]>
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); | |
59c2be1e | 541 | |
602adf40 YS |
542 | static int rbd_open(struct block_device *bdev, fmode_t mode) |
543 | { | |
f0f8cef5 | 544 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; |
b82d167b | 545 | bool removing = false; |
602adf40 | 546 | |
f84344f3 | 547 | if ((mode & FMODE_WRITE) && rbd_dev->mapping.read_only) |
602adf40 YS |
548 | return -EROFS; |
549 | ||
a14ea269 | 550 | spin_lock_irq(&rbd_dev->lock); |
b82d167b AE |
551 | if (test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) |
552 | removing = true; | |
553 | else | |
554 | rbd_dev->open_count++; | |
a14ea269 | 555 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b AE |
556 | if (removing) |
557 | return -ENOENT; | |
558 | ||
c3e946ce | 559 | (void) get_device(&rbd_dev->dev); |
340c7a2b | 560 | |
602adf40 YS |
561 | return 0; |
562 | } | |
563 | ||
db2a144b | 564 | static void rbd_release(struct gendisk *disk, fmode_t mode) |
dfc5606d YS |
565 | { |
566 | struct rbd_device *rbd_dev = disk->private_data; | |
b82d167b AE |
567 | unsigned long open_count_before; |
568 | ||
a14ea269 | 569 | spin_lock_irq(&rbd_dev->lock); |
b82d167b | 570 | open_count_before = rbd_dev->open_count--; |
a14ea269 | 571 | spin_unlock_irq(&rbd_dev->lock); |
b82d167b | 572 | rbd_assert(open_count_before > 0); |
dfc5606d | 573 | |
c3e946ce | 574 | put_device(&rbd_dev->dev); |
dfc5606d YS |
575 | } |
576 | ||
131fd9f6 GZ |
577 | static int rbd_ioctl_set_ro(struct rbd_device *rbd_dev, unsigned long arg) |
578 | { | |
77f33c03 | 579 | int ret = 0; |
131fd9f6 GZ |
580 | int val; |
581 | bool ro; | |
77f33c03 | 582 | bool ro_changed = false; |
131fd9f6 | 583 | |
77f33c03 | 584 | /* get_user() may sleep, so call it before taking rbd_dev->lock */ |
131fd9f6 GZ |
585 | if (get_user(val, (int __user *)(arg))) |
586 | return -EFAULT; | |
587 | ||
588 | ro = val ? true : false; | |
589 | /* Snapshot doesn't allow to write*/ | |
590 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP && !ro) | |
591 | return -EROFS; | |
592 | ||
77f33c03 JD |
593 | spin_lock_irq(&rbd_dev->lock); |
594 | /* prevent others open this device */ | |
595 | if (rbd_dev->open_count > 1) { | |
596 | ret = -EBUSY; | |
597 | goto out; | |
598 | } | |
599 | ||
131fd9f6 GZ |
600 | if (rbd_dev->mapping.read_only != ro) { |
601 | rbd_dev->mapping.read_only = ro; | |
77f33c03 | 602 | ro_changed = true; |
131fd9f6 GZ |
603 | } |
604 | ||
77f33c03 JD |
605 | out: |
606 | spin_unlock_irq(&rbd_dev->lock); | |
607 | /* set_disk_ro() may sleep, so call it after releasing rbd_dev->lock */ | |
608 | if (ret == 0 && ro_changed) | |
609 | set_disk_ro(rbd_dev->disk, ro ? 1 : 0); | |
610 | ||
611 | return ret; | |
131fd9f6 GZ |
612 | } |
613 | ||
614 | static int rbd_ioctl(struct block_device *bdev, fmode_t mode, | |
615 | unsigned int cmd, unsigned long arg) | |
616 | { | |
617 | struct rbd_device *rbd_dev = bdev->bd_disk->private_data; | |
618 | int ret = 0; | |
619 | ||
131fd9f6 GZ |
620 | switch (cmd) { |
621 | case BLKROSET: | |
622 | ret = rbd_ioctl_set_ro(rbd_dev, arg); | |
623 | break; | |
624 | default: | |
625 | ret = -ENOTTY; | |
626 | } | |
627 | ||
131fd9f6 GZ |
628 | return ret; |
629 | } | |
630 | ||
631 | #ifdef CONFIG_COMPAT | |
632 | static int rbd_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
633 | unsigned int cmd, unsigned long arg) | |
634 | { | |
635 | return rbd_ioctl(bdev, mode, cmd, arg); | |
636 | } | |
637 | #endif /* CONFIG_COMPAT */ | |
638 | ||
602adf40 YS |
639 | static const struct block_device_operations rbd_bd_ops = { |
640 | .owner = THIS_MODULE, | |
641 | .open = rbd_open, | |
dfc5606d | 642 | .release = rbd_release, |
131fd9f6 GZ |
643 | .ioctl = rbd_ioctl, |
644 | #ifdef CONFIG_COMPAT | |
645 | .compat_ioctl = rbd_compat_ioctl, | |
646 | #endif | |
602adf40 YS |
647 | }; |
648 | ||
649 | /* | |
7262cfca | 650 | * Initialize an rbd client instance. Success or not, this function |
cfbf6377 | 651 | * consumes ceph_opts. Caller holds client_mutex. |
602adf40 | 652 | */ |
f8c38929 | 653 | static struct rbd_client *rbd_client_create(struct ceph_options *ceph_opts) |
602adf40 YS |
654 | { |
655 | struct rbd_client *rbdc; | |
656 | int ret = -ENOMEM; | |
657 | ||
37206ee5 | 658 | dout("%s:\n", __func__); |
602adf40 YS |
659 | rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL); |
660 | if (!rbdc) | |
661 | goto out_opt; | |
662 | ||
663 | kref_init(&rbdc->kref); | |
664 | INIT_LIST_HEAD(&rbdc->node); | |
665 | ||
43ae4701 | 666 | rbdc->client = ceph_create_client(ceph_opts, rbdc, 0, 0); |
602adf40 | 667 | if (IS_ERR(rbdc->client)) |
08f75463 | 668 | goto out_rbdc; |
43ae4701 | 669 | ceph_opts = NULL; /* Now rbdc->client is responsible for ceph_opts */ |
602adf40 YS |
670 | |
671 | ret = ceph_open_session(rbdc->client); | |
672 | if (ret < 0) | |
08f75463 | 673 | goto out_client; |
602adf40 | 674 | |
432b8587 | 675 | spin_lock(&rbd_client_list_lock); |
602adf40 | 676 | list_add_tail(&rbdc->node, &rbd_client_list); |
432b8587 | 677 | spin_unlock(&rbd_client_list_lock); |
602adf40 | 678 | |
37206ee5 | 679 | dout("%s: rbdc %p\n", __func__, rbdc); |
bc534d86 | 680 | |
602adf40 | 681 | return rbdc; |
08f75463 | 682 | out_client: |
602adf40 | 683 | ceph_destroy_client(rbdc->client); |
08f75463 | 684 | out_rbdc: |
602adf40 YS |
685 | kfree(rbdc); |
686 | out_opt: | |
43ae4701 AE |
687 | if (ceph_opts) |
688 | ceph_destroy_options(ceph_opts); | |
37206ee5 AE |
689 | dout("%s: error %d\n", __func__, ret); |
690 | ||
28f259b7 | 691 | return ERR_PTR(ret); |
602adf40 YS |
692 | } |
693 | ||
2f82ee54 AE |
694 | static struct rbd_client *__rbd_get_client(struct rbd_client *rbdc) |
695 | { | |
696 | kref_get(&rbdc->kref); | |
697 | ||
698 | return rbdc; | |
699 | } | |
700 | ||
602adf40 | 701 | /* |
1f7ba331 AE |
702 | * Find a ceph client with specific addr and configuration. If |
703 | * found, bump its reference count. | |
602adf40 | 704 | */ |
1f7ba331 | 705 | static struct rbd_client *rbd_client_find(struct ceph_options *ceph_opts) |
602adf40 YS |
706 | { |
707 | struct rbd_client *client_node; | |
1f7ba331 | 708 | bool found = false; |
602adf40 | 709 | |
43ae4701 | 710 | if (ceph_opts->flags & CEPH_OPT_NOSHARE) |
602adf40 YS |
711 | return NULL; |
712 | ||
1f7ba331 AE |
713 | spin_lock(&rbd_client_list_lock); |
714 | list_for_each_entry(client_node, &rbd_client_list, node) { | |
715 | if (!ceph_compare_options(ceph_opts, client_node->client)) { | |
2f82ee54 AE |
716 | __rbd_get_client(client_node); |
717 | ||
1f7ba331 AE |
718 | found = true; |
719 | break; | |
720 | } | |
721 | } | |
722 | spin_unlock(&rbd_client_list_lock); | |
723 | ||
724 | return found ? client_node : NULL; | |
602adf40 YS |
725 | } |
726 | ||
59c2be1e | 727 | /* |
210c104c | 728 | * (Per device) rbd map options |
59c2be1e YS |
729 | */ |
730 | enum { | |
b5584180 | 731 | Opt_queue_depth, |
59c2be1e YS |
732 | Opt_last_int, |
733 | /* int args above */ | |
734 | Opt_last_string, | |
735 | /* string args above */ | |
cc0538b6 AE |
736 | Opt_read_only, |
737 | Opt_read_write, | |
210c104c | 738 | Opt_err |
59c2be1e YS |
739 | }; |
740 | ||
43ae4701 | 741 | static match_table_t rbd_opts_tokens = { |
b5584180 | 742 | {Opt_queue_depth, "queue_depth=%d"}, |
59c2be1e YS |
743 | /* int args above */ |
744 | /* string args above */ | |
be466c1c | 745 | {Opt_read_only, "read_only"}, |
cc0538b6 AE |
746 | {Opt_read_only, "ro"}, /* Alternate spelling */ |
747 | {Opt_read_write, "read_write"}, | |
748 | {Opt_read_write, "rw"}, /* Alternate spelling */ | |
210c104c | 749 | {Opt_err, NULL} |
59c2be1e YS |
750 | }; |
751 | ||
98571b5a | 752 | struct rbd_options { |
b5584180 | 753 | int queue_depth; |
98571b5a AE |
754 | bool read_only; |
755 | }; | |
756 | ||
b5584180 | 757 | #define RBD_QUEUE_DEPTH_DEFAULT BLKDEV_MAX_RQ |
98571b5a AE |
758 | #define RBD_READ_ONLY_DEFAULT false |
759 | ||
59c2be1e YS |
760 | static int parse_rbd_opts_token(char *c, void *private) |
761 | { | |
43ae4701 | 762 | struct rbd_options *rbd_opts = private; |
59c2be1e YS |
763 | substring_t argstr[MAX_OPT_ARGS]; |
764 | int token, intval, ret; | |
765 | ||
43ae4701 | 766 | token = match_token(c, rbd_opts_tokens, argstr); |
59c2be1e YS |
767 | if (token < Opt_last_int) { |
768 | ret = match_int(&argstr[0], &intval); | |
769 | if (ret < 0) { | |
210c104c | 770 | pr_err("bad mount option arg (not int) at '%s'\n", c); |
59c2be1e YS |
771 | return ret; |
772 | } | |
773 | dout("got int token %d val %d\n", token, intval); | |
774 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
210c104c | 775 | dout("got string token %d val %s\n", token, argstr[0].from); |
59c2be1e YS |
776 | } else { |
777 | dout("got token %d\n", token); | |
778 | } | |
779 | ||
780 | switch (token) { | |
b5584180 ID |
781 | case Opt_queue_depth: |
782 | if (intval < 1) { | |
783 | pr_err("queue_depth out of range\n"); | |
784 | return -EINVAL; | |
785 | } | |
786 | rbd_opts->queue_depth = intval; | |
787 | break; | |
cc0538b6 AE |
788 | case Opt_read_only: |
789 | rbd_opts->read_only = true; | |
790 | break; | |
791 | case Opt_read_write: | |
792 | rbd_opts->read_only = false; | |
793 | break; | |
59c2be1e | 794 | default: |
210c104c ID |
795 | /* libceph prints "bad option" msg */ |
796 | return -EINVAL; | |
59c2be1e | 797 | } |
210c104c | 798 | |
59c2be1e YS |
799 | return 0; |
800 | } | |
801 | ||
6d2940c8 GZ |
802 | static char* obj_op_name(enum obj_operation_type op_type) |
803 | { | |
804 | switch (op_type) { | |
805 | case OBJ_OP_READ: | |
806 | return "read"; | |
807 | case OBJ_OP_WRITE: | |
808 | return "write"; | |
90e98c52 GZ |
809 | case OBJ_OP_DISCARD: |
810 | return "discard"; | |
6d2940c8 GZ |
811 | default: |
812 | return "???"; | |
813 | } | |
814 | } | |
815 | ||
602adf40 YS |
816 | /* |
817 | * Get a ceph client with specific addr and configuration, if one does | |
7262cfca AE |
818 | * not exist create it. Either way, ceph_opts is consumed by this |
819 | * function. | |
602adf40 | 820 | */ |
9d3997fd | 821 | static struct rbd_client *rbd_get_client(struct ceph_options *ceph_opts) |
602adf40 | 822 | { |
f8c38929 | 823 | struct rbd_client *rbdc; |
59c2be1e | 824 | |
cfbf6377 | 825 | mutex_lock_nested(&client_mutex, SINGLE_DEPTH_NESTING); |
1f7ba331 | 826 | rbdc = rbd_client_find(ceph_opts); |
9d3997fd | 827 | if (rbdc) /* using an existing client */ |
43ae4701 | 828 | ceph_destroy_options(ceph_opts); |
9d3997fd | 829 | else |
f8c38929 | 830 | rbdc = rbd_client_create(ceph_opts); |
cfbf6377 | 831 | mutex_unlock(&client_mutex); |
602adf40 | 832 | |
9d3997fd | 833 | return rbdc; |
602adf40 YS |
834 | } |
835 | ||
836 | /* | |
837 | * Destroy ceph client | |
d23a4b3f | 838 | * |
432b8587 | 839 | * Caller must hold rbd_client_list_lock. |
602adf40 YS |
840 | */ |
841 | static void rbd_client_release(struct kref *kref) | |
842 | { | |
843 | struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref); | |
844 | ||
37206ee5 | 845 | dout("%s: rbdc %p\n", __func__, rbdc); |
cd9d9f5d | 846 | spin_lock(&rbd_client_list_lock); |
602adf40 | 847 | list_del(&rbdc->node); |
cd9d9f5d | 848 | spin_unlock(&rbd_client_list_lock); |
602adf40 YS |
849 | |
850 | ceph_destroy_client(rbdc->client); | |
851 | kfree(rbdc); | |
852 | } | |
853 | ||
854 | /* | |
855 | * Drop reference to ceph client node. If it's not referenced anymore, release | |
856 | * it. | |
857 | */ | |
9d3997fd | 858 | static void rbd_put_client(struct rbd_client *rbdc) |
602adf40 | 859 | { |
c53d5893 AE |
860 | if (rbdc) |
861 | kref_put(&rbdc->kref, rbd_client_release); | |
602adf40 YS |
862 | } |
863 | ||
a30b71b9 AE |
864 | static bool rbd_image_format_valid(u32 image_format) |
865 | { | |
866 | return image_format == 1 || image_format == 2; | |
867 | } | |
868 | ||
8e94af8e AE |
869 | static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) |
870 | { | |
103a150f AE |
871 | size_t size; |
872 | u32 snap_count; | |
873 | ||
874 | /* The header has to start with the magic rbd header text */ | |
875 | if (memcmp(&ondisk->text, RBD_HEADER_TEXT, sizeof (RBD_HEADER_TEXT))) | |
876 | return false; | |
877 | ||
db2388b6 AE |
878 | /* The bio layer requires at least sector-sized I/O */ |
879 | ||
880 | if (ondisk->options.order < SECTOR_SHIFT) | |
881 | return false; | |
882 | ||
883 | /* If we use u64 in a few spots we may be able to loosen this */ | |
884 | ||
885 | if (ondisk->options.order > 8 * sizeof (int) - 1) | |
886 | return false; | |
887 | ||
103a150f AE |
888 | /* |
889 | * The size of a snapshot header has to fit in a size_t, and | |
890 | * that limits the number of snapshots. | |
891 | */ | |
892 | snap_count = le32_to_cpu(ondisk->snap_count); | |
893 | size = SIZE_MAX - sizeof (struct ceph_snap_context); | |
894 | if (snap_count > size / sizeof (__le64)) | |
895 | return false; | |
896 | ||
897 | /* | |
898 | * Not only that, but the size of the entire the snapshot | |
899 | * header must also be representable in a size_t. | |
900 | */ | |
901 | size -= snap_count * sizeof (__le64); | |
902 | if ((u64) size < le64_to_cpu(ondisk->snap_names_len)) | |
903 | return false; | |
904 | ||
905 | return true; | |
8e94af8e AE |
906 | } |
907 | ||
602adf40 | 908 | /* |
bb23e37a AE |
909 | * Fill an rbd image header with information from the given format 1 |
910 | * on-disk header. | |
602adf40 | 911 | */ |
662518b1 | 912 | static int rbd_header_from_disk(struct rbd_device *rbd_dev, |
4156d998 | 913 | struct rbd_image_header_ondisk *ondisk) |
602adf40 | 914 | { |
662518b1 | 915 | struct rbd_image_header *header = &rbd_dev->header; |
bb23e37a AE |
916 | bool first_time = header->object_prefix == NULL; |
917 | struct ceph_snap_context *snapc; | |
918 | char *object_prefix = NULL; | |
919 | char *snap_names = NULL; | |
920 | u64 *snap_sizes = NULL; | |
ccece235 | 921 | u32 snap_count; |
d2bb24e5 | 922 | size_t size; |
bb23e37a | 923 | int ret = -ENOMEM; |
621901d6 | 924 | u32 i; |
602adf40 | 925 | |
bb23e37a | 926 | /* Allocate this now to avoid having to handle failure below */ |
6a52325f | 927 | |
bb23e37a AE |
928 | if (first_time) { |
929 | size_t len; | |
103a150f | 930 | |
bb23e37a AE |
931 | len = strnlen(ondisk->object_prefix, |
932 | sizeof (ondisk->object_prefix)); | |
933 | object_prefix = kmalloc(len + 1, GFP_KERNEL); | |
934 | if (!object_prefix) | |
935 | return -ENOMEM; | |
936 | memcpy(object_prefix, ondisk->object_prefix, len); | |
937 | object_prefix[len] = '\0'; | |
938 | } | |
00f1f36f | 939 | |
bb23e37a | 940 | /* Allocate the snapshot context and fill it in */ |
00f1f36f | 941 | |
bb23e37a AE |
942 | snap_count = le32_to_cpu(ondisk->snap_count); |
943 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); | |
944 | if (!snapc) | |
945 | goto out_err; | |
946 | snapc->seq = le64_to_cpu(ondisk->snap_seq); | |
602adf40 | 947 | if (snap_count) { |
bb23e37a | 948 | struct rbd_image_snap_ondisk *snaps; |
f785cc1d AE |
949 | u64 snap_names_len = le64_to_cpu(ondisk->snap_names_len); |
950 | ||
bb23e37a | 951 | /* We'll keep a copy of the snapshot names... */ |
621901d6 | 952 | |
bb23e37a AE |
953 | if (snap_names_len > (u64)SIZE_MAX) |
954 | goto out_2big; | |
955 | snap_names = kmalloc(snap_names_len, GFP_KERNEL); | |
956 | if (!snap_names) | |
6a52325f AE |
957 | goto out_err; |
958 | ||
bb23e37a | 959 | /* ...as well as the array of their sizes. */ |
621901d6 | 960 | |
d2bb24e5 | 961 | size = snap_count * sizeof (*header->snap_sizes); |
bb23e37a AE |
962 | snap_sizes = kmalloc(size, GFP_KERNEL); |
963 | if (!snap_sizes) | |
6a52325f | 964 | goto out_err; |
bb23e37a | 965 | |
f785cc1d | 966 | /* |
bb23e37a AE |
967 | * Copy the names, and fill in each snapshot's id |
968 | * and size. | |
969 | * | |
99a41ebc | 970 | * Note that rbd_dev_v1_header_info() guarantees the |
bb23e37a | 971 | * ondisk buffer we're working with has |
f785cc1d AE |
972 | * snap_names_len bytes beyond the end of the |
973 | * snapshot id array, this memcpy() is safe. | |
974 | */ | |
bb23e37a AE |
975 | memcpy(snap_names, &ondisk->snaps[snap_count], snap_names_len); |
976 | snaps = ondisk->snaps; | |
977 | for (i = 0; i < snap_count; i++) { | |
978 | snapc->snaps[i] = le64_to_cpu(snaps[i].id); | |
979 | snap_sizes[i] = le64_to_cpu(snaps[i].image_size); | |
980 | } | |
602adf40 | 981 | } |
6a52325f | 982 | |
bb23e37a | 983 | /* We won't fail any more, fill in the header */ |
621901d6 | 984 | |
bb23e37a AE |
985 | if (first_time) { |
986 | header->object_prefix = object_prefix; | |
987 | header->obj_order = ondisk->options.order; | |
988 | header->crypt_type = ondisk->options.crypt_type; | |
989 | header->comp_type = ondisk->options.comp_type; | |
990 | /* The rest aren't used for format 1 images */ | |
991 | header->stripe_unit = 0; | |
992 | header->stripe_count = 0; | |
993 | header->features = 0; | |
602adf40 | 994 | } else { |
662518b1 AE |
995 | ceph_put_snap_context(header->snapc); |
996 | kfree(header->snap_names); | |
997 | kfree(header->snap_sizes); | |
602adf40 | 998 | } |
849b4260 | 999 | |
bb23e37a | 1000 | /* The remaining fields always get updated (when we refresh) */ |
621901d6 | 1001 | |
f84344f3 | 1002 | header->image_size = le64_to_cpu(ondisk->image_size); |
bb23e37a AE |
1003 | header->snapc = snapc; |
1004 | header->snap_names = snap_names; | |
1005 | header->snap_sizes = snap_sizes; | |
468521c1 | 1006 | |
602adf40 | 1007 | return 0; |
bb23e37a AE |
1008 | out_2big: |
1009 | ret = -EIO; | |
6a52325f | 1010 | out_err: |
bb23e37a AE |
1011 | kfree(snap_sizes); |
1012 | kfree(snap_names); | |
1013 | ceph_put_snap_context(snapc); | |
1014 | kfree(object_prefix); | |
ccece235 | 1015 | |
bb23e37a | 1016 | return ret; |
602adf40 YS |
1017 | } |
1018 | ||
9682fc6d AE |
1019 | static const char *_rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, u32 which) |
1020 | { | |
1021 | const char *snap_name; | |
1022 | ||
1023 | rbd_assert(which < rbd_dev->header.snapc->num_snaps); | |
1024 | ||
1025 | /* Skip over names until we find the one we are looking for */ | |
1026 | ||
1027 | snap_name = rbd_dev->header.snap_names; | |
1028 | while (which--) | |
1029 | snap_name += strlen(snap_name) + 1; | |
1030 | ||
1031 | return kstrdup(snap_name, GFP_KERNEL); | |
1032 | } | |
1033 | ||
30d1cff8 AE |
1034 | /* |
1035 | * Snapshot id comparison function for use with qsort()/bsearch(). | |
1036 | * Note that result is for snapshots in *descending* order. | |
1037 | */ | |
1038 | static int snapid_compare_reverse(const void *s1, const void *s2) | |
1039 | { | |
1040 | u64 snap_id1 = *(u64 *)s1; | |
1041 | u64 snap_id2 = *(u64 *)s2; | |
1042 | ||
1043 | if (snap_id1 < snap_id2) | |
1044 | return 1; | |
1045 | return snap_id1 == snap_id2 ? 0 : -1; | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * Search a snapshot context to see if the given snapshot id is | |
1050 | * present. | |
1051 | * | |
1052 | * Returns the position of the snapshot id in the array if it's found, | |
1053 | * or BAD_SNAP_INDEX otherwise. | |
1054 | * | |
1055 | * Note: The snapshot array is in kept sorted (by the osd) in | |
1056 | * reverse order, highest snapshot id first. | |
1057 | */ | |
9682fc6d AE |
1058 | static u32 rbd_dev_snap_index(struct rbd_device *rbd_dev, u64 snap_id) |
1059 | { | |
1060 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
30d1cff8 | 1061 | u64 *found; |
9682fc6d | 1062 | |
30d1cff8 AE |
1063 | found = bsearch(&snap_id, &snapc->snaps, snapc->num_snaps, |
1064 | sizeof (snap_id), snapid_compare_reverse); | |
9682fc6d | 1065 | |
30d1cff8 | 1066 | return found ? (u32)(found - &snapc->snaps[0]) : BAD_SNAP_INDEX; |
9682fc6d AE |
1067 | } |
1068 | ||
2ad3d716 AE |
1069 | static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev, |
1070 | u64 snap_id) | |
9e15b77d | 1071 | { |
54cac61f | 1072 | u32 which; |
da6a6b63 | 1073 | const char *snap_name; |
9e15b77d | 1074 | |
54cac61f AE |
1075 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1076 | if (which == BAD_SNAP_INDEX) | |
da6a6b63 | 1077 | return ERR_PTR(-ENOENT); |
54cac61f | 1078 | |
da6a6b63 JD |
1079 | snap_name = _rbd_dev_v1_snap_name(rbd_dev, which); |
1080 | return snap_name ? snap_name : ERR_PTR(-ENOMEM); | |
54cac61f AE |
1081 | } |
1082 | ||
1083 | static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id) | |
1084 | { | |
9e15b77d AE |
1085 | if (snap_id == CEPH_NOSNAP) |
1086 | return RBD_SNAP_HEAD_NAME; | |
1087 | ||
54cac61f AE |
1088 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1089 | if (rbd_dev->image_format == 1) | |
1090 | return rbd_dev_v1_snap_name(rbd_dev, snap_id); | |
9e15b77d | 1091 | |
54cac61f | 1092 | return rbd_dev_v2_snap_name(rbd_dev, snap_id); |
9e15b77d AE |
1093 | } |
1094 | ||
2ad3d716 AE |
1095 | static int rbd_snap_size(struct rbd_device *rbd_dev, u64 snap_id, |
1096 | u64 *snap_size) | |
602adf40 | 1097 | { |
2ad3d716 AE |
1098 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1099 | if (snap_id == CEPH_NOSNAP) { | |
1100 | *snap_size = rbd_dev->header.image_size; | |
1101 | } else if (rbd_dev->image_format == 1) { | |
1102 | u32 which; | |
602adf40 | 1103 | |
2ad3d716 AE |
1104 | which = rbd_dev_snap_index(rbd_dev, snap_id); |
1105 | if (which == BAD_SNAP_INDEX) | |
1106 | return -ENOENT; | |
e86924a8 | 1107 | |
2ad3d716 AE |
1108 | *snap_size = rbd_dev->header.snap_sizes[which]; |
1109 | } else { | |
1110 | u64 size = 0; | |
1111 | int ret; | |
1112 | ||
1113 | ret = _rbd_dev_v2_snap_size(rbd_dev, snap_id, NULL, &size); | |
1114 | if (ret) | |
1115 | return ret; | |
1116 | ||
1117 | *snap_size = size; | |
1118 | } | |
1119 | return 0; | |
602adf40 YS |
1120 | } |
1121 | ||
2ad3d716 AE |
1122 | static int rbd_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
1123 | u64 *snap_features) | |
602adf40 | 1124 | { |
2ad3d716 AE |
1125 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); |
1126 | if (snap_id == CEPH_NOSNAP) { | |
1127 | *snap_features = rbd_dev->header.features; | |
1128 | } else if (rbd_dev->image_format == 1) { | |
1129 | *snap_features = 0; /* No features for format 1 */ | |
602adf40 | 1130 | } else { |
2ad3d716 AE |
1131 | u64 features = 0; |
1132 | int ret; | |
8b0241f8 | 1133 | |
2ad3d716 AE |
1134 | ret = _rbd_dev_v2_snap_features(rbd_dev, snap_id, &features); |
1135 | if (ret) | |
1136 | return ret; | |
1137 | ||
1138 | *snap_features = features; | |
1139 | } | |
1140 | return 0; | |
1141 | } | |
1142 | ||
1143 | static int rbd_dev_mapping_set(struct rbd_device *rbd_dev) | |
1144 | { | |
8f4b7d98 | 1145 | u64 snap_id = rbd_dev->spec->snap_id; |
2ad3d716 AE |
1146 | u64 size = 0; |
1147 | u64 features = 0; | |
1148 | int ret; | |
1149 | ||
2ad3d716 AE |
1150 | ret = rbd_snap_size(rbd_dev, snap_id, &size); |
1151 | if (ret) | |
1152 | return ret; | |
1153 | ret = rbd_snap_features(rbd_dev, snap_id, &features); | |
1154 | if (ret) | |
1155 | return ret; | |
1156 | ||
1157 | rbd_dev->mapping.size = size; | |
1158 | rbd_dev->mapping.features = features; | |
1159 | ||
8b0241f8 | 1160 | return 0; |
602adf40 YS |
1161 | } |
1162 | ||
d1cf5788 AE |
1163 | static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) |
1164 | { | |
1165 | rbd_dev->mapping.size = 0; | |
1166 | rbd_dev->mapping.features = 0; | |
200a6a8b AE |
1167 | } |
1168 | ||
7d5079aa HS |
1169 | static void rbd_segment_name_free(const char *name) |
1170 | { | |
1171 | /* The explicit cast here is needed to drop the const qualifier */ | |
1172 | ||
1173 | kmem_cache_free(rbd_segment_name_cache, (void *)name); | |
1174 | } | |
1175 | ||
98571b5a | 1176 | static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset) |
602adf40 | 1177 | { |
65ccfe21 AE |
1178 | char *name; |
1179 | u64 segment; | |
1180 | int ret; | |
3a96d5cd | 1181 | char *name_format; |
602adf40 | 1182 | |
78c2a44a | 1183 | name = kmem_cache_alloc(rbd_segment_name_cache, GFP_NOIO); |
65ccfe21 AE |
1184 | if (!name) |
1185 | return NULL; | |
1186 | segment = offset >> rbd_dev->header.obj_order; | |
3a96d5cd JD |
1187 | name_format = "%s.%012llx"; |
1188 | if (rbd_dev->image_format == 2) | |
1189 | name_format = "%s.%016llx"; | |
2d0ebc5d | 1190 | ret = snprintf(name, CEPH_MAX_OID_NAME_LEN + 1, name_format, |
65ccfe21 | 1191 | rbd_dev->header.object_prefix, segment); |
2d0ebc5d | 1192 | if (ret < 0 || ret > CEPH_MAX_OID_NAME_LEN) { |
65ccfe21 AE |
1193 | pr_err("error formatting segment name for #%llu (%d)\n", |
1194 | segment, ret); | |
7d5079aa | 1195 | rbd_segment_name_free(name); |
65ccfe21 AE |
1196 | name = NULL; |
1197 | } | |
602adf40 | 1198 | |
65ccfe21 AE |
1199 | return name; |
1200 | } | |
602adf40 | 1201 | |
65ccfe21 AE |
1202 | static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset) |
1203 | { | |
1204 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
602adf40 | 1205 | |
65ccfe21 AE |
1206 | return offset & (segment_size - 1); |
1207 | } | |
1208 | ||
1209 | static u64 rbd_segment_length(struct rbd_device *rbd_dev, | |
1210 | u64 offset, u64 length) | |
1211 | { | |
1212 | u64 segment_size = (u64) 1 << rbd_dev->header.obj_order; | |
1213 | ||
1214 | offset &= segment_size - 1; | |
1215 | ||
aafb230e | 1216 | rbd_assert(length <= U64_MAX - offset); |
65ccfe21 AE |
1217 | if (offset + length > segment_size) |
1218 | length = segment_size - offset; | |
1219 | ||
1220 | return length; | |
602adf40 YS |
1221 | } |
1222 | ||
029bcbd8 JD |
1223 | /* |
1224 | * returns the size of an object in the image | |
1225 | */ | |
1226 | static u64 rbd_obj_bytes(struct rbd_image_header *header) | |
1227 | { | |
1228 | return 1 << header->obj_order; | |
1229 | } | |
1230 | ||
602adf40 YS |
1231 | /* |
1232 | * bio helpers | |
1233 | */ | |
1234 | ||
1235 | static void bio_chain_put(struct bio *chain) | |
1236 | { | |
1237 | struct bio *tmp; | |
1238 | ||
1239 | while (chain) { | |
1240 | tmp = chain; | |
1241 | chain = chain->bi_next; | |
1242 | bio_put(tmp); | |
1243 | } | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * zeros a bio chain, starting at specific offset | |
1248 | */ | |
1249 | static void zero_bio_chain(struct bio *chain, int start_ofs) | |
1250 | { | |
7988613b KO |
1251 | struct bio_vec bv; |
1252 | struct bvec_iter iter; | |
602adf40 YS |
1253 | unsigned long flags; |
1254 | void *buf; | |
602adf40 YS |
1255 | int pos = 0; |
1256 | ||
1257 | while (chain) { | |
7988613b KO |
1258 | bio_for_each_segment(bv, chain, iter) { |
1259 | if (pos + bv.bv_len > start_ofs) { | |
602adf40 | 1260 | int remainder = max(start_ofs - pos, 0); |
7988613b | 1261 | buf = bvec_kmap_irq(&bv, &flags); |
602adf40 | 1262 | memset(buf + remainder, 0, |
7988613b KO |
1263 | bv.bv_len - remainder); |
1264 | flush_dcache_page(bv.bv_page); | |
85b5aaa6 | 1265 | bvec_kunmap_irq(buf, &flags); |
602adf40 | 1266 | } |
7988613b | 1267 | pos += bv.bv_len; |
602adf40 YS |
1268 | } |
1269 | ||
1270 | chain = chain->bi_next; | |
1271 | } | |
1272 | } | |
1273 | ||
b9434c5b AE |
1274 | /* |
1275 | * similar to zero_bio_chain(), zeros data defined by a page array, | |
1276 | * starting at the given byte offset from the start of the array and | |
1277 | * continuing up to the given end offset. The pages array is | |
1278 | * assumed to be big enough to hold all bytes up to the end. | |
1279 | */ | |
1280 | static void zero_pages(struct page **pages, u64 offset, u64 end) | |
1281 | { | |
1282 | struct page **page = &pages[offset >> PAGE_SHIFT]; | |
1283 | ||
1284 | rbd_assert(end > offset); | |
1285 | rbd_assert(end - offset <= (u64)SIZE_MAX); | |
1286 | while (offset < end) { | |
1287 | size_t page_offset; | |
1288 | size_t length; | |
1289 | unsigned long flags; | |
1290 | void *kaddr; | |
1291 | ||
491205a8 GU |
1292 | page_offset = offset & ~PAGE_MASK; |
1293 | length = min_t(size_t, PAGE_SIZE - page_offset, end - offset); | |
b9434c5b AE |
1294 | local_irq_save(flags); |
1295 | kaddr = kmap_atomic(*page); | |
1296 | memset(kaddr + page_offset, 0, length); | |
e2156054 | 1297 | flush_dcache_page(*page); |
b9434c5b AE |
1298 | kunmap_atomic(kaddr); |
1299 | local_irq_restore(flags); | |
1300 | ||
1301 | offset += length; | |
1302 | page++; | |
1303 | } | |
1304 | } | |
1305 | ||
602adf40 | 1306 | /* |
f7760dad AE |
1307 | * Clone a portion of a bio, starting at the given byte offset |
1308 | * and continuing for the number of bytes indicated. | |
602adf40 | 1309 | */ |
f7760dad AE |
1310 | static struct bio *bio_clone_range(struct bio *bio_src, |
1311 | unsigned int offset, | |
1312 | unsigned int len, | |
1313 | gfp_t gfpmask) | |
602adf40 | 1314 | { |
f7760dad AE |
1315 | struct bio *bio; |
1316 | ||
5341a627 | 1317 | bio = bio_clone(bio_src, gfpmask); |
f7760dad AE |
1318 | if (!bio) |
1319 | return NULL; /* ENOMEM */ | |
602adf40 | 1320 | |
5341a627 | 1321 | bio_advance(bio, offset); |
4f024f37 | 1322 | bio->bi_iter.bi_size = len; |
f7760dad AE |
1323 | |
1324 | return bio; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * Clone a portion of a bio chain, starting at the given byte offset | |
1329 | * into the first bio in the source chain and continuing for the | |
1330 | * number of bytes indicated. The result is another bio chain of | |
1331 | * exactly the given length, or a null pointer on error. | |
1332 | * | |
1333 | * The bio_src and offset parameters are both in-out. On entry they | |
1334 | * refer to the first source bio and the offset into that bio where | |
1335 | * the start of data to be cloned is located. | |
1336 | * | |
1337 | * On return, bio_src is updated to refer to the bio in the source | |
1338 | * chain that contains first un-cloned byte, and *offset will | |
1339 | * contain the offset of that byte within that bio. | |
1340 | */ | |
1341 | static struct bio *bio_chain_clone_range(struct bio **bio_src, | |
1342 | unsigned int *offset, | |
1343 | unsigned int len, | |
1344 | gfp_t gfpmask) | |
1345 | { | |
1346 | struct bio *bi = *bio_src; | |
1347 | unsigned int off = *offset; | |
1348 | struct bio *chain = NULL; | |
1349 | struct bio **end; | |
1350 | ||
1351 | /* Build up a chain of clone bios up to the limit */ | |
1352 | ||
4f024f37 | 1353 | if (!bi || off >= bi->bi_iter.bi_size || !len) |
f7760dad | 1354 | return NULL; /* Nothing to clone */ |
602adf40 | 1355 | |
f7760dad AE |
1356 | end = &chain; |
1357 | while (len) { | |
1358 | unsigned int bi_size; | |
1359 | struct bio *bio; | |
1360 | ||
f5400b7a AE |
1361 | if (!bi) { |
1362 | rbd_warn(NULL, "bio_chain exhausted with %u left", len); | |
f7760dad | 1363 | goto out_err; /* EINVAL; ran out of bio's */ |
f5400b7a | 1364 | } |
4f024f37 | 1365 | bi_size = min_t(unsigned int, bi->bi_iter.bi_size - off, len); |
f7760dad AE |
1366 | bio = bio_clone_range(bi, off, bi_size, gfpmask); |
1367 | if (!bio) | |
1368 | goto out_err; /* ENOMEM */ | |
1369 | ||
1370 | *end = bio; | |
1371 | end = &bio->bi_next; | |
602adf40 | 1372 | |
f7760dad | 1373 | off += bi_size; |
4f024f37 | 1374 | if (off == bi->bi_iter.bi_size) { |
f7760dad AE |
1375 | bi = bi->bi_next; |
1376 | off = 0; | |
1377 | } | |
1378 | len -= bi_size; | |
1379 | } | |
1380 | *bio_src = bi; | |
1381 | *offset = off; | |
1382 | ||
1383 | return chain; | |
1384 | out_err: | |
1385 | bio_chain_put(chain); | |
602adf40 | 1386 | |
602adf40 YS |
1387 | return NULL; |
1388 | } | |
1389 | ||
926f9b3f AE |
1390 | /* |
1391 | * The default/initial value for all object request flags is 0. For | |
1392 | * each flag, once its value is set to 1 it is never reset to 0 | |
1393 | * again. | |
1394 | */ | |
57acbaa7 | 1395 | static void obj_request_img_data_set(struct rbd_obj_request *obj_request) |
926f9b3f | 1396 | { |
57acbaa7 | 1397 | if (test_and_set_bit(OBJ_REQ_IMG_DATA, &obj_request->flags)) { |
926f9b3f AE |
1398 | struct rbd_device *rbd_dev; |
1399 | ||
57acbaa7 | 1400 | rbd_dev = obj_request->img_request->rbd_dev; |
9584d508 | 1401 | rbd_warn(rbd_dev, "obj_request %p already marked img_data", |
926f9b3f AE |
1402 | obj_request); |
1403 | } | |
1404 | } | |
1405 | ||
57acbaa7 | 1406 | static bool obj_request_img_data_test(struct rbd_obj_request *obj_request) |
926f9b3f AE |
1407 | { |
1408 | smp_mb(); | |
57acbaa7 | 1409 | return test_bit(OBJ_REQ_IMG_DATA, &obj_request->flags) != 0; |
926f9b3f AE |
1410 | } |
1411 | ||
57acbaa7 | 1412 | static void obj_request_done_set(struct rbd_obj_request *obj_request) |
6365d33a | 1413 | { |
57acbaa7 AE |
1414 | if (test_and_set_bit(OBJ_REQ_DONE, &obj_request->flags)) { |
1415 | struct rbd_device *rbd_dev = NULL; | |
6365d33a | 1416 | |
57acbaa7 AE |
1417 | if (obj_request_img_data_test(obj_request)) |
1418 | rbd_dev = obj_request->img_request->rbd_dev; | |
9584d508 | 1419 | rbd_warn(rbd_dev, "obj_request %p already marked done", |
6365d33a AE |
1420 | obj_request); |
1421 | } | |
1422 | } | |
1423 | ||
57acbaa7 | 1424 | static bool obj_request_done_test(struct rbd_obj_request *obj_request) |
6365d33a AE |
1425 | { |
1426 | smp_mb(); | |
57acbaa7 | 1427 | return test_bit(OBJ_REQ_DONE, &obj_request->flags) != 0; |
6365d33a AE |
1428 | } |
1429 | ||
5679c59f AE |
1430 | /* |
1431 | * This sets the KNOWN flag after (possibly) setting the EXISTS | |
1432 | * flag. The latter is set based on the "exists" value provided. | |
1433 | * | |
1434 | * Note that for our purposes once an object exists it never goes | |
1435 | * away again. It's possible that the response from two existence | |
1436 | * checks are separated by the creation of the target object, and | |
1437 | * the first ("doesn't exist") response arrives *after* the second | |
1438 | * ("does exist"). In that case we ignore the second one. | |
1439 | */ | |
1440 | static void obj_request_existence_set(struct rbd_obj_request *obj_request, | |
1441 | bool exists) | |
1442 | { | |
1443 | if (exists) | |
1444 | set_bit(OBJ_REQ_EXISTS, &obj_request->flags); | |
1445 | set_bit(OBJ_REQ_KNOWN, &obj_request->flags); | |
1446 | smp_mb(); | |
1447 | } | |
1448 | ||
1449 | static bool obj_request_known_test(struct rbd_obj_request *obj_request) | |
1450 | { | |
1451 | smp_mb(); | |
1452 | return test_bit(OBJ_REQ_KNOWN, &obj_request->flags) != 0; | |
1453 | } | |
1454 | ||
1455 | static bool obj_request_exists_test(struct rbd_obj_request *obj_request) | |
1456 | { | |
1457 | smp_mb(); | |
1458 | return test_bit(OBJ_REQ_EXISTS, &obj_request->flags) != 0; | |
1459 | } | |
1460 | ||
9638556a ID |
1461 | static bool obj_request_overlaps_parent(struct rbd_obj_request *obj_request) |
1462 | { | |
1463 | struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev; | |
1464 | ||
1465 | return obj_request->img_offset < | |
1466 | round_up(rbd_dev->parent_overlap, rbd_obj_bytes(&rbd_dev->header)); | |
1467 | } | |
1468 | ||
bf0d5f50 AE |
1469 | static void rbd_obj_request_get(struct rbd_obj_request *obj_request) |
1470 | { | |
37206ee5 AE |
1471 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1472 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1473 | kref_get(&obj_request->kref); |
1474 | } | |
1475 | ||
1476 | static void rbd_obj_request_destroy(struct kref *kref); | |
1477 | static void rbd_obj_request_put(struct rbd_obj_request *obj_request) | |
1478 | { | |
1479 | rbd_assert(obj_request != NULL); | |
37206ee5 AE |
1480 | dout("%s: obj %p (was %d)\n", __func__, obj_request, |
1481 | atomic_read(&obj_request->kref.refcount)); | |
bf0d5f50 AE |
1482 | kref_put(&obj_request->kref, rbd_obj_request_destroy); |
1483 | } | |
1484 | ||
0f2d5be7 AE |
1485 | static void rbd_img_request_get(struct rbd_img_request *img_request) |
1486 | { | |
1487 | dout("%s: img %p (was %d)\n", __func__, img_request, | |
1488 | atomic_read(&img_request->kref.refcount)); | |
1489 | kref_get(&img_request->kref); | |
1490 | } | |
1491 | ||
e93f3152 AE |
1492 | static bool img_request_child_test(struct rbd_img_request *img_request); |
1493 | static void rbd_parent_request_destroy(struct kref *kref); | |
bf0d5f50 AE |
1494 | static void rbd_img_request_destroy(struct kref *kref); |
1495 | static void rbd_img_request_put(struct rbd_img_request *img_request) | |
1496 | { | |
1497 | rbd_assert(img_request != NULL); | |
37206ee5 AE |
1498 | dout("%s: img %p (was %d)\n", __func__, img_request, |
1499 | atomic_read(&img_request->kref.refcount)); | |
e93f3152 AE |
1500 | if (img_request_child_test(img_request)) |
1501 | kref_put(&img_request->kref, rbd_parent_request_destroy); | |
1502 | else | |
1503 | kref_put(&img_request->kref, rbd_img_request_destroy); | |
bf0d5f50 AE |
1504 | } |
1505 | ||
1506 | static inline void rbd_img_obj_request_add(struct rbd_img_request *img_request, | |
1507 | struct rbd_obj_request *obj_request) | |
1508 | { | |
25dcf954 AE |
1509 | rbd_assert(obj_request->img_request == NULL); |
1510 | ||
b155e86c | 1511 | /* Image request now owns object's original reference */ |
bf0d5f50 | 1512 | obj_request->img_request = img_request; |
25dcf954 | 1513 | obj_request->which = img_request->obj_request_count; |
6365d33a AE |
1514 | rbd_assert(!obj_request_img_data_test(obj_request)); |
1515 | obj_request_img_data_set(obj_request); | |
bf0d5f50 | 1516 | rbd_assert(obj_request->which != BAD_WHICH); |
25dcf954 AE |
1517 | img_request->obj_request_count++; |
1518 | list_add_tail(&obj_request->links, &img_request->obj_requests); | |
37206ee5 AE |
1519 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1520 | obj_request->which); | |
bf0d5f50 AE |
1521 | } |
1522 | ||
1523 | static inline void rbd_img_obj_request_del(struct rbd_img_request *img_request, | |
1524 | struct rbd_obj_request *obj_request) | |
1525 | { | |
1526 | rbd_assert(obj_request->which != BAD_WHICH); | |
25dcf954 | 1527 | |
37206ee5 AE |
1528 | dout("%s: img %p obj %p w=%u\n", __func__, img_request, obj_request, |
1529 | obj_request->which); | |
bf0d5f50 | 1530 | list_del(&obj_request->links); |
25dcf954 AE |
1531 | rbd_assert(img_request->obj_request_count > 0); |
1532 | img_request->obj_request_count--; | |
1533 | rbd_assert(obj_request->which == img_request->obj_request_count); | |
1534 | obj_request->which = BAD_WHICH; | |
6365d33a | 1535 | rbd_assert(obj_request_img_data_test(obj_request)); |
bf0d5f50 | 1536 | rbd_assert(obj_request->img_request == img_request); |
bf0d5f50 | 1537 | obj_request->img_request = NULL; |
25dcf954 | 1538 | obj_request->callback = NULL; |
bf0d5f50 AE |
1539 | rbd_obj_request_put(obj_request); |
1540 | } | |
1541 | ||
1542 | static bool obj_request_type_valid(enum obj_request_type type) | |
1543 | { | |
1544 | switch (type) { | |
9969ebc5 | 1545 | case OBJ_REQUEST_NODATA: |
bf0d5f50 | 1546 | case OBJ_REQUEST_BIO: |
788e2df3 | 1547 | case OBJ_REQUEST_PAGES: |
bf0d5f50 AE |
1548 | return true; |
1549 | default: | |
1550 | return false; | |
1551 | } | |
1552 | } | |
1553 | ||
bf0d5f50 AE |
1554 | static int rbd_obj_request_submit(struct ceph_osd_client *osdc, |
1555 | struct rbd_obj_request *obj_request) | |
1556 | { | |
71c20a06 | 1557 | dout("%s %p\n", __func__, obj_request); |
bf0d5f50 AE |
1558 | return ceph_osdc_start_request(osdc, obj_request->osd_req, false); |
1559 | } | |
1560 | ||
71c20a06 ID |
1561 | static void rbd_obj_request_end(struct rbd_obj_request *obj_request) |
1562 | { | |
1563 | dout("%s %p\n", __func__, obj_request); | |
1564 | ceph_osdc_cancel_request(obj_request->osd_req); | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * Wait for an object request to complete. If interrupted, cancel the | |
1569 | * underlying osd request. | |
2894e1d7 ID |
1570 | * |
1571 | * @timeout: in jiffies, 0 means "wait forever" | |
71c20a06 | 1572 | */ |
2894e1d7 ID |
1573 | static int __rbd_obj_request_wait(struct rbd_obj_request *obj_request, |
1574 | unsigned long timeout) | |
71c20a06 | 1575 | { |
2894e1d7 | 1576 | long ret; |
71c20a06 ID |
1577 | |
1578 | dout("%s %p\n", __func__, obj_request); | |
2894e1d7 ID |
1579 | ret = wait_for_completion_interruptible_timeout( |
1580 | &obj_request->completion, | |
1581 | ceph_timeout_jiffies(timeout)); | |
1582 | if (ret <= 0) { | |
1583 | if (ret == 0) | |
1584 | ret = -ETIMEDOUT; | |
71c20a06 | 1585 | rbd_obj_request_end(obj_request); |
2894e1d7 ID |
1586 | } else { |
1587 | ret = 0; | |
71c20a06 ID |
1588 | } |
1589 | ||
2894e1d7 ID |
1590 | dout("%s %p ret %d\n", __func__, obj_request, (int)ret); |
1591 | return ret; | |
1592 | } | |
1593 | ||
1594 | static int rbd_obj_request_wait(struct rbd_obj_request *obj_request) | |
1595 | { | |
1596 | return __rbd_obj_request_wait(obj_request, 0); | |
1597 | } | |
1598 | ||
1599 | static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request, | |
1600 | unsigned long timeout) | |
1601 | { | |
1602 | return __rbd_obj_request_wait(obj_request, timeout); | |
71c20a06 ID |
1603 | } |
1604 | ||
bf0d5f50 AE |
1605 | static void rbd_img_request_complete(struct rbd_img_request *img_request) |
1606 | { | |
55f27e09 | 1607 | |
37206ee5 | 1608 | dout("%s: img %p\n", __func__, img_request); |
55f27e09 AE |
1609 | |
1610 | /* | |
1611 | * If no error occurred, compute the aggregate transfer | |
1612 | * count for the image request. We could instead use | |
1613 | * atomic64_cmpxchg() to update it as each object request | |
1614 | * completes; not clear which way is better off hand. | |
1615 | */ | |
1616 | if (!img_request->result) { | |
1617 | struct rbd_obj_request *obj_request; | |
1618 | u64 xferred = 0; | |
1619 | ||
1620 | for_each_obj_request(img_request, obj_request) | |
1621 | xferred += obj_request->xferred; | |
1622 | img_request->xferred = xferred; | |
1623 | } | |
1624 | ||
bf0d5f50 AE |
1625 | if (img_request->callback) |
1626 | img_request->callback(img_request); | |
1627 | else | |
1628 | rbd_img_request_put(img_request); | |
1629 | } | |
1630 | ||
0c425248 AE |
1631 | /* |
1632 | * The default/initial value for all image request flags is 0. Each | |
1633 | * is conditionally set to 1 at image request initialization time | |
1634 | * and currently never change thereafter. | |
1635 | */ | |
1636 | static void img_request_write_set(struct rbd_img_request *img_request) | |
1637 | { | |
1638 | set_bit(IMG_REQ_WRITE, &img_request->flags); | |
1639 | smp_mb(); | |
1640 | } | |
1641 | ||
1642 | static bool img_request_write_test(struct rbd_img_request *img_request) | |
1643 | { | |
1644 | smp_mb(); | |
1645 | return test_bit(IMG_REQ_WRITE, &img_request->flags) != 0; | |
1646 | } | |
1647 | ||
90e98c52 GZ |
1648 | /* |
1649 | * Set the discard flag when the img_request is an discard request | |
1650 | */ | |
1651 | static void img_request_discard_set(struct rbd_img_request *img_request) | |
1652 | { | |
1653 | set_bit(IMG_REQ_DISCARD, &img_request->flags); | |
1654 | smp_mb(); | |
1655 | } | |
1656 | ||
1657 | static bool img_request_discard_test(struct rbd_img_request *img_request) | |
1658 | { | |
1659 | smp_mb(); | |
1660 | return test_bit(IMG_REQ_DISCARD, &img_request->flags) != 0; | |
1661 | } | |
1662 | ||
9849e986 AE |
1663 | static void img_request_child_set(struct rbd_img_request *img_request) |
1664 | { | |
1665 | set_bit(IMG_REQ_CHILD, &img_request->flags); | |
1666 | smp_mb(); | |
1667 | } | |
1668 | ||
e93f3152 AE |
1669 | static void img_request_child_clear(struct rbd_img_request *img_request) |
1670 | { | |
1671 | clear_bit(IMG_REQ_CHILD, &img_request->flags); | |
1672 | smp_mb(); | |
1673 | } | |
1674 | ||
9849e986 AE |
1675 | static bool img_request_child_test(struct rbd_img_request *img_request) |
1676 | { | |
1677 | smp_mb(); | |
1678 | return test_bit(IMG_REQ_CHILD, &img_request->flags) != 0; | |
1679 | } | |
1680 | ||
d0b2e944 AE |
1681 | static void img_request_layered_set(struct rbd_img_request *img_request) |
1682 | { | |
1683 | set_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1684 | smp_mb(); | |
1685 | } | |
1686 | ||
a2acd00e AE |
1687 | static void img_request_layered_clear(struct rbd_img_request *img_request) |
1688 | { | |
1689 | clear_bit(IMG_REQ_LAYERED, &img_request->flags); | |
1690 | smp_mb(); | |
1691 | } | |
1692 | ||
d0b2e944 AE |
1693 | static bool img_request_layered_test(struct rbd_img_request *img_request) |
1694 | { | |
1695 | smp_mb(); | |
1696 | return test_bit(IMG_REQ_LAYERED, &img_request->flags) != 0; | |
1697 | } | |
1698 | ||
3b434a2a JD |
1699 | static enum obj_operation_type |
1700 | rbd_img_request_op_type(struct rbd_img_request *img_request) | |
1701 | { | |
1702 | if (img_request_write_test(img_request)) | |
1703 | return OBJ_OP_WRITE; | |
1704 | else if (img_request_discard_test(img_request)) | |
1705 | return OBJ_OP_DISCARD; | |
1706 | else | |
1707 | return OBJ_OP_READ; | |
1708 | } | |
1709 | ||
6e2a4505 AE |
1710 | static void |
1711 | rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request) | |
1712 | { | |
b9434c5b AE |
1713 | u64 xferred = obj_request->xferred; |
1714 | u64 length = obj_request->length; | |
1715 | ||
6e2a4505 AE |
1716 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, |
1717 | obj_request, obj_request->img_request, obj_request->result, | |
b9434c5b | 1718 | xferred, length); |
6e2a4505 | 1719 | /* |
17c1cc1d JD |
1720 | * ENOENT means a hole in the image. We zero-fill the entire |
1721 | * length of the request. A short read also implies zero-fill | |
1722 | * to the end of the request. An error requires the whole | |
1723 | * length of the request to be reported finished with an error | |
1724 | * to the block layer. In each case we update the xferred | |
1725 | * count to indicate the whole request was satisfied. | |
6e2a4505 | 1726 | */ |
b9434c5b | 1727 | rbd_assert(obj_request->type != OBJ_REQUEST_NODATA); |
6e2a4505 | 1728 | if (obj_request->result == -ENOENT) { |
b9434c5b AE |
1729 | if (obj_request->type == OBJ_REQUEST_BIO) |
1730 | zero_bio_chain(obj_request->bio_list, 0); | |
1731 | else | |
1732 | zero_pages(obj_request->pages, 0, length); | |
6e2a4505 | 1733 | obj_request->result = 0; |
b9434c5b AE |
1734 | } else if (xferred < length && !obj_request->result) { |
1735 | if (obj_request->type == OBJ_REQUEST_BIO) | |
1736 | zero_bio_chain(obj_request->bio_list, xferred); | |
1737 | else | |
1738 | zero_pages(obj_request->pages, xferred, length); | |
6e2a4505 | 1739 | } |
17c1cc1d | 1740 | obj_request->xferred = length; |
6e2a4505 AE |
1741 | obj_request_done_set(obj_request); |
1742 | } | |
1743 | ||
bf0d5f50 AE |
1744 | static void rbd_obj_request_complete(struct rbd_obj_request *obj_request) |
1745 | { | |
37206ee5 AE |
1746 | dout("%s: obj %p cb %p\n", __func__, obj_request, |
1747 | obj_request->callback); | |
bf0d5f50 AE |
1748 | if (obj_request->callback) |
1749 | obj_request->callback(obj_request); | |
788e2df3 AE |
1750 | else |
1751 | complete_all(&obj_request->completion); | |
bf0d5f50 AE |
1752 | } |
1753 | ||
c47f9371 | 1754 | static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request) |
39bf2c5d AE |
1755 | { |
1756 | dout("%s: obj %p\n", __func__, obj_request); | |
1757 | obj_request_done_set(obj_request); | |
1758 | } | |
1759 | ||
c47f9371 | 1760 | static void rbd_osd_read_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1761 | { |
57acbaa7 | 1762 | struct rbd_img_request *img_request = NULL; |
a9e8ba2c | 1763 | struct rbd_device *rbd_dev = NULL; |
57acbaa7 AE |
1764 | bool layered = false; |
1765 | ||
1766 | if (obj_request_img_data_test(obj_request)) { | |
1767 | img_request = obj_request->img_request; | |
1768 | layered = img_request && img_request_layered_test(img_request); | |
a9e8ba2c | 1769 | rbd_dev = img_request->rbd_dev; |
57acbaa7 | 1770 | } |
8b3e1a56 AE |
1771 | |
1772 | dout("%s: obj %p img %p result %d %llu/%llu\n", __func__, | |
1773 | obj_request, img_request, obj_request->result, | |
1774 | obj_request->xferred, obj_request->length); | |
a9e8ba2c AE |
1775 | if (layered && obj_request->result == -ENOENT && |
1776 | obj_request->img_offset < rbd_dev->parent_overlap) | |
8b3e1a56 AE |
1777 | rbd_img_parent_read(obj_request); |
1778 | else if (img_request) | |
6e2a4505 AE |
1779 | rbd_img_obj_request_read_callback(obj_request); |
1780 | else | |
1781 | obj_request_done_set(obj_request); | |
bf0d5f50 AE |
1782 | } |
1783 | ||
c47f9371 | 1784 | static void rbd_osd_write_callback(struct rbd_obj_request *obj_request) |
bf0d5f50 | 1785 | { |
1b83bef2 SW |
1786 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, |
1787 | obj_request->result, obj_request->length); | |
1788 | /* | |
8b3e1a56 AE |
1789 | * There is no such thing as a successful short write. Set |
1790 | * it to our originally-requested length. | |
1b83bef2 SW |
1791 | */ |
1792 | obj_request->xferred = obj_request->length; | |
07741308 | 1793 | obj_request_done_set(obj_request); |
bf0d5f50 AE |
1794 | } |
1795 | ||
90e98c52 GZ |
1796 | static void rbd_osd_discard_callback(struct rbd_obj_request *obj_request) |
1797 | { | |
1798 | dout("%s: obj %p result %d %llu\n", __func__, obj_request, | |
1799 | obj_request->result, obj_request->length); | |
1800 | /* | |
1801 | * There is no such thing as a successful short discard. Set | |
1802 | * it to our originally-requested length. | |
1803 | */ | |
1804 | obj_request->xferred = obj_request->length; | |
d0265de7 JD |
1805 | /* discarding a non-existent object is not a problem */ |
1806 | if (obj_request->result == -ENOENT) | |
1807 | obj_request->result = 0; | |
90e98c52 GZ |
1808 | obj_request_done_set(obj_request); |
1809 | } | |
1810 | ||
fbfab539 AE |
1811 | /* |
1812 | * For a simple stat call there's nothing to do. We'll do more if | |
1813 | * this is part of a write sequence for a layered image. | |
1814 | */ | |
c47f9371 | 1815 | static void rbd_osd_stat_callback(struct rbd_obj_request *obj_request) |
fbfab539 | 1816 | { |
37206ee5 | 1817 | dout("%s: obj %p\n", __func__, obj_request); |
fbfab539 AE |
1818 | obj_request_done_set(obj_request); |
1819 | } | |
1820 | ||
2761713d ID |
1821 | static void rbd_osd_call_callback(struct rbd_obj_request *obj_request) |
1822 | { | |
1823 | dout("%s: obj %p\n", __func__, obj_request); | |
1824 | ||
1825 | if (obj_request_img_data_test(obj_request)) | |
1826 | rbd_osd_copyup_callback(obj_request); | |
1827 | else | |
1828 | obj_request_done_set(obj_request); | |
1829 | } | |
1830 | ||
bf0d5f50 AE |
1831 | static void rbd_osd_req_callback(struct ceph_osd_request *osd_req, |
1832 | struct ceph_msg *msg) | |
1833 | { | |
1834 | struct rbd_obj_request *obj_request = osd_req->r_priv; | |
bf0d5f50 AE |
1835 | u16 opcode; |
1836 | ||
37206ee5 | 1837 | dout("%s: osd_req %p msg %p\n", __func__, osd_req, msg); |
bf0d5f50 | 1838 | rbd_assert(osd_req == obj_request->osd_req); |
57acbaa7 AE |
1839 | if (obj_request_img_data_test(obj_request)) { |
1840 | rbd_assert(obj_request->img_request); | |
1841 | rbd_assert(obj_request->which != BAD_WHICH); | |
1842 | } else { | |
1843 | rbd_assert(obj_request->which == BAD_WHICH); | |
1844 | } | |
bf0d5f50 | 1845 | |
1b83bef2 SW |
1846 | if (osd_req->r_result < 0) |
1847 | obj_request->result = osd_req->r_result; | |
bf0d5f50 | 1848 | |
c47f9371 AE |
1849 | /* |
1850 | * We support a 64-bit length, but ultimately it has to be | |
7ad18afa CH |
1851 | * passed to the block layer, which just supports a 32-bit |
1852 | * length field. | |
c47f9371 | 1853 | */ |
7665d85b | 1854 | obj_request->xferred = osd_req->r_ops[0].outdata_len; |
8b3e1a56 | 1855 | rbd_assert(obj_request->xferred < (u64)UINT_MAX); |
0ccd5926 | 1856 | |
79528734 | 1857 | opcode = osd_req->r_ops[0].op; |
bf0d5f50 AE |
1858 | switch (opcode) { |
1859 | case CEPH_OSD_OP_READ: | |
c47f9371 | 1860 | rbd_osd_read_callback(obj_request); |
bf0d5f50 | 1861 | break; |
0ccd5926 | 1862 | case CEPH_OSD_OP_SETALLOCHINT: |
e30b7577 ID |
1863 | rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE || |
1864 | osd_req->r_ops[1].op == CEPH_OSD_OP_WRITEFULL); | |
0ccd5926 | 1865 | /* fall through */ |
bf0d5f50 | 1866 | case CEPH_OSD_OP_WRITE: |
e30b7577 | 1867 | case CEPH_OSD_OP_WRITEFULL: |
c47f9371 | 1868 | rbd_osd_write_callback(obj_request); |
bf0d5f50 | 1869 | break; |
fbfab539 | 1870 | case CEPH_OSD_OP_STAT: |
c47f9371 | 1871 | rbd_osd_stat_callback(obj_request); |
fbfab539 | 1872 | break; |
90e98c52 GZ |
1873 | case CEPH_OSD_OP_DELETE: |
1874 | case CEPH_OSD_OP_TRUNCATE: | |
1875 | case CEPH_OSD_OP_ZERO: | |
1876 | rbd_osd_discard_callback(obj_request); | |
1877 | break; | |
36be9a76 | 1878 | case CEPH_OSD_OP_CALL: |
2761713d ID |
1879 | rbd_osd_call_callback(obj_request); |
1880 | break; | |
b8d70035 | 1881 | case CEPH_OSD_OP_NOTIFY_ACK: |
9969ebc5 | 1882 | case CEPH_OSD_OP_WATCH: |
c47f9371 | 1883 | rbd_osd_trivial_callback(obj_request); |
9969ebc5 | 1884 | break; |
bf0d5f50 | 1885 | default: |
9584d508 | 1886 | rbd_warn(NULL, "%s: unsupported op %hu", |
bf0d5f50 AE |
1887 | obj_request->object_name, (unsigned short) opcode); |
1888 | break; | |
1889 | } | |
1890 | ||
07741308 | 1891 | if (obj_request_done_test(obj_request)) |
bf0d5f50 AE |
1892 | rbd_obj_request_complete(obj_request); |
1893 | } | |
1894 | ||
9d4df01f | 1895 | static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request) |
430c28c3 AE |
1896 | { |
1897 | struct rbd_img_request *img_request = obj_request->img_request; | |
8c042b0d | 1898 | struct ceph_osd_request *osd_req = obj_request->osd_req; |
9d4df01f | 1899 | u64 snap_id; |
430c28c3 | 1900 | |
8c042b0d | 1901 | rbd_assert(osd_req != NULL); |
430c28c3 | 1902 | |
9d4df01f | 1903 | snap_id = img_request ? img_request->snap_id : CEPH_NOSNAP; |
8c042b0d | 1904 | ceph_osdc_build_request(osd_req, obj_request->offset, |
9d4df01f AE |
1905 | NULL, snap_id, NULL); |
1906 | } | |
1907 | ||
1908 | static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) | |
1909 | { | |
1910 | struct rbd_img_request *img_request = obj_request->img_request; | |
1911 | struct ceph_osd_request *osd_req = obj_request->osd_req; | |
1912 | struct ceph_snap_context *snapc; | |
1913 | struct timespec mtime = CURRENT_TIME; | |
1914 | ||
1915 | rbd_assert(osd_req != NULL); | |
1916 | ||
1917 | snapc = img_request ? img_request->snapc : NULL; | |
1918 | ceph_osdc_build_request(osd_req, obj_request->offset, | |
1919 | snapc, CEPH_NOSNAP, &mtime); | |
430c28c3 AE |
1920 | } |
1921 | ||
0ccd5926 ID |
1922 | /* |
1923 | * Create an osd request. A read request has one osd op (read). | |
1924 | * A write request has either one (watch) or two (hint+write) osd ops. | |
1925 | * (All rbd data writes are prefixed with an allocation hint op, but | |
1926 | * technically osd watch is a write request, hence this distinction.) | |
1927 | */ | |
bf0d5f50 AE |
1928 | static struct ceph_osd_request *rbd_osd_req_create( |
1929 | struct rbd_device *rbd_dev, | |
6d2940c8 | 1930 | enum obj_operation_type op_type, |
deb236b3 | 1931 | unsigned int num_ops, |
430c28c3 | 1932 | struct rbd_obj_request *obj_request) |
bf0d5f50 | 1933 | { |
bf0d5f50 AE |
1934 | struct ceph_snap_context *snapc = NULL; |
1935 | struct ceph_osd_client *osdc; | |
1936 | struct ceph_osd_request *osd_req; | |
bf0d5f50 | 1937 | |
90e98c52 GZ |
1938 | if (obj_request_img_data_test(obj_request) && |
1939 | (op_type == OBJ_OP_DISCARD || op_type == OBJ_OP_WRITE)) { | |
6365d33a | 1940 | struct rbd_img_request *img_request = obj_request->img_request; |
90e98c52 GZ |
1941 | if (op_type == OBJ_OP_WRITE) { |
1942 | rbd_assert(img_request_write_test(img_request)); | |
1943 | } else { | |
1944 | rbd_assert(img_request_discard_test(img_request)); | |
1945 | } | |
6d2940c8 | 1946 | snapc = img_request->snapc; |
bf0d5f50 AE |
1947 | } |
1948 | ||
6d2940c8 | 1949 | rbd_assert(num_ops == 1 || ((op_type == OBJ_OP_WRITE) && num_ops == 2)); |
deb236b3 ID |
1950 | |
1951 | /* Allocate and initialize the request, for the num_ops ops */ | |
bf0d5f50 AE |
1952 | |
1953 | osdc = &rbd_dev->rbd_client->client->osdc; | |
deb236b3 | 1954 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_ops, false, |
2224d879 | 1955 | GFP_NOIO); |
bf0d5f50 AE |
1956 | if (!osd_req) |
1957 | return NULL; /* ENOMEM */ | |
bf0d5f50 | 1958 | |
90e98c52 | 1959 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) |
bf0d5f50 | 1960 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; |
430c28c3 | 1961 | else |
bf0d5f50 | 1962 | osd_req->r_flags = CEPH_OSD_FLAG_READ; |
bf0d5f50 AE |
1963 | |
1964 | osd_req->r_callback = rbd_osd_req_callback; | |
1965 | osd_req->r_priv = obj_request; | |
1966 | ||
3c972c95 ID |
1967 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
1968 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
bf0d5f50 | 1969 | |
bf0d5f50 AE |
1970 | return osd_req; |
1971 | } | |
1972 | ||
0eefd470 | 1973 | /* |
d3246fb0 JD |
1974 | * Create a copyup osd request based on the information in the object |
1975 | * request supplied. A copyup request has two or three osd ops, a | |
1976 | * copyup method call, potentially a hint op, and a write or truncate | |
1977 | * or zero op. | |
0eefd470 AE |
1978 | */ |
1979 | static struct ceph_osd_request * | |
1980 | rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) | |
1981 | { | |
1982 | struct rbd_img_request *img_request; | |
1983 | struct ceph_snap_context *snapc; | |
1984 | struct rbd_device *rbd_dev; | |
1985 | struct ceph_osd_client *osdc; | |
1986 | struct ceph_osd_request *osd_req; | |
d3246fb0 | 1987 | int num_osd_ops = 3; |
0eefd470 AE |
1988 | |
1989 | rbd_assert(obj_request_img_data_test(obj_request)); | |
1990 | img_request = obj_request->img_request; | |
1991 | rbd_assert(img_request); | |
d3246fb0 JD |
1992 | rbd_assert(img_request_write_test(img_request) || |
1993 | img_request_discard_test(img_request)); | |
0eefd470 | 1994 | |
d3246fb0 JD |
1995 | if (img_request_discard_test(img_request)) |
1996 | num_osd_ops = 2; | |
1997 | ||
1998 | /* Allocate and initialize the request, for all the ops */ | |
0eefd470 AE |
1999 | |
2000 | snapc = img_request->snapc; | |
2001 | rbd_dev = img_request->rbd_dev; | |
2002 | osdc = &rbd_dev->rbd_client->client->osdc; | |
d3246fb0 | 2003 | osd_req = ceph_osdc_alloc_request(osdc, snapc, num_osd_ops, |
2224d879 | 2004 | false, GFP_NOIO); |
0eefd470 AE |
2005 | if (!osd_req) |
2006 | return NULL; /* ENOMEM */ | |
2007 | ||
2008 | osd_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; | |
2009 | osd_req->r_callback = rbd_osd_req_callback; | |
2010 | osd_req->r_priv = obj_request; | |
2011 | ||
3c972c95 ID |
2012 | osd_req->r_base_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout); |
2013 | ceph_oid_set_name(&osd_req->r_base_oid, obj_request->object_name); | |
0eefd470 | 2014 | |
0eefd470 AE |
2015 | return osd_req; |
2016 | } | |
2017 | ||
2018 | ||
bf0d5f50 AE |
2019 | static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req) |
2020 | { | |
2021 | ceph_osdc_put_request(osd_req); | |
2022 | } | |
2023 | ||
2024 | /* object_name is assumed to be a non-null pointer and NUL-terminated */ | |
2025 | ||
2026 | static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, | |
2027 | u64 offset, u64 length, | |
2028 | enum obj_request_type type) | |
2029 | { | |
2030 | struct rbd_obj_request *obj_request; | |
2031 | size_t size; | |
2032 | char *name; | |
2033 | ||
2034 | rbd_assert(obj_request_type_valid(type)); | |
2035 | ||
2036 | size = strlen(object_name) + 1; | |
5a60e876 | 2037 | name = kmalloc(size, GFP_NOIO); |
f907ad55 | 2038 | if (!name) |
bf0d5f50 AE |
2039 | return NULL; |
2040 | ||
5a60e876 | 2041 | obj_request = kmem_cache_zalloc(rbd_obj_request_cache, GFP_NOIO); |
f907ad55 AE |
2042 | if (!obj_request) { |
2043 | kfree(name); | |
2044 | return NULL; | |
2045 | } | |
2046 | ||
bf0d5f50 AE |
2047 | obj_request->object_name = memcpy(name, object_name, size); |
2048 | obj_request->offset = offset; | |
2049 | obj_request->length = length; | |
926f9b3f | 2050 | obj_request->flags = 0; |
bf0d5f50 AE |
2051 | obj_request->which = BAD_WHICH; |
2052 | obj_request->type = type; | |
2053 | INIT_LIST_HEAD(&obj_request->links); | |
788e2df3 | 2054 | init_completion(&obj_request->completion); |
bf0d5f50 AE |
2055 | kref_init(&obj_request->kref); |
2056 | ||
37206ee5 AE |
2057 | dout("%s: \"%s\" %llu/%llu %d -> obj %p\n", __func__, object_name, |
2058 | offset, length, (int)type, obj_request); | |
2059 | ||
bf0d5f50 AE |
2060 | return obj_request; |
2061 | } | |
2062 | ||
2063 | static void rbd_obj_request_destroy(struct kref *kref) | |
2064 | { | |
2065 | struct rbd_obj_request *obj_request; | |
2066 | ||
2067 | obj_request = container_of(kref, struct rbd_obj_request, kref); | |
2068 | ||
37206ee5 AE |
2069 | dout("%s: obj %p\n", __func__, obj_request); |
2070 | ||
bf0d5f50 AE |
2071 | rbd_assert(obj_request->img_request == NULL); |
2072 | rbd_assert(obj_request->which == BAD_WHICH); | |
2073 | ||
2074 | if (obj_request->osd_req) | |
2075 | rbd_osd_req_destroy(obj_request->osd_req); | |
2076 | ||
2077 | rbd_assert(obj_request_type_valid(obj_request->type)); | |
2078 | switch (obj_request->type) { | |
9969ebc5 AE |
2079 | case OBJ_REQUEST_NODATA: |
2080 | break; /* Nothing to do */ | |
bf0d5f50 AE |
2081 | case OBJ_REQUEST_BIO: |
2082 | if (obj_request->bio_list) | |
2083 | bio_chain_put(obj_request->bio_list); | |
2084 | break; | |
788e2df3 AE |
2085 | case OBJ_REQUEST_PAGES: |
2086 | if (obj_request->pages) | |
2087 | ceph_release_page_vector(obj_request->pages, | |
2088 | obj_request->page_count); | |
2089 | break; | |
bf0d5f50 AE |
2090 | } |
2091 | ||
f907ad55 | 2092 | kfree(obj_request->object_name); |
868311b1 AE |
2093 | obj_request->object_name = NULL; |
2094 | kmem_cache_free(rbd_obj_request_cache, obj_request); | |
bf0d5f50 AE |
2095 | } |
2096 | ||
fb65d228 AE |
2097 | /* It's OK to call this for a device with no parent */ |
2098 | ||
2099 | static void rbd_spec_put(struct rbd_spec *spec); | |
2100 | static void rbd_dev_unparent(struct rbd_device *rbd_dev) | |
2101 | { | |
2102 | rbd_dev_remove_parent(rbd_dev); | |
2103 | rbd_spec_put(rbd_dev->parent_spec); | |
2104 | rbd_dev->parent_spec = NULL; | |
2105 | rbd_dev->parent_overlap = 0; | |
2106 | } | |
2107 | ||
a2acd00e AE |
2108 | /* |
2109 | * Parent image reference counting is used to determine when an | |
2110 | * image's parent fields can be safely torn down--after there are no | |
2111 | * more in-flight requests to the parent image. When the last | |
2112 | * reference is dropped, cleaning them up is safe. | |
2113 | */ | |
2114 | static void rbd_dev_parent_put(struct rbd_device *rbd_dev) | |
2115 | { | |
2116 | int counter; | |
2117 | ||
2118 | if (!rbd_dev->parent_spec) | |
2119 | return; | |
2120 | ||
2121 | counter = atomic_dec_return_safe(&rbd_dev->parent_ref); | |
2122 | if (counter > 0) | |
2123 | return; | |
2124 | ||
2125 | /* Last reference; clean up parent data structures */ | |
2126 | ||
2127 | if (!counter) | |
2128 | rbd_dev_unparent(rbd_dev); | |
2129 | else | |
9584d508 | 2130 | rbd_warn(rbd_dev, "parent reference underflow"); |
a2acd00e AE |
2131 | } |
2132 | ||
2133 | /* | |
2134 | * If an image has a non-zero parent overlap, get a reference to its | |
2135 | * parent. | |
2136 | * | |
2137 | * Returns true if the rbd device has a parent with a non-zero | |
2138 | * overlap and a reference for it was successfully taken, or | |
2139 | * false otherwise. | |
2140 | */ | |
2141 | static bool rbd_dev_parent_get(struct rbd_device *rbd_dev) | |
2142 | { | |
ae43e9d0 | 2143 | int counter = 0; |
a2acd00e AE |
2144 | |
2145 | if (!rbd_dev->parent_spec) | |
2146 | return false; | |
2147 | ||
ae43e9d0 ID |
2148 | down_read(&rbd_dev->header_rwsem); |
2149 | if (rbd_dev->parent_overlap) | |
2150 | counter = atomic_inc_return_safe(&rbd_dev->parent_ref); | |
2151 | up_read(&rbd_dev->header_rwsem); | |
a2acd00e AE |
2152 | |
2153 | if (counter < 0) | |
9584d508 | 2154 | rbd_warn(rbd_dev, "parent reference overflow"); |
a2acd00e | 2155 | |
ae43e9d0 | 2156 | return counter > 0; |
a2acd00e AE |
2157 | } |
2158 | ||
bf0d5f50 AE |
2159 | /* |
2160 | * Caller is responsible for filling in the list of object requests | |
2161 | * that comprises the image request, and the Linux request pointer | |
2162 | * (if there is one). | |
2163 | */ | |
cc344fa1 AE |
2164 | static struct rbd_img_request *rbd_img_request_create( |
2165 | struct rbd_device *rbd_dev, | |
bf0d5f50 | 2166 | u64 offset, u64 length, |
6d2940c8 | 2167 | enum obj_operation_type op_type, |
4e752f0a | 2168 | struct ceph_snap_context *snapc) |
bf0d5f50 AE |
2169 | { |
2170 | struct rbd_img_request *img_request; | |
bf0d5f50 | 2171 | |
7a716aac | 2172 | img_request = kmem_cache_alloc(rbd_img_request_cache, GFP_NOIO); |
bf0d5f50 AE |
2173 | if (!img_request) |
2174 | return NULL; | |
2175 | ||
bf0d5f50 AE |
2176 | img_request->rq = NULL; |
2177 | img_request->rbd_dev = rbd_dev; | |
2178 | img_request->offset = offset; | |
2179 | img_request->length = length; | |
0c425248 | 2180 | img_request->flags = 0; |
90e98c52 GZ |
2181 | if (op_type == OBJ_OP_DISCARD) { |
2182 | img_request_discard_set(img_request); | |
2183 | img_request->snapc = snapc; | |
2184 | } else if (op_type == OBJ_OP_WRITE) { | |
0c425248 | 2185 | img_request_write_set(img_request); |
4e752f0a | 2186 | img_request->snapc = snapc; |
0c425248 | 2187 | } else { |
bf0d5f50 | 2188 | img_request->snap_id = rbd_dev->spec->snap_id; |
0c425248 | 2189 | } |
a2acd00e | 2190 | if (rbd_dev_parent_get(rbd_dev)) |
d0b2e944 | 2191 | img_request_layered_set(img_request); |
bf0d5f50 AE |
2192 | spin_lock_init(&img_request->completion_lock); |
2193 | img_request->next_completion = 0; | |
2194 | img_request->callback = NULL; | |
a5a337d4 | 2195 | img_request->result = 0; |
bf0d5f50 AE |
2196 | img_request->obj_request_count = 0; |
2197 | INIT_LIST_HEAD(&img_request->obj_requests); | |
2198 | kref_init(&img_request->kref); | |
2199 | ||
37206ee5 | 2200 | dout("%s: rbd_dev %p %s %llu/%llu -> img %p\n", __func__, rbd_dev, |
6d2940c8 | 2201 | obj_op_name(op_type), offset, length, img_request); |
37206ee5 | 2202 | |
bf0d5f50 AE |
2203 | return img_request; |
2204 | } | |
2205 | ||
2206 | static void rbd_img_request_destroy(struct kref *kref) | |
2207 | { | |
2208 | struct rbd_img_request *img_request; | |
2209 | struct rbd_obj_request *obj_request; | |
2210 | struct rbd_obj_request *next_obj_request; | |
2211 | ||
2212 | img_request = container_of(kref, struct rbd_img_request, kref); | |
2213 | ||
37206ee5 AE |
2214 | dout("%s: img %p\n", __func__, img_request); |
2215 | ||
bf0d5f50 AE |
2216 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) |
2217 | rbd_img_obj_request_del(img_request, obj_request); | |
25dcf954 | 2218 | rbd_assert(img_request->obj_request_count == 0); |
bf0d5f50 | 2219 | |
a2acd00e AE |
2220 | if (img_request_layered_test(img_request)) { |
2221 | img_request_layered_clear(img_request); | |
2222 | rbd_dev_parent_put(img_request->rbd_dev); | |
2223 | } | |
2224 | ||
bef95455 JD |
2225 | if (img_request_write_test(img_request) || |
2226 | img_request_discard_test(img_request)) | |
812164f8 | 2227 | ceph_put_snap_context(img_request->snapc); |
bf0d5f50 | 2228 | |
1c2a9dfe | 2229 | kmem_cache_free(rbd_img_request_cache, img_request); |
bf0d5f50 AE |
2230 | } |
2231 | ||
e93f3152 AE |
2232 | static struct rbd_img_request *rbd_parent_request_create( |
2233 | struct rbd_obj_request *obj_request, | |
2234 | u64 img_offset, u64 length) | |
2235 | { | |
2236 | struct rbd_img_request *parent_request; | |
2237 | struct rbd_device *rbd_dev; | |
2238 | ||
2239 | rbd_assert(obj_request->img_request); | |
2240 | rbd_dev = obj_request->img_request->rbd_dev; | |
2241 | ||
4e752f0a | 2242 | parent_request = rbd_img_request_create(rbd_dev->parent, img_offset, |
6d2940c8 | 2243 | length, OBJ_OP_READ, NULL); |
e93f3152 AE |
2244 | if (!parent_request) |
2245 | return NULL; | |
2246 | ||
2247 | img_request_child_set(parent_request); | |
2248 | rbd_obj_request_get(obj_request); | |
2249 | parent_request->obj_request = obj_request; | |
2250 | ||
2251 | return parent_request; | |
2252 | } | |
2253 | ||
2254 | static void rbd_parent_request_destroy(struct kref *kref) | |
2255 | { | |
2256 | struct rbd_img_request *parent_request; | |
2257 | struct rbd_obj_request *orig_request; | |
2258 | ||
2259 | parent_request = container_of(kref, struct rbd_img_request, kref); | |
2260 | orig_request = parent_request->obj_request; | |
2261 | ||
2262 | parent_request->obj_request = NULL; | |
2263 | rbd_obj_request_put(orig_request); | |
2264 | img_request_child_clear(parent_request); | |
2265 | ||
2266 | rbd_img_request_destroy(kref); | |
2267 | } | |
2268 | ||
1217857f AE |
2269 | static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request) |
2270 | { | |
6365d33a | 2271 | struct rbd_img_request *img_request; |
1217857f AE |
2272 | unsigned int xferred; |
2273 | int result; | |
8b3e1a56 | 2274 | bool more; |
1217857f | 2275 | |
6365d33a AE |
2276 | rbd_assert(obj_request_img_data_test(obj_request)); |
2277 | img_request = obj_request->img_request; | |
2278 | ||
1217857f AE |
2279 | rbd_assert(obj_request->xferred <= (u64)UINT_MAX); |
2280 | xferred = (unsigned int)obj_request->xferred; | |
2281 | result = obj_request->result; | |
2282 | if (result) { | |
2283 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
6d2940c8 GZ |
2284 | enum obj_operation_type op_type; |
2285 | ||
90e98c52 GZ |
2286 | if (img_request_discard_test(img_request)) |
2287 | op_type = OBJ_OP_DISCARD; | |
2288 | else if (img_request_write_test(img_request)) | |
2289 | op_type = OBJ_OP_WRITE; | |
2290 | else | |
2291 | op_type = OBJ_OP_READ; | |
1217857f | 2292 | |
9584d508 | 2293 | rbd_warn(rbd_dev, "%s %llx at %llx (%llx)", |
6d2940c8 GZ |
2294 | obj_op_name(op_type), obj_request->length, |
2295 | obj_request->img_offset, obj_request->offset); | |
9584d508 | 2296 | rbd_warn(rbd_dev, " result %d xferred %x", |
1217857f AE |
2297 | result, xferred); |
2298 | if (!img_request->result) | |
2299 | img_request->result = result; | |
082a75da ID |
2300 | /* |
2301 | * Need to end I/O on the entire obj_request worth of | |
2302 | * bytes in case of error. | |
2303 | */ | |
2304 | xferred = obj_request->length; | |
1217857f AE |
2305 | } |
2306 | ||
f1a4739f AE |
2307 | /* Image object requests don't own their page array */ |
2308 | ||
2309 | if (obj_request->type == OBJ_REQUEST_PAGES) { | |
2310 | obj_request->pages = NULL; | |
2311 | obj_request->page_count = 0; | |
2312 | } | |
2313 | ||
8b3e1a56 AE |
2314 | if (img_request_child_test(img_request)) { |
2315 | rbd_assert(img_request->obj_request != NULL); | |
2316 | more = obj_request->which < img_request->obj_request_count - 1; | |
2317 | } else { | |
2318 | rbd_assert(img_request->rq != NULL); | |
7ad18afa CH |
2319 | |
2320 | more = blk_update_request(img_request->rq, result, xferred); | |
2321 | if (!more) | |
2322 | __blk_mq_end_request(img_request->rq, result); | |
8b3e1a56 AE |
2323 | } |
2324 | ||
2325 | return more; | |
1217857f AE |
2326 | } |
2327 | ||
2169238d AE |
2328 | static void rbd_img_obj_callback(struct rbd_obj_request *obj_request) |
2329 | { | |
2330 | struct rbd_img_request *img_request; | |
2331 | u32 which = obj_request->which; | |
2332 | bool more = true; | |
2333 | ||
6365d33a | 2334 | rbd_assert(obj_request_img_data_test(obj_request)); |
2169238d AE |
2335 | img_request = obj_request->img_request; |
2336 | ||
2337 | dout("%s: img %p obj %p\n", __func__, img_request, obj_request); | |
2338 | rbd_assert(img_request != NULL); | |
2169238d AE |
2339 | rbd_assert(img_request->obj_request_count > 0); |
2340 | rbd_assert(which != BAD_WHICH); | |
2341 | rbd_assert(which < img_request->obj_request_count); | |
2169238d AE |
2342 | |
2343 | spin_lock_irq(&img_request->completion_lock); | |
2344 | if (which != img_request->next_completion) | |
2345 | goto out; | |
2346 | ||
2347 | for_each_obj_request_from(img_request, obj_request) { | |
2169238d AE |
2348 | rbd_assert(more); |
2349 | rbd_assert(which < img_request->obj_request_count); | |
2350 | ||
2351 | if (!obj_request_done_test(obj_request)) | |
2352 | break; | |
1217857f | 2353 | more = rbd_img_obj_end_request(obj_request); |
2169238d AE |
2354 | which++; |
2355 | } | |
2356 | ||
2357 | rbd_assert(more ^ (which == img_request->obj_request_count)); | |
2358 | img_request->next_completion = which; | |
2359 | out: | |
2360 | spin_unlock_irq(&img_request->completion_lock); | |
0f2d5be7 | 2361 | rbd_img_request_put(img_request); |
2169238d AE |
2362 | |
2363 | if (!more) | |
2364 | rbd_img_request_complete(img_request); | |
2365 | } | |
2366 | ||
3b434a2a JD |
2367 | /* |
2368 | * Add individual osd ops to the given ceph_osd_request and prepare | |
2369 | * them for submission. num_ops is the current number of | |
2370 | * osd operations already to the object request. | |
2371 | */ | |
2372 | static void rbd_img_obj_request_fill(struct rbd_obj_request *obj_request, | |
2373 | struct ceph_osd_request *osd_request, | |
2374 | enum obj_operation_type op_type, | |
2375 | unsigned int num_ops) | |
2376 | { | |
2377 | struct rbd_img_request *img_request = obj_request->img_request; | |
2378 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2379 | u64 object_size = rbd_obj_bytes(&rbd_dev->header); | |
2380 | u64 offset = obj_request->offset; | |
2381 | u64 length = obj_request->length; | |
2382 | u64 img_end; | |
2383 | u16 opcode; | |
2384 | ||
2385 | if (op_type == OBJ_OP_DISCARD) { | |
d3246fb0 JD |
2386 | if (!offset && length == object_size && |
2387 | (!img_request_layered_test(img_request) || | |
2388 | !obj_request_overlaps_parent(obj_request))) { | |
3b434a2a JD |
2389 | opcode = CEPH_OSD_OP_DELETE; |
2390 | } else if ((offset + length == object_size)) { | |
2391 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2392 | } else { | |
2393 | down_read(&rbd_dev->header_rwsem); | |
2394 | img_end = rbd_dev->header.image_size; | |
2395 | up_read(&rbd_dev->header_rwsem); | |
2396 | ||
2397 | if (obj_request->img_offset + length == img_end) | |
2398 | opcode = CEPH_OSD_OP_TRUNCATE; | |
2399 | else | |
2400 | opcode = CEPH_OSD_OP_ZERO; | |
2401 | } | |
2402 | } else if (op_type == OBJ_OP_WRITE) { | |
e30b7577 ID |
2403 | if (!offset && length == object_size) |
2404 | opcode = CEPH_OSD_OP_WRITEFULL; | |
2405 | else | |
2406 | opcode = CEPH_OSD_OP_WRITE; | |
3b434a2a JD |
2407 | osd_req_op_alloc_hint_init(osd_request, num_ops, |
2408 | object_size, object_size); | |
2409 | num_ops++; | |
2410 | } else { | |
2411 | opcode = CEPH_OSD_OP_READ; | |
2412 | } | |
2413 | ||
7e868b6e | 2414 | if (opcode == CEPH_OSD_OP_DELETE) |
144cba14 | 2415 | osd_req_op_init(osd_request, num_ops, opcode, 0); |
7e868b6e ID |
2416 | else |
2417 | osd_req_op_extent_init(osd_request, num_ops, opcode, | |
2418 | offset, length, 0, 0); | |
2419 | ||
3b434a2a JD |
2420 | if (obj_request->type == OBJ_REQUEST_BIO) |
2421 | osd_req_op_extent_osd_data_bio(osd_request, num_ops, | |
2422 | obj_request->bio_list, length); | |
2423 | else if (obj_request->type == OBJ_REQUEST_PAGES) | |
2424 | osd_req_op_extent_osd_data_pages(osd_request, num_ops, | |
2425 | obj_request->pages, length, | |
2426 | offset & ~PAGE_MASK, false, false); | |
2427 | ||
2428 | /* Discards are also writes */ | |
2429 | if (op_type == OBJ_OP_WRITE || op_type == OBJ_OP_DISCARD) | |
2430 | rbd_osd_req_format_write(obj_request); | |
2431 | else | |
2432 | rbd_osd_req_format_read(obj_request); | |
2433 | } | |
2434 | ||
f1a4739f AE |
2435 | /* |
2436 | * Split up an image request into one or more object requests, each | |
2437 | * to a different object. The "type" parameter indicates whether | |
2438 | * "data_desc" is the pointer to the head of a list of bio | |
2439 | * structures, or the base of a page array. In either case this | |
2440 | * function assumes data_desc describes memory sufficient to hold | |
2441 | * all data described by the image request. | |
2442 | */ | |
2443 | static int rbd_img_request_fill(struct rbd_img_request *img_request, | |
2444 | enum obj_request_type type, | |
2445 | void *data_desc) | |
bf0d5f50 AE |
2446 | { |
2447 | struct rbd_device *rbd_dev = img_request->rbd_dev; | |
2448 | struct rbd_obj_request *obj_request = NULL; | |
2449 | struct rbd_obj_request *next_obj_request; | |
a158073c | 2450 | struct bio *bio_list = NULL; |
f1a4739f | 2451 | unsigned int bio_offset = 0; |
a158073c | 2452 | struct page **pages = NULL; |
6d2940c8 | 2453 | enum obj_operation_type op_type; |
7da22d29 | 2454 | u64 img_offset; |
bf0d5f50 | 2455 | u64 resid; |
bf0d5f50 | 2456 | |
f1a4739f AE |
2457 | dout("%s: img %p type %d data_desc %p\n", __func__, img_request, |
2458 | (int)type, data_desc); | |
37206ee5 | 2459 | |
7da22d29 | 2460 | img_offset = img_request->offset; |
bf0d5f50 | 2461 | resid = img_request->length; |
4dda41d3 | 2462 | rbd_assert(resid > 0); |
3b434a2a | 2463 | op_type = rbd_img_request_op_type(img_request); |
f1a4739f AE |
2464 | |
2465 | if (type == OBJ_REQUEST_BIO) { | |
2466 | bio_list = data_desc; | |
4f024f37 KO |
2467 | rbd_assert(img_offset == |
2468 | bio_list->bi_iter.bi_sector << SECTOR_SHIFT); | |
90e98c52 | 2469 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2470 | pages = data_desc; |
2471 | } | |
2472 | ||
bf0d5f50 | 2473 | while (resid) { |
2fa12320 | 2474 | struct ceph_osd_request *osd_req; |
bf0d5f50 | 2475 | const char *object_name; |
bf0d5f50 AE |
2476 | u64 offset; |
2477 | u64 length; | |
2478 | ||
7da22d29 | 2479 | object_name = rbd_segment_name(rbd_dev, img_offset); |
bf0d5f50 AE |
2480 | if (!object_name) |
2481 | goto out_unwind; | |
7da22d29 AE |
2482 | offset = rbd_segment_offset(rbd_dev, img_offset); |
2483 | length = rbd_segment_length(rbd_dev, img_offset, resid); | |
bf0d5f50 | 2484 | obj_request = rbd_obj_request_create(object_name, |
f1a4739f | 2485 | offset, length, type); |
78c2a44a AE |
2486 | /* object request has its own copy of the object name */ |
2487 | rbd_segment_name_free(object_name); | |
bf0d5f50 AE |
2488 | if (!obj_request) |
2489 | goto out_unwind; | |
62054da6 | 2490 | |
03507db6 JD |
2491 | /* |
2492 | * set obj_request->img_request before creating the | |
2493 | * osd_request so that it gets the right snapc | |
2494 | */ | |
2495 | rbd_img_obj_request_add(img_request, obj_request); | |
bf0d5f50 | 2496 | |
f1a4739f AE |
2497 | if (type == OBJ_REQUEST_BIO) { |
2498 | unsigned int clone_size; | |
2499 | ||
2500 | rbd_assert(length <= (u64)UINT_MAX); | |
2501 | clone_size = (unsigned int)length; | |
2502 | obj_request->bio_list = | |
2503 | bio_chain_clone_range(&bio_list, | |
2504 | &bio_offset, | |
2505 | clone_size, | |
2224d879 | 2506 | GFP_NOIO); |
f1a4739f | 2507 | if (!obj_request->bio_list) |
62054da6 | 2508 | goto out_unwind; |
90e98c52 | 2509 | } else if (type == OBJ_REQUEST_PAGES) { |
f1a4739f AE |
2510 | unsigned int page_count; |
2511 | ||
2512 | obj_request->pages = pages; | |
2513 | page_count = (u32)calc_pages_for(offset, length); | |
2514 | obj_request->page_count = page_count; | |
2515 | if ((offset + length) & ~PAGE_MASK) | |
2516 | page_count--; /* more on last page */ | |
2517 | pages += page_count; | |
2518 | } | |
bf0d5f50 | 2519 | |
6d2940c8 GZ |
2520 | osd_req = rbd_osd_req_create(rbd_dev, op_type, |
2521 | (op_type == OBJ_OP_WRITE) ? 2 : 1, | |
2522 | obj_request); | |
2fa12320 | 2523 | if (!osd_req) |
62054da6 | 2524 | goto out_unwind; |
3b434a2a | 2525 | |
2fa12320 | 2526 | obj_request->osd_req = osd_req; |
2169238d | 2527 | obj_request->callback = rbd_img_obj_callback; |
3b434a2a | 2528 | obj_request->img_offset = img_offset; |
9d4df01f | 2529 | |
3b434a2a | 2530 | rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0); |
430c28c3 | 2531 | |
3b434a2a | 2532 | rbd_img_request_get(img_request); |
bf0d5f50 | 2533 | |
7da22d29 | 2534 | img_offset += length; |
bf0d5f50 AE |
2535 | resid -= length; |
2536 | } | |
2537 | ||
2538 | return 0; | |
2539 | ||
bf0d5f50 AE |
2540 | out_unwind: |
2541 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) | |
42dd037c | 2542 | rbd_img_obj_request_del(img_request, obj_request); |
bf0d5f50 AE |
2543 | |
2544 | return -ENOMEM; | |
2545 | } | |
2546 | ||
0eefd470 | 2547 | static void |
2761713d | 2548 | rbd_osd_copyup_callback(struct rbd_obj_request *obj_request) |
0eefd470 AE |
2549 | { |
2550 | struct rbd_img_request *img_request; | |
2551 | struct rbd_device *rbd_dev; | |
ebda6408 | 2552 | struct page **pages; |
0eefd470 AE |
2553 | u32 page_count; |
2554 | ||
2761713d ID |
2555 | dout("%s: obj %p\n", __func__, obj_request); |
2556 | ||
d3246fb0 JD |
2557 | rbd_assert(obj_request->type == OBJ_REQUEST_BIO || |
2558 | obj_request->type == OBJ_REQUEST_NODATA); | |
0eefd470 AE |
2559 | rbd_assert(obj_request_img_data_test(obj_request)); |
2560 | img_request = obj_request->img_request; | |
2561 | rbd_assert(img_request); | |
2562 | ||
2563 | rbd_dev = img_request->rbd_dev; | |
2564 | rbd_assert(rbd_dev); | |
0eefd470 | 2565 | |
ebda6408 AE |
2566 | pages = obj_request->copyup_pages; |
2567 | rbd_assert(pages != NULL); | |
0eefd470 | 2568 | obj_request->copyup_pages = NULL; |
ebda6408 AE |
2569 | page_count = obj_request->copyup_page_count; |
2570 | rbd_assert(page_count); | |
2571 | obj_request->copyup_page_count = 0; | |
2572 | ceph_release_page_vector(pages, page_count); | |
0eefd470 AE |
2573 | |
2574 | /* | |
2575 | * We want the transfer count to reflect the size of the | |
2576 | * original write request. There is no such thing as a | |
2577 | * successful short write, so if the request was successful | |
2578 | * we can just set it to the originally-requested length. | |
2579 | */ | |
2580 | if (!obj_request->result) | |
2581 | obj_request->xferred = obj_request->length; | |
2582 | ||
2761713d | 2583 | obj_request_done_set(obj_request); |
0eefd470 AE |
2584 | } |
2585 | ||
3d7efd18 AE |
2586 | static void |
2587 | rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request) | |
2588 | { | |
2589 | struct rbd_obj_request *orig_request; | |
0eefd470 AE |
2590 | struct ceph_osd_request *osd_req; |
2591 | struct ceph_osd_client *osdc; | |
2592 | struct rbd_device *rbd_dev; | |
3d7efd18 | 2593 | struct page **pages; |
d3246fb0 | 2594 | enum obj_operation_type op_type; |
ebda6408 | 2595 | u32 page_count; |
bbea1c1a | 2596 | int img_result; |
ebda6408 | 2597 | u64 parent_length; |
3d7efd18 AE |
2598 | |
2599 | rbd_assert(img_request_child_test(img_request)); | |
2600 | ||
2601 | /* First get what we need from the image request */ | |
2602 | ||
2603 | pages = img_request->copyup_pages; | |
2604 | rbd_assert(pages != NULL); | |
2605 | img_request->copyup_pages = NULL; | |
ebda6408 AE |
2606 | page_count = img_request->copyup_page_count; |
2607 | rbd_assert(page_count); | |
2608 | img_request->copyup_page_count = 0; | |
3d7efd18 AE |
2609 | |
2610 | orig_request = img_request->obj_request; | |
2611 | rbd_assert(orig_request != NULL); | |
b91f09f1 | 2612 | rbd_assert(obj_request_type_valid(orig_request->type)); |
bbea1c1a | 2613 | img_result = img_request->result; |
ebda6408 AE |
2614 | parent_length = img_request->length; |
2615 | rbd_assert(parent_length == img_request->xferred); | |
91c6febb | 2616 | rbd_img_request_put(img_request); |
3d7efd18 | 2617 | |
91c6febb AE |
2618 | rbd_assert(orig_request->img_request); |
2619 | rbd_dev = orig_request->img_request->rbd_dev; | |
0eefd470 | 2620 | rbd_assert(rbd_dev); |
0eefd470 | 2621 | |
bbea1c1a AE |
2622 | /* |
2623 | * If the overlap has become 0 (most likely because the | |
2624 | * image has been flattened) we need to free the pages | |
2625 | * and re-submit the original write request. | |
2626 | */ | |
2627 | if (!rbd_dev->parent_overlap) { | |
2628 | struct ceph_osd_client *osdc; | |
3d7efd18 | 2629 | |
bbea1c1a AE |
2630 | ceph_release_page_vector(pages, page_count); |
2631 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2632 | img_result = rbd_obj_request_submit(osdc, orig_request); | |
2633 | if (!img_result) | |
2634 | return; | |
2635 | } | |
0eefd470 | 2636 | |
bbea1c1a | 2637 | if (img_result) |
0eefd470 | 2638 | goto out_err; |
0eefd470 | 2639 | |
8785b1d4 AE |
2640 | /* |
2641 | * The original osd request is of no use to use any more. | |
0ccd5926 | 2642 | * We need a new one that can hold the three ops in a copyup |
8785b1d4 AE |
2643 | * request. Allocate the new copyup osd request for the |
2644 | * original request, and release the old one. | |
2645 | */ | |
bbea1c1a | 2646 | img_result = -ENOMEM; |
0eefd470 AE |
2647 | osd_req = rbd_osd_req_create_copyup(orig_request); |
2648 | if (!osd_req) | |
2649 | goto out_err; | |
8785b1d4 | 2650 | rbd_osd_req_destroy(orig_request->osd_req); |
0eefd470 AE |
2651 | orig_request->osd_req = osd_req; |
2652 | orig_request->copyup_pages = pages; | |
ebda6408 | 2653 | orig_request->copyup_page_count = page_count; |
3d7efd18 | 2654 | |
0eefd470 | 2655 | /* Initialize the copyup op */ |
3d7efd18 | 2656 | |
0eefd470 | 2657 | osd_req_op_cls_init(osd_req, 0, CEPH_OSD_OP_CALL, "rbd", "copyup"); |
ebda6408 | 2658 | osd_req_op_cls_request_data_pages(osd_req, 0, pages, parent_length, 0, |
0eefd470 | 2659 | false, false); |
3d7efd18 | 2660 | |
d3246fb0 | 2661 | /* Add the other op(s) */ |
0eefd470 | 2662 | |
d3246fb0 JD |
2663 | op_type = rbd_img_request_op_type(orig_request->img_request); |
2664 | rbd_img_obj_request_fill(orig_request, osd_req, op_type, 1); | |
0eefd470 AE |
2665 | |
2666 | /* All set, send it off. */ | |
2667 | ||
0eefd470 | 2668 | osdc = &rbd_dev->rbd_client->client->osdc; |
bbea1c1a AE |
2669 | img_result = rbd_obj_request_submit(osdc, orig_request); |
2670 | if (!img_result) | |
0eefd470 AE |
2671 | return; |
2672 | out_err: | |
2673 | /* Record the error code and complete the request */ | |
2674 | ||
bbea1c1a | 2675 | orig_request->result = img_result; |
0eefd470 AE |
2676 | orig_request->xferred = 0; |
2677 | obj_request_done_set(orig_request); | |
2678 | rbd_obj_request_complete(orig_request); | |
3d7efd18 AE |
2679 | } |
2680 | ||
2681 | /* | |
2682 | * Read from the parent image the range of data that covers the | |
2683 | * entire target of the given object request. This is used for | |
2684 | * satisfying a layered image write request when the target of an | |
2685 | * object request from the image request does not exist. | |
2686 | * | |
2687 | * A page array big enough to hold the returned data is allocated | |
2688 | * and supplied to rbd_img_request_fill() as the "data descriptor." | |
2689 | * When the read completes, this page array will be transferred to | |
2690 | * the original object request for the copyup operation. | |
2691 | * | |
2692 | * If an error occurs, record it as the result of the original | |
2693 | * object request and mark it done so it gets completed. | |
2694 | */ | |
2695 | static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request) | |
2696 | { | |
2697 | struct rbd_img_request *img_request = NULL; | |
2698 | struct rbd_img_request *parent_request = NULL; | |
2699 | struct rbd_device *rbd_dev; | |
2700 | u64 img_offset; | |
2701 | u64 length; | |
2702 | struct page **pages = NULL; | |
2703 | u32 page_count; | |
2704 | int result; | |
2705 | ||
2706 | rbd_assert(obj_request_img_data_test(obj_request)); | |
b91f09f1 | 2707 | rbd_assert(obj_request_type_valid(obj_request->type)); |
3d7efd18 AE |
2708 | |
2709 | img_request = obj_request->img_request; | |
2710 | rbd_assert(img_request != NULL); | |
2711 | rbd_dev = img_request->rbd_dev; | |
2712 | rbd_assert(rbd_dev->parent != NULL); | |
2713 | ||
2714 | /* | |
2715 | * Determine the byte range covered by the object in the | |
2716 | * child image to which the original request was to be sent. | |
2717 | */ | |
2718 | img_offset = obj_request->img_offset - obj_request->offset; | |
2719 | length = (u64)1 << rbd_dev->header.obj_order; | |
2720 | ||
a9e8ba2c AE |
2721 | /* |
2722 | * There is no defined parent data beyond the parent | |
2723 | * overlap, so limit what we read at that boundary if | |
2724 | * necessary. | |
2725 | */ | |
2726 | if (img_offset + length > rbd_dev->parent_overlap) { | |
2727 | rbd_assert(img_offset < rbd_dev->parent_overlap); | |
2728 | length = rbd_dev->parent_overlap - img_offset; | |
2729 | } | |
2730 | ||
3d7efd18 AE |
2731 | /* |
2732 | * Allocate a page array big enough to receive the data read | |
2733 | * from the parent. | |
2734 | */ | |
2735 | page_count = (u32)calc_pages_for(0, length); | |
2736 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2737 | if (IS_ERR(pages)) { | |
2738 | result = PTR_ERR(pages); | |
2739 | pages = NULL; | |
2740 | goto out_err; | |
2741 | } | |
2742 | ||
2743 | result = -ENOMEM; | |
e93f3152 AE |
2744 | parent_request = rbd_parent_request_create(obj_request, |
2745 | img_offset, length); | |
3d7efd18 AE |
2746 | if (!parent_request) |
2747 | goto out_err; | |
3d7efd18 AE |
2748 | |
2749 | result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages); | |
2750 | if (result) | |
2751 | goto out_err; | |
2752 | parent_request->copyup_pages = pages; | |
ebda6408 | 2753 | parent_request->copyup_page_count = page_count; |
3d7efd18 AE |
2754 | |
2755 | parent_request->callback = rbd_img_obj_parent_read_full_callback; | |
2756 | result = rbd_img_request_submit(parent_request); | |
2757 | if (!result) | |
2758 | return 0; | |
2759 | ||
2760 | parent_request->copyup_pages = NULL; | |
ebda6408 | 2761 | parent_request->copyup_page_count = 0; |
3d7efd18 AE |
2762 | parent_request->obj_request = NULL; |
2763 | rbd_obj_request_put(obj_request); | |
2764 | out_err: | |
2765 | if (pages) | |
2766 | ceph_release_page_vector(pages, page_count); | |
2767 | if (parent_request) | |
2768 | rbd_img_request_put(parent_request); | |
2769 | obj_request->result = result; | |
2770 | obj_request->xferred = 0; | |
2771 | obj_request_done_set(obj_request); | |
2772 | ||
2773 | return result; | |
2774 | } | |
2775 | ||
c5b5ef6c AE |
2776 | static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request) |
2777 | { | |
c5b5ef6c | 2778 | struct rbd_obj_request *orig_request; |
638f5abe | 2779 | struct rbd_device *rbd_dev; |
c5b5ef6c AE |
2780 | int result; |
2781 | ||
2782 | rbd_assert(!obj_request_img_data_test(obj_request)); | |
2783 | ||
2784 | /* | |
2785 | * All we need from the object request is the original | |
2786 | * request and the result of the STAT op. Grab those, then | |
2787 | * we're done with the request. | |
2788 | */ | |
2789 | orig_request = obj_request->obj_request; | |
2790 | obj_request->obj_request = NULL; | |
912c317d | 2791 | rbd_obj_request_put(orig_request); |
c5b5ef6c AE |
2792 | rbd_assert(orig_request); |
2793 | rbd_assert(orig_request->img_request); | |
2794 | ||
2795 | result = obj_request->result; | |
2796 | obj_request->result = 0; | |
2797 | ||
2798 | dout("%s: obj %p for obj %p result %d %llu/%llu\n", __func__, | |
2799 | obj_request, orig_request, result, | |
2800 | obj_request->xferred, obj_request->length); | |
2801 | rbd_obj_request_put(obj_request); | |
2802 | ||
638f5abe AE |
2803 | /* |
2804 | * If the overlap has become 0 (most likely because the | |
2805 | * image has been flattened) we need to free the pages | |
2806 | * and re-submit the original write request. | |
2807 | */ | |
2808 | rbd_dev = orig_request->img_request->rbd_dev; | |
2809 | if (!rbd_dev->parent_overlap) { | |
2810 | struct ceph_osd_client *osdc; | |
2811 | ||
638f5abe AE |
2812 | osdc = &rbd_dev->rbd_client->client->osdc; |
2813 | result = rbd_obj_request_submit(osdc, orig_request); | |
2814 | if (!result) | |
2815 | return; | |
2816 | } | |
c5b5ef6c AE |
2817 | |
2818 | /* | |
2819 | * Our only purpose here is to determine whether the object | |
2820 | * exists, and we don't want to treat the non-existence as | |
2821 | * an error. If something else comes back, transfer the | |
2822 | * error to the original request and complete it now. | |
2823 | */ | |
2824 | if (!result) { | |
2825 | obj_request_existence_set(orig_request, true); | |
2826 | } else if (result == -ENOENT) { | |
2827 | obj_request_existence_set(orig_request, false); | |
2828 | } else if (result) { | |
2829 | orig_request->result = result; | |
3d7efd18 | 2830 | goto out; |
c5b5ef6c AE |
2831 | } |
2832 | ||
2833 | /* | |
2834 | * Resubmit the original request now that we have recorded | |
2835 | * whether the target object exists. | |
2836 | */ | |
b454e36d | 2837 | orig_request->result = rbd_img_obj_request_submit(orig_request); |
3d7efd18 | 2838 | out: |
c5b5ef6c AE |
2839 | if (orig_request->result) |
2840 | rbd_obj_request_complete(orig_request); | |
c5b5ef6c AE |
2841 | } |
2842 | ||
2843 | static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request) | |
2844 | { | |
2845 | struct rbd_obj_request *stat_request; | |
2846 | struct rbd_device *rbd_dev; | |
2847 | struct ceph_osd_client *osdc; | |
2848 | struct page **pages = NULL; | |
2849 | u32 page_count; | |
2850 | size_t size; | |
2851 | int ret; | |
2852 | ||
2853 | /* | |
2854 | * The response data for a STAT call consists of: | |
2855 | * le64 length; | |
2856 | * struct { | |
2857 | * le32 tv_sec; | |
2858 | * le32 tv_nsec; | |
2859 | * } mtime; | |
2860 | */ | |
2861 | size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32); | |
2862 | page_count = (u32)calc_pages_for(0, size); | |
2863 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
2864 | if (IS_ERR(pages)) | |
2865 | return PTR_ERR(pages); | |
2866 | ||
2867 | ret = -ENOMEM; | |
2868 | stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0, | |
2869 | OBJ_REQUEST_PAGES); | |
2870 | if (!stat_request) | |
2871 | goto out; | |
2872 | ||
2873 | rbd_obj_request_get(obj_request); | |
2874 | stat_request->obj_request = obj_request; | |
2875 | stat_request->pages = pages; | |
2876 | stat_request->page_count = page_count; | |
2877 | ||
2878 | rbd_assert(obj_request->img_request); | |
2879 | rbd_dev = obj_request->img_request->rbd_dev; | |
6d2940c8 | 2880 | stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 2881 | stat_request); |
c5b5ef6c AE |
2882 | if (!stat_request->osd_req) |
2883 | goto out; | |
2884 | stat_request->callback = rbd_img_obj_exists_callback; | |
2885 | ||
144cba14 | 2886 | osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0); |
c5b5ef6c AE |
2887 | osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0, |
2888 | false, false); | |
9d4df01f | 2889 | rbd_osd_req_format_read(stat_request); |
c5b5ef6c AE |
2890 | |
2891 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2892 | ret = rbd_obj_request_submit(osdc, stat_request); | |
2893 | out: | |
2894 | if (ret) | |
2895 | rbd_obj_request_put(obj_request); | |
2896 | ||
2897 | return ret; | |
2898 | } | |
2899 | ||
70d045f6 | 2900 | static bool img_obj_request_simple(struct rbd_obj_request *obj_request) |
b454e36d AE |
2901 | { |
2902 | struct rbd_img_request *img_request; | |
a9e8ba2c | 2903 | struct rbd_device *rbd_dev; |
b454e36d AE |
2904 | |
2905 | rbd_assert(obj_request_img_data_test(obj_request)); | |
2906 | ||
2907 | img_request = obj_request->img_request; | |
2908 | rbd_assert(img_request); | |
a9e8ba2c | 2909 | rbd_dev = img_request->rbd_dev; |
b454e36d | 2910 | |
70d045f6 | 2911 | /* Reads */ |
1c220881 JD |
2912 | if (!img_request_write_test(img_request) && |
2913 | !img_request_discard_test(img_request)) | |
70d045f6 ID |
2914 | return true; |
2915 | ||
2916 | /* Non-layered writes */ | |
2917 | if (!img_request_layered_test(img_request)) | |
2918 | return true; | |
2919 | ||
b454e36d | 2920 | /* |
70d045f6 ID |
2921 | * Layered writes outside of the parent overlap range don't |
2922 | * share any data with the parent. | |
b454e36d | 2923 | */ |
70d045f6 ID |
2924 | if (!obj_request_overlaps_parent(obj_request)) |
2925 | return true; | |
b454e36d | 2926 | |
c622d226 GZ |
2927 | /* |
2928 | * Entire-object layered writes - we will overwrite whatever | |
2929 | * parent data there is anyway. | |
2930 | */ | |
2931 | if (!obj_request->offset && | |
2932 | obj_request->length == rbd_obj_bytes(&rbd_dev->header)) | |
2933 | return true; | |
2934 | ||
70d045f6 ID |
2935 | /* |
2936 | * If the object is known to already exist, its parent data has | |
2937 | * already been copied. | |
2938 | */ | |
2939 | if (obj_request_known_test(obj_request) && | |
2940 | obj_request_exists_test(obj_request)) | |
2941 | return true; | |
2942 | ||
2943 | return false; | |
2944 | } | |
2945 | ||
2946 | static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) | |
2947 | { | |
2948 | if (img_obj_request_simple(obj_request)) { | |
b454e36d AE |
2949 | struct rbd_device *rbd_dev; |
2950 | struct ceph_osd_client *osdc; | |
2951 | ||
2952 | rbd_dev = obj_request->img_request->rbd_dev; | |
2953 | osdc = &rbd_dev->rbd_client->client->osdc; | |
2954 | ||
2955 | return rbd_obj_request_submit(osdc, obj_request); | |
2956 | } | |
2957 | ||
2958 | /* | |
3d7efd18 AE |
2959 | * It's a layered write. The target object might exist but |
2960 | * we may not know that yet. If we know it doesn't exist, | |
2961 | * start by reading the data for the full target object from | |
2962 | * the parent so we can use it for a copyup to the target. | |
b454e36d | 2963 | */ |
70d045f6 | 2964 | if (obj_request_known_test(obj_request)) |
3d7efd18 AE |
2965 | return rbd_img_obj_parent_read_full(obj_request); |
2966 | ||
2967 | /* We don't know whether the target exists. Go find out. */ | |
b454e36d AE |
2968 | |
2969 | return rbd_img_obj_exists_submit(obj_request); | |
2970 | } | |
2971 | ||
bf0d5f50 AE |
2972 | static int rbd_img_request_submit(struct rbd_img_request *img_request) |
2973 | { | |
bf0d5f50 | 2974 | struct rbd_obj_request *obj_request; |
46faeed4 | 2975 | struct rbd_obj_request *next_obj_request; |
bf0d5f50 | 2976 | |
37206ee5 | 2977 | dout("%s: img %p\n", __func__, img_request); |
46faeed4 | 2978 | for_each_obj_request_safe(img_request, obj_request, next_obj_request) { |
bf0d5f50 AE |
2979 | int ret; |
2980 | ||
b454e36d | 2981 | ret = rbd_img_obj_request_submit(obj_request); |
bf0d5f50 AE |
2982 | if (ret) |
2983 | return ret; | |
bf0d5f50 AE |
2984 | } |
2985 | ||
2986 | return 0; | |
2987 | } | |
8b3e1a56 AE |
2988 | |
2989 | static void rbd_img_parent_read_callback(struct rbd_img_request *img_request) | |
2990 | { | |
2991 | struct rbd_obj_request *obj_request; | |
a9e8ba2c AE |
2992 | struct rbd_device *rbd_dev; |
2993 | u64 obj_end; | |
02c74fba AE |
2994 | u64 img_xferred; |
2995 | int img_result; | |
8b3e1a56 AE |
2996 | |
2997 | rbd_assert(img_request_child_test(img_request)); | |
2998 | ||
02c74fba AE |
2999 | /* First get what we need from the image request and release it */ |
3000 | ||
8b3e1a56 | 3001 | obj_request = img_request->obj_request; |
02c74fba AE |
3002 | img_xferred = img_request->xferred; |
3003 | img_result = img_request->result; | |
3004 | rbd_img_request_put(img_request); | |
3005 | ||
3006 | /* | |
3007 | * If the overlap has become 0 (most likely because the | |
3008 | * image has been flattened) we need to re-submit the | |
3009 | * original request. | |
3010 | */ | |
a9e8ba2c AE |
3011 | rbd_assert(obj_request); |
3012 | rbd_assert(obj_request->img_request); | |
02c74fba AE |
3013 | rbd_dev = obj_request->img_request->rbd_dev; |
3014 | if (!rbd_dev->parent_overlap) { | |
3015 | struct ceph_osd_client *osdc; | |
3016 | ||
3017 | osdc = &rbd_dev->rbd_client->client->osdc; | |
3018 | img_result = rbd_obj_request_submit(osdc, obj_request); | |
3019 | if (!img_result) | |
3020 | return; | |
3021 | } | |
a9e8ba2c | 3022 | |
02c74fba | 3023 | obj_request->result = img_result; |
a9e8ba2c AE |
3024 | if (obj_request->result) |
3025 | goto out; | |
3026 | ||
3027 | /* | |
3028 | * We need to zero anything beyond the parent overlap | |
3029 | * boundary. Since rbd_img_obj_request_read_callback() | |
3030 | * will zero anything beyond the end of a short read, an | |
3031 | * easy way to do this is to pretend the data from the | |
3032 | * parent came up short--ending at the overlap boundary. | |
3033 | */ | |
3034 | rbd_assert(obj_request->img_offset < U64_MAX - obj_request->length); | |
3035 | obj_end = obj_request->img_offset + obj_request->length; | |
a9e8ba2c AE |
3036 | if (obj_end > rbd_dev->parent_overlap) { |
3037 | u64 xferred = 0; | |
3038 | ||
3039 | if (obj_request->img_offset < rbd_dev->parent_overlap) | |
3040 | xferred = rbd_dev->parent_overlap - | |
3041 | obj_request->img_offset; | |
8b3e1a56 | 3042 | |
02c74fba | 3043 | obj_request->xferred = min(img_xferred, xferred); |
a9e8ba2c | 3044 | } else { |
02c74fba | 3045 | obj_request->xferred = img_xferred; |
a9e8ba2c AE |
3046 | } |
3047 | out: | |
8b3e1a56 AE |
3048 | rbd_img_obj_request_read_callback(obj_request); |
3049 | rbd_obj_request_complete(obj_request); | |
3050 | } | |
3051 | ||
3052 | static void rbd_img_parent_read(struct rbd_obj_request *obj_request) | |
3053 | { | |
8b3e1a56 AE |
3054 | struct rbd_img_request *img_request; |
3055 | int result; | |
3056 | ||
3057 | rbd_assert(obj_request_img_data_test(obj_request)); | |
3058 | rbd_assert(obj_request->img_request != NULL); | |
3059 | rbd_assert(obj_request->result == (s32) -ENOENT); | |
5b2ab72d | 3060 | rbd_assert(obj_request_type_valid(obj_request->type)); |
8b3e1a56 | 3061 | |
8b3e1a56 | 3062 | /* rbd_read_finish(obj_request, obj_request->length); */ |
e93f3152 | 3063 | img_request = rbd_parent_request_create(obj_request, |
8b3e1a56 | 3064 | obj_request->img_offset, |
e93f3152 | 3065 | obj_request->length); |
8b3e1a56 AE |
3066 | result = -ENOMEM; |
3067 | if (!img_request) | |
3068 | goto out_err; | |
3069 | ||
5b2ab72d AE |
3070 | if (obj_request->type == OBJ_REQUEST_BIO) |
3071 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3072 | obj_request->bio_list); | |
3073 | else | |
3074 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_PAGES, | |
3075 | obj_request->pages); | |
8b3e1a56 AE |
3076 | if (result) |
3077 | goto out_err; | |
3078 | ||
3079 | img_request->callback = rbd_img_parent_read_callback; | |
3080 | result = rbd_img_request_submit(img_request); | |
3081 | if (result) | |
3082 | goto out_err; | |
3083 | ||
3084 | return; | |
3085 | out_err: | |
3086 | if (img_request) | |
3087 | rbd_img_request_put(img_request); | |
3088 | obj_request->result = result; | |
3089 | obj_request->xferred = 0; | |
3090 | obj_request_done_set(obj_request); | |
3091 | } | |
bf0d5f50 | 3092 | |
20e0af67 | 3093 | static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id) |
b8d70035 AE |
3094 | { |
3095 | struct rbd_obj_request *obj_request; | |
2169238d | 3096 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
b8d70035 AE |
3097 | int ret; |
3098 | ||
3099 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3100 | OBJ_REQUEST_NODATA); | |
3101 | if (!obj_request) | |
3102 | return -ENOMEM; | |
3103 | ||
3104 | ret = -ENOMEM; | |
6d2940c8 | 3105 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3106 | obj_request); |
b8d70035 AE |
3107 | if (!obj_request->osd_req) |
3108 | goto out; | |
3109 | ||
c99d2d4a | 3110 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK, |
cc4a38bd | 3111 | notify_id, 0, 0); |
9d4df01f | 3112 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3113 | |
b8d70035 | 3114 | ret = rbd_obj_request_submit(osdc, obj_request); |
cf81b60e | 3115 | if (ret) |
20e0af67 JD |
3116 | goto out; |
3117 | ret = rbd_obj_request_wait(obj_request); | |
3118 | out: | |
3119 | rbd_obj_request_put(obj_request); | |
b8d70035 AE |
3120 | |
3121 | return ret; | |
3122 | } | |
3123 | ||
3124 | static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data) | |
3125 | { | |
3126 | struct rbd_device *rbd_dev = (struct rbd_device *)data; | |
e627db08 | 3127 | int ret; |
b8d70035 | 3128 | |
37206ee5 | 3129 | dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__, |
cc4a38bd AE |
3130 | rbd_dev->header_name, (unsigned long long)notify_id, |
3131 | (unsigned int)opcode); | |
52bb1f9b ID |
3132 | |
3133 | /* | |
3134 | * Until adequate refresh error handling is in place, there is | |
3135 | * not much we can do here, except warn. | |
3136 | * | |
3137 | * See http://tracker.ceph.com/issues/5040 | |
3138 | */ | |
e627db08 AE |
3139 | ret = rbd_dev_refresh(rbd_dev); |
3140 | if (ret) | |
9584d508 | 3141 | rbd_warn(rbd_dev, "refresh failed: %d", ret); |
b8d70035 | 3142 | |
52bb1f9b ID |
3143 | ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id); |
3144 | if (ret) | |
9584d508 | 3145 | rbd_warn(rbd_dev, "notify_ack ret %d", ret); |
b8d70035 AE |
3146 | } |
3147 | ||
bb040aa0 ID |
3148 | /* |
3149 | * Send a (un)watch request and wait for the ack. Return a request | |
3150 | * with a ref held on success or error. | |
3151 | */ | |
3152 | static struct rbd_obj_request *rbd_obj_watch_request_helper( | |
3153 | struct rbd_device *rbd_dev, | |
3154 | bool watch) | |
3155 | { | |
3156 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
2894e1d7 | 3157 | struct ceph_options *opts = osdc->client->options; |
bb040aa0 ID |
3158 | struct rbd_obj_request *obj_request; |
3159 | int ret; | |
3160 | ||
3161 | obj_request = rbd_obj_request_create(rbd_dev->header_name, 0, 0, | |
3162 | OBJ_REQUEST_NODATA); | |
3163 | if (!obj_request) | |
3164 | return ERR_PTR(-ENOMEM); | |
3165 | ||
6d2940c8 | 3166 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1, |
bb040aa0 ID |
3167 | obj_request); |
3168 | if (!obj_request->osd_req) { | |
3169 | ret = -ENOMEM; | |
3170 | goto out; | |
3171 | } | |
3172 | ||
3173 | osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH, | |
3174 | rbd_dev->watch_event->cookie, 0, watch); | |
3175 | rbd_osd_req_format_write(obj_request); | |
3176 | ||
3177 | if (watch) | |
3178 | ceph_osdc_set_request_linger(osdc, obj_request->osd_req); | |
3179 | ||
3180 | ret = rbd_obj_request_submit(osdc, obj_request); | |
3181 | if (ret) | |
3182 | goto out; | |
3183 | ||
2894e1d7 | 3184 | ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout); |
bb040aa0 ID |
3185 | if (ret) |
3186 | goto out; | |
3187 | ||
3188 | ret = obj_request->result; | |
3189 | if (ret) { | |
3190 | if (watch) | |
3191 | rbd_obj_request_end(obj_request); | |
3192 | goto out; | |
3193 | } | |
3194 | ||
3195 | return obj_request; | |
3196 | ||
3197 | out: | |
3198 | rbd_obj_request_put(obj_request); | |
3199 | return ERR_PTR(ret); | |
3200 | } | |
3201 | ||
9969ebc5 | 3202 | /* |
b30a01f2 | 3203 | * Initiate a watch request, synchronously. |
9969ebc5 | 3204 | */ |
b30a01f2 | 3205 | static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev) |
9969ebc5 AE |
3206 | { |
3207 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; | |
3208 | struct rbd_obj_request *obj_request; | |
9969ebc5 AE |
3209 | int ret; |
3210 | ||
b30a01f2 ID |
3211 | rbd_assert(!rbd_dev->watch_event); |
3212 | rbd_assert(!rbd_dev->watch_request); | |
9969ebc5 | 3213 | |
b30a01f2 ID |
3214 | ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev, |
3215 | &rbd_dev->watch_event); | |
3216 | if (ret < 0) | |
3217 | return ret; | |
3218 | ||
76756a51 ID |
3219 | obj_request = rbd_obj_watch_request_helper(rbd_dev, true); |
3220 | if (IS_ERR(obj_request)) { | |
3221 | ceph_osdc_cancel_event(rbd_dev->watch_event); | |
3222 | rbd_dev->watch_event = NULL; | |
3223 | return PTR_ERR(obj_request); | |
b30a01f2 | 3224 | } |
9969ebc5 | 3225 | |
8eb87565 AE |
3226 | /* |
3227 | * A watch request is set to linger, so the underlying osd | |
3228 | * request won't go away until we unregister it. We retain | |
3229 | * a pointer to the object request during that time (in | |
76756a51 ID |
3230 | * rbd_dev->watch_request), so we'll keep a reference to it. |
3231 | * We'll drop that reference after we've unregistered it in | |
3232 | * rbd_dev_header_unwatch_sync(). | |
8eb87565 | 3233 | */ |
b30a01f2 | 3234 | rbd_dev->watch_request = obj_request; |
8eb87565 | 3235 | |
b30a01f2 | 3236 | return 0; |
b30a01f2 ID |
3237 | } |
3238 | ||
3239 | /* | |
3240 | * Tear down a watch request, synchronously. | |
3241 | */ | |
76756a51 | 3242 | static void rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev) |
b30a01f2 | 3243 | { |
b30a01f2 | 3244 | struct rbd_obj_request *obj_request; |
b30a01f2 ID |
3245 | |
3246 | rbd_assert(rbd_dev->watch_event); | |
3247 | rbd_assert(rbd_dev->watch_request); | |
3248 | ||
76756a51 | 3249 | rbd_obj_request_end(rbd_dev->watch_request); |
8eb87565 AE |
3250 | rbd_obj_request_put(rbd_dev->watch_request); |
3251 | rbd_dev->watch_request = NULL; | |
b30a01f2 | 3252 | |
76756a51 ID |
3253 | obj_request = rbd_obj_watch_request_helper(rbd_dev, false); |
3254 | if (!IS_ERR(obj_request)) | |
3255 | rbd_obj_request_put(obj_request); | |
3256 | else | |
3257 | rbd_warn(rbd_dev, "unable to tear down watch request (%ld)", | |
3258 | PTR_ERR(obj_request)); | |
3259 | ||
9969ebc5 AE |
3260 | ceph_osdc_cancel_event(rbd_dev->watch_event); |
3261 | rbd_dev->watch_event = NULL; | |
811c6688 ID |
3262 | |
3263 | dout("%s flushing notifies\n", __func__); | |
3264 | ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc); | |
fca27065 ID |
3265 | } |
3266 | ||
36be9a76 | 3267 | /* |
f40eb349 AE |
3268 | * Synchronous osd object method call. Returns the number of bytes |
3269 | * returned in the outbound buffer, or a negative error code. | |
36be9a76 AE |
3270 | */ |
3271 | static int rbd_obj_method_sync(struct rbd_device *rbd_dev, | |
3272 | const char *object_name, | |
3273 | const char *class_name, | |
3274 | const char *method_name, | |
4157976b | 3275 | const void *outbound, |
36be9a76 | 3276 | size_t outbound_size, |
4157976b | 3277 | void *inbound, |
e2a58ee5 | 3278 | size_t inbound_size) |
36be9a76 | 3279 | { |
2169238d | 3280 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
36be9a76 | 3281 | struct rbd_obj_request *obj_request; |
36be9a76 AE |
3282 | struct page **pages; |
3283 | u32 page_count; | |
3284 | int ret; | |
3285 | ||
3286 | /* | |
6010a451 AE |
3287 | * Method calls are ultimately read operations. The result |
3288 | * should placed into the inbound buffer provided. They | |
3289 | * also supply outbound data--parameters for the object | |
3290 | * method. Currently if this is present it will be a | |
3291 | * snapshot id. | |
36be9a76 | 3292 | */ |
57385b51 | 3293 | page_count = (u32)calc_pages_for(0, inbound_size); |
36be9a76 AE |
3294 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); |
3295 | if (IS_ERR(pages)) | |
3296 | return PTR_ERR(pages); | |
3297 | ||
3298 | ret = -ENOMEM; | |
6010a451 | 3299 | obj_request = rbd_obj_request_create(object_name, 0, inbound_size, |
36be9a76 AE |
3300 | OBJ_REQUEST_PAGES); |
3301 | if (!obj_request) | |
3302 | goto out; | |
3303 | ||
3304 | obj_request->pages = pages; | |
3305 | obj_request->page_count = page_count; | |
3306 | ||
6d2940c8 | 3307 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3308 | obj_request); |
36be9a76 AE |
3309 | if (!obj_request->osd_req) |
3310 | goto out; | |
3311 | ||
c99d2d4a | 3312 | osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL, |
04017e29 AE |
3313 | class_name, method_name); |
3314 | if (outbound_size) { | |
3315 | struct ceph_pagelist *pagelist; | |
3316 | ||
3317 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); | |
3318 | if (!pagelist) | |
3319 | goto out; | |
3320 | ||
3321 | ceph_pagelist_init(pagelist); | |
3322 | ceph_pagelist_append(pagelist, outbound, outbound_size); | |
3323 | osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0, | |
3324 | pagelist); | |
3325 | } | |
a4ce40a9 AE |
3326 | osd_req_op_cls_response_data_pages(obj_request->osd_req, 0, |
3327 | obj_request->pages, inbound_size, | |
44cd188d | 3328 | 0, false, false); |
9d4df01f | 3329 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3330 | |
36be9a76 AE |
3331 | ret = rbd_obj_request_submit(osdc, obj_request); |
3332 | if (ret) | |
3333 | goto out; | |
3334 | ret = rbd_obj_request_wait(obj_request); | |
3335 | if (ret) | |
3336 | goto out; | |
3337 | ||
3338 | ret = obj_request->result; | |
3339 | if (ret < 0) | |
3340 | goto out; | |
57385b51 AE |
3341 | |
3342 | rbd_assert(obj_request->xferred < (u64)INT_MAX); | |
3343 | ret = (int)obj_request->xferred; | |
903bb32e | 3344 | ceph_copy_from_page_vector(pages, inbound, 0, obj_request->xferred); |
36be9a76 AE |
3345 | out: |
3346 | if (obj_request) | |
3347 | rbd_obj_request_put(obj_request); | |
3348 | else | |
3349 | ceph_release_page_vector(pages, page_count); | |
3350 | ||
3351 | return ret; | |
3352 | } | |
3353 | ||
7ad18afa | 3354 | static void rbd_queue_workfn(struct work_struct *work) |
bf0d5f50 | 3355 | { |
7ad18afa CH |
3356 | struct request *rq = blk_mq_rq_from_pdu(work); |
3357 | struct rbd_device *rbd_dev = rq->q->queuedata; | |
bc1ecc65 | 3358 | struct rbd_img_request *img_request; |
4e752f0a | 3359 | struct ceph_snap_context *snapc = NULL; |
bc1ecc65 ID |
3360 | u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT; |
3361 | u64 length = blk_rq_bytes(rq); | |
6d2940c8 | 3362 | enum obj_operation_type op_type; |
4e752f0a | 3363 | u64 mapping_size; |
bf0d5f50 AE |
3364 | int result; |
3365 | ||
7ad18afa CH |
3366 | if (rq->cmd_type != REQ_TYPE_FS) { |
3367 | dout("%s: non-fs request type %d\n", __func__, | |
3368 | (int) rq->cmd_type); | |
3369 | result = -EIO; | |
3370 | goto err; | |
3371 | } | |
3372 | ||
90e98c52 GZ |
3373 | if (rq->cmd_flags & REQ_DISCARD) |
3374 | op_type = OBJ_OP_DISCARD; | |
3375 | else if (rq->cmd_flags & REQ_WRITE) | |
6d2940c8 GZ |
3376 | op_type = OBJ_OP_WRITE; |
3377 | else | |
3378 | op_type = OBJ_OP_READ; | |
3379 | ||
bc1ecc65 | 3380 | /* Ignore/skip any zero-length requests */ |
bf0d5f50 | 3381 | |
bc1ecc65 ID |
3382 | if (!length) { |
3383 | dout("%s: zero-length request\n", __func__); | |
3384 | result = 0; | |
3385 | goto err_rq; | |
3386 | } | |
bf0d5f50 | 3387 | |
6d2940c8 | 3388 | /* Only reads are allowed to a read-only device */ |
bc1ecc65 | 3389 | |
6d2940c8 | 3390 | if (op_type != OBJ_OP_READ) { |
bc1ecc65 ID |
3391 | if (rbd_dev->mapping.read_only) { |
3392 | result = -EROFS; | |
3393 | goto err_rq; | |
4dda41d3 | 3394 | } |
bc1ecc65 ID |
3395 | rbd_assert(rbd_dev->spec->snap_id == CEPH_NOSNAP); |
3396 | } | |
4dda41d3 | 3397 | |
bc1ecc65 ID |
3398 | /* |
3399 | * Quit early if the mapped snapshot no longer exists. It's | |
3400 | * still possible the snapshot will have disappeared by the | |
3401 | * time our request arrives at the osd, but there's no sense in | |
3402 | * sending it if we already know. | |
3403 | */ | |
3404 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) { | |
3405 | dout("request for non-existent snapshot"); | |
3406 | rbd_assert(rbd_dev->spec->snap_id != CEPH_NOSNAP); | |
3407 | result = -ENXIO; | |
3408 | goto err_rq; | |
3409 | } | |
4dda41d3 | 3410 | |
bc1ecc65 ID |
3411 | if (offset && length > U64_MAX - offset + 1) { |
3412 | rbd_warn(rbd_dev, "bad request range (%llu~%llu)", offset, | |
3413 | length); | |
3414 | result = -EINVAL; | |
3415 | goto err_rq; /* Shouldn't happen */ | |
3416 | } | |
4dda41d3 | 3417 | |
7ad18afa CH |
3418 | blk_mq_start_request(rq); |
3419 | ||
4e752f0a JD |
3420 | down_read(&rbd_dev->header_rwsem); |
3421 | mapping_size = rbd_dev->mapping.size; | |
6d2940c8 | 3422 | if (op_type != OBJ_OP_READ) { |
4e752f0a JD |
3423 | snapc = rbd_dev->header.snapc; |
3424 | ceph_get_snap_context(snapc); | |
3425 | } | |
3426 | up_read(&rbd_dev->header_rwsem); | |
3427 | ||
3428 | if (offset + length > mapping_size) { | |
bc1ecc65 | 3429 | rbd_warn(rbd_dev, "beyond EOD (%llu~%llu > %llu)", offset, |
4e752f0a | 3430 | length, mapping_size); |
bc1ecc65 ID |
3431 | result = -EIO; |
3432 | goto err_rq; | |
3433 | } | |
bf0d5f50 | 3434 | |
6d2940c8 | 3435 | img_request = rbd_img_request_create(rbd_dev, offset, length, op_type, |
4e752f0a | 3436 | snapc); |
bc1ecc65 ID |
3437 | if (!img_request) { |
3438 | result = -ENOMEM; | |
3439 | goto err_rq; | |
3440 | } | |
3441 | img_request->rq = rq; | |
70b16db8 | 3442 | snapc = NULL; /* img_request consumes a ref */ |
bf0d5f50 | 3443 | |
90e98c52 GZ |
3444 | if (op_type == OBJ_OP_DISCARD) |
3445 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_NODATA, | |
3446 | NULL); | |
3447 | else | |
3448 | result = rbd_img_request_fill(img_request, OBJ_REQUEST_BIO, | |
3449 | rq->bio); | |
bc1ecc65 ID |
3450 | if (result) |
3451 | goto err_img_request; | |
bf0d5f50 | 3452 | |
bc1ecc65 ID |
3453 | result = rbd_img_request_submit(img_request); |
3454 | if (result) | |
3455 | goto err_img_request; | |
bf0d5f50 | 3456 | |
bc1ecc65 | 3457 | return; |
bf0d5f50 | 3458 | |
bc1ecc65 ID |
3459 | err_img_request: |
3460 | rbd_img_request_put(img_request); | |
3461 | err_rq: | |
3462 | if (result) | |
3463 | rbd_warn(rbd_dev, "%s %llx at %llx result %d", | |
6d2940c8 | 3464 | obj_op_name(op_type), length, offset, result); |
e96a650a | 3465 | ceph_put_snap_context(snapc); |
7ad18afa CH |
3466 | err: |
3467 | blk_mq_end_request(rq, result); | |
bc1ecc65 | 3468 | } |
bf0d5f50 | 3469 | |
7ad18afa CH |
3470 | static int rbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
3471 | const struct blk_mq_queue_data *bd) | |
bc1ecc65 | 3472 | { |
7ad18afa CH |
3473 | struct request *rq = bd->rq; |
3474 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
bf0d5f50 | 3475 | |
7ad18afa CH |
3476 | queue_work(rbd_wq, work); |
3477 | return BLK_MQ_RQ_QUEUE_OK; | |
bf0d5f50 AE |
3478 | } |
3479 | ||
602adf40 YS |
3480 | static void rbd_free_disk(struct rbd_device *rbd_dev) |
3481 | { | |
3482 | struct gendisk *disk = rbd_dev->disk; | |
3483 | ||
3484 | if (!disk) | |
3485 | return; | |
3486 | ||
a0cab924 AE |
3487 | rbd_dev->disk = NULL; |
3488 | if (disk->flags & GENHD_FL_UP) { | |
602adf40 | 3489 | del_gendisk(disk); |
a0cab924 AE |
3490 | if (disk->queue) |
3491 | blk_cleanup_queue(disk->queue); | |
7ad18afa | 3492 | blk_mq_free_tag_set(&rbd_dev->tag_set); |
a0cab924 | 3493 | } |
602adf40 YS |
3494 | put_disk(disk); |
3495 | } | |
3496 | ||
788e2df3 AE |
3497 | static int rbd_obj_read_sync(struct rbd_device *rbd_dev, |
3498 | const char *object_name, | |
7097f8df | 3499 | u64 offset, u64 length, void *buf) |
788e2df3 AE |
3500 | |
3501 | { | |
2169238d | 3502 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
788e2df3 | 3503 | struct rbd_obj_request *obj_request; |
788e2df3 AE |
3504 | struct page **pages = NULL; |
3505 | u32 page_count; | |
1ceae7ef | 3506 | size_t size; |
788e2df3 AE |
3507 | int ret; |
3508 | ||
3509 | page_count = (u32) calc_pages_for(offset, length); | |
3510 | pages = ceph_alloc_page_vector(page_count, GFP_KERNEL); | |
3511 | if (IS_ERR(pages)) | |
a8d42056 | 3512 | return PTR_ERR(pages); |
788e2df3 AE |
3513 | |
3514 | ret = -ENOMEM; | |
3515 | obj_request = rbd_obj_request_create(object_name, offset, length, | |
36be9a76 | 3516 | OBJ_REQUEST_PAGES); |
788e2df3 AE |
3517 | if (!obj_request) |
3518 | goto out; | |
3519 | ||
3520 | obj_request->pages = pages; | |
3521 | obj_request->page_count = page_count; | |
3522 | ||
6d2940c8 | 3523 | obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1, |
deb236b3 | 3524 | obj_request); |
788e2df3 AE |
3525 | if (!obj_request->osd_req) |
3526 | goto out; | |
3527 | ||
c99d2d4a AE |
3528 | osd_req_op_extent_init(obj_request->osd_req, 0, CEPH_OSD_OP_READ, |
3529 | offset, length, 0, 0); | |
406e2c9f | 3530 | osd_req_op_extent_osd_data_pages(obj_request->osd_req, 0, |
a4ce40a9 | 3531 | obj_request->pages, |
44cd188d AE |
3532 | obj_request->length, |
3533 | obj_request->offset & ~PAGE_MASK, | |
3534 | false, false); | |
9d4df01f | 3535 | rbd_osd_req_format_read(obj_request); |
430c28c3 | 3536 | |
788e2df3 AE |
3537 | ret = rbd_obj_request_submit(osdc, obj_request); |
3538 | if (ret) | |
3539 | goto out; | |
3540 | ret = rbd_obj_request_wait(obj_request); | |
3541 | if (ret) | |
3542 | goto out; | |
3543 | ||
3544 | ret = obj_request->result; | |
3545 | if (ret < 0) | |
3546 | goto out; | |
1ceae7ef AE |
3547 | |
3548 | rbd_assert(obj_request->xferred <= (u64) SIZE_MAX); | |
3549 | size = (size_t) obj_request->xferred; | |
903bb32e | 3550 | ceph_copy_from_page_vector(pages, buf, 0, size); |
7097f8df AE |
3551 | rbd_assert(size <= (size_t)INT_MAX); |
3552 | ret = (int)size; | |
788e2df3 AE |
3553 | out: |
3554 | if (obj_request) | |
3555 | rbd_obj_request_put(obj_request); | |
3556 | else | |
3557 | ceph_release_page_vector(pages, page_count); | |
3558 | ||
3559 | return ret; | |
3560 | } | |
3561 | ||
602adf40 | 3562 | /* |
662518b1 AE |
3563 | * Read the complete header for the given rbd device. On successful |
3564 | * return, the rbd_dev->header field will contain up-to-date | |
3565 | * information about the image. | |
602adf40 | 3566 | */ |
99a41ebc | 3567 | static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev) |
602adf40 | 3568 | { |
4156d998 | 3569 | struct rbd_image_header_ondisk *ondisk = NULL; |
50f7c4c9 | 3570 | u32 snap_count = 0; |
4156d998 AE |
3571 | u64 names_size = 0; |
3572 | u32 want_count; | |
3573 | int ret; | |
602adf40 | 3574 | |
00f1f36f | 3575 | /* |
4156d998 AE |
3576 | * The complete header will include an array of its 64-bit |
3577 | * snapshot ids, followed by the names of those snapshots as | |
3578 | * a contiguous block of NUL-terminated strings. Note that | |
3579 | * the number of snapshots could change by the time we read | |
3580 | * it in, in which case we re-read it. | |
00f1f36f | 3581 | */ |
4156d998 AE |
3582 | do { |
3583 | size_t size; | |
3584 | ||
3585 | kfree(ondisk); | |
3586 | ||
3587 | size = sizeof (*ondisk); | |
3588 | size += snap_count * sizeof (struct rbd_image_snap_ondisk); | |
3589 | size += names_size; | |
3590 | ondisk = kmalloc(size, GFP_KERNEL); | |
3591 | if (!ondisk) | |
662518b1 | 3592 | return -ENOMEM; |
4156d998 | 3593 | |
788e2df3 | 3594 | ret = rbd_obj_read_sync(rbd_dev, rbd_dev->header_name, |
7097f8df | 3595 | 0, size, ondisk); |
4156d998 | 3596 | if (ret < 0) |
662518b1 | 3597 | goto out; |
c0cd10db | 3598 | if ((size_t)ret < size) { |
4156d998 | 3599 | ret = -ENXIO; |
06ecc6cb AE |
3600 | rbd_warn(rbd_dev, "short header read (want %zd got %d)", |
3601 | size, ret); | |
662518b1 | 3602 | goto out; |
4156d998 AE |
3603 | } |
3604 | if (!rbd_dev_ondisk_valid(ondisk)) { | |
3605 | ret = -ENXIO; | |
06ecc6cb | 3606 | rbd_warn(rbd_dev, "invalid header"); |
662518b1 | 3607 | goto out; |
81e759fb | 3608 | } |
602adf40 | 3609 | |
4156d998 AE |
3610 | names_size = le64_to_cpu(ondisk->snap_names_len); |
3611 | want_count = snap_count; | |
3612 | snap_count = le32_to_cpu(ondisk->snap_count); | |
3613 | } while (snap_count != want_count); | |
00f1f36f | 3614 | |
662518b1 AE |
3615 | ret = rbd_header_from_disk(rbd_dev, ondisk); |
3616 | out: | |
4156d998 AE |
3617 | kfree(ondisk); |
3618 | ||
3619 | return ret; | |
602adf40 YS |
3620 | } |
3621 | ||
15228ede AE |
3622 | /* |
3623 | * Clear the rbd device's EXISTS flag if the snapshot it's mapped to | |
3624 | * has disappeared from the (just updated) snapshot context. | |
3625 | */ | |
3626 | static void rbd_exists_validate(struct rbd_device *rbd_dev) | |
3627 | { | |
3628 | u64 snap_id; | |
3629 | ||
3630 | if (!test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags)) | |
3631 | return; | |
3632 | ||
3633 | snap_id = rbd_dev->spec->snap_id; | |
3634 | if (snap_id == CEPH_NOSNAP) | |
3635 | return; | |
3636 | ||
3637 | if (rbd_dev_snap_index(rbd_dev, snap_id) == BAD_SNAP_INDEX) | |
3638 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); | |
3639 | } | |
3640 | ||
9875201e JD |
3641 | static void rbd_dev_update_size(struct rbd_device *rbd_dev) |
3642 | { | |
3643 | sector_t size; | |
9875201e JD |
3644 | |
3645 | /* | |
811c6688 ID |
3646 | * If EXISTS is not set, rbd_dev->disk may be NULL, so don't |
3647 | * try to update its size. If REMOVING is set, updating size | |
3648 | * is just useless work since the device can't be opened. | |
9875201e | 3649 | */ |
811c6688 ID |
3650 | if (test_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags) && |
3651 | !test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags)) { | |
9875201e JD |
3652 | size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE; |
3653 | dout("setting size to %llu sectors", (unsigned long long)size); | |
3654 | set_capacity(rbd_dev->disk, size); | |
3655 | revalidate_disk(rbd_dev->disk); | |
3656 | } | |
3657 | } | |
3658 | ||
cc4a38bd | 3659 | static int rbd_dev_refresh(struct rbd_device *rbd_dev) |
1fe5e993 | 3660 | { |
e627db08 | 3661 | u64 mapping_size; |
1fe5e993 AE |
3662 | int ret; |
3663 | ||
cfbf6377 | 3664 | down_write(&rbd_dev->header_rwsem); |
3b5cf2a2 | 3665 | mapping_size = rbd_dev->mapping.size; |
a720ae09 ID |
3666 | |
3667 | ret = rbd_dev_header_info(rbd_dev); | |
52bb1f9b | 3668 | if (ret) |
73e39e4d | 3669 | goto out; |
15228ede | 3670 | |
e8f59b59 ID |
3671 | /* |
3672 | * If there is a parent, see if it has disappeared due to the | |
3673 | * mapped image getting flattened. | |
3674 | */ | |
3675 | if (rbd_dev->parent) { | |
3676 | ret = rbd_dev_v2_parent_info(rbd_dev); | |
3677 | if (ret) | |
73e39e4d | 3678 | goto out; |
e8f59b59 ID |
3679 | } |
3680 | ||
5ff1108c | 3681 | if (rbd_dev->spec->snap_id == CEPH_NOSNAP) { |
73e39e4d | 3682 | rbd_dev->mapping.size = rbd_dev->header.image_size; |
5ff1108c ID |
3683 | } else { |
3684 | /* validate mapped snapshot's EXISTS flag */ | |
3685 | rbd_exists_validate(rbd_dev); | |
3686 | } | |
15228ede | 3687 | |
73e39e4d | 3688 | out: |
cfbf6377 | 3689 | up_write(&rbd_dev->header_rwsem); |
73e39e4d | 3690 | if (!ret && mapping_size != rbd_dev->mapping.size) |
9875201e | 3691 | rbd_dev_update_size(rbd_dev); |
1fe5e993 | 3692 | |
73e39e4d | 3693 | return ret; |
1fe5e993 AE |
3694 | } |
3695 | ||
7ad18afa CH |
3696 | static int rbd_init_request(void *data, struct request *rq, |
3697 | unsigned int hctx_idx, unsigned int request_idx, | |
3698 | unsigned int numa_node) | |
3699 | { | |
3700 | struct work_struct *work = blk_mq_rq_to_pdu(rq); | |
3701 | ||
3702 | INIT_WORK(work, rbd_queue_workfn); | |
3703 | return 0; | |
3704 | } | |
3705 | ||
3706 | static struct blk_mq_ops rbd_mq_ops = { | |
3707 | .queue_rq = rbd_queue_rq, | |
3708 | .map_queue = blk_mq_map_queue, | |
3709 | .init_request = rbd_init_request, | |
3710 | }; | |
3711 | ||
602adf40 YS |
3712 | static int rbd_init_disk(struct rbd_device *rbd_dev) |
3713 | { | |
3714 | struct gendisk *disk; | |
3715 | struct request_queue *q; | |
593a9e7b | 3716 | u64 segment_size; |
7ad18afa | 3717 | int err; |
602adf40 | 3718 | |
602adf40 | 3719 | /* create gendisk info */ |
7e513d43 ID |
3720 | disk = alloc_disk(single_major ? |
3721 | (1 << RBD_SINGLE_MAJOR_PART_SHIFT) : | |
3722 | RBD_MINORS_PER_MAJOR); | |
602adf40 | 3723 | if (!disk) |
1fcdb8aa | 3724 | return -ENOMEM; |
602adf40 | 3725 | |
f0f8cef5 | 3726 | snprintf(disk->disk_name, sizeof(disk->disk_name), RBD_DRV_NAME "%d", |
de71a297 | 3727 | rbd_dev->dev_id); |
602adf40 | 3728 | disk->major = rbd_dev->major; |
dd82fff1 | 3729 | disk->first_minor = rbd_dev->minor; |
7e513d43 ID |
3730 | if (single_major) |
3731 | disk->flags |= GENHD_FL_EXT_DEVT; | |
602adf40 YS |
3732 | disk->fops = &rbd_bd_ops; |
3733 | disk->private_data = rbd_dev; | |
3734 | ||
7ad18afa CH |
3735 | memset(&rbd_dev->tag_set, 0, sizeof(rbd_dev->tag_set)); |
3736 | rbd_dev->tag_set.ops = &rbd_mq_ops; | |
b5584180 | 3737 | rbd_dev->tag_set.queue_depth = rbd_dev->opts->queue_depth; |
7ad18afa | 3738 | rbd_dev->tag_set.numa_node = NUMA_NO_NODE; |
b5584180 | 3739 | rbd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
7ad18afa CH |
3740 | rbd_dev->tag_set.nr_hw_queues = 1; |
3741 | rbd_dev->tag_set.cmd_size = sizeof(struct work_struct); | |
3742 | ||
3743 | err = blk_mq_alloc_tag_set(&rbd_dev->tag_set); | |
3744 | if (err) | |
602adf40 | 3745 | goto out_disk; |
029bcbd8 | 3746 | |
7ad18afa CH |
3747 | q = blk_mq_init_queue(&rbd_dev->tag_set); |
3748 | if (IS_ERR(q)) { | |
3749 | err = PTR_ERR(q); | |
3750 | goto out_tag_set; | |
3751 | } | |
3752 | ||
d8a2c89c ID |
3753 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); |
3754 | /* QUEUE_FLAG_ADD_RANDOM is off by default for blk-mq */ | |
593a9e7b | 3755 | |
029bcbd8 | 3756 | /* set io sizes to object size */ |
593a9e7b AE |
3757 | segment_size = rbd_obj_bytes(&rbd_dev->header); |
3758 | blk_queue_max_hw_sectors(q, segment_size / SECTOR_SIZE); | |
0d9fde4f | 3759 | q->limits.max_sectors = queue_max_hw_sectors(q); |
d3834fef | 3760 | blk_queue_max_segments(q, segment_size / SECTOR_SIZE); |
593a9e7b AE |
3761 | blk_queue_max_segment_size(q, segment_size); |
3762 | blk_queue_io_min(q, segment_size); | |
3763 | blk_queue_io_opt(q, segment_size); | |
029bcbd8 | 3764 | |
90e98c52 GZ |
3765 | /* enable the discard support */ |
3766 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); | |
3767 | q->limits.discard_granularity = segment_size; | |
3768 | q->limits.discard_alignment = segment_size; | |
2bb4cd5c | 3769 | blk_queue_max_discard_sectors(q, segment_size / SECTOR_SIZE); |
b76f8239 | 3770 | q->limits.discard_zeroes_data = 1; |
90e98c52 | 3771 | |
bae818ee RH |
3772 | if (!ceph_test_opt(rbd_dev->rbd_client->client, NOCRC)) |
3773 | q->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES; | |
3774 | ||
602adf40 YS |
3775 | disk->queue = q; |
3776 | ||
3777 | q->queuedata = rbd_dev; | |
3778 | ||
3779 | rbd_dev->disk = disk; | |
602adf40 | 3780 | |
602adf40 | 3781 | return 0; |
7ad18afa CH |
3782 | out_tag_set: |
3783 | blk_mq_free_tag_set(&rbd_dev->tag_set); | |
602adf40 YS |
3784 | out_disk: |
3785 | put_disk(disk); | |
7ad18afa | 3786 | return err; |
602adf40 YS |
3787 | } |
3788 | ||
dfc5606d YS |
3789 | /* |
3790 | sysfs | |
3791 | */ | |
3792 | ||
593a9e7b AE |
3793 | static struct rbd_device *dev_to_rbd_dev(struct device *dev) |
3794 | { | |
3795 | return container_of(dev, struct rbd_device, dev); | |
3796 | } | |
3797 | ||
dfc5606d YS |
3798 | static ssize_t rbd_size_show(struct device *dev, |
3799 | struct device_attribute *attr, char *buf) | |
3800 | { | |
593a9e7b | 3801 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
a51aa0c0 | 3802 | |
fc71d833 AE |
3803 | return sprintf(buf, "%llu\n", |
3804 | (unsigned long long)rbd_dev->mapping.size); | |
dfc5606d YS |
3805 | } |
3806 | ||
34b13184 AE |
3807 | /* |
3808 | * Note this shows the features for whatever's mapped, which is not | |
3809 | * necessarily the base image. | |
3810 | */ | |
3811 | static ssize_t rbd_features_show(struct device *dev, | |
3812 | struct device_attribute *attr, char *buf) | |
3813 | { | |
3814 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3815 | ||
3816 | return sprintf(buf, "0x%016llx\n", | |
fc71d833 | 3817 | (unsigned long long)rbd_dev->mapping.features); |
34b13184 AE |
3818 | } |
3819 | ||
dfc5606d YS |
3820 | static ssize_t rbd_major_show(struct device *dev, |
3821 | struct device_attribute *attr, char *buf) | |
3822 | { | |
593a9e7b | 3823 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
602adf40 | 3824 | |
fc71d833 AE |
3825 | if (rbd_dev->major) |
3826 | return sprintf(buf, "%d\n", rbd_dev->major); | |
3827 | ||
3828 | return sprintf(buf, "(none)\n"); | |
dd82fff1 ID |
3829 | } |
3830 | ||
3831 | static ssize_t rbd_minor_show(struct device *dev, | |
3832 | struct device_attribute *attr, char *buf) | |
3833 | { | |
3834 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
fc71d833 | 3835 | |
dd82fff1 | 3836 | return sprintf(buf, "%d\n", rbd_dev->minor); |
dfc5606d YS |
3837 | } |
3838 | ||
3839 | static ssize_t rbd_client_id_show(struct device *dev, | |
3840 | struct device_attribute *attr, char *buf) | |
602adf40 | 3841 | { |
593a9e7b | 3842 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3843 | |
1dbb4399 AE |
3844 | return sprintf(buf, "client%lld\n", |
3845 | ceph_client_id(rbd_dev->rbd_client->client)); | |
602adf40 YS |
3846 | } |
3847 | ||
dfc5606d YS |
3848 | static ssize_t rbd_pool_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 | |
0d7dbfce | 3853 | return sprintf(buf, "%s\n", rbd_dev->spec->pool_name); |
dfc5606d YS |
3854 | } |
3855 | ||
9bb2f334 AE |
3856 | static ssize_t rbd_pool_id_show(struct device *dev, |
3857 | struct device_attribute *attr, char *buf) | |
3858 | { | |
3859 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3860 | ||
0d7dbfce | 3861 | return sprintf(buf, "%llu\n", |
fc71d833 | 3862 | (unsigned long long) rbd_dev->spec->pool_id); |
9bb2f334 AE |
3863 | } |
3864 | ||
dfc5606d YS |
3865 | static ssize_t rbd_name_show(struct device *dev, |
3866 | struct device_attribute *attr, char *buf) | |
3867 | { | |
593a9e7b | 3868 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3869 | |
a92ffdf8 AE |
3870 | if (rbd_dev->spec->image_name) |
3871 | return sprintf(buf, "%s\n", rbd_dev->spec->image_name); | |
3872 | ||
3873 | return sprintf(buf, "(unknown)\n"); | |
dfc5606d YS |
3874 | } |
3875 | ||
589d30e0 AE |
3876 | static ssize_t rbd_image_id_show(struct device *dev, |
3877 | struct device_attribute *attr, char *buf) | |
3878 | { | |
3879 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
3880 | ||
0d7dbfce | 3881 | return sprintf(buf, "%s\n", rbd_dev->spec->image_id); |
589d30e0 AE |
3882 | } |
3883 | ||
34b13184 AE |
3884 | /* |
3885 | * Shows the name of the currently-mapped snapshot (or | |
3886 | * RBD_SNAP_HEAD_NAME for the base image). | |
3887 | */ | |
dfc5606d YS |
3888 | static ssize_t rbd_snap_show(struct device *dev, |
3889 | struct device_attribute *attr, | |
3890 | char *buf) | |
3891 | { | |
593a9e7b | 3892 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
dfc5606d | 3893 | |
0d7dbfce | 3894 | return sprintf(buf, "%s\n", rbd_dev->spec->snap_name); |
dfc5606d YS |
3895 | } |
3896 | ||
86b00e0d | 3897 | /* |
ff96128f ID |
3898 | * For a v2 image, shows the chain of parent images, separated by empty |
3899 | * lines. For v1 images or if there is no parent, shows "(no parent | |
3900 | * image)". | |
86b00e0d AE |
3901 | */ |
3902 | static ssize_t rbd_parent_show(struct device *dev, | |
ff96128f ID |
3903 | struct device_attribute *attr, |
3904 | char *buf) | |
86b00e0d AE |
3905 | { |
3906 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
ff96128f | 3907 | ssize_t count = 0; |
86b00e0d | 3908 | |
ff96128f | 3909 | if (!rbd_dev->parent) |
86b00e0d AE |
3910 | return sprintf(buf, "(no parent image)\n"); |
3911 | ||
ff96128f ID |
3912 | for ( ; rbd_dev->parent; rbd_dev = rbd_dev->parent) { |
3913 | struct rbd_spec *spec = rbd_dev->parent_spec; | |
3914 | ||
3915 | count += sprintf(&buf[count], "%s" | |
3916 | "pool_id %llu\npool_name %s\n" | |
3917 | "image_id %s\nimage_name %s\n" | |
3918 | "snap_id %llu\nsnap_name %s\n" | |
3919 | "overlap %llu\n", | |
3920 | !count ? "" : "\n", /* first? */ | |
3921 | spec->pool_id, spec->pool_name, | |
3922 | spec->image_id, spec->image_name ?: "(unknown)", | |
3923 | spec->snap_id, spec->snap_name, | |
3924 | rbd_dev->parent_overlap); | |
3925 | } | |
3926 | ||
3927 | return count; | |
86b00e0d AE |
3928 | } |
3929 | ||
dfc5606d YS |
3930 | static ssize_t rbd_image_refresh(struct device *dev, |
3931 | struct device_attribute *attr, | |
3932 | const char *buf, | |
3933 | size_t size) | |
3934 | { | |
593a9e7b | 3935 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); |
b813623a | 3936 | int ret; |
602adf40 | 3937 | |
cc4a38bd | 3938 | ret = rbd_dev_refresh(rbd_dev); |
e627db08 | 3939 | if (ret) |
52bb1f9b | 3940 | return ret; |
b813623a | 3941 | |
52bb1f9b | 3942 | return size; |
dfc5606d | 3943 | } |
602adf40 | 3944 | |
dfc5606d | 3945 | static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL); |
34b13184 | 3946 | static DEVICE_ATTR(features, S_IRUGO, rbd_features_show, NULL); |
dfc5606d | 3947 | static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL); |
dd82fff1 | 3948 | static DEVICE_ATTR(minor, S_IRUGO, rbd_minor_show, NULL); |
dfc5606d YS |
3949 | static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL); |
3950 | static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL); | |
9bb2f334 | 3951 | static DEVICE_ATTR(pool_id, S_IRUGO, rbd_pool_id_show, NULL); |
dfc5606d | 3952 | static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL); |
589d30e0 | 3953 | static DEVICE_ATTR(image_id, S_IRUGO, rbd_image_id_show, NULL); |
dfc5606d YS |
3954 | static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh); |
3955 | static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL); | |
86b00e0d | 3956 | static DEVICE_ATTR(parent, S_IRUGO, rbd_parent_show, NULL); |
dfc5606d YS |
3957 | |
3958 | static struct attribute *rbd_attrs[] = { | |
3959 | &dev_attr_size.attr, | |
34b13184 | 3960 | &dev_attr_features.attr, |
dfc5606d | 3961 | &dev_attr_major.attr, |
dd82fff1 | 3962 | &dev_attr_minor.attr, |
dfc5606d YS |
3963 | &dev_attr_client_id.attr, |
3964 | &dev_attr_pool.attr, | |
9bb2f334 | 3965 | &dev_attr_pool_id.attr, |
dfc5606d | 3966 | &dev_attr_name.attr, |
589d30e0 | 3967 | &dev_attr_image_id.attr, |
dfc5606d | 3968 | &dev_attr_current_snap.attr, |
86b00e0d | 3969 | &dev_attr_parent.attr, |
dfc5606d | 3970 | &dev_attr_refresh.attr, |
dfc5606d YS |
3971 | NULL |
3972 | }; | |
3973 | ||
3974 | static struct attribute_group rbd_attr_group = { | |
3975 | .attrs = rbd_attrs, | |
3976 | }; | |
3977 | ||
3978 | static const struct attribute_group *rbd_attr_groups[] = { | |
3979 | &rbd_attr_group, | |
3980 | NULL | |
3981 | }; | |
3982 | ||
6cac4695 | 3983 | static void rbd_dev_release(struct device *dev); |
dfc5606d YS |
3984 | |
3985 | static struct device_type rbd_device_type = { | |
3986 | .name = "rbd", | |
3987 | .groups = rbd_attr_groups, | |
6cac4695 | 3988 | .release = rbd_dev_release, |
dfc5606d YS |
3989 | }; |
3990 | ||
8b8fb99c AE |
3991 | static struct rbd_spec *rbd_spec_get(struct rbd_spec *spec) |
3992 | { | |
3993 | kref_get(&spec->kref); | |
3994 | ||
3995 | return spec; | |
3996 | } | |
3997 | ||
3998 | static void rbd_spec_free(struct kref *kref); | |
3999 | static void rbd_spec_put(struct rbd_spec *spec) | |
4000 | { | |
4001 | if (spec) | |
4002 | kref_put(&spec->kref, rbd_spec_free); | |
4003 | } | |
4004 | ||
4005 | static struct rbd_spec *rbd_spec_alloc(void) | |
4006 | { | |
4007 | struct rbd_spec *spec; | |
4008 | ||
4009 | spec = kzalloc(sizeof (*spec), GFP_KERNEL); | |
4010 | if (!spec) | |
4011 | return NULL; | |
04077599 ID |
4012 | |
4013 | spec->pool_id = CEPH_NOPOOL; | |
4014 | spec->snap_id = CEPH_NOSNAP; | |
8b8fb99c AE |
4015 | kref_init(&spec->kref); |
4016 | ||
8b8fb99c AE |
4017 | return spec; |
4018 | } | |
4019 | ||
4020 | static void rbd_spec_free(struct kref *kref) | |
4021 | { | |
4022 | struct rbd_spec *spec = container_of(kref, struct rbd_spec, kref); | |
4023 | ||
4024 | kfree(spec->pool_name); | |
4025 | kfree(spec->image_id); | |
4026 | kfree(spec->image_name); | |
4027 | kfree(spec->snap_name); | |
4028 | kfree(spec); | |
4029 | } | |
4030 | ||
dd5ac32d ID |
4031 | static void rbd_dev_release(struct device *dev) |
4032 | { | |
4033 | struct rbd_device *rbd_dev = dev_to_rbd_dev(dev); | |
4034 | bool need_put = !!rbd_dev->opts; | |
4035 | ||
4036 | rbd_put_client(rbd_dev->rbd_client); | |
4037 | rbd_spec_put(rbd_dev->spec); | |
4038 | kfree(rbd_dev->opts); | |
4039 | kfree(rbd_dev); | |
4040 | ||
4041 | /* | |
4042 | * This is racy, but way better than putting module outside of | |
4043 | * the release callback. The race window is pretty small, so | |
4044 | * doing something similar to dm (dm-builtin.c) is overkill. | |
4045 | */ | |
4046 | if (need_put) | |
4047 | module_put(THIS_MODULE); | |
4048 | } | |
4049 | ||
cc344fa1 | 4050 | static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc, |
d147543d ID |
4051 | struct rbd_spec *spec, |
4052 | struct rbd_options *opts) | |
c53d5893 AE |
4053 | { |
4054 | struct rbd_device *rbd_dev; | |
4055 | ||
4056 | rbd_dev = kzalloc(sizeof (*rbd_dev), GFP_KERNEL); | |
4057 | if (!rbd_dev) | |
4058 | return NULL; | |
4059 | ||
4060 | spin_lock_init(&rbd_dev->lock); | |
6d292906 | 4061 | rbd_dev->flags = 0; |
a2acd00e | 4062 | atomic_set(&rbd_dev->parent_ref, 0); |
c53d5893 | 4063 | INIT_LIST_HEAD(&rbd_dev->node); |
c53d5893 AE |
4064 | init_rwsem(&rbd_dev->header_rwsem); |
4065 | ||
dd5ac32d ID |
4066 | rbd_dev->dev.bus = &rbd_bus_type; |
4067 | rbd_dev->dev.type = &rbd_device_type; | |
4068 | rbd_dev->dev.parent = &rbd_root_dev; | |
dd5ac32d ID |
4069 | device_initialize(&rbd_dev->dev); |
4070 | ||
c53d5893 | 4071 | rbd_dev->rbd_client = rbdc; |
d147543d ID |
4072 | rbd_dev->spec = spec; |
4073 | rbd_dev->opts = opts; | |
c53d5893 | 4074 | |
0903e875 AE |
4075 | /* Initialize the layout used for all rbd requests */ |
4076 | ||
4077 | rbd_dev->layout.fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4078 | rbd_dev->layout.fl_stripe_count = cpu_to_le32(1); | |
4079 | rbd_dev->layout.fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER); | |
4080 | rbd_dev->layout.fl_pg_pool = cpu_to_le32((u32) spec->pool_id); | |
4081 | ||
dd5ac32d ID |
4082 | /* |
4083 | * If this is a mapping rbd_dev (as opposed to a parent one), | |
4084 | * pin our module. We have a ref from do_rbd_add(), so use | |
4085 | * __module_get(). | |
4086 | */ | |
4087 | if (rbd_dev->opts) | |
4088 | __module_get(THIS_MODULE); | |
4089 | ||
c53d5893 AE |
4090 | return rbd_dev; |
4091 | } | |
4092 | ||
4093 | static void rbd_dev_destroy(struct rbd_device *rbd_dev) | |
4094 | { | |
dd5ac32d ID |
4095 | if (rbd_dev) |
4096 | put_device(&rbd_dev->dev); | |
c53d5893 AE |
4097 | } |
4098 | ||
9d475de5 AE |
4099 | /* |
4100 | * Get the size and object order for an image snapshot, or if | |
4101 | * snap_id is CEPH_NOSNAP, gets this information for the base | |
4102 | * image. | |
4103 | */ | |
4104 | static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id, | |
4105 | u8 *order, u64 *snap_size) | |
4106 | { | |
4107 | __le64 snapid = cpu_to_le64(snap_id); | |
4108 | int ret; | |
4109 | struct { | |
4110 | u8 order; | |
4111 | __le64 size; | |
4112 | } __attribute__ ((packed)) size_buf = { 0 }; | |
4113 | ||
36be9a76 | 4114 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
9d475de5 | 4115 | "rbd", "get_size", |
4157976b | 4116 | &snapid, sizeof (snapid), |
e2a58ee5 | 4117 | &size_buf, sizeof (size_buf)); |
36be9a76 | 4118 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
9d475de5 AE |
4119 | if (ret < 0) |
4120 | return ret; | |
57385b51 AE |
4121 | if (ret < sizeof (size_buf)) |
4122 | return -ERANGE; | |
9d475de5 | 4123 | |
c3545579 | 4124 | if (order) { |
c86f86e9 | 4125 | *order = size_buf.order; |
c3545579 JD |
4126 | dout(" order %u", (unsigned int)*order); |
4127 | } | |
9d475de5 AE |
4128 | *snap_size = le64_to_cpu(size_buf.size); |
4129 | ||
c3545579 JD |
4130 | dout(" snap_id 0x%016llx snap_size = %llu\n", |
4131 | (unsigned long long)snap_id, | |
57385b51 | 4132 | (unsigned long long)*snap_size); |
9d475de5 AE |
4133 | |
4134 | return 0; | |
4135 | } | |
4136 | ||
4137 | static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev) | |
4138 | { | |
4139 | return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP, | |
4140 | &rbd_dev->header.obj_order, | |
4141 | &rbd_dev->header.image_size); | |
4142 | } | |
4143 | ||
1e130199 AE |
4144 | static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev) |
4145 | { | |
4146 | void *reply_buf; | |
4147 | int ret; | |
4148 | void *p; | |
4149 | ||
4150 | reply_buf = kzalloc(RBD_OBJ_PREFIX_LEN_MAX, GFP_KERNEL); | |
4151 | if (!reply_buf) | |
4152 | return -ENOMEM; | |
4153 | ||
36be9a76 | 4154 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4155 | "rbd", "get_object_prefix", NULL, 0, |
e2a58ee5 | 4156 | reply_buf, RBD_OBJ_PREFIX_LEN_MAX); |
36be9a76 | 4157 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
1e130199 AE |
4158 | if (ret < 0) |
4159 | goto out; | |
4160 | ||
4161 | p = reply_buf; | |
4162 | rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p, | |
57385b51 AE |
4163 | p + ret, NULL, GFP_NOIO); |
4164 | ret = 0; | |
1e130199 AE |
4165 | |
4166 | if (IS_ERR(rbd_dev->header.object_prefix)) { | |
4167 | ret = PTR_ERR(rbd_dev->header.object_prefix); | |
4168 | rbd_dev->header.object_prefix = NULL; | |
4169 | } else { | |
4170 | dout(" object_prefix = %s\n", rbd_dev->header.object_prefix); | |
4171 | } | |
1e130199 AE |
4172 | out: |
4173 | kfree(reply_buf); | |
4174 | ||
4175 | return ret; | |
4176 | } | |
4177 | ||
b1b5402a AE |
4178 | static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id, |
4179 | u64 *snap_features) | |
4180 | { | |
4181 | __le64 snapid = cpu_to_le64(snap_id); | |
4182 | struct { | |
4183 | __le64 features; | |
4184 | __le64 incompat; | |
4157976b | 4185 | } __attribute__ ((packed)) features_buf = { 0 }; |
d3767f0f | 4186 | u64 unsup; |
b1b5402a AE |
4187 | int ret; |
4188 | ||
36be9a76 | 4189 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b1b5402a | 4190 | "rbd", "get_features", |
4157976b | 4191 | &snapid, sizeof (snapid), |
e2a58ee5 | 4192 | &features_buf, sizeof (features_buf)); |
36be9a76 | 4193 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
b1b5402a AE |
4194 | if (ret < 0) |
4195 | return ret; | |
57385b51 AE |
4196 | if (ret < sizeof (features_buf)) |
4197 | return -ERANGE; | |
d889140c | 4198 | |
d3767f0f ID |
4199 | unsup = le64_to_cpu(features_buf.incompat) & ~RBD_FEATURES_SUPPORTED; |
4200 | if (unsup) { | |
4201 | rbd_warn(rbd_dev, "image uses unsupported features: 0x%llx", | |
4202 | unsup); | |
b8f5c6ed | 4203 | return -ENXIO; |
d3767f0f | 4204 | } |
d889140c | 4205 | |
b1b5402a AE |
4206 | *snap_features = le64_to_cpu(features_buf.features); |
4207 | ||
4208 | dout(" snap_id 0x%016llx features = 0x%016llx incompat = 0x%016llx\n", | |
57385b51 AE |
4209 | (unsigned long long)snap_id, |
4210 | (unsigned long long)*snap_features, | |
4211 | (unsigned long long)le64_to_cpu(features_buf.incompat)); | |
b1b5402a AE |
4212 | |
4213 | return 0; | |
4214 | } | |
4215 | ||
4216 | static int rbd_dev_v2_features(struct rbd_device *rbd_dev) | |
4217 | { | |
4218 | return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP, | |
4219 | &rbd_dev->header.features); | |
4220 | } | |
4221 | ||
86b00e0d AE |
4222 | static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) |
4223 | { | |
4224 | struct rbd_spec *parent_spec; | |
4225 | size_t size; | |
4226 | void *reply_buf = NULL; | |
4227 | __le64 snapid; | |
4228 | void *p; | |
4229 | void *end; | |
642a2537 | 4230 | u64 pool_id; |
86b00e0d | 4231 | char *image_id; |
3b5cf2a2 | 4232 | u64 snap_id; |
86b00e0d | 4233 | u64 overlap; |
86b00e0d AE |
4234 | int ret; |
4235 | ||
4236 | parent_spec = rbd_spec_alloc(); | |
4237 | if (!parent_spec) | |
4238 | return -ENOMEM; | |
4239 | ||
4240 | size = sizeof (__le64) + /* pool_id */ | |
4241 | sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX + /* image_id */ | |
4242 | sizeof (__le64) + /* snap_id */ | |
4243 | sizeof (__le64); /* overlap */ | |
4244 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4245 | if (!reply_buf) { | |
4246 | ret = -ENOMEM; | |
4247 | goto out_err; | |
4248 | } | |
4249 | ||
4d9b67cd | 4250 | snapid = cpu_to_le64(rbd_dev->spec->snap_id); |
36be9a76 | 4251 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
86b00e0d | 4252 | "rbd", "get_parent", |
4157976b | 4253 | &snapid, sizeof (snapid), |
e2a58ee5 | 4254 | reply_buf, size); |
36be9a76 | 4255 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
86b00e0d AE |
4256 | if (ret < 0) |
4257 | goto out_err; | |
4258 | ||
86b00e0d | 4259 | p = reply_buf; |
57385b51 AE |
4260 | end = reply_buf + ret; |
4261 | ret = -ERANGE; | |
642a2537 | 4262 | ceph_decode_64_safe(&p, end, pool_id, out_err); |
392a9dad AE |
4263 | if (pool_id == CEPH_NOPOOL) { |
4264 | /* | |
4265 | * Either the parent never existed, or we have | |
4266 | * record of it but the image got flattened so it no | |
4267 | * longer has a parent. When the parent of a | |
4268 | * layered image disappears we immediately set the | |
4269 | * overlap to 0. The effect of this is that all new | |
4270 | * requests will be treated as if the image had no | |
4271 | * parent. | |
4272 | */ | |
4273 | if (rbd_dev->parent_overlap) { | |
4274 | rbd_dev->parent_overlap = 0; | |
392a9dad AE |
4275 | rbd_dev_parent_put(rbd_dev); |
4276 | pr_info("%s: clone image has been flattened\n", | |
4277 | rbd_dev->disk->disk_name); | |
4278 | } | |
4279 | ||
86b00e0d | 4280 | goto out; /* No parent? No problem. */ |
392a9dad | 4281 | } |
86b00e0d | 4282 | |
0903e875 AE |
4283 | /* The ceph file layout needs to fit pool id in 32 bits */ |
4284 | ||
4285 | ret = -EIO; | |
642a2537 | 4286 | if (pool_id > (u64)U32_MAX) { |
9584d508 | 4287 | rbd_warn(NULL, "parent pool id too large (%llu > %u)", |
642a2537 | 4288 | (unsigned long long)pool_id, U32_MAX); |
57385b51 | 4289 | goto out_err; |
c0cd10db | 4290 | } |
0903e875 | 4291 | |
979ed480 | 4292 | image_id = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
86b00e0d AE |
4293 | if (IS_ERR(image_id)) { |
4294 | ret = PTR_ERR(image_id); | |
4295 | goto out_err; | |
4296 | } | |
3b5cf2a2 | 4297 | ceph_decode_64_safe(&p, end, snap_id, out_err); |
86b00e0d AE |
4298 | ceph_decode_64_safe(&p, end, overlap, out_err); |
4299 | ||
3b5cf2a2 AE |
4300 | /* |
4301 | * The parent won't change (except when the clone is | |
4302 | * flattened, already handled that). So we only need to | |
4303 | * record the parent spec we have not already done so. | |
4304 | */ | |
4305 | if (!rbd_dev->parent_spec) { | |
4306 | parent_spec->pool_id = pool_id; | |
4307 | parent_spec->image_id = image_id; | |
4308 | parent_spec->snap_id = snap_id; | |
70cf49cf AE |
4309 | rbd_dev->parent_spec = parent_spec; |
4310 | parent_spec = NULL; /* rbd_dev now owns this */ | |
fbba11b3 ID |
4311 | } else { |
4312 | kfree(image_id); | |
3b5cf2a2 AE |
4313 | } |
4314 | ||
4315 | /* | |
cf32bd9c ID |
4316 | * We always update the parent overlap. If it's zero we issue |
4317 | * a warning, as we will proceed as if there was no parent. | |
3b5cf2a2 | 4318 | */ |
3b5cf2a2 | 4319 | if (!overlap) { |
3b5cf2a2 | 4320 | if (parent_spec) { |
cf32bd9c ID |
4321 | /* refresh, careful to warn just once */ |
4322 | if (rbd_dev->parent_overlap) | |
4323 | rbd_warn(rbd_dev, | |
4324 | "clone now standalone (overlap became 0)"); | |
3b5cf2a2 | 4325 | } else { |
cf32bd9c ID |
4326 | /* initial probe */ |
4327 | rbd_warn(rbd_dev, "clone is standalone (overlap 0)"); | |
3b5cf2a2 | 4328 | } |
70cf49cf | 4329 | } |
cf32bd9c ID |
4330 | rbd_dev->parent_overlap = overlap; |
4331 | ||
86b00e0d AE |
4332 | out: |
4333 | ret = 0; | |
4334 | out_err: | |
4335 | kfree(reply_buf); | |
4336 | rbd_spec_put(parent_spec); | |
4337 | ||
4338 | return ret; | |
4339 | } | |
4340 | ||
cc070d59 AE |
4341 | static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev) |
4342 | { | |
4343 | struct { | |
4344 | __le64 stripe_unit; | |
4345 | __le64 stripe_count; | |
4346 | } __attribute__ ((packed)) striping_info_buf = { 0 }; | |
4347 | size_t size = sizeof (striping_info_buf); | |
4348 | void *p; | |
4349 | u64 obj_size; | |
4350 | u64 stripe_unit; | |
4351 | u64 stripe_count; | |
4352 | int ret; | |
4353 | ||
4354 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, | |
4355 | "rbd", "get_stripe_unit_count", NULL, 0, | |
e2a58ee5 | 4356 | (char *)&striping_info_buf, size); |
cc070d59 AE |
4357 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
4358 | if (ret < 0) | |
4359 | return ret; | |
4360 | if (ret < size) | |
4361 | return -ERANGE; | |
4362 | ||
4363 | /* | |
4364 | * We don't actually support the "fancy striping" feature | |
4365 | * (STRIPINGV2) yet, but if the striping sizes are the | |
4366 | * defaults the behavior is the same as before. So find | |
4367 | * out, and only fail if the image has non-default values. | |
4368 | */ | |
4369 | ret = -EINVAL; | |
4370 | obj_size = (u64)1 << rbd_dev->header.obj_order; | |
4371 | p = &striping_info_buf; | |
4372 | stripe_unit = ceph_decode_64(&p); | |
4373 | if (stripe_unit != obj_size) { | |
4374 | rbd_warn(rbd_dev, "unsupported stripe unit " | |
4375 | "(got %llu want %llu)", | |
4376 | stripe_unit, obj_size); | |
4377 | return -EINVAL; | |
4378 | } | |
4379 | stripe_count = ceph_decode_64(&p); | |
4380 | if (stripe_count != 1) { | |
4381 | rbd_warn(rbd_dev, "unsupported stripe count " | |
4382 | "(got %llu want 1)", stripe_count); | |
4383 | return -EINVAL; | |
4384 | } | |
500d0c0f AE |
4385 | rbd_dev->header.stripe_unit = stripe_unit; |
4386 | rbd_dev->header.stripe_count = stripe_count; | |
cc070d59 AE |
4387 | |
4388 | return 0; | |
4389 | } | |
4390 | ||
9e15b77d AE |
4391 | static char *rbd_dev_image_name(struct rbd_device *rbd_dev) |
4392 | { | |
4393 | size_t image_id_size; | |
4394 | char *image_id; | |
4395 | void *p; | |
4396 | void *end; | |
4397 | size_t size; | |
4398 | void *reply_buf = NULL; | |
4399 | size_t len = 0; | |
4400 | char *image_name = NULL; | |
4401 | int ret; | |
4402 | ||
4403 | rbd_assert(!rbd_dev->spec->image_name); | |
4404 | ||
69e7a02f AE |
4405 | len = strlen(rbd_dev->spec->image_id); |
4406 | image_id_size = sizeof (__le32) + len; | |
9e15b77d AE |
4407 | image_id = kmalloc(image_id_size, GFP_KERNEL); |
4408 | if (!image_id) | |
4409 | return NULL; | |
4410 | ||
4411 | p = image_id; | |
4157976b | 4412 | end = image_id + image_id_size; |
57385b51 | 4413 | ceph_encode_string(&p, end, rbd_dev->spec->image_id, (u32)len); |
9e15b77d AE |
4414 | |
4415 | size = sizeof (__le32) + RBD_IMAGE_NAME_LEN_MAX; | |
4416 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4417 | if (!reply_buf) | |
4418 | goto out; | |
4419 | ||
36be9a76 | 4420 | ret = rbd_obj_method_sync(rbd_dev, RBD_DIRECTORY, |
9e15b77d AE |
4421 | "rbd", "dir_get_name", |
4422 | image_id, image_id_size, | |
e2a58ee5 | 4423 | reply_buf, size); |
9e15b77d AE |
4424 | if (ret < 0) |
4425 | goto out; | |
4426 | p = reply_buf; | |
f40eb349 AE |
4427 | end = reply_buf + ret; |
4428 | ||
9e15b77d AE |
4429 | image_name = ceph_extract_encoded_string(&p, end, &len, GFP_KERNEL); |
4430 | if (IS_ERR(image_name)) | |
4431 | image_name = NULL; | |
4432 | else | |
4433 | dout("%s: name is %s len is %zd\n", __func__, image_name, len); | |
4434 | out: | |
4435 | kfree(reply_buf); | |
4436 | kfree(image_id); | |
4437 | ||
4438 | return image_name; | |
4439 | } | |
4440 | ||
2ad3d716 AE |
4441 | static u64 rbd_v1_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) |
4442 | { | |
4443 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4444 | const char *snap_name; | |
4445 | u32 which = 0; | |
4446 | ||
4447 | /* Skip over names until we find the one we are looking for */ | |
4448 | ||
4449 | snap_name = rbd_dev->header.snap_names; | |
4450 | while (which < snapc->num_snaps) { | |
4451 | if (!strcmp(name, snap_name)) | |
4452 | return snapc->snaps[which]; | |
4453 | snap_name += strlen(snap_name) + 1; | |
4454 | which++; | |
4455 | } | |
4456 | return CEPH_NOSNAP; | |
4457 | } | |
4458 | ||
4459 | static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4460 | { | |
4461 | struct ceph_snap_context *snapc = rbd_dev->header.snapc; | |
4462 | u32 which; | |
4463 | bool found = false; | |
4464 | u64 snap_id; | |
4465 | ||
4466 | for (which = 0; !found && which < snapc->num_snaps; which++) { | |
4467 | const char *snap_name; | |
4468 | ||
4469 | snap_id = snapc->snaps[which]; | |
4470 | snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id); | |
efadc98a JD |
4471 | if (IS_ERR(snap_name)) { |
4472 | /* ignore no-longer existing snapshots */ | |
4473 | if (PTR_ERR(snap_name) == -ENOENT) | |
4474 | continue; | |
4475 | else | |
4476 | break; | |
4477 | } | |
2ad3d716 AE |
4478 | found = !strcmp(name, snap_name); |
4479 | kfree(snap_name); | |
4480 | } | |
4481 | return found ? snap_id : CEPH_NOSNAP; | |
4482 | } | |
4483 | ||
4484 | /* | |
4485 | * Assumes name is never RBD_SNAP_HEAD_NAME; returns CEPH_NOSNAP if | |
4486 | * no snapshot by that name is found, or if an error occurs. | |
4487 | */ | |
4488 | static u64 rbd_snap_id_by_name(struct rbd_device *rbd_dev, const char *name) | |
4489 | { | |
4490 | if (rbd_dev->image_format == 1) | |
4491 | return rbd_v1_snap_id_by_name(rbd_dev, name); | |
4492 | ||
4493 | return rbd_v2_snap_id_by_name(rbd_dev, name); | |
4494 | } | |
4495 | ||
9e15b77d | 4496 | /* |
04077599 ID |
4497 | * An image being mapped will have everything but the snap id. |
4498 | */ | |
4499 | static int rbd_spec_fill_snap_id(struct rbd_device *rbd_dev) | |
4500 | { | |
4501 | struct rbd_spec *spec = rbd_dev->spec; | |
4502 | ||
4503 | rbd_assert(spec->pool_id != CEPH_NOPOOL && spec->pool_name); | |
4504 | rbd_assert(spec->image_id && spec->image_name); | |
4505 | rbd_assert(spec->snap_name); | |
4506 | ||
4507 | if (strcmp(spec->snap_name, RBD_SNAP_HEAD_NAME)) { | |
4508 | u64 snap_id; | |
4509 | ||
4510 | snap_id = rbd_snap_id_by_name(rbd_dev, spec->snap_name); | |
4511 | if (snap_id == CEPH_NOSNAP) | |
4512 | return -ENOENT; | |
4513 | ||
4514 | spec->snap_id = snap_id; | |
4515 | } else { | |
4516 | spec->snap_id = CEPH_NOSNAP; | |
4517 | } | |
4518 | ||
4519 | return 0; | |
4520 | } | |
4521 | ||
4522 | /* | |
4523 | * A parent image will have all ids but none of the names. | |
e1d4213f | 4524 | * |
04077599 ID |
4525 | * All names in an rbd spec are dynamically allocated. It's OK if we |
4526 | * can't figure out the name for an image id. | |
9e15b77d | 4527 | */ |
04077599 | 4528 | static int rbd_spec_fill_names(struct rbd_device *rbd_dev) |
9e15b77d | 4529 | { |
2e9f7f1c AE |
4530 | struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; |
4531 | struct rbd_spec *spec = rbd_dev->spec; | |
4532 | const char *pool_name; | |
4533 | const char *image_name; | |
4534 | const char *snap_name; | |
9e15b77d AE |
4535 | int ret; |
4536 | ||
04077599 ID |
4537 | rbd_assert(spec->pool_id != CEPH_NOPOOL); |
4538 | rbd_assert(spec->image_id); | |
4539 | rbd_assert(spec->snap_id != CEPH_NOSNAP); | |
9e15b77d | 4540 | |
2e9f7f1c | 4541 | /* Get the pool name; we have to make our own copy of this */ |
9e15b77d | 4542 | |
2e9f7f1c AE |
4543 | pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, spec->pool_id); |
4544 | if (!pool_name) { | |
4545 | rbd_warn(rbd_dev, "no pool with id %llu", spec->pool_id); | |
935dc89f AE |
4546 | return -EIO; |
4547 | } | |
2e9f7f1c AE |
4548 | pool_name = kstrdup(pool_name, GFP_KERNEL); |
4549 | if (!pool_name) | |
9e15b77d AE |
4550 | return -ENOMEM; |
4551 | ||
4552 | /* Fetch the image name; tolerate failure here */ | |
4553 | ||
2e9f7f1c AE |
4554 | image_name = rbd_dev_image_name(rbd_dev); |
4555 | if (!image_name) | |
06ecc6cb | 4556 | rbd_warn(rbd_dev, "unable to get image name"); |
9e15b77d | 4557 | |
04077599 | 4558 | /* Fetch the snapshot name */ |
9e15b77d | 4559 | |
2e9f7f1c | 4560 | snap_name = rbd_snap_name(rbd_dev, spec->snap_id); |
da6a6b63 JD |
4561 | if (IS_ERR(snap_name)) { |
4562 | ret = PTR_ERR(snap_name); | |
9e15b77d | 4563 | goto out_err; |
2e9f7f1c AE |
4564 | } |
4565 | ||
4566 | spec->pool_name = pool_name; | |
4567 | spec->image_name = image_name; | |
4568 | spec->snap_name = snap_name; | |
9e15b77d AE |
4569 | |
4570 | return 0; | |
04077599 | 4571 | |
9e15b77d | 4572 | out_err: |
2e9f7f1c AE |
4573 | kfree(image_name); |
4574 | kfree(pool_name); | |
9e15b77d AE |
4575 | return ret; |
4576 | } | |
4577 | ||
cc4a38bd | 4578 | static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev) |
35d489f9 AE |
4579 | { |
4580 | size_t size; | |
4581 | int ret; | |
4582 | void *reply_buf; | |
4583 | void *p; | |
4584 | void *end; | |
4585 | u64 seq; | |
4586 | u32 snap_count; | |
4587 | struct ceph_snap_context *snapc; | |
4588 | u32 i; | |
4589 | ||
4590 | /* | |
4591 | * We'll need room for the seq value (maximum snapshot id), | |
4592 | * snapshot count, and array of that many snapshot ids. | |
4593 | * For now we have a fixed upper limit on the number we're | |
4594 | * prepared to receive. | |
4595 | */ | |
4596 | size = sizeof (__le64) + sizeof (__le32) + | |
4597 | RBD_MAX_SNAP_COUNT * sizeof (__le64); | |
4598 | reply_buf = kzalloc(size, GFP_KERNEL); | |
4599 | if (!reply_buf) | |
4600 | return -ENOMEM; | |
4601 | ||
36be9a76 | 4602 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
4157976b | 4603 | "rbd", "get_snapcontext", NULL, 0, |
e2a58ee5 | 4604 | reply_buf, size); |
36be9a76 | 4605 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
35d489f9 AE |
4606 | if (ret < 0) |
4607 | goto out; | |
4608 | ||
35d489f9 | 4609 | p = reply_buf; |
57385b51 AE |
4610 | end = reply_buf + ret; |
4611 | ret = -ERANGE; | |
35d489f9 AE |
4612 | ceph_decode_64_safe(&p, end, seq, out); |
4613 | ceph_decode_32_safe(&p, end, snap_count, out); | |
4614 | ||
4615 | /* | |
4616 | * Make sure the reported number of snapshot ids wouldn't go | |
4617 | * beyond the end of our buffer. But before checking that, | |
4618 | * make sure the computed size of the snapshot context we | |
4619 | * allocate is representable in a size_t. | |
4620 | */ | |
4621 | if (snap_count > (SIZE_MAX - sizeof (struct ceph_snap_context)) | |
4622 | / sizeof (u64)) { | |
4623 | ret = -EINVAL; | |
4624 | goto out; | |
4625 | } | |
4626 | if (!ceph_has_room(&p, end, snap_count * sizeof (__le64))) | |
4627 | goto out; | |
468521c1 | 4628 | ret = 0; |
35d489f9 | 4629 | |
812164f8 | 4630 | snapc = ceph_create_snap_context(snap_count, GFP_KERNEL); |
35d489f9 AE |
4631 | if (!snapc) { |
4632 | ret = -ENOMEM; | |
4633 | goto out; | |
4634 | } | |
35d489f9 | 4635 | snapc->seq = seq; |
35d489f9 AE |
4636 | for (i = 0; i < snap_count; i++) |
4637 | snapc->snaps[i] = ceph_decode_64(&p); | |
4638 | ||
49ece554 | 4639 | ceph_put_snap_context(rbd_dev->header.snapc); |
35d489f9 AE |
4640 | rbd_dev->header.snapc = snapc; |
4641 | ||
4642 | dout(" snap context seq = %llu, snap_count = %u\n", | |
57385b51 | 4643 | (unsigned long long)seq, (unsigned int)snap_count); |
35d489f9 AE |
4644 | out: |
4645 | kfree(reply_buf); | |
4646 | ||
57385b51 | 4647 | return ret; |
35d489f9 AE |
4648 | } |
4649 | ||
54cac61f AE |
4650 | static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev, |
4651 | u64 snap_id) | |
b8b1e2db AE |
4652 | { |
4653 | size_t size; | |
4654 | void *reply_buf; | |
54cac61f | 4655 | __le64 snapid; |
b8b1e2db AE |
4656 | int ret; |
4657 | void *p; | |
4658 | void *end; | |
b8b1e2db AE |
4659 | char *snap_name; |
4660 | ||
4661 | size = sizeof (__le32) + RBD_MAX_SNAP_NAME_LEN; | |
4662 | reply_buf = kmalloc(size, GFP_KERNEL); | |
4663 | if (!reply_buf) | |
4664 | return ERR_PTR(-ENOMEM); | |
4665 | ||
54cac61f | 4666 | snapid = cpu_to_le64(snap_id); |
36be9a76 | 4667 | ret = rbd_obj_method_sync(rbd_dev, rbd_dev->header_name, |
b8b1e2db | 4668 | "rbd", "get_snapshot_name", |
54cac61f | 4669 | &snapid, sizeof (snapid), |
e2a58ee5 | 4670 | reply_buf, size); |
36be9a76 | 4671 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
f40eb349 AE |
4672 | if (ret < 0) { |
4673 | snap_name = ERR_PTR(ret); | |
b8b1e2db | 4674 | goto out; |
f40eb349 | 4675 | } |
b8b1e2db AE |
4676 | |
4677 | p = reply_buf; | |
f40eb349 | 4678 | end = reply_buf + ret; |
e5c35534 | 4679 | snap_name = ceph_extract_encoded_string(&p, end, NULL, GFP_KERNEL); |
f40eb349 | 4680 | if (IS_ERR(snap_name)) |
b8b1e2db | 4681 | goto out; |
b8b1e2db | 4682 | |
f40eb349 | 4683 | dout(" snap_id 0x%016llx snap_name = %s\n", |
54cac61f | 4684 | (unsigned long long)snap_id, snap_name); |
b8b1e2db AE |
4685 | out: |
4686 | kfree(reply_buf); | |
4687 | ||
f40eb349 | 4688 | return snap_name; |
b8b1e2db AE |
4689 | } |
4690 | ||
2df3fac7 | 4691 | static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev) |
117973fb | 4692 | { |
2df3fac7 | 4693 | bool first_time = rbd_dev->header.object_prefix == NULL; |
117973fb | 4694 | int ret; |
117973fb | 4695 | |
1617e40c JD |
4696 | ret = rbd_dev_v2_image_size(rbd_dev); |
4697 | if (ret) | |
cfbf6377 | 4698 | return ret; |
1617e40c | 4699 | |
2df3fac7 AE |
4700 | if (first_time) { |
4701 | ret = rbd_dev_v2_header_onetime(rbd_dev); | |
4702 | if (ret) | |
cfbf6377 | 4703 | return ret; |
2df3fac7 AE |
4704 | } |
4705 | ||
cc4a38bd | 4706 | ret = rbd_dev_v2_snap_context(rbd_dev); |
d194cd1d ID |
4707 | if (ret && first_time) { |
4708 | kfree(rbd_dev->header.object_prefix); | |
4709 | rbd_dev->header.object_prefix = NULL; | |
4710 | } | |
117973fb AE |
4711 | |
4712 | return ret; | |
4713 | } | |
4714 | ||
a720ae09 ID |
4715 | static int rbd_dev_header_info(struct rbd_device *rbd_dev) |
4716 | { | |
4717 | rbd_assert(rbd_image_format_valid(rbd_dev->image_format)); | |
4718 | ||
4719 | if (rbd_dev->image_format == 1) | |
4720 | return rbd_dev_v1_header_info(rbd_dev); | |
4721 | ||
4722 | return rbd_dev_v2_header_info(rbd_dev); | |
4723 | } | |
4724 | ||
1ddbe94e | 4725 | /* |
499afd5b | 4726 | * Get a unique rbd identifier for the given new rbd_dev, and add |
f8a22fc2 | 4727 | * the rbd_dev to the global list. |
1ddbe94e | 4728 | */ |
f8a22fc2 | 4729 | static int rbd_dev_id_get(struct rbd_device *rbd_dev) |
b7f23c36 | 4730 | { |
f8a22fc2 ID |
4731 | int new_dev_id; |
4732 | ||
9b60e70b ID |
4733 | new_dev_id = ida_simple_get(&rbd_dev_id_ida, |
4734 | 0, minor_to_rbd_dev_id(1 << MINORBITS), | |
4735 | GFP_KERNEL); | |
f8a22fc2 ID |
4736 | if (new_dev_id < 0) |
4737 | return new_dev_id; | |
4738 | ||
4739 | rbd_dev->dev_id = new_dev_id; | |
499afd5b AE |
4740 | |
4741 | spin_lock(&rbd_dev_list_lock); | |
4742 | list_add_tail(&rbd_dev->node, &rbd_dev_list); | |
4743 | spin_unlock(&rbd_dev_list_lock); | |
f8a22fc2 | 4744 | |
70eebd20 | 4745 | dout("rbd_dev %p given dev id %d\n", rbd_dev, rbd_dev->dev_id); |
f8a22fc2 ID |
4746 | |
4747 | return 0; | |
1ddbe94e | 4748 | } |
b7f23c36 | 4749 | |
1ddbe94e | 4750 | /* |
499afd5b AE |
4751 | * Remove an rbd_dev from the global list, and record that its |
4752 | * identifier is no longer in use. | |
1ddbe94e | 4753 | */ |
e2839308 | 4754 | static void rbd_dev_id_put(struct rbd_device *rbd_dev) |
1ddbe94e | 4755 | { |
499afd5b AE |
4756 | spin_lock(&rbd_dev_list_lock); |
4757 | list_del_init(&rbd_dev->node); | |
4758 | spin_unlock(&rbd_dev_list_lock); | |
b7f23c36 | 4759 | |
f8a22fc2 ID |
4760 | ida_simple_remove(&rbd_dev_id_ida, rbd_dev->dev_id); |
4761 | ||
4762 | dout("rbd_dev %p released dev id %d\n", rbd_dev, rbd_dev->dev_id); | |
b7f23c36 AE |
4763 | } |
4764 | ||
e28fff26 AE |
4765 | /* |
4766 | * Skips over white space at *buf, and updates *buf to point to the | |
4767 | * first found non-space character (if any). Returns the length of | |
593a9e7b AE |
4768 | * the token (string of non-white space characters) found. Note |
4769 | * that *buf must be terminated with '\0'. | |
e28fff26 AE |
4770 | */ |
4771 | static inline size_t next_token(const char **buf) | |
4772 | { | |
4773 | /* | |
4774 | * These are the characters that produce nonzero for | |
4775 | * isspace() in the "C" and "POSIX" locales. | |
4776 | */ | |
4777 | const char *spaces = " \f\n\r\t\v"; | |
4778 | ||
4779 | *buf += strspn(*buf, spaces); /* Find start of token */ | |
4780 | ||
4781 | return strcspn(*buf, spaces); /* Return token length */ | |
4782 | } | |
4783 | ||
ea3352f4 AE |
4784 | /* |
4785 | * Finds the next token in *buf, dynamically allocates a buffer big | |
4786 | * enough to hold a copy of it, and copies the token into the new | |
4787 | * buffer. The copy is guaranteed to be terminated with '\0'. Note | |
4788 | * that a duplicate buffer is created even for a zero-length token. | |
4789 | * | |
4790 | * Returns a pointer to the newly-allocated duplicate, or a null | |
4791 | * pointer if memory for the duplicate was not available. If | |
4792 | * the lenp argument is a non-null pointer, the length of the token | |
4793 | * (not including the '\0') is returned in *lenp. | |
4794 | * | |
4795 | * If successful, the *buf pointer will be updated to point beyond | |
4796 | * the end of the found token. | |
4797 | * | |
4798 | * Note: uses GFP_KERNEL for allocation. | |
4799 | */ | |
4800 | static inline char *dup_token(const char **buf, size_t *lenp) | |
4801 | { | |
4802 | char *dup; | |
4803 | size_t len; | |
4804 | ||
4805 | len = next_token(buf); | |
4caf35f9 | 4806 | dup = kmemdup(*buf, len + 1, GFP_KERNEL); |
ea3352f4 AE |
4807 | if (!dup) |
4808 | return NULL; | |
ea3352f4 AE |
4809 | *(dup + len) = '\0'; |
4810 | *buf += len; | |
4811 | ||
4812 | if (lenp) | |
4813 | *lenp = len; | |
4814 | ||
4815 | return dup; | |
4816 | } | |
4817 | ||
a725f65e | 4818 | /* |
859c31df AE |
4819 | * Parse the options provided for an "rbd add" (i.e., rbd image |
4820 | * mapping) request. These arrive via a write to /sys/bus/rbd/add, | |
4821 | * and the data written is passed here via a NUL-terminated buffer. | |
4822 | * Returns 0 if successful or an error code otherwise. | |
d22f76e7 | 4823 | * |
859c31df AE |
4824 | * The information extracted from these options is recorded in |
4825 | * the other parameters which return dynamically-allocated | |
4826 | * structures: | |
4827 | * ceph_opts | |
4828 | * The address of a pointer that will refer to a ceph options | |
4829 | * structure. Caller must release the returned pointer using | |
4830 | * ceph_destroy_options() when it is no longer needed. | |
4831 | * rbd_opts | |
4832 | * Address of an rbd options pointer. Fully initialized by | |
4833 | * this function; caller must release with kfree(). | |
4834 | * spec | |
4835 | * Address of an rbd image specification pointer. Fully | |
4836 | * initialized by this function based on parsed options. | |
4837 | * Caller must release with rbd_spec_put(). | |
4838 | * | |
4839 | * The options passed take this form: | |
4840 | * <mon_addrs> <options> <pool_name> <image_name> [<snap_id>] | |
4841 | * where: | |
4842 | * <mon_addrs> | |
4843 | * A comma-separated list of one or more monitor addresses. | |
4844 | * A monitor address is an ip address, optionally followed | |
4845 | * by a port number (separated by a colon). | |
4846 | * I.e.: ip1[:port1][,ip2[:port2]...] | |
4847 | * <options> | |
4848 | * A comma-separated list of ceph and/or rbd options. | |
4849 | * <pool_name> | |
4850 | * The name of the rados pool containing the rbd image. | |
4851 | * <image_name> | |
4852 | * The name of the image in that pool to map. | |
4853 | * <snap_id> | |
4854 | * An optional snapshot id. If provided, the mapping will | |
4855 | * present data from the image at the time that snapshot was | |
4856 | * created. The image head is used if no snapshot id is | |
4857 | * provided. Snapshot mappings are always read-only. | |
a725f65e | 4858 | */ |
859c31df | 4859 | static int rbd_add_parse_args(const char *buf, |
dc79b113 | 4860 | struct ceph_options **ceph_opts, |
859c31df AE |
4861 | struct rbd_options **opts, |
4862 | struct rbd_spec **rbd_spec) | |
e28fff26 | 4863 | { |
d22f76e7 | 4864 | size_t len; |
859c31df | 4865 | char *options; |
0ddebc0c | 4866 | const char *mon_addrs; |
ecb4dc22 | 4867 | char *snap_name; |
0ddebc0c | 4868 | size_t mon_addrs_size; |
859c31df | 4869 | struct rbd_spec *spec = NULL; |
4e9afeba | 4870 | struct rbd_options *rbd_opts = NULL; |
859c31df | 4871 | struct ceph_options *copts; |
dc79b113 | 4872 | int ret; |
e28fff26 AE |
4873 | |
4874 | /* The first four tokens are required */ | |
4875 | ||
7ef3214a | 4876 | len = next_token(&buf); |
4fb5d671 AE |
4877 | if (!len) { |
4878 | rbd_warn(NULL, "no monitor address(es) provided"); | |
4879 | return -EINVAL; | |
4880 | } | |
0ddebc0c | 4881 | mon_addrs = buf; |
f28e565a | 4882 | mon_addrs_size = len + 1; |
7ef3214a | 4883 | buf += len; |
a725f65e | 4884 | |
dc79b113 | 4885 | ret = -EINVAL; |
f28e565a AE |
4886 | options = dup_token(&buf, NULL); |
4887 | if (!options) | |
dc79b113 | 4888 | return -ENOMEM; |
4fb5d671 AE |
4889 | if (!*options) { |
4890 | rbd_warn(NULL, "no options provided"); | |
4891 | goto out_err; | |
4892 | } | |
e28fff26 | 4893 | |
859c31df AE |
4894 | spec = rbd_spec_alloc(); |
4895 | if (!spec) | |
f28e565a | 4896 | goto out_mem; |
859c31df AE |
4897 | |
4898 | spec->pool_name = dup_token(&buf, NULL); | |
4899 | if (!spec->pool_name) | |
4900 | goto out_mem; | |
4fb5d671 AE |
4901 | if (!*spec->pool_name) { |
4902 | rbd_warn(NULL, "no pool name provided"); | |
4903 | goto out_err; | |
4904 | } | |
e28fff26 | 4905 | |
69e7a02f | 4906 | spec->image_name = dup_token(&buf, NULL); |
859c31df | 4907 | if (!spec->image_name) |
f28e565a | 4908 | goto out_mem; |
4fb5d671 AE |
4909 | if (!*spec->image_name) { |
4910 | rbd_warn(NULL, "no image name provided"); | |
4911 | goto out_err; | |
4912 | } | |
d4b125e9 | 4913 | |
f28e565a AE |
4914 | /* |
4915 | * Snapshot name is optional; default is to use "-" | |
4916 | * (indicating the head/no snapshot). | |
4917 | */ | |
3feeb894 | 4918 | len = next_token(&buf); |
820a5f3e | 4919 | if (!len) { |
3feeb894 AE |
4920 | buf = RBD_SNAP_HEAD_NAME; /* No snapshot supplied */ |
4921 | len = sizeof (RBD_SNAP_HEAD_NAME) - 1; | |
f28e565a | 4922 | } else if (len > RBD_MAX_SNAP_NAME_LEN) { |
dc79b113 | 4923 | ret = -ENAMETOOLONG; |
f28e565a | 4924 | goto out_err; |
849b4260 | 4925 | } |
ecb4dc22 AE |
4926 | snap_name = kmemdup(buf, len + 1, GFP_KERNEL); |
4927 | if (!snap_name) | |
f28e565a | 4928 | goto out_mem; |
ecb4dc22 AE |
4929 | *(snap_name + len) = '\0'; |
4930 | spec->snap_name = snap_name; | |
e5c35534 | 4931 | |
0ddebc0c | 4932 | /* Initialize all rbd options to the defaults */ |
e28fff26 | 4933 | |
4e9afeba AE |
4934 | rbd_opts = kzalloc(sizeof (*rbd_opts), GFP_KERNEL); |
4935 | if (!rbd_opts) | |
4936 | goto out_mem; | |
4937 | ||
4938 | rbd_opts->read_only = RBD_READ_ONLY_DEFAULT; | |
b5584180 | 4939 | rbd_opts->queue_depth = RBD_QUEUE_DEPTH_DEFAULT; |
d22f76e7 | 4940 | |
859c31df | 4941 | copts = ceph_parse_options(options, mon_addrs, |
0ddebc0c | 4942 | mon_addrs + mon_addrs_size - 1, |
4e9afeba | 4943 | parse_rbd_opts_token, rbd_opts); |
859c31df AE |
4944 | if (IS_ERR(copts)) { |
4945 | ret = PTR_ERR(copts); | |
dc79b113 AE |
4946 | goto out_err; |
4947 | } | |
859c31df AE |
4948 | kfree(options); |
4949 | ||
4950 | *ceph_opts = copts; | |
4e9afeba | 4951 | *opts = rbd_opts; |
859c31df | 4952 | *rbd_spec = spec; |
0ddebc0c | 4953 | |
dc79b113 | 4954 | return 0; |
f28e565a | 4955 | out_mem: |
dc79b113 | 4956 | ret = -ENOMEM; |
d22f76e7 | 4957 | out_err: |
859c31df AE |
4958 | kfree(rbd_opts); |
4959 | rbd_spec_put(spec); | |
f28e565a | 4960 | kfree(options); |
d22f76e7 | 4961 | |
dc79b113 | 4962 | return ret; |
a725f65e AE |
4963 | } |
4964 | ||
30ba1f02 ID |
4965 | /* |
4966 | * Return pool id (>= 0) or a negative error code. | |
4967 | */ | |
4968 | static int rbd_add_get_pool_id(struct rbd_client *rbdc, const char *pool_name) | |
4969 | { | |
a319bf56 | 4970 | struct ceph_options *opts = rbdc->client->options; |
30ba1f02 | 4971 | u64 newest_epoch; |
30ba1f02 ID |
4972 | int tries = 0; |
4973 | int ret; | |
4974 | ||
4975 | again: | |
4976 | ret = ceph_pg_poolid_by_name(rbdc->client->osdc.osdmap, pool_name); | |
4977 | if (ret == -ENOENT && tries++ < 1) { | |
4978 | ret = ceph_monc_do_get_version(&rbdc->client->monc, "osdmap", | |
4979 | &newest_epoch); | |
4980 | if (ret < 0) | |
4981 | return ret; | |
4982 | ||
4983 | if (rbdc->client->osdc.osdmap->epoch < newest_epoch) { | |
4984 | ceph_monc_request_next_osdmap(&rbdc->client->monc); | |
4985 | (void) ceph_monc_wait_osdmap(&rbdc->client->monc, | |
a319bf56 ID |
4986 | newest_epoch, |
4987 | opts->mount_timeout); | |
30ba1f02 ID |
4988 | goto again; |
4989 | } else { | |
4990 | /* the osdmap we have is new enough */ | |
4991 | return -ENOENT; | |
4992 | } | |
4993 | } | |
4994 | ||
4995 | return ret; | |
4996 | } | |
4997 | ||
589d30e0 AE |
4998 | /* |
4999 | * An rbd format 2 image has a unique identifier, distinct from the | |
5000 | * name given to it by the user. Internally, that identifier is | |
5001 | * what's used to specify the names of objects related to the image. | |
5002 | * | |
5003 | * A special "rbd id" object is used to map an rbd image name to its | |
5004 | * id. If that object doesn't exist, then there is no v2 rbd image | |
5005 | * with the supplied name. | |
5006 | * | |
5007 | * This function will record the given rbd_dev's image_id field if | |
5008 | * it can be determined, and in that case will return 0. If any | |
5009 | * errors occur a negative errno will be returned and the rbd_dev's | |
5010 | * image_id field will be unchanged (and should be NULL). | |
5011 | */ | |
5012 | static int rbd_dev_image_id(struct rbd_device *rbd_dev) | |
5013 | { | |
5014 | int ret; | |
5015 | size_t size; | |
5016 | char *object_name; | |
5017 | void *response; | |
c0fba368 | 5018 | char *image_id; |
2f82ee54 | 5019 | |
2c0d0a10 AE |
5020 | /* |
5021 | * When probing a parent image, the image id is already | |
5022 | * known (and the image name likely is not). There's no | |
c0fba368 AE |
5023 | * need to fetch the image id again in this case. We |
5024 | * do still need to set the image format though. | |
2c0d0a10 | 5025 | */ |
c0fba368 AE |
5026 | if (rbd_dev->spec->image_id) { |
5027 | rbd_dev->image_format = *rbd_dev->spec->image_id ? 2 : 1; | |
5028 | ||
2c0d0a10 | 5029 | return 0; |
c0fba368 | 5030 | } |
2c0d0a10 | 5031 | |
589d30e0 AE |
5032 | /* |
5033 | * First, see if the format 2 image id file exists, and if | |
5034 | * so, get the image's persistent id from it. | |
5035 | */ | |
69e7a02f | 5036 | size = sizeof (RBD_ID_PREFIX) + strlen(rbd_dev->spec->image_name); |
589d30e0 AE |
5037 | object_name = kmalloc(size, GFP_NOIO); |
5038 | if (!object_name) | |
5039 | return -ENOMEM; | |
0d7dbfce | 5040 | sprintf(object_name, "%s%s", RBD_ID_PREFIX, rbd_dev->spec->image_name); |
589d30e0 AE |
5041 | dout("rbd id object name is %s\n", object_name); |
5042 | ||
5043 | /* Response will be an encoded string, which includes a length */ | |
5044 | ||
5045 | size = sizeof (__le32) + RBD_IMAGE_ID_LEN_MAX; | |
5046 | response = kzalloc(size, GFP_NOIO); | |
5047 | if (!response) { | |
5048 | ret = -ENOMEM; | |
5049 | goto out; | |
5050 | } | |
5051 | ||
c0fba368 AE |
5052 | /* If it doesn't exist we'll assume it's a format 1 image */ |
5053 | ||
36be9a76 | 5054 | ret = rbd_obj_method_sync(rbd_dev, object_name, |
4157976b | 5055 | "rbd", "get_id", NULL, 0, |
e2a58ee5 | 5056 | response, RBD_IMAGE_ID_LEN_MAX); |
36be9a76 | 5057 | dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret); |
c0fba368 AE |
5058 | if (ret == -ENOENT) { |
5059 | image_id = kstrdup("", GFP_KERNEL); | |
5060 | ret = image_id ? 0 : -ENOMEM; | |
5061 | if (!ret) | |
5062 | rbd_dev->image_format = 1; | |
7dd440c9 | 5063 | } else if (ret >= 0) { |
c0fba368 AE |
5064 | void *p = response; |
5065 | ||
5066 | image_id = ceph_extract_encoded_string(&p, p + ret, | |
979ed480 | 5067 | NULL, GFP_NOIO); |
461f758a | 5068 | ret = PTR_ERR_OR_ZERO(image_id); |
c0fba368 AE |
5069 | if (!ret) |
5070 | rbd_dev->image_format = 2; | |
c0fba368 AE |
5071 | } |
5072 | ||
5073 | if (!ret) { | |
5074 | rbd_dev->spec->image_id = image_id; | |
5075 | dout("image_id is %s\n", image_id); | |
589d30e0 AE |
5076 | } |
5077 | out: | |
5078 | kfree(response); | |
5079 | kfree(object_name); | |
5080 | ||
5081 | return ret; | |
5082 | } | |
5083 | ||
3abef3b3 AE |
5084 | /* |
5085 | * Undo whatever state changes are made by v1 or v2 header info | |
5086 | * call. | |
5087 | */ | |
6fd48b3b AE |
5088 | static void rbd_dev_unprobe(struct rbd_device *rbd_dev) |
5089 | { | |
5090 | struct rbd_image_header *header; | |
5091 | ||
e69b8d41 | 5092 | rbd_dev_parent_put(rbd_dev); |
6fd48b3b AE |
5093 | |
5094 | /* Free dynamic fields from the header, then zero it out */ | |
5095 | ||
5096 | header = &rbd_dev->header; | |
812164f8 | 5097 | ceph_put_snap_context(header->snapc); |
6fd48b3b AE |
5098 | kfree(header->snap_sizes); |
5099 | kfree(header->snap_names); | |
5100 | kfree(header->object_prefix); | |
5101 | memset(header, 0, sizeof (*header)); | |
5102 | } | |
5103 | ||
2df3fac7 | 5104 | static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) |
a30b71b9 AE |
5105 | { |
5106 | int ret; | |
a30b71b9 | 5107 | |
1e130199 | 5108 | ret = rbd_dev_v2_object_prefix(rbd_dev); |
57385b51 | 5109 | if (ret) |
b1b5402a AE |
5110 | goto out_err; |
5111 | ||
2df3fac7 AE |
5112 | /* |
5113 | * Get the and check features for the image. Currently the | |
5114 | * features are assumed to never change. | |
5115 | */ | |
b1b5402a | 5116 | ret = rbd_dev_v2_features(rbd_dev); |
57385b51 | 5117 | if (ret) |
9d475de5 | 5118 | goto out_err; |
35d489f9 | 5119 | |
cc070d59 AE |
5120 | /* If the image supports fancy striping, get its parameters */ |
5121 | ||
5122 | if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) { | |
5123 | ret = rbd_dev_v2_striping_info(rbd_dev); | |
5124 | if (ret < 0) | |
5125 | goto out_err; | |
5126 | } | |
2df3fac7 | 5127 | /* No support for crypto and compression type format 2 images */ |
a30b71b9 | 5128 | |
35152979 | 5129 | return 0; |
9d475de5 | 5130 | out_err: |
642a2537 | 5131 | rbd_dev->header.features = 0; |
1e130199 AE |
5132 | kfree(rbd_dev->header.object_prefix); |
5133 | rbd_dev->header.object_prefix = NULL; | |
9d475de5 AE |
5134 | |
5135 | return ret; | |
a30b71b9 AE |
5136 | } |
5137 | ||
6d69bb53 ID |
5138 | /* |
5139 | * @depth is rbd_dev_image_probe() -> rbd_dev_probe_parent() -> | |
5140 | * rbd_dev_image_probe() recursion depth, which means it's also the | |
5141 | * length of the already discovered part of the parent chain. | |
5142 | */ | |
5143 | static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth) | |
83a06263 | 5144 | { |
2f82ee54 | 5145 | struct rbd_device *parent = NULL; |
124afba2 AE |
5146 | int ret; |
5147 | ||
5148 | if (!rbd_dev->parent_spec) | |
5149 | return 0; | |
124afba2 | 5150 | |
6d69bb53 ID |
5151 | if (++depth > RBD_MAX_PARENT_CHAIN_LEN) { |
5152 | pr_info("parent chain is too long (%d)\n", depth); | |
5153 | ret = -EINVAL; | |
5154 | goto out_err; | |
5155 | } | |
5156 | ||
1f2c6651 ID |
5157 | parent = rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec, |
5158 | NULL); | |
5159 | if (!parent) { | |
5160 | ret = -ENOMEM; | |
124afba2 | 5161 | goto out_err; |
1f2c6651 ID |
5162 | } |
5163 | ||
5164 | /* | |
5165 | * Images related by parent/child relationships always share | |
5166 | * rbd_client and spec/parent_spec, so bump their refcounts. | |
5167 | */ | |
5168 | __rbd_get_client(rbd_dev->rbd_client); | |
5169 | rbd_spec_get(rbd_dev->parent_spec); | |
124afba2 | 5170 | |
6d69bb53 | 5171 | ret = rbd_dev_image_probe(parent, depth); |
124afba2 AE |
5172 | if (ret < 0) |
5173 | goto out_err; | |
1f2c6651 | 5174 | |
124afba2 | 5175 | rbd_dev->parent = parent; |
a2acd00e | 5176 | atomic_set(&rbd_dev->parent_ref, 1); |
124afba2 | 5177 | return 0; |
1f2c6651 | 5178 | |
124afba2 | 5179 | out_err: |
1f2c6651 | 5180 | rbd_dev_unparent(rbd_dev); |
1761b229 | 5181 | rbd_dev_destroy(parent); |
124afba2 AE |
5182 | return ret; |
5183 | } | |
5184 | ||
811c6688 ID |
5185 | /* |
5186 | * rbd_dev->header_rwsem must be locked for write and will be unlocked | |
5187 | * upon return. | |
5188 | */ | |
200a6a8b | 5189 | static int rbd_dev_device_setup(struct rbd_device *rbd_dev) |
124afba2 | 5190 | { |
83a06263 | 5191 | int ret; |
d1cf5788 | 5192 | |
f8a22fc2 ID |
5193 | /* Get an id and fill in device name. */ |
5194 | ||
5195 | ret = rbd_dev_id_get(rbd_dev); | |
5196 | if (ret) | |
811c6688 | 5197 | goto err_out_unlock; |
83a06263 | 5198 | |
83a06263 AE |
5199 | BUILD_BUG_ON(DEV_NAME_LEN |
5200 | < sizeof (RBD_DRV_NAME) + MAX_INT_FORMAT_WIDTH); | |
5201 | sprintf(rbd_dev->name, "%s%d", RBD_DRV_NAME, rbd_dev->dev_id); | |
5202 | ||
9b60e70b | 5203 | /* Record our major and minor device numbers. */ |
83a06263 | 5204 | |
9b60e70b ID |
5205 | if (!single_major) { |
5206 | ret = register_blkdev(0, rbd_dev->name); | |
5207 | if (ret < 0) | |
5208 | goto err_out_id; | |
5209 | ||
5210 | rbd_dev->major = ret; | |
5211 | rbd_dev->minor = 0; | |
5212 | } else { | |
5213 | rbd_dev->major = rbd_major; | |
5214 | rbd_dev->minor = rbd_dev_id_to_minor(rbd_dev->dev_id); | |
5215 | } | |
83a06263 AE |
5216 | |
5217 | /* Set up the blkdev mapping. */ | |
5218 | ||
5219 | ret = rbd_init_disk(rbd_dev); | |
5220 | if (ret) | |
5221 | goto err_out_blkdev; | |
5222 | ||
f35a4dee | 5223 | ret = rbd_dev_mapping_set(rbd_dev); |
83a06263 AE |
5224 | if (ret) |
5225 | goto err_out_disk; | |
bc1ecc65 | 5226 | |
f35a4dee | 5227 | set_capacity(rbd_dev->disk, rbd_dev->mapping.size / SECTOR_SIZE); |
22001f61 | 5228 | set_disk_ro(rbd_dev->disk, rbd_dev->mapping.read_only); |
f35a4dee | 5229 | |
dd5ac32d ID |
5230 | dev_set_name(&rbd_dev->dev, "%d", rbd_dev->dev_id); |
5231 | ret = device_add(&rbd_dev->dev); | |
f35a4dee | 5232 | if (ret) |
f5ee37bd | 5233 | goto err_out_mapping; |
83a06263 | 5234 | |
83a06263 AE |
5235 | /* Everything's ready. Announce the disk to the world. */ |
5236 | ||
129b79d4 | 5237 | set_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
811c6688 | 5238 | up_write(&rbd_dev->header_rwsem); |
83a06263 | 5239 | |
811c6688 | 5240 | add_disk(rbd_dev->disk); |
83a06263 AE |
5241 | pr_info("%s: added with size 0x%llx\n", rbd_dev->disk->disk_name, |
5242 | (unsigned long long) rbd_dev->mapping.size); | |
5243 | ||
5244 | return ret; | |
2f82ee54 | 5245 | |
f35a4dee AE |
5246 | err_out_mapping: |
5247 | rbd_dev_mapping_clear(rbd_dev); | |
83a06263 AE |
5248 | err_out_disk: |
5249 | rbd_free_disk(rbd_dev); | |
5250 | err_out_blkdev: | |
9b60e70b ID |
5251 | if (!single_major) |
5252 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
83a06263 AE |
5253 | err_out_id: |
5254 | rbd_dev_id_put(rbd_dev); | |
811c6688 ID |
5255 | err_out_unlock: |
5256 | up_write(&rbd_dev->header_rwsem); | |
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 | |
811c6688 | 5447 | down_write(&rbd_dev->header_rwsem); |
6d69bb53 | 5448 | rc = rbd_dev_image_probe(rbd_dev, 0); |
a30b71b9 | 5449 | if (rc < 0) |
c53d5893 | 5450 | goto err_out_rbd_dev; |
05fd6f6f | 5451 | |
7ce4eef7 AE |
5452 | /* If we are mapping a snapshot it must be marked read-only */ |
5453 | ||
d147543d | 5454 | read_only = rbd_dev->opts->read_only; |
7ce4eef7 AE |
5455 | if (rbd_dev->spec->snap_id != CEPH_NOSNAP) |
5456 | read_only = true; | |
5457 | rbd_dev->mapping.read_only = read_only; | |
5458 | ||
b536f69a | 5459 | rc = rbd_dev_device_setup(rbd_dev); |
3abef3b3 | 5460 | if (rc) { |
e37180c0 ID |
5461 | /* |
5462 | * rbd_dev_header_unwatch_sync() can't be moved into | |
5463 | * rbd_dev_image_release() without refactoring, see | |
5464 | * commit 1f3ef78861ac. | |
5465 | */ | |
5466 | rbd_dev_header_unwatch_sync(rbd_dev); | |
3abef3b3 | 5467 | rbd_dev_image_release(rbd_dev); |
dd5ac32d | 5468 | goto out; |
3abef3b3 AE |
5469 | } |
5470 | ||
dd5ac32d ID |
5471 | rc = count; |
5472 | out: | |
5473 | module_put(THIS_MODULE); | |
5474 | return rc; | |
b536f69a | 5475 | |
c53d5893 | 5476 | err_out_rbd_dev: |
811c6688 | 5477 | up_write(&rbd_dev->header_rwsem); |
c53d5893 | 5478 | rbd_dev_destroy(rbd_dev); |
bd4ba655 | 5479 | err_out_client: |
9d3997fd | 5480 | rbd_put_client(rbdc); |
0ddebc0c | 5481 | err_out_args: |
859c31df | 5482 | rbd_spec_put(spec); |
d147543d | 5483 | kfree(rbd_opts); |
dd5ac32d | 5484 | goto out; |
602adf40 YS |
5485 | } |
5486 | ||
9b60e70b ID |
5487 | static ssize_t rbd_add(struct bus_type *bus, |
5488 | const char *buf, | |
5489 | size_t count) | |
5490 | { | |
5491 | if (single_major) | |
5492 | return -EINVAL; | |
5493 | ||
5494 | return do_rbd_add(bus, buf, count); | |
5495 | } | |
5496 | ||
5497 | static ssize_t rbd_add_single_major(struct bus_type *bus, | |
5498 | const char *buf, | |
5499 | size_t count) | |
5500 | { | |
5501 | return do_rbd_add(bus, buf, count); | |
5502 | } | |
5503 | ||
dd5ac32d | 5504 | static void rbd_dev_device_release(struct rbd_device *rbd_dev) |
602adf40 | 5505 | { |
602adf40 | 5506 | rbd_free_disk(rbd_dev); |
200a6a8b | 5507 | clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags); |
dd5ac32d | 5508 | device_del(&rbd_dev->dev); |
6d80b130 | 5509 | rbd_dev_mapping_clear(rbd_dev); |
9b60e70b ID |
5510 | if (!single_major) |
5511 | unregister_blkdev(rbd_dev->major, rbd_dev->name); | |
e2839308 | 5512 | rbd_dev_id_put(rbd_dev); |
602adf40 YS |
5513 | } |
5514 | ||
05a46afd AE |
5515 | static void rbd_dev_remove_parent(struct rbd_device *rbd_dev) |
5516 | { | |
ad945fc1 | 5517 | while (rbd_dev->parent) { |
05a46afd AE |
5518 | struct rbd_device *first = rbd_dev; |
5519 | struct rbd_device *second = first->parent; | |
5520 | struct rbd_device *third; | |
5521 | ||
5522 | /* | |
5523 | * Follow to the parent with no grandparent and | |
5524 | * remove it. | |
5525 | */ | |
5526 | while (second && (third = second->parent)) { | |
5527 | first = second; | |
5528 | second = third; | |
5529 | } | |
ad945fc1 | 5530 | rbd_assert(second); |
8ad42cd0 | 5531 | rbd_dev_image_release(second); |
ad945fc1 AE |
5532 | first->parent = NULL; |
5533 | first->parent_overlap = 0; | |
5534 | ||
5535 | rbd_assert(first->parent_spec); | |
05a46afd AE |
5536 | rbd_spec_put(first->parent_spec); |
5537 | first->parent_spec = NULL; | |
05a46afd AE |
5538 | } |
5539 | } | |
5540 | ||
9b60e70b ID |
5541 | static ssize_t do_rbd_remove(struct bus_type *bus, |
5542 | const char *buf, | |
5543 | size_t count) | |
602adf40 YS |
5544 | { |
5545 | struct rbd_device *rbd_dev = NULL; | |
751cc0e3 AE |
5546 | struct list_head *tmp; |
5547 | int dev_id; | |
602adf40 | 5548 | unsigned long ul; |
82a442d2 | 5549 | bool already = false; |
0d8189e1 | 5550 | int ret; |
602adf40 | 5551 | |
bb8e0e84 | 5552 | ret = kstrtoul(buf, 10, &ul); |
0d8189e1 AE |
5553 | if (ret) |
5554 | return ret; | |
602adf40 YS |
5555 | |
5556 | /* convert to int; abort if we lost anything in the conversion */ | |
751cc0e3 AE |
5557 | dev_id = (int)ul; |
5558 | if (dev_id != ul) | |
602adf40 YS |
5559 | return -EINVAL; |
5560 | ||
751cc0e3 AE |
5561 | ret = -ENOENT; |
5562 | spin_lock(&rbd_dev_list_lock); | |
5563 | list_for_each(tmp, &rbd_dev_list) { | |
5564 | rbd_dev = list_entry(tmp, struct rbd_device, node); | |
5565 | if (rbd_dev->dev_id == dev_id) { | |
5566 | ret = 0; | |
5567 | break; | |
5568 | } | |
42382b70 | 5569 | } |
751cc0e3 AE |
5570 | if (!ret) { |
5571 | spin_lock_irq(&rbd_dev->lock); | |
5572 | if (rbd_dev->open_count) | |
5573 | ret = -EBUSY; | |
5574 | else | |
82a442d2 AE |
5575 | already = test_and_set_bit(RBD_DEV_FLAG_REMOVING, |
5576 | &rbd_dev->flags); | |
751cc0e3 AE |
5577 | spin_unlock_irq(&rbd_dev->lock); |
5578 | } | |
5579 | spin_unlock(&rbd_dev_list_lock); | |
82a442d2 | 5580 | if (ret < 0 || already) |
1ba0f1e7 | 5581 | return ret; |
751cc0e3 | 5582 | |
fca27065 | 5583 | rbd_dev_header_unwatch_sync(rbd_dev); |
fca27065 | 5584 | |
9875201e JD |
5585 | /* |
5586 | * Don't free anything from rbd_dev->disk until after all | |
5587 | * notifies are completely processed. Otherwise | |
5588 | * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting | |
5589 | * in a potential use after free of rbd_dev->disk or rbd_dev. | |
5590 | */ | |
dd5ac32d | 5591 | rbd_dev_device_release(rbd_dev); |
8ad42cd0 | 5592 | rbd_dev_image_release(rbd_dev); |
aafb230e | 5593 | |
1ba0f1e7 | 5594 | return count; |
602adf40 YS |
5595 | } |
5596 | ||
9b60e70b ID |
5597 | static ssize_t rbd_remove(struct bus_type *bus, |
5598 | const char *buf, | |
5599 | size_t count) | |
5600 | { | |
5601 | if (single_major) | |
5602 | return -EINVAL; | |
5603 | ||
5604 | return do_rbd_remove(bus, buf, count); | |
5605 | } | |
5606 | ||
5607 | static ssize_t rbd_remove_single_major(struct bus_type *bus, | |
5608 | const char *buf, | |
5609 | size_t count) | |
5610 | { | |
5611 | return do_rbd_remove(bus, buf, count); | |
5612 | } | |
5613 | ||
602adf40 YS |
5614 | /* |
5615 | * create control files in sysfs | |
dfc5606d | 5616 | * /sys/bus/rbd/... |
602adf40 YS |
5617 | */ |
5618 | static int rbd_sysfs_init(void) | |
5619 | { | |
dfc5606d | 5620 | int ret; |
602adf40 | 5621 | |
fed4c143 | 5622 | ret = device_register(&rbd_root_dev); |
21079786 | 5623 | if (ret < 0) |
dfc5606d | 5624 | return ret; |
602adf40 | 5625 | |
fed4c143 AE |
5626 | ret = bus_register(&rbd_bus_type); |
5627 | if (ret < 0) | |
5628 | device_unregister(&rbd_root_dev); | |
602adf40 | 5629 | |
602adf40 YS |
5630 | return ret; |
5631 | } | |
5632 | ||
5633 | static void rbd_sysfs_cleanup(void) | |
5634 | { | |
dfc5606d | 5635 | bus_unregister(&rbd_bus_type); |
fed4c143 | 5636 | device_unregister(&rbd_root_dev); |
602adf40 YS |
5637 | } |
5638 | ||
1c2a9dfe AE |
5639 | static int rbd_slab_init(void) |
5640 | { | |
5641 | rbd_assert(!rbd_img_request_cache); | |
03d94406 | 5642 | rbd_img_request_cache = KMEM_CACHE(rbd_img_request, 0); |
868311b1 AE |
5643 | if (!rbd_img_request_cache) |
5644 | return -ENOMEM; | |
5645 | ||
5646 | rbd_assert(!rbd_obj_request_cache); | |
03d94406 | 5647 | rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0); |
78c2a44a AE |
5648 | if (!rbd_obj_request_cache) |
5649 | goto out_err; | |
5650 | ||
5651 | rbd_assert(!rbd_segment_name_cache); | |
5652 | rbd_segment_name_cache = kmem_cache_create("rbd_segment_name", | |
2d0ebc5d | 5653 | CEPH_MAX_OID_NAME_LEN + 1, 1, 0, NULL); |
78c2a44a | 5654 | if (rbd_segment_name_cache) |
1c2a9dfe | 5655 | return 0; |
78c2a44a | 5656 | out_err: |
13bf2834 JL |
5657 | kmem_cache_destroy(rbd_obj_request_cache); |
5658 | rbd_obj_request_cache = NULL; | |
1c2a9dfe | 5659 | |
868311b1 AE |
5660 | kmem_cache_destroy(rbd_img_request_cache); |
5661 | rbd_img_request_cache = NULL; | |
5662 | ||
1c2a9dfe AE |
5663 | return -ENOMEM; |
5664 | } | |
5665 | ||
5666 | static void rbd_slab_exit(void) | |
5667 | { | |
78c2a44a AE |
5668 | rbd_assert(rbd_segment_name_cache); |
5669 | kmem_cache_destroy(rbd_segment_name_cache); | |
5670 | rbd_segment_name_cache = NULL; | |
5671 | ||
868311b1 AE |
5672 | rbd_assert(rbd_obj_request_cache); |
5673 | kmem_cache_destroy(rbd_obj_request_cache); | |
5674 | rbd_obj_request_cache = NULL; | |
5675 | ||
1c2a9dfe AE |
5676 | rbd_assert(rbd_img_request_cache); |
5677 | kmem_cache_destroy(rbd_img_request_cache); | |
5678 | rbd_img_request_cache = NULL; | |
5679 | } | |
5680 | ||
cc344fa1 | 5681 | static int __init rbd_init(void) |
602adf40 YS |
5682 | { |
5683 | int rc; | |
5684 | ||
1e32d34c AE |
5685 | if (!libceph_compatible(NULL)) { |
5686 | rbd_warn(NULL, "libceph incompatibility (quitting)"); | |
1e32d34c AE |
5687 | return -EINVAL; |
5688 | } | |
e1b4d96d | 5689 | |
1c2a9dfe | 5690 | rc = rbd_slab_init(); |
602adf40 YS |
5691 | if (rc) |
5692 | return rc; | |
e1b4d96d | 5693 | |
f5ee37bd ID |
5694 | /* |
5695 | * The number of active work items is limited by the number of | |
f77303bd | 5696 | * rbd devices * queue depth, so leave @max_active at default. |
f5ee37bd ID |
5697 | */ |
5698 | rbd_wq = alloc_workqueue(RBD_DRV_NAME, WQ_MEM_RECLAIM, 0); | |
5699 | if (!rbd_wq) { | |
5700 | rc = -ENOMEM; | |
5701 | goto err_out_slab; | |
5702 | } | |
5703 | ||
9b60e70b ID |
5704 | if (single_major) { |
5705 | rbd_major = register_blkdev(0, RBD_DRV_NAME); | |
5706 | if (rbd_major < 0) { | |
5707 | rc = rbd_major; | |
f5ee37bd | 5708 | goto err_out_wq; |
9b60e70b ID |
5709 | } |
5710 | } | |
5711 | ||
1c2a9dfe AE |
5712 | rc = rbd_sysfs_init(); |
5713 | if (rc) | |
9b60e70b ID |
5714 | goto err_out_blkdev; |
5715 | ||
5716 | if (single_major) | |
5717 | pr_info("loaded (major %d)\n", rbd_major); | |
5718 | else | |
5719 | pr_info("loaded\n"); | |
1c2a9dfe | 5720 | |
e1b4d96d ID |
5721 | return 0; |
5722 | ||
9b60e70b ID |
5723 | err_out_blkdev: |
5724 | if (single_major) | |
5725 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd ID |
5726 | err_out_wq: |
5727 | destroy_workqueue(rbd_wq); | |
e1b4d96d ID |
5728 | err_out_slab: |
5729 | rbd_slab_exit(); | |
1c2a9dfe | 5730 | return rc; |
602adf40 YS |
5731 | } |
5732 | ||
cc344fa1 | 5733 | static void __exit rbd_exit(void) |
602adf40 | 5734 | { |
ffe312cf | 5735 | ida_destroy(&rbd_dev_id_ida); |
602adf40 | 5736 | rbd_sysfs_cleanup(); |
9b60e70b ID |
5737 | if (single_major) |
5738 | unregister_blkdev(rbd_major, RBD_DRV_NAME); | |
f5ee37bd | 5739 | destroy_workqueue(rbd_wq); |
1c2a9dfe | 5740 | rbd_slab_exit(); |
602adf40 YS |
5741 | } |
5742 | ||
5743 | module_init(rbd_init); | |
5744 | module_exit(rbd_exit); | |
5745 | ||
d552c619 | 5746 | MODULE_AUTHOR("Alex Elder <elder@inktank.com>"); |
602adf40 YS |
5747 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); |
5748 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
602adf40 YS |
5749 | /* following authorship retained from original osdblk.c */ |
5750 | MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>"); | |
5751 | ||
90da258b | 5752 | MODULE_DESCRIPTION("RADOS Block Device (RBD) driver"); |
602adf40 | 5753 | MODULE_LICENSE("GPL"); |