Commit | Line | Data |
---|---|---|
e2a58ee5 | 1 | |
602adf40 YS |
2 | /* |
3 | rbd.c -- Export ceph rados objects as a Linux block device | |
4 | ||
5 | ||
6 | based on drivers/block/osdblk.c: | |
7 | ||
8 | Copyright 2009 Red Hat, Inc. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; see the file COPYING. If not, write to | |
21 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
22 | ||
23 | ||
24 | ||
dfc5606d | 25 | For usage instructions, please refer to: |
602adf40 | 26 | |
dfc5606d | 27 | Documentation/ABI/testing/sysfs-bus-rbd |
602adf40 YS |
28 | |
29 | */ | |
30 | ||
31 | #include <linux/ceph/libceph.h> | |
32 | #include <linux/ceph/osd_client.h> | |
33 | #include <linux/ceph/mon_client.h> | |
34 | #include <linux/ceph/decode.h> | |
59c2be1e | 35 | #include <linux/parser.h> |
30d1cff8 | 36 | #include <linux/bsearch.h> |
602adf40 YS |
37 | |
38 | #include <linux/kernel.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/module.h> | |
7ad18afa | 41 | #include <linux/blk-mq.h> |
602adf40 YS |
42 | #include <linux/fs.h> |
43 | #include <linux/blkdev.h> | |
1c2a9dfe | 44 | #include <linux/slab.h> |
f8a22fc2 | 45 | #include <linux/idr.h> |
bc1ecc65 | 46 | #include <linux/workqueue.h> |
602adf40 YS |
47 | |
48 | #include "rbd_types.h" | |
49 | ||
aafb230e AE |
50 | #define RBD_DEBUG /* Activate rbd_assert() calls */ |
51 | ||
593a9e7b AE |
52 | /* |
53 | * The basic unit of block I/O is a sector. It is interpreted in a | |
54 | * number of contexts in Linux (blk, bio, genhd), but the default is | |
55 | * universally 512 bytes. These symbols are just slightly more | |
56 | * meaningful than the bare numbers they represent. | |
57 | */ | |
58 | #define SECTOR_SHIFT 9 | |
59 | #define SECTOR_SIZE (1ULL << SECTOR_SHIFT) | |
60 | ||
a2acd00e AE |
61 | /* |
62 | * Increment the given counter and return its updated value. | |
63 | * If the counter is already 0 it will not be incremented. | |
64 | * If the counter is already at its maximum value returns | |
65 | * -EINVAL without updating it. | |
66 | */ | |
67 | static int atomic_inc_return_safe(atomic_t *v) | |
68 | { | |
69 | unsigned int counter; | |
70 | ||
71 | counter = (unsigned int)__atomic_add_unless(v, 1, 0); | |
72 | if (counter <= (unsigned int)INT_MAX) | |
73 | return (int)counter; | |
74 | ||
75 | atomic_dec(v); | |
76 | ||
77 | return -EINVAL; | |
78 | } | |
79 | ||
80 | /* Decrement the counter. Return the resulting value, or -EINVAL */ | |
81 | static int atomic_dec_return_safe(atomic_t *v) | |
82 | { | |
83 | int counter; | |
84 | ||
85 | counter = atomic_dec_return(v); | |
86 | if (counter >= 0) | |
87 | return counter; | |
88 | ||
89 | atomic_inc(v); | |
90 | ||
91 | return -EINVAL; | |
92 | } | |
93 | ||
f0f8cef5 | 94 | #define RBD_DRV_NAME "rbd" |
602adf40 | 95 | |
7e513d43 ID |
96 | #define RBD_MINORS_PER_MAJOR 256 |
97 | #define RBD_SINGLE_MAJOR_PART_SHIFT 4 | |
602adf40 | 98 | |
6d69bb53 ID |
99 | #define RBD_MAX_PARENT_CHAIN_LEN 16 |
100 | ||
d4b125e9 AE |
101 | #define RBD_SNAP_DEV_NAME_PREFIX "snap_" |
102 | #define RBD_MAX_SNAP_NAME_LEN \ | |
103 | (NAME_MAX - (sizeof (RBD_SNAP_DEV_NAME_PREFIX) - 1)) | |
104 | ||
35d489f9 | 105 | #define RBD_MAX_SNAP_COUNT 510 /* allows max snapc to fit in 4KB */ |
602adf40 YS |
106 | |
107 | #define RBD_SNAP_HEAD_NAME "-" | |
108 | ||
9682fc6d AE |
109 | #define BAD_SNAP_INDEX U32_MAX /* invalid index into snap array */ |
110 | ||
9e15b77d AE |
111 | /* This allows a single page to hold an image name sent by OSD */ |
112 | #define RBD_IMAGE_NAME_LEN_MAX (PAGE_SIZE - sizeof (__le32) - 1) | |
1e130199 | 113 | #define RBD_IMAGE_ID_LEN_MAX 64 |
9e15b77d | 114 | |
1e130199 | 115 | #define RBD_OBJ_PREFIX_LEN_MAX 64 |
589d30e0 | 116 | |
d889140c AE |
117 | /* Feature bits */ |
118 | ||
5cbf6f12 AE |
119 | #define RBD_FEATURE_LAYERING (1<<0) |
120 | #define RBD_FEATURE_STRIPINGV2 (1<<1) | |
121 | #define RBD_FEATURES_ALL \ | |
122 | (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2) | |
d889140c AE |
123 | |
124 | /* Features supported by this (client software) implementation. */ | |
125 | ||
770eba6e | 126 | #define RBD_FEATURES_SUPPORTED (RBD_FEATURES_ALL) |
d889140c | 127 | |
81a89793 AE |
128 | /* |
129 | * An RBD device name will be "rbd#", where the "rbd" comes from | |
130 | * RBD_DRV_NAME above, and # is a unique integer identifier. | |
131 | * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big | |
132 | * enough to hold all possible device names. | |
133 | */ | |
602adf40 | 134 | #define DEV_NAME_LEN 32 |
81a89793 | 135 | #define MAX_INT_FORMAT_WIDTH ((5 * sizeof (int)) / 2 + 1) |
602adf40 YS |
136 | |
137 | /* | |
138 | * block device image metadata (in-memory version) | |
139 | */ | |
140 | struct rbd_image_header { | |
f35a4dee | 141 | /* These six fields never change for a given rbd image */ |
849b4260 | 142 | char *object_prefix; |
602adf40 YS |
143 | __u8 obj_order; |
144 | __u8 crypt_type; | |
145 | __u8 comp_type; | |
f35a4dee AE |
146 | u64 stripe_unit; |
147 | u64 stripe_count; | |
148 | u64 features; /* Might be changeable someday? */ | |
602adf40 | 149 | |
f84344f3 AE |
150 | /* The remaining fields need to be updated occasionally */ |
151 | u64 image_size; | |
152 | struct ceph_snap_context *snapc; | |
f35a4dee AE |
153 | char *snap_names; /* format 1 only */ |
154 | u64 *snap_sizes; /* format 1 only */ | |
59c2be1e YS |
155 | }; |
156 | ||
0d7dbfce AE |
157 | /* |
158 | * An rbd image specification. | |
159 | * | |
160 | * The tuple (pool_id, image_id, snap_id) is sufficient to uniquely | |
c66c6e0c AE |
161 | * identify an image. Each rbd_dev structure includes a pointer to |
162 | * an rbd_spec structure that encapsulates this identity. | |
163 | * | |
164 | * Each of the id's in an rbd_spec has an associated name. For a | |
165 | * user-mapped image, the names are supplied and the id's associated | |
166 | * with them are looked up. For a layered image, a parent image is | |
167 | * defined by the tuple, and the names are looked up. | |
168 | * | |
169 | * An rbd_dev structure contains a parent_spec pointer which is | |
170 | * non-null if the image it represents is a child in a layered | |
171 | * image. This pointer will refer to the rbd_spec structure used | |
172 | * by the parent rbd_dev for its own identity (i.e., the structure | |
173 | * is shared between the parent and child). | |
174 | * | |
175 | * Since these structures are populated once, during the discovery | |
176 | * phase of image construction, they are effectively immutable so | |
177 | * we make no effort to synchronize access to them. | |
178 | * | |
179 | * Note that code herein does not assume the image name is known (it | |
180 | * could be a null pointer). | |
0d7dbfce AE |
181 | */ |
182 | struct rbd_spec { | |
183 | u64 pool_id; | |
ecb4dc22 | 184 | const char *pool_name; |
0d7dbfce | 185 | |
ecb4dc22 AE |
186 | const char *image_id; |
187 | const char *image_name; | |
0d7dbfce AE |
188 | |
189 | u64 snap_id; | |
ecb4dc22 | 190 | const char *snap_name; |
0d7dbfce AE |
191 | |
192 | struct kref kref; | |
193 | }; | |
194 | ||
602adf40 | 195 | /* |
f0f8cef5 | 196 | * an instance of the client. multiple devices may share an rbd client. |
602adf40 YS |
197 | */ |
198 | struct rbd_client { | |
199 | struct ceph_client *client; | |
200 | struct kref kref; | |
201 | struct list_head node; | |
202 | }; | |
203 | ||
bf0d5f50 AE |
204 | struct rbd_img_request; |
205 | typedef void (*rbd_img_callback_t)(struct rbd_img_request *); | |
206 | ||
207 | #define BAD_WHICH U32_MAX /* Good which or bad which, which? */ | |
208 | ||
209 | struct rbd_obj_request; | |
210 | typedef void (*rbd_obj_callback_t)(struct rbd_obj_request *); | |
211 | ||
9969ebc5 AE |
212 | enum obj_request_type { |
213 | OBJ_REQUEST_NODATA, OBJ_REQUEST_BIO, OBJ_REQUEST_PAGES | |
214 | }; | |
bf0d5f50 | 215 | |
6d2940c8 GZ |
216 | enum obj_operation_type { |
217 | OBJ_OP_WRITE, | |
218 | OBJ_OP_READ, | |
90e98c52 | 219 | OBJ_OP_DISCARD, |
6d2940c8 GZ |
220 | }; |
221 | ||
926f9b3f AE |
222 | enum obj_req_flags { |
223 | OBJ_REQ_DONE, /* completion flag: not done = 0, done = 1 */ | |
6365d33a | 224 | OBJ_REQ_IMG_DATA, /* object usage: standalone = 0, image = 1 */ |
5679c59f AE |
225 | OBJ_REQ_KNOWN, /* EXISTS flag valid: no = 0, yes = 1 */ |
226 | OBJ_REQ_EXISTS, /* target exists: no = 0, yes = 1 */ | |
926f9b3f AE |
227 | }; |
228 | ||
bf0d5f50 AE |
229 | struct rbd_obj_request { |
230 | const char *object_name; | |
231 | u64 offset; /* object start byte */ | |
232 | u64 length; /* bytes from offset */ | |
926f9b3f | 233 | unsigned long flags; |
bf0d5f50 | 234 | |
c5b5ef6c AE |
235 | /* |
236 | * An object request associated with an image will have its | |
237 | * img_data flag set; a standalone object request will not. | |
238 | * | |
239 | * A standalone object request will have which == BAD_WHICH | |
240 | * and a null obj_request pointer. | |
241 | * | |
242 | * An object request initiated in support of a layered image | |
243 | * object (to check for its existence before a write) will | |
244 | * have which == BAD_WHICH and a non-null obj_request pointer. | |
245 | * | |
246 | * Finally, an object request for rbd image data will have | |
247 | * which != BAD_WHICH, and will have a non-null img_request | |
248 | * pointer. The value of which will be in the range | |
249 | * 0..(img_request->obj_request_count-1). | |
250 | */ | |
251 | union { | |
252 | struct rbd_obj_request *obj_request; /* STAT op */ | |
253 | struct { | |
254 | struct rbd_img_request *img_request; | |
255 | u64 img_offset; | |
256 | /* links for img_request->obj_requests list */ | |
257 | struct list_head links; | |
258 | }; | |
259 | }; | |
bf0d5f50 AE |
260 | u32 which; /* posn image request list */ |
261 | ||
262 | enum obj_request_type type; | |
788e2df3 AE |
263 | union { |
264 | struct bio *bio_list; | |
265 | struct { | |
266 | struct page **pages; | |
267 | u32 page_count; | |
268 | }; | |
269 | }; | |
0eefd470 | 270 | struct page **copyup_pages; |
ebda6408 | 271 | u32 copyup_page_count; |
bf0d5f50 AE |
272 | |
273 | struct ceph_osd_request *osd_req; | |
274 | ||
275 | u64 xferred; /* bytes transferred */ | |
1b83bef2 | 276 | int result; |
bf0d5f50 AE |
277 | |
278 | rbd_obj_callback_t callback; | |
788e2df3 | 279 | struct completion completion; |
bf0d5f50 AE |
280 | |
281 | struct kref kref; | |
282 | }; | |
283 | ||
0c425248 | 284 | enum img_req_flags { |
9849e986 AE |
285 | IMG_REQ_WRITE, /* I/O direction: read = 0, write = 1 */ |
286 | IMG_REQ_CHILD, /* initiator: block = 0, child image = 1 */ | |
d0b2e944 | 287 | IMG_REQ_LAYERED, /* ENOENT handling: normal = 0, layered = 1 */ |
90e98c52 | 288 | IMG_REQ_DISCARD, /* discard: normal = 0, discard request = 1 */ |
0c425248 AE |
289 | }; |
290 | ||
bf0d5f50 | 291 | struct rbd_img_request { |
bf0d5f50 AE |
292 | struct rbd_device *rbd_dev; |
293 | u64 offset; /* starting image byte offset */ | |
294 | u64 length; /* byte count from offset */ | |
0c425248 | 295 | unsigned long flags; |
bf0d5f50 | 296 | union { |
9849e986 | 297 | u64 snap_id; /* for reads */ |
bf0d5f50 | 298 | struct ceph_snap_context *snapc; /* for writes */ |
9849e986 AE |
299 | }; |
300 | union { | |
301 | struct request *rq; /* block request */ | |
302 | struct rbd_obj_request *obj_request; /* obj req initiator */ | |
bf0d5f50 | 303 | }; |
3d7efd18 | 304 | struct page **copyup_pages; |
ebda6408 | 305 | u32 copyup_page_count; |
bf0d5f50 AE |
306 | spinlock_t completion_lock;/* protects next_completion */ |
307 | u32 next_completion; | |
308 | rbd_img_callback_t callback; | |
55f27e09 | 309 | u64 xferred;/* aggregate bytes transferred */ |
a5a337d4 | 310 | int result; /* first nonzero obj_request result */ |
bf0d5f50 AE |
311 | |
312 | u32 obj_request_count; | |
313 | struct list_head obj_requests; /* rbd_obj_request structs */ | |
314 | ||
315 | struct kref kref; | |
316 | }; | |
317 | ||
318 | #define for_each_obj_request(ireq, oreq) \ | |
ef06f4d3 | 319 | list_for_each_entry(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 320 | #define for_each_obj_request_from(ireq, oreq) \ |
ef06f4d3 | 321 | list_for_each_entry_from(oreq, &(ireq)->obj_requests, links) |
bf0d5f50 | 322 | #define for_each_obj_request_safe(ireq, oreq, n) \ |
ef06f4d3 | 323 | list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links) |
bf0d5f50 | 324 | |
f84344f3 | 325 | struct rbd_mapping { |
99c1f08f | 326 | u64 size; |
34b13184 | 327 | u64 features; |
f84344f3 AE |
328 | bool read_only; |
329 | }; | |
330 | ||
602adf40 YS |
331 | /* |
332 | * a single device | |
333 | */ | |
334 | struct rbd_device { | |
de71a297 | 335 | int dev_id; /* blkdev unique id */ |
602adf40 YS |
336 | |
337 | int major; /* blkdev assigned major */ | |
dd82fff1 | 338 | int minor; |
602adf40 | 339 | struct gendisk *disk; /* blkdev's gendisk and rq */ |
602adf40 | 340 | |
a30b71b9 | 341 | u32 image_format; /* Either 1 or 2 */ |
602adf40 YS |
342 | struct rbd_client *rbd_client; |
343 | ||
344 | char name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */ | |
345 | ||
b82d167b | 346 | spinlock_t lock; /* queue, flags, open_count */ |
602adf40 YS |
347 | |
348 | struct rbd_image_header header; | |
b82d167b | 349 | unsigned long flags; /* possibly lock protected */ |
0d7dbfce | 350 | struct rbd_spec *spec; |
d147543d | 351 | struct rbd_options *opts; |
602adf40 | 352 | |
0d7dbfce | 353 | char *header_name; |
971f839a | 354 | |
0903e875 AE |
355 | struct ceph_file_layout layout; |
356 | ||
59c2be1e | 357 | struct ceph_osd_event *watch_event; |
975241af | 358 | struct rbd_obj_request *watch_request; |
59c2be1e | 359 | |
86b00e0d AE |
360 | struct rbd_spec *parent_spec; |
361 | u64 parent_overlap; | |
a2acd00e | 362 | atomic_t parent_ref; |
2f82ee54 | 363 | struct rbd_device *parent; |
86b00e0d | 364 | |
7ad18afa CH |
365 | /* Block layer tags. */ |
366 | struct blk_mq_tag_set tag_set; | |
367 | ||
c666601a JD |
368 | /* protects updating the header */ |
369 | struct rw_semaphore header_rwsem; | |
f84344f3 AE |
370 | |
371 | struct rbd_mapping mapping; | |
602adf40 YS |
372 | |
373 | struct list_head node; | |
dfc5606d | 374 | |
dfc5606d YS |
375 | /* sysfs related */ |
376 | struct device dev; | |
b82d167b | 377 | unsigned long open_count; /* protected by lock */ |
dfc5606d YS |
378 | }; |
379 | ||
b82d167b AE |
380 | /* |
381 | * Flag bits for rbd_dev->flags. If atomicity is required, | |
382 | * rbd_dev->lock is used to protect access. | |
383 | * | |
384 | * Currently, only the "removing" flag (which is coupled with the | |
385 | * "open_count" field) requires atomic access. | |
386 | */ | |
6d292906 AE |
387 | enum rbd_dev_flags { |
388 | RBD_DEV_FLAG_EXISTS, /* mapped snapshot has not been deleted */ | |
b82d167b | 389 | RBD_DEV_FLAG_REMOVING, /* this mapping is being removed */ |
6d292906 AE |
390 | }; |
391 | ||
cfbf6377 | 392 | static DEFINE_MUTEX(client_mutex); /* Serialize client creation */ |
e124a82f | 393 | |
602adf40 | 394 | static LIST_HEAD(rbd_dev_list); /* devices */ |
e124a82f AE |
395 | static DEFINE_SPINLOCK(rbd_dev_list_lock); |
396 | ||
432b8587 AE |
397 | static LIST_HEAD(rbd_client_list); /* clients */ |
398 | static DEFINE_SPINLOCK(rbd_client_list_lock); | |
602adf40 | 399 | |
78c2a44a AE |
400 | /* Slab caches for frequently-allocated structures */ |
401 | ||
1c2a9dfe | 402 | static struct kmem_cache *rbd_img_request_cache; |
868311b1 | 403 | static struct kmem_cache *rbd_obj_request_cache; |
78c2a44a | 404 | static struct kmem_cache *rbd_segment_name_cache; |
1c2a9dfe | 405 | |
9b60e70b | 406 | static int rbd_major; |
f8a22fc2 ID |
407 | static DEFINE_IDA(rbd_dev_id_ida); |
408 | ||
f5ee37bd ID |
409 | static struct workqueue_struct *rbd_wq; |
410 | ||
9b60e70b ID |
411 | /* |
412 | * Default to false for now, as single-major requires >= 0.75 version of | |
413 | * userspace rbd utility. | |
414 | */ | |
415 | static bool single_major = false; | |
416 | module_param(single_major, bool, S_IRUGO); | |
417 | MODULE_PARM_DESC(single_major, "Use a single major number for all rbd devices (default: false)"); | |
418 | ||
3d7efd18 AE |
419 | static int rbd_img_request_submit(struct rbd_img_request *img_request); |
420 | ||
f0f8cef5 AE |
421 | static ssize_t rbd_add(struct bus_type *bus, const char *buf, |
422 | size_t count); | |
423 | static ssize_t rbd_remove(struct bus_type *bus, const char *buf, | |
424 | size_t count); | |
9b60e70b ID |
425 | static ssize_t rbd_add_single_major(struct bus_type *bus, const char *buf, |
426 | size_t count); | |
427 | static ssize_t rbd_remove_single_major(struct bus_type *bus, const char *buf, | |
428 | size_t count); | |
6d69bb53 | 429 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth); |
a2acd00e | 430 | static void rbd_spec_put(struct rbd_spec *spec); |
f0f8cef5 | 431 | |
9b60e70b ID |
432 | static int rbd_dev_id_to_minor(int dev_id) |
433 | { | |
7e513d43 | 434 | return dev_id << RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
435 | } |
436 | ||
437 | static int minor_to_rbd_dev_id(int minor) | |
438 | { | |
7e513d43 | 439 | return minor >> RBD_SINGLE_MAJOR_PART_SHIFT; |
9b60e70b ID |
440 | } |
441 | ||
b15a21dd GKH |
442 | static BUS_ATTR(add, S_IWUSR, NULL, rbd_add); |
443 | static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove); | |
9b60e70b ID |
444 | static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major); |
445 | static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major); | |
b15a21dd GKH |
446 | |
447 | static struct attribute *rbd_bus_attrs[] = { | |
448 | &bus_attr_add.attr, | |
449 | &bus_attr_remove.attr, | |
9b60e70b ID |
450 | &bus_attr_add_single_major.attr, |
451 | &bus_attr_remove_single_major.attr, | |
b15a21dd | 452 | NULL, |
f0f8cef5 | 453 | }; |
92c76dc0 ID |
454 | |
455 | static umode_t rbd_bus_is_visible(struct kobject *kobj, | |
456 | struct attribute *attr, int index) | |
457 | { | |
9b60e70b ID |
458 | if (!single_major && |
459 | (attr == &bus_attr_add_single_major.attr || | |
460 | attr == &bus_attr_remove_single_major.attr)) | |
461 | return 0; | |
462 | ||
92c76dc0 ID |
463 | return attr->mode; |
464 | } | |
465 | ||
466 | static const struct attribute_group rbd_bus_group = { | |
467 | .attrs = rbd_bus_attrs, | |
468 | .is_visible = rbd_bus_is_visible, | |
469 | }; | |
470 | __ATTRIBUTE_GROUPS(rbd_bus); | |
f0f8cef5 AE |
471 | |
472 | static struct bus_type rbd_bus_type = { | |
473 | .name = "rbd", | |
b15a21dd | 474 | .bus_groups = rbd_bus_groups, |
f0f8cef5 AE |
475 | }; |
476 | ||
477 | static void rbd_root_dev_release(struct device *dev) | |
478 | { | |
479 | } | |
480 | ||
481 | static struct device rbd_root_dev = { | |
482 | .init_name = "rbd", | |
483 | .release = rbd_root_dev_release, | |
484 | }; | |
485 | ||
06ecc6cb AE |
486 | static __printf(2, 3) |
487 | void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...) | |
488 | { | |
489 | struct va_format vaf; | |
490 | va_list args; | |
491 | ||
492 | va_start(args, fmt); | |
493 | vaf.fmt = fmt; | |
494 | vaf.va = &args; | |
495 | ||
496 | if (!rbd_dev) | |
497 | printk(KERN_WARNING "%s: %pV\n", RBD_DRV_NAME, &vaf); | |
498 | else if (rbd_dev->disk) | |
499 | printk(KERN_WARNING "%s: %s: %pV\n", | |
500 | RBD_DRV_NAME, rbd_dev->disk->disk_name, &vaf); | |
501 | else if (rbd_dev->spec && rbd_dev->spec->image_name) | |
502 | printk(KERN_WARNING "%s: image %s: %pV\n", | |
503 | RBD_DRV_NAME, rbd_dev->spec->image_name, &vaf); | |
504 | else if (rbd_dev->spec && rbd_dev->spec->image_id) | |
505 | printk(KERN_WARNING "%s: id %s: %pV\n", | |
506 | RBD_DRV_NAME, rbd_dev->spec->image_id, &vaf); | |
507 | else /* punt */ | |
508 | printk(KERN_WARNING "%s: rbd_dev %p: %pV\n", | |
509 | RBD_DRV_NAME, rbd_dev, &vaf); | |
510 | va_end(args); | |
511 | } | |
512 | ||
aafb230e AE |
513 | #ifdef RBD_DEBUG |
514 | #define rbd_assert(expr) \ | |
515 | if (unlikely(!(expr))) { \ | |
516 | printk(KERN_ERR "\nAssertion failure in %s() " \ | |
517 | "at line %d:\n\n" \ | |
518 | "\trbd_assert(%s);\n\n", \ | |
519 | __func__, __LINE__, #expr); \ | |
520 | BUG(); \ | |
521 | } | |
522 | #else /* !RBD_DEBUG */ | |
523 | # define rbd_assert(expr) ((void) 0) | |
524 | #endif /* !RBD_DEBUG */ | |
dfc5606d | 525 | |
2761713d | 526 | static void rbd_osd_copyup_callback(struct rbd_obj_request *obj_request); |
b454e36d | 527 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request); |
05a46afd AE |
528 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request); |
529 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev); | |
8b3e1a56 | 530 | |
cc4a38bd | 531 | static int rbd_dev_refresh(struct rbd_device *rbd_dev); |
2df3fac7 | 532 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev); |
a720ae09 | 533 | static int rbd_dev_header_info(struct rbd_device *rbd_dev); |
e8f59b59 | 534 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev); |
54cac61f AE |
535 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
536 | u64 snap_id); | |
2ad3d716 AE |
537 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
538 | u8 *order, u64 *snap_size); | |
539 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, | |
540 | u64 *snap_features); | |
541 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name); | |
59c2be1e | 542 | |
602adf40 YS |
543 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
544 | { | |
f0f8cef5 | 545 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 546 | bool removing = false; |
602adf40 | 547 | |
f84344f3 | 548 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
549 | return -EROFS; |
550 | ||
a14ea269 | 551 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
552 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
553 | removing = true; | |
554 | else | |
555 | rbd_dev->open_count++; | |
a14ea269 | 556 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
557 | if (removing) |
558 | return -ENOENT; | |
559 | ||
c3e946ce | 560 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 561 | |
602adf40 YS |
562 | return 0; |
563 | } | |
564 | ||
db2a144b | 565 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
566 | { |
567 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
568 | unsigned long open_count_before; |
569 | ||
a14ea269 | 570 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 571 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 572 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 573 | rbd_assert(open_count_before > 0); |
dfc5606d | 574 | |
c3e946ce | 575 | put_device(&rbd_dev->dev); |
dfc5606d YS |
576 | } |
577 | ||
131fd9f6 GZ |
578 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
579 | { | |
77f33c03 | 580 | int ret = 0; |
131fd9f6 GZ |
581 | int val; |
582 | bool ro; | |
77f33c03 | 583 | bool ro_changed = false; |
131fd9f6 | 584 | |
77f33c03 | 585 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
586 | if (get_user(val, (int __user *)(arg))) |
587 | return -EFAULT; | |
588 | ||
589 | ro = val ? true : false; | |
590 | /* Snapshot doesn't allow to write*/ | |
591 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
592 | return -EROFS; | |
593 | ||
77f33c03 JD |
594 | spin_lock_irq(&rbd_dev->lock); |
595 | /* prevent others open this device */ | |
596 | if (rbd_dev->open_count > 1) { | |
597 | ret = -EBUSY; | |
598 | goto out; | |
599 | } | |
600 | ||
131fd9f6 GZ |
601 | if (rbd_dev->mapping.read_only != ro) { |
602 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 603 | ro_changed = true; |
131fd9f6 GZ |
604 | } |
605 | ||
77f33c03 JD |
606 | out: |
607 | spin_unlock_irq(&rbd_dev->lock); | |
608 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
609 | if (ret == 0 && ro_changed) | |
610 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
611 | ||
612 | return ret; | |
131fd9f6 GZ |
613 | } |
614 | ||
615 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
616 | unsigned int cmd, unsigned long arg) | |
617 | { | |
618 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
619 | int ret = 0; | |
620 | ||
131fd9f6 GZ |
621 | switch (cmd) { |
622 | case BLKROSET: | |
623 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
624 | break; | |
625 | default: | |
626 | ret = -ENOTTY; | |
627 | } | |
628 | ||
131fd9f6 GZ |
629 | return ret; |
630 | } | |
631 | ||
632 | #ifdef CONFIG_COMPAT | |
633 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
634 | unsigned int cmd, unsigned long arg) | |
635 | { | |
636 | return rbd_ioctl(bdev, mode, cmd, arg); | |
637 | } | |
638 | #endif /* CONFIG_COMPAT */ | |
639 | ||
602adf40 YS |
640 | static const struct block_device_operations rbd_bd_ops = { |
641 | .owner = THIS_MODULE, | |
642 | .open = rbd_open, | |
dfc5606d | 643 | .release = rbd_release, |
131fd9f6 GZ |
644 | .ioctl = rbd_ioctl, |
645 | #ifdef CONFIG_COMPAT | |
646 | .compat_ioctl = rbd_compat_ioctl, | |
647 | #endif | |
602adf40 YS |
648 | }; |
649 | ||
650 | /* | |
7262cfca | 651 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 652 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 653 | */ |
f8c38929 | 654 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
655 | { |
656 | struct rbd_client *rbdc; | |
657 | int ret = -ENOMEM; | |
658 | ||
37206ee5 | 659 | dout("%s:\n", __func__); |
602adf40 YS |
660 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
661 | if (!rbdc) | |
662 | goto out_opt; | |
663 | ||
664 | kref_init(&rbdc->kref); | |
665 | INIT_LIST_HEAD(&rbdc->node); | |
666 | ||
43ae4701 | 667 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 668 | if (IS_ERR(rbdc->client)) |
08f75463 | 669 | goto out_rbdc; |
43ae4701 | 670 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
671 | |
672 | ret = ceph_open_session(rbdc->client); | |
673 | if (ret < 0) | |
08f75463 | 674 | goto out_client; |
602adf40 | 675 | |
432b8587 | 676 | spin_lock(&rbd_client_list_lock); |
602adf40 | 677 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 678 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 679 | |
37206ee5 | 680 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 681 | |
602adf40 | 682 | return rbdc; |
08f75463 | 683 | out_client: |
602adf40 | 684 | ceph_destroy_client(rbdc->client); |
08f75463 | 685 | out_rbdc: |
602adf40 YS |
686 | kfree(rbdc); |
687 | out_opt: | |
43ae4701 AE |
688 | if (ceph_opts) |
689 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
690 | dout("%s: error %d\n", __func__, ret); |
691 | ||
28f259b7 | 692 | return ERR_PTR(ret); |
602adf40 YS |
693 | } |
694 | ||
2f82ee54 AE |
695 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
696 | { | |
697 | kref_get(&rbdc->kref); | |
698 | ||
699 | return rbdc; | |
700 | } | |
701 | ||
602adf40 | 702 | /* |
1f7ba331 AE |
703 | * Find a ceph client with specific addr and configuration. If |
704 | * found, bump its reference count. | |
602adf40 | 705 | */ |
1f7ba331 | 706 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
707 | { |
708 | struct rbd_client *client_node; | |
1f7ba331 | 709 | bool found = false; |
602adf40 | 710 | |
43ae4701 | 711 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
712 | return NULL; |
713 | ||
1f7ba331 AE |
714 | spin_lock(&rbd_client_list_lock); |
715 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
716 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
717 | __rbd_get_client(client_node); |
718 | ||
1f7ba331 AE |
719 | found = true; |
720 | break; | |
721 | } | |
722 | } | |
723 | spin_unlock(&rbd_client_list_lock); | |
724 | ||
725 | return found ? client_node : NULL; | |
602adf40 YS |
726 | } |
727 | ||
59c2be1e | 728 | /* |
210c104c | 729 | * (Per device) rbd map options |
59c2be1e YS |
730 | */ |
731 | enum { | |
b5584180 | 732 | Opt_queue_depth, |
59c2be1e YS |
733 | Opt_last_int, |
734 | /* int args above */ | |
735 | Opt_last_string, | |
736 | /* string args above */ | |
cc0538b6 AE |
737 | Opt_read_only, |
738 | Opt_read_write, | |
210c104c | 739 | Opt_err |
59c2be1e YS |
740 | }; |
741 | ||
43ae4701 | 742 | static match_table_t rbd_opts_tokens = { |
b5584180 | 743 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
744 | /* int args above */ |
745 | /* string args above */ | |
be466c1c | 746 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
747 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
748 | {Opt_read_write, "read_write"}, | |
749 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 750 | {Opt_err, NULL} |
59c2be1e YS |
751 | }; |
752 | ||
98571b5a | 753 | struct rbd_options { |
b5584180 | 754 | int queue_depth; |
98571b5a AE |
755 | bool read_only; |
756 | }; | |
757 | ||
b5584180 | 758 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a AE |
759 | #define RBD_READ_ONLY_DEFAULT false |
760 | ||
59c2be1e YS |
761 | static int parse_rbd_opts_token(char *c, void *private) |
762 | { | |
43ae4701 | 763 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
764 | substring_t argstr[MAX_OPT_ARGS]; |
765 | int token, intval, ret; | |
766 | ||
43ae4701 | 767 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
768 | if (token < Opt_last_int) { |
769 | ret = match_int(&argstr[0], &intval); | |
770 | if (ret < 0) { | |
210c104c | 771 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
772 | return ret; |
773 | } | |
774 | dout("got int token %d val %d\n", token, intval); | |
775 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 776 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
777 | } else { |
778 | dout("got token %d\n", token); | |
779 | } | |
780 | ||
781 | switch (token) { | |
b5584180 ID |
782 | case Opt_queue_depth: |
783 | if (intval < 1) { | |
784 | pr_err("queue_depth out of range\n"); | |
785 | return -EINVAL; | |
786 | } | |
787 | rbd_opts->queue_depth = intval; | |
788 | break; | |
cc0538b6 AE |
789 | case Opt_read_only: |
790 | rbd_opts->read_only = true; | |
791 | break; | |
792 | case Opt_read_write: | |
793 | rbd_opts->read_only = false; | |
794 | break; | |
59c2be1e | 795 | default: |
210c104c ID |
796 | /* libceph prints "bad option" msg */ |
797 | return -EINVAL; | |
59c2be1e | 798 | } |
210c104c | 799 | |
59c2be1e YS |
800 | return 0; |
801 | } | |
802 | ||
6d2940c8 GZ |
803 | static char* obj_op_name(enum obj_operation_type op_type) |
804 | { | |
805 | switch (op_type) { | |
806 | case OBJ_OP_READ: | |
807 | return "read"; | |
808 | case OBJ_OP_WRITE: | |
809 | return "write"; | |
90e98c52 GZ |
810 | case OBJ_OP_DISCARD: |
811 | return "discard"; | |
6d2940c8 GZ |
812 | default: |
813 | return "???"; | |
814 | } | |
815 | } | |
816 | ||
602adf40 YS |
817 | /* |
818 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
819 | * not exist create it. Either way, ceph_opts is consumed by this |
820 | * function. | |
602adf40 | 821 | */ |
9d3997fd | 822 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 823 | { |
f8c38929 | 824 | struct rbd_client *rbdc; |
59c2be1e | 825 | |
cfbf6377 | 826 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 827 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 828 | if (rbdc) /* using an existing client */ |
43ae4701 | 829 | ceph_destroy_options(ceph_opts); |
9d3997fd | 830 | else |
f8c38929 | 831 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 832 | mutex_unlock(&client_mutex); |
602adf40 | 833 | |
9d3997fd | 834 | return rbdc; |
602adf40 YS |
835 | } |
836 | ||
837 | /* | |
838 | * Destroy ceph client | |
d23a4b3f | 839 | * |
432b8587 | 840 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
841 | */ |
842 | static void rbd_client_release(struct kref *kref) | |
843 | { | |
844 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
845 | ||
37206ee5 | 846 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 847 | spin_lock(&rbd_client_list_lock); |
602adf40 | 848 | list_del(&rbdc->node); |
cd9d9f5d | 849 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
850 | |
851 | ceph_destroy_client(rbdc->client); | |
852 | kfree(rbdc); | |
853 | } | |
854 | ||
855 | /* | |
856 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
857 | * it. | |
858 | */ | |
9d3997fd | 859 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 860 | { |
c53d5893 AE |
861 | if (rbdc) |
862 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
863 | } |
864 | ||
a30b71b9 AE |
865 | static bool rbd_image_format_valid(u32 image_format) |
866 | { | |
867 | return image_format == 1 || image_format == 2; | |
868 | } | |
869 | ||
8e94af8e AE |
870 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
871 | { | |
103a150f AE |
872 | size_t size; |
873 | u32 snap_count; | |
874 | ||
875 | /* The header has to start with the magic rbd header text */ | |
876 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
877 | return false; | |
878 | ||
db2388b6 AE |
879 | /* The bio layer requires at least sector-sized I/O */ |
880 | ||
881 | if (ondisk->options.order < SECTOR_SHIFT) | |
882 | return false; | |
883 | ||
884 | /* If we use u64 in a few spots we may be able to loosen this */ | |
885 | ||
886 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
887 | return false; | |
888 | ||
103a150f AE |
889 | /* |
890 | * The size of a snapshot header has to fit in a size_t, and | |
891 | * that limits the number of snapshots. | |
892 | */ | |
893 | snap_count = le32_to_cpu(ondisk->snap_count); | |
894 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
895 | if (snap_count > size / sizeof (__le64)) | |
896 | return false; | |
897 | ||
898 | /* | |
899 | * Not only that, but the size of the entire the snapshot | |
900 | * header must also be representable in a size_t. | |
901 | */ | |
902 | size -= snap_count * sizeof (__le64); | |
903 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
904 | return false; | |
905 | ||
906 | return true; | |
8e94af8e AE |
907 | } |
908 | ||
602adf40 | 909 | /* |
bb23e37a AE |
910 | * Fill an rbd image header with information from the given format 1 |
911 | * on-disk header. | |
602adf40 | 912 | */ |
662518b1 | 913 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 914 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 915 | { |
662518b1 | 916 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
917 | bool first_time = header->object_prefix == NULL; |
918 | struct ceph_snap_context *snapc; | |
919 | char *object_prefix = NULL; | |
920 | char *snap_names = NULL; | |
921 | u64 *snap_sizes = NULL; | |
ccece235 | 922 | u32 snap_count; |
d2bb24e5 | 923 | size_t size; |
bb23e37a | 924 | int ret = -ENOMEM; |
621901d6 | 925 | u32 i; |
602adf40 | 926 | |
bb23e37a | 927 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 928 | |
bb23e37a AE |
929 | if (first_time) { |
930 | size_t len; | |
103a150f | 931 | |
bb23e37a AE |
932 | len = strnlen(ondisk->object_prefix, |
933 | sizeof (ondisk->object_prefix)); | |
934 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
935 | if (!object_prefix) | |
936 | return -ENOMEM; | |
937 | memcpy(object_prefix, ondisk->object_prefix, len); | |
938 | object_prefix[len] = '\0'; | |
939 | } | |
00f1f36f | 940 | |
bb23e37a | 941 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 942 | |
bb23e37a AE |
943 | snap_count = le32_to_cpu(ondisk->snap_count); |
944 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
945 | if (!snapc) | |
946 | goto out_err; | |
947 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 948 | if (snap_count) { |
bb23e37a | 949 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
950 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
951 | ||
bb23e37a | 952 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 953 | |
bb23e37a AE |
954 | if (snap_names_len > (u64)SIZE_MAX) |
955 | goto out_2big; | |
956 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
957 | if (!snap_names) | |
6a52325f AE |
958 | goto out_err; |
959 | ||
bb23e37a | 960 | /* ...as well as the array of their sizes. */ |
621901d6 | 961 | |
d2bb24e5 | 962 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
963 | snap_sizes = kmalloc(size, GFP_KERNEL); |
964 | if (!snap_sizes) | |
6a52325f | 965 | goto out_err; |
bb23e37a | 966 | |
f785cc1d | 967 | /* |
bb23e37a AE |
968 | * Copy the names, and fill in each snapshot's id |
969 | * and size. | |
970 | * | |
99a41ebc | 971 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 972 | * ondisk buffer we're working with has |
f785cc1d AE |
973 | * snap_names_len bytes beyond the end of the |
974 | * snapshot id array, this memcpy() is safe. | |
975 | */ | |
bb23e37a AE |
976 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
977 | snaps = ondisk->snaps; | |
978 | for (i = 0; i < snap_count; i++) { | |
979 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
980 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
981 | } | |
602adf40 | 982 | } |
6a52325f | 983 | |
bb23e37a | 984 | /* We won't fail any more, fill in the header */ |
621901d6 | 985 | |
bb23e37a AE |
986 | if (first_time) { |
987 | header->object_prefix = object_prefix; | |
988 | header->obj_order = ondisk->options.order; | |
989 | header->crypt_type = ondisk->options.crypt_type; | |
990 | header->comp_type = ondisk->options.comp_type; | |
991 | /* The rest aren't used for format 1 images */ | |
992 | header->stripe_unit = 0; | |
993 | header->stripe_count = 0; | |
994 | header->features = 0; | |
602adf40 | 995 | } else { |
662518b1 AE |
996 | ceph_put_snap_context(header->snapc); |
997 | kfree(header->snap_names); | |
998 | kfree(header->snap_sizes); | |
602adf40 | 999 | } |
849b4260 | 1000 | |
bb23e37a | 1001 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1002 | |
f84344f3 | 1003 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1004 | header->snapc = snapc; |
1005 | header->snap_names = snap_names; | |
1006 | header->snap_sizes = snap_sizes; | |
468521c1 | 1007 | |
602adf40 | 1008 | return 0; |
bb23e37a AE |
1009 | out_2big: |
1010 | ret = -EIO; | |
6a52325f | 1011 | out_err: |
bb23e37a AE |
1012 | kfree(snap_sizes); |
1013 | kfree(snap_names); | |
1014 | ceph_put_snap_context(snapc); | |
1015 | kfree(object_prefix); | |
ccece235 | 1016 | |
bb23e37a | 1017 | return ret; |
602adf40 YS |
1018 | } |
1019 | ||
9682fc6d AE |
1020 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1021 | { | |
1022 | const char *snap_name; | |
1023 | ||
1024 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1025 | ||
1026 | /* Skip over names until we find the one we are looking for */ | |
1027 | ||
1028 | snap_name = rbd_dev->header.snap_names; | |
1029 | while (which--) | |
1030 | snap_name += strlen(snap_name) + 1; | |
1031 | ||
1032 | return kstrdup(snap_name, GFP_KERNEL); | |
1033 | } | |
1034 | ||
30d1cff8 AE |
1035 | /* |
1036 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1037 | * Note that result is for snapshots in *descending* order. | |
1038 | */ | |
1039 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1040 | { | |
1041 | u64 snap_id1 = *(u64 *)s1; | |
1042 | u64 snap_id2 = *(u64 *)s2; | |
1043 | ||
1044 | if (snap_id1 < snap_id2) | |
1045 | return 1; | |
1046 | return snap_id1 == snap_id2 ? 0 : -1; | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Search a snapshot context to see if the given snapshot id is | |
1051 | * present. | |
1052 | * | |
1053 | * Returns the position of the snapshot id in the array if it's found, | |
1054 | * or BAD_SNAP_INDEX otherwise. | |
1055 | * | |
1056 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1057 | * reverse order, highest snapshot id first. | |
1058 | */ | |
9682fc6d AE |
1059 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1060 | { | |
1061 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1062 | u64 *found; |
9682fc6d | 1063 | |
30d1cff8 AE |
1064 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1065 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1066 | |
30d1cff8 | 1067 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1068 | } |
1069 | ||
2ad3d716 AE |
1070 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1071 | u64 snap_id) | |
9e15b77d | 1072 | { |
54cac61f | 1073 | u32 which; |
da6a6b63 | 1074 | const char *snap_name; |
9e15b77d | 1075 | |
54cac61f AE |
1076 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1077 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1078 | return ERR_PTR(-ENOENT); |
54cac61f | 1079 | |
da6a6b63 JD |
1080 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1081 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1082 | } |
1083 | ||
1084 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1085 | { | |
9e15b77d AE |
1086 | if (snap_id == CEPH_NOSNAP) |
1087 | return RBD_SNAP_HEAD_NAME; | |
1088 | ||
54cac61f AE |
1089 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1090 | if (rbd_dev->image_format == 1) | |
1091 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1092 | |
54cac61f | 1093 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1094 | } |
1095 | ||
2ad3d716 AE |
1096 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1097 | u64 *snap_size) | |
602adf40 | 1098 | { |
2ad3d716 AE |
1099 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1100 | if (snap_id == CEPH_NOSNAP) { | |
1101 | *snap_size = rbd_dev->header.image_size; | |
1102 | } else if (rbd_dev->image_format == 1) { | |
1103 | u32 which; | |
602adf40 | 1104 | |
2ad3d716 AE |
1105 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1106 | if (which == BAD_SNAP_INDEX) | |
1107 | return -ENOENT; | |
e86924a8 | 1108 | |
2ad3d716 AE |
1109 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1110 | } else { | |
1111 | u64 size = 0; | |
1112 | int ret; | |
1113 | ||
1114 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1115 | if (ret) | |
1116 | return ret; | |
1117 | ||
1118 | *snap_size = size; | |
1119 | } | |
1120 | return 0; | |
602adf40 YS |
1121 | } |
1122 | ||
2ad3d716 AE |
1123 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1124 | u64 *snap_features) | |
602adf40 | 1125 | { |
2ad3d716 AE |
1126 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1127 | if (snap_id == CEPH_NOSNAP) { | |
1128 | *snap_features = rbd_dev->header.features; | |
1129 | } else if (rbd_dev->image_format == 1) { | |
1130 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1131 | } else { |
2ad3d716 AE |
1132 | u64 features = 0; |
1133 | int ret; | |
8b0241f8 | 1134 | |
2ad3d716 AE |
1135 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1136 | if (ret) | |
1137 | return ret; | |
1138 | ||
1139 | *snap_features = features; | |
1140 | } | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1145 | { | |
8f4b7d98 | 1146 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1147 | u64 size = 0; |
1148 | u64 features = 0; | |
1149 | int ret; | |
1150 | ||
2ad3d716 AE |
1151 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1152 | if (ret) | |
1153 | return ret; | |
1154 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1155 | if (ret) | |
1156 | return ret; | |
1157 | ||
1158 | rbd_dev->mapping.size = size; | |
1159 | rbd_dev->mapping.features = features; | |
1160 | ||
8b0241f8 | 1161 | return 0; |
602adf40 YS |
1162 | } |
1163 | ||
d1cf5788 AE |
1164 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1165 | { | |
1166 | rbd_dev->mapping.size = 0; | |
1167 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1168 | } |
1169 | ||
7d5079aa HS |
1170 | static void rbd_segment_name_free(const char *name) |
1171 | { | |
1172 | /* The explicit cast here is needed to drop the const qualifier */ | |
1173 | ||
1174 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1175 | } | |
1176 | ||
98571b5a | 1177 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1178 | { |
65ccfe21 AE |
1179 | char *name; |
1180 | u64 segment; | |
1181 | int ret; | |
3a96d5cd | 1182 | char *name_format; |
602adf40 | 1183 | |
78c2a44a | 1184 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1185 | if (!name) |
1186 | return NULL; | |
1187 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1188 | name_format = "%s.%012llx"; |
1189 | if (rbd_dev->image_format == 2) | |
1190 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1191 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1192 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1193 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1194 | pr_err("error formatting segment name for #%llu (%d)\n", |
1195 | segment, ret); | |
7d5079aa | 1196 | rbd_segment_name_free(name); |
65ccfe21 AE |
1197 | name = NULL; |
1198 | } | |
602adf40 | 1199 | |
65ccfe21 AE |
1200 | return name; |
1201 | } | |
602adf40 | 1202 | |
65ccfe21 AE |
1203 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1204 | { | |
1205 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1206 | |
65ccfe21 AE |
1207 | return offset & (segment_size - 1); |
1208 | } | |
1209 | ||
1210 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1211 | u64 offset, u64 length) | |
1212 | { | |
1213 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1214 | ||
1215 | offset &= segment_size - 1; | |
1216 | ||
aafb230e | 1217 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1218 | if (offset + length > segment_size) |
1219 | length = segment_size - offset; | |
1220 | ||
1221 | return length; | |
602adf40 YS |
1222 | } |
1223 | ||
029bcbd8 JD |
1224 | /* |
1225 | * returns the size of an object in the image | |
1226 | */ | |
1227 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1228 | { | |
1229 | return 1 << header->obj_order; | |
1230 | } | |
1231 | ||
602adf40 YS |
1232 | /* |
1233 | * bio helpers | |
1234 | */ | |
1235 | ||
1236 | static void bio_chain_put(struct bio *chain) | |
1237 | { | |
1238 | struct bio *tmp; | |
1239 | ||
1240 | while (chain) { | |
1241 | tmp = chain; | |
1242 | chain = chain->bi_next; | |
1243 | bio_put(tmp); | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * zeros a bio chain, starting at specific offset | |
1249 | */ | |
1250 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1251 | { | |
7988613b KO |
1252 | struct bio_vec bv; |
1253 | struct bvec_iter iter; | |
602adf40 YS |
1254 | unsigned long flags; |
1255 | void *buf; | |
602adf40 YS |
1256 | int pos = 0; |
1257 | ||
1258 | while (chain) { | |
7988613b KO |
1259 | bio_for_each_segment(bv, chain, iter) { |
1260 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1261 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1262 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1263 | memset(buf + remainder, 0, |
7988613b KO |
1264 | bv.bv_len - remainder); |
1265 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1266 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1267 | } |
7988613b | 1268 | pos += bv.bv_len; |
602adf40 YS |
1269 | } |
1270 | ||
1271 | chain = chain->bi_next; | |
1272 | } | |
1273 | } | |
1274 | ||
b9434c5b AE |
1275 | /* |
1276 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1277 | * starting at the given byte offset from the start of the array and | |
1278 | * continuing up to the given end offset. The pages array is | |
1279 | * assumed to be big enough to hold all bytes up to the end. | |
1280 | */ | |
1281 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1282 | { | |
1283 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1284 | ||
1285 | rbd_assert(end > offset); | |
1286 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1287 | while (offset < end) { | |
1288 | size_t page_offset; | |
1289 | size_t length; | |
1290 | unsigned long flags; | |
1291 | void *kaddr; | |
1292 | ||
491205a8 GU |
1293 | page_offset = offset & ~PAGE_MASK; |
1294 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1295 | local_irq_save(flags); |
1296 | kaddr = kmap_atomic(*page); | |
1297 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1298 | flush_dcache_page(*page); |
b9434c5b AE |
1299 | kunmap_atomic(kaddr); |
1300 | local_irq_restore(flags); | |
1301 | ||
1302 | offset += length; | |
1303 | page++; | |
1304 | } | |
1305 | } | |
1306 | ||
602adf40 | 1307 | /* |
f7760dad AE |
1308 | * Clone a portion of a bio, starting at the given byte offset |
1309 | * and continuing for the number of bytes indicated. | |
602adf40 | 1310 | */ |
f7760dad AE |
1311 | static struct bio *bio_clone_range(struct bio *bio_src, |
1312 | unsigned int offset, | |
1313 | unsigned int len, | |
1314 | gfp_t gfpmask) | |
602adf40 | 1315 | { |
f7760dad AE |
1316 | struct bio *bio; |
1317 | ||
5341a627 | 1318 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1319 | if (!bio) |
1320 | return NULL; /* ENOMEM */ | |
602adf40 | 1321 | |
5341a627 | 1322 | bio_advance(bio, offset); |
4f024f37 | 1323 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1324 | |
1325 | return bio; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * Clone a portion of a bio chain, starting at the given byte offset | |
1330 | * into the first bio in the source chain and continuing for the | |
1331 | * number of bytes indicated. The result is another bio chain of | |
1332 | * exactly the given length, or a null pointer on error. | |
1333 | * | |
1334 | * The bio_src and offset parameters are both in-out. On entry they | |
1335 | * refer to the first source bio and the offset into that bio where | |
1336 | * the start of data to be cloned is located. | |
1337 | * | |
1338 | * On return, bio_src is updated to refer to the bio in the source | |
1339 | * chain that contains first un-cloned byte, and *offset will | |
1340 | * contain the offset of that byte within that bio. | |
1341 | */ | |
1342 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1343 | unsigned int *offset, | |
1344 | unsigned int len, | |
1345 | gfp_t gfpmask) | |
1346 | { | |
1347 | struct bio *bi = *bio_src; | |
1348 | unsigned int off = *offset; | |
1349 | struct bio *chain = NULL; | |
1350 | struct bio **end; | |
1351 | ||
1352 | /* Build up a chain of clone bios up to the limit */ | |
1353 | ||
4f024f37 | 1354 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1355 | return NULL; /* Nothing to clone */ |
602adf40 | 1356 | |
f7760dad AE |
1357 | end = &chain; |
1358 | while (len) { | |
1359 | unsigned int bi_size; | |
1360 | struct bio *bio; | |
1361 | ||
f5400b7a AE |
1362 | if (!bi) { |
1363 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1364 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1365 | } |
4f024f37 | 1366 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1367 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1368 | if (!bio) | |
1369 | goto out_err; /* ENOMEM */ | |
1370 | ||
1371 | *end = bio; | |
1372 | end = &bio->bi_next; | |
602adf40 | 1373 | |
f7760dad | 1374 | off += bi_size; |
4f024f37 | 1375 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1376 | bi = bi->bi_next; |
1377 | off = 0; | |
1378 | } | |
1379 | len -= bi_size; | |
1380 | } | |
1381 | *bio_src = bi; | |
1382 | *offset = off; | |
1383 | ||
1384 | return chain; | |
1385 | out_err: | |
1386 | bio_chain_put(chain); | |
602adf40 | 1387 | |
602adf40 YS |
1388 | return NULL; |
1389 | } | |
1390 | ||
926f9b3f AE |
1391 | /* |
1392 | * The default/initial value for all object request flags is 0. For | |
1393 | * each flag, once its value is set to 1 it is never reset to 0 | |
1394 | * again. | |
1395 | */ | |
57acbaa7 | 1396 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1397 | { |
57acbaa7 | 1398 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1399 | struct rbd_device *rbd_dev; |
1400 | ||
57acbaa7 | 1401 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1402 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1403 | obj_request); |
1404 | } | |
1405 | } | |
1406 | ||
57acbaa7 | 1407 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1408 | { |
1409 | smp_mb(); | |
57acbaa7 | 1410 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1411 | } |
1412 | ||
57acbaa7 | 1413 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1414 | { |
57acbaa7 AE |
1415 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1416 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1417 | |
57acbaa7 AE |
1418 | if (obj_request_img_data_test(obj_request)) |
1419 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1420 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1421 | obj_request); |
1422 | } | |
1423 | } | |
1424 | ||
57acbaa7 | 1425 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1426 | { |
1427 | smp_mb(); | |
57acbaa7 | 1428 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1429 | } |
1430 | ||
5679c59f AE |
1431 | /* |
1432 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1433 | * flag. The latter is set based on the "exists" value provided. | |
1434 | * | |
1435 | * Note that for our purposes once an object exists it never goes | |
1436 | * away again. It's possible that the response from two existence | |
1437 | * checks are separated by the creation of the target object, and | |
1438 | * the first ("doesn't exist") response arrives *after* the second | |
1439 | * ("does exist"). In that case we ignore the second one. | |
1440 | */ | |
1441 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1442 | bool exists) | |
1443 | { | |
1444 | if (exists) | |
1445 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1446 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1447 | smp_mb(); | |
1448 | } | |
1449 | ||
1450 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1451 | { | |
1452 | smp_mb(); | |
1453 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1454 | } | |
1455 | ||
1456 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1457 | { | |
1458 | smp_mb(); | |
1459 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1460 | } | |
1461 | ||
9638556a ID |
1462 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1463 | { | |
1464 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1465 | ||
1466 | return obj_request->img_offset < | |
1467 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1468 | } | |
1469 | ||
bf0d5f50 AE |
1470 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1471 | { | |
37206ee5 AE |
1472 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1473 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1474 | kref_get(&obj_request->kref); |
1475 | } | |
1476 | ||
1477 | static void rbd_obj_request_destroy(struct kref *kref); | |
1478 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1479 | { | |
1480 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1481 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1482 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1483 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1484 | } | |
1485 | ||
0f2d5be7 AE |
1486 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1487 | { | |
1488 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1489 | atomic_read(&img_request->kref.refcount)); | |
1490 | kref_get(&img_request->kref); | |
1491 | } | |
1492 | ||
e93f3152 AE |
1493 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1494 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1495 | static void rbd_img_request_destroy(struct kref *kref); |
1496 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1497 | { | |
1498 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1499 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1500 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1501 | if (img_request_child_test(img_request)) |
1502 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1503 | else | |
1504 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1505 | } |
1506 | ||
1507 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1508 | struct rbd_obj_request *obj_request) | |
1509 | { | |
25dcf954 AE |
1510 | rbd_assert(obj_request->img_request == NULL); |
1511 | ||
b155e86c | 1512 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1513 | obj_request->img_request = img_request; |
25dcf954 | 1514 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1515 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1516 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1517 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1518 | img_request->obj_request_count++; |
1519 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1520 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1521 | obj_request->which); | |
bf0d5f50 AE |
1522 | } |
1523 | ||
1524 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1525 | struct rbd_obj_request *obj_request) | |
1526 | { | |
1527 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1528 | |
37206ee5 AE |
1529 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1530 | obj_request->which); | |
bf0d5f50 | 1531 | list_del(&obj_request->links); |
25dcf954 AE |
1532 | rbd_assert(img_request->obj_request_count > 0); |
1533 | img_request->obj_request_count--; | |
1534 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1535 | obj_request->which = BAD_WHICH; | |
6365d33a | 1536 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1537 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1538 | obj_request->img_request = NULL; |
25dcf954 | 1539 | obj_request->callback = NULL; |
bf0d5f50 AE |
1540 | rbd_obj_request_put(obj_request); |
1541 | } | |
1542 | ||
1543 | static bool obj_request_type_valid(enum obj_request_type type) | |
1544 | { | |
1545 | switch (type) { | |
9969ebc5 | 1546 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1547 | case OBJ_REQUEST_BIO: |
788e2df3 | 1548 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1549 | return true; |
1550 | default: | |
1551 | return false; | |
1552 | } | |
1553 | } | |
1554 | ||
bf0d5f50 AE |
1555 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1556 | struct rbd_obj_request *obj_request) | |
1557 | { | |
71c20a06 | 1558 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1559 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1560 | } | |
1561 | ||
71c20a06 ID |
1562 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1563 | { | |
1564 | dout("%s %p\n", __func__, obj_request); | |
1565 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * Wait for an object request to complete. If interrupted, cancel the | |
1570 | * underlying osd request. | |
2894e1d7 ID |
1571 | * |
1572 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1573 | */ |
2894e1d7 ID |
1574 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1575 | unsigned long timeout) | |
71c20a06 | 1576 | { |
2894e1d7 | 1577 | long ret; |
71c20a06 ID |
1578 | |
1579 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1580 | ret = wait_for_completion_interruptible_timeout( |
1581 | &obj_request->completion, | |
1582 | ceph_timeout_jiffies(timeout)); | |
1583 | if (ret <= 0) { | |
1584 | if (ret == 0) | |
1585 | ret = -ETIMEDOUT; | |
71c20a06 | 1586 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1587 | } else { |
1588 | ret = 0; | |
71c20a06 ID |
1589 | } |
1590 | ||
2894e1d7 ID |
1591 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1592 | return ret; | |
1593 | } | |
1594 | ||
1595 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1596 | { | |
1597 | return __rbd_obj_request_wait(obj_request, 0); | |
1598 | } | |
1599 | ||
1600 | static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request, | |
1601 | unsigned long timeout) | |
1602 | { | |
1603 | return __rbd_obj_request_wait(obj_request, timeout); | |
71c20a06 ID |
1604 | } |
1605 | ||
bf0d5f50 AE |
1606 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1607 | { | |
55f27e09 | 1608 | |
37206ee5 | 1609 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1610 | |
1611 | /* | |
1612 | * If no error occurred, compute the aggregate transfer | |
1613 | * count for the image request. We could instead use | |
1614 | * atomic64_cmpxchg() to update it as each object request | |
1615 | * completes; not clear which way is better off hand. | |
1616 | */ | |
1617 | if (!img_request->result) { | |
1618 | struct rbd_obj_request *obj_request; | |
1619 | u64 xferred = 0; | |
1620 | ||
1621 | for_each_obj_request(img_request, obj_request) | |
1622 | xferred += obj_request->xferred; | |
1623 | img_request->xferred = xferred; | |
1624 | } | |
1625 | ||
bf0d5f50 AE |
1626 | if (img_request->callback) |
1627 | img_request->callback(img_request); | |
1628 | else | |
1629 | rbd_img_request_put(img_request); | |
1630 | } | |
1631 | ||
0c425248 AE |
1632 | /* |
1633 | * The default/initial value for all image request flags is 0. Each | |
1634 | * is conditionally set to 1 at image request initialization time | |
1635 | * and currently never change thereafter. | |
1636 | */ | |
1637 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1638 | { | |
1639 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1640 | smp_mb(); | |
1641 | } | |
1642 | ||
1643 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1644 | { | |
1645 | smp_mb(); | |
1646 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1647 | } | |
1648 | ||
90e98c52 GZ |
1649 | /* |
1650 | * Set the discard flag when the img_request is an discard request | |
1651 | */ | |
1652 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1653 | { | |
1654 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1655 | smp_mb(); | |
1656 | } | |
1657 | ||
1658 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1659 | { | |
1660 | smp_mb(); | |
1661 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1662 | } | |
1663 | ||
9849e986 AE |
1664 | static void img_request_child_set(struct rbd_img_request *img_request) |
1665 | { | |
1666 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1667 | smp_mb(); | |
1668 | } | |
1669 | ||
e93f3152 AE |
1670 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1671 | { | |
1672 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1673 | smp_mb(); | |
1674 | } | |
1675 | ||
9849e986 AE |
1676 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1677 | { | |
1678 | smp_mb(); | |
1679 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1680 | } | |
1681 | ||
d0b2e944 AE |
1682 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1683 | { | |
1684 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1685 | smp_mb(); | |
1686 | } | |
1687 | ||
a2acd00e AE |
1688 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1689 | { | |
1690 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1691 | smp_mb(); | |
1692 | } | |
1693 | ||
d0b2e944 AE |
1694 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1695 | { | |
1696 | smp_mb(); | |
1697 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1698 | } | |
1699 | ||
3b434a2a JD |
1700 | static enum obj_operation_type |
1701 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1702 | { | |
1703 | if (img_request_write_test(img_request)) | |
1704 | return OBJ_OP_WRITE; | |
1705 | else if (img_request_discard_test(img_request)) | |
1706 | return OBJ_OP_DISCARD; | |
1707 | else | |
1708 | return OBJ_OP_READ; | |
1709 | } | |
1710 | ||
6e2a4505 AE |
1711 | static void |
1712 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1713 | { | |
b9434c5b AE |
1714 | u64 xferred = obj_request->xferred; |
1715 | u64 length = obj_request->length; | |
1716 | ||
6e2a4505 AE |
1717 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1718 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1719 | xferred, length); |
6e2a4505 | 1720 | /* |
17c1cc1d JD |
1721 | * ENOENT means a hole in the image. We zero-fill the entire |
1722 | * length of the request. A short read also implies zero-fill | |
1723 | * to the end of the request. An error requires the whole | |
1724 | * length of the request to be reported finished with an error | |
1725 | * to the block layer. In each case we update the xferred | |
1726 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1727 | */ |
b9434c5b | 1728 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1729 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1730 | if (obj_request->type == OBJ_REQUEST_BIO) |
1731 | zero_bio_chain(obj_request->bio_list, 0); | |
1732 | else | |
1733 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1734 | obj_request->result = 0; |
b9434c5b AE |
1735 | } else if (xferred < length && !obj_request->result) { |
1736 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1737 | zero_bio_chain(obj_request->bio_list, xferred); | |
1738 | else | |
1739 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1740 | } |
17c1cc1d | 1741 | obj_request->xferred = length; |
6e2a4505 AE |
1742 | obj_request_done_set(obj_request); |
1743 | } | |
1744 | ||
bf0d5f50 AE |
1745 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1746 | { | |
37206ee5 AE |
1747 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1748 | obj_request->callback); | |
bf0d5f50 AE |
1749 | if (obj_request->callback) |
1750 | obj_request->callback(obj_request); | |
788e2df3 AE |
1751 | else |
1752 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1753 | } |
1754 | ||
c47f9371 | 1755 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1756 | { |
1757 | dout("%s: obj %p\n", __func__, obj_request); | |
1758 | obj_request_done_set(obj_request); | |
1759 | } | |
1760 | ||
c47f9371 | 1761 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1762 | { |
57acbaa7 | 1763 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1764 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1765 | bool layered = false; |
1766 | ||
1767 | if (obj_request_img_data_test(obj_request)) { | |
1768 | img_request = obj_request->img_request; | |
1769 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1770 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1771 | } |
8b3e1a56 AE |
1772 | |
1773 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1774 | obj_request, img_request, obj_request->result, | |
1775 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1776 | if (layered && obj_request->result == -ENOENT && |
1777 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1778 | rbd_img_parent_read(obj_request); |
1779 | else if (img_request) | |
6e2a4505 AE |
1780 | rbd_img_obj_request_read_callback(obj_request); |
1781 | else | |
1782 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1783 | } |
1784 | ||
c47f9371 | 1785 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1786 | { |
1b83bef2 SW |
1787 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1788 | obj_request->result, obj_request->length); | |
1789 | /* | |
8b3e1a56 AE |
1790 | * There is no such thing as a successful short write. Set |
1791 | * it to our originally-requested length. | |
1b83bef2 SW |
1792 | */ |
1793 | obj_request->xferred = obj_request->length; | |
07741308 | 1794 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1795 | } |
1796 | ||
90e98c52 GZ |
1797 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1798 | { | |
1799 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1800 | obj_request->result, obj_request->length); | |
1801 | /* | |
1802 | * There is no such thing as a successful short discard. Set | |
1803 | * it to our originally-requested length. | |
1804 | */ | |
1805 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1806 | /* discarding a non-existent object is not a problem */ |
1807 | if (obj_request->result == -ENOENT) | |
1808 | obj_request->result = 0; | |
90e98c52 GZ |
1809 | obj_request_done_set(obj_request); |
1810 | } | |
1811 | ||
fbfab539 AE |
1812 | /* |
1813 | * For a simple stat call there's nothing to do. We'll do more if | |
1814 | * this is part of a write sequence for a layered image. | |
1815 | */ | |
c47f9371 | 1816 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1817 | { |
37206ee5 | 1818 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1819 | obj_request_done_set(obj_request); |
1820 | } | |
1821 | ||
2761713d ID |
1822 | static void rbd_osd_call_callback(struct rbd_obj_request *obj_request) |
1823 | { | |
1824 | dout("%s: obj %p\n", __func__, obj_request); | |
1825 | ||
1826 | if (obj_request_img_data_test(obj_request)) | |
1827 | rbd_osd_copyup_callback(obj_request); | |
1828 | else | |
1829 | obj_request_done_set(obj_request); | |
1830 | } | |
1831 | ||
bf0d5f50 AE |
1832 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1833 | struct ceph_msg *msg) | |
1834 | { | |
1835 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1836 | u16 opcode; |
1837 | ||
37206ee5 | 1838 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1839 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1840 | if (obj_request_img_data_test(obj_request)) { |
1841 | rbd_assert(obj_request->img_request); | |
1842 | rbd_assert(obj_request->which != BAD_WHICH); | |
1843 | } else { | |
1844 | rbd_assert(obj_request->which == BAD_WHICH); | |
1845 | } | |
bf0d5f50 | 1846 | |
1b83bef2 SW |
1847 | if (osd_req->r_result < 0) |
1848 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1849 | |
7cc69d42 | 1850 | rbd_assert(osd_req->r_num_ops <= CEPH_OSD_MAX_OP); |
bf0d5f50 | 1851 | |
c47f9371 AE |
1852 | /* |
1853 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1854 | * passed to the block layer, which just supports a 32-bit |
1855 | * length field. | |
c47f9371 | 1856 | */ |
1b83bef2 | 1857 | obj_request->xferred = osd_req->r_reply_op_len[0]; |
8b3e1a56 | 1858 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1859 | |
79528734 | 1860 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1861 | switch (opcode) { |
1862 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1863 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1864 | break; |
0ccd5926 | 1865 | case CEPH_OSD_OP_SETALLOCHINT: |
e30b7577 ID |
1866 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE || |
1867 | osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL); | |
0ccd5926 | 1868 | /* fall through */ |
bf0d5f50 | 1869 | case CEPH_OSD_OP_WRITE: |
e30b7577 | 1870 | case CEPH_OSD_OP_WRITEFULL: |
c47f9371 | 1871 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1872 | break; |
fbfab539 | 1873 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1874 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1875 | break; |
90e98c52 GZ |
1876 | case CEPH_OSD_OP_DELETE: |
1877 | case CEPH_OSD_OP_TRUNCATE: | |
1878 | case CEPH_OSD_OP_ZERO: | |
1879 | rbd_osd_discard_callback(obj_request); | |
1880 | break; | |
36be9a76 | 1881 | case CEPH_OSD_OP_CALL: |
2761713d ID |
1882 | rbd_osd_call_callback(obj_request); |
1883 | break; | |
b8d70035 | 1884 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1885 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1886 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1887 | break; |
bf0d5f50 | 1888 | default: |
9584d508 | 1889 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1890 | obj_request->object_name, (unsigned short) opcode); |
1891 | break; | |
1892 | } | |
1893 | ||
07741308 | 1894 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1895 | rbd_obj_request_complete(obj_request); |
1896 | } | |
1897 | ||
9d4df01f | 1898 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1899 | { |
1900 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1901 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1902 | u64 snap_id; |
430c28c3 | 1903 | |
8c042b0d | 1904 | rbd_assert(osd_req != NULL); |
430c28c3 | 1905 | |
9d4df01f | 1906 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1907 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1908 | NULL, snap_id, NULL); |
1909 | } | |
1910 | ||
1911 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1912 | { | |
1913 | struct rbd_img_request *img_request = obj_request->img_request; | |
1914 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1915 | struct ceph_snap_context *snapc; | |
1916 | struct timespec mtime = CURRENT_TIME; | |
1917 | ||
1918 | rbd_assert(osd_req != NULL); | |
1919 | ||
1920 | snapc = img_request ? img_request->snapc : NULL; | |
1921 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1922 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1923 | } |
1924 | ||
0ccd5926 ID |
1925 | /* |
1926 | * Create an osd request. A read request has one osd op (read). | |
1927 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1928 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1929 | * technically osd watch is a write request, hence this distinction.) | |
1930 | */ | |
bf0d5f50 AE |
1931 | static struct ceph_osd_request *rbd_osd_req_create( |
1932 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1933 | enum obj_operation_type op_type, |
deb236b3 | 1934 | unsigned int num_ops, |
430c28c3 | 1935 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1936 | { |
bf0d5f50 AE |
1937 | struct ceph_snap_context *snapc = NULL; |
1938 | struct ceph_osd_client *osdc; | |
1939 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1940 | |
90e98c52 GZ |
1941 | if (obj_request_img_data_test(obj_request) && |
1942 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1943 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1944 | if (op_type == OBJ_OP_WRITE) { |
1945 | rbd_assert(img_request_write_test(img_request)); | |
1946 | } else { | |
1947 | rbd_assert(img_request_discard_test(img_request)); | |
1948 | } | |
6d2940c8 | 1949 | snapc = img_request->snapc; |
bf0d5f50 AE |
1950 | } |
1951 | ||
6d2940c8 | 1952 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1953 | |
1954 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1955 | |
1956 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 ID |
1957 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
1958 | GFP_ATOMIC); | |
bf0d5f50 AE |
1959 | if (!osd_req) |
1960 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1961 | |
90e98c52 | 1962 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1963 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1964 | else |
bf0d5f50 | 1965 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1966 | |
1967 | osd_req->r_callback = rbd_osd_req_callback; | |
1968 | osd_req->r_priv = obj_request; | |
1969 | ||
3c972c95 ID |
1970 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1971 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1972 | |
bf0d5f50 AE |
1973 | return osd_req; |
1974 | } | |
1975 | ||
0eefd470 | 1976 | /* |
d3246fb0 JD |
1977 | * Create a copyup osd request based on the information in the object |
1978 | * request supplied. A copyup request has two or three osd ops, a | |
1979 | * copyup method call, potentially a hint op, and a write or truncate | |
1980 | * or zero op. | |
0eefd470 AE |
1981 | */ |
1982 | static struct ceph_osd_request * | |
1983 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1984 | { | |
1985 | struct rbd_img_request *img_request; | |
1986 | struct ceph_snap_context *snapc; | |
1987 | struct rbd_device *rbd_dev; | |
1988 | struct ceph_osd_client *osdc; | |
1989 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1990 | int num_osd_ops = 3; |
0eefd470 AE |
1991 | |
1992 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1993 | img_request = obj_request->img_request; | |
1994 | rbd_assert(img_request); | |
d3246fb0 JD |
1995 | rbd_assert(img_request_write_test(img_request) || |
1996 | img_request_discard_test(img_request)); | |
0eefd470 | 1997 | |
d3246fb0 JD |
1998 | if (img_request_discard_test(img_request)) |
1999 | num_osd_ops = 2; | |
2000 | ||
2001 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
2002 | |
2003 | snapc = img_request->snapc; | |
2004 | rbd_dev = img_request->rbd_dev; | |
2005 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 JD |
2006 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
2007 | false, GFP_ATOMIC); | |
0eefd470 AE |
2008 | if (!osd_req) |
2009 | return NULL; /* ENOMEM */ | |
2010 | ||
2011 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
2012 | osd_req->r_callback = rbd_osd_req_callback; | |
2013 | osd_req->r_priv = obj_request; | |
2014 | ||
3c972c95 ID |
2015 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
2016 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 2017 | |
0eefd470 AE |
2018 | return osd_req; |
2019 | } | |
2020 | ||
2021 | ||
bf0d5f50 AE |
2022 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2023 | { | |
2024 | ceph_osdc_put_request(osd_req); | |
2025 | } | |
2026 | ||
2027 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2028 | ||
2029 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2030 | u64 offset, u64 length, | |
2031 | enum obj_request_type type) | |
2032 | { | |
2033 | struct rbd_obj_request *obj_request; | |
2034 | size_t size; | |
2035 | char *name; | |
2036 | ||
2037 | rbd_assert(obj_request_type_valid(type)); | |
2038 | ||
2039 | size = strlen(object_name) + 1; | |
5a60e876 | 2040 | name = kmalloc(size, GFP_NOIO); |
f907ad55 | 2041 | if (!name) |
bf0d5f50 AE |
2042 | return NULL; |
2043 | ||
5a60e876 | 2044 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
f907ad55 AE |
2045 | if (!obj_request) { |
2046 | kfree(name); | |
2047 | return NULL; | |
2048 | } | |
2049 | ||
bf0d5f50 AE |
2050 | obj_request->object_name = memcpy(name, object_name, size); |
2051 | obj_request->offset = offset; | |
2052 | obj_request->length = length; | |
926f9b3f | 2053 | obj_request->flags = 0; |
bf0d5f50 AE |
2054 | obj_request->which = BAD_WHICH; |
2055 | obj_request->type = type; | |
2056 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2057 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2058 | kref_init(&obj_request->kref); |
2059 | ||
37206ee5 AE |
2060 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2061 | offset, length, (int)type, obj_request); | |
2062 | ||
bf0d5f50 AE |
2063 | return obj_request; |
2064 | } | |
2065 | ||
2066 | static void rbd_obj_request_destroy(struct kref *kref) | |
2067 | { | |
2068 | struct rbd_obj_request *obj_request; | |
2069 | ||
2070 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2071 | ||
37206ee5 AE |
2072 | dout("%s: obj %p\n", __func__, obj_request); |
2073 | ||
bf0d5f50 AE |
2074 | rbd_assert(obj_request->img_request == NULL); |
2075 | rbd_assert(obj_request->which == BAD_WHICH); | |
2076 | ||
2077 | if (obj_request->osd_req) | |
2078 | rbd_osd_req_destroy(obj_request->osd_req); | |
2079 | ||
2080 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2081 | switch (obj_request->type) { | |
9969ebc5 AE |
2082 | case OBJ_REQUEST_NODATA: |
2083 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2084 | case OBJ_REQUEST_BIO: |
2085 | if (obj_request->bio_list) | |
2086 | bio_chain_put(obj_request->bio_list); | |
2087 | break; | |
788e2df3 AE |
2088 | case OBJ_REQUEST_PAGES: |
2089 | if (obj_request->pages) | |
2090 | ceph_release_page_vector(obj_request->pages, | |
2091 | obj_request->page_count); | |
2092 | break; | |
bf0d5f50 AE |
2093 | } |
2094 | ||
f907ad55 | 2095 | kfree(obj_request->object_name); |
868311b1 AE |
2096 | obj_request->object_name = NULL; |
2097 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2098 | } |
2099 | ||
fb65d228 AE |
2100 | /* It's OK to call this for a device with no parent */ |
2101 | ||
2102 | static void rbd_spec_put(struct rbd_spec *spec); | |
2103 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2104 | { | |
2105 | rbd_dev_remove_parent(rbd_dev); | |
2106 | rbd_spec_put(rbd_dev->parent_spec); | |
2107 | rbd_dev->parent_spec = NULL; | |
2108 | rbd_dev->parent_overlap = 0; | |
2109 | } | |
2110 | ||
a2acd00e AE |
2111 | /* |
2112 | * Parent image reference counting is used to determine when an | |
2113 | * image's parent fields can be safely torn down--after there are no | |
2114 | * more in-flight requests to the parent image. When the last | |
2115 | * reference is dropped, cleaning them up is safe. | |
2116 | */ | |
2117 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2118 | { | |
2119 | int counter; | |
2120 | ||
2121 | if (!rbd_dev->parent_spec) | |
2122 | return; | |
2123 | ||
2124 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2125 | if (counter > 0) | |
2126 | return; | |
2127 | ||
2128 | /* Last reference; clean up parent data structures */ | |
2129 | ||
2130 | if (!counter) | |
2131 | rbd_dev_unparent(rbd_dev); | |
2132 | else | |
9584d508 | 2133 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2134 | } |
2135 | ||
2136 | /* | |
2137 | * If an image has a non-zero parent overlap, get a reference to its | |
2138 | * parent. | |
2139 | * | |
2140 | * Returns true if the rbd device has a parent with a non-zero | |
2141 | * overlap and a reference for it was successfully taken, or | |
2142 | * false otherwise. | |
2143 | */ | |
2144 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2145 | { | |
ae43e9d0 | 2146 | int counter = 0; |
a2acd00e AE |
2147 | |
2148 | if (!rbd_dev->parent_spec) | |
2149 | return false; | |
2150 | ||
ae43e9d0 ID |
2151 | down_read(&rbd_dev->header_rwsem); |
2152 | if (rbd_dev->parent_overlap) | |
2153 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2154 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2155 | |
2156 | if (counter < 0) | |
9584d508 | 2157 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2158 | |
ae43e9d0 | 2159 | return counter > 0; |
a2acd00e AE |
2160 | } |
2161 | ||
bf0d5f50 AE |
2162 | /* |
2163 | * Caller is responsible for filling in the list of object requests | |
2164 | * that comprises the image request, and the Linux request pointer | |
2165 | * (if there is one). | |
2166 | */ | |
cc344fa1 AE |
2167 | static struct rbd_img_request *rbd_img_request_create( |
2168 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2169 | u64 offset, u64 length, |
6d2940c8 | 2170 | enum obj_operation_type op_type, |
4e752f0a | 2171 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2172 | { |
2173 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2174 | |
7a716aac | 2175 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2176 | if (!img_request) |
2177 | return NULL; | |
2178 | ||
bf0d5f50 AE |
2179 | img_request->rq = NULL; |
2180 | img_request->rbd_dev = rbd_dev; | |
2181 | img_request->offset = offset; | |
2182 | img_request->length = length; | |
0c425248 | 2183 | img_request->flags = 0; |
90e98c52 GZ |
2184 | if (op_type == OBJ_OP_DISCARD) { |
2185 | img_request_discard_set(img_request); | |
2186 | img_request->snapc = snapc; | |
2187 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2188 | img_request_write_set(img_request); |
4e752f0a | 2189 | img_request->snapc = snapc; |
0c425248 | 2190 | } else { |
bf0d5f50 | 2191 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2192 | } |
a2acd00e | 2193 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2194 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2195 | spin_lock_init(&img_request->completion_lock); |
2196 | img_request->next_completion = 0; | |
2197 | img_request->callback = NULL; | |
a5a337d4 | 2198 | img_request->result = 0; |
bf0d5f50 AE |
2199 | img_request->obj_request_count = 0; |
2200 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2201 | kref_init(&img_request->kref); | |
2202 | ||
37206ee5 | 2203 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2204 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2205 | |
bf0d5f50 AE |
2206 | return img_request; |
2207 | } | |
2208 | ||
2209 | static void rbd_img_request_destroy(struct kref *kref) | |
2210 | { | |
2211 | struct rbd_img_request *img_request; | |
2212 | struct rbd_obj_request *obj_request; | |
2213 | struct rbd_obj_request *next_obj_request; | |
2214 | ||
2215 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2216 | ||
37206ee5 AE |
2217 | dout("%s: img %p\n", __func__, img_request); |
2218 | ||
bf0d5f50 AE |
2219 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2220 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2221 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2222 | |
a2acd00e AE |
2223 | if (img_request_layered_test(img_request)) { |
2224 | img_request_layered_clear(img_request); | |
2225 | rbd_dev_parent_put(img_request->rbd_dev); | |
2226 | } | |
2227 | ||
bef95455 JD |
2228 | if (img_request_write_test(img_request) || |
2229 | img_request_discard_test(img_request)) | |
812164f8 | 2230 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2231 | |
1c2a9dfe | 2232 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2233 | } |
2234 | ||
e93f3152 AE |
2235 | static struct rbd_img_request *rbd_parent_request_create( |
2236 | struct rbd_obj_request *obj_request, | |
2237 | u64 img_offset, u64 length) | |
2238 | { | |
2239 | struct rbd_img_request *parent_request; | |
2240 | struct rbd_device *rbd_dev; | |
2241 | ||
2242 | rbd_assert(obj_request->img_request); | |
2243 | rbd_dev = obj_request->img_request->rbd_dev; | |
2244 | ||
4e752f0a | 2245 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2246 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2247 | if (!parent_request) |
2248 | return NULL; | |
2249 | ||
2250 | img_request_child_set(parent_request); | |
2251 | rbd_obj_request_get(obj_request); | |
2252 | parent_request->obj_request = obj_request; | |
2253 | ||
2254 | return parent_request; | |
2255 | } | |
2256 | ||
2257 | static void rbd_parent_request_destroy(struct kref *kref) | |
2258 | { | |
2259 | struct rbd_img_request *parent_request; | |
2260 | struct rbd_obj_request *orig_request; | |
2261 | ||
2262 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2263 | orig_request = parent_request->obj_request; | |
2264 | ||
2265 | parent_request->obj_request = NULL; | |
2266 | rbd_obj_request_put(orig_request); | |
2267 | img_request_child_clear(parent_request); | |
2268 | ||
2269 | rbd_img_request_destroy(kref); | |
2270 | } | |
2271 | ||
1217857f AE |
2272 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2273 | { | |
6365d33a | 2274 | struct rbd_img_request *img_request; |
1217857f AE |
2275 | unsigned int xferred; |
2276 | int result; | |
8b3e1a56 | 2277 | bool more; |
1217857f | 2278 | |
6365d33a AE |
2279 | rbd_assert(obj_request_img_data_test(obj_request)); |
2280 | img_request = obj_request->img_request; | |
2281 | ||
1217857f AE |
2282 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2283 | xferred = (unsigned int)obj_request->xferred; | |
2284 | result = obj_request->result; | |
2285 | if (result) { | |
2286 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2287 | enum obj_operation_type op_type; |
2288 | ||
90e98c52 GZ |
2289 | if (img_request_discard_test(img_request)) |
2290 | op_type = OBJ_OP_DISCARD; | |
2291 | else if (img_request_write_test(img_request)) | |
2292 | op_type = OBJ_OP_WRITE; | |
2293 | else | |
2294 | op_type = OBJ_OP_READ; | |
1217857f | 2295 | |
9584d508 | 2296 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2297 | obj_op_name(op_type), obj_request->length, |
2298 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2299 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2300 | result, xferred); |
2301 | if (!img_request->result) | |
2302 | img_request->result = result; | |
082a75da ID |
2303 | /* |
2304 | * Need to end I/O on the entire obj_request worth of | |
2305 | * bytes in case of error. | |
2306 | */ | |
2307 | xferred = obj_request->length; | |
1217857f AE |
2308 | } |
2309 | ||
f1a4739f AE |
2310 | /* Image object requests don't own their page array */ |
2311 | ||
2312 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2313 | obj_request->pages = NULL; | |
2314 | obj_request->page_count = 0; | |
2315 | } | |
2316 | ||
8b3e1a56 AE |
2317 | if (img_request_child_test(img_request)) { |
2318 | rbd_assert(img_request->obj_request != NULL); | |
2319 | more = obj_request->which < img_request->obj_request_count - 1; | |
2320 | } else { | |
2321 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2322 | |
2323 | more = blk_update_request(img_request->rq, result, xferred); | |
2324 | if (!more) | |
2325 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2326 | } |
2327 | ||
2328 | return more; | |
1217857f AE |
2329 | } |
2330 | ||
2169238d AE |
2331 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2332 | { | |
2333 | struct rbd_img_request *img_request; | |
2334 | u32 which = obj_request->which; | |
2335 | bool more = true; | |
2336 | ||
6365d33a | 2337 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2338 | img_request = obj_request->img_request; |
2339 | ||
2340 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2341 | rbd_assert(img_request != NULL); | |
2169238d AE |
2342 | rbd_assert(img_request->obj_request_count > 0); |
2343 | rbd_assert(which != BAD_WHICH); | |
2344 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2345 | |
2346 | spin_lock_irq(&img_request->completion_lock); | |
2347 | if (which != img_request->next_completion) | |
2348 | goto out; | |
2349 | ||
2350 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2351 | rbd_assert(more); |
2352 | rbd_assert(which < img_request->obj_request_count); | |
2353 | ||
2354 | if (!obj_request_done_test(obj_request)) | |
2355 | break; | |
1217857f | 2356 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2357 | which++; |
2358 | } | |
2359 | ||
2360 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2361 | img_request->next_completion = which; | |
2362 | out: | |
2363 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2364 | rbd_img_request_put(img_request); |
2169238d AE |
2365 | |
2366 | if (!more) | |
2367 | rbd_img_request_complete(img_request); | |
2368 | } | |
2369 | ||
3b434a2a JD |
2370 | /* |
2371 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2372 | * them for submission. num_ops is the current number of | |
2373 | * osd operations already to the object request. | |
2374 | */ | |
2375 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2376 | struct ceph_osd_request *osd_request, | |
2377 | enum obj_operation_type op_type, | |
2378 | unsigned int num_ops) | |
2379 | { | |
2380 | struct rbd_img_request *img_request = obj_request->img_request; | |
2381 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2382 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2383 | u64 offset = obj_request->offset; | |
2384 | u64 length = obj_request->length; | |
2385 | u64 img_end; | |
2386 | u16 opcode; | |
2387 | ||
2388 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2389 | if (!offset && length == object_size && |
2390 | (!img_request_layered_test(img_request) || | |
2391 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2392 | opcode = CEPH_OSD_OP_DELETE; |
2393 | } else if ((offset + length == object_size)) { | |
2394 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2395 | } else { | |
2396 | down_read(&rbd_dev->header_rwsem); | |
2397 | img_end = rbd_dev->header.image_size; | |
2398 | up_read(&rbd_dev->header_rwsem); | |
2399 | ||
2400 | if (obj_request->img_offset + length == img_end) | |
2401 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2402 | else | |
2403 | opcode = CEPH_OSD_OP_ZERO; | |
2404 | } | |
2405 | } else if (op_type == OBJ_OP_WRITE) { | |
e30b7577 ID |
2406 | if (!offset && length == object_size) |
2407 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2408 | else | |
2409 | opcode = CEPH_OSD_OP_WRITE; | |
3b434a2a JD |
2410 | osd_req_op_alloc_hint_init(osd_request, num_ops, |
2411 | object_size, object_size); | |
2412 | num_ops++; | |
2413 | } else { | |
2414 | opcode = CEPH_OSD_OP_READ; | |
2415 | } | |
2416 | ||
7e868b6e | 2417 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2418 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2419 | else |
2420 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2421 | offset, length, 0, 0); | |
2422 | ||
3b434a2a JD |
2423 | if (obj_request->type == OBJ_REQUEST_BIO) |
2424 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2425 | obj_request->bio_list, length); | |
2426 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2427 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2428 | obj_request->pages, length, | |
2429 | offset & ~PAGE_MASK, false, false); | |
2430 | ||
2431 | /* Discards are also writes */ | |
2432 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2433 | rbd_osd_req_format_write(obj_request); | |
2434 | else | |
2435 | rbd_osd_req_format_read(obj_request); | |
2436 | } | |
2437 | ||
f1a4739f AE |
2438 | /* |
2439 | * Split up an image request into one or more object requests, each | |
2440 | * to a different object. The "type" parameter indicates whether | |
2441 | * "data_desc" is the pointer to the head of a list of bio | |
2442 | * structures, or the base of a page array. In either case this | |
2443 | * function assumes data_desc describes memory sufficient to hold | |
2444 | * all data described by the image request. | |
2445 | */ | |
2446 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2447 | enum obj_request_type type, | |
2448 | void *data_desc) | |
bf0d5f50 AE |
2449 | { |
2450 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2451 | struct rbd_obj_request *obj_request = NULL; | |
2452 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2453 | struct bio *bio_list = NULL; |
f1a4739f | 2454 | unsigned int bio_offset = 0; |
a158073c | 2455 | struct page **pages = NULL; |
6d2940c8 | 2456 | enum obj_operation_type op_type; |
7da22d29 | 2457 | u64 img_offset; |
bf0d5f50 | 2458 | u64 resid; |
bf0d5f50 | 2459 | |
f1a4739f AE |
2460 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2461 | (int)type, data_desc); | |
37206ee5 | 2462 | |
7da22d29 | 2463 | img_offset = img_request->offset; |
bf0d5f50 | 2464 | resid = img_request->length; |
4dda41d3 | 2465 | rbd_assert(resid > 0); |
3b434a2a | 2466 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2467 | |
2468 | if (type == OBJ_REQUEST_BIO) { | |
2469 | bio_list = data_desc; | |
4f024f37 KO |
2470 | rbd_assert(img_offset == |
2471 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2472 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2473 | pages = data_desc; |
2474 | } | |
2475 | ||
bf0d5f50 | 2476 | while (resid) { |
2fa12320 | 2477 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2478 | const char *object_name; |
bf0d5f50 AE |
2479 | u64 offset; |
2480 | u64 length; | |
2481 | ||
7da22d29 | 2482 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2483 | if (!object_name) |
2484 | goto out_unwind; | |
7da22d29 AE |
2485 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2486 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2487 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2488 | offset, length, type); |
78c2a44a AE |
2489 | /* object request has its own copy of the object name */ |
2490 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2491 | if (!obj_request) |
2492 | goto out_unwind; | |
62054da6 | 2493 | |
03507db6 JD |
2494 | /* |
2495 | * set obj_request->img_request before creating the | |
2496 | * osd_request so that it gets the right snapc | |
2497 | */ | |
2498 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2499 | |
f1a4739f AE |
2500 | if (type == OBJ_REQUEST_BIO) { |
2501 | unsigned int clone_size; | |
2502 | ||
2503 | rbd_assert(length <= (u64)UINT_MAX); | |
2504 | clone_size = (unsigned int)length; | |
2505 | obj_request->bio_list = | |
2506 | bio_chain_clone_range(&bio_list, | |
2507 | &bio_offset, | |
2508 | clone_size, | |
2509 | GFP_ATOMIC); | |
2510 | if (!obj_request->bio_list) | |
62054da6 | 2511 | goto out_unwind; |
90e98c52 | 2512 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2513 | unsigned int page_count; |
2514 | ||
2515 | obj_request->pages = pages; | |
2516 | page_count = (u32)calc_pages_for(offset, length); | |
2517 | obj_request->page_count = page_count; | |
2518 | if ((offset + length) & ~PAGE_MASK) | |
2519 | page_count--; /* more on last page */ | |
2520 | pages += page_count; | |
2521 | } | |
bf0d5f50 | 2522 | |
6d2940c8 GZ |
2523 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2524 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2525 | obj_request); | |
2fa12320 | 2526 | if (!osd_req) |
62054da6 | 2527 | goto out_unwind; |
3b434a2a | 2528 | |
2fa12320 | 2529 | obj_request->osd_req = osd_req; |
2169238d | 2530 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2531 | obj_request->img_offset = img_offset; |
9d4df01f | 2532 | |
3b434a2a | 2533 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2534 | |
3b434a2a | 2535 | rbd_img_request_get(img_request); |
bf0d5f50 | 2536 | |
7da22d29 | 2537 | img_offset += length; |
bf0d5f50 AE |
2538 | resid -= length; |
2539 | } | |
2540 | ||
2541 | return 0; | |
2542 | ||
bf0d5f50 AE |
2543 | out_unwind: |
2544 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2545 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2546 | |
2547 | return -ENOMEM; | |
2548 | } | |
2549 | ||
0eefd470 | 2550 | static void |
2761713d | 2551 | rbd_osd_copyup_callback(struct rbd_obj_request *obj_request) |
0eefd470 AE |
2552 | { |
2553 | struct rbd_img_request *img_request; | |
2554 | struct rbd_device *rbd_dev; | |
ebda6408 | 2555 | struct page **pages; |
0eefd470 AE |
2556 | u32 page_count; |
2557 | ||
2761713d ID |
2558 | dout("%s: obj %p\n", __func__, obj_request); |
2559 | ||
d3246fb0 JD |
2560 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2561 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2562 | rbd_assert(obj_request_img_data_test(obj_request)); |
2563 | img_request = obj_request->img_request; | |
2564 | rbd_assert(img_request); | |
2565 | ||
2566 | rbd_dev = img_request->rbd_dev; | |
2567 | rbd_assert(rbd_dev); | |
0eefd470 | 2568 | |
ebda6408 AE |
2569 | pages = obj_request->copyup_pages; |
2570 | rbd_assert(pages != NULL); | |
0eefd470 | 2571 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2572 | page_count = obj_request->copyup_page_count; |
2573 | rbd_assert(page_count); | |
2574 | obj_request->copyup_page_count = 0; | |
2575 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2576 | |
2577 | /* | |
2578 | * We want the transfer count to reflect the size of the | |
2579 | * original write request. There is no such thing as a | |
2580 | * successful short write, so if the request was successful | |
2581 | * we can just set it to the originally-requested length. | |
2582 | */ | |
2583 | if (!obj_request->result) | |
2584 | obj_request->xferred = obj_request->length; | |
2585 | ||
2761713d | 2586 | obj_request_done_set(obj_request); |
0eefd470 AE |
2587 | } |
2588 | ||
3d7efd18 AE |
2589 | static void |
2590 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2591 | { | |
2592 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2593 | struct ceph_osd_request *osd_req; |
2594 | struct ceph_osd_client *osdc; | |
2595 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2596 | struct page **pages; |
d3246fb0 | 2597 | enum obj_operation_type op_type; |
ebda6408 | 2598 | u32 page_count; |
bbea1c1a | 2599 | int img_result; |
ebda6408 | 2600 | u64 parent_length; |
3d7efd18 AE |
2601 | |
2602 | rbd_assert(img_request_child_test(img_request)); | |
2603 | ||
2604 | /* First get what we need from the image request */ | |
2605 | ||
2606 | pages = img_request->copyup_pages; | |
2607 | rbd_assert(pages != NULL); | |
2608 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2609 | page_count = img_request->copyup_page_count; |
2610 | rbd_assert(page_count); | |
2611 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2612 | |
2613 | orig_request = img_request->obj_request; | |
2614 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2615 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2616 | img_result = img_request->result; |
ebda6408 AE |
2617 | parent_length = img_request->length; |
2618 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2619 | rbd_img_request_put(img_request); |
3d7efd18 | 2620 | |
91c6febb AE |
2621 | rbd_assert(orig_request->img_request); |
2622 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2623 | rbd_assert(rbd_dev); |
0eefd470 | 2624 | |
bbea1c1a AE |
2625 | /* |
2626 | * If the overlap has become 0 (most likely because the | |
2627 | * image has been flattened) we need to free the pages | |
2628 | * and re-submit the original write request. | |
2629 | */ | |
2630 | if (!rbd_dev->parent_overlap) { | |
2631 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2632 | |
bbea1c1a AE |
2633 | ceph_release_page_vector(pages, page_count); |
2634 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2635 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2636 | if (!img_result) | |
2637 | return; | |
2638 | } | |
0eefd470 | 2639 | |
bbea1c1a | 2640 | if (img_result) |
0eefd470 | 2641 | goto out_err; |
0eefd470 | 2642 | |
8785b1d4 AE |
2643 | /* |
2644 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2645 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2646 | * request. Allocate the new copyup osd request for the |
2647 | * original request, and release the old one. | |
2648 | */ | |
bbea1c1a | 2649 | img_result = -ENOMEM; |
0eefd470 AE |
2650 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2651 | if (!osd_req) | |
2652 | goto out_err; | |
8785b1d4 | 2653 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2654 | orig_request->osd_req = osd_req; |
2655 | orig_request->copyup_pages = pages; | |
ebda6408 | 2656 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2657 | |
0eefd470 | 2658 | /* Initialize the copyup op */ |
3d7efd18 | 2659 | |
0eefd470 | 2660 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2661 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2662 | false, false); |
3d7efd18 | 2663 | |
d3246fb0 | 2664 | /* Add the other op(s) */ |
0eefd470 | 2665 | |
d3246fb0 JD |
2666 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2667 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2668 | |
2669 | /* All set, send it off. */ | |
2670 | ||
0eefd470 | 2671 | osdc = &rbd_dev->rbd_client->client->osdc; |
bbea1c1a AE |
2672 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2673 | if (!img_result) | |
0eefd470 AE |
2674 | return; |
2675 | out_err: | |
2676 | /* Record the error code and complete the request */ | |
2677 | ||
bbea1c1a | 2678 | orig_request->result = img_result; |
0eefd470 AE |
2679 | orig_request->xferred = 0; |
2680 | obj_request_done_set(orig_request); | |
2681 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2682 | } |
2683 | ||
2684 | /* | |
2685 | * Read from the parent image the range of data that covers the | |
2686 | * entire target of the given object request. This is used for | |
2687 | * satisfying a layered image write request when the target of an | |
2688 | * object request from the image request does not exist. | |
2689 | * | |
2690 | * A page array big enough to hold the returned data is allocated | |
2691 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2692 | * When the read completes, this page array will be transferred to | |
2693 | * the original object request for the copyup operation. | |
2694 | * | |
2695 | * If an error occurs, record it as the result of the original | |
2696 | * object request and mark it done so it gets completed. | |
2697 | */ | |
2698 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2699 | { | |
2700 | struct rbd_img_request *img_request = NULL; | |
2701 | struct rbd_img_request *parent_request = NULL; | |
2702 | struct rbd_device *rbd_dev; | |
2703 | u64 img_offset; | |
2704 | u64 length; | |
2705 | struct page **pages = NULL; | |
2706 | u32 page_count; | |
2707 | int result; | |
2708 | ||
2709 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2710 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2711 | |
2712 | img_request = obj_request->img_request; | |
2713 | rbd_assert(img_request != NULL); | |
2714 | rbd_dev = img_request->rbd_dev; | |
2715 | rbd_assert(rbd_dev->parent != NULL); | |
2716 | ||
2717 | /* | |
2718 | * Determine the byte range covered by the object in the | |
2719 | * child image to which the original request was to be sent. | |
2720 | */ | |
2721 | img_offset = obj_request->img_offset - obj_request->offset; | |
2722 | length = (u64)1 << rbd_dev->header.obj_order; | |
2723 | ||
a9e8ba2c AE |
2724 | /* |
2725 | * There is no defined parent data beyond the parent | |
2726 | * overlap, so limit what we read at that boundary if | |
2727 | * necessary. | |
2728 | */ | |
2729 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2730 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2731 | length = rbd_dev->parent_overlap - img_offset; | |
2732 | } | |
2733 | ||
3d7efd18 AE |
2734 | /* |
2735 | * Allocate a page array big enough to receive the data read | |
2736 | * from the parent. | |
2737 | */ | |
2738 | page_count = (u32)calc_pages_for(0, length); | |
2739 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2740 | if (IS_ERR(pages)) { | |
2741 | result = PTR_ERR(pages); | |
2742 | pages = NULL; | |
2743 | goto out_err; | |
2744 | } | |
2745 | ||
2746 | result = -ENOMEM; | |
e93f3152 AE |
2747 | parent_request = rbd_parent_request_create(obj_request, |
2748 | img_offset, length); | |
3d7efd18 AE |
2749 | if (!parent_request) |
2750 | goto out_err; | |
3d7efd18 AE |
2751 | |
2752 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2753 | if (result) | |
2754 | goto out_err; | |
2755 | parent_request->copyup_pages = pages; | |
ebda6408 | 2756 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2757 | |
2758 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2759 | result = rbd_img_request_submit(parent_request); | |
2760 | if (!result) | |
2761 | return 0; | |
2762 | ||
2763 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2764 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2765 | parent_request->obj_request = NULL; |
2766 | rbd_obj_request_put(obj_request); | |
2767 | out_err: | |
2768 | if (pages) | |
2769 | ceph_release_page_vector(pages, page_count); | |
2770 | if (parent_request) | |
2771 | rbd_img_request_put(parent_request); | |
2772 | obj_request->result = result; | |
2773 | obj_request->xferred = 0; | |
2774 | obj_request_done_set(obj_request); | |
2775 | ||
2776 | return result; | |
2777 | } | |
2778 | ||
c5b5ef6c AE |
2779 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2780 | { | |
c5b5ef6c | 2781 | struct rbd_obj_request *orig_request; |
638f5abe | 2782 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2783 | int result; |
2784 | ||
2785 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2786 | ||
2787 | /* | |
2788 | * All we need from the object request is the original | |
2789 | * request and the result of the STAT op. Grab those, then | |
2790 | * we're done with the request. | |
2791 | */ | |
2792 | orig_request = obj_request->obj_request; | |
2793 | obj_request->obj_request = NULL; | |
912c317d | 2794 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2795 | rbd_assert(orig_request); |
2796 | rbd_assert(orig_request->img_request); | |
2797 | ||
2798 | result = obj_request->result; | |
2799 | obj_request->result = 0; | |
2800 | ||
2801 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2802 | obj_request, orig_request, result, | |
2803 | obj_request->xferred, obj_request->length); | |
2804 | rbd_obj_request_put(obj_request); | |
2805 | ||
638f5abe AE |
2806 | /* |
2807 | * If the overlap has become 0 (most likely because the | |
2808 | * image has been flattened) we need to free the pages | |
2809 | * and re-submit the original write request. | |
2810 | */ | |
2811 | rbd_dev = orig_request->img_request->rbd_dev; | |
2812 | if (!rbd_dev->parent_overlap) { | |
2813 | struct ceph_osd_client *osdc; | |
2814 | ||
638f5abe AE |
2815 | osdc = &rbd_dev->rbd_client->client->osdc; |
2816 | result = rbd_obj_request_submit(osdc, orig_request); | |
2817 | if (!result) | |
2818 | return; | |
2819 | } | |
c5b5ef6c AE |
2820 | |
2821 | /* | |
2822 | * Our only purpose here is to determine whether the object | |
2823 | * exists, and we don't want to treat the non-existence as | |
2824 | * an error. If something else comes back, transfer the | |
2825 | * error to the original request and complete it now. | |
2826 | */ | |
2827 | if (!result) { | |
2828 | obj_request_existence_set(orig_request, true); | |
2829 | } else if (result == -ENOENT) { | |
2830 | obj_request_existence_set(orig_request, false); | |
2831 | } else if (result) { | |
2832 | orig_request->result = result; | |
3d7efd18 | 2833 | goto out; |
c5b5ef6c AE |
2834 | } |
2835 | ||
2836 | /* | |
2837 | * Resubmit the original request now that we have recorded | |
2838 | * whether the target object exists. | |
2839 | */ | |
b454e36d | 2840 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2841 | out: |
c5b5ef6c AE |
2842 | if (orig_request->result) |
2843 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2844 | } |
2845 | ||
2846 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2847 | { | |
2848 | struct rbd_obj_request *stat_request; | |
2849 | struct rbd_device *rbd_dev; | |
2850 | struct ceph_osd_client *osdc; | |
2851 | struct page **pages = NULL; | |
2852 | u32 page_count; | |
2853 | size_t size; | |
2854 | int ret; | |
2855 | ||
2856 | /* | |
2857 | * The response data for a STAT call consists of: | |
2858 | * le64 length; | |
2859 | * struct { | |
2860 | * le32 tv_sec; | |
2861 | * le32 tv_nsec; | |
2862 | * } mtime; | |
2863 | */ | |
2864 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2865 | page_count = (u32)calc_pages_for(0, size); | |
2866 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2867 | if (IS_ERR(pages)) | |
2868 | return PTR_ERR(pages); | |
2869 | ||
2870 | ret = -ENOMEM; | |
2871 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2872 | OBJ_REQUEST_PAGES); | |
2873 | if (!stat_request) | |
2874 | goto out; | |
2875 | ||
2876 | rbd_obj_request_get(obj_request); | |
2877 | stat_request->obj_request = obj_request; | |
2878 | stat_request->pages = pages; | |
2879 | stat_request->page_count = page_count; | |
2880 | ||
2881 | rbd_assert(obj_request->img_request); | |
2882 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2883 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2884 | stat_request); |
c5b5ef6c AE |
2885 | if (!stat_request->osd_req) |
2886 | goto out; | |
2887 | stat_request->callback = rbd_img_obj_exists_callback; | |
2888 | ||
144cba14 | 2889 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2890 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2891 | false, false); | |
9d4df01f | 2892 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2893 | |
2894 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2895 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2896 | out: | |
2897 | if (ret) | |
2898 | rbd_obj_request_put(obj_request); | |
2899 | ||
2900 | return ret; | |
2901 | } | |
2902 | ||
70d045f6 | 2903 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2904 | { |
2905 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2906 | struct rbd_device *rbd_dev; |
b454e36d AE |
2907 | |
2908 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2909 | ||
2910 | img_request = obj_request->img_request; | |
2911 | rbd_assert(img_request); | |
a9e8ba2c | 2912 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2913 | |
70d045f6 | 2914 | /* Reads */ |
1c220881 JD |
2915 | if (!img_request_write_test(img_request) && |
2916 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2917 | return true; |
2918 | ||
2919 | /* Non-layered writes */ | |
2920 | if (!img_request_layered_test(img_request)) | |
2921 | return true; | |
2922 | ||
b454e36d | 2923 | /* |
70d045f6 ID |
2924 | * Layered writes outside of the parent overlap range don't |
2925 | * share any data with the parent. | |
b454e36d | 2926 | */ |
70d045f6 ID |
2927 | if (!obj_request_overlaps_parent(obj_request)) |
2928 | return true; | |
b454e36d | 2929 | |
c622d226 GZ |
2930 | /* |
2931 | * Entire-object layered writes - we will overwrite whatever | |
2932 | * parent data there is anyway. | |
2933 | */ | |
2934 | if (!obj_request->offset && | |
2935 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2936 | return true; | |
2937 | ||
70d045f6 ID |
2938 | /* |
2939 | * If the object is known to already exist, its parent data has | |
2940 | * already been copied. | |
2941 | */ | |
2942 | if (obj_request_known_test(obj_request) && | |
2943 | obj_request_exists_test(obj_request)) | |
2944 | return true; | |
2945 | ||
2946 | return false; | |
2947 | } | |
2948 | ||
2949 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2950 | { | |
2951 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2952 | struct rbd_device *rbd_dev; |
2953 | struct ceph_osd_client *osdc; | |
2954 | ||
2955 | rbd_dev = obj_request->img_request->rbd_dev; | |
2956 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2957 | ||
2958 | return rbd_obj_request_submit(osdc, obj_request); | |
2959 | } | |
2960 | ||
2961 | /* | |
3d7efd18 AE |
2962 | * It's a layered write. The target object might exist but |
2963 | * we may not know that yet. If we know it doesn't exist, | |
2964 | * start by reading the data for the full target object from | |
2965 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2966 | */ |
70d045f6 | 2967 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2968 | return rbd_img_obj_parent_read_full(obj_request); |
2969 | ||
2970 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2971 | |
2972 | return rbd_img_obj_exists_submit(obj_request); | |
2973 | } | |
2974 | ||
bf0d5f50 AE |
2975 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2976 | { | |
bf0d5f50 | 2977 | struct rbd_obj_request *obj_request; |
46faeed4 | 2978 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2979 | |
37206ee5 | 2980 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2981 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2982 | int ret; |
2983 | ||
b454e36d | 2984 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2985 | if (ret) |
2986 | return ret; | |
bf0d5f50 AE |
2987 | } |
2988 | ||
2989 | return 0; | |
2990 | } | |
8b3e1a56 AE |
2991 | |
2992 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2993 | { | |
2994 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2995 | struct rbd_device *rbd_dev; |
2996 | u64 obj_end; | |
02c74fba AE |
2997 | u64 img_xferred; |
2998 | int img_result; | |
8b3e1a56 AE |
2999 | |
3000 | rbd_assert(img_request_child_test(img_request)); | |
3001 | ||
02c74fba AE |
3002 | /* First get what we need from the image request and release it */ |
3003 | ||
8b3e1a56 | 3004 | obj_request = img_request->obj_request; |
02c74fba AE |
3005 | img_xferred = img_request->xferred; |
3006 | img_result = img_request->result; | |
3007 | rbd_img_request_put(img_request); | |
3008 | ||
3009 | /* | |
3010 | * If the overlap has become 0 (most likely because the | |
3011 | * image has been flattened) we need to re-submit the | |
3012 | * original request. | |
3013 | */ | |
a9e8ba2c AE |
3014 | rbd_assert(obj_request); |
3015 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
3016 | rbd_dev = obj_request->img_request->rbd_dev; |
3017 | if (!rbd_dev->parent_overlap) { | |
3018 | struct ceph_osd_client *osdc; | |
3019 | ||
3020 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3021 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
3022 | if (!img_result) | |
3023 | return; | |
3024 | } | |
a9e8ba2c | 3025 | |
02c74fba | 3026 | obj_request->result = img_result; |
a9e8ba2c AE |
3027 | if (obj_request->result) |
3028 | goto out; | |
3029 | ||
3030 | /* | |
3031 | * We need to zero anything beyond the parent overlap | |
3032 | * boundary. Since rbd_img_obj_request_read_callback() | |
3033 | * will zero anything beyond the end of a short read, an | |
3034 | * easy way to do this is to pretend the data from the | |
3035 | * parent came up short--ending at the overlap boundary. | |
3036 | */ | |
3037 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3038 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3039 | if (obj_end > rbd_dev->parent_overlap) { |
3040 | u64 xferred = 0; | |
3041 | ||
3042 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3043 | xferred = rbd_dev->parent_overlap - | |
3044 | obj_request->img_offset; | |
8b3e1a56 | 3045 | |
02c74fba | 3046 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3047 | } else { |
02c74fba | 3048 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3049 | } |
3050 | out: | |
8b3e1a56 AE |
3051 | rbd_img_obj_request_read_callback(obj_request); |
3052 | rbd_obj_request_complete(obj_request); | |
3053 | } | |
3054 | ||
3055 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3056 | { | |
8b3e1a56 AE |
3057 | struct rbd_img_request *img_request; |
3058 | int result; | |
3059 | ||
3060 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3061 | rbd_assert(obj_request->img_request != NULL); | |
3062 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3063 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3064 | |
8b3e1a56 | 3065 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3066 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3067 | obj_request->img_offset, |
e93f3152 | 3068 | obj_request->length); |
8b3e1a56 AE |
3069 | result = -ENOMEM; |
3070 | if (!img_request) | |
3071 | goto out_err; | |
3072 | ||
5b2ab72d AE |
3073 | if (obj_request->type == OBJ_REQUEST_BIO) |
3074 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3075 | obj_request->bio_list); | |
3076 | else | |
3077 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3078 | obj_request->pages); | |
8b3e1a56 AE |
3079 | if (result) |
3080 | goto out_err; | |
3081 | ||
3082 | img_request->callback = rbd_img_parent_read_callback; | |
3083 | result = rbd_img_request_submit(img_request); | |
3084 | if (result) | |
3085 | goto out_err; | |
3086 | ||
3087 | return; | |
3088 | out_err: | |
3089 | if (img_request) | |
3090 | rbd_img_request_put(img_request); | |
3091 | obj_request->result = result; | |
3092 | obj_request->xferred = 0; | |
3093 | obj_request_done_set(obj_request); | |
3094 | } | |
bf0d5f50 | 3095 | |
20e0af67 | 3096 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3097 | { |
3098 | struct rbd_obj_request *obj_request; | |
2169238d | 3099 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3100 | int ret; |
3101 | ||
3102 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3103 | OBJ_REQUEST_NODATA); | |
3104 | if (!obj_request) | |
3105 | return -ENOMEM; | |
3106 | ||
3107 | ret = -ENOMEM; | |
6d2940c8 | 3108 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3109 | obj_request); |
b8d70035 AE |
3110 | if (!obj_request->osd_req) |
3111 | goto out; | |
3112 | ||
c99d2d4a | 3113 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3114 | notify_id, 0, 0); |
9d4df01f | 3115 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3116 | |
b8d70035 | 3117 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3118 | if (ret) |
20e0af67 JD |
3119 | goto out; |
3120 | ret = rbd_obj_request_wait(obj_request); | |
3121 | out: | |
3122 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3123 | |
3124 | return ret; | |
3125 | } | |
3126 | ||
3127 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3128 | { | |
3129 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3130 | int ret; |
b8d70035 AE |
3131 | |
3132 | if (!rbd_dev) | |
3133 | return; | |
3134 | ||
37206ee5 | 3135 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3136 | rbd_dev->header_name, (unsigned long long)notify_id, |
3137 | (unsigned int)opcode); | |
52bb1f9b ID |
3138 | |
3139 | /* | |
3140 | * Until adequate refresh error handling is in place, there is | |
3141 | * not much we can do here, except warn. | |
3142 | * | |
3143 | * See http://tracker.ceph.com/issues/5040 | |
3144 | */ | |
e627db08 AE |
3145 | ret = rbd_dev_refresh(rbd_dev); |
3146 | if (ret) | |
9584d508 | 3147 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3148 | |
52bb1f9b ID |
3149 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3150 | if (ret) | |
9584d508 | 3151 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3152 | } |
3153 | ||
bb040aa0 ID |
3154 | /* |
3155 | * Send a (un)watch request and wait for the ack. Return a request | |
3156 | * with a ref held on success or error. | |
3157 | */ | |
3158 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3159 | struct rbd_device *rbd_dev, | |
3160 | bool watch) | |
3161 | { | |
3162 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2894e1d7 | 3163 | struct ceph_options *opts = osdc->client->options; |
bb040aa0 ID |
3164 | struct rbd_obj_request *obj_request; |
3165 | int ret; | |
3166 | ||
3167 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3168 | OBJ_REQUEST_NODATA); | |
3169 | if (!obj_request) | |
3170 | return ERR_PTR(-ENOMEM); | |
3171 | ||
6d2940c8 | 3172 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3173 | obj_request); |
3174 | if (!obj_request->osd_req) { | |
3175 | ret = -ENOMEM; | |
3176 | goto out; | |
3177 | } | |
3178 | ||
3179 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3180 | rbd_dev->watch_event->cookie, 0, watch); | |
3181 | rbd_osd_req_format_write(obj_request); | |
3182 | ||
3183 | if (watch) | |
3184 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3185 | ||
3186 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3187 | if (ret) | |
3188 | goto out; | |
3189 | ||
2894e1d7 | 3190 | ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout); |
bb040aa0 ID |
3191 | if (ret) |
3192 | goto out; | |
3193 | ||
3194 | ret = obj_request->result; | |
3195 | if (ret) { | |
3196 | if (watch) | |
3197 | rbd_obj_request_end(obj_request); | |
3198 | goto out; | |
3199 | } | |
3200 | ||
3201 | return obj_request; | |
3202 | ||
3203 | out: | |
3204 | rbd_obj_request_put(obj_request); | |
3205 | return ERR_PTR(ret); | |
3206 | } | |
3207 | ||
9969ebc5 | 3208 | /* |
b30a01f2 | 3209 | * Initiate a watch request, synchronously. |
9969ebc5 | 3210 | */ |
b30a01f2 | 3211 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3212 | { |
3213 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3214 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3215 | int ret; |
3216 | ||
b30a01f2 ID |
3217 | rbd_assert(!rbd_dev->watch_event); |
3218 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3219 | |
b30a01f2 ID |
3220 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3221 | &rbd_dev->watch_event); | |
3222 | if (ret < 0) | |
3223 | return ret; | |
3224 | ||
76756a51 ID |
3225 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3226 | if (IS_ERR(obj_request)) { | |
3227 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3228 | rbd_dev->watch_event = NULL; | |
3229 | return PTR_ERR(obj_request); | |
b30a01f2 | 3230 | } |
9969ebc5 | 3231 | |
8eb87565 AE |
3232 | /* |
3233 | * A watch request is set to linger, so the underlying osd | |
3234 | * request won't go away until we unregister it. We retain | |
3235 | * a pointer to the object request during that time (in | |
76756a51 ID |
3236 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3237 | * We'll drop that reference after we've unregistered it in | |
3238 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3239 | */ |
b30a01f2 | 3240 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3241 | |
b30a01f2 | 3242 | return 0; |
b30a01f2 ID |
3243 | } |
3244 | ||
3245 | /* | |
3246 | * Tear down a watch request, synchronously. | |
3247 | */ | |
76756a51 | 3248 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3249 | { |
b30a01f2 | 3250 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3251 | |
3252 | rbd_assert(rbd_dev->watch_event); | |
3253 | rbd_assert(rbd_dev->watch_request); | |
3254 | ||
76756a51 | 3255 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3256 | rbd_obj_request_put(rbd_dev->watch_request); |
3257 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3258 | |
76756a51 ID |
3259 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3260 | if (!IS_ERR(obj_request)) | |
3261 | rbd_obj_request_put(obj_request); | |
3262 | else | |
3263 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3264 | PTR_ERR(obj_request)); | |
3265 | ||
9969ebc5 AE |
3266 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3267 | rbd_dev->watch_event = NULL; | |
fca27065 ID |
3268 | } |
3269 | ||
36be9a76 | 3270 | /* |
f40eb349 AE |
3271 | * Synchronous osd object method call. Returns the number of bytes |
3272 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3273 | */ |
3274 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3275 | const char *object_name, | |
3276 | const char *class_name, | |
3277 | const char *method_name, | |
4157976b | 3278 | const void *outbound, |
36be9a76 | 3279 | size_t outbound_size, |
4157976b | 3280 | void *inbound, |
e2a58ee5 | 3281 | size_t inbound_size) |
36be9a76 | 3282 | { |
2169238d | 3283 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3284 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3285 | struct page **pages; |
3286 | u32 page_count; | |
3287 | int ret; | |
3288 | ||
3289 | /* | |
6010a451 AE |
3290 | * Method calls are ultimately read operations. The result |
3291 | * should placed into the inbound buffer provided. They | |
3292 | * also supply outbound data--parameters for the object | |
3293 | * method. Currently if this is present it will be a | |
3294 | * snapshot id. | |
36be9a76 | 3295 | */ |
57385b51 | 3296 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3297 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3298 | if (IS_ERR(pages)) | |
3299 | return PTR_ERR(pages); | |
3300 | ||
3301 | ret = -ENOMEM; | |
6010a451 | 3302 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3303 | OBJ_REQUEST_PAGES); |
3304 | if (!obj_request) | |
3305 | goto out; | |
3306 | ||
3307 | obj_request->pages = pages; | |
3308 | obj_request->page_count = page_count; | |
3309 | ||
6d2940c8 | 3310 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3311 | obj_request); |
36be9a76 AE |
3312 | if (!obj_request->osd_req) |
3313 | goto out; | |
3314 | ||
c99d2d4a | 3315 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3316 | class_name, method_name); |
3317 | if (outbound_size) { | |
3318 | struct ceph_pagelist *pagelist; | |
3319 | ||
3320 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3321 | if (!pagelist) | |
3322 | goto out; | |
3323 | ||
3324 | ceph_pagelist_init(pagelist); | |
3325 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3326 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3327 | pagelist); | |
3328 | } | |
a4ce40a9 AE |
3329 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3330 | obj_request->pages, inbound_size, | |
44cd188d | 3331 | 0, false, false); |
9d4df01f | 3332 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3333 | |
36be9a76 AE |
3334 | ret = rbd_obj_request_submit(osdc, obj_request); |
3335 | if (ret) | |
3336 | goto out; | |
3337 | ret = rbd_obj_request_wait(obj_request); | |
3338 | if (ret) | |
3339 | goto out; | |
3340 | ||
3341 | ret = obj_request->result; | |
3342 | if (ret < 0) | |
3343 | goto out; | |
57385b51 AE |
3344 | |
3345 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3346 | ret = (int)obj_request->xferred; | |
903bb32e | 3347 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3348 | out: |
3349 | if (obj_request) | |
3350 | rbd_obj_request_put(obj_request); | |
3351 | else | |
3352 | ceph_release_page_vector(pages, page_count); | |
3353 | ||
3354 | return ret; | |
3355 | } | |
3356 | ||
7ad18afa | 3357 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3358 | { |
7ad18afa CH |
3359 | struct request *rq = blk_mq_rq_from_pdu(work); |
3360 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3361 | struct rbd_img_request *img_request; |
4e752f0a | 3362 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3363 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3364 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3365 | enum obj_operation_type op_type; |
4e752f0a | 3366 | u64 mapping_size; |
bf0d5f50 AE |
3367 | int result; |
3368 | ||
7ad18afa CH |
3369 | if (rq->cmd_type != REQ_TYPE_FS) { |
3370 | dout("%s: non-fs request type %d\n", __func__, | |
3371 | (int) rq->cmd_type); | |
3372 | result = -EIO; | |
3373 | goto err; | |
3374 | } | |
3375 | ||
90e98c52 GZ |
3376 | if (rq->cmd_flags & REQ_DISCARD) |
3377 | op_type = OBJ_OP_DISCARD; | |
3378 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3379 | op_type = OBJ_OP_WRITE; |
3380 | else | |
3381 | op_type = OBJ_OP_READ; | |
3382 | ||
bc1ecc65 | 3383 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3384 | |
bc1ecc65 ID |
3385 | if (!length) { |
3386 | dout("%s: zero-length request\n", __func__); | |
3387 | result = 0; | |
3388 | goto err_rq; | |
3389 | } | |
bf0d5f50 | 3390 | |
6d2940c8 | 3391 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3392 | |
6d2940c8 | 3393 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3394 | if (rbd_dev->mapping.read_only) { |
3395 | result = -EROFS; | |
3396 | goto err_rq; | |
4dda41d3 | 3397 | } |
bc1ecc65 ID |
3398 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3399 | } | |
4dda41d3 | 3400 | |
bc1ecc65 ID |
3401 | /* |
3402 | * Quit early if the mapped snapshot no longer exists. It's | |
3403 | * still possible the snapshot will have disappeared by the | |
3404 | * time our request arrives at the osd, but there's no sense in | |
3405 | * sending it if we already know. | |
3406 | */ | |
3407 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3408 | dout("request for non-existent snapshot"); | |
3409 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3410 | result = -ENXIO; | |
3411 | goto err_rq; | |
3412 | } | |
4dda41d3 | 3413 | |
bc1ecc65 ID |
3414 | if (offset && length > U64_MAX - offset + 1) { |
3415 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3416 | length); | |
3417 | result = -EINVAL; | |
3418 | goto err_rq; /* Shouldn't happen */ | |
3419 | } | |
4dda41d3 | 3420 | |
7ad18afa CH |
3421 | blk_mq_start_request(rq); |
3422 | ||
4e752f0a JD |
3423 | down_read(&rbd_dev->header_rwsem); |
3424 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3425 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3426 | snapc = rbd_dev->header.snapc; |
3427 | ceph_get_snap_context(snapc); | |
3428 | } | |
3429 | up_read(&rbd_dev->header_rwsem); | |
3430 | ||
3431 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3432 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3433 | length, mapping_size); |
bc1ecc65 ID |
3434 | result = -EIO; |
3435 | goto err_rq; | |
3436 | } | |
bf0d5f50 | 3437 | |
6d2940c8 | 3438 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3439 | snapc); |
bc1ecc65 ID |
3440 | if (!img_request) { |
3441 | result = -ENOMEM; | |
3442 | goto err_rq; | |
3443 | } | |
3444 | img_request->rq = rq; | |
bf0d5f50 | 3445 | |
90e98c52 GZ |
3446 | if (op_type == OBJ_OP_DISCARD) |
3447 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3448 | NULL); | |
3449 | else | |
3450 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3451 | rq->bio); | |
bc1ecc65 ID |
3452 | if (result) |
3453 | goto err_img_request; | |
bf0d5f50 | 3454 | |
bc1ecc65 ID |
3455 | result = rbd_img_request_submit(img_request); |
3456 | if (result) | |
3457 | goto err_img_request; | |
bf0d5f50 | 3458 | |
bc1ecc65 | 3459 | return; |
bf0d5f50 | 3460 | |
bc1ecc65 ID |
3461 | err_img_request: |
3462 | rbd_img_request_put(img_request); | |
3463 | err_rq: | |
3464 | if (result) | |
3465 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3466 | obj_op_name(op_type), length, offset, result); |
e96a650a | 3467 | ceph_put_snap_context(snapc); |
7ad18afa CH |
3468 | err: |
3469 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 3470 | } |
bf0d5f50 | 3471 | |
7ad18afa CH |
3472 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
3473 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 3474 | { |
7ad18afa CH |
3475 | struct request *rq = bd->rq; |
3476 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 3477 | |
7ad18afa CH |
3478 | queue_work(rbd_wq, work); |
3479 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
3480 | } |
3481 | ||
602adf40 YS |
3482 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
3483 | { | |
3484 | struct gendisk *disk = rbd_dev->disk; | |
3485 | ||
3486 | if (!disk) | |
3487 | return; | |
3488 | ||
a0cab924 AE |
3489 | rbd_dev->disk = NULL; |
3490 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3491 | del_gendisk(disk); |
a0cab924 AE |
3492 | if (disk->queue) |
3493 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 3494 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 3495 | } |
602adf40 YS |
3496 | put_disk(disk); |
3497 | } | |
3498 | ||
788e2df3 AE |
3499 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3500 | const char *object_name, | |
7097f8df | 3501 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3502 | |
3503 | { | |
2169238d | 3504 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3505 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3506 | struct page **pages = NULL; |
3507 | u32 page_count; | |
1ceae7ef | 3508 | size_t size; |
788e2df3 AE |
3509 | int ret; |
3510 | ||
3511 | page_count = (u32) calc_pages_for(offset, length); | |
3512 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3513 | if (IS_ERR(pages)) | |
a8d42056 | 3514 | return PTR_ERR(pages); |
788e2df3 AE |
3515 | |
3516 | ret = -ENOMEM; | |
3517 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3518 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3519 | if (!obj_request) |
3520 | goto out; | |
3521 | ||
3522 | obj_request->pages = pages; | |
3523 | obj_request->page_count = page_count; | |
3524 | ||
6d2940c8 | 3525 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3526 | obj_request); |
788e2df3 AE |
3527 | if (!obj_request->osd_req) |
3528 | goto out; | |
3529 | ||
c99d2d4a AE |
3530 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3531 | offset, length, 0, 0); | |
406e2c9f | 3532 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3533 | obj_request->pages, |
44cd188d AE |
3534 | obj_request->length, |
3535 | obj_request->offset & ~PAGE_MASK, | |
3536 | false, false); | |
9d4df01f | 3537 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3538 | |
788e2df3 AE |
3539 | ret = rbd_obj_request_submit(osdc, obj_request); |
3540 | if (ret) | |
3541 | goto out; | |
3542 | ret = rbd_obj_request_wait(obj_request); | |
3543 | if (ret) | |
3544 | goto out; | |
3545 | ||
3546 | ret = obj_request->result; | |
3547 | if (ret < 0) | |
3548 | goto out; | |
1ceae7ef AE |
3549 | |
3550 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3551 | size = (size_t) obj_request->xferred; | |
903bb32e | 3552 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3553 | rbd_assert(size <= (size_t)INT_MAX); |
3554 | ret = (int)size; | |
788e2df3 AE |
3555 | out: |
3556 | if (obj_request) | |
3557 | rbd_obj_request_put(obj_request); | |
3558 | else | |
3559 | ceph_release_page_vector(pages, page_count); | |
3560 | ||
3561 | return ret; | |
3562 | } | |
3563 | ||
602adf40 | 3564 | /* |
662518b1 AE |
3565 | * Read the complete header for the given rbd device. On successful |
3566 | * return, the rbd_dev->header field will contain up-to-date | |
3567 | * information about the image. | |
602adf40 | 3568 | */ |
99a41ebc | 3569 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3570 | { |
4156d998 | 3571 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3572 | u32 snap_count = 0; |
4156d998 AE |
3573 | u64 names_size = 0; |
3574 | u32 want_count; | |
3575 | int ret; | |
602adf40 | 3576 | |
00f1f36f | 3577 | /* |
4156d998 AE |
3578 | * The complete header will include an array of its 64-bit |
3579 | * snapshot ids, followed by the names of those snapshots as | |
3580 | * a contiguous block of NUL-terminated strings. Note that | |
3581 | * the number of snapshots could change by the time we read | |
3582 | * it in, in which case we re-read it. | |
00f1f36f | 3583 | */ |
4156d998 AE |
3584 | do { |
3585 | size_t size; | |
3586 | ||
3587 | kfree(ondisk); | |
3588 | ||
3589 | size = sizeof (*ondisk); | |
3590 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3591 | size += names_size; | |
3592 | ondisk = kmalloc(size, GFP_KERNEL); | |
3593 | if (!ondisk) | |
662518b1 | 3594 | return -ENOMEM; |
4156d998 | 3595 | |
788e2df3 | 3596 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3597 | 0, size, ondisk); |
4156d998 | 3598 | if (ret < 0) |
662518b1 | 3599 | goto out; |
c0cd10db | 3600 | if ((size_t)ret < size) { |
4156d998 | 3601 | ret = -ENXIO; |
06ecc6cb AE |
3602 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3603 | size, ret); | |
662518b1 | 3604 | goto out; |
4156d998 AE |
3605 | } |
3606 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3607 | ret = -ENXIO; | |
06ecc6cb | 3608 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3609 | goto out; |
81e759fb | 3610 | } |
602adf40 | 3611 | |
4156d998 AE |
3612 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3613 | want_count = snap_count; | |
3614 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3615 | } while (snap_count != want_count); | |
00f1f36f | 3616 | |
662518b1 AE |
3617 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3618 | out: | |
4156d998 AE |
3619 | kfree(ondisk); |
3620 | ||
3621 | return ret; | |
602adf40 YS |
3622 | } |
3623 | ||
15228ede AE |
3624 | /* |
3625 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3626 | * has disappeared from the (just updated) snapshot context. | |
3627 | */ | |
3628 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3629 | { | |
3630 | u64 snap_id; | |
3631 | ||
3632 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3633 | return; | |
3634 | ||
3635 | snap_id = rbd_dev->spec->snap_id; | |
3636 | if (snap_id == CEPH_NOSNAP) | |
3637 | return; | |
3638 | ||
3639 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3640 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3641 | } | |
3642 | ||
9875201e JD |
3643 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3644 | { | |
3645 | sector_t size; | |
3646 | bool removing; | |
3647 | ||
3648 | /* | |
3649 | * Don't hold the lock while doing disk operations, | |
3650 | * or lock ordering will conflict with the bdev mutex via: | |
3651 | * rbd_add() -> blkdev_get() -> rbd_open() | |
3652 | */ | |
3653 | spin_lock_irq(&rbd_dev->lock); | |
3654 | removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags); | |
3655 | spin_unlock_irq(&rbd_dev->lock); | |
3656 | /* | |
3657 | * If the device is being removed, rbd_dev->disk has | |
3658 | * been destroyed, so don't try to update its size | |
3659 | */ | |
3660 | if (!removing) { | |
3661 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; | |
3662 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3663 | set_capacity(rbd_dev->disk, size); | |
3664 | revalidate_disk(rbd_dev->disk); | |
3665 | } | |
3666 | } | |
3667 | ||
cc4a38bd | 3668 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3669 | { |
e627db08 | 3670 | u64 mapping_size; |
1fe5e993 AE |
3671 | int ret; |
3672 | ||
cfbf6377 | 3673 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3674 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3675 | |
3676 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 3677 | if (ret) |
73e39e4d | 3678 | goto out; |
15228ede | 3679 | |
e8f59b59 ID |
3680 | /* |
3681 | * If there is a parent, see if it has disappeared due to the | |
3682 | * mapped image getting flattened. | |
3683 | */ | |
3684 | if (rbd_dev->parent) { | |
3685 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3686 | if (ret) | |
73e39e4d | 3687 | goto out; |
e8f59b59 ID |
3688 | } |
3689 | ||
5ff1108c | 3690 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 3691 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
3692 | } else { |
3693 | /* validate mapped snapshot's EXISTS flag */ | |
3694 | rbd_exists_validate(rbd_dev); | |
3695 | } | |
15228ede | 3696 | |
73e39e4d | 3697 | out: |
cfbf6377 | 3698 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 3699 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 3700 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3701 | |
73e39e4d | 3702 | return ret; |
1fe5e993 AE |
3703 | } |
3704 | ||
7ad18afa CH |
3705 | static int rbd_init_request(void *data, struct request *rq, |
3706 | unsigned int hctx_idx, unsigned int request_idx, | |
3707 | unsigned int numa_node) | |
3708 | { | |
3709 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
3710 | ||
3711 | INIT_WORK(work, rbd_queue_workfn); | |
3712 | return 0; | |
3713 | } | |
3714 | ||
3715 | static struct blk_mq_ops rbd_mq_ops = { | |
3716 | .queue_rq = rbd_queue_rq, | |
3717 | .map_queue = blk_mq_map_queue, | |
3718 | .init_request = rbd_init_request, | |
3719 | }; | |
3720 | ||
602adf40 YS |
3721 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3722 | { | |
3723 | struct gendisk *disk; | |
3724 | struct request_queue *q; | |
593a9e7b | 3725 | u64 segment_size; |
7ad18afa | 3726 | int err; |
602adf40 | 3727 | |
602adf40 | 3728 | /* create gendisk info */ |
7e513d43 ID |
3729 | disk = alloc_disk(single_major ? |
3730 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3731 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3732 | if (!disk) |
1fcdb8aa | 3733 | return -ENOMEM; |
602adf40 | 3734 | |
f0f8cef5 | 3735 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3736 | rbd_dev->dev_id); |
602adf40 | 3737 | disk->major = rbd_dev->major; |
dd82fff1 | 3738 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3739 | if (single_major) |
3740 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3741 | disk->fops = &rbd_bd_ops; |
3742 | disk->private_data = rbd_dev; | |
3743 | ||
7ad18afa CH |
3744 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
3745 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 3746 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 3747 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 3748 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
3749 | rbd_dev->tag_set.nr_hw_queues = 1; |
3750 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
3751 | ||
3752 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
3753 | if (err) | |
602adf40 | 3754 | goto out_disk; |
029bcbd8 | 3755 | |
7ad18afa CH |
3756 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
3757 | if (IS_ERR(q)) { | |
3758 | err = PTR_ERR(q); | |
3759 | goto out_tag_set; | |
3760 | } | |
3761 | ||
d8a2c89c ID |
3762 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
3763 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 3764 | |
029bcbd8 | 3765 | /* set io sizes to object size */ |
593a9e7b AE |
3766 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3767 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
0d9fde4f | 3768 | q->limits.max_sectors = queue_max_hw_sectors(q); |
d3834fef | 3769 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
3770 | blk_queue_max_segment_size(q, segment_size); |
3771 | blk_queue_io_min(q, segment_size); | |
3772 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3773 | |
90e98c52 GZ |
3774 | /* enable the discard support */ |
3775 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3776 | q->limits.discard_granularity = segment_size; | |
3777 | q->limits.discard_alignment = segment_size; | |
2bb4cd5c | 3778 | blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE); |
b76f8239 | 3779 | q->limits.discard_zeroes_data = 1; |
90e98c52 | 3780 | |
bae818ee RH |
3781 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
3782 | q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES; | |
3783 | ||
602adf40 YS |
3784 | disk->queue = q; |
3785 | ||
3786 | q->queuedata = rbd_dev; | |
3787 | ||
3788 | rbd_dev->disk = disk; | |
602adf40 | 3789 | |
602adf40 | 3790 | return 0; |
7ad18afa CH |
3791 | out_tag_set: |
3792 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
3793 | out_disk: |
3794 | put_disk(disk); | |
7ad18afa | 3795 | return err; |
602adf40 YS |
3796 | } |
3797 | ||
dfc5606d YS |
3798 | /* |
3799 | sysfs | |
3800 | */ | |
3801 | ||
593a9e7b AE |
3802 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3803 | { | |
3804 | return container_of(dev, struct rbd_device, dev); | |
3805 | } | |
3806 | ||
dfc5606d YS |
3807 | static ssize_t rbd_size_show(struct device *dev, |
3808 | struct device_attribute *attr, char *buf) | |
3809 | { | |
593a9e7b | 3810 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3811 | |
fc71d833 AE |
3812 | return sprintf(buf, "%llu\n", |
3813 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3814 | } |
3815 | ||
34b13184 AE |
3816 | /* |
3817 | * Note this shows the features for whatever's mapped, which is not | |
3818 | * necessarily the base image. | |
3819 | */ | |
3820 | static ssize_t rbd_features_show(struct device *dev, | |
3821 | struct device_attribute *attr, char *buf) | |
3822 | { | |
3823 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3824 | ||
3825 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3826 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3827 | } |
3828 | ||
dfc5606d YS |
3829 | static ssize_t rbd_major_show(struct device *dev, |
3830 | struct device_attribute *attr, char *buf) | |
3831 | { | |
593a9e7b | 3832 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3833 | |
fc71d833 AE |
3834 | if (rbd_dev->major) |
3835 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3836 | ||
3837 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3838 | } |
3839 | ||
3840 | static ssize_t rbd_minor_show(struct device *dev, | |
3841 | struct device_attribute *attr, char *buf) | |
3842 | { | |
3843 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3844 | |
dd82fff1 | 3845 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3846 | } |
3847 | ||
3848 | static ssize_t rbd_client_id_show(struct device *dev, | |
3849 | struct device_attribute *attr, char *buf) | |
602adf40 | 3850 | { |
593a9e7b | 3851 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3852 | |
1dbb4399 AE |
3853 | return sprintf(buf, "client%lld\n", |
3854 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3855 | } |
3856 | ||
dfc5606d YS |
3857 | static ssize_t rbd_pool_show(struct device *dev, |
3858 | struct device_attribute *attr, char *buf) | |
602adf40 | 3859 | { |
593a9e7b | 3860 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3861 | |
0d7dbfce | 3862 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3863 | } |
3864 | ||
9bb2f334 AE |
3865 | static ssize_t rbd_pool_id_show(struct device *dev, |
3866 | struct device_attribute *attr, char *buf) | |
3867 | { | |
3868 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3869 | ||
0d7dbfce | 3870 | return sprintf(buf, "%llu\n", |
fc71d833 | 3871 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3872 | } |
3873 | ||
dfc5606d YS |
3874 | static ssize_t rbd_name_show(struct device *dev, |
3875 | struct device_attribute *attr, char *buf) | |
3876 | { | |
593a9e7b | 3877 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3878 | |
a92ffdf8 AE |
3879 | if (rbd_dev->spec->image_name) |
3880 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3881 | ||
3882 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3883 | } |
3884 | ||
589d30e0 AE |
3885 | static ssize_t rbd_image_id_show(struct device *dev, |
3886 | struct device_attribute *attr, char *buf) | |
3887 | { | |
3888 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3889 | ||
0d7dbfce | 3890 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3891 | } |
3892 | ||
34b13184 AE |
3893 | /* |
3894 | * Shows the name of the currently-mapped snapshot (or | |
3895 | * RBD_SNAP_HEAD_NAME for the base image). | |
3896 | */ | |
dfc5606d YS |
3897 | static ssize_t rbd_snap_show(struct device *dev, |
3898 | struct device_attribute *attr, | |
3899 | char *buf) | |
3900 | { | |
593a9e7b | 3901 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3902 | |
0d7dbfce | 3903 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3904 | } |
3905 | ||
86b00e0d | 3906 | /* |
ff96128f ID |
3907 | * For a v2 image, shows the chain of parent images, separated by empty |
3908 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3909 | * image)". | |
86b00e0d AE |
3910 | */ |
3911 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3912 | struct device_attribute *attr, |
3913 | char *buf) | |
86b00e0d AE |
3914 | { |
3915 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3916 | ssize_t count = 0; |
86b00e0d | 3917 | |
ff96128f | 3918 | if (!rbd_dev->parent) |
86b00e0d AE |
3919 | return sprintf(buf, "(no parent image)\n"); |
3920 | ||
ff96128f ID |
3921 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3922 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3923 | ||
3924 | count += sprintf(&buf[count], "%s" | |
3925 | "pool_id %llu\npool_name %s\n" | |
3926 | "image_id %s\nimage_name %s\n" | |
3927 | "snap_id %llu\nsnap_name %s\n" | |
3928 | "overlap %llu\n", | |
3929 | !count ? "" : "\n", /* first? */ | |
3930 | spec->pool_id, spec->pool_name, | |
3931 | spec->image_id, spec->image_name ?: "(unknown)", | |
3932 | spec->snap_id, spec->snap_name, | |
3933 | rbd_dev->parent_overlap); | |
3934 | } | |
3935 | ||
3936 | return count; | |
86b00e0d AE |
3937 | } |
3938 | ||
dfc5606d YS |
3939 | static ssize_t rbd_image_refresh(struct device *dev, |
3940 | struct device_attribute *attr, | |
3941 | const char *buf, | |
3942 | size_t size) | |
3943 | { | |
593a9e7b | 3944 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3945 | int ret; |
602adf40 | 3946 | |
cc4a38bd | 3947 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3948 | if (ret) |
52bb1f9b | 3949 | return ret; |
b813623a | 3950 | |
52bb1f9b | 3951 | return size; |
dfc5606d | 3952 | } |
602adf40 | 3953 | |
dfc5606d | 3954 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3955 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3956 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3957 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3958 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3959 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3960 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3961 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3962 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3963 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3964 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3965 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3966 | |
3967 | static struct attribute *rbd_attrs[] = { | |
3968 | &dev_attr_size.attr, | |
34b13184 | 3969 | &dev_attr_features.attr, |
dfc5606d | 3970 | &dev_attr_major.attr, |
dd82fff1 | 3971 | &dev_attr_minor.attr, |
dfc5606d YS |
3972 | &dev_attr_client_id.attr, |
3973 | &dev_attr_pool.attr, | |
9bb2f334 | 3974 | &dev_attr_pool_id.attr, |
dfc5606d | 3975 | &dev_attr_name.attr, |
589d30e0 | 3976 | &dev_attr_image_id.attr, |
dfc5606d | 3977 | &dev_attr_current_snap.attr, |
86b00e0d | 3978 | &dev_attr_parent.attr, |
dfc5606d | 3979 | &dev_attr_refresh.attr, |
dfc5606d YS |
3980 | NULL |
3981 | }; | |
3982 | ||
3983 | static struct attribute_group rbd_attr_group = { | |
3984 | .attrs = rbd_attrs, | |
3985 | }; | |
3986 | ||
3987 | static const struct attribute_group *rbd_attr_groups[] = { | |
3988 | &rbd_attr_group, | |
3989 | NULL | |
3990 | }; | |
3991 | ||
6cac4695 | 3992 | static void rbd_dev_release(struct device *dev); |
dfc5606d YS |
3993 | |
3994 | static struct device_type rbd_device_type = { | |
3995 | .name = "rbd", | |
3996 | .groups = rbd_attr_groups, | |
6cac4695 | 3997 | .release = rbd_dev_release, |
dfc5606d YS |
3998 | }; |
3999 | ||
8b8fb99c AE |
4000 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
4001 | { | |
4002 | kref_get(&spec->kref); | |
4003 | ||
4004 | return spec; | |
4005 | } | |
4006 | ||
4007 | static void rbd_spec_free(struct kref *kref); | |
4008 | static void rbd_spec_put(struct rbd_spec *spec) | |
4009 | { | |
4010 | if (spec) | |
4011 | kref_put(&spec->kref, rbd_spec_free); | |
4012 | } | |
4013 | ||
4014 | static struct rbd_spec *rbd_spec_alloc(void) | |
4015 | { | |
4016 | struct rbd_spec *spec; | |
4017 | ||
4018 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4019 | if (!spec) | |
4020 | return NULL; | |
04077599 ID |
4021 | |
4022 | spec->pool_id = CEPH_NOPOOL; | |
4023 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4024 | kref_init(&spec->kref); |
4025 | ||
8b8fb99c AE |
4026 | return spec; |
4027 | } | |
4028 | ||
4029 | static void rbd_spec_free(struct kref *kref) | |
4030 | { | |
4031 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4032 | ||
4033 | kfree(spec->pool_name); | |
4034 | kfree(spec->image_id); | |
4035 | kfree(spec->image_name); | |
4036 | kfree(spec->snap_name); | |
4037 | kfree(spec); | |
4038 | } | |
4039 | ||
dd5ac32d ID |
4040 | static void rbd_dev_release(struct device *dev) |
4041 | { | |
4042 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4043 | bool need_put = !!rbd_dev->opts; | |
4044 | ||
4045 | rbd_put_client(rbd_dev->rbd_client); | |
4046 | rbd_spec_put(rbd_dev->spec); | |
4047 | kfree(rbd_dev->opts); | |
4048 | kfree(rbd_dev); | |
4049 | ||
4050 | /* | |
4051 | * This is racy, but way better than putting module outside of | |
4052 | * the release callback. The race window is pretty small, so | |
4053 | * doing something similar to dm (dm-builtin.c) is overkill. | |
4054 | */ | |
4055 | if (need_put) | |
4056 | module_put(THIS_MODULE); | |
4057 | } | |
4058 | ||
cc344fa1 | 4059 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
d147543d ID |
4060 | struct rbd_spec *spec, |
4061 | struct rbd_options *opts) | |
c53d5893 AE |
4062 | { |
4063 | struct rbd_device *rbd_dev; | |
4064 | ||
4065 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4066 | if (!rbd_dev) | |
4067 | return NULL; | |
4068 | ||
4069 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 4070 | rbd_dev->flags = 0; |
a2acd00e | 4071 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4072 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4073 | init_rwsem(&rbd_dev->header_rwsem); |
4074 | ||
dd5ac32d ID |
4075 | rbd_dev->dev.bus = &rbd_bus_type; |
4076 | rbd_dev->dev.type = &rbd_device_type; | |
4077 | rbd_dev->dev.parent = &rbd_root_dev; | |
dd5ac32d ID |
4078 | device_initialize(&rbd_dev->dev); |
4079 | ||
c53d5893 | 4080 | rbd_dev->rbd_client = rbdc; |
d147543d ID |
4081 | rbd_dev->spec = spec; |
4082 | rbd_dev->opts = opts; | |
c53d5893 | 4083 | |
0903e875 AE |
4084 | /* Initialize the layout used for all rbd requests */ |
4085 | ||
4086 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4087 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4088 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4089 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4090 | ||
dd5ac32d ID |
4091 | /* |
4092 | * If this is a mapping rbd_dev (as opposed to a parent one), | |
4093 | * pin our module. We have a ref from do_rbd_add(), so use | |
4094 | * __module_get(). | |
4095 | */ | |
4096 | if (rbd_dev->opts) | |
4097 | __module_get(THIS_MODULE); | |
4098 | ||
c53d5893 AE |
4099 | return rbd_dev; |
4100 | } | |
4101 | ||
4102 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4103 | { | |
dd5ac32d ID |
4104 | if (rbd_dev) |
4105 | put_device(&rbd_dev->dev); | |
c53d5893 AE |
4106 | } |
4107 | ||
9d475de5 AE |
4108 | /* |
4109 | * Get the size and object order for an image snapshot, or if | |
4110 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4111 | * image. | |
4112 | */ | |
4113 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4114 | u8 *order, u64 *snap_size) | |
4115 | { | |
4116 | __le64 snapid = cpu_to_le64(snap_id); | |
4117 | int ret; | |
4118 | struct { | |
4119 | u8 order; | |
4120 | __le64 size; | |
4121 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4122 | ||
36be9a76 | 4123 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4124 | "rbd", "get_size", |
4157976b | 4125 | &snapid, sizeof (snapid), |
e2a58ee5 | 4126 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4127 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4128 | if (ret < 0) |
4129 | return ret; | |
57385b51 AE |
4130 | if (ret < sizeof (size_buf)) |
4131 | return -ERANGE; | |
9d475de5 | 4132 | |
c3545579 | 4133 | if (order) { |
c86f86e9 | 4134 | *order = size_buf.order; |
c3545579 JD |
4135 | dout(" order %u", (unsigned int)*order); |
4136 | } | |
9d475de5 AE |
4137 | *snap_size = le64_to_cpu(size_buf.size); |
4138 | ||
c3545579 JD |
4139 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4140 | (unsigned long long)snap_id, | |
57385b51 | 4141 | (unsigned long long)*snap_size); |
9d475de5 AE |
4142 | |
4143 | return 0; | |
4144 | } | |
4145 | ||
4146 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4147 | { | |
4148 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4149 | &rbd_dev->header.obj_order, | |
4150 | &rbd_dev->header.image_size); | |
4151 | } | |
4152 | ||
1e130199 AE |
4153 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4154 | { | |
4155 | void *reply_buf; | |
4156 | int ret; | |
4157 | void *p; | |
4158 | ||
4159 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4160 | if (!reply_buf) | |
4161 | return -ENOMEM; | |
4162 | ||
36be9a76 | 4163 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4164 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4165 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4166 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4167 | if (ret < 0) |
4168 | goto out; | |
4169 | ||
4170 | p = reply_buf; | |
4171 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4172 | p + ret, NULL, GFP_NOIO); |
4173 | ret = 0; | |
1e130199 AE |
4174 | |
4175 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4176 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4177 | rbd_dev->header.object_prefix = NULL; | |
4178 | } else { | |
4179 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4180 | } | |
1e130199 AE |
4181 | out: |
4182 | kfree(reply_buf); | |
4183 | ||
4184 | return ret; | |
4185 | } | |
4186 | ||
b1b5402a AE |
4187 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4188 | u64 *snap_features) | |
4189 | { | |
4190 | __le64 snapid = cpu_to_le64(snap_id); | |
4191 | struct { | |
4192 | __le64 features; | |
4193 | __le64 incompat; | |
4157976b | 4194 | } __attribute__ ((packed)) features_buf = { 0 }; |
d889140c | 4195 | u64 incompat; |
b1b5402a AE |
4196 | int ret; |
4197 | ||
36be9a76 | 4198 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4199 | "rbd", "get_features", |
4157976b | 4200 | &snapid, sizeof (snapid), |
e2a58ee5 | 4201 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4202 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4203 | if (ret < 0) |
4204 | return ret; | |
57385b51 AE |
4205 | if (ret < sizeof (features_buf)) |
4206 | return -ERANGE; | |
d889140c AE |
4207 | |
4208 | incompat = le64_to_cpu(features_buf.incompat); | |
5cbf6f12 | 4209 | if (incompat & ~RBD_FEATURES_SUPPORTED) |
b8f5c6ed | 4210 | return -ENXIO; |
d889140c | 4211 | |
b1b5402a AE |
4212 | *snap_features = le64_to_cpu(features_buf.features); |
4213 | ||
4214 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4215 | (unsigned long long)snap_id, |
4216 | (unsigned long long)*snap_features, | |
4217 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4218 | |
4219 | return 0; | |
4220 | } | |
4221 | ||
4222 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4223 | { | |
4224 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4225 | &rbd_dev->header.features); | |
4226 | } | |
4227 | ||
86b00e0d AE |
4228 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4229 | { | |
4230 | struct rbd_spec *parent_spec; | |
4231 | size_t size; | |
4232 | void *reply_buf = NULL; | |
4233 | __le64 snapid; | |
4234 | void *p; | |
4235 | void *end; | |
642a2537 | 4236 | u64 pool_id; |
86b00e0d | 4237 | char *image_id; |
3b5cf2a2 | 4238 | u64 snap_id; |
86b00e0d | 4239 | u64 overlap; |
86b00e0d AE |
4240 | int ret; |
4241 | ||
4242 | parent_spec = rbd_spec_alloc(); | |
4243 | if (!parent_spec) | |
4244 | return -ENOMEM; | |
4245 | ||
4246 | size = sizeof (__le64) + /* pool_id */ | |
4247 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4248 | sizeof (__le64) + /* snap_id */ | |
4249 | sizeof (__le64); /* overlap */ | |
4250 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4251 | if (!reply_buf) { | |
4252 | ret = -ENOMEM; | |
4253 | goto out_err; | |
4254 | } | |
4255 | ||
4d9b67cd | 4256 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4257 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4258 | "rbd", "get_parent", |
4157976b | 4259 | &snapid, sizeof (snapid), |
e2a58ee5 | 4260 | reply_buf, size); |
36be9a76 | 4261 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4262 | if (ret < 0) |
4263 | goto out_err; | |
4264 | ||
86b00e0d | 4265 | p = reply_buf; |
57385b51 AE |
4266 | end = reply_buf + ret; |
4267 | ret = -ERANGE; | |
642a2537 | 4268 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4269 | if (pool_id == CEPH_NOPOOL) { |
4270 | /* | |
4271 | * Either the parent never existed, or we have | |
4272 | * record of it but the image got flattened so it no | |
4273 | * longer has a parent. When the parent of a | |
4274 | * layered image disappears we immediately set the | |
4275 | * overlap to 0. The effect of this is that all new | |
4276 | * requests will be treated as if the image had no | |
4277 | * parent. | |
4278 | */ | |
4279 | if (rbd_dev->parent_overlap) { | |
4280 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4281 | rbd_dev_parent_put(rbd_dev); |
4282 | pr_info("%s: clone image has been flattened\n", | |
4283 | rbd_dev->disk->disk_name); | |
4284 | } | |
4285 | ||
86b00e0d | 4286 | goto out; /* No parent? No problem. */ |
392a9dad | 4287 | } |
86b00e0d | 4288 | |
0903e875 AE |
4289 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4290 | ||
4291 | ret = -EIO; | |
642a2537 | 4292 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4293 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4294 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4295 | goto out_err; |
c0cd10db | 4296 | } |
0903e875 | 4297 | |
979ed480 | 4298 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4299 | if (IS_ERR(image_id)) { |
4300 | ret = PTR_ERR(image_id); | |
4301 | goto out_err; | |
4302 | } | |
3b5cf2a2 | 4303 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4304 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4305 | ||
3b5cf2a2 AE |
4306 | /* |
4307 | * The parent won't change (except when the clone is | |
4308 | * flattened, already handled that). So we only need to | |
4309 | * record the parent spec we have not already done so. | |
4310 | */ | |
4311 | if (!rbd_dev->parent_spec) { | |
4312 | parent_spec->pool_id = pool_id; | |
4313 | parent_spec->image_id = image_id; | |
4314 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4315 | rbd_dev->parent_spec = parent_spec; |
4316 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4317 | } else { |
4318 | kfree(image_id); | |
3b5cf2a2 AE |
4319 | } |
4320 | ||
4321 | /* | |
cf32bd9c ID |
4322 | * We always update the parent overlap. If it's zero we issue |
4323 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 4324 | */ |
3b5cf2a2 | 4325 | if (!overlap) { |
3b5cf2a2 | 4326 | if (parent_spec) { |
cf32bd9c ID |
4327 | /* refresh, careful to warn just once */ |
4328 | if (rbd_dev->parent_overlap) | |
4329 | rbd_warn(rbd_dev, | |
4330 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 4331 | } else { |
cf32bd9c ID |
4332 | /* initial probe */ |
4333 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 4334 | } |
70cf49cf | 4335 | } |
cf32bd9c ID |
4336 | rbd_dev->parent_overlap = overlap; |
4337 | ||
86b00e0d AE |
4338 | out: |
4339 | ret = 0; | |
4340 | out_err: | |
4341 | kfree(reply_buf); | |
4342 | rbd_spec_put(parent_spec); | |
4343 | ||
4344 | return ret; | |
4345 | } | |
4346 | ||
cc070d59 AE |
4347 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4348 | { | |
4349 | struct { | |
4350 | __le64 stripe_unit; | |
4351 | __le64 stripe_count; | |
4352 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4353 | size_t size = sizeof (striping_info_buf); | |
4354 | void *p; | |
4355 | u64 obj_size; | |
4356 | u64 stripe_unit; | |
4357 | u64 stripe_count; | |
4358 | int ret; | |
4359 | ||
4360 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4361 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4362 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4363 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4364 | if (ret < 0) | |
4365 | return ret; | |
4366 | if (ret < size) | |
4367 | return -ERANGE; | |
4368 | ||
4369 | /* | |
4370 | * We don't actually support the "fancy striping" feature | |
4371 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4372 | * defaults the behavior is the same as before. So find | |
4373 | * out, and only fail if the image has non-default values. | |
4374 | */ | |
4375 | ret = -EINVAL; | |
4376 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4377 | p = &striping_info_buf; | |
4378 | stripe_unit = ceph_decode_64(&p); | |
4379 | if (stripe_unit != obj_size) { | |
4380 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4381 | "(got %llu want %llu)", | |
4382 | stripe_unit, obj_size); | |
4383 | return -EINVAL; | |
4384 | } | |
4385 | stripe_count = ceph_decode_64(&p); | |
4386 | if (stripe_count != 1) { | |
4387 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4388 | "(got %llu want 1)", stripe_count); | |
4389 | return -EINVAL; | |
4390 | } | |
500d0c0f AE |
4391 | rbd_dev->header.stripe_unit = stripe_unit; |
4392 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4393 | |
4394 | return 0; | |
4395 | } | |
4396 | ||
9e15b77d AE |
4397 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4398 | { | |
4399 | size_t image_id_size; | |
4400 | char *image_id; | |
4401 | void *p; | |
4402 | void *end; | |
4403 | size_t size; | |
4404 | void *reply_buf = NULL; | |
4405 | size_t len = 0; | |
4406 | char *image_name = NULL; | |
4407 | int ret; | |
4408 | ||
4409 | rbd_assert(!rbd_dev->spec->image_name); | |
4410 | ||
69e7a02f AE |
4411 | len = strlen(rbd_dev->spec->image_id); |
4412 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4413 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4414 | if (!image_id) | |
4415 | return NULL; | |
4416 | ||
4417 | p = image_id; | |
4157976b | 4418 | end = image_id + image_id_size; |
57385b51 | 4419 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4420 | |
4421 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4422 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4423 | if (!reply_buf) | |
4424 | goto out; | |
4425 | ||
36be9a76 | 4426 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4427 | "rbd", "dir_get_name", |
4428 | image_id, image_id_size, | |
e2a58ee5 | 4429 | reply_buf, size); |
9e15b77d AE |
4430 | if (ret < 0) |
4431 | goto out; | |
4432 | p = reply_buf; | |
f40eb349 AE |
4433 | end = reply_buf + ret; |
4434 | ||
9e15b77d AE |
4435 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4436 | if (IS_ERR(image_name)) | |
4437 | image_name = NULL; | |
4438 | else | |
4439 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4440 | out: | |
4441 | kfree(reply_buf); | |
4442 | kfree(image_id); | |
4443 | ||
4444 | return image_name; | |
4445 | } | |
4446 | ||
2ad3d716 AE |
4447 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4448 | { | |
4449 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4450 | const char *snap_name; | |
4451 | u32 which = 0; | |
4452 | ||
4453 | /* Skip over names until we find the one we are looking for */ | |
4454 | ||
4455 | snap_name = rbd_dev->header.snap_names; | |
4456 | while (which < snapc->num_snaps) { | |
4457 | if (!strcmp(name, snap_name)) | |
4458 | return snapc->snaps[which]; | |
4459 | snap_name += strlen(snap_name) + 1; | |
4460 | which++; | |
4461 | } | |
4462 | return CEPH_NOSNAP; | |
4463 | } | |
4464 | ||
4465 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4466 | { | |
4467 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4468 | u32 which; | |
4469 | bool found = false; | |
4470 | u64 snap_id; | |
4471 | ||
4472 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4473 | const char *snap_name; | |
4474 | ||
4475 | snap_id = snapc->snaps[which]; | |
4476 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4477 | if (IS_ERR(snap_name)) { |
4478 | /* ignore no-longer existing snapshots */ | |
4479 | if (PTR_ERR(snap_name) == -ENOENT) | |
4480 | continue; | |
4481 | else | |
4482 | break; | |
4483 | } | |
2ad3d716 AE |
4484 | found = !strcmp(name, snap_name); |
4485 | kfree(snap_name); | |
4486 | } | |
4487 | return found ? snap_id : CEPH_NOSNAP; | |
4488 | } | |
4489 | ||
4490 | /* | |
4491 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4492 | * no snapshot by that name is found, or if an error occurs. | |
4493 | */ | |
4494 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4495 | { | |
4496 | if (rbd_dev->image_format == 1) | |
4497 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4498 | ||
4499 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4500 | } | |
4501 | ||
9e15b77d | 4502 | /* |
04077599 ID |
4503 | * An image being mapped will have everything but the snap id. |
4504 | */ | |
4505 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4506 | { | |
4507 | struct rbd_spec *spec = rbd_dev->spec; | |
4508 | ||
4509 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4510 | rbd_assert(spec->image_id && spec->image_name); | |
4511 | rbd_assert(spec->snap_name); | |
4512 | ||
4513 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4514 | u64 snap_id; | |
4515 | ||
4516 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4517 | if (snap_id == CEPH_NOSNAP) | |
4518 | return -ENOENT; | |
4519 | ||
4520 | spec->snap_id = snap_id; | |
4521 | } else { | |
4522 | spec->snap_id = CEPH_NOSNAP; | |
4523 | } | |
4524 | ||
4525 | return 0; | |
4526 | } | |
4527 | ||
4528 | /* | |
4529 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4530 | * |
04077599 ID |
4531 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4532 | * can't figure out the name for an image id. | |
9e15b77d | 4533 | */ |
04077599 | 4534 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4535 | { |
2e9f7f1c AE |
4536 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4537 | struct rbd_spec *spec = rbd_dev->spec; | |
4538 | const char *pool_name; | |
4539 | const char *image_name; | |
4540 | const char *snap_name; | |
9e15b77d AE |
4541 | int ret; |
4542 | ||
04077599 ID |
4543 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4544 | rbd_assert(spec->image_id); | |
4545 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4546 | |
2e9f7f1c | 4547 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4548 | |
2e9f7f1c AE |
4549 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4550 | if (!pool_name) { | |
4551 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4552 | return -EIO; |
4553 | } | |
2e9f7f1c AE |
4554 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4555 | if (!pool_name) | |
9e15b77d AE |
4556 | return -ENOMEM; |
4557 | ||
4558 | /* Fetch the image name; tolerate failure here */ | |
4559 | ||
2e9f7f1c AE |
4560 | image_name = rbd_dev_image_name(rbd_dev); |
4561 | if (!image_name) | |
06ecc6cb | 4562 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4563 | |
04077599 | 4564 | /* Fetch the snapshot name */ |
9e15b77d | 4565 | |
2e9f7f1c | 4566 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4567 | if (IS_ERR(snap_name)) { |
4568 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4569 | goto out_err; |
2e9f7f1c AE |
4570 | } |
4571 | ||
4572 | spec->pool_name = pool_name; | |
4573 | spec->image_name = image_name; | |
4574 | spec->snap_name = snap_name; | |
9e15b77d AE |
4575 | |
4576 | return 0; | |
04077599 | 4577 | |
9e15b77d | 4578 | out_err: |
2e9f7f1c AE |
4579 | kfree(image_name); |
4580 | kfree(pool_name); | |
9e15b77d AE |
4581 | return ret; |
4582 | } | |
4583 | ||
cc4a38bd | 4584 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4585 | { |
4586 | size_t size; | |
4587 | int ret; | |
4588 | void *reply_buf; | |
4589 | void *p; | |
4590 | void *end; | |
4591 | u64 seq; | |
4592 | u32 snap_count; | |
4593 | struct ceph_snap_context *snapc; | |
4594 | u32 i; | |
4595 | ||
4596 | /* | |
4597 | * We'll need room for the seq value (maximum snapshot id), | |
4598 | * snapshot count, and array of that many snapshot ids. | |
4599 | * For now we have a fixed upper limit on the number we're | |
4600 | * prepared to receive. | |
4601 | */ | |
4602 | size = sizeof (__le64) + sizeof (__le32) + | |
4603 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4604 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4605 | if (!reply_buf) | |
4606 | return -ENOMEM; | |
4607 | ||
36be9a76 | 4608 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4609 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4610 | reply_buf, size); |
36be9a76 | 4611 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4612 | if (ret < 0) |
4613 | goto out; | |
4614 | ||
35d489f9 | 4615 | p = reply_buf; |
57385b51 AE |
4616 | end = reply_buf + ret; |
4617 | ret = -ERANGE; | |
35d489f9 AE |
4618 | ceph_decode_64_safe(&p, end, seq, out); |
4619 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4620 | ||
4621 | /* | |
4622 | * Make sure the reported number of snapshot ids wouldn't go | |
4623 | * beyond the end of our buffer. But before checking that, | |
4624 | * make sure the computed size of the snapshot context we | |
4625 | * allocate is representable in a size_t. | |
4626 | */ | |
4627 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4628 | / sizeof (u64)) { | |
4629 | ret = -EINVAL; | |
4630 | goto out; | |
4631 | } | |
4632 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4633 | goto out; | |
468521c1 | 4634 | ret = 0; |
35d489f9 | 4635 | |
812164f8 | 4636 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4637 | if (!snapc) { |
4638 | ret = -ENOMEM; | |
4639 | goto out; | |
4640 | } | |
35d489f9 | 4641 | snapc->seq = seq; |
35d489f9 AE |
4642 | for (i = 0; i < snap_count; i++) |
4643 | snapc->snaps[i] = ceph_decode_64(&p); | |
4644 | ||
49ece554 | 4645 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4646 | rbd_dev->header.snapc = snapc; |
4647 | ||
4648 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4649 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4650 | out: |
4651 | kfree(reply_buf); | |
4652 | ||
57385b51 | 4653 | return ret; |
35d489f9 AE |
4654 | } |
4655 | ||
54cac61f AE |
4656 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4657 | u64 snap_id) | |
b8b1e2db AE |
4658 | { |
4659 | size_t size; | |
4660 | void *reply_buf; | |
54cac61f | 4661 | __le64 snapid; |
b8b1e2db AE |
4662 | int ret; |
4663 | void *p; | |
4664 | void *end; | |
b8b1e2db AE |
4665 | char *snap_name; |
4666 | ||
4667 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4668 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4669 | if (!reply_buf) | |
4670 | return ERR_PTR(-ENOMEM); | |
4671 | ||
54cac61f | 4672 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4673 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4674 | "rbd", "get_snapshot_name", |
54cac61f | 4675 | &snapid, sizeof (snapid), |
e2a58ee5 | 4676 | reply_buf, size); |
36be9a76 | 4677 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4678 | if (ret < 0) { |
4679 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4680 | goto out; |
f40eb349 | 4681 | } |
b8b1e2db AE |
4682 | |
4683 | p = reply_buf; | |
f40eb349 | 4684 | end = reply_buf + ret; |
e5c35534 | 4685 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4686 | if (IS_ERR(snap_name)) |
b8b1e2db | 4687 | goto out; |
b8b1e2db | 4688 | |
f40eb349 | 4689 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4690 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4691 | out: |
4692 | kfree(reply_buf); | |
4693 | ||
f40eb349 | 4694 | return snap_name; |
b8b1e2db AE |
4695 | } |
4696 | ||
2df3fac7 | 4697 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4698 | { |
2df3fac7 | 4699 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4700 | int ret; |
117973fb | 4701 | |
1617e40c JD |
4702 | ret = rbd_dev_v2_image_size(rbd_dev); |
4703 | if (ret) | |
cfbf6377 | 4704 | return ret; |
1617e40c | 4705 | |
2df3fac7 AE |
4706 | if (first_time) { |
4707 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4708 | if (ret) | |
cfbf6377 | 4709 | return ret; |
2df3fac7 AE |
4710 | } |
4711 | ||
cc4a38bd | 4712 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
4713 | if (ret && first_time) { |
4714 | kfree(rbd_dev->header.object_prefix); | |
4715 | rbd_dev->header.object_prefix = NULL; | |
4716 | } | |
117973fb AE |
4717 | |
4718 | return ret; | |
4719 | } | |
4720 | ||
a720ae09 ID |
4721 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4722 | { | |
4723 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4724 | ||
4725 | if (rbd_dev->image_format == 1) | |
4726 | return rbd_dev_v1_header_info(rbd_dev); | |
4727 | ||
4728 | return rbd_dev_v2_header_info(rbd_dev); | |
4729 | } | |
4730 | ||
1ddbe94e | 4731 | /* |
499afd5b | 4732 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4733 | * the rbd_dev to the global list. |
1ddbe94e | 4734 | */ |
f8a22fc2 | 4735 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4736 | { |
f8a22fc2 ID |
4737 | int new_dev_id; |
4738 | ||
9b60e70b ID |
4739 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4740 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4741 | GFP_KERNEL); | |
f8a22fc2 ID |
4742 | if (new_dev_id < 0) |
4743 | return new_dev_id; | |
4744 | ||
4745 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4746 | |
4747 | spin_lock(&rbd_dev_list_lock); | |
4748 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4749 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4750 | |
70eebd20 | 4751 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4752 | |
4753 | return 0; | |
1ddbe94e | 4754 | } |
b7f23c36 | 4755 | |
1ddbe94e | 4756 | /* |
499afd5b AE |
4757 | * Remove an rbd_dev from the global list, and record that its |
4758 | * identifier is no longer in use. | |
1ddbe94e | 4759 | */ |
e2839308 | 4760 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4761 | { |
499afd5b AE |
4762 | spin_lock(&rbd_dev_list_lock); |
4763 | list_del_init(&rbd_dev->node); | |
4764 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4765 | |
f8a22fc2 ID |
4766 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4767 | ||
4768 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4769 | } |
4770 | ||
e28fff26 AE |
4771 | /* |
4772 | * Skips over white space at *buf, and updates *buf to point to the | |
4773 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4774 | * the token (string of non-white space characters) found. Note |
4775 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4776 | */ |
4777 | static inline size_t next_token(const char **buf) | |
4778 | { | |
4779 | /* | |
4780 | * These are the characters that produce nonzero for | |
4781 | * isspace() in the "C" and "POSIX" locales. | |
4782 | */ | |
4783 | const char *spaces = " \f\n\r\t\v"; | |
4784 | ||
4785 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4786 | ||
4787 | return strcspn(*buf, spaces); /* Return token length */ | |
4788 | } | |
4789 | ||
ea3352f4 AE |
4790 | /* |
4791 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4792 | * enough to hold a copy of it, and copies the token into the new | |
4793 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4794 | * that a duplicate buffer is created even for a zero-length token. | |
4795 | * | |
4796 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4797 | * pointer if memory for the duplicate was not available. If | |
4798 | * the lenp argument is a non-null pointer, the length of the token | |
4799 | * (not including the '\0') is returned in *lenp. | |
4800 | * | |
4801 | * If successful, the *buf pointer will be updated to point beyond | |
4802 | * the end of the found token. | |
4803 | * | |
4804 | * Note: uses GFP_KERNEL for allocation. | |
4805 | */ | |
4806 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4807 | { | |
4808 | char *dup; | |
4809 | size_t len; | |
4810 | ||
4811 | len = next_token(buf); | |
4caf35f9 | 4812 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4813 | if (!dup) |
4814 | return NULL; | |
ea3352f4 AE |
4815 | *(dup + len) = '\0'; |
4816 | *buf += len; | |
4817 | ||
4818 | if (lenp) | |
4819 | *lenp = len; | |
4820 | ||
4821 | return dup; | |
4822 | } | |
4823 | ||
a725f65e | 4824 | /* |
859c31df AE |
4825 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4826 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4827 | * and the data written is passed here via a NUL-terminated buffer. | |
4828 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4829 | * |
859c31df AE |
4830 | * The information extracted from these options is recorded in |
4831 | * the other parameters which return dynamically-allocated | |
4832 | * structures: | |
4833 | * ceph_opts | |
4834 | * The address of a pointer that will refer to a ceph options | |
4835 | * structure. Caller must release the returned pointer using | |
4836 | * ceph_destroy_options() when it is no longer needed. | |
4837 | * rbd_opts | |
4838 | * Address of an rbd options pointer. Fully initialized by | |
4839 | * this function; caller must release with kfree(). | |
4840 | * spec | |
4841 | * Address of an rbd image specification pointer. Fully | |
4842 | * initialized by this function based on parsed options. | |
4843 | * Caller must release with rbd_spec_put(). | |
4844 | * | |
4845 | * The options passed take this form: | |
4846 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4847 | * where: | |
4848 | * <mon_addrs> | |
4849 | * A comma-separated list of one or more monitor addresses. | |
4850 | * A monitor address is an ip address, optionally followed | |
4851 | * by a port number (separated by a colon). | |
4852 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4853 | * <options> | |
4854 | * A comma-separated list of ceph and/or rbd options. | |
4855 | * <pool_name> | |
4856 | * The name of the rados pool containing the rbd image. | |
4857 | * <image_name> | |
4858 | * The name of the image in that pool to map. | |
4859 | * <snap_id> | |
4860 | * An optional snapshot id. If provided, the mapping will | |
4861 | * present data from the image at the time that snapshot was | |
4862 | * created. The image head is used if no snapshot id is | |
4863 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4864 | */ |
859c31df | 4865 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4866 | struct ceph_options **ceph_opts, |
859c31df AE |
4867 | struct rbd_options **opts, |
4868 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4869 | { |
d22f76e7 | 4870 | size_t len; |
859c31df | 4871 | char *options; |
0ddebc0c | 4872 | const char *mon_addrs; |
ecb4dc22 | 4873 | char *snap_name; |
0ddebc0c | 4874 | size_t mon_addrs_size; |
859c31df | 4875 | struct rbd_spec *spec = NULL; |
4e9afeba | 4876 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4877 | struct ceph_options *copts; |
dc79b113 | 4878 | int ret; |
e28fff26 AE |
4879 | |
4880 | /* The first four tokens are required */ | |
4881 | ||
7ef3214a | 4882 | len = next_token(&buf); |
4fb5d671 AE |
4883 | if (!len) { |
4884 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4885 | return -EINVAL; | |
4886 | } | |
0ddebc0c | 4887 | mon_addrs = buf; |
f28e565a | 4888 | mon_addrs_size = len + 1; |
7ef3214a | 4889 | buf += len; |
a725f65e | 4890 | |
dc79b113 | 4891 | ret = -EINVAL; |
f28e565a AE |
4892 | options = dup_token(&buf, NULL); |
4893 | if (!options) | |
dc79b113 | 4894 | return -ENOMEM; |
4fb5d671 AE |
4895 | if (!*options) { |
4896 | rbd_warn(NULL, "no options provided"); | |
4897 | goto out_err; | |
4898 | } | |
e28fff26 | 4899 | |
859c31df AE |
4900 | spec = rbd_spec_alloc(); |
4901 | if (!spec) | |
f28e565a | 4902 | goto out_mem; |
859c31df AE |
4903 | |
4904 | spec->pool_name = dup_token(&buf, NULL); | |
4905 | if (!spec->pool_name) | |
4906 | goto out_mem; | |
4fb5d671 AE |
4907 | if (!*spec->pool_name) { |
4908 | rbd_warn(NULL, "no pool name provided"); | |
4909 | goto out_err; | |
4910 | } | |
e28fff26 | 4911 | |
69e7a02f | 4912 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4913 | if (!spec->image_name) |
f28e565a | 4914 | goto out_mem; |
4fb5d671 AE |
4915 | if (!*spec->image_name) { |
4916 | rbd_warn(NULL, "no image name provided"); | |
4917 | goto out_err; | |
4918 | } | |
d4b125e9 | 4919 | |
f28e565a AE |
4920 | /* |
4921 | * Snapshot name is optional; default is to use "-" | |
4922 | * (indicating the head/no snapshot). | |
4923 | */ | |
3feeb894 | 4924 | len = next_token(&buf); |
820a5f3e | 4925 | if (!len) { |
3feeb894 AE |
4926 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4927 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4928 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4929 | ret = -ENAMETOOLONG; |
f28e565a | 4930 | goto out_err; |
849b4260 | 4931 | } |
ecb4dc22 AE |
4932 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4933 | if (!snap_name) | |
f28e565a | 4934 | goto out_mem; |
ecb4dc22 AE |
4935 | *(snap_name + len) = '\0'; |
4936 | spec->snap_name = snap_name; | |
e5c35534 | 4937 | |
0ddebc0c | 4938 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4939 | |
4e9afeba AE |
4940 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4941 | if (!rbd_opts) | |
4942 | goto out_mem; | |
4943 | ||
4944 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 4945 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
d22f76e7 | 4946 | |
859c31df | 4947 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4948 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4949 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4950 | if (IS_ERR(copts)) { |
4951 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4952 | goto out_err; |
4953 | } | |
859c31df AE |
4954 | kfree(options); |
4955 | ||
4956 | *ceph_opts = copts; | |
4e9afeba | 4957 | *opts = rbd_opts; |
859c31df | 4958 | *rbd_spec = spec; |
0ddebc0c | 4959 | |
dc79b113 | 4960 | return 0; |
f28e565a | 4961 | out_mem: |
dc79b113 | 4962 | ret = -ENOMEM; |
d22f76e7 | 4963 | out_err: |
859c31df AE |
4964 | kfree(rbd_opts); |
4965 | rbd_spec_put(spec); | |
f28e565a | 4966 | kfree(options); |
d22f76e7 | 4967 | |
dc79b113 | 4968 | return ret; |
a725f65e AE |
4969 | } |
4970 | ||
30ba1f02 ID |
4971 | /* |
4972 | * Return pool id (>= 0) or a negative error code. | |
4973 | */ | |
4974 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4975 | { | |
a319bf56 | 4976 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 4977 | u64 newest_epoch; |
30ba1f02 ID |
4978 | int tries = 0; |
4979 | int ret; | |
4980 | ||
4981 | again: | |
4982 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4983 | if (ret == -ENOENT && tries++ < 1) { | |
4984 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4985 | &newest_epoch); | |
4986 | if (ret < 0) | |
4987 | return ret; | |
4988 | ||
4989 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4990 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4991 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
a319bf56 ID |
4992 | newest_epoch, |
4993 | opts->mount_timeout); | |
30ba1f02 ID |
4994 | goto again; |
4995 | } else { | |
4996 | /* the osdmap we have is new enough */ | |
4997 | return -ENOENT; | |
4998 | } | |
4999 | } | |
5000 | ||
5001 | return ret; | |
5002 | } | |
5003 | ||
589d30e0 AE |
5004 | /* |
5005 | * An rbd format 2 image has a unique identifier, distinct from the | |
5006 | * name given to it by the user. Internally, that identifier is | |
5007 | * what's used to specify the names of objects related to the image. | |
5008 | * | |
5009 | * A special "rbd id" object is used to map an rbd image name to its | |
5010 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5011 | * with the supplied name. | |
5012 | * | |
5013 | * This function will record the given rbd_dev's image_id field if | |
5014 | * it can be determined, and in that case will return 0. If any | |
5015 | * errors occur a negative errno will be returned and the rbd_dev's | |
5016 | * image_id field will be unchanged (and should be NULL). | |
5017 | */ | |
5018 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5019 | { | |
5020 | int ret; | |
5021 | size_t size; | |
5022 | char *object_name; | |
5023 | void *response; | |
c0fba368 | 5024 | char *image_id; |
2f82ee54 | 5025 | |
2c0d0a10 AE |
5026 | /* |
5027 | * When probing a parent image, the image id is already | |
5028 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5029 | * need to fetch the image id again in this case. We |
5030 | * do still need to set the image format though. | |
2c0d0a10 | 5031 | */ |
c0fba368 AE |
5032 | if (rbd_dev->spec->image_id) { |
5033 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5034 | ||
2c0d0a10 | 5035 | return 0; |
c0fba368 | 5036 | } |
2c0d0a10 | 5037 | |
589d30e0 AE |
5038 | /* |
5039 | * First, see if the format 2 image id file exists, and if | |
5040 | * so, get the image's persistent id from it. | |
5041 | */ | |
69e7a02f | 5042 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5043 | object_name = kmalloc(size, GFP_NOIO); |
5044 | if (!object_name) | |
5045 | return -ENOMEM; | |
0d7dbfce | 5046 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5047 | dout("rbd id object name is %s\n", object_name); |
5048 | ||
5049 | /* Response will be an encoded string, which includes a length */ | |
5050 | ||
5051 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5052 | response = kzalloc(size, GFP_NOIO); | |
5053 | if (!response) { | |
5054 | ret = -ENOMEM; | |
5055 | goto out; | |
5056 | } | |
5057 | ||
c0fba368 AE |
5058 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5059 | ||
36be9a76 | 5060 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5061 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5062 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5063 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5064 | if (ret == -ENOENT) { |
5065 | image_id = kstrdup("", GFP_KERNEL); | |
5066 | ret = image_id ? 0 : -ENOMEM; | |
5067 | if (!ret) | |
5068 | rbd_dev->image_format = 1; | |
7dd440c9 | 5069 | } else if (ret >= 0) { |
c0fba368 AE |
5070 | void *p = response; |
5071 | ||
5072 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5073 | NULL, GFP_NOIO); |
461f758a | 5074 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5075 | if (!ret) |
5076 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5077 | } |
5078 | ||
5079 | if (!ret) { | |
5080 | rbd_dev->spec->image_id = image_id; | |
5081 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5082 | } |
5083 | out: | |
5084 | kfree(response); | |
5085 | kfree(object_name); | |
5086 | ||
5087 | return ret; | |
5088 | } | |
5089 | ||
3abef3b3 AE |
5090 | /* |
5091 | * Undo whatever state changes are made by v1 or v2 header info | |
5092 | * call. | |
5093 | */ | |
6fd48b3b AE |
5094 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5095 | { | |
5096 | struct rbd_image_header *header; | |
5097 | ||
e69b8d41 | 5098 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5099 | |
5100 | /* Free dynamic fields from the header, then zero it out */ | |
5101 | ||
5102 | header = &rbd_dev->header; | |
812164f8 | 5103 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5104 | kfree(header->snap_sizes); |
5105 | kfree(header->snap_names); | |
5106 | kfree(header->object_prefix); | |
5107 | memset(header, 0, sizeof (*header)); | |
5108 | } | |
5109 | ||
2df3fac7 | 5110 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5111 | { |
5112 | int ret; | |
a30b71b9 | 5113 | |
1e130199 | 5114 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5115 | if (ret) |
b1b5402a AE |
5116 | goto out_err; |
5117 | ||
2df3fac7 AE |
5118 | /* |
5119 | * Get the and check features for the image. Currently the | |
5120 | * features are assumed to never change. | |
5121 | */ | |
b1b5402a | 5122 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5123 | if (ret) |
9d475de5 | 5124 | goto out_err; |
35d489f9 | 5125 | |
cc070d59 AE |
5126 | /* If the image supports fancy striping, get its parameters */ |
5127 | ||
5128 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5129 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5130 | if (ret < 0) | |
5131 | goto out_err; | |
5132 | } | |
2df3fac7 | 5133 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5134 | |
35152979 | 5135 | return 0; |
9d475de5 | 5136 | out_err: |
642a2537 | 5137 | rbd_dev->header.features = 0; |
1e130199 AE |
5138 | kfree(rbd_dev->header.object_prefix); |
5139 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5140 | |
5141 | return ret; | |
a30b71b9 AE |
5142 | } |
5143 | ||
6d69bb53 ID |
5144 | /* |
5145 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
5146 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
5147 | * length of the already discovered part of the parent chain. | |
5148 | */ | |
5149 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 5150 | { |
2f82ee54 | 5151 | struct rbd_device *parent = NULL; |
124afba2 AE |
5152 | int ret; |
5153 | ||
5154 | if (!rbd_dev->parent_spec) | |
5155 | return 0; | |
124afba2 | 5156 | |
6d69bb53 ID |
5157 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
5158 | pr_info("parent chain is too long (%d)\n", depth); | |
5159 | ret = -EINVAL; | |
5160 | goto out_err; | |
5161 | } | |
5162 | ||
1f2c6651 ID |
5163 | parent = rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec, |
5164 | NULL); | |
5165 | if (!parent) { | |
5166 | ret = -ENOMEM; | |
124afba2 | 5167 | goto out_err; |
1f2c6651 ID |
5168 | } |
5169 | ||
5170 | /* | |
5171 | * Images related by parent/child relationships always share | |
5172 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
5173 | */ | |
5174 | __rbd_get_client(rbd_dev->rbd_client); | |
5175 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 5176 | |
6d69bb53 | 5177 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
5178 | if (ret < 0) |
5179 | goto out_err; | |
1f2c6651 | 5180 | |
124afba2 | 5181 | rbd_dev->parent = parent; |
a2acd00e | 5182 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 5183 | return 0; |
1f2c6651 | 5184 | |
124afba2 | 5185 | out_err: |
1f2c6651 ID |
5186 | rbd_dev_unparent(rbd_dev); |
5187 | if (parent) | |
124afba2 | 5188 | rbd_dev_destroy(parent); |
124afba2 AE |
5189 | return ret; |
5190 | } | |
5191 | ||
200a6a8b | 5192 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5193 | { |
83a06263 | 5194 | int ret; |
d1cf5788 | 5195 | |
f8a22fc2 ID |
5196 | /* Get an id and fill in device name. */ |
5197 | ||
5198 | ret = rbd_dev_id_get(rbd_dev); | |
5199 | if (ret) | |
5200 | return ret; | |
83a06263 | 5201 | |
83a06263 AE |
5202 | BUILD_BUG_ON(DEV_NAME_LEN |
5203 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5204 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5205 | ||
9b60e70b | 5206 | /* Record our major and minor device numbers. */ |
83a06263 | 5207 | |
9b60e70b ID |
5208 | if (!single_major) { |
5209 | ret = register_blkdev(0, rbd_dev->name); | |
5210 | if (ret < 0) | |
5211 | goto err_out_id; | |
5212 | ||
5213 | rbd_dev->major = ret; | |
5214 | rbd_dev->minor = 0; | |
5215 | } else { | |
5216 | rbd_dev->major = rbd_major; | |
5217 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5218 | } | |
83a06263 AE |
5219 | |
5220 | /* Set up the blkdev mapping. */ | |
5221 | ||
5222 | ret = rbd_init_disk(rbd_dev); | |
5223 | if (ret) | |
5224 | goto err_out_blkdev; | |
5225 | ||
f35a4dee | 5226 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5227 | if (ret) |
5228 | goto err_out_disk; | |
bc1ecc65 | 5229 | |
f35a4dee | 5230 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5231 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 5232 | |
dd5ac32d ID |
5233 | dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id); |
5234 | ret = device_add(&rbd_dev->dev); | |
f35a4dee | 5235 | if (ret) |
f5ee37bd | 5236 | goto err_out_mapping; |
83a06263 | 5237 | |
83a06263 AE |
5238 | /* Everything's ready. Announce the disk to the world. */ |
5239 | ||
129b79d4 | 5240 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
83a06263 AE |
5241 | add_disk(rbd_dev->disk); |
5242 | ||
5243 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, | |
5244 | (unsigned long long) rbd_dev->mapping.size); | |
5245 | ||
5246 | return ret; | |
2f82ee54 | 5247 | |
f35a4dee AE |
5248 | err_out_mapping: |
5249 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5250 | err_out_disk: |
5251 | rbd_free_disk(rbd_dev); | |
5252 | err_out_blkdev: | |
9b60e70b ID |
5253 | if (!single_major) |
5254 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5255 | err_out_id: |
5256 | rbd_dev_id_put(rbd_dev); | |
83a06263 AE |
5257 | return ret; |
5258 | } | |
5259 | ||
332bb12d AE |
5260 | static int rbd_dev_header_name(struct rbd_device *rbd_dev) |
5261 | { | |
5262 | struct rbd_spec *spec = rbd_dev->spec; | |
5263 | size_t size; | |
5264 | ||
5265 | /* Record the header object name for this rbd image. */ | |
5266 | ||
5267 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
5268 | ||
5269 | if (rbd_dev->image_format == 1) | |
5270 | size = strlen(spec->image_name) + sizeof (RBD_SUFFIX); | |
5271 | else | |
5272 | size = sizeof (RBD_HEADER_PREFIX) + strlen(spec->image_id); | |
5273 | ||
5274 | rbd_dev->header_name = kmalloc(size, GFP_KERNEL); | |
5275 | if (!rbd_dev->header_name) | |
5276 | return -ENOMEM; | |
5277 | ||
5278 | if (rbd_dev->image_format == 1) | |
5279 | sprintf(rbd_dev->header_name, "%s%s", | |
5280 | spec->image_name, RBD_SUFFIX); | |
5281 | else | |
5282 | sprintf(rbd_dev->header_name, "%s%s", | |
5283 | RBD_HEADER_PREFIX, spec->image_id); | |
5284 | return 0; | |
5285 | } | |
5286 | ||
200a6a8b AE |
5287 | static void rbd_dev_image_release(struct rbd_device *rbd_dev) |
5288 | { | |
6fd48b3b | 5289 | rbd_dev_unprobe(rbd_dev); |
200a6a8b | 5290 | kfree(rbd_dev->header_name); |
6fd48b3b AE |
5291 | rbd_dev->header_name = NULL; |
5292 | rbd_dev->image_format = 0; | |
5293 | kfree(rbd_dev->spec->image_id); | |
5294 | rbd_dev->spec->image_id = NULL; | |
5295 | ||
200a6a8b AE |
5296 | rbd_dev_destroy(rbd_dev); |
5297 | } | |
5298 | ||
a30b71b9 AE |
5299 | /* |
5300 | * Probe for the existence of the header object for the given rbd | |
1f3ef788 AE |
5301 | * device. If this image is the one being mapped (i.e., not a |
5302 | * parent), initiate a watch on its header object before using that | |
5303 | * object to get detailed information about the rbd image. | |
a30b71b9 | 5304 | */ |
6d69bb53 | 5305 | static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth) |
a30b71b9 AE |
5306 | { |
5307 | int ret; | |
5308 | ||
5309 | /* | |
3abef3b3 AE |
5310 | * Get the id from the image id object. Unless there's an |
5311 | * error, rbd_dev->spec->image_id will be filled in with | |
5312 | * a dynamically-allocated string, and rbd_dev->image_format | |
5313 | * will be set to either 1 or 2. | |
a30b71b9 AE |
5314 | */ |
5315 | ret = rbd_dev_image_id(rbd_dev); | |
5316 | if (ret) | |
c0fba368 | 5317 | return ret; |
c0fba368 | 5318 | |
332bb12d AE |
5319 | ret = rbd_dev_header_name(rbd_dev); |
5320 | if (ret) | |
5321 | goto err_out_format; | |
5322 | ||
6d69bb53 | 5323 | if (!depth) { |
fca27065 | 5324 | ret = rbd_dev_header_watch_sync(rbd_dev); |
1fe48023 ID |
5325 | if (ret) { |
5326 | if (ret == -ENOENT) | |
5327 | pr_info("image %s/%s does not exist\n", | |
5328 | rbd_dev->spec->pool_name, | |
5329 | rbd_dev->spec->image_name); | |
1f3ef788 | 5330 | goto out_header_name; |
1fe48023 | 5331 | } |
1f3ef788 | 5332 | } |
b644de2b | 5333 | |
a720ae09 | 5334 | ret = rbd_dev_header_info(rbd_dev); |
5655c4d9 | 5335 | if (ret) |
b644de2b | 5336 | goto err_out_watch; |
83a06263 | 5337 | |
04077599 ID |
5338 | /* |
5339 | * If this image is the one being mapped, we have pool name and | |
5340 | * id, image name and id, and snap name - need to fill snap id. | |
5341 | * Otherwise this is a parent image, identified by pool, image | |
5342 | * and snap ids - need to fill in names for those ids. | |
5343 | */ | |
6d69bb53 | 5344 | if (!depth) |
04077599 ID |
5345 | ret = rbd_spec_fill_snap_id(rbd_dev); |
5346 | else | |
5347 | ret = rbd_spec_fill_names(rbd_dev); | |
1fe48023 ID |
5348 | if (ret) { |
5349 | if (ret == -ENOENT) | |
5350 | pr_info("snap %s/%s@%s does not exist\n", | |
5351 | rbd_dev->spec->pool_name, | |
5352 | rbd_dev->spec->image_name, | |
5353 | rbd_dev->spec->snap_name); | |
33dca39f | 5354 | goto err_out_probe; |
1fe48023 | 5355 | } |
9bb81c9b | 5356 | |
e8f59b59 ID |
5357 | if (rbd_dev->header.features & RBD_FEATURE_LAYERING) { |
5358 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
5359 | if (ret) | |
5360 | goto err_out_probe; | |
5361 | ||
5362 | /* | |
5363 | * Need to warn users if this image is the one being | |
5364 | * mapped and has a parent. | |
5365 | */ | |
6d69bb53 | 5366 | if (!depth && rbd_dev->parent_spec) |
e8f59b59 ID |
5367 | rbd_warn(rbd_dev, |
5368 | "WARNING: kernel layering is EXPERIMENTAL!"); | |
5369 | } | |
5370 | ||
6d69bb53 | 5371 | ret = rbd_dev_probe_parent(rbd_dev, depth); |
30d60ba2 AE |
5372 | if (ret) |
5373 | goto err_out_probe; | |
5374 | ||
5375 | dout("discovered format %u image, header name is %s\n", | |
5376 | rbd_dev->image_format, rbd_dev->header_name); | |
30d60ba2 | 5377 | return 0; |
e8f59b59 | 5378 | |
6fd48b3b AE |
5379 | err_out_probe: |
5380 | rbd_dev_unprobe(rbd_dev); | |
b644de2b | 5381 | err_out_watch: |
6d69bb53 | 5382 | if (!depth) |
fca27065 | 5383 | rbd_dev_header_unwatch_sync(rbd_dev); |
332bb12d AE |
5384 | out_header_name: |
5385 | kfree(rbd_dev->header_name); | |
5386 | rbd_dev->header_name = NULL; | |
5387 | err_out_format: | |
5388 | rbd_dev->image_format = 0; | |
5655c4d9 AE |
5389 | kfree(rbd_dev->spec->image_id); |
5390 | rbd_dev->spec->image_id = NULL; | |
a30b71b9 AE |
5391 | return ret; |
5392 | } | |
5393 | ||
9b60e70b ID |
5394 | static ssize_t do_rbd_add(struct bus_type *bus, |
5395 | const char *buf, | |
5396 | size_t count) | |
602adf40 | 5397 | { |
cb8627c7 | 5398 | struct rbd_device *rbd_dev = NULL; |
dc79b113 | 5399 | struct ceph_options *ceph_opts = NULL; |
4e9afeba | 5400 | struct rbd_options *rbd_opts = NULL; |
859c31df | 5401 | struct rbd_spec *spec = NULL; |
9d3997fd | 5402 | struct rbd_client *rbdc; |
51344a38 | 5403 | bool read_only; |
b51c83c2 | 5404 | int rc; |
602adf40 YS |
5405 | |
5406 | if (!try_module_get(THIS_MODULE)) | |
5407 | return -ENODEV; | |
5408 | ||
602adf40 | 5409 | /* parse add command */ |
859c31df | 5410 | rc = rbd_add_parse_args(buf, &ceph_opts, &rbd_opts, &spec); |
dc79b113 | 5411 | if (rc < 0) |
dd5ac32d | 5412 | goto out; |
78cea76e | 5413 | |
9d3997fd AE |
5414 | rbdc = rbd_get_client(ceph_opts); |
5415 | if (IS_ERR(rbdc)) { | |
5416 | rc = PTR_ERR(rbdc); | |
0ddebc0c | 5417 | goto err_out_args; |
9d3997fd | 5418 | } |
602adf40 | 5419 | |
602adf40 | 5420 | /* pick the pool */ |
30ba1f02 | 5421 | rc = rbd_add_get_pool_id(rbdc, spec->pool_name); |
1fe48023 ID |
5422 | if (rc < 0) { |
5423 | if (rc == -ENOENT) | |
5424 | pr_info("pool %s does not exist\n", spec->pool_name); | |
602adf40 | 5425 | goto err_out_client; |
1fe48023 | 5426 | } |
c0cd10db | 5427 | spec->pool_id = (u64)rc; |
859c31df | 5428 | |
0903e875 AE |
5429 | /* The ceph file layout needs to fit pool id in 32 bits */ |
5430 | ||
c0cd10db | 5431 | if (spec->pool_id > (u64)U32_MAX) { |
9584d508 | 5432 | rbd_warn(NULL, "pool id too large (%llu > %u)", |
c0cd10db | 5433 | (unsigned long long)spec->pool_id, U32_MAX); |
0903e875 AE |
5434 | rc = -EIO; |
5435 | goto err_out_client; | |
5436 | } | |
5437 | ||
d147543d | 5438 | rbd_dev = rbd_dev_create(rbdc, spec, rbd_opts); |
b51c83c2 ID |
5439 | if (!rbd_dev) { |
5440 | rc = -ENOMEM; | |
bd4ba655 | 5441 | goto err_out_client; |
b51c83c2 | 5442 | } |
c53d5893 AE |
5443 | rbdc = NULL; /* rbd_dev now owns this */ |
5444 | spec = NULL; /* rbd_dev now owns this */ | |
d147543d | 5445 | rbd_opts = NULL; /* rbd_dev now owns this */ |
602adf40 | 5446 | |
6d69bb53 | 5447 | rc = rbd_dev_image_probe(rbd_dev, 0); |
a30b71b9 | 5448 | if (rc < 0) |
c53d5893 | 5449 | goto err_out_rbd_dev; |
05fd6f6f | 5450 | |
7ce4eef7 AE |
5451 | /* If we are mapping a snapshot it must be marked read-only */ |
5452 | ||
d147543d | 5453 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
5454 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
5455 | read_only = true; | |
5456 | rbd_dev->mapping.read_only = read_only; | |
5457 | ||
b536f69a | 5458 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5459 | if (rc) { |
e37180c0 ID |
5460 | /* |
5461 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5462 | * rbd_dev_image_release() without refactoring, see | |
5463 | * commit 1f3ef78861ac. | |
5464 | */ | |
5465 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 | 5466 | rbd_dev_image_release(rbd_dev); |
dd5ac32d | 5467 | goto out; |
3abef3b3 AE |
5468 | } |
5469 | ||
dd5ac32d ID |
5470 | rc = count; |
5471 | out: | |
5472 | module_put(THIS_MODULE); | |
5473 | return rc; | |
b536f69a | 5474 | |
c53d5893 AE |
5475 | err_out_rbd_dev: |
5476 | rbd_dev_destroy(rbd_dev); | |
bd4ba655 | 5477 | err_out_client: |
9d3997fd | 5478 | rbd_put_client(rbdc); |
0ddebc0c | 5479 | err_out_args: |
859c31df | 5480 | rbd_spec_put(spec); |
d147543d | 5481 | kfree(rbd_opts); |
dd5ac32d | 5482 | goto out; |
602adf40 YS |
5483 | } |
5484 | ||
9b60e70b ID |
5485 | static ssize_t rbd_add(struct bus_type *bus, |
5486 | const char *buf, | |
5487 | size_t count) | |
5488 | { | |
5489 | if (single_major) | |
5490 | return -EINVAL; | |
5491 | ||
5492 | return do_rbd_add(bus, buf, count); | |
5493 | } | |
5494 | ||
5495 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5496 | const char *buf, | |
5497 | size_t count) | |
5498 | { | |
5499 | return do_rbd_add(bus, buf, count); | |
5500 | } | |
5501 | ||
dd5ac32d | 5502 | static void rbd_dev_device_release(struct rbd_device *rbd_dev) |
602adf40 | 5503 | { |
602adf40 | 5504 | rbd_free_disk(rbd_dev); |
200a6a8b | 5505 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
dd5ac32d | 5506 | device_del(&rbd_dev->dev); |
6d80b130 | 5507 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5508 | if (!single_major) |
5509 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5510 | rbd_dev_id_put(rbd_dev); |
602adf40 YS |
5511 | } |
5512 | ||
05a46afd AE |
5513 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5514 | { | |
ad945fc1 | 5515 | while (rbd_dev->parent) { |
05a46afd AE |
5516 | struct rbd_device *first = rbd_dev; |
5517 | struct rbd_device *second = first->parent; | |
5518 | struct rbd_device *third; | |
5519 | ||
5520 | /* | |
5521 | * Follow to the parent with no grandparent and | |
5522 | * remove it. | |
5523 | */ | |
5524 | while (second && (third = second->parent)) { | |
5525 | first = second; | |
5526 | second = third; | |
5527 | } | |
ad945fc1 | 5528 | rbd_assert(second); |
8ad42cd0 | 5529 | rbd_dev_image_release(second); |
ad945fc1 AE |
5530 | first->parent = NULL; |
5531 | first->parent_overlap = 0; | |
5532 | ||
5533 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5534 | rbd_spec_put(first->parent_spec); |
5535 | first->parent_spec = NULL; | |
05a46afd AE |
5536 | } |
5537 | } | |
5538 | ||
9b60e70b ID |
5539 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5540 | const char *buf, | |
5541 | size_t count) | |
602adf40 YS |
5542 | { |
5543 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5544 | struct list_head *tmp; |
5545 | int dev_id; | |
602adf40 | 5546 | unsigned long ul; |
82a442d2 | 5547 | bool already = false; |
0d8189e1 | 5548 | int ret; |
602adf40 | 5549 | |
bb8e0e84 | 5550 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5551 | if (ret) |
5552 | return ret; | |
602adf40 YS |
5553 | |
5554 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5555 | dev_id = (int)ul; |
5556 | if (dev_id != ul) | |
602adf40 YS |
5557 | return -EINVAL; |
5558 | ||
751cc0e3 AE |
5559 | ret = -ENOENT; |
5560 | spin_lock(&rbd_dev_list_lock); | |
5561 | list_for_each(tmp, &rbd_dev_list) { | |
5562 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5563 | if (rbd_dev->dev_id == dev_id) { | |
5564 | ret = 0; | |
5565 | break; | |
5566 | } | |
42382b70 | 5567 | } |
751cc0e3 AE |
5568 | if (!ret) { |
5569 | spin_lock_irq(&rbd_dev->lock); | |
5570 | if (rbd_dev->open_count) | |
5571 | ret = -EBUSY; | |
5572 | else | |
82a442d2 AE |
5573 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5574 | &rbd_dev->flags); | |
751cc0e3 AE |
5575 | spin_unlock_irq(&rbd_dev->lock); |
5576 | } | |
5577 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5578 | if (ret < 0 || already) |
1ba0f1e7 | 5579 | return ret; |
751cc0e3 | 5580 | |
fca27065 | 5581 | rbd_dev_header_unwatch_sync(rbd_dev); |
9abc5990 JD |
5582 | /* |
5583 | * flush remaining watch callbacks - these must be complete | |
5584 | * before the osd_client is shutdown | |
5585 | */ | |
5586 | dout("%s: flushing notifies", __func__); | |
5587 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 | 5588 | |
9875201e JD |
5589 | /* |
5590 | * Don't free anything from rbd_dev->disk until after all | |
5591 | * notifies are completely processed. Otherwise | |
5592 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5593 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5594 | */ | |
dd5ac32d | 5595 | rbd_dev_device_release(rbd_dev); |
8ad42cd0 | 5596 | rbd_dev_image_release(rbd_dev); |
aafb230e | 5597 | |
1ba0f1e7 | 5598 | return count; |
602adf40 YS |
5599 | } |
5600 | ||
9b60e70b ID |
5601 | static ssize_t rbd_remove(struct bus_type *bus, |
5602 | const char *buf, | |
5603 | size_t count) | |
5604 | { | |
5605 | if (single_major) | |
5606 | return -EINVAL; | |
5607 | ||
5608 | return do_rbd_remove(bus, buf, count); | |
5609 | } | |
5610 | ||
5611 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5612 | const char *buf, | |
5613 | size_t count) | |
5614 | { | |
5615 | return do_rbd_remove(bus, buf, count); | |
5616 | } | |
5617 | ||
602adf40 YS |
5618 | /* |
5619 | * create control files in sysfs | |
dfc5606d | 5620 | * /sys/bus/rbd/... |
602adf40 YS |
5621 | */ |
5622 | static int rbd_sysfs_init(void) | |
5623 | { | |
dfc5606d | 5624 | int ret; |
602adf40 | 5625 | |
fed4c143 | 5626 | ret = device_register(&rbd_root_dev); |
21079786 | 5627 | if (ret < 0) |
dfc5606d | 5628 | return ret; |
602adf40 | 5629 | |
fed4c143 AE |
5630 | ret = bus_register(&rbd_bus_type); |
5631 | if (ret < 0) | |
5632 | device_unregister(&rbd_root_dev); | |
602adf40 | 5633 | |
602adf40 YS |
5634 | return ret; |
5635 | } | |
5636 | ||
5637 | static void rbd_sysfs_cleanup(void) | |
5638 | { | |
dfc5606d | 5639 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5640 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5641 | } |
5642 | ||
1c2a9dfe AE |
5643 | static int rbd_slab_init(void) |
5644 | { | |
5645 | rbd_assert(!rbd_img_request_cache); | |
5646 | rbd_img_request_cache = kmem_cache_create("rbd_img_request", | |
5647 | sizeof (struct rbd_img_request), | |
5648 | __alignof__(struct rbd_img_request), | |
5649 | 0, NULL); | |
868311b1 AE |
5650 | if (!rbd_img_request_cache) |
5651 | return -ENOMEM; | |
5652 | ||
5653 | rbd_assert(!rbd_obj_request_cache); | |
5654 | rbd_obj_request_cache = kmem_cache_create("rbd_obj_request", | |
5655 | sizeof (struct rbd_obj_request), | |
5656 | __alignof__(struct rbd_obj_request), | |
5657 | 0, NULL); | |
78c2a44a AE |
5658 | if (!rbd_obj_request_cache) |
5659 | goto out_err; | |
5660 | ||
5661 | rbd_assert(!rbd_segment_name_cache); | |
5662 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5663 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5664 | if (rbd_segment_name_cache) |
1c2a9dfe | 5665 | return 0; |
78c2a44a | 5666 | out_err: |
13bf2834 JL |
5667 | kmem_cache_destroy(rbd_obj_request_cache); |
5668 | rbd_obj_request_cache = NULL; | |
1c2a9dfe | 5669 | |
868311b1 AE |
5670 | kmem_cache_destroy(rbd_img_request_cache); |
5671 | rbd_img_request_cache = NULL; | |
5672 | ||
1c2a9dfe AE |
5673 | return -ENOMEM; |
5674 | } | |
5675 | ||
5676 | static void rbd_slab_exit(void) | |
5677 | { | |
78c2a44a AE |
5678 | rbd_assert(rbd_segment_name_cache); |
5679 | kmem_cache_destroy(rbd_segment_name_cache); | |
5680 | rbd_segment_name_cache = NULL; | |
5681 | ||
868311b1 AE |
5682 | rbd_assert(rbd_obj_request_cache); |
5683 | kmem_cache_destroy(rbd_obj_request_cache); | |
5684 | rbd_obj_request_cache = NULL; | |
5685 | ||
1c2a9dfe AE |
5686 | rbd_assert(rbd_img_request_cache); |
5687 | kmem_cache_destroy(rbd_img_request_cache); | |
5688 | rbd_img_request_cache = NULL; | |
5689 | } | |
5690 | ||
cc344fa1 | 5691 | static int __init rbd_init(void) |
602adf40 YS |
5692 | { |
5693 | int rc; | |
5694 | ||
1e32d34c AE |
5695 | if (!libceph_compatible(NULL)) { |
5696 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5697 | return -EINVAL; |
5698 | } | |
e1b4d96d | 5699 | |
1c2a9dfe | 5700 | rc = rbd_slab_init(); |
602adf40 YS |
5701 | if (rc) |
5702 | return rc; | |
e1b4d96d | 5703 | |
f5ee37bd ID |
5704 | /* |
5705 | * The number of active work items is limited by the number of | |
f77303bd | 5706 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
5707 | */ |
5708 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
5709 | if (!rbd_wq) { | |
5710 | rc = -ENOMEM; | |
5711 | goto err_out_slab; | |
5712 | } | |
5713 | ||
9b60e70b ID |
5714 | if (single_major) { |
5715 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5716 | if (rbd_major < 0) { | |
5717 | rc = rbd_major; | |
f5ee37bd | 5718 | goto err_out_wq; |
9b60e70b ID |
5719 | } |
5720 | } | |
5721 | ||
1c2a9dfe AE |
5722 | rc = rbd_sysfs_init(); |
5723 | if (rc) | |
9b60e70b ID |
5724 | goto err_out_blkdev; |
5725 | ||
5726 | if (single_major) | |
5727 | pr_info("loaded (major %d)\n", rbd_major); | |
5728 | else | |
5729 | pr_info("loaded\n"); | |
1c2a9dfe | 5730 | |
e1b4d96d ID |
5731 | return 0; |
5732 | ||
9b60e70b ID |
5733 | err_out_blkdev: |
5734 | if (single_major) | |
5735 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
5736 | err_out_wq: |
5737 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
5738 | err_out_slab: |
5739 | rbd_slab_exit(); | |
1c2a9dfe | 5740 | return rc; |
602adf40 YS |
5741 | } |
5742 | ||
cc344fa1 | 5743 | static void __exit rbd_exit(void) |
602adf40 | 5744 | { |
ffe312cf | 5745 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5746 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5747 | if (single_major) |
5748 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 5749 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 5750 | rbd_slab_exit(); |
602adf40 YS |
5751 | } |
5752 | ||
5753 | module_init(rbd_init); | |
5754 | module_exit(rbd_exit); | |
5755 | ||
d552c619 | 5756 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5757 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5758 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5759 | /* following authorship retained from original osdblk.c */ |
5760 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5761 | ||
90da258b | 5762 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5763 | MODULE_LICENSE("GPL"); |