f7760dad AE |
1218 | static struct bio *bio_clone_range(struct bio *bio_src, |
1219 | unsigned int offset, | |
1220 | unsigned int len, | |
1221 | gfp_t gfpmask) | |
602adf40 | 1222 | { |
f7760dad AE |
1223 | struct bio *bio; |
1224 | ||
5341a627 | 1225 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1226 | if (!bio) |
1227 | return NULL; /* ENOMEM */ | |
602adf40 | 1228 | |
5341a627 | 1229 | bio_advance(bio, offset); |
4f024f37 | 1230 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1231 | |
1232 | return bio; | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * Clone a portion of a bio chain, starting at the given byte offset | |
1237 | * into the first bio in the source chain and continuing for the | |
1238 | * number of bytes indicated. The result is another bio chain of | |
1239 | * exactly the given length, or a null pointer on error. | |
1240 | * | |
1241 | * The bio_src and offset parameters are both in-out. On entry they | |
1242 | * refer to the first source bio and the offset into that bio where | |
1243 | * the start of data to be cloned is located. | |
1244 | * | |
1245 | * On return, bio_src is updated to refer to the bio in the source | |
1246 | * chain that contains first un-cloned byte, and *offset will | |
1247 | * contain the offset of that byte within that bio. | |
1248 | */ | |
1249 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1250 | unsigned int *offset, | |
1251 | unsigned int len, | |
1252 | gfp_t gfpmask) | |
1253 | { | |
1254 | struct bio *bi = *bio_src; | |
1255 | unsigned int off = *offset; | |
1256 | struct bio *chain = NULL; | |
1257 | struct bio **end; | |
1258 | ||
1259 | /* Build up a chain of clone bios up to the limit */ | |
1260 | ||
4f024f37 | 1261 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1262 | return NULL; /* Nothing to clone */ |
602adf40 | 1263 | |
f7760dad AE |
1264 | end = &chain; |
1265 | while (len) { | |
1266 | unsigned int bi_size; | |
1267 | struct bio *bio; | |
1268 | ||
f5400b7a AE |
1269 | if (!bi) { |
1270 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1271 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1272 | } |
4f024f37 | 1273 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1274 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1275 | if (!bio) | |
1276 | goto out_err; /* ENOMEM */ | |
1277 | ||
1278 | *end = bio; | |
1279 | end = &bio->bi_next; | |
602adf40 | 1280 | |
f7760dad | 1281 | off += bi_size; |
4f024f37 | 1282 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1283 | bi = bi->bi_next; |
1284 | off = 0; | |
1285 | } | |
1286 | len -= bi_size; | |
1287 | } | |
1288 | *bio_src = bi; | |
1289 | *offset = off; | |
1290 | ||
1291 | return chain; | |
1292 | out_err: | |
1293 | bio_chain_put(chain); | |
602adf40 | 1294 | |
602adf40 YS |
1295 | return NULL; |
1296 | } | |
1297 | ||
926f9b3f AE |
1298 | /* |
1299 | * The default/initial value for all object request flags is 0. For | |
1300 | * each flag, once its value is set to 1 it is never reset to 0 | |
1301 | * again. | |
1302 | */ | |
57acbaa7 | 1303 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1304 | { |
57acbaa7 | 1305 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1306 | struct rbd_device *rbd_dev; |
1307 | ||
57acbaa7 AE |
1308 | rbd_dev = obj_request->img_request->rbd_dev; |
1309 | rbd_warn(rbd_dev, "obj_request %p already marked img_data\n", | |
926f9b3f AE |
1310 | obj_request); |
1311 | } | |
1312 | } | |
1313 | ||
57acbaa7 | 1314 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1315 | { |
1316 | smp_mb(); | |
57acbaa7 | 1317 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1318 | } |
1319 | ||
57acbaa7 | 1320 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1321 | { |
57acbaa7 AE |
1322 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1323 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1324 | |
57acbaa7 AE |
1325 | if (obj_request_img_data_test(obj_request)) |
1326 | rbd_dev = obj_request->img_request->rbd_dev; | |
1327 | rbd_warn(rbd_dev, "obj_request %p already marked done\n", | |
6365d33a AE |
1328 | obj_request); |
1329 | } | |
1330 | } | |
1331 | ||
57acbaa7 | 1332 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1333 | { |
1334 | smp_mb(); | |
57acbaa7 | 1335 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1336 | } |
1337 | ||
5679c59f AE |
1338 | /* |
1339 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1340 | * flag. The latter is set based on the "exists" value provided. | |
1341 | * | |
1342 | * Note that for our purposes once an object exists it never goes | |
1343 | * away again. It's possible that the response from two existence | |
1344 | * checks are separated by the creation of the target object, and | |
1345 | * the first ("doesn't exist") response arrives *after* the second | |
1346 | * ("does exist"). In that case we ignore the second one. | |
1347 | */ | |
1348 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1349 | bool exists) | |
1350 | { | |
1351 | if (exists) | |
1352 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1353 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1354 | smp_mb(); | |
1355 | } | |
1356 | ||
1357 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1358 | { | |
1359 | smp_mb(); | |
1360 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1361 | } | |
1362 | ||
1363 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1364 | { | |
1365 | smp_mb(); | |
1366 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1367 | } | |
1368 | ||
bf0d5f50 AE |
1369 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1370 | { | |
37206ee5 AE |
1371 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1372 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1373 | kref_get(&obj_request->kref); |
1374 | } | |
1375 | ||
1376 | static void rbd_obj_request_destroy(struct kref *kref); | |
1377 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1378 | { | |
1379 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1380 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1381 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1382 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1383 | } | |
1384 | ||
e93f3152 AE |
1385 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1386 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1387 | static void rbd_img_request_destroy(struct kref *kref); |
1388 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1389 | { | |
1390 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1391 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1392 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1393 | if (img_request_child_test(img_request)) |
1394 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1395 | else | |
1396 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1397 | } |
1398 | ||
1399 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1400 | struct rbd_obj_request *obj_request) | |
1401 | { | |
25dcf954 AE |
1402 | rbd_assert(obj_request->img_request == NULL); |
1403 | ||
b155e86c | 1404 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1405 | obj_request->img_request = img_request; |
25dcf954 | 1406 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1407 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1408 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1409 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1410 | img_request->obj_request_count++; |
1411 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1412 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1413 | obj_request->which); | |
bf0d5f50 AE |
1414 | } |
1415 | ||
1416 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1417 | struct rbd_obj_request *obj_request) | |
1418 | { | |
1419 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1420 | |
37206ee5 AE |
1421 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1422 | obj_request->which); | |
bf0d5f50 | 1423 | list_del(&obj_request->links); |
25dcf954 AE |
1424 | rbd_assert(img_request->obj_request_count > 0); |
1425 | img_request->obj_request_count--; | |
1426 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1427 | obj_request->which = BAD_WHICH; | |
6365d33a | 1428 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1429 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1430 | obj_request->img_request = NULL; |
25dcf954 | 1431 | obj_request->callback = NULL; |
bf0d5f50 AE |
1432 | rbd_obj_request_put(obj_request); |
1433 | } | |
1434 | ||
1435 | static bool obj_request_type_valid(enum obj_request_type type) | |
1436 | { | |
1437 | switch (type) { | |
9969ebc5 | 1438 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1439 | case OBJ_REQUEST_BIO: |
788e2df3 | 1440 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1441 | return true; |
1442 | default: | |
1443 | return false; | |
1444 | } | |
1445 | } | |
1446 | ||
bf0d5f50 AE |
1447 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1448 | struct rbd_obj_request *obj_request) | |
1449 | { | |
37206ee5 AE |
1450 | dout("%s: osdc %p obj %p\n", __func__, osdc, obj_request); |
1451 | ||
bf0d5f50 AE |
1452 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1453 | } | |
1454 | ||
1455 | static void rbd_img_request_complete(struct rbd_img_request *img_request) | |
1456 | { | |
55f27e09 | 1457 | |
37206ee5 | 1458 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1459 | |
1460 | /* | |
1461 | * If no error occurred, compute the aggregate transfer | |
1462 | * count for the image request. We could instead use | |
1463 | * atomic64_cmpxchg() to update it as each object request | |
1464 | * completes; not clear which way is better off hand. | |
1465 | */ | |
1466 | if (!img_request->result) { | |
1467 | struct rbd_obj_request *obj_request; | |
1468 | u64 xferred = 0; | |
1469 | ||
1470 | for_each_obj_request(img_request, obj_request) | |
1471 | xferred += obj_request->xferred; | |
1472 | img_request->xferred = xferred; | |
1473 | } | |
1474 | ||
bf0d5f50 AE |
1475 | if (img_request->callback) |
1476 | img_request->callback(img_request); | |
1477 | else | |
1478 | rbd_img_request_put(img_request); | |
1479 | } | |
1480 | ||
788e2df3 AE |
1481 | /* Caller is responsible for rbd_obj_request_destroy(obj_request) */ |
1482 | ||
1483 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1484 | { | |
37206ee5 AE |
1485 | dout("%s: obj %p\n", __func__, obj_request); |
1486 | ||
788e2df3 AE |
1487 | return wait_for_completion_interruptible(&obj_request->completion); |
1488 | } | |
1489 | ||
0c425248 AE |
1490 | /* |
1491 | * The default/initial value for all image request flags is 0. Each | |
1492 | * is conditionally set to 1 at image request initialization time | |
1493 | * and currently never change thereafter. | |
1494 | */ | |
1495 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1496 | { | |
1497 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1498 | smp_mb(); | |
1499 | } | |
1500 | ||
1501 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1502 | { | |
1503 | smp_mb(); | |
1504 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1505 | } | |
1506 | ||
9849e986 AE |
1507 | static void img_request_child_set(struct rbd_img_request *img_request) |
1508 | { | |
1509 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1510 | smp_mb(); | |
1511 | } | |
1512 | ||
e93f3152 AE |
1513 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1514 | { | |
1515 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1516 | smp_mb(); | |
1517 | } | |
1518 | ||
9849e986 AE |
1519 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1520 | { | |
1521 | smp_mb(); | |
1522 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1523 | } | |
1524 | ||
d0b2e944 AE |
1525 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1526 | { | |
1527 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1528 | smp_mb(); | |
1529 | } | |
1530 | ||
a2acd00e AE |
1531 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1532 | { | |
1533 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1534 | smp_mb(); | |
1535 | } | |
1536 | ||
d0b2e944 AE |
1537 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1538 | { | |
1539 | smp_mb(); | |
1540 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1541 | } | |
1542 | ||
6e2a4505 AE |
1543 | static void |
1544 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1545 | { | |
b9434c5b AE |
1546 | u64 xferred = obj_request->xferred; |
1547 | u64 length = obj_request->length; | |
1548 | ||
6e2a4505 AE |
1549 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1550 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1551 | xferred, length); |
6e2a4505 | 1552 | /* |
17c1cc1d JD |
1553 | * ENOENT means a hole in the image. We zero-fill the entire |
1554 | * length of the request. A short read also implies zero-fill | |
1555 | * to the end of the request. An error requires the whole | |
1556 | * length of the request to be reported finished with an error | |
1557 | * to the block layer. In each case we update the xferred | |
1558 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1559 | */ |
b9434c5b | 1560 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1561 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1562 | if (obj_request->type == OBJ_REQUEST_BIO) |
1563 | zero_bio_chain(obj_request->bio_list, 0); | |
1564 | else | |
1565 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1566 | obj_request->result = 0; |
b9434c5b AE |
1567 | } else if (xferred < length && !obj_request->result) { |
1568 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1569 | zero_bio_chain(obj_request->bio_list, xferred); | |
1570 | else | |
1571 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1572 | } |
17c1cc1d | 1573 | obj_request->xferred = length; |
6e2a4505 AE |
1574 | obj_request_done_set(obj_request); |
1575 | } | |
1576 | ||
bf0d5f50 AE |
1577 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1578 | { | |
37206ee5 AE |
1579 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1580 | obj_request->callback); | |
bf0d5f50 AE |
1581 | if (obj_request->callback) |
1582 | obj_request->callback(obj_request); | |
788e2df3 AE |
1583 | else |
1584 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1585 | } |
1586 | ||
c47f9371 | 1587 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1588 | { |
1589 | dout("%s: obj %p\n", __func__, obj_request); | |
1590 | obj_request_done_set(obj_request); | |
1591 | } | |
1592 | ||
c47f9371 | 1593 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1594 | { |
57acbaa7 | 1595 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1596 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1597 | bool layered = false; |
1598 | ||
1599 | if (obj_request_img_data_test(obj_request)) { | |
1600 | img_request = obj_request->img_request; | |
1601 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1602 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1603 | } |
8b3e1a56 AE |
1604 | |
1605 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1606 | obj_request, img_request, obj_request->result, | |
1607 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1608 | if (layered && obj_request->result == -ENOENT && |
1609 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1610 | rbd_img_parent_read(obj_request); |
1611 | else if (img_request) | |
6e2a4505 AE |
1612 | rbd_img_obj_request_read_callback(obj_request); |
1613 | else | |
1614 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1615 | } |
1616 | ||
c47f9371 | 1617 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1618 | { |
1b83bef2 SW |
1619 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1620 | obj_request->result, obj_request->length); | |
1621 | /* | |
8b3e1a56 AE |
1622 | * There is no such thing as a successful short write. Set |
1623 | * it to our originally-requested length. | |
1b83bef2 SW |
1624 | */ |
1625 | obj_request->xferred = obj_request->length; | |
07741308 | 1626 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1627 | } |
1628 | ||
fbfab539 AE |
1629 | /* |
1630 | * For a simple stat call there's nothing to do. We'll do more if | |
1631 | * this is part of a write sequence for a layered image. | |
1632 | */ | |
c47f9371 | 1633 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1634 | { |
37206ee5 | 1635 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1636 | obj_request_done_set(obj_request); |
1637 | } | |
1638 | ||
bf0d5f50 AE |
1639 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1640 | struct ceph_msg *msg) | |
1641 | { | |
1642 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1643 | u16 opcode; |
1644 | ||
37206ee5 | 1645 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1646 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1647 | if (obj_request_img_data_test(obj_request)) { |
1648 | rbd_assert(obj_request->img_request); | |
1649 | rbd_assert(obj_request->which != BAD_WHICH); | |
1650 | } else { | |
1651 | rbd_assert(obj_request->which == BAD_WHICH); | |
1652 | } | |
bf0d5f50 | 1653 | |
1b83bef2 SW |
1654 | if (osd_req->r_result < 0) |
1655 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1656 | |
0eefd470 | 1657 | BUG_ON(osd_req->r_num_ops > 2); |
bf0d5f50 | 1658 | |
c47f9371 AE |
1659 | /* |
1660 | * We support a 64-bit length, but ultimately it has to be | |
1661 | * passed to blk_end_request(), which takes an unsigned int. | |
1662 | */ | |
1b83bef2 | 1663 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1664 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
79528734 | 1665 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1666 | switch (opcode) { |
1667 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1668 | rbd_osd_read_callback(obj_request); |
bf0d5f50 AE |
1669 | break; |
1670 | case CEPH_OSD_OP_WRITE: | |
c47f9371 | 1671 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1672 | break; |
fbfab539 | 1673 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1674 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1675 | break; |
36be9a76 | 1676 | case CEPH_OSD_OP_CALL: |
b8d70035 | 1677 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1678 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1679 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1680 | break; |
bf0d5f50 AE |
1681 | default: |
1682 | rbd_warn(NULL, "%s: unsupported op %hu\n", | |
1683 | obj_request->object_name, (unsigned short) opcode); | |
1684 | break; | |
1685 | } | |
1686 | ||
07741308 | 1687 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1688 | rbd_obj_request_complete(obj_request); |
1689 | } | |
1690 | ||
9d4df01f | 1691 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1692 | { |
1693 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1694 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1695 | u64 snap_id; |
430c28c3 | 1696 | |
8c042b0d | 1697 | rbd_assert(osd_req != NULL); |
430c28c3 | 1698 | |
9d4df01f | 1699 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1700 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1701 | NULL, snap_id, NULL); |
1702 | } | |
1703 | ||
1704 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1705 | { | |
1706 | struct rbd_img_request *img_request = obj_request->img_request; | |
1707 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1708 | struct ceph_snap_context *snapc; | |
1709 | struct timespec mtime = CURRENT_TIME; | |
1710 | ||
1711 | rbd_assert(osd_req != NULL); | |
1712 | ||
1713 | snapc = img_request ? img_request->snapc : NULL; | |
1714 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1715 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1716 | } |
1717 | ||
bf0d5f50 AE |
1718 | static struct ceph_osd_request *rbd_osd_req_create( |
1719 | struct rbd_device *rbd_dev, | |
1720 | bool write_request, | |
430c28c3 | 1721 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1722 | { |
bf0d5f50 AE |
1723 | struct ceph_snap_context *snapc = NULL; |
1724 | struct ceph_osd_client *osdc; | |
1725 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1726 | |
6365d33a AE |
1727 | if (obj_request_img_data_test(obj_request)) { |
1728 | struct rbd_img_request *img_request = obj_request->img_request; | |
1729 | ||
0c425248 AE |
1730 | rbd_assert(write_request == |
1731 | img_request_write_test(img_request)); | |
1732 | if (write_request) | |
bf0d5f50 | 1733 | snapc = img_request->snapc; |
bf0d5f50 AE |
1734 | } |
1735 | ||
1736 | /* Allocate and initialize the request, for the single op */ | |
1737 | ||
1738 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1739 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 1, false, GFP_ATOMIC); | |
1740 | if (!osd_req) | |
1741 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1742 | |
430c28c3 | 1743 | if (write_request) |
bf0d5f50 | 1744 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1745 | else |
bf0d5f50 | 1746 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1747 | |
1748 | osd_req->r_callback = rbd_osd_req_callback; | |
1749 | osd_req->r_priv = obj_request; | |
1750 | ||
3c972c95 ID |
1751 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1752 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1753 | |
bf0d5f50 AE |
1754 | return osd_req; |
1755 | } | |
1756 | ||
0eefd470 AE |
1757 | /* |
1758 | * Create a copyup osd request based on the information in the | |
1759 | * object request supplied. A copyup request has two osd ops, | |
1760 | * a copyup method call, and a "normal" write request. | |
1761 | */ | |
1762 | static struct ceph_osd_request * | |
1763 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1764 | { | |
1765 | struct rbd_img_request *img_request; | |
1766 | struct ceph_snap_context *snapc; | |
1767 | struct rbd_device *rbd_dev; | |
1768 | struct ceph_osd_client *osdc; | |
1769 | struct ceph_osd_request *osd_req; | |
1770 | ||
1771 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1772 | img_request = obj_request->img_request; | |
1773 | rbd_assert(img_request); | |
1774 | rbd_assert(img_request_write_test(img_request)); | |
1775 | ||
1776 | /* Allocate and initialize the request, for the two ops */ | |
1777 | ||
1778 | snapc = img_request->snapc; | |
1779 | rbd_dev = img_request->rbd_dev; | |
1780 | osdc = &rbd_dev->rbd_client->client->osdc; | |
1781 | osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC); | |
1782 | if (!osd_req) | |
1783 | return NULL; /* ENOMEM */ | |
1784 | ||
1785 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
1786 | osd_req->r_callback = rbd_osd_req_callback; | |
1787 | osd_req->r_priv = obj_request; | |
1788 | ||
3c972c95 ID |
1789 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1790 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 1791 | |
0eefd470 AE |
1792 | return osd_req; |
1793 | } | |
1794 | ||
1795 | ||
bf0d5f50 AE |
1796 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
1797 | { | |
1798 | ceph_osdc_put_request(osd_req); | |
1799 | } | |
1800 | ||
1801 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
1802 | ||
1803 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
1804 | u64 offset, u64 length, | |
1805 | enum obj_request_type type) | |
1806 | { | |
1807 | struct rbd_obj_request *obj_request; | |
1808 | size_t size; | |
1809 | char *name; | |
1810 | ||
1811 | rbd_assert(obj_request_type_valid(type)); | |
1812 | ||
1813 | size = strlen(object_name) + 1; | |
f907ad55 AE |
1814 | name = kmalloc(size, GFP_KERNEL); |
1815 | if (!name) | |
bf0d5f50 AE |
1816 | return NULL; |
1817 | ||
868311b1 | 1818 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_KERNEL); |
f907ad55 AE |
1819 | if (!obj_request) { |
1820 | kfree(name); | |
1821 | return NULL; | |
1822 | } | |
1823 | ||
bf0d5f50 AE |
1824 | obj_request->object_name = memcpy(name, object_name, size); |
1825 | obj_request->offset = offset; | |
1826 | obj_request->length = length; | |
926f9b3f | 1827 | obj_request->flags = 0; |
bf0d5f50 AE |
1828 | obj_request->which = BAD_WHICH; |
1829 | obj_request->type = type; | |
1830 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 1831 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
1832 | kref_init(&obj_request->kref); |
1833 | ||
37206ee5 AE |
1834 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
1835 | offset, length, (int)type, obj_request); | |
1836 | ||
bf0d5f50 AE |
1837 | return obj_request; |
1838 | } | |
1839 | ||
1840 | static void rbd_obj_request_destroy(struct kref *kref) | |
1841 | { | |
1842 | struct rbd_obj_request *obj_request; | |
1843 | ||
1844 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
1845 | ||
37206ee5 AE |
1846 | dout("%s: obj %p\n", __func__, obj_request); |
1847 | ||
bf0d5f50 AE |
1848 | rbd_assert(obj_request->img_request == NULL); |
1849 | rbd_assert(obj_request->which == BAD_WHICH); | |
1850 | ||
1851 | if (obj_request->osd_req) | |
1852 | rbd_osd_req_destroy(obj_request->osd_req); | |
1853 | ||
1854 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
1855 | switch (obj_request->type) { | |
9969ebc5 AE |
1856 | case OBJ_REQUEST_NODATA: |
1857 | break; /* Nothing to do */ | |
bf0d5f50 AE |
1858 | case OBJ_REQUEST_BIO: |
1859 | if (obj_request->bio_list) | |
1860 | bio_chain_put(obj_request->bio_list); | |
1861 | break; | |
788e2df3 AE |
1862 | case OBJ_REQUEST_PAGES: |
1863 | if (obj_request->pages) | |
1864 | ceph_release_page_vector(obj_request->pages, | |
1865 | obj_request->page_count); | |
1866 | break; | |
bf0d5f50 AE |
1867 | } |
1868 | ||
f907ad55 | 1869 | kfree(obj_request->object_name); |
868311b1 AE |
1870 | obj_request->object_name = NULL; |
1871 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
1872 | } |
1873 | ||
fb65d228 AE |
1874 | /* It's OK to call this for a device with no parent */ |
1875 | ||
1876 | static void rbd_spec_put(struct rbd_spec *spec); | |
1877 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
1878 | { | |
1879 | rbd_dev_remove_parent(rbd_dev); | |
1880 | rbd_spec_put(rbd_dev->parent_spec); | |
1881 | rbd_dev->parent_spec = NULL; | |
1882 | rbd_dev->parent_overlap = 0; | |
1883 | } | |
1884 | ||
a2acd00e AE |
1885 | /* |
1886 | * Parent image reference counting is used to determine when an | |
1887 | * image's parent fields can be safely torn down--after there are no | |
1888 | * more in-flight requests to the parent image. When the last | |
1889 | * reference is dropped, cleaning them up is safe. | |
1890 | */ | |
1891 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
1892 | { | |
1893 | int counter; | |
1894 | ||
1895 | if (!rbd_dev->parent_spec) | |
1896 | return; | |
1897 | ||
1898 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
1899 | if (counter > 0) | |
1900 | return; | |
1901 | ||
1902 | /* Last reference; clean up parent data structures */ | |
1903 | ||
1904 | if (!counter) | |
1905 | rbd_dev_unparent(rbd_dev); | |
1906 | else | |
1907 | rbd_warn(rbd_dev, "parent reference underflow\n"); | |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * If an image has a non-zero parent overlap, get a reference to its | |
1912 | * parent. | |
1913 | * | |
392a9dad AE |
1914 | * We must get the reference before checking for the overlap to |
1915 | * coordinate properly with zeroing the parent overlap in | |
1916 | * rbd_dev_v2_parent_info() when an image gets flattened. We | |
1917 | * drop it again if there is no overlap. | |
1918 | * | |
a2acd00e AE |
1919 | * Returns true if the rbd device has a parent with a non-zero |
1920 | * overlap and a reference for it was successfully taken, or | |
1921 | * false otherwise. | |
1922 | */ | |
1923 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
1924 | { | |
1925 | int counter; | |
1926 | ||
1927 | if (!rbd_dev->parent_spec) | |
1928 | return false; | |
1929 | ||
1930 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
1931 | if (counter > 0 && rbd_dev->parent_overlap) | |
1932 | return true; | |
1933 | ||
1934 | /* Image was flattened, but parent is not yet torn down */ | |
1935 | ||
1936 | if (counter < 0) | |
1937 | rbd_warn(rbd_dev, "parent reference overflow\n"); | |
1938 | ||
1939 | return false; | |
1940 | } | |
1941 | ||
bf0d5f50 AE |
1942 | /* |
1943 | * Caller is responsible for filling in the list of object requests | |
1944 | * that comprises the image request, and the Linux request pointer | |
1945 | * (if there is one). | |
1946 | */ | |
cc344fa1 AE |
1947 | static struct rbd_img_request *rbd_img_request_create( |
1948 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 1949 | u64 offset, u64 length, |
e93f3152 | 1950 | bool write_request) |
bf0d5f50 AE |
1951 | { |
1952 | struct rbd_img_request *img_request; | |
bf0d5f50 | 1953 | |
1c2a9dfe | 1954 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_ATOMIC); |
bf0d5f50 AE |
1955 | if (!img_request) |
1956 | return NULL; | |
1957 | ||
1958 | if (write_request) { | |
1959 | down_read(&rbd_dev->header_rwsem); | |
812164f8 | 1960 | ceph_get_snap_context(rbd_dev->header.snapc); |
bf0d5f50 | 1961 | up_read(&rbd_dev->header_rwsem); |
bf0d5f50 AE |
1962 | } |
1963 | ||
1964 | img_request->rq = NULL; | |
1965 | img_request->rbd_dev = rbd_dev; | |
1966 | img_request->offset = offset; | |
1967 | img_request->length = length; | |
0c425248 AE |
1968 | img_request->flags = 0; |
1969 | if (write_request) { | |
1970 | img_request_write_set(img_request); | |
468521c1 | 1971 | img_request->snapc = rbd_dev->header.snapc; |
0c425248 | 1972 | } else { |
bf0d5f50 | 1973 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 1974 | } |
a2acd00e | 1975 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 1976 | img_request_layered_set(img_request); |
bf0d5f50 AE |
1977 | spin_lock_init(&img_request->completion_lock); |
1978 | img_request->next_completion = 0; | |
1979 | img_request->callback = NULL; | |
a5a337d4 | 1980 | img_request->result = 0; |
bf0d5f50 AE |
1981 | img_request->obj_request_count = 0; |
1982 | INIT_LIST_HEAD(&img_request->obj_requests); | |
1983 | kref_init(&img_request->kref); | |
1984 | ||
37206ee5 AE |
1985 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
1986 | write_request ? "write" : "read", offset, length, | |
1987 | img_request); | |
1988 | ||
bf0d5f50 AE |
1989 | return img_request; |
1990 | } | |
1991 | ||
1992 | static void rbd_img_request_destroy(struct kref *kref) | |
1993 | { | |
1994 | struct rbd_img_request *img_request; | |
1995 | struct rbd_obj_request *obj_request; | |
1996 | struct rbd_obj_request *next_obj_request; | |
1997 | ||
1998 | img_request = container_of(kref, struct rbd_img_request, kref); | |
1999 | ||
37206ee5 AE |
2000 | dout("%s: img %p\n", __func__, img_request); |
2001 | ||
bf0d5f50 AE |
2002 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2003 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2004 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2005 | |
a2acd00e AE |
2006 | if (img_request_layered_test(img_request)) { |
2007 | img_request_layered_clear(img_request); | |
2008 | rbd_dev_parent_put(img_request->rbd_dev); | |
2009 | } | |
2010 | ||
0c425248 | 2011 | if (img_request_write_test(img_request)) |
812164f8 | 2012 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2013 | |
1c2a9dfe | 2014 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2015 | } |
2016 | ||
e93f3152 AE |
2017 | static struct rbd_img_request *rbd_parent_request_create( |
2018 | struct rbd_obj_request *obj_request, | |
2019 | u64 img_offset, u64 length) | |
2020 | { | |
2021 | struct rbd_img_request *parent_request; | |
2022 | struct rbd_device *rbd_dev; | |
2023 | ||
2024 | rbd_assert(obj_request->img_request); | |
2025 | rbd_dev = obj_request->img_request->rbd_dev; | |
2026 | ||
2027 | parent_request = rbd_img_request_create(rbd_dev->parent, | |
2028 | img_offset, length, false); | |
2029 | if (!parent_request) | |
2030 | return NULL; | |
2031 | ||
2032 | img_request_child_set(parent_request); | |
2033 | rbd_obj_request_get(obj_request); | |
2034 | parent_request->obj_request = obj_request; | |
2035 | ||
2036 | return parent_request; | |
2037 | } | |
2038 | ||
2039 | static void rbd_parent_request_destroy(struct kref *kref) | |
2040 | { | |
2041 | struct rbd_img_request *parent_request; | |
2042 | struct rbd_obj_request *orig_request; | |
2043 | ||
2044 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2045 | orig_request = parent_request->obj_request; | |
2046 | ||
2047 | parent_request->obj_request = NULL; | |
2048 | rbd_obj_request_put(orig_request); | |
2049 | img_request_child_clear(parent_request); | |
2050 | ||
2051 | rbd_img_request_destroy(kref); | |
2052 | } | |
2053 | ||
1217857f AE |
2054 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2055 | { | |
6365d33a | 2056 | struct rbd_img_request *img_request; |
1217857f AE |
2057 | unsigned int xferred; |
2058 | int result; | |
8b3e1a56 | 2059 | bool more; |
1217857f | 2060 | |
6365d33a AE |
2061 | rbd_assert(obj_request_img_data_test(obj_request)); |
2062 | img_request = obj_request->img_request; | |
2063 | ||
1217857f AE |
2064 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2065 | xferred = (unsigned int)obj_request->xferred; | |
2066 | result = obj_request->result; | |
2067 | if (result) { | |
2068 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2069 | ||
2070 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)\n", | |
2071 | img_request_write_test(img_request) ? "write" : "read", | |
2072 | obj_request->length, obj_request->img_offset, | |
2073 | obj_request->offset); | |
2074 | rbd_warn(rbd_dev, " result %d xferred %x\n", | |
2075 | result, xferred); | |
2076 | if (!img_request->result) | |
2077 | img_request->result = result; | |
2078 | } | |
2079 | ||
f1a4739f AE |
2080 | /* Image object requests don't own their page array */ |
2081 | ||
2082 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2083 | obj_request->pages = NULL; | |
2084 | obj_request->page_count = 0; | |
2085 | } | |
2086 | ||
8b3e1a56 AE |
2087 | if (img_request_child_test(img_request)) { |
2088 | rbd_assert(img_request->obj_request != NULL); | |
2089 | more = obj_request->which < img_request->obj_request_count - 1; | |
2090 | } else { | |
2091 | rbd_assert(img_request->rq != NULL); | |
2092 | more = blk_end_request(img_request->rq, result, xferred); | |
2093 | } | |
2094 | ||
2095 | return more; | |
1217857f AE |
2096 | } |
2097 | ||
2169238d AE |
2098 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2099 | { | |
2100 | struct rbd_img_request *img_request; | |
2101 | u32 which = obj_request->which; | |
2102 | bool more = true; | |
2103 | ||
6365d33a | 2104 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2105 | img_request = obj_request->img_request; |
2106 | ||
2107 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2108 | rbd_assert(img_request != NULL); | |
2169238d AE |
2109 | rbd_assert(img_request->obj_request_count > 0); |
2110 | rbd_assert(which != BAD_WHICH); | |
2111 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2112 | |
2113 | spin_lock_irq(&img_request->completion_lock); | |
2114 | if (which != img_request->next_completion) | |
2115 | goto out; | |
2116 | ||
2117 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2118 | rbd_assert(more); |
2119 | rbd_assert(which < img_request->obj_request_count); | |
2120 | ||
2121 | if (!obj_request_done_test(obj_request)) | |
2122 | break; | |
1217857f | 2123 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2124 | which++; |
2125 | } | |
2126 | ||
2127 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2128 | img_request->next_completion = which; | |
2129 | out: | |
2130 | spin_unlock_irq(&img_request->completion_lock); | |
2131 | ||
2132 | if (!more) | |
2133 | rbd_img_request_complete(img_request); | |
2134 | } | |
2135 | ||
f1a4739f AE |
2136 | /* |
2137 | * Split up an image request into one or more object requests, each | |
2138 | * to a different object. The "type" parameter indicates whether | |
2139 | * "data_desc" is the pointer to the head of a list of bio | |
2140 | * structures, or the base of a page array. In either case this | |
2141 | * function assumes data_desc describes memory sufficient to hold | |
2142 | * all data described by the image request. | |
2143 | */ | |
2144 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2145 | enum obj_request_type type, | |
2146 | void *data_desc) | |
bf0d5f50 AE |
2147 | { |
2148 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2149 | struct rbd_obj_request *obj_request = NULL; | |
2150 | struct rbd_obj_request *next_obj_request; | |
0c425248 | 2151 | bool write_request = img_request_write_test(img_request); |
a158073c | 2152 | struct bio *bio_list = NULL; |
f1a4739f | 2153 | unsigned int bio_offset = 0; |
a158073c | 2154 | struct page **pages = NULL; |
7da22d29 | 2155 | u64 img_offset; |
bf0d5f50 AE |
2156 | u64 resid; |
2157 | u16 opcode; | |
2158 | ||
f1a4739f AE |
2159 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2160 | (int)type, data_desc); | |
37206ee5 | 2161 | |
430c28c3 | 2162 | opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ; |
7da22d29 | 2163 | img_offset = img_request->offset; |
bf0d5f50 | 2164 | resid = img_request->length; |
4dda41d3 | 2165 | rbd_assert(resid > 0); |
f1a4739f AE |
2166 | |
2167 | if (type == OBJ_REQUEST_BIO) { | |
2168 | bio_list = data_desc; | |
4f024f37 KO |
2169 | rbd_assert(img_offset == |
2170 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
f1a4739f AE |
2171 | } else { |
2172 | rbd_assert(type == OBJ_REQUEST_PAGES); | |
2173 | pages = data_desc; | |
2174 | } | |
2175 | ||
bf0d5f50 | 2176 | while (resid) { |
2fa12320 | 2177 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2178 | const char *object_name; |
bf0d5f50 AE |
2179 | u64 offset; |
2180 | u64 length; | |
2181 | ||
7da22d29 | 2182 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2183 | if (!object_name) |
2184 | goto out_unwind; | |
7da22d29 AE |
2185 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2186 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2187 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2188 | offset, length, type); |
78c2a44a AE |
2189 | /* object request has its own copy of the object name */ |
2190 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2191 | if (!obj_request) |
2192 | goto out_unwind; | |
62054da6 | 2193 | |
03507db6 JD |
2194 | /* |
2195 | * set obj_request->img_request before creating the | |
2196 | * osd_request so that it gets the right snapc | |
2197 | */ | |
2198 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2199 | |
f1a4739f AE |
2200 | if (type == OBJ_REQUEST_BIO) { |
2201 | unsigned int clone_size; | |
2202 | ||
2203 | rbd_assert(length <= (u64)UINT_MAX); | |
2204 | clone_size = (unsigned int)length; | |
2205 | obj_request->bio_list = | |
2206 | bio_chain_clone_range(&bio_list, | |
2207 | &bio_offset, | |
2208 | clone_size, | |
2209 | GFP_ATOMIC); | |
2210 | if (!obj_request->bio_list) | |
62054da6 | 2211 | goto out_unwind; |
f1a4739f AE |
2212 | } else { |
2213 | unsigned int page_count; | |
2214 | ||
2215 | obj_request->pages = pages; | |
2216 | page_count = (u32)calc_pages_for(offset, length); | |
2217 | obj_request->page_count = page_count; | |
2218 | if ((offset + length) & ~PAGE_MASK) | |
2219 | page_count--; /* more on last page */ | |
2220 | pages += page_count; | |
2221 | } | |
bf0d5f50 | 2222 | |
2fa12320 AE |
2223 | osd_req = rbd_osd_req_create(rbd_dev, write_request, |
2224 | obj_request); | |
2225 | if (!osd_req) | |
62054da6 | 2226 | goto out_unwind; |
2fa12320 | 2227 | obj_request->osd_req = osd_req; |
2169238d | 2228 | obj_request->callback = rbd_img_obj_callback; |
430c28c3 | 2229 | |
2fa12320 AE |
2230 | osd_req_op_extent_init(osd_req, 0, opcode, offset, length, |
2231 | 0, 0); | |
f1a4739f AE |
2232 | if (type == OBJ_REQUEST_BIO) |
2233 | osd_req_op_extent_osd_data_bio(osd_req, 0, | |
2234 | obj_request->bio_list, length); | |
2235 | else | |
2236 | osd_req_op_extent_osd_data_pages(osd_req, 0, | |
2237 | obj_request->pages, length, | |
2238 | offset & ~PAGE_MASK, false, false); | |
9d4df01f AE |
2239 | |
2240 | if (write_request) | |
2241 | rbd_osd_req_format_write(obj_request); | |
2242 | else | |
2243 | rbd_osd_req_format_read(obj_request); | |
430c28c3 | 2244 | |
7da22d29 | 2245 | obj_request->img_offset = img_offset; |
bf0d5f50 | 2246 | |
7da22d29 | 2247 | img_offset += length; |
bf0d5f50 AE |
2248 | resid -= length; |
2249 | } | |
2250 | ||
2251 | return 0; | |
2252 | ||
bf0d5f50 AE |
2253 | out_unwind: |
2254 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
2255 | rbd_obj_request_put(obj_request); | |
2256 | ||
2257 | return -ENOMEM; | |
2258 | } | |
2259 | ||
0eefd470 AE |
2260 | static void |
2261 | rbd_img_obj_copyup_callback(struct rbd_obj_request *obj_request) | |
2262 | { | |
2263 | struct rbd_img_request *img_request; | |
2264 | struct rbd_device *rbd_dev; | |
ebda6408 | 2265 | struct page **pages; |
0eefd470 AE |
2266 | u32 page_count; |
2267 | ||
2268 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO); | |
2269 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2270 | img_request = obj_request->img_request; | |
2271 | rbd_assert(img_request); | |
2272 | ||
2273 | rbd_dev = img_request->rbd_dev; | |
2274 | rbd_assert(rbd_dev); | |
0eefd470 | 2275 | |
ebda6408 AE |
2276 | pages = obj_request->copyup_pages; |
2277 | rbd_assert(pages != NULL); | |
0eefd470 | 2278 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2279 | page_count = obj_request->copyup_page_count; |
2280 | rbd_assert(page_count); | |
2281 | obj_request->copyup_page_count = 0; | |
2282 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2283 | |
2284 | /* | |
2285 | * We want the transfer count to reflect the size of the | |
2286 | * original write request. There is no such thing as a | |
2287 | * successful short write, so if the request was successful | |
2288 | * we can just set it to the originally-requested length. | |
2289 | */ | |
2290 | if (!obj_request->result) | |
2291 | obj_request->xferred = obj_request->length; | |
2292 | ||
2293 | /* Finish up with the normal image object callback */ | |
2294 | ||
2295 | rbd_img_obj_callback(obj_request); | |
2296 | } | |
2297 | ||
3d7efd18 AE |
2298 | static void |
2299 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2300 | { | |
2301 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2302 | struct ceph_osd_request *osd_req; |
2303 | struct ceph_osd_client *osdc; | |
2304 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2305 | struct page **pages; |
ebda6408 | 2306 | u32 page_count; |
bbea1c1a | 2307 | int img_result; |
ebda6408 | 2308 | u64 parent_length; |
b91f09f1 AE |
2309 | u64 offset; |
2310 | u64 length; | |
3d7efd18 AE |
2311 | |
2312 | rbd_assert(img_request_child_test(img_request)); | |
2313 | ||
2314 | /* First get what we need from the image request */ | |
2315 | ||
2316 | pages = img_request->copyup_pages; | |
2317 | rbd_assert(pages != NULL); | |
2318 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2319 | page_count = img_request->copyup_page_count; |
2320 | rbd_assert(page_count); | |
2321 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2322 | |
2323 | orig_request = img_request->obj_request; | |
2324 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2325 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2326 | img_result = img_request->result; |
ebda6408 AE |
2327 | parent_length = img_request->length; |
2328 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2329 | rbd_img_request_put(img_request); |
3d7efd18 | 2330 | |
91c6febb AE |
2331 | rbd_assert(orig_request->img_request); |
2332 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2333 | rbd_assert(rbd_dev); |
0eefd470 | 2334 | |
bbea1c1a AE |
2335 | /* |
2336 | * If the overlap has become 0 (most likely because the | |
2337 | * image has been flattened) we need to free the pages | |
2338 | * and re-submit the original write request. | |
2339 | */ | |
2340 | if (!rbd_dev->parent_overlap) { | |
2341 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2342 | |
bbea1c1a AE |
2343 | ceph_release_page_vector(pages, page_count); |
2344 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2345 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2346 | if (!img_result) | |
2347 | return; | |
2348 | } | |
0eefd470 | 2349 | |
bbea1c1a | 2350 | if (img_result) |
0eefd470 | 2351 | goto out_err; |
0eefd470 | 2352 | |
8785b1d4 AE |
2353 | /* |
2354 | * The original osd request is of no use to use any more. | |
2355 | * We need a new one that can hold the two ops in a copyup | |
2356 | * request. Allocate the new copyup osd request for the | |
2357 | * original request, and release the old one. | |
2358 | */ | |
bbea1c1a | 2359 | img_result = -ENOMEM; |
0eefd470 AE |
2360 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2361 | if (!osd_req) | |
2362 | goto out_err; | |
8785b1d4 | 2363 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2364 | orig_request->osd_req = osd_req; |
2365 | orig_request->copyup_pages = pages; | |
ebda6408 | 2366 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2367 | |
0eefd470 | 2368 | /* Initialize the copyup op */ |
3d7efd18 | 2369 | |
0eefd470 | 2370 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2371 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2372 | false, false); |
3d7efd18 | 2373 | |
0eefd470 AE |
2374 | /* Then the original write request op */ |
2375 | ||
b91f09f1 AE |
2376 | offset = orig_request->offset; |
2377 | length = orig_request->length; | |
0eefd470 | 2378 | osd_req_op_extent_init(osd_req, 1, CEPH_OSD_OP_WRITE, |
b91f09f1 AE |
2379 | offset, length, 0, 0); |
2380 | if (orig_request->type == OBJ_REQUEST_BIO) | |
2381 | osd_req_op_extent_osd_data_bio(osd_req, 1, | |
2382 | orig_request->bio_list, length); | |
2383 | else | |
2384 | osd_req_op_extent_osd_data_pages(osd_req, 1, | |
2385 | orig_request->pages, length, | |
2386 | offset & ~PAGE_MASK, false, false); | |
0eefd470 AE |
2387 | |
2388 | rbd_osd_req_format_write(orig_request); | |
2389 | ||
2390 | /* All set, send it off. */ | |
2391 | ||
2392 | orig_request->callback = rbd_img_obj_copyup_callback; | |
2393 | osdc = &rbd_dev->rbd_client->client->osdc; | |
bbea1c1a AE |
2394 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2395 | if (!img_result) | |
0eefd470 AE |
2396 | return; |
2397 | out_err: | |
2398 | /* Record the error code and complete the request */ | |
2399 | ||
bbea1c1a | 2400 | orig_request->result = img_result; |
0eefd470 AE |
2401 | orig_request->xferred = 0; |
2402 | obj_request_done_set(orig_request); | |
2403 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2404 | } |
2405 | ||
2406 | /* | |
2407 | * Read from the parent image the range of data that covers the | |
2408 | * entire target of the given object request. This is used for | |
2409 | * satisfying a layered image write request when the target of an | |
2410 | * object request from the image request does not exist. | |
2411 | * | |
2412 | * A page array big enough to hold the returned data is allocated | |
2413 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2414 | * When the read completes, this page array will be transferred to | |
2415 | * the original object request for the copyup operation. | |
2416 | * | |
2417 | * If an error occurs, record it as the result of the original | |
2418 | * object request and mark it done so it gets completed. | |
2419 | */ | |
2420 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2421 | { | |
2422 | struct rbd_img_request *img_request = NULL; | |
2423 | struct rbd_img_request *parent_request = NULL; | |
2424 | struct rbd_device *rbd_dev; | |
2425 | u64 img_offset; | |
2426 | u64 length; | |
2427 | struct page **pages = NULL; | |
2428 | u32 page_count; | |
2429 | int result; | |
2430 | ||
2431 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2432 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2433 | |
2434 | img_request = obj_request->img_request; | |
2435 | rbd_assert(img_request != NULL); | |
2436 | rbd_dev = img_request->rbd_dev; | |
2437 | rbd_assert(rbd_dev->parent != NULL); | |
2438 | ||
2439 | /* | |
2440 | * Determine the byte range covered by the object in the | |
2441 | * child image to which the original request was to be sent. | |
2442 | */ | |
2443 | img_offset = obj_request->img_offset - obj_request->offset; | |
2444 | length = (u64)1 << rbd_dev->header.obj_order; | |
2445 | ||
a9e8ba2c AE |
2446 | /* |
2447 | * There is no defined parent data beyond the parent | |
2448 | * overlap, so limit what we read at that boundary if | |
2449 | * necessary. | |
2450 | */ | |
2451 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2452 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2453 | length = rbd_dev->parent_overlap - img_offset; | |
2454 | } | |
2455 | ||
3d7efd18 AE |
2456 | /* |
2457 | * Allocate a page array big enough to receive the data read | |
2458 | * from the parent. | |
2459 | */ | |
2460 | page_count = (u32)calc_pages_for(0, length); | |
2461 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2462 | if (IS_ERR(pages)) { | |
2463 | result = PTR_ERR(pages); | |
2464 | pages = NULL; | |
2465 | goto out_err; | |
2466 | } | |
2467 | ||
2468 | result = -ENOMEM; | |
e93f3152 AE |
2469 | parent_request = rbd_parent_request_create(obj_request, |
2470 | img_offset, length); | |
3d7efd18 AE |
2471 | if (!parent_request) |
2472 | goto out_err; | |
3d7efd18 AE |
2473 | |
2474 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2475 | if (result) | |
2476 | goto out_err; | |
2477 | parent_request->copyup_pages = pages; | |
ebda6408 | 2478 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2479 | |
2480 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2481 | result = rbd_img_request_submit(parent_request); | |
2482 | if (!result) | |
2483 | return 0; | |
2484 | ||
2485 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2486 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2487 | parent_request->obj_request = NULL; |
2488 | rbd_obj_request_put(obj_request); | |
2489 | out_err: | |
2490 | if (pages) | |
2491 | ceph_release_page_vector(pages, page_count); | |
2492 | if (parent_request) | |
2493 | rbd_img_request_put(parent_request); | |
2494 | obj_request->result = result; | |
2495 | obj_request->xferred = 0; | |
2496 | obj_request_done_set(obj_request); | |
2497 | ||
2498 | return result; | |
2499 | } | |
2500 | ||
c5b5ef6c AE |
2501 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2502 | { | |
c5b5ef6c | 2503 | struct rbd_obj_request *orig_request; |
638f5abe | 2504 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2505 | int result; |
2506 | ||
2507 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2508 | ||
2509 | /* | |
2510 | * All we need from the object request is the original | |
2511 | * request and the result of the STAT op. Grab those, then | |
2512 | * we're done with the request. | |
2513 | */ | |
2514 | orig_request = obj_request->obj_request; | |
2515 | obj_request->obj_request = NULL; | |
912c317d | 2516 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2517 | rbd_assert(orig_request); |
2518 | rbd_assert(orig_request->img_request); | |
2519 | ||
2520 | result = obj_request->result; | |
2521 | obj_request->result = 0; | |
2522 | ||
2523 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2524 | obj_request, orig_request, result, | |
2525 | obj_request->xferred, obj_request->length); | |
2526 | rbd_obj_request_put(obj_request); | |
2527 | ||
638f5abe AE |
2528 | /* |
2529 | * If the overlap has become 0 (most likely because the | |
2530 | * image has been flattened) we need to free the pages | |
2531 | * and re-submit the original write request. | |
2532 | */ | |
2533 | rbd_dev = orig_request->img_request->rbd_dev; | |
2534 | if (!rbd_dev->parent_overlap) { | |
2535 | struct ceph_osd_client *osdc; | |
2536 | ||
638f5abe AE |
2537 | osdc = &rbd_dev->rbd_client->client->osdc; |
2538 | result = rbd_obj_request_submit(osdc, orig_request); | |
2539 | if (!result) | |
2540 | return; | |
2541 | } | |
c5b5ef6c AE |
2542 | |
2543 | /* | |
2544 | * Our only purpose here is to determine whether the object | |
2545 | * exists, and we don't want to treat the non-existence as | |
2546 | * an error. If something else comes back, transfer the | |
2547 | * error to the original request and complete it now. | |
2548 | */ | |
2549 | if (!result) { | |
2550 | obj_request_existence_set(orig_request, true); | |
2551 | } else if (result == -ENOENT) { | |
2552 | obj_request_existence_set(orig_request, false); | |
2553 | } else if (result) { | |
2554 | orig_request->result = result; | |
3d7efd18 | 2555 | goto out; |
c5b5ef6c AE |
2556 | } |
2557 | ||
2558 | /* | |
2559 | * Resubmit the original request now that we have recorded | |
2560 | * whether the target object exists. | |
2561 | */ | |
b454e36d | 2562 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2563 | out: |
c5b5ef6c AE |
2564 | if (orig_request->result) |
2565 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2566 | } |
2567 | ||
2568 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2569 | { | |
2570 | struct rbd_obj_request *stat_request; | |
2571 | struct rbd_device *rbd_dev; | |
2572 | struct ceph_osd_client *osdc; | |
2573 | struct page **pages = NULL; | |
2574 | u32 page_count; | |
2575 | size_t size; | |
2576 | int ret; | |
2577 | ||
2578 | /* | |
2579 | * The response data for a STAT call consists of: | |
2580 | * le64 length; | |
2581 | * struct { | |
2582 | * le32 tv_sec; | |
2583 | * le32 tv_nsec; | |
2584 | * } mtime; | |
2585 | */ | |
2586 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2587 | page_count = (u32)calc_pages_for(0, size); | |
2588 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2589 | if (IS_ERR(pages)) | |
2590 | return PTR_ERR(pages); | |
2591 | ||
2592 | ret = -ENOMEM; | |
2593 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2594 | OBJ_REQUEST_PAGES); | |
2595 | if (!stat_request) | |
2596 | goto out; | |
2597 | ||
2598 | rbd_obj_request_get(obj_request); | |
2599 | stat_request->obj_request = obj_request; | |
2600 | stat_request->pages = pages; | |
2601 | stat_request->page_count = page_count; | |
2602 | ||
2603 | rbd_assert(obj_request->img_request); | |
2604 | rbd_dev = obj_request->img_request->rbd_dev; | |
2605 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, false, | |
2606 | stat_request); | |
2607 | if (!stat_request->osd_req) | |
2608 | goto out; | |
2609 | stat_request->callback = rbd_img_obj_exists_callback; | |
2610 | ||
2611 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT); | |
2612 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, | |
2613 | false, false); | |
9d4df01f | 2614 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2615 | |
2616 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2617 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2618 | out: | |
2619 | if (ret) | |
2620 | rbd_obj_request_put(obj_request); | |
2621 | ||
2622 | return ret; | |
2623 | } | |
2624 | ||
b454e36d AE |
2625 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) |
2626 | { | |
2627 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2628 | struct rbd_device *rbd_dev; |
3d7efd18 | 2629 | bool known; |
b454e36d AE |
2630 | |
2631 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2632 | ||
2633 | img_request = obj_request->img_request; | |
2634 | rbd_assert(img_request); | |
a9e8ba2c | 2635 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2636 | |
b454e36d | 2637 | /* |
a9e8ba2c AE |
2638 | * Only writes to layered images need special handling. |
2639 | * Reads and non-layered writes are simple object requests. | |
2640 | * Layered writes that start beyond the end of the overlap | |
2641 | * with the parent have no parent data, so they too are | |
2642 | * simple object requests. Finally, if the target object is | |
2643 | * known to already exist, its parent data has already been | |
2644 | * copied, so a write to the object can also be handled as a | |
2645 | * simple object request. | |
b454e36d AE |
2646 | */ |
2647 | if (!img_request_write_test(img_request) || | |
2648 | !img_request_layered_test(img_request) || | |
a9e8ba2c | 2649 | rbd_dev->parent_overlap <= obj_request->img_offset || |
3d7efd18 AE |
2650 | ((known = obj_request_known_test(obj_request)) && |
2651 | obj_request_exists_test(obj_request))) { | |
b454e36d AE |
2652 | |
2653 | struct rbd_device *rbd_dev; | |
2654 | struct ceph_osd_client *osdc; | |
2655 | ||
2656 | rbd_dev = obj_request->img_request->rbd_dev; | |
2657 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2658 | ||
2659 | return rbd_obj_request_submit(osdc, obj_request); | |
2660 | } | |
2661 | ||
2662 | /* | |
3d7efd18 AE |
2663 | * It's a layered write. The target object might exist but |
2664 | * we may not know that yet. If we know it doesn't exist, | |
2665 | * start by reading the data for the full target object from | |
2666 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2667 | */ |
3d7efd18 AE |
2668 | if (known) |
2669 | return rbd_img_obj_parent_read_full(obj_request); | |
2670 | ||
2671 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2672 | |
2673 | return rbd_img_obj_exists_submit(obj_request); | |
2674 | } | |
2675 | ||
bf0d5f50 AE |
2676 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2677 | { | |
bf0d5f50 | 2678 | struct rbd_obj_request *obj_request; |
46faeed4 | 2679 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2680 | |
37206ee5 | 2681 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2682 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2683 | int ret; |
2684 | ||
b454e36d | 2685 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2686 | if (ret) |
2687 | return ret; | |
bf0d5f50 AE |
2688 | } |
2689 | ||
2690 | return 0; | |
2691 | } | |
8b3e1a56 AE |
2692 | |
2693 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2694 | { | |
2695 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2696 | struct rbd_device *rbd_dev; |
2697 | u64 obj_end; | |
02c74fba AE |
2698 | u64 img_xferred; |
2699 | int img_result; | |
8b3e1a56 AE |
2700 | |
2701 | rbd_assert(img_request_child_test(img_request)); | |
2702 | ||
02c74fba AE |
2703 | /* First get what we need from the image request and release it */ |
2704 | ||
8b3e1a56 | 2705 | obj_request = img_request->obj_request; |
02c74fba AE |
2706 | img_xferred = img_request->xferred; |
2707 | img_result = img_request->result; | |
2708 | rbd_img_request_put(img_request); | |
2709 | ||
2710 | /* | |
2711 | * If the overlap has become 0 (most likely because the | |
2712 | * image has been flattened) we need to re-submit the | |
2713 | * original request. | |
2714 | */ | |
a9e8ba2c AE |
2715 | rbd_assert(obj_request); |
2716 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
2717 | rbd_dev = obj_request->img_request->rbd_dev; |
2718 | if (!rbd_dev->parent_overlap) { | |
2719 | struct ceph_osd_client *osdc; | |
2720 | ||
2721 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2722 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
2723 | if (!img_result) | |
2724 | return; | |
2725 | } | |
a9e8ba2c | 2726 | |
02c74fba | 2727 | obj_request->result = img_result; |
a9e8ba2c AE |
2728 | if (obj_request->result) |
2729 | goto out; | |
2730 | ||
2731 | /* | |
2732 | * We need to zero anything beyond the parent overlap | |
2733 | * boundary. Since rbd_img_obj_request_read_callback() | |
2734 | * will zero anything beyond the end of a short read, an | |
2735 | * easy way to do this is to pretend the data from the | |
2736 | * parent came up short--ending at the overlap boundary. | |
2737 | */ | |
2738 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
2739 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
2740 | if (obj_end > rbd_dev->parent_overlap) { |
2741 | u64 xferred = 0; | |
2742 | ||
2743 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
2744 | xferred = rbd_dev->parent_overlap - | |
2745 | obj_request->img_offset; | |
8b3e1a56 | 2746 | |
02c74fba | 2747 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 2748 | } else { |
02c74fba | 2749 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
2750 | } |
2751 | out: | |
8b3e1a56 AE |
2752 | rbd_img_obj_request_read_callback(obj_request); |
2753 | rbd_obj_request_complete(obj_request); | |
2754 | } | |
2755 | ||
2756 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
2757 | { | |
8b3e1a56 AE |
2758 | struct rbd_img_request *img_request; |
2759 | int result; | |
2760 | ||
2761 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2762 | rbd_assert(obj_request->img_request != NULL); | |
2763 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 2764 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 2765 | |
8b3e1a56 | 2766 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 2767 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 2768 | obj_request->img_offset, |
e93f3152 | 2769 | obj_request->length); |
8b3e1a56 AE |
2770 | result = -ENOMEM; |
2771 | if (!img_request) | |
2772 | goto out_err; | |
2773 | ||
5b2ab72d AE |
2774 | if (obj_request->type == OBJ_REQUEST_BIO) |
2775 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
2776 | obj_request->bio_list); | |
2777 | else | |
2778 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
2779 | obj_request->pages); | |
8b3e1a56 AE |
2780 | if (result) |
2781 | goto out_err; | |
2782 | ||
2783 | img_request->callback = rbd_img_parent_read_callback; | |
2784 | result = rbd_img_request_submit(img_request); | |
2785 | if (result) | |
2786 | goto out_err; | |
2787 | ||
2788 | return; | |
2789 | out_err: | |
2790 | if (img_request) | |
2791 | rbd_img_request_put(img_request); | |
2792 | obj_request->result = result; | |
2793 | obj_request->xferred = 0; | |
2794 | obj_request_done_set(obj_request); | |
2795 | } | |
bf0d5f50 | 2796 | |
20e0af67 | 2797 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
2798 | { |
2799 | struct rbd_obj_request *obj_request; | |
2169238d | 2800 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
2801 | int ret; |
2802 | ||
2803 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2804 | OBJ_REQUEST_NODATA); | |
2805 | if (!obj_request) | |
2806 | return -ENOMEM; | |
2807 | ||
2808 | ret = -ENOMEM; | |
430c28c3 | 2809 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
b8d70035 AE |
2810 | if (!obj_request->osd_req) |
2811 | goto out; | |
2812 | ||
c99d2d4a | 2813 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 2814 | notify_id, 0, 0); |
9d4df01f | 2815 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 2816 | |
b8d70035 | 2817 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 2818 | if (ret) |
20e0af67 JD |
2819 | goto out; |
2820 | ret = rbd_obj_request_wait(obj_request); | |
2821 | out: | |
2822 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
2823 | |
2824 | return ret; | |
2825 | } | |
2826 | ||
2827 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
2828 | { | |
2829 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 2830 | int ret; |
b8d70035 AE |
2831 | |
2832 | if (!rbd_dev) | |
2833 | return; | |
2834 | ||
37206ee5 | 2835 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
2836 | rbd_dev->header_name, (unsigned long long)notify_id, |
2837 | (unsigned int)opcode); | |
e627db08 AE |
2838 | ret = rbd_dev_refresh(rbd_dev); |
2839 | if (ret) | |
3b5cf2a2 | 2840 | rbd_warn(rbd_dev, "header refresh error (%d)\n", ret); |
b8d70035 | 2841 | |
20e0af67 | 2842 | rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
b8d70035 AE |
2843 | } |
2844 | ||
9969ebc5 AE |
2845 | /* |
2846 | * Request sync osd watch/unwatch. The value of "start" determines | |
2847 | * whether a watch request is being initiated or torn down. | |
2848 | */ | |
fca27065 | 2849 | static int __rbd_dev_header_watch_sync(struct rbd_device *rbd_dev, bool start) |
9969ebc5 AE |
2850 | { |
2851 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2852 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
2853 | int ret; |
2854 | ||
2855 | rbd_assert(start ^ !!rbd_dev->watch_event); | |
2856 | rbd_assert(start ^ !!rbd_dev->watch_request); | |
2857 | ||
2858 | if (start) { | |
3c663bbd | 2859 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
9969ebc5 AE |
2860 | &rbd_dev->watch_event); |
2861 | if (ret < 0) | |
2862 | return ret; | |
8eb87565 | 2863 | rbd_assert(rbd_dev->watch_event != NULL); |
9969ebc5 AE |
2864 | } |
2865 | ||
2866 | ret = -ENOMEM; | |
2867 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
2868 | OBJ_REQUEST_NODATA); | |
2869 | if (!obj_request) | |
2870 | goto out_cancel; | |
2871 | ||
430c28c3 AE |
2872 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, true, obj_request); |
2873 | if (!obj_request->osd_req) | |
2874 | goto out_cancel; | |
2875 | ||
8eb87565 | 2876 | if (start) |
975241af | 2877 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); |
8eb87565 | 2878 | else |
6977c3f9 | 2879 | ceph_osdc_unregister_linger_request(osdc, |
975241af | 2880 | rbd_dev->watch_request->osd_req); |
2169238d AE |
2881 | |
2882 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
1f3ef788 | 2883 | rbd_dev->watch_event->cookie, 0, start ? 1 : 0); |
9d4df01f | 2884 | rbd_osd_req_format_write(obj_request); |
2169238d | 2885 | |
9969ebc5 AE |
2886 | ret = rbd_obj_request_submit(osdc, obj_request); |
2887 | if (ret) | |
2888 | goto out_cancel; | |
2889 | ret = rbd_obj_request_wait(obj_request); | |
2890 | if (ret) | |
2891 | goto out_cancel; | |
9969ebc5 AE |
2892 | ret = obj_request->result; |
2893 | if (ret) | |
2894 | goto out_cancel; | |
2895 | ||
8eb87565 AE |
2896 | /* |
2897 | * A watch request is set to linger, so the underlying osd | |
2898 | * request won't go away until we unregister it. We retain | |
2899 | * a pointer to the object request during that time (in | |
2900 | * rbd_dev->watch_request), so we'll keep a reference to | |
2901 | * it. We'll drop that reference (below) after we've | |
2902 | * unregistered it. | |
2903 | */ | |
2904 | if (start) { | |
2905 | rbd_dev->watch_request = obj_request; | |
2906 | ||
2907 | return 0; | |
2908 | } | |
2909 | ||
2910 | /* We have successfully torn down the watch request */ | |
2911 | ||
2912 | rbd_obj_request_put(rbd_dev->watch_request); | |
2913 | rbd_dev->watch_request = NULL; | |
9969ebc5 AE |
2914 | out_cancel: |
2915 | /* Cancel the event if we're tearing down, or on error */ | |
2916 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
2917 | rbd_dev->watch_event = NULL; | |
9969ebc5 AE |
2918 | if (obj_request) |
2919 | rbd_obj_request_put(obj_request); | |
2920 | ||
2921 | return ret; | |
2922 | } | |
2923 | ||
fca27065 ID |
2924 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
2925 | { | |
2926 | return __rbd_dev_header_watch_sync(rbd_dev, true); | |
2927 | } | |
2928 | ||
2929 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) | |
2930 | { | |
2931 | int ret; | |
2932 | ||
2933 | ret = __rbd_dev_header_watch_sync(rbd_dev, false); | |
2934 | if (ret) { | |
2935 | rbd_warn(rbd_dev, "unable to tear down watch request: %d\n", | |
2936 | ret); | |
2937 | } | |
2938 | } | |
2939 | ||
36be9a76 | 2940 | /* |
f40eb349 AE |
2941 | * Synchronous osd object method call. Returns the number of bytes |
2942 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
2943 | */ |
2944 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
2945 | const char *object_name, | |
2946 | const char *class_name, | |
2947 | const char *method_name, | |
4157976b | 2948 | const void *outbound, |
36be9a76 | 2949 | size_t outbound_size, |
4157976b | 2950 | void *inbound, |
e2a58ee5 | 2951 | size_t inbound_size) |
36be9a76 | 2952 | { |
2169238d | 2953 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 2954 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
2955 | struct page **pages; |
2956 | u32 page_count; | |
2957 | int ret; | |
2958 | ||
2959 | /* | |
6010a451 AE |
2960 | * Method calls are ultimately read operations. The result |
2961 | * should placed into the inbound buffer provided. They | |
2962 | * also supply outbound data--parameters for the object | |
2963 | * method. Currently if this is present it will be a | |
2964 | * snapshot id. | |
36be9a76 | 2965 | */ |
57385b51 | 2966 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
2967 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
2968 | if (IS_ERR(pages)) | |
2969 | return PTR_ERR(pages); | |
2970 | ||
2971 | ret = -ENOMEM; | |
6010a451 | 2972 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
2973 | OBJ_REQUEST_PAGES); |
2974 | if (!obj_request) | |
2975 | goto out; | |
2976 | ||
2977 | obj_request->pages = pages; | |
2978 | obj_request->page_count = page_count; | |
2979 | ||
430c28c3 | 2980 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
36be9a76 AE |
2981 | if (!obj_request->osd_req) |
2982 | goto out; | |
2983 | ||
c99d2d4a | 2984 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
2985 | class_name, method_name); |
2986 | if (outbound_size) { | |
2987 | struct ceph_pagelist *pagelist; | |
2988 | ||
2989 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
2990 | if (!pagelist) | |
2991 | goto out; | |
2992 | ||
2993 | ceph_pagelist_init(pagelist); | |
2994 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
2995 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
2996 | pagelist); | |
2997 | } | |
a4ce40a9 AE |
2998 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
2999 | obj_request->pages, inbound_size, | |
44cd188d | 3000 | 0, false, false); |
9d4df01f | 3001 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3002 | |
36be9a76 AE |
3003 | ret = rbd_obj_request_submit(osdc, obj_request); |
3004 | if (ret) | |
3005 | goto out; | |
3006 | ret = rbd_obj_request_wait(obj_request); | |
3007 | if (ret) | |
3008 | goto out; | |
3009 | ||
3010 | ret = obj_request->result; | |
3011 | if (ret < 0) | |
3012 | goto out; | |
57385b51 AE |
3013 | |
3014 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3015 | ret = (int)obj_request->xferred; | |
903bb32e | 3016 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3017 | out: |
3018 | if (obj_request) | |
3019 | rbd_obj_request_put(obj_request); | |
3020 | else | |
3021 | ceph_release_page_vector(pages, page_count); | |
3022 | ||
3023 | return ret; | |
3024 | } | |
3025 | ||
bf0d5f50 | 3026 | static void rbd_request_fn(struct request_queue *q) |
cc344fa1 | 3027 | __releases(q->queue_lock) __acquires(q->queue_lock) |
bf0d5f50 AE |
3028 | { |
3029 | struct rbd_device *rbd_dev = q->queuedata; | |
3030 | bool read_only = rbd_dev->mapping.read_only; | |
3031 | struct request *rq; | |
3032 | int result; | |
3033 | ||
3034 | while ((rq = blk_fetch_request(q))) { | |
3035 | bool write_request = rq_data_dir(rq) == WRITE; | |
3036 | struct rbd_img_request *img_request; | |
3037 | u64 offset; | |
3038 | u64 length; | |
3039 | ||
3040 | /* Ignore any non-FS requests that filter through. */ | |
3041 | ||
3042 | if (rq->cmd_type != REQ_TYPE_FS) { | |
4dda41d3 AE |
3043 | dout("%s: non-fs request type %d\n", __func__, |
3044 | (int) rq->cmd_type); | |
3045 | __blk_end_request_all(rq, 0); | |
3046 | continue; | |
3047 | } | |
3048 | ||
3049 | /* Ignore/skip any zero-length requests */ | |
3050 | ||
3051 | offset = (u64) blk_rq_pos(rq) << SECTOR_SHIFT; | |
3052 | length = (u64) blk_rq_bytes(rq); | |
3053 | ||
3054 | if (!length) { | |
3055 | dout("%s: zero-length request\n", __func__); | |
bf0d5f50 AE |
3056 | __blk_end_request_all(rq, 0); |
3057 | continue; | |
3058 | } | |
3059 | ||
3060 | spin_unlock_irq(q->queue_lock); | |
3061 | ||
3062 | /* Disallow writes to a read-only device */ | |
3063 | ||
3064 | if (write_request) { | |
3065 | result = -EROFS; | |
3066 | if (read_only) | |
3067 | goto end_request; | |
3068 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); | |
3069 | } | |
3070 | ||
6d292906 AE |
3071 | /* |
3072 | * Quit early if the mapped snapshot no longer | |
3073 | * exists. It's still possible the snapshot will | |
3074 | * have disappeared by the time our request arrives | |
3075 | * at the osd, but there's no sense in sending it if | |
3076 | * we already know. | |
3077 | */ | |
3078 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
bf0d5f50 AE |
3079 | dout("request for non-existent snapshot"); |
3080 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3081 | result = -ENXIO; | |
3082 | goto end_request; | |
3083 | } | |
3084 | ||
bf0d5f50 | 3085 | result = -EINVAL; |
c0cd10db AE |
3086 | if (offset && length > U64_MAX - offset + 1) { |
3087 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)\n", | |
3088 | offset, length); | |
bf0d5f50 | 3089 | goto end_request; /* Shouldn't happen */ |
c0cd10db | 3090 | } |
bf0d5f50 | 3091 | |
00a653e2 AE |
3092 | result = -EIO; |
3093 | if (offset + length > rbd_dev->mapping.size) { | |
3094 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)\n", | |
3095 | offset, length, rbd_dev->mapping.size); | |
3096 | goto end_request; | |
3097 | } | |
3098 | ||
bf0d5f50 AE |
3099 | result = -ENOMEM; |
3100 | img_request = rbd_img_request_create(rbd_dev, offset, length, | |
e93f3152 | 3101 | write_request); |
bf0d5f50 AE |
3102 | if (!img_request) |
3103 | goto end_request; | |
3104 | ||
3105 | img_request->rq = rq; | |
3106 | ||
f1a4739f AE |
3107 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, |
3108 | rq->bio); | |
bf0d5f50 AE |
3109 | if (!result) |
3110 | result = rbd_img_request_submit(img_request); | |
3111 | if (result) | |
3112 | rbd_img_request_put(img_request); | |
3113 | end_request: | |
3114 | spin_lock_irq(q->queue_lock); | |
3115 | if (result < 0) { | |
7da22d29 AE |
3116 | rbd_warn(rbd_dev, "%s %llx at %llx result %d\n", |
3117 | write_request ? "write" : "read", | |
3118 | length, offset, result); | |
3119 | ||
bf0d5f50 AE |
3120 | __blk_end_request_all(rq, result); |
3121 | } | |
3122 | } | |
3123 | } | |
3124 | ||
602adf40 YS |
3125 | /* |
3126 | * a queue callback. Makes sure that we don't create a bio that spans across | |
3127 | * multiple osd objects. One exception would be with a single page bios, | |
f7760dad | 3128 | * which we handle later at bio_chain_clone_range() |
602adf40 YS |
3129 | */ |
3130 | static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd, | |
3131 | struct bio_vec *bvec) | |
3132 | { | |
3133 | struct rbd_device *rbd_dev = q->queuedata; | |
e5cfeed2 AE |
3134 | sector_t sector_offset; |
3135 | sector_t sectors_per_obj; | |
3136 | sector_t obj_sector_offset; | |
3137 | int ret; | |
3138 | ||
3139 | /* | |
3140 | * Find how far into its rbd object the partition-relative | |
3141 | * bio start sector is to offset relative to the enclosing | |
3142 | * device. | |
3143 | */ | |
3144 | sector_offset = get_start_sect(bmd->bi_bdev) + bmd->bi_sector; | |
3145 | sectors_per_obj = 1 << (rbd_dev->header.obj_order - SECTOR_SHIFT); | |
3146 | obj_sector_offset = sector_offset & (sectors_per_obj - 1); | |
3147 | ||
3148 | /* | |
3149 | * Compute the number of bytes from that offset to the end | |
3150 | * of the object. Account for what's already used by the bio. | |
3151 | */ | |
3152 | ret = (int) (sectors_per_obj - obj_sector_offset) << SECTOR_SHIFT; | |
3153 | if (ret > bmd->bi_size) | |
3154 | ret -= bmd->bi_size; | |
3155 | else | |
3156 | ret = 0; | |
3157 | ||
3158 | /* | |
3159 | * Don't send back more than was asked for. And if the bio | |
3160 | * was empty, let the whole thing through because: "Note | |
3161 | * that a block device *must* allow a single page to be | |
3162 | * added to an empty bio." | |
3163 | */ | |
3164 | rbd_assert(bvec->bv_len <= PAGE_SIZE); | |
3165 | if (ret > (int) bvec->bv_len || !bmd->bi_size) | |
3166 | ret = (int) bvec->bv_len; | |
3167 | ||
3168 | return ret; | |
602adf40 YS |
3169 | } |
3170 | ||
3171 | static void rbd_free_disk(struct rbd_device *rbd_dev) | |
3172 | { | |
3173 | struct gendisk *disk = rbd_dev->disk; | |
3174 | ||
3175 | if (!disk) | |
3176 | return; | |
3177 | ||
a0cab924 AE |
3178 | rbd_dev->disk = NULL; |
3179 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3180 | del_gendisk(disk); |
a0cab924 AE |
3181 | if (disk->queue) |
3182 | blk_cleanup_queue(disk->queue); | |
3183 | } | |
602adf40 YS |
3184 | put_disk(disk); |
3185 | } | |
3186 | ||
788e2df3 AE |
3187 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3188 | const char *object_name, | |
7097f8df | 3189 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3190 | |
3191 | { | |
2169238d | 3192 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3193 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3194 | struct page **pages = NULL; |
3195 | u32 page_count; | |
1ceae7ef | 3196 | size_t size; |
788e2df3 AE |
3197 | int ret; |
3198 | ||
3199 | page_count = (u32) calc_pages_for(offset, length); | |
3200 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3201 | if (IS_ERR(pages)) | |
3202 | ret = PTR_ERR(pages); | |
3203 | ||
3204 | ret = -ENOMEM; | |
3205 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3206 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3207 | if (!obj_request) |
3208 | goto out; | |
3209 | ||
3210 | obj_request->pages = pages; | |
3211 | obj_request->page_count = page_count; | |
3212 | ||
430c28c3 | 3213 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request); |
788e2df3 AE |
3214 | if (!obj_request->osd_req) |
3215 | goto out; | |
3216 | ||
c99d2d4a AE |
3217 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3218 | offset, length, 0, 0); | |
406e2c9f | 3219 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3220 | obj_request->pages, |
44cd188d AE |
3221 | obj_request->length, |
3222 | obj_request->offset & ~PAGE_MASK, | |
3223 | false, false); | |
9d4df01f | 3224 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3225 | |
788e2df3 AE |
3226 | ret = rbd_obj_request_submit(osdc, obj_request); |
3227 | if (ret) | |
3228 | goto out; | |
3229 | ret = rbd_obj_request_wait(obj_request); | |
3230 | if (ret) | |
3231 | goto out; | |
3232 | ||
3233 | ret = obj_request->result; | |
3234 | if (ret < 0) | |
3235 | goto out; | |
1ceae7ef AE |
3236 | |
3237 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3238 | size = (size_t) obj_request->xferred; | |
903bb32e | 3239 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3240 | rbd_assert(size <= (size_t)INT_MAX); |
3241 | ret = (int)size; | |
788e2df3 AE |
3242 | out: |
3243 | if (obj_request) | |
3244 | rbd_obj_request_put(obj_request); | |
3245 | else | |
3246 | ceph_release_page_vector(pages, page_count); | |
3247 | ||
3248 | return ret; | |
3249 | } | |
3250 | ||
602adf40 | 3251 | /* |
662518b1 AE |
3252 | * Read the complete header for the given rbd device. On successful |
3253 | * return, the rbd_dev->header field will contain up-to-date | |
3254 | * information about the image. | |
602adf40 | 3255 | */ |
99a41ebc | 3256 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3257 | { |
4156d998 | 3258 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3259 | u32 snap_count = 0; |
4156d998 AE |
3260 | u64 names_size = 0; |
3261 | u32 want_count; | |
3262 | int ret; | |
602adf40 | 3263 | |
00f1f36f | 3264 | /* |
4156d998 AE |
3265 | * The complete header will include an array of its 64-bit |
3266 | * snapshot ids, followed by the names of those snapshots as | |
3267 | * a contiguous block of NUL-terminated strings. Note that | |
3268 | * the number of snapshots could change by the time we read | |
3269 | * it in, in which case we re-read it. | |
00f1f36f | 3270 | */ |
4156d998 AE |
3271 | do { |
3272 | size_t size; | |
3273 | ||
3274 | kfree(ondisk); | |
3275 | ||
3276 | size = sizeof (*ondisk); | |
3277 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3278 | size += names_size; | |
3279 | ondisk = kmalloc(size, GFP_KERNEL); | |
3280 | if (!ondisk) | |
662518b1 | 3281 | return -ENOMEM; |
4156d998 | 3282 | |
788e2df3 | 3283 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3284 | 0, size, ondisk); |
4156d998 | 3285 | if (ret < 0) |
662518b1 | 3286 | goto out; |
c0cd10db | 3287 | if ((size_t)ret < size) { |
4156d998 | 3288 | ret = -ENXIO; |
06ecc6cb AE |
3289 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3290 | size, ret); | |
662518b1 | 3291 | goto out; |
4156d998 AE |
3292 | } |
3293 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3294 | ret = -ENXIO; | |
06ecc6cb | 3295 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3296 | goto out; |
81e759fb | 3297 | } |
602adf40 | 3298 | |
4156d998 AE |
3299 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3300 | want_count = snap_count; | |
3301 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3302 | } while (snap_count != want_count); | |
00f1f36f | 3303 | |
662518b1 AE |
3304 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3305 | out: | |
4156d998 AE |
3306 | kfree(ondisk); |
3307 | ||
3308 | return ret; | |
602adf40 YS |
3309 | } |
3310 | ||
15228ede AE |
3311 | /* |
3312 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3313 | * has disappeared from the (just updated) snapshot context. | |
3314 | */ | |
3315 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3316 | { | |
3317 | u64 snap_id; | |
3318 | ||
3319 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3320 | return; | |
3321 | ||
3322 | snap_id = rbd_dev->spec->snap_id; | |
3323 | if (snap_id == CEPH_NOSNAP) | |
3324 | return; | |
3325 | ||
3326 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3327 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3328 | } | |
3329 | ||
9875201e JD |
3330 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3331 | { | |
3332 | sector_t size; | |
3333 | bool removing; | |
3334 | ||
3335 | /* | |
3336 | * Don't hold the lock while doing disk operations, | |
3337 | * or lock ordering will conflict with the bdev mutex via: | |
3338 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3339 | */ | |
3340 | spin_lock_irq(&rbd_dev->lock); | |
3341 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3342 | spin_unlock_irq(&rbd_dev->lock); | |
3343 | /* | |
3344 | * If the device is being removed, rbd_dev->disk has | |
3345 | * been destroyed, so don't try to update its size | |
3346 | */ | |
3347 | if (!removing) { | |
3348 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3349 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3350 | set_capacity(rbd_dev->disk, size); | |
3351 | revalidate_disk(rbd_dev->disk); | |
3352 | } | |
3353 | } | |
3354 | ||
cc4a38bd | 3355 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3356 | { |
e627db08 | 3357 | u64 mapping_size; |
1fe5e993 AE |
3358 | int ret; |
3359 | ||
117973fb | 3360 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
cfbf6377 | 3361 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3362 | mapping_size = rbd_dev->mapping.size; |
117973fb | 3363 | if (rbd_dev->image_format == 1) |
99a41ebc | 3364 | ret = rbd_dev_v1_header_info(rbd_dev); |
117973fb | 3365 | else |
2df3fac7 | 3366 | ret = rbd_dev_v2_header_info(rbd_dev); |
15228ede AE |
3367 | |
3368 | /* If it's a mapped snapshot, validate its EXISTS flag */ | |
3369 | ||
3370 | rbd_exists_validate(rbd_dev); | |
cfbf6377 AE |
3371 | up_write(&rbd_dev->header_rwsem); |
3372 | ||
00a653e2 | 3373 | if (mapping_size != rbd_dev->mapping.size) { |
9875201e | 3374 | rbd_dev_update_size(rbd_dev); |
00a653e2 | 3375 | } |
1fe5e993 AE |
3376 | |
3377 | return ret; | |
3378 | } | |
3379 | ||
602adf40 YS |
3380 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3381 | { | |
3382 | struct gendisk *disk; | |
3383 | struct request_queue *q; | |
593a9e7b | 3384 | u64 segment_size; |
602adf40 | 3385 | |
602adf40 | 3386 | /* create gendisk info */ |
7e513d43 ID |
3387 | disk = alloc_disk(single_major ? |
3388 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3389 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3390 | if (!disk) |
1fcdb8aa | 3391 | return -ENOMEM; |
602adf40 | 3392 | |
f0f8cef5 | 3393 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3394 | rbd_dev->dev_id); |
602adf40 | 3395 | disk->major = rbd_dev->major; |
dd82fff1 | 3396 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3397 | if (single_major) |
3398 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3399 | disk->fops = &rbd_bd_ops; |
3400 | disk->private_data = rbd_dev; | |
3401 | ||
bf0d5f50 | 3402 | q = blk_init_queue(rbd_request_fn, &rbd_dev->lock); |
602adf40 YS |
3403 | if (!q) |
3404 | goto out_disk; | |
029bcbd8 | 3405 | |
593a9e7b AE |
3406 | /* We use the default size, but let's be explicit about it. */ |
3407 | blk_queue_physical_block_size(q, SECTOR_SIZE); | |
3408 | ||
029bcbd8 | 3409 | /* set io sizes to object size */ |
593a9e7b AE |
3410 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3411 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
3412 | blk_queue_max_segment_size(q, segment_size); | |
3413 | blk_queue_io_min(q, segment_size); | |
3414 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3415 | |
602adf40 YS |
3416 | blk_queue_merge_bvec(q, rbd_merge_bvec); |
3417 | disk->queue = q; | |
3418 | ||
3419 | q->queuedata = rbd_dev; | |
3420 | ||
3421 | rbd_dev->disk = disk; | |
602adf40 | 3422 | |
602adf40 | 3423 | return 0; |
602adf40 YS |
3424 | out_disk: |
3425 | put_disk(disk); | |
1fcdb8aa AE |
3426 | |
3427 | return -ENOMEM; | |
602adf40 YS |
3428 | } |
3429 | ||
dfc5606d YS |
3430 | /* |
3431 | sysfs | |
3432 | */ | |
3433 | ||
593a9e7b AE |
3434 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3435 | { | |
3436 | return container_of(dev, struct rbd_device, dev); | |
3437 | } | |
3438 | ||
dfc5606d YS |
3439 | static ssize_t rbd_size_show(struct device *dev, |
3440 | struct device_attribute *attr, char *buf) | |
3441 | { | |
593a9e7b | 3442 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3443 | |
fc71d833 AE |
3444 | return sprintf(buf, "%llu\n", |
3445 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3446 | } |
3447 | ||
34b13184 AE |
3448 | /* |
3449 | * Note this shows the features for whatever's mapped, which is not | |
3450 | * necessarily the base image. | |
3451 | */ | |
3452 | static ssize_t rbd_features_show(struct device *dev, | |
3453 | struct device_attribute *attr, char *buf) | |
3454 | { | |
3455 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3456 | ||
3457 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3458 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3459 | } |
3460 | ||
dfc5606d YS |
3461 | static ssize_t rbd_major_show(struct device *dev, |
3462 | struct device_attribute *attr, char *buf) | |
3463 | { | |
593a9e7b | 3464 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3465 | |
fc71d833 AE |
3466 | if (rbd_dev->major) |
3467 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3468 | ||
3469 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3470 | } |
3471 | ||
3472 | static ssize_t rbd_minor_show(struct device *dev, | |
3473 | struct device_attribute *attr, char *buf) | |
3474 | { | |
3475 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3476 | |
dd82fff1 | 3477 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3478 | } |
3479 | ||
3480 | static ssize_t rbd_client_id_show(struct device *dev, | |
3481 | struct device_attribute *attr, char *buf) | |
602adf40 | 3482 | { |
593a9e7b | 3483 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3484 | |
1dbb4399 AE |
3485 | return sprintf(buf, "client%lld\n", |
3486 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3487 | } |
3488 | ||
dfc5606d YS |
3489 | static ssize_t rbd_pool_show(struct device *dev, |
3490 | struct device_attribute *attr, char *buf) | |
602adf40 | 3491 | { |
593a9e7b | 3492 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3493 | |
0d7dbfce | 3494 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3495 | } |
3496 | ||
9bb2f334 AE |
3497 | static ssize_t rbd_pool_id_show(struct device *dev, |
3498 | struct device_attribute *attr, char *buf) | |
3499 | { | |
3500 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3501 | ||
0d7dbfce | 3502 | return sprintf(buf, "%llu\n", |
fc71d833 | 3503 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3504 | } |
3505 | ||
dfc5606d YS |
3506 | static ssize_t rbd_name_show(struct device *dev, |
3507 | struct device_attribute *attr, char *buf) | |
3508 | { | |
593a9e7b | 3509 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3510 | |
a92ffdf8 AE |
3511 | if (rbd_dev->spec->image_name) |
3512 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3513 | ||
3514 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3515 | } |
3516 | ||
589d30e0 AE |
3517 | static ssize_t rbd_image_id_show(struct device *dev, |
3518 | struct device_attribute *attr, char *buf) | |
3519 | { | |
3520 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3521 | ||
0d7dbfce | 3522 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3523 | } |
3524 | ||
34b13184 AE |
3525 | /* |
3526 | * Shows the name of the currently-mapped snapshot (or | |
3527 | * RBD_SNAP_HEAD_NAME for the base image). | |
3528 | */ | |
dfc5606d YS |
3529 | static ssize_t rbd_snap_show(struct device *dev, |
3530 | struct device_attribute *attr, | |
3531 | char *buf) | |
3532 | { | |
593a9e7b | 3533 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3534 | |
0d7dbfce | 3535 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3536 | } |
3537 | ||
86b00e0d AE |
3538 | /* |
3539 | * For an rbd v2 image, shows the pool id, image id, and snapshot id | |
3540 | * for the parent image. If there is no parent, simply shows | |
3541 | * "(no parent image)". | |
3542 | */ | |
3543 | static ssize_t rbd_parent_show(struct device *dev, | |
3544 | struct device_attribute *attr, | |
3545 | char *buf) | |
3546 | { | |
3547 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3548 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3549 | int count; | |
3550 | char *bufp = buf; | |
3551 | ||
3552 | if (!spec) | |
3553 | return sprintf(buf, "(no parent image)\n"); | |
3554 | ||
3555 | count = sprintf(bufp, "pool_id %llu\npool_name %s\n", | |
3556 | (unsigned long long) spec->pool_id, spec->pool_name); | |
3557 | if (count < 0) | |
3558 | return count; | |
3559 | bufp += count; | |
3560 | ||
3561 | count = sprintf(bufp, "image_id %s\nimage_name %s\n", spec->image_id, | |
3562 | spec->image_name ? spec->image_name : "(unknown)"); | |
3563 | if (count < 0) | |
3564 | return count; | |
3565 | bufp += count; | |
3566 | ||
3567 | count = sprintf(bufp, "snap_id %llu\nsnap_name %s\n", | |
3568 | (unsigned long long) spec->snap_id, spec->snap_name); | |
3569 | if (count < 0) | |
3570 | return count; | |
3571 | bufp += count; | |
3572 | ||
3573 | count = sprintf(bufp, "overlap %llu\n", rbd_dev->parent_overlap); | |
3574 | if (count < 0) | |
3575 | return count; | |
3576 | bufp += count; | |
3577 | ||
3578 | return (ssize_t) (bufp - buf); | |
3579 | } | |
3580 | ||
dfc5606d YS |
3581 | static ssize_t rbd_image_refresh(struct device *dev, |
3582 | struct device_attribute *attr, | |
3583 | const char *buf, | |
3584 | size_t size) | |
3585 | { | |
593a9e7b | 3586 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3587 | int ret; |
602adf40 | 3588 | |
cc4a38bd | 3589 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 AE |
3590 | if (ret) |
3591 | rbd_warn(rbd_dev, ": manual header refresh error (%d)\n", ret); | |
b813623a AE |
3592 | |
3593 | return ret < 0 ? ret : size; | |
dfc5606d | 3594 | } |
602adf40 | 3595 | |
dfc5606d | 3596 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3597 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3598 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3599 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3600 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3601 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3602 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3603 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3604 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3605 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3606 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3607 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3608 | |
3609 | static struct attribute *rbd_attrs[] = { | |
3610 | &dev_attr_size.attr, | |
34b13184 | 3611 | &dev_attr_features.attr, |
dfc5606d | 3612 | &dev_attr_major.attr, |
dd82fff1 | 3613 | &dev_attr_minor.attr, |
dfc5606d YS |
3614 | &dev_attr_client_id.attr, |
3615 | &dev_attr_pool.attr, | |
9bb2f334 | 3616 | &dev_attr_pool_id.attr, |
dfc5606d | 3617 | &dev_attr_name.attr, |
589d30e0 | 3618 | &dev_attr_image_id.attr, |
dfc5606d | 3619 | &dev_attr_current_snap.attr, |
86b00e0d | 3620 | &dev_attr_parent.attr, |
dfc5606d | 3621 | &dev_attr_refresh.attr, |
dfc5606d YS |
3622 | NULL |
3623 | }; | |
3624 | ||
3625 | static struct attribute_group rbd_attr_group = { | |
3626 | .attrs = rbd_attrs, | |
3627 | }; | |
3628 | ||
3629 | static const struct attribute_group *rbd_attr_groups[] = { | |
3630 | &rbd_attr_group, | |
3631 | NULL | |
3632 | }; | |
3633 | ||
3634 | static void rbd_sysfs_dev_release(struct device *dev) | |
3635 | { | |
3636 | } | |
3637 | ||
3638 | static struct device_type rbd_device_type = { | |
3639 | .name = "rbd", | |
3640 | .groups = rbd_attr_groups, | |
3641 | .release = rbd_sysfs_dev_release, | |
3642 | }; | |
3643 | ||
8b8fb99c AE |
3644 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3645 | { | |
3646 | kref_get(&spec->kref); | |
3647 | ||
3648 | return spec; | |
3649 | } | |
3650 | ||
3651 | static void rbd_spec_free(struct kref *kref); | |
3652 | static void rbd_spec_put(struct rbd_spec *spec) | |
3653 | { | |
3654 | if (spec) | |
3655 | kref_put(&spec->kref, rbd_spec_free); | |
3656 | } | |
3657 | ||
3658 | static struct rbd_spec *rbd_spec_alloc(void) | |
3659 | { | |
3660 | struct rbd_spec *spec; | |
3661 | ||
3662 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
3663 | if (!spec) | |
3664 | return NULL; | |
3665 | kref_init(&spec->kref); | |
3666 | ||
8b8fb99c AE |
3667 | return spec; |
3668 | } | |
3669 | ||
3670 | static void rbd_spec_free(struct kref *kref) | |
3671 | { | |
3672 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
3673 | ||
3674 | kfree(spec->pool_name); | |
3675 | kfree(spec->image_id); | |
3676 | kfree(spec->image_name); | |
3677 | kfree(spec->snap_name); | |
3678 | kfree(spec); | |
3679 | } | |
3680 | ||
cc344fa1 | 3681 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
c53d5893 AE |
3682 | struct rbd_spec *spec) |
3683 | { | |
3684 | struct rbd_device *rbd_dev; | |
3685 | ||
3686 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
3687 | if (!rbd_dev) | |
3688 | return NULL; | |
3689 | ||
3690 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 3691 | rbd_dev->flags = 0; |
a2acd00e | 3692 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 3693 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
3694 | init_rwsem(&rbd_dev->header_rwsem); |
3695 | ||
3696 | rbd_dev->spec = spec; | |
3697 | rbd_dev->rbd_client = rbdc; | |
3698 | ||
0903e875 AE |
3699 | /* Initialize the layout used for all rbd requests */ |
3700 | ||
3701 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3702 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
3703 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
3704 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
3705 | ||
c53d5893 AE |
3706 | return rbd_dev; |
3707 | } | |
3708 | ||
3709 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
3710 | { | |
c53d5893 AE |
3711 | rbd_put_client(rbd_dev->rbd_client); |
3712 | rbd_spec_put(rbd_dev->spec); | |
3713 | kfree(rbd_dev); | |
3714 | } | |
3715 | ||
9d475de5 AE |
3716 | /* |
3717 | * Get the size and object order for an image snapshot, or if | |
3718 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
3719 | * image. | |
3720 | */ | |
3721 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
3722 | u8 *order, u64 *snap_size) | |
3723 | { | |
3724 | __le64 snapid = cpu_to_le64(snap_id); | |
3725 | int ret; | |
3726 | struct { | |
3727 | u8 order; | |
3728 | __le64 size; | |
3729 | } __attribute__ ((packed)) size_buf = { 0 }; | |
3730 | ||
36be9a76 | 3731 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 3732 | "rbd", "get_size", |
4157976b | 3733 | &snapid, sizeof (snapid), |
e2a58ee5 | 3734 | &size_buf, sizeof (size_buf)); |
36be9a76 | 3735 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
3736 | if (ret < 0) |
3737 | return ret; | |
57385b51 AE |
3738 | if (ret < sizeof (size_buf)) |
3739 | return -ERANGE; | |
9d475de5 | 3740 | |
c3545579 | 3741 | if (order) { |
c86f86e9 | 3742 | *order = size_buf.order; |
c3545579 JD |
3743 | dout(" order %u", (unsigned int)*order); |
3744 | } | |
9d475de5 AE |
3745 | *snap_size = le64_to_cpu(size_buf.size); |
3746 | ||
c3545579 JD |
3747 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
3748 | (unsigned long long)snap_id, | |
57385b51 | 3749 | (unsigned long long)*snap_size); |
9d475de5 AE |
3750 | |
3751 | return 0; | |
3752 | } | |
3753 | ||
3754 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
3755 | { | |
3756 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
3757 | &rbd_dev->header.obj_order, | |
3758 | &rbd_dev->header.image_size); | |
3759 | } | |
3760 | ||
1e130199 AE |
3761 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
3762 | { | |
3763 | void *reply_buf; | |
3764 | int ret; | |
3765 | void *p; | |
3766 | ||
3767 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
3768 | if (!reply_buf) | |
3769 | return -ENOMEM; | |
3770 | ||
36be9a76 | 3771 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 3772 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 3773 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 3774 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
3775 | if (ret < 0) |
3776 | goto out; | |
3777 | ||
3778 | p = reply_buf; | |
3779 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
3780 | p + ret, NULL, GFP_NOIO); |
3781 | ret = 0; | |
1e130199 AE |
3782 | |
3783 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
3784 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
3785 | rbd_dev->header.object_prefix = NULL; | |
3786 | } else { | |
3787 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
3788 | } | |
1e130199 AE |
3789 | out: |
3790 | kfree(reply_buf); | |
3791 | ||
3792 | return ret; | |
3793 | } | |
3794 | ||
b1b5402a AE |
3795 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
3796 | u64 *snap_features) | |
3797 | { | |
3798 | __le64 snapid = cpu_to_le64(snap_id); | |
3799 | struct { | |
3800 | __le64 features; | |
3801 | __le64 incompat; | |
4157976b | 3802 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 3803 | u64 incompat; |
b1b5402a AE |
3804 | int ret; |
3805 | ||
36be9a76 | 3806 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 3807 | "rbd", "get_features", |
4157976b | 3808 | &snapid, sizeof (snapid), |
e2a58ee5 | 3809 | &features_buf, sizeof (features_buf)); |
36be9a76 | 3810 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
3811 | if (ret < 0) |
3812 | return ret; | |
57385b51 AE |
3813 | if (ret < sizeof (features_buf)) |
3814 | return -ERANGE; | |
d889140c AE |
3815 | |
3816 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 3817 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 3818 | return -ENXIO; |
d889140c | 3819 | |
b1b5402a AE |
3820 | *snap_features = le64_to_cpu(features_buf.features); |
3821 | ||
3822 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
3823 | (unsigned long long)snap_id, |
3824 | (unsigned long long)*snap_features, | |
3825 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
3826 | |
3827 | return 0; | |
3828 | } | |
3829 | ||
3830 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
3831 | { | |
3832 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
3833 | &rbd_dev->header.features); | |
3834 | } | |
3835 | ||
86b00e0d AE |
3836 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
3837 | { | |
3838 | struct rbd_spec *parent_spec; | |
3839 | size_t size; | |
3840 | void *reply_buf = NULL; | |
3841 | __le64 snapid; | |
3842 | void *p; | |
3843 | void *end; | |
642a2537 | 3844 | u64 pool_id; |
86b00e0d | 3845 | char *image_id; |
3b5cf2a2 | 3846 | u64 snap_id; |
86b00e0d | 3847 | u64 overlap; |
86b00e0d AE |
3848 | int ret; |
3849 | ||
3850 | parent_spec = rbd_spec_alloc(); | |
3851 | if (!parent_spec) | |
3852 | return -ENOMEM; | |
3853 | ||
3854 | size = sizeof (__le64) + /* pool_id */ | |
3855 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
3856 | sizeof (__le64) + /* snap_id */ | |
3857 | sizeof (__le64); /* overlap */ | |
3858 | reply_buf = kmalloc(size, GFP_KERNEL); | |
3859 | if (!reply_buf) { | |
3860 | ret = -ENOMEM; | |
3861 | goto out_err; | |
3862 | } | |
3863 | ||
3864 | snapid = cpu_to_le64(CEPH_NOSNAP); | |
36be9a76 | 3865 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 3866 | "rbd", "get_parent", |
4157976b | 3867 | &snapid, sizeof (snapid), |
e2a58ee5 | 3868 | reply_buf, size); |
36be9a76 | 3869 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
3870 | if (ret < 0) |
3871 | goto out_err; | |
3872 | ||
86b00e0d | 3873 | p = reply_buf; |
57385b51 AE |
3874 | end = reply_buf + ret; |
3875 | ret = -ERANGE; | |
642a2537 | 3876 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
3877 | if (pool_id == CEPH_NOPOOL) { |
3878 | /* | |
3879 | * Either the parent never existed, or we have | |
3880 | * record of it but the image got flattened so it no | |
3881 | * longer has a parent. When the parent of a | |
3882 | * layered image disappears we immediately set the | |
3883 | * overlap to 0. The effect of this is that all new | |
3884 | * requests will be treated as if the image had no | |
3885 | * parent. | |
3886 | */ | |
3887 | if (rbd_dev->parent_overlap) { | |
3888 | rbd_dev->parent_overlap = 0; | |
3889 | smp_mb(); | |
3890 | rbd_dev_parent_put(rbd_dev); | |
3891 | pr_info("%s: clone image has been flattened\n", | |
3892 | rbd_dev->disk->disk_name); | |
3893 | } | |
3894 | ||
86b00e0d | 3895 | goto out; /* No parent? No problem. */ |
392a9dad | 3896 | } |
86b00e0d | 3897 | |
0903e875 AE |
3898 | /* The ceph file layout needs to fit pool id in 32 bits */ |
3899 | ||
3900 | ret = -EIO; | |
642a2537 | 3901 | if (pool_id > (u64)U32_MAX) { |
c0cd10db | 3902 | rbd_warn(NULL, "parent pool id too large (%llu > %u)\n", |
642a2537 | 3903 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 3904 | goto out_err; |
c0cd10db | 3905 | } |
0903e875 | 3906 | |
979ed480 | 3907 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
3908 | if (IS_ERR(image_id)) { |
3909 | ret = PTR_ERR(image_id); | |
3910 | goto out_err; | |
3911 | } | |
3b5cf2a2 | 3912 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
3913 | ceph_decode_64_safe(&p, end, overlap, out_err); |
3914 | ||
3b5cf2a2 AE |
3915 | /* |
3916 | * The parent won't change (except when the clone is | |
3917 | * flattened, already handled that). So we only need to | |
3918 | * record the parent spec we have not already done so. | |
3919 | */ | |
3920 | if (!rbd_dev->parent_spec) { | |
3921 | parent_spec->pool_id = pool_id; | |
3922 | parent_spec->image_id = image_id; | |
3923 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
3924 | rbd_dev->parent_spec = parent_spec; |
3925 | parent_spec = NULL; /* rbd_dev now owns this */ | |
3b5cf2a2 AE |
3926 | } |
3927 | ||
3928 | /* | |
3929 | * We always update the parent overlap. If it's zero we | |
3930 | * treat it specially. | |
3931 | */ | |
3932 | rbd_dev->parent_overlap = overlap; | |
3933 | smp_mb(); | |
3934 | if (!overlap) { | |
3935 | ||
3936 | /* A null parent_spec indicates it's the initial probe */ | |
3937 | ||
3938 | if (parent_spec) { | |
3939 | /* | |
3940 | * The overlap has become zero, so the clone | |
3941 | * must have been resized down to 0 at some | |
3942 | * point. Treat this the same as a flatten. | |
3943 | */ | |
3944 | rbd_dev_parent_put(rbd_dev); | |
3945 | pr_info("%s: clone image now standalone\n", | |
3946 | rbd_dev->disk->disk_name); | |
3947 | } else { | |
3948 | /* | |
3949 | * For the initial probe, if we find the | |
3950 | * overlap is zero we just pretend there was | |
3951 | * no parent image. | |
3952 | */ | |
3953 | rbd_warn(rbd_dev, "ignoring parent of " | |
3954 | "clone with overlap 0\n"); | |
3955 | } | |
70cf49cf | 3956 | } |
86b00e0d AE |
3957 | out: |
3958 | ret = 0; | |
3959 | out_err: | |
3960 | kfree(reply_buf); | |
3961 | rbd_spec_put(parent_spec); | |
3962 | ||
3963 | return ret; | |
3964 | } | |
3965 | ||
cc070d59 AE |
3966 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
3967 | { | |
3968 | struct { | |
3969 | __le64 stripe_unit; | |
3970 | __le64 stripe_count; | |
3971 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
3972 | size_t size = sizeof (striping_info_buf); | |
3973 | void *p; | |
3974 | u64 obj_size; | |
3975 | u64 stripe_unit; | |
3976 | u64 stripe_count; | |
3977 | int ret; | |
3978 | ||
3979 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
3980 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 3981 | (char *)&striping_info_buf, size); |
cc070d59 AE |
3982 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
3983 | if (ret < 0) | |
3984 | return ret; | |
3985 | if (ret < size) | |
3986 | return -ERANGE; | |
3987 | ||
3988 | /* | |
3989 | * We don't actually support the "fancy striping" feature | |
3990 | * (STRIPINGV2) yet, but if the striping sizes are the | |
3991 | * defaults the behavior is the same as before. So find | |
3992 | * out, and only fail if the image has non-default values. | |
3993 | */ | |
3994 | ret = -EINVAL; | |
3995 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
3996 | p = &striping_info_buf; | |
3997 | stripe_unit = ceph_decode_64(&p); | |
3998 | if (stripe_unit != obj_size) { | |
3999 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4000 | "(got %llu want %llu)", | |
4001 | stripe_unit, obj_size); | |
4002 | return -EINVAL; | |
4003 | } | |
4004 | stripe_count = ceph_decode_64(&p); | |
4005 | if (stripe_count != 1) { | |
4006 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4007 | "(got %llu want 1)", stripe_count); | |
4008 | return -EINVAL; | |
4009 | } | |
500d0c0f AE |
4010 | rbd_dev->header.stripe_unit = stripe_unit; |
4011 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4012 | |
4013 | return 0; | |
4014 | } | |
4015 | ||
9e15b77d AE |
4016 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4017 | { | |
4018 | size_t image_id_size; | |
4019 | char *image_id; | |
4020 | void *p; | |
4021 | void *end; | |
4022 | size_t size; | |
4023 | void *reply_buf = NULL; | |
4024 | size_t len = 0; | |
4025 | char *image_name = NULL; | |
4026 | int ret; | |
4027 | ||
4028 | rbd_assert(!rbd_dev->spec->image_name); | |
4029 | ||
69e7a02f AE |
4030 | len = strlen(rbd_dev->spec->image_id); |
4031 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4032 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4033 | if (!image_id) | |
4034 | return NULL; | |
4035 | ||
4036 | p = image_id; | |
4157976b | 4037 | end = image_id + image_id_size; |
57385b51 | 4038 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4039 | |
4040 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4041 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4042 | if (!reply_buf) | |
4043 | goto out; | |
4044 | ||
36be9a76 | 4045 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4046 | "rbd", "dir_get_name", |
4047 | image_id, image_id_size, | |
e2a58ee5 | 4048 | reply_buf, size); |
9e15b77d AE |
4049 | if (ret < 0) |
4050 | goto out; | |
4051 | p = reply_buf; | |
f40eb349 AE |
4052 | end = reply_buf + ret; |
4053 | ||
9e15b77d AE |
4054 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4055 | if (IS_ERR(image_name)) | |
4056 | image_name = NULL; | |
4057 | else | |
4058 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4059 | out: | |
4060 | kfree(reply_buf); | |
4061 | kfree(image_id); | |
4062 | ||
4063 | return image_name; | |
4064 | } | |
4065 | ||
2ad3d716 AE |
4066 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4067 | { | |
4068 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4069 | const char *snap_name; | |
4070 | u32 which = 0; | |
4071 | ||
4072 | /* Skip over names until we find the one we are looking for */ | |
4073 | ||
4074 | snap_name = rbd_dev->header.snap_names; | |
4075 | while (which < snapc->num_snaps) { | |
4076 | if (!strcmp(name, snap_name)) | |
4077 | return snapc->snaps[which]; | |
4078 | snap_name += strlen(snap_name) + 1; | |
4079 | which++; | |
4080 | } | |
4081 | return CEPH_NOSNAP; | |
4082 | } | |
4083 | ||
4084 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4085 | { | |
4086 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4087 | u32 which; | |
4088 | bool found = false; | |
4089 | u64 snap_id; | |
4090 | ||
4091 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4092 | const char *snap_name; | |
4093 | ||
4094 | snap_id = snapc->snaps[which]; | |
4095 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4096 | if (IS_ERR(snap_name)) { |
4097 | /* ignore no-longer existing snapshots */ | |
4098 | if (PTR_ERR(snap_name) == -ENOENT) | |
4099 | continue; | |
4100 | else | |
4101 | break; | |
4102 | } | |
2ad3d716 AE |
4103 | found = !strcmp(name, snap_name); |
4104 | kfree(snap_name); | |
4105 | } | |
4106 | return found ? snap_id : CEPH_NOSNAP; | |
4107 | } | |
4108 | ||
4109 | /* | |
4110 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4111 | * no snapshot by that name is found, or if an error occurs. | |
4112 | */ | |
4113 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4114 | { | |
4115 | if (rbd_dev->image_format == 1) | |
4116 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4117 | ||
4118 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4119 | } | |
4120 | ||
9e15b77d | 4121 | /* |
2e9f7f1c AE |
4122 | * When an rbd image has a parent image, it is identified by the |
4123 | * pool, image, and snapshot ids (not names). This function fills | |
4124 | * in the names for those ids. (It's OK if we can't figure out the | |
4125 | * name for an image id, but the pool and snapshot ids should always | |
4126 | * exist and have names.) All names in an rbd spec are dynamically | |
4127 | * allocated. | |
e1d4213f AE |
4128 | * |
4129 | * When an image being mapped (not a parent) is probed, we have the | |
4130 | * pool name and pool id, image name and image id, and the snapshot | |
4131 | * name. The only thing we're missing is the snapshot id. | |
9e15b77d | 4132 | */ |
2e9f7f1c | 4133 | static int rbd_dev_spec_update(struct rbd_device *rbd_dev) |
9e15b77d | 4134 | { |
2e9f7f1c AE |
4135 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4136 | struct rbd_spec *spec = rbd_dev->spec; | |
4137 | const char *pool_name; | |
4138 | const char *image_name; | |
4139 | const char *snap_name; | |
9e15b77d AE |
4140 | int ret; |
4141 | ||
e1d4213f AE |
4142 | /* |
4143 | * An image being mapped will have the pool name (etc.), but | |
4144 | * we need to look up the snapshot id. | |
4145 | */ | |
2e9f7f1c AE |
4146 | if (spec->pool_name) { |
4147 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
2ad3d716 | 4148 | u64 snap_id; |
e1d4213f | 4149 | |
2ad3d716 AE |
4150 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); |
4151 | if (snap_id == CEPH_NOSNAP) | |
e1d4213f | 4152 | return -ENOENT; |
2ad3d716 | 4153 | spec->snap_id = snap_id; |
e1d4213f | 4154 | } else { |
2e9f7f1c | 4155 | spec->snap_id = CEPH_NOSNAP; |
e1d4213f AE |
4156 | } |
4157 | ||
4158 | return 0; | |
4159 | } | |
9e15b77d | 4160 | |
2e9f7f1c | 4161 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4162 | |
2e9f7f1c AE |
4163 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4164 | if (!pool_name) { | |
4165 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4166 | return -EIO; |
4167 | } | |
2e9f7f1c AE |
4168 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4169 | if (!pool_name) | |
9e15b77d AE |
4170 | return -ENOMEM; |
4171 | ||
4172 | /* Fetch the image name; tolerate failure here */ | |
4173 | ||
2e9f7f1c AE |
4174 | image_name = rbd_dev_image_name(rbd_dev); |
4175 | if (!image_name) | |
06ecc6cb | 4176 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4177 | |
2e9f7f1c | 4178 | /* Look up the snapshot name, and make a copy */ |
9e15b77d | 4179 | |
2e9f7f1c | 4180 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4181 | if (IS_ERR(snap_name)) { |
4182 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4183 | goto out_err; |
2e9f7f1c AE |
4184 | } |
4185 | ||
4186 | spec->pool_name = pool_name; | |
4187 | spec->image_name = image_name; | |
4188 | spec->snap_name = snap_name; | |
9e15b77d AE |
4189 | |
4190 | return 0; | |
4191 | out_err: | |
2e9f7f1c AE |
4192 | kfree(image_name); |
4193 | kfree(pool_name); | |
9e15b77d AE |
4194 | |
4195 | return ret; | |
4196 | } | |
4197 | ||
cc4a38bd | 4198 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4199 | { |
4200 | size_t size; | |
4201 | int ret; | |
4202 | void *reply_buf; | |
4203 | void *p; | |
4204 | void *end; | |
4205 | u64 seq; | |
4206 | u32 snap_count; | |
4207 | struct ceph_snap_context *snapc; | |
4208 | u32 i; | |
4209 | ||
4210 | /* | |
4211 | * We'll need room for the seq value (maximum snapshot id), | |
4212 | * snapshot count, and array of that many snapshot ids. | |
4213 | * For now we have a fixed upper limit on the number we're | |
4214 | * prepared to receive. | |
4215 | */ | |
4216 | size = sizeof (__le64) + sizeof (__le32) + | |
4217 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4218 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4219 | if (!reply_buf) | |
4220 | return -ENOMEM; | |
4221 | ||
36be9a76 | 4222 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4223 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4224 | reply_buf, size); |
36be9a76 | 4225 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4226 | if (ret < 0) |
4227 | goto out; | |
4228 | ||
35d489f9 | 4229 | p = reply_buf; |
57385b51 AE |
4230 | end = reply_buf + ret; |
4231 | ret = -ERANGE; | |
35d489f9 AE |
4232 | ceph_decode_64_safe(&p, end, seq, out); |
4233 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4234 | ||
4235 | /* | |
4236 | * Make sure the reported number of snapshot ids wouldn't go | |
4237 | * beyond the end of our buffer. But before checking that, | |
4238 | * make sure the computed size of the snapshot context we | |
4239 | * allocate is representable in a size_t. | |
4240 | */ | |
4241 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4242 | / sizeof (u64)) { | |
4243 | ret = -EINVAL; | |
4244 | goto out; | |
4245 | } | |
4246 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4247 | goto out; | |
468521c1 | 4248 | ret = 0; |
35d489f9 | 4249 | |
812164f8 | 4250 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4251 | if (!snapc) { |
4252 | ret = -ENOMEM; | |
4253 | goto out; | |
4254 | } | |
35d489f9 | 4255 | snapc->seq = seq; |
35d489f9 AE |
4256 | for (i = 0; i < snap_count; i++) |
4257 | snapc->snaps[i] = ceph_decode_64(&p); | |
4258 | ||
49ece554 | 4259 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4260 | rbd_dev->header.snapc = snapc; |
4261 | ||
4262 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4263 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4264 | out: |
4265 | kfree(reply_buf); | |
4266 | ||
57385b51 | 4267 | return ret; |
35d489f9 AE |
4268 | } |
4269 | ||
54cac61f AE |
4270 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4271 | u64 snap_id) | |
b8b1e2db AE |
4272 | { |
4273 | size_t size; | |
4274 | void *reply_buf; | |
54cac61f | 4275 | __le64 snapid; |
b8b1e2db AE |
4276 | int ret; |
4277 | void *p; | |
4278 | void *end; | |
b8b1e2db AE |
4279 | char *snap_name; |
4280 | ||
4281 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4282 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4283 | if (!reply_buf) | |
4284 | return ERR_PTR(-ENOMEM); | |
4285 | ||
54cac61f | 4286 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4287 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4288 | "rbd", "get_snapshot_name", |
54cac61f | 4289 | &snapid, sizeof (snapid), |
e2a58ee5 | 4290 | reply_buf, size); |
36be9a76 | 4291 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4292 | if (ret < 0) { |
4293 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4294 | goto out; |
f40eb349 | 4295 | } |
b8b1e2db AE |
4296 | |
4297 | p = reply_buf; | |
f40eb349 | 4298 | end = reply_buf + ret; |
e5c35534 | 4299 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4300 | if (IS_ERR(snap_name)) |
b8b1e2db | 4301 | goto out; |
b8b1e2db | 4302 | |
f40eb349 | 4303 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4304 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4305 | out: |
4306 | kfree(reply_buf); | |
4307 | ||
f40eb349 | 4308 | return snap_name; |
b8b1e2db AE |
4309 | } |
4310 | ||
2df3fac7 | 4311 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4312 | { |
2df3fac7 | 4313 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4314 | int ret; |
117973fb | 4315 | |
1617e40c JD |
4316 | ret = rbd_dev_v2_image_size(rbd_dev); |
4317 | if (ret) | |
cfbf6377 | 4318 | return ret; |
1617e40c | 4319 | |
2df3fac7 AE |
4320 | if (first_time) { |
4321 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4322 | if (ret) | |
cfbf6377 | 4323 | return ret; |
2df3fac7 AE |
4324 | } |
4325 | ||
642a2537 AE |
4326 | /* |
4327 | * If the image supports layering, get the parent info. We | |
4328 | * need to probe the first time regardless. Thereafter we | |
4329 | * only need to if there's a parent, to see if it has | |
4330 | * disappeared due to the mapped image getting flattened. | |
4331 | */ | |
4332 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING && | |
4333 | (first_time || rbd_dev->parent_spec)) { | |
4334 | bool warn; | |
4335 | ||
4336 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
4337 | if (ret) | |
cfbf6377 | 4338 | return ret; |
642a2537 AE |
4339 | |
4340 | /* | |
4341 | * Print a warning if this is the initial probe and | |
4342 | * the image has a parent. Don't print it if the | |
4343 | * image now being probed is itself a parent. We | |
4344 | * can tell at this point because we won't know its | |
4345 | * pool name yet (just its pool id). | |
4346 | */ | |
4347 | warn = rbd_dev->parent_spec && rbd_dev->spec->pool_name; | |
4348 | if (first_time && warn) | |
4349 | rbd_warn(rbd_dev, "WARNING: kernel layering " | |
4350 | "is EXPERIMENTAL!"); | |
4351 | } | |
4352 | ||
29334ba4 AE |
4353 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) |
4354 | if (rbd_dev->mapping.size != rbd_dev->header.image_size) | |
4355 | rbd_dev->mapping.size = rbd_dev->header.image_size; | |
117973fb | 4356 | |
cc4a38bd | 4357 | ret = rbd_dev_v2_snap_context(rbd_dev); |
117973fb | 4358 | dout("rbd_dev_v2_snap_context returned %d\n", ret); |
117973fb AE |
4359 | |
4360 | return ret; | |
4361 | } | |
4362 | ||
dfc5606d YS |
4363 | static int rbd_bus_add_dev(struct rbd_device *rbd_dev) |
4364 | { | |
dfc5606d | 4365 | struct device *dev; |
cd789ab9 | 4366 | int ret; |
dfc5606d | 4367 | |
cd789ab9 | 4368 | dev = &rbd_dev->dev; |
dfc5606d YS |
4369 | dev->bus = &rbd_bus_type; |
4370 | dev->type = &rbd_device_type; | |
4371 | dev->parent = &rbd_root_dev; | |
200a6a8b | 4372 | dev->release = rbd_dev_device_release; |
de71a297 | 4373 | dev_set_name(dev, "%d", rbd_dev->dev_id); |
dfc5606d | 4374 | ret = device_register(dev); |
dfc5606d | 4375 | |
dfc5606d | 4376 | return ret; |
602adf40 YS |
4377 | } |
4378 | ||
dfc5606d YS |
4379 | static void rbd_bus_del_dev(struct rbd_device *rbd_dev) |
4380 | { | |
4381 | device_unregister(&rbd_dev->dev); | |
4382 | } | |
4383 | ||
1ddbe94e | 4384 | /* |
499afd5b | 4385 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4386 | * the rbd_dev to the global list. |
1ddbe94e | 4387 | */ |
f8a22fc2 | 4388 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4389 | { |
f8a22fc2 ID |
4390 | int new_dev_id; |
4391 | ||
9b60e70b ID |
4392 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4393 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4394 | GFP_KERNEL); | |
f8a22fc2 ID |
4395 | if (new_dev_id < 0) |
4396 | return new_dev_id; | |
4397 | ||
4398 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4399 | |
4400 | spin_lock(&rbd_dev_list_lock); | |
4401 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4402 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4403 | |
70eebd20 | 4404 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4405 | |
4406 | return 0; | |
1ddbe94e | 4407 | } |
b7f23c36 | 4408 | |
1ddbe94e | 4409 | /* |
499afd5b AE |
4410 | * Remove an rbd_dev from the global list, and record that its |
4411 | * identifier is no longer in use. | |
1ddbe94e | 4412 | */ |
e2839308 | 4413 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4414 | { |
499afd5b AE |
4415 | spin_lock(&rbd_dev_list_lock); |
4416 | list_del_init(&rbd_dev->node); | |
4417 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4418 | |
f8a22fc2 ID |
4419 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4420 | ||
4421 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4422 | } |
4423 | ||
e28fff26 AE |
4424 | /* |
4425 | * Skips over white space at *buf, and updates *buf to point to the | |
4426 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4427 | * the token (string of non-white space characters) found. Note |
4428 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4429 | */ |
4430 | static inline size_t next_token(const char **buf) | |
4431 | { | |
4432 | /* | |
4433 | * These are the characters that produce nonzero for | |
4434 | * isspace() in the "C" and "POSIX" locales. | |
4435 | */ | |
4436 | const char *spaces = " \f\n\r\t\v"; | |
4437 | ||
4438 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4439 | ||
4440 | return strcspn(*buf, spaces); /* Return token length */ | |
4441 | } | |
4442 | ||
4443 | /* | |
4444 | * Finds the next token in *buf, and if the provided token buffer is | |
4445 | * big enough, copies the found token into it. The result, if | |
593a9e7b AE |
4446 | * copied, is guaranteed to be terminated with '\0'. Note that *buf |
4447 | * must be terminated with '\0' on entry. | |
e28fff26 AE |
4448 | * |
4449 | * Returns the length of the token found (not including the '\0'). | |
4450 | * Return value will be 0 if no token is found, and it will be >= | |
4451 | * token_size if the token would not fit. | |
4452 | * | |
593a9e7b | 4453 | * The *buf pointer will be updated to point beyond the end of the |
e28fff26 AE |
4454 | * found token. Note that this occurs even if the token buffer is |
4455 | * too small to hold it. | |
4456 | */ | |
4457 | static inline size_t copy_token(const char **buf, | |
4458 | char *token, | |
4459 | size_t token_size) | |
4460 | { | |
4461 | size_t len; | |
4462 | ||
4463 | len = next_token(buf); | |
4464 | if (len < token_size) { | |
4465 | memcpy(token, *buf, len); | |
4466 | *(token + len) = '\0'; | |
4467 | } | |
4468 | *buf += len; | |
4469 | ||
4470 | return len; | |
4471 | } | |
4472 | ||
ea3352f4 AE |
4473 | /* |
4474 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4475 | * enough to hold a copy of it, and copies the token into the new | |
4476 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4477 | * that a duplicate buffer is created even for a zero-length token. | |
4478 | * | |
4479 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4480 | * pointer if memory for the duplicate was not available. If | |
4481 | * the lenp argument is a non-null pointer, the length of the token | |
4482 | * (not including the '\0') is returned in *lenp. | |
4483 | * | |
4484 | * If successful, the *buf pointer will be updated to point beyond | |
4485 | * the end of the found token. | |
4486 | * | |
4487 | * Note: uses GFP_KERNEL for allocation. | |
4488 | */ | |
4489 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4490 | { | |
4491 | char *dup; | |
4492 | size_t len; | |
4493 | ||
4494 | len = next_token(buf); | |
4caf35f9 | 4495 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4496 | if (!dup) |
4497 | return NULL; | |
ea3352f4 AE |
4498 | *(dup + len) = '\0'; |
4499 | *buf += len; | |
4500 | ||
4501 | if (lenp) | |
4502 | *lenp = len; | |
4503 | ||
4504 | return dup; | |
4505 | } | |
4506 | ||
a725f65e | 4507 | /* |
859c31df AE |
4508 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4509 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4510 | * and the data written is passed here via a NUL-terminated buffer. | |
4511 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4512 | * |
859c31df AE |
4513 | * The information extracted from these options is recorded in |
4514 | * the other parameters which return dynamically-allocated | |
4515 | * structures: | |
4516 | * ceph_opts | |
4517 | * The address of a pointer that will refer to a ceph options | |
4518 | * structure. Caller must release the returned pointer using | |
4519 | * ceph_destroy_options() when it is no longer needed. | |
4520 | * rbd_opts | |
4521 | * Address of an rbd options pointer. Fully initialized by | |
4522 | * this function; caller must release with kfree(). | |
4523 | * spec | |
4524 | * Address of an rbd image specification pointer. Fully | |
4525 | * initialized by this function based on parsed options. | |
4526 | * Caller must release with rbd_spec_put(). | |
4527 | * | |
4528 | * The options passed take this form: | |
4529 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4530 | * where: | |
4531 | * <mon_addrs> | |
4532 | * A comma-separated list of one or more monitor addresses. | |
4533 | * A monitor address is an ip address, optionally followed | |
4534 | * by a port number (separated by a colon). | |
4535 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4536 | * <options> | |
4537 | * A comma-separated list of ceph and/or rbd options. | |
4538 | * <pool_name> | |
4539 | * The name of the rados pool containing the rbd image. | |
4540 | * <image_name> | |
4541 | * The name of the image in that pool to map. | |
4542 | * <snap_id> | |
4543 | * An optional snapshot id. If provided, the mapping will | |
4544 | * present data from the image at the time that snapshot was | |
4545 | * created. The image head is used if no snapshot id is | |
4546 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4547 | */ |
859c31df | 4548 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4549 | struct ceph_options **ceph_opts, |
859c31df AE |
4550 | struct rbd_options **opts, |
4551 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4552 | { |
d22f76e7 | 4553 | size_t len; |
859c31df | 4554 | char *options; |
0ddebc0c | 4555 | const char *mon_addrs; |
ecb4dc22 | 4556 | char *snap_name; |
0ddebc0c | 4557 | size_t mon_addrs_size; |
859c31df | 4558 | struct rbd_spec *spec = NULL; |
4e9afeba | 4559 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4560 | struct ceph_options *copts; |
dc79b113 | 4561 | int ret; |
e28fff26 AE |
4562 | |
4563 | /* The first four tokens are required */ | |
4564 | ||
7ef3214a | 4565 | len = next_token(&buf); |
4fb5d671 AE |
4566 | if (!len) { |
4567 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4568 | return -EINVAL; | |
4569 | } | |
0ddebc0c | 4570 | mon_addrs = buf; |
f28e565a | 4571 | mon_addrs_size = len + 1; |
7ef3214a | 4572 | buf += len; |
a725f65e | 4573 | |
dc79b113 | 4574 | ret = -EINVAL; |
f28e565a AE |
4575 | options = dup_token(&buf, NULL); |
4576 | if (!options) | |
dc79b113 | 4577 | return -ENOMEM; |
4fb5d671 AE |
4578 | if (!*options) { |
4579 | rbd_warn(NULL, "no options provided"); | |
4580 | goto out_err; | |
4581 | } | |
e28fff26 | 4582 | |
859c31df AE |
4583 | spec = rbd_spec_alloc(); |
4584 | if (!spec) | |
f28e565a | 4585 | goto out_mem; |
859c31df AE |
4586 | |
4587 | spec->pool_name = dup_token(&buf, NULL); | |
4588 | if (!spec->pool_name) | |
4589 | goto out_mem; | |
4fb5d671 AE |
4590 | if (!*spec->pool_name) { |
4591 | rbd_warn(NULL, "no pool name provided"); | |
4592 | goto out_err; | |
4593 | } | |
e28fff26 | 4594 | |
69e7a02f | 4595 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4596 | if (!spec->image_name) |
f28e565a | 4597 | goto out_mem; |
4fb5d671 AE |
4598 | if (!*spec->image_name) { |
4599 | rbd_warn(NULL, "no image name provided"); | |
4600 | goto out_err; | |
4601 | } | |
d4b125e9 | 4602 | |
f28e565a AE |
4603 | /* |
4604 | * Snapshot name is optional; default is to use "-" | |
4605 | * (indicating the head/no snapshot). | |
4606 | */ | |
3feeb894 | 4607 | len = next_token(&buf); |
820a5f3e | 4608 | if (!len) { |
3feeb894 AE |
4609 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4610 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4611 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4612 | ret = -ENAMETOOLONG; |
f28e565a | 4613 | goto out_err; |
849b4260 | 4614 | } |
ecb4dc22 AE |
4615 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4616 | if (!snap_name) | |
f28e565a | 4617 | goto out_mem; |
ecb4dc22 AE |
4618 | *(snap_name + len) = '\0'; |
4619 | spec->snap_name = snap_name; | |
e5c35534 | 4620 | |
0ddebc0c | 4621 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4622 | |
4e9afeba AE |
4623 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4624 | if (!rbd_opts) | |
4625 | goto out_mem; | |
4626 | ||
4627 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
d22f76e7 | 4628 | |
859c31df | 4629 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4630 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4631 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4632 | if (IS_ERR(copts)) { |
4633 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4634 | goto out_err; |
4635 | } | |
859c31df AE |
4636 | kfree(options); |
4637 | ||
4638 | *ceph_opts = copts; | |
4e9afeba | 4639 | *opts = rbd_opts; |
859c31df | 4640 | *rbd_spec = spec; |
0ddebc0c | 4641 | |
dc79b113 | 4642 | return 0; |
f28e565a | 4643 | out_mem: |
dc79b113 | 4644 | ret = -ENOMEM; |
d22f76e7 | 4645 | out_err: |
859c31df AE |
4646 | kfree(rbd_opts); |
4647 | rbd_spec_put(spec); | |
f28e565a | 4648 | kfree(options); |
d22f76e7 | 4649 | |
dc79b113 | 4650 | return ret; |
a725f65e AE |
4651 | } |
4652 | ||
589d30e0 AE |
4653 | /* |
4654 | * An rbd format 2 image has a unique identifier, distinct from the | |
4655 | * name given to it by the user. Internally, that identifier is | |
4656 | * what's used to specify the names of objects related to the image. | |
4657 | * | |
4658 | * A special "rbd id" object is used to map an rbd image name to its | |
4659 | * id. If that object doesn't exist, then there is no v2 rbd image | |
4660 | * with the supplied name. | |
4661 | * | |
4662 | * This function will record the given rbd_dev's image_id field if | |
4663 | * it can be determined, and in that case will return 0. If any | |
4664 | * errors occur a negative errno will be returned and the rbd_dev's | |
4665 | * image_id field will be unchanged (and should be NULL). | |
4666 | */ | |
4667 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
4668 | { | |
4669 | int ret; | |
4670 | size_t size; | |
4671 | char *object_name; | |
4672 | void *response; | |
c0fba368 | 4673 | char *image_id; |
2f82ee54 | 4674 | |
2c0d0a10 AE |
4675 | /* |
4676 | * When probing a parent image, the image id is already | |
4677 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
4678 | * need to fetch the image id again in this case. We |
4679 | * do still need to set the image format though. | |
2c0d0a10 | 4680 | */ |
c0fba368 AE |
4681 | if (rbd_dev->spec->image_id) { |
4682 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
4683 | ||
2c0d0a10 | 4684 | return 0; |
c0fba368 | 4685 | } |
2c0d0a10 | 4686 | |
589d30e0 AE |
4687 | /* |
4688 | * First, see if the format 2 image id file exists, and if | |
4689 | * so, get the image's persistent id from it. | |
4690 | */ | |
69e7a02f | 4691 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
4692 | object_name = kmalloc(size, GFP_NOIO); |
4693 | if (!object_name) | |
4694 | return -ENOMEM; | |
0d7dbfce | 4695 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
4696 | dout("rbd id object name is %s\n", object_name); |
4697 | ||
4698 | /* Response will be an encoded string, which includes a length */ | |
4699 | ||
4700 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
4701 | response = kzalloc(size, GFP_NOIO); | |
4702 | if (!response) { | |
4703 | ret = -ENOMEM; | |
4704 | goto out; | |
4705 | } | |
4706 | ||
c0fba368 AE |
4707 | /* If it doesn't exist we'll assume it's a format 1 image */ |
4708 | ||
36be9a76 | 4709 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 4710 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 4711 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 4712 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
4713 | if (ret == -ENOENT) { |
4714 | image_id = kstrdup("", GFP_KERNEL); | |
4715 | ret = image_id ? 0 : -ENOMEM; | |
4716 | if (!ret) | |
4717 | rbd_dev->image_format = 1; | |
4718 | } else if (ret > sizeof (__le32)) { | |
4719 | void *p = response; | |
4720 | ||
4721 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 4722 | NULL, GFP_NOIO); |
c0fba368 AE |
4723 | ret = IS_ERR(image_id) ? PTR_ERR(image_id) : 0; |
4724 | if (!ret) | |
4725 | rbd_dev->image_format = 2; | |
589d30e0 | 4726 | } else { |
c0fba368 AE |
4727 | ret = -EINVAL; |
4728 | } | |
4729 | ||
4730 | if (!ret) { | |
4731 | rbd_dev->spec->image_id = image_id; | |
4732 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
4733 | } |
4734 | out: | |
4735 | kfree(response); | |
4736 | kfree(object_name); | |
4737 | ||
4738 | return ret; | |
4739 | } | |
4740 | ||
3abef3b3 AE |
4741 | /* |
4742 | * Undo whatever state changes are made by v1 or v2 header info | |
4743 | * call. | |
4744 | */ | |
6fd48b3b AE |
4745 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
4746 | { | |
4747 | struct rbd_image_header *header; | |
4748 | ||
392a9dad AE |
4749 | /* Drop parent reference unless it's already been done (or none) */ |
4750 | ||
4751 | if (rbd_dev->parent_overlap) | |
4752 | rbd_dev_parent_put(rbd_dev); | |
6fd48b3b AE |
4753 | |
4754 | /* Free dynamic fields from the header, then zero it out */ | |
4755 | ||
4756 | header = &rbd_dev->header; | |
812164f8 | 4757 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
4758 | kfree(header->snap_sizes); |
4759 | kfree(header->snap_names); | |
4760 | kfree(header->object_prefix); | |
4761 | memset(header, 0, sizeof (*header)); | |
4762 | } | |
4763 | ||
2df3fac7 | 4764 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
4765 | { |
4766 | int ret; | |
a30b71b9 | 4767 | |
1e130199 | 4768 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 4769 | if (ret) |
b1b5402a AE |
4770 | goto out_err; |
4771 | ||
2df3fac7 AE |
4772 | /* |
4773 | * Get the and check features for the image. Currently the | |
4774 | * features are assumed to never change. | |
4775 | */ | |
b1b5402a | 4776 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 4777 | if (ret) |
9d475de5 | 4778 | goto out_err; |
35d489f9 | 4779 | |
cc070d59 AE |
4780 | /* If the image supports fancy striping, get its parameters */ |
4781 | ||
4782 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
4783 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
4784 | if (ret < 0) | |
4785 | goto out_err; | |
4786 | } | |
2df3fac7 | 4787 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 4788 | |
35152979 | 4789 | return 0; |
9d475de5 | 4790 | out_err: |
642a2537 | 4791 | rbd_dev->header.features = 0; |
1e130199 AE |
4792 | kfree(rbd_dev->header.object_prefix); |
4793 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
4794 | |
4795 | return ret; | |
a30b71b9 AE |
4796 | } |
4797 | ||
124afba2 | 4798 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev) |
83a06263 | 4799 | { |
2f82ee54 | 4800 | struct rbd_device *parent = NULL; |
124afba2 AE |
4801 | struct rbd_spec *parent_spec; |
4802 | struct rbd_client *rbdc; | |
4803 | int ret; | |
4804 | ||
4805 | if (!rbd_dev->parent_spec) | |
4806 | return 0; | |
4807 | /* | |
4808 | * We need to pass a reference to the client and the parent | |
4809 | * spec when creating the parent rbd_dev. Images related by | |
4810 | * parent/child relationships always share both. | |
4811 | */ | |
4812 | parent_spec = rbd_spec_get(rbd_dev->parent_spec); | |
4813 | rbdc = __rbd_get_client(rbd_dev->rbd_client); | |
4814 | ||
4815 | ret = -ENOMEM; | |
4816 | parent = rbd_dev_create(rbdc, parent_spec); | |
4817 | if (!parent) | |
4818 | goto out_err; | |
4819 | ||
1f3ef788 | 4820 | ret = rbd_dev_image_probe(parent, false); |
124afba2 AE |
4821 | if (ret < 0) |
4822 | goto out_err; | |
4823 | rbd_dev->parent = parent; | |
a2acd00e | 4824 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 AE |
4825 | |
4826 | return 0; | |
4827 | out_err: | |
4828 | if (parent) { | |
fb65d228 | 4829 | rbd_dev_unparent(rbd_dev); |
124afba2 AE |
4830 | kfree(rbd_dev->header_name); |
4831 | rbd_dev_destroy(parent); | |
4832 | } else { | |
4833 | rbd_put_client(rbdc); | |
4834 | rbd_spec_put(parent_spec); | |
4835 | } | |
4836 | ||
4837 | return ret; | |
4838 | } | |
4839 | ||
200a6a8b | 4840 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 4841 | { |
83a06263 | 4842 | int ret; |
d1cf5788 | 4843 | |
f8a22fc2 ID |
4844 | /* Get an id and fill in device name. */ |
4845 | ||
4846 | ret = rbd_dev_id_get(rbd_dev); | |
4847 | if (ret) | |
4848 | return ret; | |
83a06263 | 4849 | |
83a06263 AE |
4850 | BUILD_BUG_ON(DEV_NAME_LEN |
4851 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
4852 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
4853 | ||
9b60e70b | 4854 | /* Record our major and minor device numbers. */ |
83a06263 | 4855 | |
9b60e70b ID |
4856 | if (!single_major) { |
4857 | ret = register_blkdev(0, rbd_dev->name); | |
4858 | if (ret < 0) | |
4859 | goto err_out_id; | |
4860 | ||
4861 | rbd_dev->major = ret; | |
4862 | rbd_dev->minor = 0; | |
4863 | } else { | |
4864 | rbd_dev->major = rbd_major; | |
4865 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
4866 | } | |
83a06263 AE |
4867 | |
4868 | /* Set up the blkdev mapping. */ | |
4869 | ||
4870 | ret = rbd_init_disk(rbd_dev); | |
4871 | if (ret) | |
4872 | goto err_out_blkdev; | |
4873 | ||
f35a4dee | 4874 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
4875 | if (ret) |
4876 | goto err_out_disk; | |
f35a4dee AE |
4877 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
4878 | ||
4879 | ret = rbd_bus_add_dev(rbd_dev); | |
4880 | if (ret) | |
4881 | goto err_out_mapping; | |
83a06263 | 4882 | |
83a06263 AE |
4883 | /* Everything's ready. Announce the disk to the world. */ |
4884 | ||
129b79d4 | 4885 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
4886 | add_disk(rbd_dev->disk); |
4887 | ||
4888 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
4889 | (unsigned long long) rbd_dev->mapping.size); | |
4890 | ||
4891 | return ret; | |
2f82ee54 | 4892 | |
f35a4dee AE |
4893 | err_out_mapping: |
4894 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
4895 | err_out_disk: |
4896 | rbd_free_disk(rbd_dev); | |
4897 | err_out_blkdev: | |
9b60e70b ID |
4898 | if (!single_major) |
4899 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
4900 | err_out_id: |
4901 | rbd_dev_id_put(rbd_dev); | |
d1cf5788 | 4902 | rbd_dev_mapping_clear(rbd_dev); |
83a06263 AE |
4903 | |
4904 | return ret; | |
4905 | } | |
4906 | ||
332bb12d AE |
4907 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
4908 | { | |
4909 | struct rbd_spec *spec = rbd_dev->spec; | |
4910 | size_t size; | |
4911 | ||
4912 | /* Record the header object name for this rbd image. */ | |
4913 | ||
4914 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4915 | ||
4916 | if (rbd_dev->image_format == 1) | |
4917 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
4918 | else | |
4919 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
4920 | ||
4921 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
4922 | if (!rbd_dev->header_name) | |
4923 | return -ENOMEM; | |
4924 | ||
4925 | if (rbd_dev->image_format == 1) | |
4926 | sprintf(rbd_dev->header_name, "%s%s", | |
4927 | spec->image_name, RBD_SUFFIX); | |
4928 | else | |
4929 | sprintf(rbd_dev->header_name, "%s%s", | |
4930 | RBD_HEADER_PREFIX, spec->image_id); | |
4931 | return 0; | |
4932 | } | |
4933 | ||
200a6a8b AE |
4934 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
4935 | { | |
6fd48b3b | 4936 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 4937 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
4938 | rbd_dev->header_name = NULL; |
4939 | rbd_dev->image_format = 0; | |
4940 | kfree(rbd_dev->spec->image_id); | |
4941 | rbd_dev->spec->image_id = NULL; | |
4942 | ||
200a6a8b AE |
4943 | rbd_dev_destroy(rbd_dev); |
4944 | } | |
4945 | ||
a30b71b9 AE |
4946 | /* |
4947 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
4948 | * device. If this image is the one being mapped (i.e., not a |
4949 | * parent), initiate a watch on its header object before using that | |
4950 | * object to get detailed information about the rbd image. | |
a30b71b9 | 4951 | */ |
1f3ef788 | 4952 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping) |
a30b71b9 AE |
4953 | { |
4954 | int ret; | |
4955 | ||
4956 | /* | |
3abef3b3 AE |
4957 | * Get the id from the image id object. Unless there's an |
4958 | * error, rbd_dev->spec->image_id will be filled in with | |
4959 | * a dynamically-allocated string, and rbd_dev->image_format | |
4960 | * will be set to either 1 or 2. | |
a30b71b9 AE |
4961 | */ |
4962 | ret = rbd_dev_image_id(rbd_dev); | |
4963 | if (ret) | |
c0fba368 AE |
4964 | return ret; |
4965 | rbd_assert(rbd_dev->spec->image_id); | |
4966 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4967 | ||
332bb12d AE |
4968 | ret = rbd_dev_header_name(rbd_dev); |
4969 | if (ret) | |
4970 | goto err_out_format; | |
4971 | ||
1f3ef788 | 4972 | if (mapping) { |
fca27065 | 4973 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1f3ef788 AE |
4974 | if (ret) |
4975 | goto out_header_name; | |
4976 | } | |
b644de2b | 4977 | |
c0fba368 | 4978 | if (rbd_dev->image_format == 1) |
99a41ebc | 4979 | ret = rbd_dev_v1_header_info(rbd_dev); |
a30b71b9 | 4980 | else |
2df3fac7 | 4981 | ret = rbd_dev_v2_header_info(rbd_dev); |
5655c4d9 | 4982 | if (ret) |
b644de2b | 4983 | goto err_out_watch; |
83a06263 | 4984 | |
9bb81c9b AE |
4985 | ret = rbd_dev_spec_update(rbd_dev); |
4986 | if (ret) | |
33dca39f | 4987 | goto err_out_probe; |
9bb81c9b AE |
4988 | |
4989 | ret = rbd_dev_probe_parent(rbd_dev); | |
30d60ba2 AE |
4990 | if (ret) |
4991 | goto err_out_probe; | |
4992 | ||
4993 | dout("discovered format %u image, header name is %s\n", | |
4994 | rbd_dev->image_format, rbd_dev->header_name); | |
83a06263 | 4995 | |
30d60ba2 | 4996 | return 0; |
6fd48b3b AE |
4997 | err_out_probe: |
4998 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 4999 | err_out_watch: |
fca27065 ID |
5000 | if (mapping) |
5001 | rbd_dev_header_unwatch_sync(rbd_dev); | |
332bb12d AE |
5002 | out_header_name: |
5003 | kfree(rbd_dev->header_name); | |
5004 | rbd_dev->header_name = NULL; | |
5005 | err_out_format: | |
5006 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5007 | kfree(rbd_dev->spec->image_id); |
5008 | rbd_dev->spec->image_id = NULL; | |
5009 | ||
5010 | dout("probe failed, returning %d\n", ret); | |
5011 | ||
a30b71b9 AE |
5012 | return ret; |
5013 | } | |
5014 | ||
9b60e70b ID |
5015 | static ssize_t do_rbd_add(struct bus_type *bus, |
5016 | const char *buf, | |
5017 | size_t count) | |
602adf40 | 5018 | { |
cb8627c7 | 5019 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5020 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5021 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5022 | struct rbd_spec *spec = NULL; |
9d3997fd | 5023 | struct rbd_client *rbdc; |
27cc2594 | 5024 | struct ceph_osd_client *osdc; |
51344a38 | 5025 | bool read_only; |
27cc2594 | 5026 | int rc = -ENOMEM; |
602adf40 YS |
5027 | |
5028 | if (!try_module_get(THIS_MODULE)) | |
5029 | return -ENODEV; | |
5030 | ||
602adf40 | 5031 | /* parse add command */ |
859c31df | 5032 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5033 | if (rc < 0) |
bd4ba655 | 5034 | goto err_out_module; |
51344a38 AE |
5035 | read_only = rbd_opts->read_only; |
5036 | kfree(rbd_opts); | |
5037 | rbd_opts = NULL; /* done with this */ | |
78cea76e | 5038 | |
9d3997fd AE |
5039 | rbdc = rbd_get_client(ceph_opts); |
5040 | if (IS_ERR(rbdc)) { | |
5041 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5042 | goto err_out_args; |
9d3997fd | 5043 | } |
602adf40 | 5044 | |
602adf40 | 5045 | /* pick the pool */ |
9d3997fd | 5046 | osdc = &rbdc->client->osdc; |
859c31df | 5047 | rc = ceph_pg_poolid_by_name(osdc->osdmap, spec->pool_name); |
602adf40 YS |
5048 | if (rc < 0) |
5049 | goto err_out_client; | |
c0cd10db | 5050 | spec->pool_id = (u64)rc; |
859c31df | 5051 | |
0903e875 AE |
5052 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5053 | ||
c0cd10db AE |
5054 | if (spec->pool_id > (u64)U32_MAX) { |
5055 | rbd_warn(NULL, "pool id too large (%llu > %u)\n", | |
5056 | (unsigned long long)spec->pool_id, U32_MAX); | |
0903e875 AE |
5057 | rc = -EIO; |
5058 | goto err_out_client; | |
5059 | } | |
5060 | ||
c53d5893 | 5061 | rbd_dev = rbd_dev_create(rbdc, spec); |
bd4ba655 AE |
5062 | if (!rbd_dev) |
5063 | goto err_out_client; | |
c53d5893 AE |
5064 | rbdc = NULL; /* rbd_dev now owns this */ |
5065 | spec = NULL; /* rbd_dev now owns this */ | |
602adf40 | 5066 | |
1f3ef788 | 5067 | rc = rbd_dev_image_probe(rbd_dev, true); |
a30b71b9 | 5068 | if (rc < 0) |
c53d5893 | 5069 | goto err_out_rbd_dev; |
05fd6f6f | 5070 | |
7ce4eef7 AE |
5071 | /* If we are mapping a snapshot it must be marked read-only */ |
5072 | ||
5073 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) | |
5074 | read_only = true; | |
5075 | rbd_dev->mapping.read_only = read_only; | |
5076 | ||
b536f69a | 5077 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5078 | if (rc) { |
e37180c0 ID |
5079 | /* |
5080 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5081 | * rbd_dev_image_release() without refactoring, see | |
5082 | * commit 1f3ef78861ac. | |
5083 | */ | |
5084 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 AE |
5085 | rbd_dev_image_release(rbd_dev); |
5086 | goto err_out_module; | |
5087 | } | |
5088 | ||
5089 | return count; | |
b536f69a | 5090 | |
c53d5893 AE |
5091 | err_out_rbd_dev: |
5092 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5093 | err_out_client: |
9d3997fd | 5094 | rbd_put_client(rbdc); |
0ddebc0c | 5095 | err_out_args: |
859c31df | 5096 | rbd_spec_put(spec); |
bd4ba655 AE |
5097 | err_out_module: |
5098 | module_put(THIS_MODULE); | |
27cc2594 | 5099 | |
602adf40 | 5100 | dout("Error adding device %s\n", buf); |
27cc2594 | 5101 | |
c0cd10db | 5102 | return (ssize_t)rc; |
602adf40 YS |
5103 | } |
5104 | ||
9b60e70b ID |
5105 | static ssize_t rbd_add(struct bus_type *bus, |
5106 | const char *buf, | |
5107 | size_t count) | |
5108 | { | |
5109 | if (single_major) | |
5110 | return -EINVAL; | |
5111 | ||
5112 | return do_rbd_add(bus, buf, count); | |
5113 | } | |
5114 | ||
5115 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5116 | const char *buf, | |
5117 | size_t count) | |
5118 | { | |
5119 | return do_rbd_add(bus, buf, count); | |
5120 | } | |
5121 | ||
200a6a8b | 5122 | static void rbd_dev_device_release(struct device *dev) |
602adf40 | 5123 | { |
593a9e7b | 5124 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 5125 | |
602adf40 | 5126 | rbd_free_disk(rbd_dev); |
200a6a8b | 5127 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
6d80b130 | 5128 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5129 | if (!single_major) |
5130 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5131 | rbd_dev_id_put(rbd_dev); |
d1cf5788 | 5132 | rbd_dev_mapping_clear(rbd_dev); |
602adf40 YS |
5133 | } |
5134 | ||
05a46afd AE |
5135 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5136 | { | |
ad945fc1 | 5137 | while (rbd_dev->parent) { |
05a46afd AE |
5138 | struct rbd_device *first = rbd_dev; |
5139 | struct rbd_device *second = first->parent; | |
5140 | struct rbd_device *third; | |
5141 | ||
5142 | /* | |
5143 | * Follow to the parent with no grandparent and | |
5144 | * remove it. | |
5145 | */ | |
5146 | while (second && (third = second->parent)) { | |
5147 | first = second; | |
5148 | second = third; | |
5149 | } | |
ad945fc1 | 5150 | rbd_assert(second); |
8ad42cd0 | 5151 | rbd_dev_image_release(second); |
ad945fc1 AE |
5152 | first->parent = NULL; |
5153 | first->parent_overlap = 0; | |
5154 | ||
5155 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5156 | rbd_spec_put(first->parent_spec); |
5157 | first->parent_spec = NULL; | |
05a46afd AE |
5158 | } |
5159 | } | |
5160 | ||
9b60e70b ID |
5161 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5162 | const char *buf, | |
5163 | size_t count) | |
602adf40 YS |
5164 | { |
5165 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5166 | struct list_head *tmp; |
5167 | int dev_id; | |
602adf40 | 5168 | unsigned long ul; |
82a442d2 | 5169 | bool already = false; |
0d8189e1 | 5170 | int ret; |
602adf40 | 5171 | |
bb8e0e84 | 5172 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5173 | if (ret) |
5174 | return ret; | |
602adf40 YS |
5175 | |
5176 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5177 | dev_id = (int)ul; |
5178 | if (dev_id != ul) | |
602adf40 YS |
5179 | return -EINVAL; |
5180 | ||
751cc0e3 AE |
5181 | ret = -ENOENT; |
5182 | spin_lock(&rbd_dev_list_lock); | |
5183 | list_for_each(tmp, &rbd_dev_list) { | |
5184 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5185 | if (rbd_dev->dev_id == dev_id) { | |
5186 | ret = 0; | |
5187 | break; | |
5188 | } | |
42382b70 | 5189 | } |
751cc0e3 AE |
5190 | if (!ret) { |
5191 | spin_lock_irq(&rbd_dev->lock); | |
5192 | if (rbd_dev->open_count) | |
5193 | ret = -EBUSY; | |
5194 | else | |
82a442d2 AE |
5195 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5196 | &rbd_dev->flags); | |
751cc0e3 AE |
5197 | spin_unlock_irq(&rbd_dev->lock); |
5198 | } | |
5199 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5200 | if (ret < 0 || already) |
1ba0f1e7 | 5201 | return ret; |
751cc0e3 | 5202 | |
fca27065 | 5203 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5204 | /* |
5205 | * flush remaining watch callbacks - these must be complete | |
5206 | * before the osd_client is shutdown | |
5207 | */ | |
5208 | dout("%s: flushing notifies", __func__); | |
5209 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5210 | |
9875201e JD |
5211 | /* |
5212 | * Don't free anything from rbd_dev->disk until after all | |
5213 | * notifies are completely processed. Otherwise | |
5214 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5215 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5216 | */ | |
5217 | rbd_bus_del_dev(rbd_dev); | |
8ad42cd0 | 5218 | rbd_dev_image_release(rbd_dev); |
79ab7558 | 5219 | module_put(THIS_MODULE); |
aafb230e | 5220 | |
1ba0f1e7 | 5221 | return count; |
602adf40 YS |
5222 | } |
5223 | ||
9b60e70b ID |
5224 | static ssize_t rbd_remove(struct bus_type *bus, |
5225 | const char *buf, | |
5226 | size_t count) | |
5227 | { | |
5228 | if (single_major) | |
5229 | return -EINVAL; | |
5230 | ||
5231 | return do_rbd_remove(bus, buf, count); | |
5232 | } | |
5233 | ||
5234 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5235 | const char *buf, | |
5236 | size_t count) | |
5237 | { | |
5238 | return do_rbd_remove(bus, buf, count); | |
5239 | } | |
5240 | ||
602adf40 YS |
5241 | /* |
5242 | * create control files in sysfs | |
dfc5606d | 5243 | * /sys/bus/rbd/... |
602adf40 YS |
5244 | */ |
5245 | static int rbd_sysfs_init(void) | |
5246 | { | |
dfc5606d | 5247 | int ret; |
602adf40 | 5248 | |
fed4c143 | 5249 | ret = device_register(&rbd_root_dev); |
21079786 | 5250 | if (ret < 0) |
dfc5606d | 5251 | return ret; |
602adf40 | 5252 | |
fed4c143 AE |
5253 | ret = bus_register(&rbd_bus_type); |
5254 | if (ret < 0) | |
5255 | device_unregister(&rbd_root_dev); | |
602adf40 | 5256 | |
602adf40 YS |
5257 | return ret; |
5258 | } | |
5259 | ||
5260 | static void rbd_sysfs_cleanup(void) | |
5261 | { | |
dfc5606d | 5262 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5263 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5264 | } |
5265 | ||
1c2a9dfe AE |
5266 | static int rbd_slab_init(void) |
5267 | { | |
5268 | rbd_assert(!rbd_img_request_cache); | |
5269 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5270 | sizeof (struct rbd_img_request), | |
5271 | __alignof__(struct rbd_img_request), | |
5272 | 0, NULL); | |
868311b1 AE |
5273 | if (!rbd_img_request_cache) |
5274 | return -ENOMEM; | |
5275 | ||
5276 | rbd_assert(!rbd_obj_request_cache); | |
5277 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5278 | sizeof (struct rbd_obj_request), | |
5279 | __alignof__(struct rbd_obj_request), | |
5280 | 0, NULL); | |
78c2a44a AE |
5281 | if (!rbd_obj_request_cache) |
5282 | goto out_err; | |
5283 | ||
5284 | rbd_assert(!rbd_segment_name_cache); | |
5285 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5286 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5287 | if (rbd_segment_name_cache) |
1c2a9dfe | 5288 | return 0; |
78c2a44a AE |
5289 | out_err: |
5290 | if (rbd_obj_request_cache) { | |
5291 | kmem_cache_destroy(rbd_obj_request_cache); | |
5292 | rbd_obj_request_cache = NULL; | |
5293 | } | |
1c2a9dfe | 5294 | |
868311b1 AE |
5295 | kmem_cache_destroy(rbd_img_request_cache); |
5296 | rbd_img_request_cache = NULL; | |
5297 | ||
1c2a9dfe AE |
5298 | return -ENOMEM; |
5299 | } | |
5300 | ||
5301 | static void rbd_slab_exit(void) | |
5302 | { | |
78c2a44a AE |
5303 | rbd_assert(rbd_segment_name_cache); |
5304 | kmem_cache_destroy(rbd_segment_name_cache); | |
5305 | rbd_segment_name_cache = NULL; | |
5306 | ||
868311b1 AE |
5307 | rbd_assert(rbd_obj_request_cache); |
5308 | kmem_cache_destroy(rbd_obj_request_cache); | |
5309 | rbd_obj_request_cache = NULL; | |
5310 | ||
1c2a9dfe AE |
5311 | rbd_assert(rbd_img_request_cache); |
5312 | kmem_cache_destroy(rbd_img_request_cache); | |
5313 | rbd_img_request_cache = NULL; | |
5314 | } | |
5315 | ||
cc344fa1 | 5316 | static int __init rbd_init(void) |
602adf40 YS |
5317 | { |
5318 | int rc; | |
5319 | ||
1e32d34c AE |
5320 | if (!libceph_compatible(NULL)) { |
5321 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5322 | return -EINVAL; |
5323 | } | |
e1b4d96d | 5324 | |
1c2a9dfe | 5325 | rc = rbd_slab_init(); |
602adf40 YS |
5326 | if (rc) |
5327 | return rc; | |
e1b4d96d | 5328 | |
9b60e70b ID |
5329 | if (single_major) { |
5330 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5331 | if (rbd_major < 0) { | |
5332 | rc = rbd_major; | |
5333 | goto err_out_slab; | |
5334 | } | |
5335 | } | |
5336 | ||
1c2a9dfe AE |
5337 | rc = rbd_sysfs_init(); |
5338 | if (rc) | |
9b60e70b ID |
5339 | goto err_out_blkdev; |
5340 | ||
5341 | if (single_major) | |
5342 | pr_info("loaded (major %d)\n", rbd_major); | |
5343 | else | |
5344 | pr_info("loaded\n"); | |
1c2a9dfe | 5345 | |
e1b4d96d ID |
5346 | return 0; |
5347 | ||
9b60e70b ID |
5348 | err_out_blkdev: |
5349 | if (single_major) | |
5350 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
e1b4d96d ID |
5351 | err_out_slab: |
5352 | rbd_slab_exit(); | |
1c2a9dfe | 5353 | return rc; |
602adf40 YS |
5354 | } |
5355 | ||
cc344fa1 | 5356 | static void __exit rbd_exit(void) |
602adf40 YS |
5357 | { |
5358 | rbd_sysfs_cleanup(); | |
9b60e70b ID |
5359 | if (single_major) |
5360 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
1c2a9dfe | 5361 | rbd_slab_exit(); |
602adf40 YS |
5362 | } |
5363 | ||
5364 | module_init(rbd_init); | |
5365 | module_exit(rbd_exit); | |
5366 | ||
d552c619 | 5367 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5368 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5369 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5370 | /* following authorship retained from original osdblk.c */ |
5371 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5372 | ||
90da258b | 5373 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5374 | MODULE_LICENSE("GPL"); |